UNIVERSIDAD SIMN BOLVAR

DECANATO DE ESTUDIOS PROFESIONALES

COORDINACIN DE INGENIERA ELCTRICA
APLICACIN DE FLUJO DE CARGA DIRECTO A
REDES DE DISTRIBUCIN DE GRAN TAMAO
POR
HENRY ROMAN ESCOBAR MELGAREJO
JOS RAFAEL PREZ ZORRILLA
PROYECTO DE GRADO
PRESENTADO ANTE LA ILUSTRE UNIVERSIDAD SIMN BOLVAR
COMO REQUISITO PARCIAL PARA OPTAR AL TTULO DE
INGENIERO ELECTRICISTA
Sartenejas, Noviembre de 2010
UNIVERSIDAD SIMN BOLVAR
DECANATO DE ESTUDIOS PROFESIONALES
COORDINACIN DE INGENIERA ELCTRICA
APLICACIN DE FLUJO DE CARGA DIRECTO A
REDES DE DISTRIBUCIN DE GRAN TAMAO
POR
HENRY ROMAN ESCOBAR MELGAREJO
JOS RAFAEL PREZ ZORRILLA
TUTOR:PROF. PAULO DE OLIVEIRA
PROYECTO DE GRADO
PRESENTADO ANTE LA ILUSTRE UNIVERSIDAD SIMN BOLVAR
COMO REQUISITO PARCIAL PARA OPTAR AL TTULO DE
INGENIERO ELECTRICISTA
Sartenejas, Noviembre de 2010

UNIVERSIDAD SIMN BOLVAR

Decanato de Estudios Profesionales
Coordinacin de Ingeniera Elctrica
ACTA DE EVALUACIN DEL PROYECTO DE GRADO
CDIGO DE LA ASIGNATURA: EP 370 (
ESTUDIANTES:1:l EN ~ 1 ESO~A 'P1 . CARNET: D<{ 3(; 04 (
JOSf 'ft ~b2 CARNET: 03 3(,313
TTULODELTRABAJO:+~\\Ce--.C\; JQ -C\u~o ~e. ce-...~Oc...
\.ce..c\o c;-... '~~<2-S e. ,sA\\~uc,;<') ~Q 'n,c....", ~.,-..f'\c..;\C>
--:::::>
TUTOR: Prof.\co,-..>\.o ~-'< a\ '-.)Q.\,c...... CO-TUTOR: Prof. _
JURADo:profs.~A1/1JO t\GUl?1AS y ftP~O 'E_A_\V~A __
APROBADO: ~ REPROBADO: D
OBSERV ACIONES:

~: ,1
i
Co-Tutor
Jurado
Nota: Colocar los sellos de los respectiyos Departamentos. Para jurados externos, usar el sello de la
Coordinacin.
APLICACIN DE FLUJO DE CARGA DIRECTO A REDES DE DISTRIBUCIN DE
GRAN TAMAO
Por:
HENRY ROMAN ESCOBAR MELGAREJO
JOS RAFAEL PREZ ZORRILA
RESUMEN
En este trabajo se implementa un anlisis de ujo de carga, (FDC), para ser aplicado
en redes de distribucin de gran tamao. El FDC implementado es un mtodo de barrido
unidireccional directo denido por el uso de una matriz nica que caracteriza la resistencia y
admitancia de las lneas y la topologa de la red. El algoritmo fue codicado en MATLAB. El
mismo fue desarrollado en forma modular utilizando un patrn de diseo de software Modelo-
Vista-Controlador, (MVC). El FDC implementado en el presente trabajo est basado en Teng
(2003) y se incluye la modicacin propuesta por De Oliveira (2010) para utilizar nmeros
reales en lugar de nmeros complejos. Como contribucin principal se establece un esquema de
datos en RAM (Random Access Memory), lo cual optimiza el proceso de clculo del FDC. El
algoritmo fue validado con redes de 4, 7, 12 y 69 barras. Adems, se vericaron los resultados
del circuito de 7 barras con un Newton-Raphson (NR), (MATPOWER). Finalmente, se aplic
exitosamente en una muestra signicativa de la red de la Gran Caracas; conformada por 530
circuitos distribuidos en 78 S/E y 64.251 nodos, que corresponden al 70 % de la demanda de
las Gran Caracas. El resultado del FDC para el 70 % de la Gran Caracas se obtuvo en 0, 76
segundos. El FDC directo estudiado es un mtodo eciente con un gran potencial para ser
aplicado en el anlisis y planicacin del Sistema Elctrico de Distribucin (SDEE).
iv
A mi madre; Gloria, por su innito amor y sabidura.
a mi padre; Henry, por su esfuerzo.
Henry Escobar.
A mi madre; Olivia, por su apoyo y constancia.
Jos Prez.
v
AGRADECIMIENTOS
Quisiera agradecer a Dios y a las tres personas que inuyeron en quien soy hoy en da: mi
madre, mi padre y mi abuela Pancha. Gracias a su esfuerzo es que he llegado hasta aqu.
A mis tas Carmen y Zulay que me vieron crecer y siempre han estado apoyndome.
A mi abuelo Santiago y mis tos con los que siempre he contado y podr contar en cualquier
momento.
A mi hermano, mis primos y mi sobrina con los que he crecido y he compartido valiosos
momentos de mi vida.
A mis amigos y compaeros del equipo de rugby subacutico CONGRIOS USB con los
que comparto el vicio por el agua y la adrenalina.
A mis amigos: Martha, Gabriela, Debora, Leopold, Ariaam, Fedora y Jos Rafael. Por ser
unos excelentes amigos y personas.
A todos; no existen palabras sucientes para agradecer lo que han hecho por m.
Todo es cuestin de actitud!!!
Henry Escobar.
Quisiera agradecer a mi madre por ser la persona que siempre ha estado conmigo.
Gracias!!!
Jos Rafael Prez.
Gracias al Prof. Paulo De Oliveira por ayudarnos y apoyarnos en todo momento durante la
tesis. Gracias a Benicia, Mara Teresa y al Prof. Miguel Martnez por estar a la orden cada
vez que necesitbamos ayuda. Gracias a Angel y Luis Gerardo por darnos una mano justo
cuando era necesario.
Gracias Totales.
vi
LISTA DE SMBOLOS
i, j Nodos del sistema.
n Nmero de nodos del sistema.
P
i
Potencia activa en el nodo i.
Q
i
Potencia reactiva en el nodo i.
V
i
Tensin en el nodo i.
V
j
Tensin en el nodo j.
G
ii
Conductancia propia en el nodo i.
B
ii
Susceptancia propia en el nodo i.
Y
ij
Admitancia entre los nodos i-j.

ij
ngulo de la admitancia de una lnea entre los nodos i-j.

i
ngulo de la tensin en el nodo i.

j
ngulo de la tensin en el nodo j.
s Nodo emisor.
r Nodo receptor.
P
r
Potencia activa en el nodo r.
Q
r
Potencia reactiva en el nodo r.
P
s
Potencia activa en el nodo s.
Q
s
Potencia reactiva en el nodo s.
V
s
Tensin en el nodo s.
V
r
Tensin en el nodo r.
Z Impedancia de la lnea entre los nodos s-r.
vii
R Resistencia de la lnea entre los nodos s-r.
X Reactancia de la lnea entre los nodos s-r.

Z
ngulo de la impedancia de la lnea entre los nodos s-r.

s
ngulo de la tensin en el nodo s.

r
ngulo de la tensin en el nodo r.
V
s
Tensin en el nodo s forma compleja.
V
r
Tensin en el nodo r forma compleja.
I
s
Corriente en el nodo s forma compleja.

Z Impedancia de la lnea entre los nodos s-r en forma compleja.

BIBC Matriz de corrientes inyectadas a corrientes de rama (Bus Injection to Branch
Current), [19].
BCBV Matriz de corrientes de rama a voltajes nodales (Branch Current to Bus Voltage),
[19].
DLF Matriz de ujo de carga (Distribution Load Flow), [19].
K Iteracin K-sima.
[I
K
] Vector columna de corrientes inyectadas.
V
K+1
Vector de variacin de tensiones nodales en la iteracin K + 1.
V
K+1
Vector de tensiones nodales en la iteracin K + 1.
V
0
Vector de tensiones nodales iniciales.
(p.u) Por Unidad.
[Y ] Matriz de Admitancias Nodales.
I
K
i
Corriente inyectada en el nodo i durante la iteracin K
P
i
Potencia activa demandada en el nodo i.
viii
Q
i
Potencia reactiva demandada en el nodo i.
V
K
i
Tensin en el nodo i durante la iteracin K.
V
K+1
i
Tensin en el nodo i durante la iteracin K + 1.
Tolerancia utilizada como criterio de convergencia.
S Vector de potencias inyectadas.
[V
0
x
] Vector de la componente real de las tensiones iniciales, [20].
[V
0
y
] Vector de la componente imaginaria de las tensiones iniciales, [20].
V
0
xi
Componente real de la tensin inicial en la barra i, [20].
V
0
yi
Componente imaginaria de la tensin inicial en la barra i, [20].
V
K
xi
Componente real de la tensin en la barra i durante la iteracin K, [20].
V
K
yi
Componente imaginaria de la tensin en la barra i durante la iteracin K, [20].
[T] Matriz de corrientes inyectadas a corrientes de rama, [20].
[D
R
] Matriz diagonal de resistencias de lneas, [20].
[D
X
] Matriz diagonal de reactancias de lneas, [20].
[TRX] Matriz de ujo de carga, [20].
[I] Matriz de corrientes inyectadas, [20].
I
K
xi
Componente real de la corriente en la barra i durante la iteracin K, [20].
I
K
yi
Componente imaginaria de la tensin inicial en la barra i, [20].
V
Slack
Tensin en la barra de referencia o slack.
V
i
Cada de tensin en el nodo i respecto a la barra slack ( %).
L
mp
Prdidas activas totales.
L
mq
Prdidas reactivas totales.
ix
G
ij
Conductancia entre los nodos i-j.
B
ij
Susceptancia entre los nodos i-j.
R
ij
Resistencia entre los nodos i-j.
X
ij
Reactancia entre los nodos i-j.
L
S/E
Prdidas de potencia en una subestacin
L
T
Prdidas totales en la red.
P
ij
Flujo de potencia activa entre los nodos i-j.
Q
ij
Flujo de potencia reactiva entre los nodos i-j.
S
ij
Flujo de potencia aparente entre los nodos i-j.
S
max
ij
Potencia nominal mxima permitida por un conductor.
Factor de seguridad.
CF Capacidad Firme.
C Factor de escalamiento.
N Nmero de transformadores operativos.
kV A
m
Potencia nominal del transformador.
kV A
max
Potencia del transformador de mayor capacidad.
V
BASE
Base de voltaje.
S
BASE
Base de potencia.
kV Kilovoltio.
MV A Mega voltio-Amper.
Pgen
0
Valor inicial de la potencia activa generada (p.u).
Qgen
0
Valor inicial de la potencia reactiva generada (p.u).
x
P
load
Potencia activa demandada (p.u).
Q
load
Potencia reactiva demandada (p.u).
fp Factor de potencia.
I
r
Componente real de la corriente en el nodo i (p.u).
I
j
Componente imaginaria de la corriente en el nodo i (p.u).
V
r
Componente real de la tensin en el nodo i (p.u).
V
j
Componente imaginaria de la tensin en el nodo i (p.u).
.V
r
Diferencia de la componente real de tensin entre iteraciones.
.V
j
Diferencia de la componente imaginaria de tensin entre iteraciones.
xi
LISTA DE ABREVIATURAS
FDC Flujo de Carga.
MVC Patrn de Diseo Modelo-Vista-Controlador.
MATLAB Herramienta Computacional de Clculo.
NR Newton-Raphson.
SDEE Sistema Elctrico de Distribucin.
GS Gauss-Seidel.
FDCB Flujo de Carga de Barrido.
INDENE Instituto de Energa de la Universidad Simn Bolvar.
SEDDGE Sistemas Elctricos de Distribucin a Gran Escala.
FDCT Flujo de Carga Directo desarrollado por Teng, [19].
TRX Flujo de Carga Directo desarrollado por De Oliveira, [20].
SEDDGT Sistema Elctrico de Distribucin de Gran Tamao.
EDC Electricidad de Caracas.
SEP Sistema Elctrico de Potencia.
SEDT Sistema Elctrico de Transmisin.
NRD Newton-Raphson Desacoplado.
slack Barra de Referencia del Sistema.
backward sweep Barrido Hacia Atrs en el FDCB.
forward sweep Barrido Hacia Adelante en el FDCB.
KVL Ley de Kirchho de Voltajes.
KCL Ley de Kirchho de Corrientes.
xii
BIBC Matriz de Corrientes Inyectadas a Corrientes de Rama o Bus Injection to Branch
Current.
BCBV Matriz de Corrientes de Rama a Voltajes Nodales o Branch Current to Bus
Voltage.
DLF Matriz de Flujo de Carga o Distribution Load Flow.
backward and forward sweep Proceso de Barrido Hacia Adelante y Hacia Atrs.
MATPOWER Paquete de Simulacin de Sistemas de Potencia de MATLAB.
ASP Sistema de Anlisis y Simulacin de Redes Primarias.
FDCO Flujo de Carga ptimo.
NRDR Newton-Raphson Desacoplado Rpido.
LP Programacin Lineal.
CE Capacidad de Emergencia .
CN Capacidad Nominal.
S/E Subestacin.
CF Capacidad Firme.
SAP Software de Gestin y Estrategia.
GIS Software de Estimacin de Demanda.
SCADA Software de registro de carga y data del sistema.
ORM Mapeadores Objeto-Relacionales.
.DAT Formato Original de la Base Datos.
.MAT Formato de los Archivos Generados con el Algoritmo Implementado.
RAM Memoria de Acceso Aleatorio (Random Access Memory).
GW Gigavatio.
xiii
ndice general
INTRODUCCIN 1
1. ANTECEDENTES 5
1.1. Caractersticas del Sistema de Distribucin . . . . . . . . . . . . . . . . . . . 5
1.2. El Flujo de Carga de Distribucin . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.1. FDC para Topologas Malladas . . . . . . . . . . . . . . . . . . . . . 8
1.2.2. FDC para Topologas Radiales . . . . . . . . . . . . . . . . . . . . . 9
1.2.3. Flujos de Carga de Barrido (FDCB) o forward and backward sweep 10
2. METODOLOGA 14
2.1. Algoritmos Implementados . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.1.1. Flujo de Carga Directo Complejo, (FDCDC), [19] . . . . . . . . . . . 15
2.1.2. Flujo de Carga Directo TRX . . . . . . . . . . . . . . . . . . . . . . 18
2.2. Herramientas Computacionales . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2.1. MATPOWER, [52] . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2.2. ASP, [54] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3. Condiciones de Operacin de la Red . . . . . . . . . . . . . . . . . . . . . . . 24
xiv
2.3.1. Anlisis por Cada de Tensin . . . . . . . . . . . . . . . . . . . . . . 25
2.3.2. Anlisis por Prdidas de Potencia . . . . . . . . . . . . . . . . . . . . 25
2.3.3. Anlisis por Capacidad Amperimtrica de Conductores . . . . . . . . 26
2.3.4. Anlisis de Capacidad Firme de una S/E . . . . . . . . . . . . . . . . 28
3. IMPLEMENTACIN 29
3.1. Arquitectura del Sistema . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2. Descripcin del Algoritmo Implementado . . . . . . . . . . . . . . . . . . . . 34
3.3. Datos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.4. Estructura del Programa . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.4.1. Adquisicin de Datos . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.4.2. Proceso Iterativo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4. CASOS DE ESTUDIOS Y ANLISIS DE RESULTADOS 39
4.1. Aplicacin en Circuito de 7 Barras . . . . . . . . . . . . . . . . . . . . . . . 39
4.1.1. Clculo de la Matriz TRX . . . . . . . . . . . . . . . . . . . . . . . . 40
4.1.2. Proceso Iterativo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.2. Validacin de Resultados . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.3. Estudio Comparativo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.4. Aplicacin en Redes de Gran Tamao . . . . . . . . . . . . . . . . . . . . . . 51
4.4.1. Tiempo de cmputo . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.4.2. Anlisis de Condiciones Operacionales de una Red de Gran Tamao . 54
xv
CONCLUSIONES 63
REFERENCIAS BIBLIOGRFICAS 65
A. CASO EXPLICATIVO: MATRICES BIBC Y BCBV 71
A.1. Matriz BIBC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
A.2. Matriz BCBV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
B. ALGORITMO IMPLEMENTADO TRX 74
B.1. Algoritmo de Adquisicin de Datos . . . . . . . . . . . . . . . . . . . . . . . 74
B.2. Flujo de Carga TRX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
C. EJEMPLO DE REPORTE DEL ALGORITMO DESARROLLADO 87
D. CASO DE ESTUDIO: CIRCUITO DE 4 BARRAS 98
E. REPORTE DEL MATPOWER PARA EL CASO DE 7 BARRAS 100
F. A COMPENSATION-BASED POWER FLOW METHOD FOR WEAK-
LY MESHED DISTRIBUTION AND TRANSMISSION NETWORKS,
SHIRMOHAMMADI ET AL. 1988, [11] 103
G. A DIRECT APPROACH FOR DISTRIBUTION SYSTEM LOAD FLOW
SOLUTION, TENG. 2003, [19] 114
H. THE DISTRIBUTION TRX-POWER FLOW METHOD, DE OLIVEIRA.
2010, [20] 121
I. OPTIMAL SIZING OF CAPACITORS PLACED ON RADIAL DISTRIBUTION
SYSTEM, BARAN, M. E. ET AL. 1989, [31] 139
xvi
J. SIMPLE AND EFFICIENT COMPUTER ALGORITHM TO SOLVE RA-
DIAL DISTRIBUTION NETWORKS, RANJAN, R. ET AL. 2003, [27] 150
xvii
ndice de tablas
2.1. Cada de Tensin Mxima Permitida, [56] . . . . . . . . . . . . . . . . . . . 25
4.1. Datos de Lnea (Ldat) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.2. Datos de Nodos, (Bdat) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.3. Valores de potencia y tensin inicial (V
0
, P
0
) . . . . . . . . . . . . . . . . . . 44
4.4. Resultados de la Primera Iteracin . . . . . . . . . . . . . . . . . . . . . . . 45
4.5. Resultados de la Segunda Iteracin . . . . . . . . . . . . . . . . . . . . . . . 45
4.6. Resultados de la Tercera Iteracin . . . . . . . . . . . . . . . . . . . . . . . . 46
4.7. Tensiones nodales en mdulo y ngulo . . . . . . . . . . . . . . . . . . . . . 46
4.8. Tensiones Nodales Validadas en Mdulo y ngulo . . . . . . . . . . . . . . . 47
4.9. Tiempo de cmputo de [19] y [20] para circuitos de 4, 7, 12 y 69 barras . . . 50
4.10. Tiempo de cmputo del ASP para red de 530 circuitos de la EDC . . . . . . 51
4.11. Tiempo de cmputo del algoritmo implementado para red de 530 circuitos de
la EDC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.12. Espacio Ocupado en Disco por redes de 10, 50, 100, 200 y 530 circuitos . . . 54
4.13. Condiciones de Operacin de la Red, 530 circuitos. . . . . . . . . . . . . . . 62
D.1. Datos de lnea (Ldat) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
D.2. Datos de nodos (Bdat) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
xviii
ndice de guras
1.1. Red de Distribucin. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.1. Diagrama de Flujo del Algoritmo Propuesto por Teng, [19]. . . . . . . . . . . 19
2.2. Diagrama de Flujo del Algoritmo Propuesto por De Oliviera, [20]. . . . . . . 22
2.3. Porcentaje de Carga de un Cable Subterrneo, [56]. . . . . . . . . . . . . . . 27
3.1. Base de Datos Unicada, [49]. . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.2. Sincronizacin de Base de Datos, [49]. . . . . . . . . . . . . . . . . . . . . . . 30
3.3. Modelo MVC, [50]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.4. Esquema General de Herramienta de Planicacin Corto-Mediano Plazo (HPCMP)
- Visualizacin de Largo Plazo, [50]. . . . . . . . . . . . . . . . . . . . . . . . 33
3.5. Esquema MVC del programa implementado. . . . . . . . . . . . . . . . . . . 35
3.6. Diagrama de Flujo del Algoritmo Desarrollado. . . . . . . . . . . . . . . . . 36
4.1. Estudio Comparativo: Tiempo de Cmputo, [20]. . . . . . . . . . . . . . . . 49
4.2. Tiempo de Cmputo del ASP para la Red de la Gran Caracas. . . . . . . . . 52
4.3. Tiempo de Cmputo del Algoritmo Implementado para la Red de la Gran
Caracas. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.4. Perl de Cada de Tensin en ( %) del circuito ANT_A01. . . . . . . . . . . 55
xix
4.5. Potencia Activa Demandada y Prdidas Tcnicas del Circuito ANT_A01. . 56
4.6. Potencia Reactiva Demandada y Prdidas Tcnicas del Circuito ANT_A01. 56
4.7. Capacidad Amperimtrica de los conductores del circuito ANT_A01. . . . . 57
4.8. Potencia Entregada y Perdida por Circuito de la S/E ANT. . . . . . . . . . 58
4.9. Capacidad Amperimtrica de los conductores de la S/E ANT . . . . . . . . 59
4.10. Capacidad Firme de las S/E ANT y SRO. . . . . . . . . . . . . . . . . . . . 60
4.11. Potencia Activa Demandada y Prdidas Tcnicas de la red. . . . . . . . . . . 61
4.12. Capacidad Amperimtrica de los conductores de la red. . . . . . . . . . . . . 62
A.1. Sistema de Distribucin de Ejemplo. . . . . . . . . . . . . . . . . . . . . . . 71
xx
INTRODUCCIN
El Flujo de Carga, FDC, es un mtodo de anlisis numrico para la determinacin del
estado de los Sistemas Elctricos de Potencia, SEP. El FDC, como herramienta de anlisis,
cobra mayor importancia en el sistema de gestin elctrica actual. Esta herramienta no slo
permite conocer el estado de la red en un momento determinado sino que tambin permite
realizar aplicaciones como despacho de generacin distribuida, equilibrio de fases, control de
tensiones, colocacin ptima de capacitores y estudios de planicacin de corto y mediano
plazo, [20].
El FDC es un estudio que requiere la resolucin de sistemas de ecuaciones no lineales. Por
lo tanto, los clculos implicados se caracterizan por presentar un alto grado de complejidad y
resulta necesario el uso de procesos iterativos para encontrar la solucin con la mayor exacti-
tud posible. An ms, el nmero de variables a manejar aumenta en funcin del tamao del
sistema elctrico en estudio y con esto aumenta la complejidad del clculo a realizar. Adems,
es bien conocido que los Sistemas Elctricos de Distribucin, SEDD, actuales se caracterizan
por ser de gran tamao; lo cual implica que el nmero de variables a manejar en el FDC
tambin lo es. Debido a esto, se hace imprescindible el uso de herramientas computacionales
para poder estudiar dichos sistemas de forma eciente.
En la actualidad hay diversas aplicaciones computacionales que son utilizadas como he-
rramientas para el anlisis de ujo de carga. La mayora de stas estn basadas en mtodos
numricos, como el Newton-Raphson (NR), [1]. Dicha metodologa fue desarrollada y usada
para la operacin, control y planicacin de sistemas de transmisin; cuya conguracin es
tpicamente mallada [2, 3]. Pero al ser empleado en SDEE presenta problemas de convergencia
1
2
y eciencia ya que se produce singularidad en la matriz Jacobiana. Debido, principalmente,
a su estructura radial, baja relacin x/r, lneas no traspuestas y cargas desbalanceadas entre
fases [2, 4, 5].
Tomando en cuenta lo anterior y considerando la topologa radial del sistema de distribu-
cin, se han desarrollado nuevas metodologas de ujo de carga que explotan la radialidad
del circuito; es decir, no requieren inversin de Jacobiano al utilizar tcnicas conocidas como
el Gauss-Seidel GS o de barrido, [18, 22, 31, 33, 44]. Finalmente, dichos algoritmos se ex-
tendieron para poder ser aplicados en sistemas dbilmente mallados [11, 12, 33].
Los mtodos desarrollados para solucionar el FDC aplicado a sistemas radiales de dis-
tribucin, llamados tpicamente ujos de carga de barrido (FDCB), pueden ser catalogados
en dos grupos [2]. En el primer grupo, aquellos que hacen ciertas modicaciones a tcnicas
ya existentes como el NR [6]-[10]. Y en el segundo, estn aquellos que hacen un proceso
de barrido hacia delante y hacia atrs, (forward and backward sweep), usando las leyes de
Kirchho [11]-[21] o utilizando la ecuacin bicuadrtica [22]-[36].
Tpicamente, los FDCB presentan una tasa lenta de convergencia. Sin embargo, son al-
tamente ecientes desde el punto de vista computacional porque no es necesario invertir
matrices [20]; hecho que resulta importante al resolver redes de gran tamao. Con las he-
rramientas computacionales disponibles en la actualidad se ha reducido considerablemente
el tiempo empleado en obtener una solucin del FDC en redes de distribucin de gran tamao.
El presente trabajo se realiz partiendo de una lnea de investigacin desarrollada por el
Instituto de Energa de la Universidad Simn Bolvar (INDENE) para el desarrollo de una
herramienta computacional para el anlisis de Sistemas Elctricos de Distribucin a Gran Es-
cala, SEDDGE. El proyecto general surge por la necesidad de desarrollar nuevos sistemas de
anlisis con una estructura abierta (cdigo abierto) a nes de integrar efectivamente los pro-
cesos atendiendo los siguientes criterios: cdigo modular, programacin orientada a objeto,
3
aplicacin web multiplataforma, control de versiones mediante un esquema colaborativo, [50].
Este trabajo aborda la resolucin de Sistema Elctrico de Distribucin de Gran Tamao,
SEDDGT, mediante la aplicacin de la metodologa de FDC unidireccional denominada
Flujo de Carga Directo. Se denomina directo por cuanto se obtiene el estado de la red
iterativamente a partir de una matriz nica que contiene informacin del sistema: dimen-
sionamiento y topologa. El algoritmo se basa en Teng, [19], el cual ha sido modicado por
De oliveira [20] con el objetivo de realizar operaciones con nmeros reales y, de esta forma,
mejorar la eciencia del algoritmo una vez implementado. El algoritmo se caracteriza por una
lgica de clculo sencilla (suma y multiplicacin de matrices), con lo cual se evitan opera-
ciones que requieran invertir matrices como en el caso de la matriz Jacobiana del NR. An
ms, las matrices requeridas para realizar el FDC estarn disponibles en memoria RAM cada
vez que se requiera. Con esto, se optimiza el algoritmo y es posible obtener resultados para
SEDDGT en el orden de los segundos. Razones que hacen del FDC un algoritmo poderoso
y muy til en el rea de planicacin de SDEE.
Objetivos
Para lograr el alcance propuesto, se plantearon un objetivo general y varios especcos,
stos se muestran a continuacin:
Objetivo General: Implementar un ujo de carga en base a la tcnica de barrido unidi-
reccional, basada en la construccin de una matriz nica que incluye impedancias de lnea y
la topologa del sistema, en redes de distribucin de gran tamao.
Objetivos Especcos:
Desarrollar un ujo de carga prototipo.
Realizar un estudio comparativo de los tiempos de cmputo entre el FDC implementado
y otros FDC ecientes como el NR y el Teng.
4
Implementar ecientemente el programa en plataforma computacional.
Aplicacin en redes ejemplo reportadas en literatura y redes de distribucin reales de
gran tamao.
Justicacin de la Tesis
El proyecto surge de la necesidad de hacer estudios de red a gran escala en tiempo til
para cumplir los nes de la operacin y planicacin de SDEE.
Organizacin de la Tesis
El presente trabajo se divide en 4 captulos. En el primer captulo se introducen los
antecedentes de los FDC aplicados a redes de distribucin. En el captulo 2, se presenta el
marco metodolgico desarrollado a lo largo del proyecto. En el tercer captulo, se describe el
proceso de implementacin del algoritmo: empezando por la arquitectura de diseo utilizada
y la descripcin de los mdulos desarrollados. En el captulo 4 se habla de los casos de estudio:
el primero, es un estudio comparativo de los tiempos de cmputo del FDCT y TRX al ser
aplicados en redes de 4, 7, 12 y 69 nodos. Se mostrar la eciencia del segundo respecto
al primero; en segundo lugar, se realiza el caso de 7 barras paso por paso para ilustrar el
proceso de clculo del TRX; en tercer lugar, se validan los resultados obtenidos en el caso
de 7 barras utilizando un NR eciente, (MATPOWER); en cuarto lugar, se implement a
gran escala en SDEE y se compararn los tiempos de cmputo del TRX con el programa
utilizado por la EDC para este tipo de estudios. Finalmente, se presentan las conclusiones y
recomendaciones pertinentes. Los anexos se encuentran al nal del documento.
CAPTULO 1
ANTECEDENTES
El Sistema Elctrico de Potencia, SEP, se encarga de la generacin, transmisin y dis-
tribucin de la energa elctrica. Como consecuencia del gran tamao y la alta complejidad
en funciones de dicho sistema; ste se encuentra dividido en dos grandes sub-sistemas:
Sistema Elctrico de Transmisin, SEDT. Conformado por el rea de generacin
y la red de transmisin; los cuales operan a grandes niveles de tensin. Se encarga de
generar y transmitir grandes bloques de potencia.
Sistema Elctrico de Distribucin, SDEE. Conformado por las subestaciones y
la red de distribucin; los cuales operan a niveles intermedios y bajos de tensin. La
funcin de este sistema es distribuir los grandes bloques de potencia a los consumidores
nales.
1.1. Caractersticas del Sistema de Distribucin
El SDEE presenta una serie de caractersticas especcas que lo diferencian considerable-
mente del sistema de transmisin. Estas caractersticas se presentan a continuacin:
Relacin x/r. Los conductores de la red de distribucin presentan una baja relacin
x/r, ya que x r. Mientras que, en los SEDT se tiene que normalmente x >> r; razn
6
sobre la cual se basa el Principio de Desacople:
los cambios en la potencia activa deben manifestarse sobre los ngulos de fases del
sistema (...) los cambios de potencia reactiva deben reejarse en las magnitudes de
tensin [38, pg. 159].
Debido a la baja relacin x/r presente en las redes de distribucin, no se cumple dicho
principio. Razn por la cual, el Newton-Raphson, NR, tiene problemas de convergencia
y el Newton-Raphson Desacoplado, NRD, no es funcional.
Diversidad de Cargas. Debido a la diversidad de consumidores en la red es posible
encontrar diversos tipos de demanda, desde zonas rurales con densidades del orden de
kV A/km
2
, hasta zonas urbanas con densidades en el orden de MV A/km
2
[3]. Adems,
el mismo circuito puede ser usado para suplir cargas residenciales, comerciales y/o
industriales.
Cargas Desbalanceadas. Usualmente se emplean acometidas de dos, tres y cuatro
hilos para alimentar cargas tanto trifsicas como bifsicas y/o monofsicas. Esta car-
acterstica tiende a desbalancear las redes de distribucin.
1.2. El Flujo de Carga de Distribucin
El FDC es un algoritmo que permite calcular las tensiones nodales, en mdulo y ngulo,
en rgimen permanente de un SEP. El rgimen permanente es aquel estado cuasiestacionario
en el cual existe un equilibrio de las potencias y las variables de tensin y frecuencia no pre-
sentan variaciones signicativas [38]. Es bien conocido que la carga vara constantemente en
el tiempo, razn por la cual se utiliza el termino cuasiestacionario. Sin embargo, se entiende
que el FDC analiza el sistema en estados puntuales [3]. Tambin es utilizado en el rea de
planicacin por la capacidad de introducir suposiciones de estados futuros de la red. Razn
por la cual la herramienta posee una gran versatilidad y extraordinario potencial en el anlisis
de SEP.
7
Existen dos tipos de datos de entrada para el algoritmo, stos son:
Datos de Red. Hacen referencia a las especicaciones de las conexiones y a los
parmetros de las lneas, transformadores y compensadores presentes en el circuito.
Datos de Nodos. Hacen referencia a las especicaciones de los datos operacionales
de cada nodo, como tipo de nodo, su potencia consumida o generada, su magnitud de
tensin, etc.
El problema es no-lineal y no puede ser resuelto analticamente. Razn por la cual, se
recurre a tcnicas numricas iterativas para hallar la solucin [3]. Las ecuaciones que describen
al problema se pueden plantear como se presenta en las ecuaciones 1.1 y 1.2, [37].
P
i
= |V
i
|
2
G
ii
+
n

j=1
|V
i
||V
j
||Y
ij
| cos(
ij
+
j

i
) (1.1)
Q
i
= |V
i
|
2
B
ii

j=1
|V
i
||V
j
||Y
ij
| sin(
ij
+
j

i
) (1.2)
Donde, i, j son los nodos del sistema. n representa el nmero de nodos del sistema. P
i
y
Q
i
es la potencia activa y reactiva en el nodo i. V
i
y V
j
son las tensiones en los nodos i y j.
G
ii
y B
ii
son la conductancia y la susceptancia propias del nodo i. Y
ij
es la admitancia entre
los nodos i-j.
ij
es el ngulo de la admitancia de una lnea entre los nodos i-j.
i
y
j
son los
ngulos de las tensiones en los nodos i y j, respectivamente. En [37] se encuentra con mayor
detalle las ecuaciones planteadas en los FDC.
Una vez calculadas las tensiones, es posible calcular las corrientes por el circuito, ujos
de potencia, capacidades amperimtricas de conductores y prdidas en la red.
Para realizar ciertos anlisis no es estrictamente necesario considerar el desbalance de la
red. Cuando este es el caso, es posible asumir la red balanceada y realizar la modelacin de la
red mediante un equivalente unilar [11]. Para efecto del anlisis realizado, las modelaciones
fueron realizadas considerando redes balanceadas.
8
Tipos de Nodos Dependiendo de las condiciones de contorno que se especiquen, se
pueden clasicar los nodos del sistema en tres grupos; [37]:
Nodo de Referencia o Slack. Es una barra de generacin, en la que se asume
conocido el mdulo y ngulo de la tensin.
Nodos PQ. Son aquellos nodos en los que se especican las potencias activa y reactiva
netas inyectadas. Los nodos PQ son los ms abundantes en los sistemas de distribucin.
Nodos PV. Son nodos en los que se especican la potencia activa y el mdulo de la
tensin. Este tipo de nodos son poco comunes en sistemas de distribucin.
1.2.1. FDC para Topologas Malladas
Los primeros algoritmos de FDC, como el Newton-Raphson (NR) o el Gauss-Seidel (GS),
fueron desarrollados para la operacin, control y planicacin de SEDT, [2, 37, 1]. La prin-
cipal diferencia entre estos ltimos respecto a los FDC para topologas radiales, es que los
mismos pueden ser aplicados independientemente de la conguracin de la red [3].
El NR es un mtodo basado en un slido fundamento matemtico como lo son las series
de Taylor [3]. Este mtodo utiliza un proceso iterativo mediante el cual se aproxima a la
solucin linealizando las ecuaciones de potencia (ver ecuaciones 1.1 y 1.2). En [3, 37, 38] es
posible encontrar la descripcin del proceso algortmico de este FDC. Adems, este mtodo
requiere de un gran esfuerzo computacional y un elevado tiempo de cmputo.
La aplicacin del mtodo a un sistema no lineal de ecuaciones de orden (n) va a implicar
en cada iteracin, la formacin del Jacobiano de orden (nxn) y su subsiguiente inversin, lo
cual toma un tiempo no despreciable [38, pg. 129].
El mtodo GS presenta la ventaja de no utilizar matrices Jacobianas; lo cual lo convierte
en un algoritmo mucho ms sencillo y requiere un menor esfuerzo computacional que el NR.
9
Su simplicidad matemtica es notoria, no requirindose la inversin de matrices en lo absolu-
to por lo cual su programacin digital es muy rpida [38, pg. 109]. En [3, 38] se encuentra la
descripcin de las ecuaciones utilizadas por el FDC y la descripcin del proceso algortmico
del mismo.
Sin embargo, el GS requiere de un gran nmero de iteraciones para converger. Por lo
general, realiza un nmero de iteraciones aproximadamente igual al nmero de nodos del
sistema estudiado [38].
En la literatura [4, 11, 14, 18, 19, 33, 39, 40, 41, 42, 43, 44] es posible encontrar una
amplia discusin sobre los problemas de convergencia o tiempos inecientes de convergencia
de estos mtodos al ser aplicados en redes del tipo enfermas. Las redes de distribucin, por
su estructura radial y baja relacin x/r entran dentro de esta categora.
1.2.2. FDC para Topologas Radiales
Debido a los problemas presentados por los FDC NR y GS al ser aplicados en redes de
distribucin, se ha hecho ms frecuente el uso del FDC de barrido [16]. Estos algoritmos
sacan provecho de la topologa, tpicamente radial, de la red. Este mtodo consiste en un
proceso iterativo en el cual se computan tensiones y corrientes mediante evaluaciones se-
cuenciales, llamados barridos. Est compuesto por dos etapas: barrido hacia atrs (backward
sweep) desde las cargas hacia la fuente; y el barrido hacia adelante (forward sweep) desde la
fuente hacia las cargas [2].
En el backward sweep se calculan las corrientes de rama y/o potencias del sistema, par-
tiendo desde el nodo ms lejano a la fuente hasta llegar al ms cercano a sta. En el forward
sweep se calculan los voltajes en las barras del sistema, partiendo desde la barra ms cercana
a la fuente hasta la ms lejana a sta [2]. De esta forma, se utiliza el backward sweep para
actualizar las corrientes y/o potencias del sistema, y con el forward sweep se actualizan los
voltajes en las barras.
10
1.2.3. Flujos de Carga de Barrido (FDCB) o forward and backward
sweep
Clculo de Voltaje en Redes de Distribucin
Como se describe en [2], se va a considerar el circuito de distribucin mostrado en la gura
1.1. La potencia activa y reactiva en la barra receptora, puede escribirse como se indica en
las ecuaciones 1.3 y 1.4
Figura 1.1: Red de Distribucin.
P
r
=
V
s
V
r
Z
cos(
Z

s
+
r
)
V
2
r
Z
cos(
Z
) (1.3)
Q
r
=
V
s
V
r
Z
sin(
Z

s
+
r
)
V
2
r
Z
sin(
Z
) (1.4)
Usando la identidad trigonomtrica, (ecuacin 1.5):
cos
2
(
Z

s
+
r
) + sin
2
(
Z

s
+
r
) = 1 (1.5)
Despejando los trminos cos(
Z

s
+
r
) y sin(
Z

s
+
r
) de 1.3 y 1.4, respectivamente
y sustituyndolos en 1.5 se obtiene la forma general de la ecuacin bi-cuadrtica, dada en
1.6. Con la mxima raz real de la ecuacin 1.6 se obtiene la magnitud del voltaje en el nodo
receptor. El voltaje en el nodo receptor se puede escribir en funcin de la impedancia de la
lnea, de la potencia y voltaje del nodo emisor, tal como se muestra en 1.7.
V
4
r
+ 2V
r
(P
r
R +Q
r
X) V
2
s
V
2
r
+ (P
2
r
+Q
2
r
)Z
2
= 0 (1.6)
11
V
r
=

V
2
s
2(P
s
R +Q
s
X) +
(P
2
s
+Q
2
s
Z
2
)
V
2
s
(1.7)
Tambin, se puede aplicar la ley de Kirchho de voltajes (KVL) para obtener los voltajes
en las barras del sistema. Aplicando KVL en el circuito de la gura 1.1. En la literatura
es posible encontrar varios algoritmos desarrollados en base a la ecuacin cuadrtica. En el
presente trabajo se enfoca la investigacin en los FDCB basados en las leyes de Kirchho.
V
s
= V
r
+I
s
.Z (1.8)
V
r
= V
s
I
s
.Z (1.9)
Descripcin de FDCB basados en las leyes de Kirchho, [2]
Muchos de los algoritmos usados en sistemas de distribucin ([11]-[21]) usan las leyes
de Kirchho (KVL y KCL, por sus siglas en ingls) para calcular las corrientes de rama
y los voltajes nodales en el backward y forward sweep, respectivamente. En [11], los auto-
res presentan un mtodo de compensacin para redes de distribucin balanceadas radiales
y/o dbilmente malladas. Para esto, utilizan una tcnica de compensacin multipuertos y
la formulacin bsica de las leyes de Kirchho. Los sistemas radiales son resueltos usando
un procedimiento de dos pasos: las corrientes de rama son calculadas (backward sweep) y
luego se actualizan los voltajes en los nodos usando la ecuacin 1.8 para cada rama (forward
sweep). Se utiliza la diferencia de la potencia activa y reactiva en las cargas entre iteraciones
como criterio de convergencia. En [12] se aplica el algoritmo de [11] en redes desbalanceadas
(radiales o dbilmente malladas).
En los algoritmos [13] y [14] se hacen pequeas modicaciones a los algoritmos menciona-
dos anteriormente. En stos, los circuitos radiales son resueltos tal como lo explica [11] y se
usa la diferencia de las tensiones entre iteraciones como criterio de convergencia. Sin embargo,
en [13] se incorpora el modelo de trasformadores trifsicos, propuesto por [45], en el anlisis.
Mientras que, en [14] se describe un algoritmo computacional que permite encontrar el valor
12
exacto de las corrientes en todas las ramas del circuito.
En [15], se propone una versin modicada del mtodo de barrido para redes radiales
balanceadas o desbalanceadas. En el backward sweep cada corriente de rama es calcula usan-
do KCL. Luego, conociendo las corrientes, se calculan los voltajes en las barras, usando la
ecuacin 1.9, durante el forward sweep. La magnitud del voltaje en cada barra es comparada
con el valor de la iteracin anterior. Si el error est dentro de determinada tolerancia el pro-
ceso se detiene (criterio de convergencia). En caso contrario, el proceso de barrido contina
hasta que todas las tensiones cumplan dicho criterio. En el algoritmo propuesto en [16] se
calculan las corrientes de cada rama y, usando 1.8, se calculan los voltajes de barra. El voltaje
en la barra slack es calculado y se compara con un valor denido anteriormente. Si el error
est dentro de cierta tolerancia el proceso iterativo se detiene.
Liu et al, en [17], desarrollan un algoritmo aplicable a redes radiales y/o redes dbilmente
malladas. La parte radial es resuelta como se describe en [16]. Sin embargo, [17] se diferencia
de [16] porque primero se ajustan los voltajes en las barras a travs de un rango obtenido
por medio de la relacin entre el voltaje computado en la fuente y el especicado. Se utiliza
la diferencia de los voltajes entre iteraciones como criterio de convergencia.
Un algoritmo para sistemas de distribucin desbalanceadas es dado por [18]. En este
algoritmo, se utiliza la topologa de la red para resolver el problema. Propone la creacin de
dos matrices: una matriz que relaciona las corrientes inyectadas con las corrientes de rama,
BIBC (Bus Injection to Branch Current), y una matriz que relaciona las corrientes de rama
con los voltajes de nodo, BCBV (Branch Current to Bus Voltage). Luego, se obtiene la matriz
de ujo de carga de distribucin, DLF (Distribution Load Flow) al multiplicar BCBV.BIBC.
Posteriormente, se calculan las cadas de tensin en las lneas multiplicando DLF y la matriz
de corrientes inyectadas, I
K
, 1.10. Las tensiones nodales se obtienen restando la tensin en
la fuente menos la cada de tensin en las lneas 1.11.
.V
K+1
= DLF.I
K
(1.10)
13
V
K+1
= V
0
.V
K+1
(1.11)
Este algoritmo es mejorado en [19] para redes radiales y redes dbilmente malladas. Para
esto se realizan algunas modicaciones en las matrices BIBC y BCBV. El criterio de conver-
gencia utilizado es la diferencia de voltajes entre iteraciones. Una modicacin al algoritmo
anterior es realizada por [20]. En sta se sigue la misma lgica utilizada por [19] pero se in-
troduce un cambio para poder realizar todos los clculos con nmeros reales, dicho algoritmo
es denominado TRX. Con esta modicacin se mejora la eciencia del algoritmo propuesto
por [19].
Un algoritmo basado en la KVL para sistemas de distribucin monofsicos es propuesto
por [21]. El problema es resuelto considerando las cargas como impedancias constantes du-
rante el backward sweep. El voltaje en cada nodo es calculado usando la ecuacin 1.8 durante
el backward sweep. Luego, se usa la relacin entre la tensin impuesta en el nodo slack y su
nuevo valor calculado para obtener el valor actual del voltaje en las barras y de las corrientes
inyectadas y de rama.
CAPTULO 2
METODOLOGA
En este captulo se describen en detalle los algoritmos implementados: en primer lugar, el
ujo de carga, FDC, desarrollado por Teng, denominado (FDCT), [19], y en segundo lugar,
el FDC desarrollado por De Oliveira, denominado (TRX), [20]. Adems se describen dos
herramientas computacionales utilizadas para comparar y validar el algoritmo implementado;
es decir, el (TRX). Luego, se muestran las ecuaciones requeridas para realizar el anlisis del
sistema posterior a la solucin del ujo de carga.
2.1. Algoritmos Implementados
Partiendo de una lnea de investigacin desarrollada por el profesor Paulo de Oliveira se
trabaj con el TRX. El mismo utiliza la misma lgica de clculo que el FDCT pero incluye
una modicacin para trabajar en nmeros reales y no en complejos como lo hace FDCT.
Razn por la cual, es vlido armar que el TRX es ms eciente que el FDCT. Dicha ar-
macin se comprueba en la seccin 4.3 del captulo 4
En la implementacin de dichos algoritmos se asumieron las siguientes premisas:
Sistema balanceado.
Datos del sistema transformado en por unidad (p.u) en una base comn.
15
Sistema exento de nodos PV.
Numeracin del nodo slack como el nmero 1.
Redes de distribucin radiales.
Para todos los algoritmos implementados, el proceso de numeracin de los nodos se realiza
como se describe en [11].
2.1.1. Flujo de Carga Directo Complejo, (FDCDC), [19]
Este algoritmo se caracteriza por realizar los clculos en nmeros complejos, como la
mayora de los algoritmos de FDC. Adems, utiliza una matriz nica (DLF) para determinar
el estado del sistema y el proceso se realiza de forma unidireccional. La matriz [DLF] reeja
las impedancias del sistema asociadas a la topologa del mismo.
En el mismo, se utilizan dos matrices -la matriz de inyecciones de corriente a corriente de
rama (Bus Injection to Branch Current), BIBC, y la matriz de corrientes de rama a voltajes
nodales (Branch Current to Bus Voltage, BCBV- y una simple multiplicacin de matrices
son usadas para obtener la solucin del FDC.
Debido a las tcnicas utilizadas en este algoritmo; se tiene que procesos como el de in-
versin de la matriz Jacobiana o la construccin de la matriz de admitancias nodales ([Y ])
requerido en los FDC tradicionales ya no son necesarios. Lo que hace del FDCT un algoritmo
robusto y eciente en tiempo, [19]. De acuerdo a [19], el FDCT presenta un gran potencial
para ser utilizado en aplicaciones de automatizacin de redes de distribucin. El artculo se
encuentra disponible en el apndice G.
Descripcin General
El proceso inicia numerando los nodos del circuito con el mtodo propuesto por [11]; segn
el cual se divide el circuito en capas y la primera va de la fuente hacia el(los) nodo(s) ms
16
cercano(s) a la misma, la segunda capa va desde stos ltimos hacia los nodos ms cercanos
a los mismos, se repite el proceso hasta haber dividido todo el circuito en capas. Luego se
enumeran los nodos empezando desde la capa nmero 1 hasta llegar a la ltima pero no se
puede realizar cambio de capas hasta que todos los nodos de la capa previa sean enumerados,
ver apndice F.
Luego, se calculan las corrientes inyectadas en los nodos para la iteracin K-sima, I
K
i
,
(ecuacin 2.1). Donde, P
i
y Q
i
en la potencia activa y reactiva demandada en el nodo i, la
cual no cambia enter iteraciones. V
K
i
es la tensin en el nodo i durante la iteracin K.
I
K
i
= (
P
i
+jQ
i
V
K
i
)

(2.1)
Luego, se construyen las matrices BIBC y BCBV. En el apndice A se ilustra el proceso de
armado de ambas matrices. BIBC est relacionada con la topologa del circuito y BCBV est
relacionada con las impedancias de las lneas. Posteriormente, se multiplican ambas matrices
como lo muestra la ecuacin 2.2 para obtener la matriz de impedancias de lnea asociadas a
la topologa del sistema. La matriz [DLF] es de orden (n1)x(n1), donde n es el nmero
de nodos del sistema.
[DLF] = [BCBV ][BIBC] (2.2)
En el siguiente paso, se calcula la diferencia de voltaje entre iteraciones, [.V
K+1
], uti-
lizando la ecuacin 2.3. Dicho clculo se realiza en forma matricial. [I
K
] es el vector columna
de inyecciones de corriente.
[.V
K+1
] = [DLF][I
K
] (2.3)
Finalmente, se calculan las tensiones nodales 2.4. Donde, V
K+1
es el vector de tensiones
nodales en la iteracin K + 1 y V
0
es el vector de tensiones nodales iniciales, tpicamente la
tensin inicial en cada nodo se asume 1 p.u.
17
[V
K+1
] = [V
0
] + [.V
K+1
] (2.4)
El proceso se detiene cuando la diferencia de voltaje entre iteraciones es menor a cierta
tolerancia, ecuacin 2.5. Tpicamente, se asume igual a 0, 001 p.u.
||V
K
i
| |V
K1
i
|| (2.5)
Procesos Algortmicos
Matriz BIBC. En el apndice A se muestra un ejemplo ilustrativo para la construccin
de BIBC.
Paso 1. Para un sistema de m ramas y n nodos, la dimensin de la matriz BIBC es
m(n 1)
Paso 2. Si una lnea (B
l
) est entre los nodos i y j. Se copia la columna de la barra i,
se pega en la columna de la barra j y se coloca +1 en la la l y la columna de la barra
j.
Paso 3. Repetir el paso 2 hasta que todas las lneas estn en la matriz BIBC.
Matriz BCBV. En el apndice A se muestra un ejemplo ilustrativo para la construccin
de BCBV.
Paso 1. Para un sistema de m ramas y n nodos, la dimensin de la matriz BCBV es
(n 1) m
Paso 2. Si una lnea (B
l
) est entre los nodos i y j. Se copia la la de la barra i, se
pega en la la de la barra j y se coloca la impedancia de la lnea Z
ij
en la la de la
barra j y la columna l.
Paso 3. Repetir el paso 2 hasta que todas las lneas estn en la matriz BCBV.
18
Solucin del FDC
Paso 1. Enumerar los nodos del circuito utilizando el mtodo propuesto por Shirmo-
hammadi, [11].
Paso 2. Leer datos referentes a las conexiones del sistema, potencia aparente en los
nodos e impedancias de las lneas.
Paso 3. Construir matrices BIBC y BCBV.
Paso 4. Pre-especicar V
0
= V
slack
e inicializar todas las tensiones al valor del nodo
slack.
Paso 5. Calcular I
K
i
(2.1).
Paso 6. Calcular la matriz DLF (2.2).
Paso 7. Calcular la variacin de voltaje en la iteracin K + 1 (2.3).
Paso 8. Calcular la tensin en la iteracin K + 1 (2.4).
Paso 9. Vericar si se cumple (2.5).
Paso 10. En caso de no cumplirse (2.5), volver al paso 5.
En la gura 2.1 se encuentra el diagrama del FDCT, [19].
2.1.2. Flujo de Carga Directo TRX
Este algoritmo se caracteriza por realizar los clculos en nmeros reales, lo cual optimiza
el proceso de cmputo. Adems, utiliza una matriz nica (TRX) para determinar el esta-
do del sistema y el proceso se realiza de forma unidireccional. La matriz [TRX] reeja las
impedancias del sistema asociadas a la topologa del mismo.
Este algoritmo es til al ser aplicado con propsitos de planicacin y evaluacin del sis-
tema de distribucin en tiempo real. El estado del sistema se obtiene usando el histrico de
19
Figura 2.1: Diagrama de Flujo del Algoritmo Propuesto por Teng, [19].
las medidas tomadas al mismo; teniendo en cuenta su topologa en el presente y en el futuro.
El TRX resulta ms eciente que el FDCT porque el mismo trabaja con nmeros reales a
diferencia del segundo lo hace en nmeros complejos. Desde un punto de vista computacional,
consume implica ms tiempo realizar operaciones en nmeros complejos que en nmeros
reales. El artculo se encuentra disponible en el apndice H.
Descripcin General
Para este algoritmo se requiere la numeracin nodal desarrollada por Shirmohammadi
(explicada en la seccin 2.1.1, ver apndice F). El FDC utiliza el vector de potencias inyec-
tadas (S), la topologa del circuito y la impedancia de las lneas (Z = R+jX). En este FDC
se trabaja con la parte real y la parte imaginaria de los datos por separado. De esta forma,
se parte de un vector de voltajes iniciales ([V
0
]) separado en parte real e imaginaria, ecuacin
(2.6).
20
V
0
x
= [V
0
x1
...V
0
xi
...V
0
xn
] (2.6)
V
0
y
= [V
0
y1
...V
0
yi
...V
0
yn
]
Luego, se calcula la matriz [T] siguiendo el mismo procedimiento utilizado para construir
la matriz BIBC en el FDCT, ver apndice A. Adems, utilizan dos matrices diagonales: en
la primera se denen las resistencias, [D
R
], y en la segunda las reactancias de las lneas del
circuito, [D
X
], ecuaciones 2.7 y 2.8, respectivamente.
D
R
=

D
R1
0 . . . 0
0 D
R2
. . . 0
0 0
.
.
.
.
.
.
0 0 D
Rn

(2.7)
D
X
=

D
X1
0 . . . 0
0 D
X2
. . . 0
0 0
.
.
.
.
.
.
0 0 D
Xn

(2.8)
Posteriormente, se multiplican las matrices T, D
R
y D
x
, como lo muestra la ecuacin 2.9,
para obtener la matriz de impedancias de lnea asociadas a la topologa del sistema, TRX.
La matriz [TRX] es de orden 2(n 1)x2(n 1), donde n es el nmero de nodos del sistema.
TRX =

T
T
D
R
T T
T
D
X
T
T
T
D
X
T T
T
D
R
T

(2.9)
Luego, se calcula la matriz de corrientes inyectadas [I], ecuacin 2.10.
I =

I
K
xi
I
K
yi

(2.10)
Donde, I
K
xi
y I
K
yi
se obtienen de las ecuaciones, (2.11) y (2.12), respectivamente. I
K
xi
y I
K
yi
es la componente real e imaginaria de la corriente en el nodo i durante la iteracin K. P
i
y
21
Q
i
son las potencias activa y reactiva, respectivamente. V
K
xi
y V
K
yi
son las componentes real
e imaginaria de la tensin en la barra i durante la iteracin K.
I
K
xi
= Re{I
K
i
} =
P
i
V
K
xi
Q
i
V
K
yi
(V
K
xi
)
2
+ (V
K
yi
)
2
(2.11)
I
K
yi
= Im{I
K
i
} =
Q
i
V
K
xi
P
i
V
K
yi
(V
K
xi
)
2
+ (V
K
yi
)
2
(2.12)
Finalmente, se calculan las tensiones en los nodos, ecuacin (2.13).
V = V
0
TRX.I (2.13)
El proceso se detiene cuando la diferencia de voltaje entre iteraciones es menor a deter-
minada tolerancia, ecuacin (2.14).
||V
K
i
| |V
K1
i
|| (2.14)
Para los casos estudiados, se asumi V
0
xi
= 1 y V
0
yi
= 0 para i = 1, ..., n.
Procesos Algortmicos
1. Paso 1. Leer datos referentes a las conexiones del sistema, potencia aparente en los
nodos e impedancias de las lneas.
2. Paso 2. Construir matrices T, D
R
y D
X
.
3. Paso 3. Pre-especicar V
0
= V
slack
e inicializar todas las tensiones al valor del nodo
slack, (2.6).
4. Paso 4. Calcular matriz de corrientes inyectadas, I.
5. Paso 5. Calcular la matriz TRX.
6. Paso 6. Calcular la tensin en la iteracin K + 1 (2.13).
22
7. Paso 7. Vericar si se cumple (2.14).
8. Paso 8. En caso de no cumplirse (2.14), volver al paso 4.
En la gura 2.2 se muestra el diagrama de ujo del algoritmo propuesto por De Oliveira
[20].
Figura 2.2: Diagrama de Flujo del Algoritmo Propuesto por De Oliviera, [20].
2.2. Herramientas Computacionales
A continuacin se describirn las herramientas computacionales utilizadas para comparar
y validar el algoritmo implementado.
2.2.1. MATPOWER, [52]
MATPOWER es un paquete de Matlab para resolver problemas de FDC y ujo de carga
ptimo, FDCO. MATPOWER est diseado para dar el mejor desempeo posible mante-
23
niendo un cdigo simple de entender y modicar.
MATPOWER presenta tres algoritmos para resolver FDC: un FDC estndar y dos FD-
CO). El FDC estndar est basado en el mtodo Newton-Raphson (NR), en el cual la matriz
Jacobiana es actualizada en cada iteracin. Este mtodo se describe con detalle en [37, 38].
Los otros algoritmos son variaciones del NR desacoplado rpido, NRDR, [46]. El algoritmo
basado en el NR tradicional presente un excelente rendimiento al ser aplicado en sistemas
de potencia de gran escala. Esto se debe a que el algoritmo trabaja con la esparsidad de
la matriz Jacobiana. Es decir, con esta tcnica se evita construir la matriz Jacobiana y su
posterior inversin.
El primer FDCO est basado en la funcin constr de Matlab, la cual usa una tcnica
de programacin cuadrtica sucesiva para trabajar con la matriz Hessiana del sistema. El
segundo algoritmo est basado en programacin lineal (LP, por sus siglas en ingls), [53].
Sin embargo, el desempeo de los FDCO del MATPOWER depende de muchos factores.
En primer lugar, la funcin constr utiliza un algoritmo que no preserva la esparsidad de la
matriz. Por lo tanto, el primer FDCO queda limitado a sistemas de potencia de poco tamao.
Por otro lado, el algoritmo basado en LP preserva la esparsidad de la matriz pero no le saca
provecho, [53]. El programa se describe con mayor detalle en [52].
2.2.2. ASP, [54]
El Sistema de Anlisis y Simulacin de Redes Primarias, ASP, est orientado al ingeniero
de distribucin especializado en planicacin y proyectos de redes primarias. Las capacidades
de esta aplicacin son, [55]:
Analizar y editar circuitos.
Simulacin de crecimiento de redes.
24
Compensacin capacitiva para mnima perdida y correccin de bajo voltaje.
Anlisis de sensibilidad de parmetros.
Simulacin de interrupciones y recuperacin con otros circuitos interconectados.
Conguracin para mnima perdida.
Con el ASP se pueden obtener los siguientes resultados, [54]:
Muestra el nodo con mayor cada de tensin y los nodos en los que la tensin es menor a
0,95 p.u. El programa seala la ubicacin grcamente de los nodos con la caracterstica
arriba mencionada e indica su tensin.
Muestra el tramo con mayor Capacidad de Emergencia, CE, y seala con colores difer-
entes aquellos tramos en los que la demanda es superior a un 67 % de su CE, los que
superan su Capacidad Nominal, CN y aquellos que superan el 100 % de la CE.
Muestra un reporte en el que se encuentran las prdidas totales en kVAR y kW, tanto
en valores reales como en porcentaje ( %).
2.3. Condiciones de Operacin de la Red
Al conocer el estado del sistema, obtenido al aplicar el FDC, es posible calcular las
caractersticas de operacin del sistema estudiado; entre otros anlisis se puede estudiar la
operacin del sistema ante condiciones de fallas, disear y/o planicar posibles expansiones
o mejoras en la red, ente otros estudios. Slo se har nfasis en el clculo de las condiciones
de operacin de la red. En la literatura ([16], [37], [38]) es posible encontrar informacin
referente a otros estudios que se pueden realizar al conocer el estado de la red.
25
2.3.1. Anlisis por Cada de Tensin
Criterio de Cada de Tensin Mxima, [56]
Este criterio indica la mxima cada de tensin que puede ocurrir en circuitos primarios,
tanto areos como subterrneos. Los lmites permitidos se encuentran en la tabla 2.1.
Tabla 2.1: Cada de Tensin Mxima Permitida, [56]
Condiciones de Operacin VMAX( %) Banda Permitida (p.u)
Normal 5 0, 95 < V < 1, 05
Emergencia 8 0, 92 < V < 1, 08
La ecuacin 2.15 muestra, en trminos porcentuales, la cada de tensin en cada nodo
respecto a la tensin de la barra Slack, [56].
.V
i
=
V
Slack
V
i
V
Slack
100 (2.15)
2.3.2. Anlisis por Prdidas de Potencia
Con las ecuaciones 2.16 y 2.17 se pueden obtener las prdidas activas (L
mp
) y reactivas
(L
mq
) totales en el circuito de estudio, [48].
L
mp
=
1
2
n

i=1
n

j=1
G
ij
[V
2
i
V
i
V
j
cos
ij
] (2.16)
L
mq
=
1
2
n

i=1
n

j=1
B
ij
[V
2
i
V
i
V
j
sin
ij
] (2.17)
i, j = 1, 2, ..., n
i = j
26
Donde, i, j y n representan el nodo de salida, de llegada y el nmero de nodos del circuito,
respectivamente. Adems, G
ij
, B
ij
y
ij
se obtienen de 2.18, 2.19 y 2.20, respectivamente,
[48]. G
ij
, B
ij
, R
ij
y X
ij
son la conductancia, susceptancia, resistencia y reactancia entre los
nodos i-j.
ij
es el ngulo de la admitancia de una lnea entre los nodos i-j.
G
ij
=
R
ij
(R
ij
)
2
+ (X
ij
)
2
(2.18)
B
ij
=
X
ij
(R
ij
)
2
+ (X
ij
)
2
(2.19)

ij
=
i

j
(2.20)
Las prdidas en una subestacin (S/E), L
S/E
, corresponde a la suma de las prdidas de
cada circuito de dicha S/E, ecuacin 2.21. En esta ecuacin, C corresponde al nmero total
de circuitos que la componen.
L
S/E
=
C

k=1
L
m
(2.21)
Las prdidas totales en la red, L
T
, corresponden a la suma de las perdidas por subestacin,
ecuacin 2.22. En esta ecuacin, S corresponde al nmero total de S/E que componen toda
la red.
L
T
=
S

k=1
L
m
(2.22)
2.3.3. Anlisis por Capacidad Amperimtrica de Conductores
Criterio de Capacidad de Carga, [56]
Todo circuito debe tener un porcentaje mximo de carga igual al 67 % de su capacidad
de emergencia, ya que debe cumplirse que cada circuito primario pueda ser asistido por
dos o ms circuitos. En la gura 2.3 se puede observar el porcentaje de carga de un cable
27
Figura 2.3: Porcentaje de Carga de un Cable Subterrneo, [56].
subterrneo en condiciones normales y de emergencia, as como tambin la reserva que puede
ser utilizada a mediano o a largo plazo.
Para una lnea conectada entre los nodos i y j, el ujo de potencia que va de la barra i a
la j, puede ser obtenida de la siguiente forma, [48].
P
ij
= G
ij
[V
i
V
j
cos
ij
V
2
i
] +B
ij
V
i
V
j
sin
ij
(2.23)
Q
ij
= B
ij
[V
2
i
V
i
V
j
cos
ij
] +G
ij
V
i
V
j
sin
ij
(2.24)
S
ij
=

(P
ij
)
2
+ (Q
ij
)
2
(2.25)
G
ij
, B
ij
y
ij
se obtienen de 2.18, 2.19 y 2.20, respectivamente. S
ij
corresponde al ujo
de potencia aparente del conductor en condiciones de operacin.
Para vericar que un conductor est operando en condiciones normales se debe cumplir
la relacin de la ecuacin 2.26, [56].
.S
max
ij
S
ij
(2.26)
28
corresponde a un factor de seguridad; tpicamente corresponde a 67 %. S
max
ij
corresponde
al ujo mximo de potencia de un conductor ubicado entre los nodos i y j.
2.3.4. Anlisis de Capacidad Firme de una S/E
Criterio de Capacidad de Firme, [54]
La capacidad rme (CF) es la que se debe manejar cuando se disea una S/E, para que
ante una posible salida forzada de algn transformador se pueda seguir supliendo la carga
demandada de una forma segura y continua, sin necesidad de realizar interconexiones con
otros circuitos para suplir la demanda.
La demanda actual y la proyectada para el futuro no deben exceder la CF de la S/E.
Adems, se debe tratar que todos los transformadores de la misma S/E operen a la misma
capacidad nominal con el n de tener la mayor CF, lo cual proporciona una mayor CF ante
posibles contingencias. La Electricidad de Caracas, EDC, opera con un mximo de 4 trans-
formadores por S/E.
Utilizando la ecuacin 2.27 es posible calcular la CF de una S/E.
CF = C.(
N

m=1
kV A
m
kV A
max
) (2.27)
Donde, N es el nmero de transformadores operando en la S/E. kV A
m
es la potencia
nominal del transformador. kV A
max
es la potencia del transformador de mayor capacidad.
C es un factor que puede ser:
120 por ciento para transformadores de distribucin cuyo tiempo til sea menor o igual
a 40 aos.
100 por ciento para transformadores de distribucin cuyo tiempo til sea mayor o igual
a 40 aos.
CAPTULO 3
IMPLEMENTACIN
3.1. Arquitectura del Sistema
El estado actual del sistema de adquisicin y procesamiento de datos utilizado en la
Electricidad de Caracas, EDC, presenta las siguientes caractersticas, [49]:
Incompatibilidad de datos.
Duplicidad de Informacin.
Mayor trabajo del necesario.
Retardos administrativos.
Ineciencia operativa.
Efectos acumulativos con tendencia al caos.
Debido a esto, se busca unicar y sincronizar la base de datos con la que cuenta dicha
empresa. Al unicar la base de datos se busca tener en la misma base de datos la informacin
suministrada por los sistemas SAP (Software de Gestin y Estrategia), GIS (Software de
Estimacin de Demanda) y SCADA (Software de registro de carga y data del sistema), ver
gura 3.1.
30
Figura 3.1: Base de Datos Unicada, [49].
Al sincronizar la base de datos es posible accesar a la misma desde cualquier computador
que haya sido sincronizado a dicha red. Esto elimina la necesidad de un computador central
que contenga toda la informacin de la base de datos, ver gura (3.2).
Figura 3.2: Sincronizacin de Base de Datos, [49].
El Instituto de Energa de la Universidad Simn Bolvar, (INDENE), en conjunto con
otras organizaciones y empresas, ha venido desarrollando un proyecto con el cual se busca
automatizar el proceso de estudios de la red de distribucin y desarrollar una plataforma
basada en cdigo abierto que permita, [49]:
31
1. Acceso inmediato y able a la informacin requerida para el proceso.
2. Desarrollo de herramientas tcnicas especcas en forma modular, garantizando sosteni-
bilidad y escalabilidad de la solucin tecnolgica.
3. Manejo de informacin tcnica a gran escala integrando los distintos sistemas existentes.
4. Eliminar dependencias en cuanto a plataformas de cdigo cerrado o propietario.
En, [50], se ha visualizado la elaboracin del proyecto en tres etapas:
Etapa 1. Implementacin del prototipo.
Etapa 2. Implementacin de Funciones Bsicas.
Etapa 3. Implementacin de Funciones Avanzadas.
El presente trabajo forma parte de la primera etapa del proyecto; siendo el ujo de carga
(FDC) desarrollado uno de los mdulos a implementar en el prototipo. En la segunda y
tercera etapa se denirn otras funciones de la herramienta computacional.
Para la aplicacin del proyecto se ha considerado que una arquitectura adecuada es el
patrn de diseo Modelo-Vista-Controlador (MVC), mostrado en la gura 3.3. Este patrn
ha tomado especial relevancia a raz de las nuevas implementaciones que se han logrado hacer
del mismo y la explosin en cuanto a programacin orientada a objetos.
Como se explica en [50], este patrn se compone de tres capas que permiten separar los
mbitos de trabajo de la aplicacin. El nivel superior es la Vista, la misma corresponde a la
interfaz de usuario en la cual se realizan las interacciones con el operador del programa. En la
parte ms baja se encuentra la capa correspondiente al Modelo. En ste se gestionan todas
las interacciones y validaciones con las fuentes de datos. Usualmente este proceso se delega en
paquetes conocidos como Mapeadores Objeto-Relacionales (o ORM por sus siglas en ingls)
que proveen un nivel de acceso de alto nivel convirtiendo las interacciones con las bases de
32
Figura 3.3: Modelo MVC, [50].
datos en algo ms acorde con la programacin orientada a objetos y liberando la denicin
de los datos de la implementacin (gestor de bases de datos seleccionado). Por ltimo, entre
las capas arriba mencionadas se encuentra la capa de los Controladores, stos se encargan
de las funciones inherentes a la lgica del programa, negociar las solicitudes de acciones de
las vistas y gestionar los datos en los modelos. En la gura 3.4 se puede apreciar el esquema
general del sistema que se est diseando, [49].
De acuerdo a [50], esta segmentacin de funciones permite mejorar el rendimiento a la
hora de hacer mantenimiento a la aplicacin o el control de cambios solicitados. Por ejemplo,
si se requieren cambios en la interfaz de usuario solamente es necesario realizar cambios en
el diseo de la capa correspondiente a la vista. De igual forma, si existe una reorganizacin
de la estructura de los datos o una implementacin en un nuevo gestor de bases de datos
se ver afectada solamente la capa del modelo. Del mismo modo, una nueva opcin dentro
del programa implicar la incorporacin de un controlador adecuado que gestione esta nueva
funcionalidad.
33
Figura 3.4: Esquema General de Herramienta de Planicacin Corto-Mediano Plazo
(HPCMP) - Visualizacin de Largo Plazo, [50].
34
3.2. Descripcin del Algoritmo Implementado
El algoritmo implementado fue desarrollado bajo la arquitectura MVC, descrita en la
seccin 3.1 del presente captulo. El programa est desarrollado de forma modular para opti-
mizar el tiempo de cmputo del mismo. ste est compuesto por dos mdulos: en el primero
se realiza el proceso de adquisicin de datos; en el segundo, se realiza el proceso iterativo y,
con las tensiones nodales, se realiza el anlisis de las condiciones de operacin de la red. Con
el primer mdulo se ltra la informacin de los archivos .DAT y se obtienen las matrices
TRX, S y los datos de nodos y lneas para clculos futuros; las cuales se guardan en archivos
.MAT y pueden ser almacenadas en disco o en memoria RAM. En el segundo mdulo se
realiza el proceso iterativo, se obtienen los voltajes nodales y se analizan los parmetros
correspondientes al estudio de prdidas tcnicas en conductores, cadas de tensin y capaci-
dad amperimtrica de los conductores en condiciones normales de operacin. Es importante
destacar que MATLAB es utilizado como programa interpretador del algoritmo implementa-
do.
En la gura 3.5 se muestra el esquema basado en la arquitectura MVC del programa
desarrollado en el presente proyecto. En el mdulo M se realiza el proceso de la adquisicin
de datos. En el mdulo C se lleva a cabo el proceso iterativo. Finalmente, en el mdulo V
se realiza el anlisis posterior de las condiciones de operacin de la red.
En la gura 3.6 se observa el diagrama de ujo del programa implementado.
3.3. Datos
Los archivos utilizados para la implementacin del programa en redes de gran tamao
fueron obtenidos de una data suministrada por la Electricidad de Caracas (EDC) en el ao
2006; dichos archivos se encuentran en formato .DAT. Los mismos son archivos de datos
generados automticamente por los sistemas de adquisicin de datos y mediciones, como
35
Figura 3.5: Esquema MVC del programa implementado.
el SCADA y el GIS, y en los mismos se almacena informacin referente a dicho programa
para uso interno del mismo. Los datos estn separados por comas (,) y la informacin de la
red est agrupada en bloques; cada bloque termina con la palabra END [51]. La estruc-
tura de los archivos DAT, su explicacin exhaustiva y un ejemplo se pueden encontrar en [51].
En la EDC, los archivos .DAT son utilizados por el programa ASP, desarrollado por
el profesor Alberto Naranjo, el cual tiene como funciones principales realizar un anlisis
consecuente con el (FDC), y clculo de corto circuito. Una vez que se ejecuta el (FDC) se
muestran las magnitudes de variables elctricas como tensin y corrientes pertenecientes al
sistema elctrico ordenadas en columnas, [50].
36
Figura 3.6: Diagrama de Flujo del Algoritmo Desarrollado.
37
3.4. Estructura del Programa
3.4.1. Adquisicin de Datos
Inicialmente, se desarroll un algoritmo para ltrar de los archivos .DAT los datos co-
rrespondientes a la potencia de las cargas, impedancias de lneas, tipos de nodo, distancia de
las lneas, factor de potencia, tensin nominal de operacin y topologa de la red.
Con este programa se obtiene la matriz de impedancias asociadas a la topologa de la red
(TRX, [20]) y la matriz de potencia aparente (S). Las matrices pueden ser almacenadas en
el disco o en memoria RAM. Para optimizar el proceso de ujo de carga se busca que las
matrices TRX y S estn disponibles en la memoria RAM del sistema y se actualizan oine
con cierta frecuencia. Es importante el proceso de actualizacin de ambas matrices porque
en la primera, (TRX), se registran los cambios en la red y en la segunda, (S), se registran
los cambios en la potencia demandada. La segunda matriz requiere un proceso de actualiza-
cin ms frecuente que la primera ya que la potencia demandada vara con ms frecuencia
que la conexin de la red. De esa forma se obtienen resultados ms objetivos y reales de la red.
En la seccin B.1 del apndice B se puede encontrar el cdigo del algoritmo desarrollado.
3.4.2. Proceso Iterativo
Con las matrices TRX y S de todos los circuitos disponibles en memoria RAM se realiza
el proceso iterativo, descrito en la seccin 2.1.2 del captulo 2. Como es de esperar, por la
simplicidad de los clculos, los resultados son obtenidos en pocas iteraciones y en un perodo
de tiempo considerablemente reducido en comparacin con otros algoritmos de FDC. Como
resultado de este modulo se obtienen las tensiones en mdulo y ngulo de todos los nodos de
la(s) red(es) estudiada(s).
38
Conocidas las tensiones de la red se realiza el anlisis correspondiente al estudio de ca-
das de tensin (seccin 2.3.1), prdidas tcnicas de potencia (ver seccin 2.3.2), capacidad
amperimtrica de conductores (ver seccin 2.3.3) y capacidad rme (seccin 2.3.4) en condi-
ciones normales de operacin. Adems, se muestra un sumario de los resultados obtenidos,
incluyendo tensiones nodales, prdidas tcnicas, cadas de tensin, cantidad de conductores
que operan por debajo del 67 % de su capacidad, cantidad de conductores que superan el
67 % y 100 % de su capacidad. Los resultados son almacenados en un archivo con la ex-
tensin .MAT. En el apndice C se encuentra el reporte que muestra el programa para un
circuito ejemplo. En este caso el circuito es el GRA_A01 correspondiente al circuito A1 de
la subestacin (S/E) Granada.
En la seccin B.2 del apndice B se puede encontrar el cdigo del algoritmo desarrollado.
CAPTULO 4
CASOS DE ESTUDIOS Y ANLISIS DE RESULTADOS
El algoritmo implementado fue probado en diferentes casos de estudio. En primer lugar,
se hizo un estudio en un circuito de 7 barras. Luego, se realiz la validacin de los resultados
del caso anterior con un programa de uso comercial. Despus, se realiz un estudio compa-
rativo entre los algoritmos propuestos por Teng (FDCT, [19]) y por De Oliveira (TRX, [20])
para comprobar la eciencia del ltimo respecto al primero. Finalmente, se implement el
algoritmo en circuitos de gran tamao de la Electricidad de Caracas, EDC. Con los resultados
obtenidos del ujo de carga (FDC) se realiz un anlisis de las condiciones de operacin de
la red estudiada.
4.1. Aplicacin en Circuito de 7 Barras
Para ilustrar el proceso de clculo del algoritmo se utilizar un circuito de 7 barras. El
proceso de clculo se realizar paso por paso para demostrar la lgica del mismo, dicho pro-
ceso se describe en la seccin 2.1.2 que se encuentra en el captulo 2. El sistema incluye 1
generador y 6 cargas equivalentes, conectadas a una red de distribucin radial. Los datos de
lneas y de nodos del circuito se muestran en las tablas 4.1 y 4.2, respectivamente.
Datos de Lnea El circuito est conformado por 6 lneas, descritas en la tabla 4.1. Los datos
se presentan en el siguiente orden: nodo de salida, nodo de llegada, resistencia y reactancia
40
de cada lnea en p.u. Las bases utilizadas son las siguientes: V
BASE
= 12, 47kV y S
BASE
=
100MVA
Tabla 4.1: Datos de Lnea (Ldat)
salida llegada r(p.u) x(p.u)
1 2 0,0265 0,0462
1 3 0,1005 0,0693
3 4 0,0670 0,0462
3 5 0,0265 0,0462
5 6 0,1005 0,0693
5 7 0,0670 0,0462
Datos de Nodo Los datos de los 7 nodos se encuentran en la tabla 4.2. Los mismos se
presentan en el siguiente orden: 1. n es el nmero de la barra. 2. Tipo corresponde al tipo
de barra: 1 barra slack, 2 barra PV y 3 barra PQ. 3. Pgen
0
es el valor inicial de la potencia
activa generada (p.u). 4. Qgen
0
es el el valor inicial de la potencia reactiva generada (p.u).
5. P
load
es la potencia activa demandada (p.u). 6. Q
load
es la potencia reactiva demandada
(p.u). 7. V
0
es la tensin inicial de la barra (p.u). 8. fcap es el factor de capacidad de la barra.
4.1.1. Clculo de la Matriz TRX
Como se describe en la seccin 2.1.2 del captulo 2, para calcular la matriz es necesario
hallar la matriz que relaciona las corrientes inyectadas con las corrientes de rama, [T], y las
matrices diagonales de resistencias de lnea, [Dr], y reactancias de lneas, [Dx].
Matriz [T]
En el apndice A se explica paso a paso cmo construir la matriz [T]. En esta matriz se
representa la topologa del circuito estudiado. El nmero 1 implica la existencia de una lnea
41
Tabla 4.2: Datos de Nodos, (Bdat)
n Tipo Pgen
0
(p.u) Qgen
0
(p.u) P
load
(p.u) Q
load
(p.u) V
0
fcap
1 1 0 0 0 0 1 1
2 3 0 0 0,1017 0,0635 1 1
3 3 0 0 0,0547 0,0342 1 1
4 3 0 0 0,0809 0,0596 1 1
5 3 0 0 0,1017 0,0635 1 1
6 3 0 0 0,0547 0,0342 1 1
7 3 0 0 0,0809 0,0596 1 1
entre el nodo correspondiente a la la y la columna respectiva.
T =

1 0 0 0 0 0
0 1 1 1 1 1
0 0 1 0 0 0
0 0 0 1 1 1
0 0 0 0 1 0
0 0 0 0 0 1

Matriz de Resistencias y Reactancias de lnea, Dr y Dx

Matriz diagonal de resistencias de lneas, Dr. En la ecuacin 2.7 se observa la forma de
la matriz.
Dr =

0,0265 0 0 0 0 0
0 0,1005 0 0 0 0
0 0 0,0670 0 0 0
0 0 0 0,0265 0 0
0 0 0 0 0,1005 0
0 0 0 0 0 0,0670

42
Matriz diagonal de reactancias de lneas, Dx. En la ecuacin 2.8 se observa la forma de
la matriz.
Dx =

0,0462 0 0 0 0 0
0 0,0693 0 0 0 0
0 0 0,0462 0 0 0
0 0 0 0,0462 0 0
0 0 0 0 0,0693 0
0 0 0 0 0 0,0462

Matriz TRX
La estructura de la matriz TRX se muestra en la ecuacin . Por motivos de espacio,
se presentarn las cuatro submatrices que conforman la matriz TRX por separado. De esta
forma,
T
T
D
R
T =

0.0265 0 0 0 0 0
0 0,1005 0,1005 0,1005 0,1005 0,1005
0 0,1005 0,1675 0,1005 0,1005 0,1005
0 0,1005 0,1005 0,127 0,127 0,127
0 0,1005 0,1005 0,127 0,2275 0,127
0 0,1005 0,1005 0,127 0,127 0,194

T
T
D
X
T =

-0.0462 0 0 0 0 0
0 -0,0693 -0,0693 -0,0693 -0,0693 -0,0693
0 -0,0693 -0,1155 -0,0693 -0,0693 -0,0693
0 -0,0693 -0,0693 -0,1155 -0,1155 -0,1155
0 -0,0693 -0,0693 -0,1155 -0,1848 -0,1156
0 -0,0693 -0,0693 -0,1155 -0,1156 -0,1617

43
T
T
D
X
T =

0.0462 0 0 0 0 0
0 0,0693 0,0693 0,0693 0,0693 0,0693
0 0,0693 0,1155 0,0693 0,0693 0,0693
0 0,0693 0,0693 0,1155 0,1155 0,1155
0 0,0693 0,0693 0,1155 0,1848 0,1156
0 0,0693 0,0693 0,1155 0,1156 0,1617

T
T
D
R
T =

0.0265 0 0 0 0 0
0 0,1005 0,1005 0,1005 0,1005 0,1005
0 0,1005 0,1675 0,1005 0,1005 0,1005
0 0,1005 0,1005 0,127 0,127 0,127
0 0,1005 0,1005 0,127 0,2275 0,127
0 0,1005 0,1005 0,127 0,127 0,194

4.1.2. Proceso Iterativo

Valores iniciales
Los valores de potencia y tensiones iniciales utilizados para empezar proceso algortmico
se muestran en la tabla 4.3. Los datos se presentan en el siguiente orden: 1. nmero del nodo
2. Potencia activa demandada en p.u. (P
load
) 3. Potencia activa consumida en p.u. (Q
load
) 4.
Componente real de la tensin inicial en p.u. (V
R0
) 5. Componente imaginaria de la tensin
inicial en p.u. (V
I0
).
Resultados de las Iteraciones
Construida la matriz TRX y conociendo el vector de potencias y tensiones iniciales es
posible realizar el FDC tal como se describe en la seccin 2.1.2 en el captulo 2. Para el cir-
cuito de 7 barras se lleg a un resultado en 3 iteraciones. El criterio de convergencia utilizado
44
Tabla 4.3: Valores de potencia y tensin inicial (V
0
, P
0
)
Nodo P
load
(p.u) Q
load
(p.u) V
R0
(p.u) V
I0
(p.u)
2 0,1017 0,0635 1 0
3 0,0547 0,0342 1 0
4 0,0809 0,0596 1 0
5 0,1017 0,0635 1 0
6 0,0544 0,0342 1 0
7 0,0809 0,0596 1 0
fue 0, 001. Se debe tomar en cuenta que se utiliz un criterio de convergencia bastante
conservador por motivos acadmicos. Los resultados de las iteraciones se muestran en las
tablas 4.4, 4.5 y 4.6, respectivamente.
Los resultados se presentan con la siguiente nomenclatura:
I
r
(p.u) y I
j
(p.u). Componente real e imaginaria de la corriente en el nodo i (p.u).
V
r
(p.u) y V
j
(p.u). Componente real e imaginaria de la tensin en el nodo i (p.u).
.V
r
y .V
j
. Diferencia de la componente real e imaginaria de tensin entre iteraciones.
Primera Iteracin
En la tabla 4.4 se muestran los resultados de la primera iteracin. Al terminar la primera
iteracin se observa que no se cumple el criterio de convergencia. En la mayora de los casos
la variacin es mayor a la tolerancia predeterminada.
45
Tabla 4.4: Resultados de la Primera Iteracin
Nodo I
r
(p.u) I
j
(p.u) V
r
(p.u) V
j
(p.u) V
r
V
j
2 -0,1017 0,0635 0,9944 -0,003 0,0056 0,003
3 -0,0547 0,0342 0,9452 -0,0005 0,054 0,0006
4 -0,0809 0,0596 0,9370 -0,0053 0,0631 0,0052
5 -0,1017 0,0635 0,9316 -0,0074 0,0684 0,0074
6 -0,0544 0,0342 0,9237 -0,0077 0,0762 0,0077
7 -0,0809 0,0596 0,9234 -0,0071 0,0766 0,0071
Segunda Iteracin
Tabla 4.5: Resultados de la Segunda Iteracin
Nodo I
r
(p.u) I
j
(p.u) V
r
(p.u) V
j
(p.u) .V
r
.V
j
2 -0,1025 0,0642 0,9943 -0,003 0,0001 0,0001
3 -0,0579 0,0362 0,9411 -0,0005 0,0040 0,0003
4 -0,0867 0,0641 0,9324 -0,0059 0,0631 0,0008
5 -0,1097 0,0690 0,9265 -0,0074 0,0051 0,0001
6 -0,0592 0,0375 0,9180 -0,0082 0,0058 0,0001
7 -0,0881 0,0652 0,9172 -0,0075 0,0058 0,0001
Al terminar la segunda iteracin se observa que no se cumple el criterio de convergencia
en una de las barras. La diferencia de tensin entre iteraciones del nodo 4 es mayor a la
tolerancia predeterminada.
Tercera Iteracin
Al terminar la tercera iteracin se observa que se cumple el criterio de convergencia en
todas las barras.
46
Tabla 4.6: Resultados de la Tercera Iteracin
Nodo I
r
(p.u) I
j
(p.u) V
r
(p.u) V
j
(p.u) .V
r
.V
j
2 -0,1042 0,0636 0,9943 -0,003 0,0001 0,0001
3 -0,0582 0,0363 0,9411 -0,0005 0,0001 0,0001
4 -0,0868 0,0642 0,9323 -0,0059 0,0001 0,0001
5 -0,1092 0,0676 0,9265 -0,0078 0,0001 0,0001
6 -0,0590 0,0367 0,9175 -0,0082 0,0005 0,0001
7 -0,0877 0,0643 0,9171 -0,0075 0,0001 0,0001
La tabla 4.7, muestra las tensiones del circuito en mdulo y ngulo. Adems se muestra
el porcentaje de cada de tensin respecto a la barra slack.
Tabla 4.7: Tensiones nodales en mdulo y ngulo
Nodo |V |(p.u)

(V )(grados) V( %)
1 1 0 0
2 0,9943 -0,1728 0,57
3 0,9411 -0,0204 5,89
4 0,9323 -0,0036 6,77
5 0,9265 -0,4823 7,35
6 0,9175 -0,5121 8,25
7 0,9171 -0,4685 8,29
Se observa que 5 nodos estn por debajo del criterio del 5 % de cada de tensin permitido
en condiciones de operacin normal, ver tabla 2.1. La mxima cada de tensin en el circuito
es de 8, 29 %.
47
4.2. Validacin de Resultados
Para validar los resultados se utiliz el programa MATPOWER. En la seccin E del
captulo 2 se describe dicho programa. Este programa est basado en el algoritmo Newton-
Raphson, NR. Sin embargo, ste trabaja con la esparsidad de las matrices para evitar con-
struir la matriz Jacobiana, tpica en el NR. Principio que hace del programa un algoritmo
bastante poderoso y eciente. De esta forma, se evitan problemas de convergencia tpicos en
redes de distribucin, (ver seccin 1.1 del captulo 1). Para mayor informacin referente al
programa se recomienda revisar el manual de usuario, [52].
Para realizar la validacin se utiliz como caso de ejemplo el circuito explicado en la sec-
cin 4.1 del presente captulo. Las tensiones nodales se muestran en la tabla 4.8. Los datos de
presentan de la siguiente forma: 1. Nombre del nodo. 2. Mdulo de tensin en p.u. (|V |). 3.
ngulo de tensin en grados (

(V )). 4. Error porcentual del mdulo de las tensiones obtenidas

con el algoritmo implementado respecto a las del MATPOWER. 5. Error porcentual del ngu-
lo de las tensiones obtenidas con el algoritmo implementado respecto a las del MATPOWER.
Tabla 4.8: Tensiones Nodales Validadas en Mdulo y ngulo
Nodo |V |(p.u)

(V )(grados) Error |V |( %) Error

(V )( %)
1 1 0 0 NA
2 0,9943 -0,1737 0 0,5181
3 0,9407 -0,0206 0,0425 0,9708
4 0,9319 -0,0039 0,0429 7,6923
5 0,9260 -0,4699 0,0540 2,6389
6 0,9174 -0,4937 0,0109 3,7269
7 0,9171 -0,4526 0 3,5130
Se observa que el mayor error en el mdulo de las tensiones es slo del 0, 0540 %; mientras
48
que en el ngulo de las tensiones se tuve un error mximo correspondiente al 7, 6923 %. En el
apndice E, se encuentra el reporte generado por el MATPOWER como resultado del ujo
de carga del circuito de 7 barras explicado en la seccin 4.1.
4.3. Estudio Comparativo
En [20], se estudia el tiempo de proceso de CPU utilizando tres algoritmos:
1. El algoritmo propuesto en [20], (TRX), (ver seccin 2.1.2 en el captulo 2).
2. El algoritmo propuesto en [19], (FDCT), (ver seccin 2.1.1 en el captulo 2).
3. Newton-Raphson, (NR), utilizado por el programa MATPOWER (ver seccin E en el
captulo 2).
Para dicho estudio no se consider el tiempo de entrada/salida de datos. Esto ltimo se
justica porque en la aplicacin del mismo se planea que un programa centinela mantenga
las matrices de los datos de entrada en memoria RAM y actualizados peridicamente; razn
por la cual, no se invertira tiempo en cargar los datos sino que se tendran disponibles en
cualquier momento que se requiera.
El caso de estudio fue resuelto utilizando una red de n nodos variando la variable n de
1.000 hasta 3.000 nodos. La gura 4.1, obtenida de dicho artculo, muestra el tiempo de
convergencia utilizado en el proceso iterativo de cada algoritmo. El grco muestra que el
TRX presenta un mejor tiempo de cmputo que el FDCT y el NR.
El TRX es desde un punto de vista computacional ms eciente que los otros dos algo-
ritmos porque el proceso est basado en la suma y multiplicacin de nmeros reales que se
encuentran previamente alojados en memoria RAM. La matriz TRX no requiere ser actuali-
zada en cada iteracin tal como se hace con la matriz Jacobiana del NR. Motivo por el cual,
49
Figura 4.1: Estudio Comparativo: Tiempo de Cmputo, [20].
el TRX es un algoritmo altamente competitivo respecto al utilizado por el MATPOWER,
el cual utiliza la esparsidad de las matrices para no armar la matriz Jacobiana y, en conse-
cuencia, reducir el tiempo de computo al evitar tener que invertir y actualizar dicha matriz
en cada iteracin.
En el caso propuesto, el NR tuvo el segundo mejor comportamiento por lo anteriormente
descrito. El peor tiempo se registr al utilizar el algoritmo FDCT en nmeros complejos ya
que el tiempo requerido para realizar las operaciones se incrementa al trabajar con nmeros
complejos, [20].
Por otro lado, se decidi corroborar dicho estudio de forma independiente. Slo se com-
prob la eciencia del (TRX) respecto al FDCT. Dicho estudio se realiz tomando en cuenta
el nmero de iteraciones y el tiempo de computo de ambos algoritmos.
Para esto se utilizaron cuatro (4) circuitos de diferentes tamaos: 4, 7, 12 y 69 barras.
Los datos del circuito de 4 barras se encuentra en el apndice D, el circuito de 7 se describi
en la seccin 4.1 y los circuitos de 12 y 69 barras se encuentran disponibles en los apndices
J e I, respectivamente. El nmero de iteraciones para la convergencia es el mismo en ambos
50
algoritmos ya que se basan en el mismo principio.
Para poder estudiar el tiempo de respuesta de ambos algoritmos, se forz el ciclo de itera-
ciones para que realizaran un total de 5.000 iteraciones. De esta forma, fue posible hacer la
medicin del tiempo en el que los dos algoritmos realizan las operaciones matemticas. Esto
se fundamenta en el hecho de que los procesadores actuales realizan estos procesos en tiempos
muy pequeos, lo cual diculta su medicin. Sin embargo, al aumentar apreciablemente el
nmero de iteraciones se puede medir el tiempo del proceso.
En la tabla 4.9 se muestran los resultados obtenidos. Se observa que los tiempos se reducen
aproximadamente a la mitad al aplicar la modicacin de [20]. Para este estudio se utiliz
un computador de 2GB de memoria RAM y un procesador Intel Core 2 Duo.
Tabla 4.9: Tiempo de cmputo de [19] y [20] para circuitos de 4, 7, 12 y 69 barras
Estudio con 5000 iteraciones
N Barras [20] (mseg) [19] (mseg)
69 243,61 432,03
12 45,48 84,84
7 35,61 79,54
4 30,33 74,22
Los resultados demuestran que el cmputo en nmeros reales utilizado por el TRX es
ms eciente en comparacin con el clculo en nmeros complejos utilizado por FDCT, a
pesar de que las matrices son de mayor dimensin en el TRX.
51
4.4. Aplicacin en Redes de Gran Tamao
4.4.1. Tiempo de cmputo
Utilizando la data proporcionada por la EDC se implement el programa desarrollado en
un total de 530 circuitos distribuidos en 78 S/E, compuestos por un total de 64.251 nodos
que corresponden a un 80 % de la demanda mxima de la Gran Caracas (3, 28GW).
Para probar la eciencia del algoritmo se compar el tiempo de cmputo del mismo con
el utilizado por el programa ASP; programa desarrollado por el profesor Alberto Naranjo.
Dicho programa es utilizado actualmente por la EDC para realizar estudios de planicacin
a corto y mediano plazo. La utilizacin de esta herramienta permite con ciertas limitaciones
obtener resultados en tiempo til, [50]. En la tabla 4.2 se muestra el tiempo de cmputo del
ASP. Para este estudio se utiliz un computador de 2GB de memoria RAM y un procesador
Intel Core 2 Duo.
Tabla 4.10: Tiempo de cmputo del ASP para red de 530 circuitos de la EDC
Tiempo Total(seg)
ASP 438,82
En la gura 4.2 se muestra de forma esquemtica el proceso algortmico que utiliza el
ASP y el tiempo requerido para obtener la solucin del ujo de carga para el caso planteado
anteriormente.
En contraste, el tiempo utilizado por el algoritmo implementado se muestra en la tabla
4.11. Los resultados presentados son de cada uno de los mdulos que componen el algoritmo.
Para este estudio, se dividi el algoritmo mostrado en la seccin B.2 del apndice B en dos
submdulos: el primero correspondiente al proceso iterativo y el segundo correspondiente al
52
Figura 4.2: Tiempo de Cmputo del ASP para la Red de la Gran Caracas.
anlisis de condiciones de la red.
Tabla 4.11: Tiempo de cmputo del algoritmo implementado para red de 530 circuitos de la
EDC
Mdulo Tiempo Total(seg)
Adquisicin de Datos 1.700
Proceso Iterativo 0,76
Resultados 3,22
En la gura 4.3 se muestra de forma esquemtica el proceso algortmico que utiliza el
algoritmo desarrollado y el tiempo requerido para obtener la solucin del ujo de carga para
el caso planteado anteriormente.
A primera vista se observa que, en trminos generales, el ASP requiri de menor tiempo
para obtener un resultado del FDC. Sin embargo, es importante resaltar que varios hechos
relevantes que demostrarn que la armacin inicial no es del todo valida.
En primer lugar, hacer una comparacin en igualdad de condiciones ente ambos algorit-
53
Figura 4.3: Tiempo de Cmputo del Algoritmo Implementado para la Red de la Gran Caracas.
mos no es del todo vlido ya que el programa del profesor Naranjo es un programa compilado
en un lenguaje de bajo nivel como lo es DELPHI a diferencia del algoritmo implementado
que no ha sido compilado y utiliza MATLAB como interpretador.
Por otro lado, el mdulo que requiri de mayor tiempo de proceso fue el modulo de
adquisicin de datos. El tiempo utilizado fue de 1.700 segundos; es decir, el mdulo tard en
promedio 3, 21 segundos en procesar la data de cada circuito. No obstante, como se explic
en la seccin 3.4.1 del captulo 3, en este mdulo se realiza el ltrado de los datos y se colocan
las matrices TRX y S en memoria RAM para que estn disponibles en cualquier momento
que se requiera, lo cual optimiza el proceso iterativo.
An ms, se debe recordar que el proceso de adquisicin de datos y actualizacin de las
matrices arriba mencionadas se realiza oine. Razn por la cual no resulta afectado el pro-
ceso iterativo del programa. A diferencia del ASP que tiene que recurrir al disco duro para
adquirir la informacin y luego realizar el ujo de carga.
Adems, es vlido acotar que el espacio ocupado por las matrices TRX y S de los 530
circuitos es de 40, 6MB. Es decir, el espacio requerido para almacenar las matrices de los
54
circuitos de una red de gran tamao es insignicante en comparacin con la capacidad que
poseen los procesadores actualmente. En la tabla 4.12, se muestra el espacio ocupado en disco
por 10, 50, 100, 200 y 530 circuitos elegidos aleatoriamente. Los datos son presentados de la
siguiente forma: 1. Nmero de circuitos. 2. Espacio ocupado en disco en MB. 3. Nmero de
nodos totales.
Tabla 4.12: Espacio Ocupado en Disco por redes de 10, 50, 100, 200 y 530 circuitos
Nmero de Circuitos Espacio en Disco(MB) Nmero de Nodos
10 2,029 1.168
50 3,78 5.934
100 6,6 11.346
200 20 23.115
530 40,6 64.251
Por otra parte, se observa que el tiempo requerido para obtener el resultado del FDC
de toda la red estudiada fue 0, 76 segundos utilizando un computador de uso comn. Es
decir, en promedio se obtuvo el resultado del FDC de cada circuito en 1, 43 milisegundos.
Al compararlo con el ASP, que en promedio tena la solucin en 0, 75 segundos por circuito,
se puede concluir que el TRX es ms eciente. Dicha eciencia se debe a la sencillez de la
lgica de clculo, al hecho de que los clculos se realizan en nmeros reales y a la disposicin
de la informacin de los circuitos (en forma matricial) en memoria RAM.
4.4.2. Anlisis de Condiciones Operacionales de una Red de Gran
Tamao
Anlisis Ejemplo para 1 Circuito
Conocidas las tensiones nodales en mdulo y ngulo se realiz un anlisis de las condi-
ciones operacionales del circuito ANT_A04 correspondiente al circuito A04 de la S/E An-
55
tmano. Para este circuito se estudi la cada de tensin, la capacidad amperimtrica y las
prdidas tcnicas.
La gura 4.4 muestra el perl de la cada de tensin en el circuito Antmano A04, confor-
mado por un total de 318 nodos. Es lgico obtener que a medida que los nodos se alejan de la
barra Slack, stos tengan un mayor porcentaje de cada de tensin. Sin embargo, la mxima
cada de tensin de este circuito es de 15 %, valor que supera el criterio de cada de tensin
del 5 % de la EDC, ver seccin 2.3.1 del captulo 2.
Figura 4.4: Perl de Cada de Tensin en ( %) del circuito ANT_A01.
La gura 4.5 muestra la potencia activa demandada y las prdidas tcnicas en relacin
con la potencia activa total entregada. Los clculos se realizaron como se explica en la seccin
2.3.2 del captulo 2. La potencia activa entregada al circuito es de 16, 59MW, de los cuales
un 8 % representa las prdidas tcnicas en el circuito, valor que est por encima del 4 %
permitido, [56]. En promedio el valor de las prdidas para toda la red oscila alrededor del
2 %, [54].
La gura 4.6 muestra la potencia reactiva demandada y las prdidas tcnicas en relacin
con la potencia activa total entregada. La potencia reactiva entregada al circuito es de
11, 72MVAr, de los cuales un 20 % representa las prdidas tcnicas en el circuito, valor que
est por encima del 4 % permitido, [56].
56
Figura 4.5: Potencia Activa Demandada y Prdidas Tcnicas del Circuito ANT_A01.
Figura 4.6: Potencia Reactiva Demandada y Prdidas Tcnicas del Circuito ANT_A01.
57
La gura 4.7 muestra la cantidad de conductores del circuito que estn operando por de-
bajo del 67 % de la capacidad de emergencia (CE), aquellos que operan por encima del 67 %
de la CE y aquellos que operan por encima del 100 % de la CE. La capacidad amperimtrica
de los conductores se calcul como se explica en la seccin 2.3.3 del captulo 2. Para un total
de 318 ramas, se observa que 7 (2 %) de los conductores superan el 100 % de la CE, 141
(44 %) de los conductores operan por encima del 67 % de la CE. Slo 171 (54 %) operan por
debajo del 67 % de la CE. Esto indica que se debe tratar de equilibrar la distribucin de las
cargas en el circuito para reducir el nmero de conductores que operan por encima 67 % o
del 100 % de la CE.
Figura 4.7: Capacidad Amperimtrica de los conductores del circuito ANT_A01.
Tambin, se analiz la potencia entregada por circuito por la S/E comparndola con las
prdidas de los mismos. Adems, se veric la carga de los conductores de todos los circuitos
de la S/E.
58
Anlisis Ejemplo para 1 S/E
Conocidas las tensiones nodales en mdulo y ngulo se realiz un anlisis de las condi-
ciones operacionales de la S/E Antmano (ANT). Para esta S/E se estudi la capacidad
amperimtrica de los conductores de los circuitos que conforman dicha S/E, se compar la
potencia entregada por circuito respecto a las prdidas tcnicas de los mismos en ( %) y
estudi la capacidad rme de dos S/E: Antmano ANT y Santa Rosa SRO.
En la gura 4.8 se muestra la potencia entregada junto con el porcentaje de prdidas por
circuito de la S/E Antmano. Se observa que el circuito con mayor potencia entregada es el
ANT_B04 con 22MVA; mientras que, el que menos potencia demanda es el ANT_B02 con
2, 3MVA. Por otro lado, el circuito que presenta ms prdidas es el ANT_A012 con 3, 1MVA
en prdidas tcnicas. Para este anlisis no se incluyen prdidas en los transformadores.
Figura 4.8: Potencia Entregada y Perdida por Circuito de la S/E ANT.
En la gura 4.9 se muestra la cantidad de conductores del circuito que estn operando
por debajo del 67 % de la capacidad de emergencia (CE), aquellos que operan por encima
del 67 % de la CE y aquellos que operan por encima del 100 % de la CE.
La capacidad amperimtrica de los conductores se calcul como se explica en la seccin
59
Figura 4.9: Capacidad Amperimtrica de los conductores de la S/E ANT
2.3.3 del captulo 2. Para un total de 1,649 ramas, se observa que 20 (1 %) de los conductores
superan el 100 % de la CE, 678 (41 %) de los conductores operan por encima del 67 % de la
CE. Slo 952 (58 %) conductores operan por debajo del 67 % de la CE. En trminos generales
ms del 40 % de los conductores trabajan por encima del 67 % de su CE.
Para nalizar, en la gura 4.10 se muestra la potencia entregada por 2 S/E: Antmano y
Santa Rosa. La potencia total de cada circuito se calcul sumando la potencia suministrada
por cada circuito perteneciente a dicha S/E. Como caso ejemplo, se asume que cada S/E
cuenta con 4 transformadores de 33, 3MVA. Aplicando el criterio de capacidad rme CF,
explicado en la seccin 2.3.4 del captulo 2, se obtiene que la CF de las S/E debe ser apro-
ximadamente 100MVA. Las potencias suministradas por los circuitos se compararon con el
valor terico de CF por S/E.
La potencia suministrada por las S/E ANT es 114MVA y SRO es 49, 5MVA, respecti-
vamente. Se observa que la S/E ANT supera el valor terico por 14MVA mientras que la
S/E SRO est aproximadamente 50MVA por debajo de dicha capacidad. Se podra pensar
60
Figura 4.10: Capacidad Firme de las S/E ANT y SRO.
en una mejor distribucin de la carga.
Anlisis Ejemplo para Toda la Red
Conocidas las tensiones nodales en mdulo y ngulo se realiz un anlisis de las condi-
ciones operacionales de toda la red de la Gran Caracas; conformada por 530 circuitos corres-
pondientes a 78 S/E. La cantidad total de nodos de la red estudiada es 64,251. Se estudi la
capacidad amperimtrica de todos los conductores de la red y las prdidas activas en com-
paracin con la potencia activada entregada.
La gura 4.11 muestra la potencia activa demandada y las prdidas tcnicas en relacin
con la potencia activa total entregada. Los clculos se realizaron como se explica en la seccin
2.3.2 del captulo 2. La potencia activa entregada al circuito es de 3, 28GW, de los cuales un
558MW representa las prdidas tcnicas en el circuito, valor que est por encima del por-
centaje de prdidas permitido. En promedio el valor de las prdidas para toda la red oscila
alrededor del 2 %, [54]. Sin embargo, se debe recordar que la data con la que se trabaj estaba
corrupta. Por lo tanto, los resultados ac mostrados se vieron afectados por el estado de la
61
data con la que se trabaj.
Figura 4.11: Potencia Activa Demandada y Prdidas Tcnicas de la red.
En la gura 4.12 se muestra la cantidad de conductores de la red que estn operando por
debajo del 67 % de la capacidad de emergencia (CE), aquellos que operan por encima del
67 % y aquellos que operan por encima del 100 % de CE.
La capacidad amperimtrica de los conductores se calcul como se explica en la seccin
2.3.3 del captulo 2. Para un total de 63.721 ramas, se observa que 22.130 (3 %) de los con-
ductores superan el 100 % de la CE, 24.227 (35 %) de los conductores operan por encima del
67 % de la CE. Slo 39.494 (62 %) conductores operan por debajo del 67 % de la CE. En
trminos generales ms del 35 % de los conductores trabajan por encima del 67 % de su CE.
En la tabla 4.13 se muestra un reporte general de las condiciones de operacin de la red.
62
Figura 4.12: Capacidad Amperimtrica de los conductores de la red.
Tabla 4.13: Condiciones de Operacin de la Red, 530 circuitos.
Parmetro Analizado Resultado
Prdidas Activas Totales 557, 27MW
Prdidas Reactivas Totales 808, 25MVAr
Potencia Activas Totales Demandada 3, 28GW
Potencia Reactivas Totales Demandada 2, 03GVAr
Conductores que violan 67 % de la CE 24.227
Conductores que violan 100 % de la CE 22.130
Total de Nodos Analizados 64.251
CONCLUSIONES
A continuacin se plantean las conclusiones de este trabajo de investigacin:
Se implement un ujo de carga de barrido unidireccional directo basado en el BIBC
(desarrollado por Teng, [19]). Incluyendo la modicacin propuesta por De Oliveira
(TRX), [20], para hacer ms eciente al algoritmo inicial.
El algoritmo fue codicado en MATLAB y se caracteriza por presentar una lgica de
clculo de gran sencillez: se basa en una suma y multiplicacin de matrices. Adems, el
mismo fue desarrollado de forma modular, compuesto por dos mdulos diferentes: en
el primero se adquieren los datos y en el segundo se realiza el proceso de clculo.
El proyecto se aplic con un patrn Modelo-Vista-Controlador (MVC).
Se demostr que la modicacin del ujo de carga propuesta por De Oliveria, [20], con-
vierte al algoritmo original en un algoritmo ms eciente; reduciendo aproximadamente
a la mitad los tiempos de cmputo, ver seccin 4.3 del captulo 4.
El algoritmo desarrollado fue validado utilizando un ujo de carga directo complejo,
[19], y un ujo de carga basado en el Newton-Raphson, [52].
Para hacer an ms eciente al algoritmo se estableci un esquema de datos en RAM,
en el cual se colocaron las matrices que caracterizan al circuito (TRX y S). De esta
forma, estarn disponibles en cualquier momento que se requiera hacer el ujo de Carga.
An ms, las matrices en el caso de estudio de 530 circuitos slo ocuparon un espacio
en memoria de 40MB.
El algoritmo fue implementado ecientemente en una red de 530 circuitos correspon-
dientes a 78 S/E de la EDC. Las tensiones nodales para los 64.251 nodos se obtuvo en
un tiempo de 0, 76 segundos.
63
64
Comentarios y Recomendaciones
Debido a todos los problemas ocasionados por la data de los circuitos de la EDC se
recomienda depurar la data existente, haciendo especial nfasis en los datos de las
impedancias.
Se recomienda ir a un modelo comn de datos en XML.
A n de tener un registro de las cargas y poder tener disponibles sus respectivas curvas
se recomienda tipicar el consumo de los clientes.
REFERENCIAS BIBLIOGRFICAS
[1] Tinney, W. F., Hart, C. E., Power Flow Solution by Newtons Method, IEEE
Transactions on Power Apparatus and Systems, Vol. 86, No. 11, pp. 1449-1460, Nov.
1967.
[2] Eminoglu, U., Hocaoglu, M. H., Distribution Systems Forward/Backward Sweep-based
Flow Algorithms: A Review and Comparison Study, Electric Power Components and
Systems, Vol. 37, No. 1, 2009, pp. 91-110.
[3] Rypin Cazal, F., Barrenechea Gruber, R., Flujo de Carga en los Sistemas Integrales
de Distribucin, Proyecto de Grado presentado a la Universidad Simn Bolvar, Mar.
2006.
[4] Iwamoto, S., Tamura, Y., A Load Flow Calculation Method for Ill-Conditioned Power
Systems, IEEE Transactions on Power Apparatus and Systems, Vol. PAS-100, No. 4,
pp. 1736-1743, Abr. 1981.
[5] Tripathy, S. C., Durga, G., Malik, O. P., Hope, G. S., Load-Flow Solutions for Ill-
Conditioned Power Systems by Newton-Like Method, IEEE Transactions on Power
Apparatus and Systems, Vol. PAS-101, No. 10, pp. 3648-3657, Oct. 1982.
[6] Zhang, F., Cheng, C. S., A Modied Newton Method for Radial Distribution System
Power Flow Analysis, IEEE Transactions on Power Systems, Vol. 12, No. 1, pp. 389-397,
Feb. 1997.
[7] Da Costa, V. M., Martins, N., Pereira, J. L., Development in the Newton Raphson
Power Flow Formulation Based on Current Injections, IEEE Transactions on Power
Systems, Vol. 14, No. 4, pp. 1320-1326, 1999.
65
66
[8] Garcia, P. A. N., Pereira, J. L. R., Carneiro, S., Da Kosta, V. M., Martins, N., Three-
Phase Power Flow Calculations Using the Current Injection Method, IEEE Transactions
on Power Systems, Vol. 15, No. 2, pp. 508-514, May. 2000.
[9] Lin, W. M., Teng, J. H., Three-Phase Distribution Network Fast-Decoupled Power Flow
Solutions, Electrical Power and Energy Systems, Vol. 22, No. 5, pp. 375-380, 2000.
[10] Teng, J. H., A Modied Gauss-Seidel Algorithm of Three-Phase Power Flow Analysis
in Distribution Networks, Electrical Power and Energy Systems, Vol. 24, No. 2, pp.
97-102, 2002.
[11] Shirmohammadi, D., Hong, H. W., Semlyen, A., Luo, G. X., A Compensation-Based
Power Flow Method for Weakly Meshed Distribution and Transmission Networks, IEEE
Transactions on Power Systems, Vol. 3, No. 2, pp. 753-762, May. 1988.
[12] Cheng, C. S., Shirmohammadi, D., A Three-Phase Power Flow Method for Real-Time
Distribution System Analysis, IEEE Transactions on Power Systems, Vol. 10, No. 2,
pp. 671-679, May. 1995.
[13] Zhu, Y., Tomsovic, K., Adaptive Power Flow Method for Distribution Systems with
Dispersed Generation, IEEE Transactions on Power Delivery, Vol. 17, No. 3, pp. 822-
827, Jul. 2002.
[14] Ghosh, S., Das, D., Method for Load-Flow Solution of Radial Distribution Networks,
IEE Proc. Generation Transmission and Distribution, Vol. 146, No. 6, pp. 641-648, 1996.
[15] Thukaram, D., Banda, H. M. W., Jerome, J., A robust Three Phase Power Algorithm
for Radial Distribution Systems, Electrical Power Systems Research, Vol. 50, No. 3, pp.
227-236, 1999.
[16] Kersting, W. H., Distribution System Modeling and Analysis, CRC Press, Nuevo M-
xico, pp. 269-302 (2002).
[17] Liu, J., Salama, M. M. A., Mansour, R. R., An Ecient Power Flow Algorithm
for Distribution Systems with Polynomial Load, International Journal of Electrical
Engineering Education", Vol. 39, No. 4, pp. 371-386, 2002.
67
[18] Teng, J. H., A Network-Topology Based Three-Phase Load Flow for Distribution
Systems, Proc. Natl. Sci. Counc. ROC (A), Vol. 24, No. 4, pp. 259-264, 2000.
[19] Teng, J. H., A Direct Approach for Distribution System Load Flow solution, IEEE
Transactions on Power Delivery, Vol. 18, No. 3, pp. 882-887, Jul. 2003.
[20] De Oliveira, P. M., The Distribution TRX-Power Flow Method, Universidad Simn
Bolvar, artculo aprobado para publicacin, 2010.
[21] Augugliaro, A., Dusonchet, L., Favuzza, S., Ippolito, M. G., and Sanseverino, E. R.,
A New Backward/Forward Method for Solving Radial Distribution Network with PV
Nodes, Electrical Power Systems Research, Vol. 78, No. 3, pp. 330-336, Mar. 2008.
[22] Cspedes, G. R., New Method for the Analysis of Distribution Networks, IEEE
Transactions on Power Delivery, Vol. 5, No. 1, pp. 391-396, Ene. 1990.
[23] Haque, M. H., Load Flow Solution of Distribution Systems with Voltage Dependent
Load Models, Electrical Power Systems Research, Vol. 36, No. 3, pp. 151-156, 1996.
[24] Haque, M. H., Ecient Load Flow Method for Distribution Systems with Radial or
Mesh Conguration, IEE Proc. Generation Transmission and Distribution, Vol. 143,
No. 1, pp. 33-38, 1996.
[25] Haque, M. H., A General Load Flow Method for Distribution Systems, Electrical Power
Systems Research, Vol. 54, No. 1, pp. 106-114, 2000.
[26] Asfari, M., Singh, S. P., Raju, G. S., Rao, G. K., A Fast Power Flow Solution of Radial
Distribution Networks, Electric Power Components and Systems, Vol. 31, No. 1, pp.
1065-1074, 2002.
[27] Ranjan, R., Das, D., Simple and Ecient Computer Algorithm to Solve Radial
Distribution Networks, Electric Power Components and Systems, Vol. 31, No. 1, pp.
95-107, 2003.
[28] Das, D., Nagi, H. S., Kothari, D. P., Novel Method for Solving Radial Distribution
Networks, IEE. Proc. Generation Transmission and Distribution, Vol. 141, No. 4, pp.
291-298, 1994.
68
[29] Das, D., Kothari, D. P., Kalam, A., Simple and Ecient Method for Load Flow Solution
of Radial Distribution Networks, Electrical Power and Energy Systems, Vol. 17, No. 5,
pp. 335-346, 1995.
[30] Eminoglu, U., Hocaoglu, M. H., A New Power Flow Method for Radial Distribution
Systems Including Voltage Dependent Load Models, Electrical Power Systems Research,
Vol. 76, No. 1-3, pp. 106-114, 2005.
[31] Baran, M. E., Wu, F. F., Optimal Sizing of Capacitors Placed on a Radial Distribution
System, IEEE Transactions on Power Delivery, Vol. 4, No. 1, pp. 735-742, 1989.
[32] Baran, M. E., Wu, F. F., Network Reconguration in Distribution Systems for Loss
Reduction and Load Balancing, IEEE Transactions on Power Delivery, Vol. 4, No. 2,
pp. 1401-1407, Abr. 1989.
[33] Luo, G. X., Semlyen, A., Ecient Load Flow for Large Weakly Meshed Networks,
IEEE Transactions on Power Systems, Vol. 5, No. 4, pp. 1309-1313,1990.
[34] Jasmon, G. B., Lee, L. H. C., Distribution Network reduction for Voltage Stability
Analysis and Load Flow Calculations, Electrical Power and Energy Systems, Vol. 13,
No. 1, pp. 9-13, 1991.
[35] Rajicic, D., Ackovski, R., Taleski, R., Voltage Correction power Flow, IEEE
Transactions on Power Delivery, Vol. 9, No. 2, pp. 1056-1062, 1994.
[36] Satyanarayana, S., Ramana, T., Sivanagaraju, S., and Rao, G. K., An Ecient Load
Flow Solution for Radial Distribution Network Including Voltage Dependent Load
Models, Electric Power Components and Systems, Vol. 35, No. 5, pp. 539-551, 2007.
[37] Graigner, J. J., Stevenson, W. D., Anlisis de Sistemas de Potencia, MC Graw Hill,
Mxico, pp. 309-357 (2004).
[38] Bermdez, J. F., Anlisis de Flujo de Carga en sistemas Elctricos de Potencia, Versin
B, Mar. 1994.
[39] Rajicic, D., Bose, A., A Modication to the Fast decoupled Power Flow for Methods
with High R/X Ratios, IEEE Transactions on Power Systems, Vol. 3, No. 2, pp. 743-
746,1988.
69
[40] Baldwin, T. L., Lewis, S. A., Distribution Load Flow Methods for Shipboard Power
Systems, Vol. 40, No. 5, pp. 812-819, Sept. 2004.
[41] Lin, W. M., Su, Y. S., Chin, H. C., Teng, J. H., Three-phase unbalanced distribution
power ow solutions with minimum data preparation, IEEE Transactions on Power
Systems, Vol. 14, No. 3, pp. 1178-1183, Aug. 1999.
[42] Expsito. G. A., Augmented Rectangular Flow Model, IEEE Transactions on Power
Systems, Vol. 17, No. 2, May. 2002.
[43] Expsito. G. A., Ramos, E. R., Reliable Load Flow Technique for Radial Distribution
Networks, IEEE Transactions on Power Systems, Vol. 14, No. 3, pp. 1063-1069, Aug.
1999.
[44] Jovanovic, S., Milicevic, F., Triangular Distribution Load ow, IEEE Power
Engineering Review, pp. 60-62, May. 2000.
[45] Baran, M. E., Staton, E. A., Distribution Transformer Models for Branch Current Based
Feeder Analysis, IEEE Transactions on Power Systems, Vol. 12, pp. 698-703, 1997.
[46] Stott, B., Alsac, O., Fast Decoupled Load Flow, IEEE Transactions on Power
Apparatus and Systems, pp. 859-869, May/Jun. 1974.
[47] Younes, M., Rahli, M., Abdelhakem K, L., Economic Power Dispatch Using Evolu-
tionary Algorithm, Journal of ELECTRICAL ENGINEERING, Vol. 57, No. 4, pp.
211-217, 2006.
[48] Expsito. G. A., Conejo. J. A., Caizares. C.,Electric Energy Systems, Analysis and
Operation, CRC Press, Florida, pp. 95-126 (2009).
[49] Alvarez, M., Anteproyecto para el Desarrollo de Herramienta Computacional de Sis-
temas Elctricos de Distribucin, INDENE-Parque Tecnolgico Sartenejas, 2010.
[50] Arteaga, A., Rojas, A., Ruiz C., Informe Final: CONTRATO N NCO/ 0909-589:
ANTEPROYECTO PARA EL DESARROLLO DE HERRAMIENTA COMPUTA-
CIONAL PARA ESTUDIOS DE PLANIFICACIN CORTO-MEDIANO PLAZO DE
70
SISTEMAS ELCTRICOS DE DISTRIBUCIN"DOC. No. EDC2010 10 001,
INDENE-Parque Tecnolgico Sartenejas, 2010.
[51] PSS/ADEPT 4, User Manual, Power Technologies INC, Nueva York,pp. B1-B22
(Febrero 2002).
[52] Zimmerman, R., Murillo S., C. E., MATPOWER: A MATLAB Power System
Simulation Package, Users Manual, Power Systems Engineering Research Center
(PSerc), Nueva York, Versin 3.2 (Septiembre 2007).
[53] Younes, M., Rahli, M., Abdelhakem K, L., Economic Power Dispatch Using Evolu-
tionary Algorithm, Journal of ELECTRICAL ENGINEERING, Vol. 57, No. 4, pp.
211-217, 2006.
[54] Rodrgues Goncalves, C., Estudio de Planicacin a Mediano Plazo de la S/E Esme-
ralda de 12,47KV de la Electricidad de Caracas, Proyecto de Grado presentado a la
Universidad Simn Bolvar, Ene. 2008.
[55] Naranjo, A., Gua de Comandos del Porgrama ASPV7A, Manual, Caracas, 2003.
[56] Pia, Y., Estudio de Planicacin a Mediano Plazo de dos Subestaciones de 100MVA
de la Regin Este de La Electricidad de Caracas, Proyecto de Grado presentado a la
Universidad Central de Venezuela, Nov. 2005.
Apndice A
CASO EXPLICATIVO: MATRICES BIBC Y BCBV
Para ilustar el proceso de armado de la matriz BIBC se utiliza el circuito de 6 barras
disponible [19]. El sistema incluye 1 generador y 5 cargas equivalentes, conectadas a una red
de distribucin radial. El circuito se muestra en la gura C.
Figura A.1: Sistema de Distribucin de Ejemplo.
A.1. Matriz BIBC
Utilizando la Ley de Corriente de Kirchho (KLC) se pueden obtener las corrientes de
rama en funcin de las corrientes inyectadas. Para este caso,
B
1
= I
2
+I
3
+I
4
+I
5
+I
6
(A.1)
71
72
B
2
= I
3
+I
4
+I
5
+I
6
B
3
= I
4
+I
5
B
4
= I
5
B
5
= I
6
De esta forma, la relacin entre las corrientes de rama y las corrientes de inyectadas se
puede escribir de forma matricial como se muestra en la ecuacin A.2.
B =

B
1
B
2
B
3
B
4
B
5

1 1 1 1 1
0 1 1 1 1
0 0 1 1 0
0 0 0 1 0
0 0 0 0 1

I
2
I
3
I
4
I
5
I
6

(A.2)
La ecuacin A.2 puede ser reescrita en forma general como se muestra en la ecuacin A.3
[B] = [BIBC][I] (A.3)
BIBC es una matriz triangular superior que slo contiene 1 0. Adems, la matriz T del
TRX se calcula de igual forma que la matriz BIBC del FDCT.
A.2. Matriz BCBV
Utilizando la Ley de Voltajes de Kirchho (KLV) se pueden obtener los voltajes nodales
en funcin de las corrientes de rama. Para este caso,
V
2
= V
1
B
1
Z
12
(A.4)
V
3
= V
1
B
1
Z
12
B
2
Z
23
73
V
4
= V
1
B
1
Z
12
B
2
Z
23
B
3
Z
34
V
5
= V
1
B
1
Z
12
B
2
Z
23
B
3
Z
34
B
4
Z
45
V
6
= V
1
B
1
Z
12
B
2
Z
23
B
5
Z
36
De esta forma, la relacin entre los voltajes nodales y las corrientes de rama se puede
escribir de forma matricial como se muestra en la ecuacin A.5.
.V =

V
1
V
1
V
1
V
1
V
1

V
2
V
3
V
4
V
5
V
6

Z
12
0 0 0 0
Z
12
Z
23
0 0 0
Z
12
Z
23
Z
34
0 0
Z
12
Z
23
Z
34
Z
45
0
Z
12
Z
23
0 0 Z
36

B
1
B
2
B
3
B
4
B
5

(A.5)
La ecuacin A.5 puede ser reescrita en forma general como se muestra en la ecuacin A.6.
[.V ] = [BCBV ][B] (A.6)
Apndice B
ALGORITMO IMPLEMENTADO TRX
A continuacin se mostrarn las lneas de cdigo de los algoritmos desarrollados a lo largo
del proyecto:
1. Algoritmo de adquisicin de datos.
2. Algoritmo iterativo [20] y algoritmo de anlisis de resultados. .
Los algoritmos mostrados a continuacin fueron desarrollados e implementados utilizan-
do un programa comercial; los mismos estn en formato .MAT. En el captulo 3 se puede
encontrar el diagrama de ujo del programa implementado.
B.1. Algoritmo de Adquisicin de Datos
Algoritmo desarrollado para el ltrado y adquisicin de datos de los archivos .DAT pre-
viamente transformados a archivos .MAT. Adems, arma las matrices TRX y S; las guarda
en un archivo .MAT y las almacena en disco.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%% DECLARACIN DE VARIABLES %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
clear all clc delete CCS.mat path=C:\Documents and
Settings\JR\Escritorio\integracion\; directorio=dir(path);
74
75
directorio(1)=[]; directorio(1)=[]; [T0 T1]=size(directorio); [Nc Tc
Rc]=xlsread(C:\Documents and
Settings\JR\Escritorio\integracion\conductor.xlsx,); zk=0; t=0;
w1=0; Vbase=0;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%% CICLO QUE ABRE LOS ARCHIVOS DE LA CARPETA %%%%%%%%%%%%%%%%
for w=1:T0
tic;
clear N T R nodos0 lineas0
if strcmp(finfo(directorio(w).name),xlsx)
w1=w1+1;
archivo=directorio(w).name;
[N T R]=xlsread(archivo,);
pro=0;
i=0; fc=1; fp=0.85;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%% INICIO DE CILCO QUE BUSCA LOS DATOS %%%%%%%%%%%%
for j=1:length(T)
%%%%% Busco datos de Tension, potencia y factor de potencia del sistema. %%
if strcmp(T(j,1),END/TITLE)
Sbase=cell2mat(R(j+2,1));
Vbase=cell2mat(R(j+2,2));
fp=cell2mat(R(j+2,3));
if fp==0 || fp==1
fp=0.85;
end
end
if Vbase > 100
Vbase=Vbase/1000;
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
76
%%%%%%%%%%%%%%%% SE BUSCAN DATOS DE NODOS DEL SISTEMA %%%%%%%%%%%%%%%%%%%%%
if strcmp(T(j,1),END/PARAMS)
j=j+1;
while strcmp(T(j,1),END/NODES)==0
if cell2mat(R(j,2))==Vbase
i=i+1;
nodos0{i,2}=cell2mat(R(j,1));
end
j=j+1;
end
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%% ASIGNA VALORES Y TIPO A LOS NODOS %%%%%%%%%%%%%%%%%%%%%%%%
if strcmp(T(j,1),END/NODES)
%Encuentra y crea la variable con la barra Slak
slak=T(j+1,1);
Q=size(nodos0,1);
for i0=1:Q
%potencia activa Generada
nodos0{i0,4}=0;
%potencia reactiva Generada
nodos0{i0,5}=0;
nodos0{i0,6}=0;
nodos0{i0,7}=0;
%Le pone numero a los nodos
nodos0{i0,1}=i0;
%Definir Barra Slak
if strcmp(nodos0{i0,2},slak)
nodos0{i0,3}=1;
else
nodos0{i0,3}=3;
77
end
end
end
%Armar la matriz de lineas
i=0;
i2=0;
if strcmp(T(j,1),END/SOURCE)
j=j+1;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%% ENCUENTRA LOS DATOS DE LINEAS Y CARGAS %%%%%%%%%%%%%%%%%%%
while strcmp(T(j,1),END/BRANCH)==0
if strcmp(T(j,3),T)
i2=i2+1;
pro=1;
trans{i2,1}=R(j,1);
trans{i2,2}=R(j,2);
trans{i2,3}=(cell2mat(R(j,7)))*3;
trans{i2,4}=trans{i2,3}*fp*fc;
trans{i2,5}=sqrt((trans{i2,3})^2-(trans{i2,4})^2);
else
i=i+1;
lineas0{i,1}=cell2mat(R(j,1));
lineas0{i,2}=cell2mat(R(j,2));
if strcmp(T(j,3),L)
lineas0{i,3}=cell2mat(R(j,7));
lineas0{i,4}=R(j,6);
lineas0{i,5}=0;
lineas0{i,6}=0;
else
lineas0{i,3}=0;
lineas0{i,4}=0;
78
lineas0{i,5}=0;
end
end
j=j+1;
end
end
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%% ENCUENTRA Y ASIGNA LA CARGA AL NODO ADECUADO %%%%%%%%%%%%%
clear i i0 j slak i2 K=0; if pro==1 [q y]=size(trans);
for i=1:size(nodos0,1)
for j=1:q
if strcmp(nodos0{i,2},trans{j,1})
nodos0{i,6}=trans{j,4}+nodos0{i,6};
nodos0{i,7}=trans{j,5}+nodos0{i,7};
K=K+1;
end
end
end
clear i j
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%% ENCUENTRA CONDUCTORES Y ADJUNTA VALORES DE IMPEDANCIA %%%%%
% Y %
%%%%%%%%%%%%%%%%%% CAPACIDAD AMPERIMETRICA DEL CONDUCTOR. %%%%%%%%%%%%%%%%
[wl rl]=size(lineas0);
for i=1:wl
for j=1:length(Tc)
if lineas0{i,3}~=0
if strcmp(lineas0{i,4},Tc{j,2})
lineas0{i,5}=Nc(j,3);
lineas0{i,6}=Nc(j,4);
79
lineas0{i,7}=(Nc(j,7))/((Sbase)/(Vbase*sqrt(3)));
end
end
end
if lineas0{i,5}==0
lineas0{i,5}=0.18026;
lineas0{i,6}=0.12779;
lineas0{i,7}=324.09/((Sbase)/(Vbase*sqrt(3)));
end
end
clear fc fp i j pro q rl trans u y wl K N Q R T T1
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%% INICIO DEL FORMATEO DE LAS MATRICES BDAT Y LDAT %%%%%%%%%%
nl=size(lineas0,1); %nmero de lneas
nn=size(nodos0,1); %nmero de nodos
for k=1:nn
if cell2mat(nodos0(k,3))==1
bdat0{1,1}=1;
bdat0{1,2}=nodos0(k,2);
bdat0{1,3}=nodos0{k,3};
bdat0{1,4}=(cell2mat(nodos0(k,4)))/Sbase;
bdat0{1,5}=(cell2mat(nodos0(k,5)))/Sbase;
bdat0{1,6}=(cell2mat(nodos0(k,6)))/Sbase;
bdat0{1,7}=(cell2mat(nodos0(k,7)))/Sbase;
bdat0{1,8}=Vbase/Vbase;
end
end
%ciclo que ordena los datos de lnea y construye la matriz bdat0
%con base a la metodologa propuesta por Shirmohammadi[10].
%Esta metodologa es la base para la matriz BIBC del flujo de
%carga implementado.
80
l=2; k=0; for u=1:nn
for g=1:nl
if strcmp(bdat0{u,2},lineas0{g,2})
nlle=lineas0{g,2};
nsal=lineas0{g,1};
lineas0{g,1}=nlle;
lineas0{g,2}=nsal;
end
end
for k=1:nl
if strcmp(bdat0{u,2},lineas0{k,1})
for y=1:nn
if strcmp(lineas0{k,2},nodos0{y,2})
bdat0{l,1}=l;
bdat0{l,2}=nodos0(y,2);
bdat0{l,3}=nodos0{y,3};
bdat0{l,4}=nodos0{y,4}/Sbase;
bdat0{l,5}=(nodos0{y,5})/Sbase;
bdat0{l,6}=(nodos0{y,6})/Sbase;
bdat0{l,7}=(nodos0{y,7})/Sbase;
bdat0{l,8}=Vbase/Vbase;
ldat0{l-1,1}=bdat0{u,1};
ldat0{l-1,2}=bdat0{l,1};
ldat0{l-1,3}=lineas0{k,7};
ldat0{l-1,4}=(lineas0{k,5}*lineas0{k,3})/
(Vbase^2/(Sbase/1000));
ldat0{l-1,5}=(lineas0{k,6}*lineas0{k,3})/
(Vbase^2/(Sbase/1000));
if ldat0{l-1,4}==0
ldat0{l-1,3}=0;
end
81
lineas0{k,1}=0;
lineas0{k,2}=0;
l=l+1;
end
end
end
end
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%% CONVIERTE LAS CELDAS EN MATRICES %%%%%%%%%%%%%%%%%%%%%%%%%
clear g k l nl nlle nn nsal u y nll ns bdat ldat
bdat(:,1)=cell2mat(bdat0(:,1)); bdat(:,3)=cell2mat(bdat0(:,3));
bdat(:,4)=cell2mat(bdat0(:,4)); bdat(:,5)=cell2mat(bdat0(:,5));
bdat(:,6)=cell2mat(bdat0(:,6)); bdat(:,7)=cell2mat(bdat0(:,7));
bdat(:,8)=cell2mat(bdat0(:,8)); ldat(:,1)=cell2mat(ldat0(:,1));
ldat(:,2)=cell2mat(ldat0(:,2)); ldat(:,3)=cell2mat(ldat0(:,3));
ldat(:,4)=cell2mat(ldat0(:,4)); ldat(:,5)=cell2mat(ldat0(:,5));
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%% SE INICIA EL ARMADO DE LA MATRIZ "T" Y LA "DLF" %%%%%%%%%%
m=size(ldat,1); %nmero de lneas
n=size(bdat,1); %nmero de nodos
T1(m,n)=0; % matriz previa a T
V0((2*n)-2,1)=0; %inicializar vector Voltaje con "0"
Vectx=ldat(:,5); %Definimos el vector de X de las ramas.
Vectr=ldat(:,4); %Definimos el vector de R de las ramas.
for k=1:m %ciclo que crea la matriz T1
ns=ldat(k,1); % nodo de salida
nll=ldat(k,2); % nodo de llegada
B(:,1)=T1(:,ns);
B(nll-1,1)=1;
T1(:,k+1)=B(:,1);
82
end
for k=2:n %ciclo que crea la matriz T
T(:,k-1)=T1(:,k);
end
TR=T*diag(Vectr)*T; %Sub-matriz de Rs de la matriz TRX
TX=T*diag(Vectx)*T; %Sub-matriz de xs de la matriz TRX
TRX=vertcat(horzcat(TR,-TX),horzcat(TX,TR)); %Matriz TRX Matriz real.
DLF=TRX; S(:,1)=bdat(:,6); S(:,2)=bdat(:,7); for ll=1:size(bdat0,1)
nno(ll,1)=bdat0{ll,2}; end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%% SE CREA LA VARIABLE "RED" CON LOS DATOS DEL SISTEMA %%%%%%%%%%
RED{w1,1}=(directorio(w).name);
RED{w1,2}=DLF;
RED{w1,3}=S;
RED{w1,4}=nno;
RED{w1,5}=T;
RED{w1,6}=ldat;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
else
zk=zk+1;
malos{zk,1}=directorio(w).name;
end
clear nno B BCBV BCBVI BIBC BIBCI DLF I S V0 bdat k m n nll ns z
ll nodos0 lineas0 ldat0 bdat0 t=toc+t;
end
end save CCS RED clear all
Load_flow % SE LLAMA AL MODULO QUE EJECUTA EL FLUJO DE CARGA%
83
B.2. Flujo de Carga TRX
Cdigo desarrollado para el algoritmo de ujo de carga implementado. El mismo coloca
las matrices TRX y S en memoria RAM y realiza el proceso iterativo descrito en la seccin
2.1.2 del captulo 2. Luego, almacena las tensiones en archivos .MAT. Con los resultados de
tensin se calculan las prdidas de potencia en los circuitos, las capacidades amperimtricas
de los conductores y las cadas de tensin; todos los resultados son almacenados en el mismo
archivo .MAT.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%% DECLARACIN DE VARIABLES %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
clear all clc
load CCS % SE CARGA LA MATRIZ DE DATOS DEL SISTEMA A LA RAM
iter=10; %definir nmero mximo de iteraciones
con=0.0001; %definir criterio de convergencia
t=0; t1=0; Y=size(RED,1); ntot=0;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%% CICLO QUE INICIA Y CARGA CADA CIRCUITO %%%%%%%%%%%%%%%%%%%
for qp=1:Y
iter2=0;
TRX=RED{qp,2};
S=RED{qp,3};
n=size(S,1);
m=n-1;
V0((2*n)-2,1)=0;
T=RED{qp,5};
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%% CICLO QUE CREA LOS VECTORES DE "S" E "I" INICIALES %%%%%%%%%%
for k=1:n-1
V0(k,1)=1;
I(k,1)=((-S(k+1,1)*V0(k,1))-(S(k+1,2)*V0(k+(n-1),1)))/(((V0(k,1))^2+
84
(V0(k+(n-1),1))^2));
I(k+(n-1),1)=(-S(k+1,2)*V0(k,1)+S(k+1,1)*V0(k+(n-1),1))/(((V0(k,1))^2+
(V0(k+(n-1),1))^2));
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%% OPERACIONES PARA INICIALIZAR VECTORES DE ITERACION %%%%%%%%%%
DV=TRX*I; %Multiplicacin inicial para hallar el delta de tensin inicial
%para corregir Vo.
V1=V0-DV; %V1 coregida, V0-DV
conve=1; % inicializacion de variabel para convergencia.
Vp(1,1)=1; V(1,1)=1; ang(1,1)=0; DV2=DV; k=0;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%% CICLO ITERATIVO DEL FLUJO DE CARGA %%%%%%%%%%%%%%%%%%
tic; % INICIA EL CONTEO DE TIEMPO
while conve>con & iter>iter2
iter2=iter2+1; %Contador de Iteraciones.
Vk=V1;%Definimos una variable Vk para actualizar los voltajes corregidos.
for k=1:n-1 %Proceso para corregir las corrientes del sistema.
I(k,1)=((S(k+1,1)*Vk(k,1))-(S(k+1,2)*Vk(k+(n-1),1)))/(((Vk(k,1))^2+
(Vk(k+(n-1),1))^2));
I(k+(n-1),1)=(-S(k+1,2)*Vk(k,1)+S(k+1,1)*Vk(k+(n-1),1))/(((Vk(k,1))^2+
(Vk(k+(n-1),1))^2));
end
DV=TRX*I; %corrige el delta de tensin para la prxima iteracin.
V1=V0-DV; %corrige el voltaje final antes de comprobar la convergencia.
conve=max(abs(DV2-DV)); %Define la convergencia midiendo la variacin
%de DV
DV2=DV;
end toc; t=t+toc; tic;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%% CICLO QUE CONVIERTE A "I" EN IMAGINARIO %%%%%%%%%%%%%%%%%%
85
ni=(size(I,1))/2; for L=1:ni
Ii(L,1)=I(L,1);
Ii(L,1)=Ii(L,1)+I(L+ni,1)*i;
end Ip=abs(T*Ii);
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%% CICLO QUE ORDENA LOS VOLTAJES Y ANGULOS %%%%%%%%%%%%%%%
for M=2:n
V(M,1)=abs((V1(M-1+(n-1),1))+i*(V1(M-1,1)));
ang(M,1)=rad2deg(angle((V1(M-1+(n-1 ),1))*i+(V1(M-1,1))));
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%% GUARDA RESULTADOS EN LA VARIABLE "RED" %%%%%%%%%%%%%%%%%
RED{qp,6}(:,6)=Ip; RED{qp,7}(:,1)=V; RED{qp,7}(:,2)=ang(:,1);
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%% CALCULO DE PERDIDAS POR CIRCUITOS, CAIDAS EN LAS RAMAS%%%%%%%%%%
K=size(RED{qp,6},1); for J=1:K
RED{qp,6}(J,7)=(RED{qp,6}(J,6))^2*(RED{qp,6}(J,4));
RED{qp,6}(J,8)=(RED{qp,6}(J,6))^2*(RED{qp,6}(J,5));
end J=0;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%% CALCULO DE VARIACION DE TENSION EN CADA UNO DE LOS NODOS%%%%%%%%
for J=2:n
RED{qp,7}(J,3)=((RED{qp,7}(1,1)-RED{qp,7}(J,1))/(RED{qp,7}(1,1)))*100;
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%% CALCULO DE NODOS POR CIRCUITOS Y TOTALES DE LA RED%%%%%%%%%%%%
K2=size(RED{qp,7},1);
%Calculamos el nmero de nodos totales del sistema.
npc=size(RED{qp,4},1); RED{qp,8}=npc; ntot=ntot+npc;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
86
clear DV DV2 I M S TRX V V0 V1 Vk Vp ang conve k m n K J npc ni Ii L
Ip T toc; t1=t1+toc; end ntot t t1 save CCS2 RED clear all
Apndice C
EJEMPLO DE REPORTE DEL ALGORITMO DESARROLLADO
NOMBRE DEL CIRCUITO: GRA_A01.xlsx
VOLTAJES EN LOS NODOS
***************************************************************
# Nodo: Nodo: Tensin: ngulo: Cada:
1 5201_1 1.000000 0.000000 0.000000
2 5201_2 0.996748 -0.011527 0.325188
3 5201_70 1.000000 0.000000 0.000000
4 5201_43 0.994400 -0.019907 0.559986
5 5201_3 0.993386 -0.023515 0.661434
6 5201_89 0.996708 -0.010766 0.329241
7 5201_71 1.000000 0.000000 0.000000
8 5201_51 0.993381 -0.023554 0.661860
9 5201_44 0.994213 -0.020578 0.578715
10 5201_5 0.993386 -0.023515 0.661434
11 5201_90 0.996708 -0.010766 0.329241
12 5201_55 0.993381 -0.023554 0.661860
13 5201_52 0.993381 -0.023554 0.661860
14 5201_45 0.994213 -0.020578 0.578715
15 5201_68 0.994213 -0.020578 0.578715
87
88
16 5201_46 0.994213 -0.020578 0.578715
17 5201_7 0.990692 -0.033176 0.930784
18 5201_54 0.993381 -0.023554 0.661860
19 5201_53 0.993381 -0.023554 0.661860
20 5201_52 0.993381 -0.023554 0.661860
21 5201_53 0.993381 -0.023554 0.661860
22 5201_47 0.994213 -0.020578 0.578715
23 5201_49 0.994213 -0.020578 0.578715
24 5202_133 0.994213 -0.020578 0.578715
25 5201_8 0.990648 -0.032334 0.935247
26 5201_10 0.989429 -0.037725 1.057105
27 5201_53 0.993381 -0.023554 0.661860
28 5201_56 0.992600 -0.026356 0.740000
29 5201_48 0.994190 -0.020377 0.580982
30 5201_50 0.994207 -0.020526 0.579297
31 5201_9 0.990648 -0.032334 0.935247
32 5201_13 0.989429 -0.037725 1.057105
33 5201_57 0.992581 -0.025996 0.741908
34 5201_59 0.991759 -0.029377 0.824109
35 5201_11 0.989429 -0.037725 1.057105
36 5201_12 0.989429 -0.037725 1.057105
37 5201_58 0.992581 -0.025996 0.741908
38 5201_60 0.991530 -0.025054 0.847034
39 5201_62 0.991054 -0.031910 0.894558
40 5201_80 0.989352 -0.037042 1.064767
41 5201_19 0.988295 -0.041816 1.170520
42 5201_61 0.991530 -0.025054 0.847034
43 5201_63 0.991039 -0.031965 0.896062
44 5201_64 0.991054 -0.031910 0.894558
45 5201_14 0.989250 -0.036132 1.074988
46 5201_81 0.989144 -0.035187 1.085595
89
47 5201_20 0.988290 -0.041732 1.170967
48 5201_67 0.988237 -0.040723 1.176300
49 5201_23 0.987049 -0.046320 1.295086
50 5201_6 0.990475 -0.033996 0.952472
51 5202_49 0.991039 -0.031965 0.896062
52 5202_47 0.991039 -0.031965 0.896062
53 5201_65 0.991054 -0.031910 0.894558
54 5201_15 0.989249 -0.036102 1.075148
55 5201_17 0.989152 -0.035257 1.084806
56 5201_82 0.989144 -0.035187 1.085595
57 5201_21 0.988290 -0.041732 1.170967
58 5201_22 0.988237 -0.040723 1.176300
59 5201_24 0.987049 -0.046320 1.295086
60 5201_86 0.990454 -0.033595 0.954597
61 5201_25 0.990475 -0.033996 0.952472
62 5201_16 0.989249 -0.036102 1.075148
63 5201_18 0.989152 -0.035257 1.084806
64 5201_84 0.989144 -0.035187 1.085595
65 5201_83 0.989144 -0.035187 1.085595
66 5201_27 0.986254 -0.049202 1.374616
67 5201_87 0.990454 -0.033595 0.954597
68 5201_85 0.990475 -0.033996 0.952472
69 5201_28 0.986203 -0.048231 1.379741
70 5201_30 0.985663 -0.051343 1.433652
71 5201_88 0.990454 -0.033595 0.954597
72 5201_29 0.986203 -0.048231 1.379741
73 5201_32 0.985527 -0.050125 1.447291
74 5201_72 0.985575 -0.051665 1.442519
75 5201_34 0.985365 -0.052426 1.463477
76 5201_33 0.985527 -0.050125 1.447291
77 5201_69 0.985575 -0.051665 1.442519
90
78 5201_73 0.985569 -0.051562 1.443064
79 5201_35 0.985365 -0.052426 1.463477
80 5201_31 0.985575 -0.051665 1.442519
81 5201_74 0.985569 -0.051562 1.443064
82 5201_36 0.985365 -0.052426 1.463477
83 5201_37 0.985349 -0.052485 1.465098
84 5201_38 0.985342 -0.052511 1.465818
85 5201_75 0.985019 -0.053683 1.498064
86 5201_39 0.985342 -0.052511 1.465818
87 5201_76 0.984992 -0.053441 1.500764
88 5201_40 0.984910 -0.054080 1.508981
89 5201_77 0.984992 -0.053441 1.500764
90 5201_41 0.984856 -0.053045 1.514433
91 5201_79 0.984992 -0.053441 1.500764
92 5201_78 0.984992 -0.053441 1.500764
93 5201_42 0.984856 -0.053045 1.514433
PERDIDAS EN LAS RAMAS
***************************************************************
# Nodo sal # Nodo lle: Perdida P: Perdida Q:
1 2 2.770122e-004 1.963796e-004
1 3 0.000000e+000 0.000000e+000
2 4 7.354838e-005 5.213995e-005
2 5 1.728360e-004 1.225270e-004
2 6 1.247080e-007 3.039511e-008
3 7 0.000000e+000 0.000000e+000
4 8 2.143848e-005 1.519818e-005
4 9 1.925370e-006 1.364934e-006
5 10 0.000000e+000 0.000000e+000
91
6 11 0.000000e+000 0.000000e+000
8 12 0.000000e+000 0.000000e+000
8 13 0.000000e+000 0.000000e+000
9 14 0.000000e+000 0.000000e+000
9 15 0.000000e+000 0.000000e+000
9 16 0.000000e+000 0.000000e+000
10 17 1.384522e-004 9.815158e-005
12 18 0.000000e+000 0.000000e+000
12 19 0.000000e+000 0.000000e+000
12 20 0.000000e+000 0.000000e+000
13 21 0.000000e+000 0.000000e+000
16 22 0.000000e+000 0.000000e+000
16 23 0.000000e+000 0.000000e+000
16 24 0.000000e+000 0.000000e+000
17 25 2.302853e-007 5.612751e-008
17 26 5.971812e-005 4.233540e-005
18 27 0.000000e+000 0.000000e+000
19 28 1.644385e-005 1.165738e-005
22 29 1.593068e-007 6.769693e-008
23 30 2.723222e-008 1.157225e-008
25 31 0.000000e+000 0.000000e+000
26 32 0.000000e+000 0.000000e+000
28 33 5.897191e-008 1.437324e-008
28 34 1.562133e-005 1.107428e-005
32 35 0.000000e+000 0.000000e+000
32 36 0.000000e+000 0.000000e+000
33 37 0.000000e+000 0.000000e+000
34 38 1.181952e-006 2.880776e-007
34 39 1.017943e-005 7.216409e-006
35 40 1.244838e-006 5.289901e-007
36 41 3.744148e-005 2.654303e-005
92
38 42 0.000000e+000 0.000000e+000
39 43 2.173545e-007 1.540870e-007
39 44 0.000000e+000 0.000000e+000
40 45 7.942518e-007 3.375149e-007
40 46 1.765661e-006 7.503123e-007
41 47 6.943301e-009 1.692293e-009
41 48 3.005762e-007 7.325951e-008
41 49 3.439619e-005 2.438416e-005
43 50 8.150792e-006 5.778263e-006
43 51 0.000000e+000 0.000000e+000
43 52 0.000000e+000 0.000000e+000
44 53 0.000000e+000 0.000000e+000
45 54 1.237258e-009 3.015572e-010
45 55 6.935722e-007 2.947314e-007
46 56 0.000000e+000 0.000000e+000
47 57 0.000000e+000 0.000000e+000
48 58 0.000000e+000 0.000000e+000
49 59 0.000000e+000 0.000000e+000
50 60 2.194010e-007 5.347467e-008
50 61 0.000000e+000 0.000000e+000
54 62 0.000000e+000 0.000000e+000
55 63 0.000000e+000 0.000000e+000
56 64 0.000000e+000 0.000000e+000
56 65 0.000000e+000 0.000000e+000
59 66 2.196064e-005 1.556835e-005
60 67 0.000000e+000 0.000000e+000
61 68 0.000000e+000 0.000000e+000
66 69 2.656918e-007 6.475715e-008
66 70 1.385303e-005 9.820693e-006
67 71 0.000000e+000 0.000000e+000
69 72 0.000000e+000 0.000000e+000
93
70 73 6.461058e-007 2.745607e-007
70 74 5.520172e-008 3.913363e-008
70 75 5.572751e-006 3.950637e-006
73 76 0.000000e+000 0.000000e+000
74 77 0.000000e+000 0.000000e+000
74 78 4.240176e-009 1.033459e-009
75 79 0.000000e+000 0.000000e+000
77 80 0.000000e+000 0.000000e+000
78 81 0.000000e+000 0.000000e+000
79 82 0.000000e+000 0.000000e+000
82 83 3.028669e-007 2.147085e-007
83 84 5.979455e-008 4.238958e-008
83 85 3.422693e-006 2.426417e-006
84 86 0.000000e+000 0.000000e+000
85 87 1.277531e-007 5.428831e-008
85 88 6.801111e-007 4.821447e-007
87 89 0.000000e+000 0.000000e+000
88 90 4.246517e-007 1.035005e-007
89 91 0.000000e+000 0.000000e+000
89 92 0.000000e+000 0.000000e+000
90 93 0.000000e+000 0.000000e+000
CAPACIDADES AMPERIMETRICAS
***************************************************************
#Nodo sal #Nodo lle: I(max): CAP(67) CAP (max)
1 2 1.042209e-001 4.689942e-002 6.999914e-002
1 3 0.000000e+000 4.689942e-002 6.999914e-002
2 4 3.832317e-002 4.689942e-002 6.999914e-002
2 5 6.288734e-002 4.689942e-002 6.999914e-002
94
2 6 3.010414e-003 1.852592e-002 2.765062e-002
3 7 0.000000e+000 0.000000e+000 0.000000e+000
4 8 2.574611e-002 4.689942e-002 6.999914e-002
4 9 1.257706e-002 4.689942e-002 6.999914e-002
5 10 6.288734e-002 0.000000e+000 0.000000e+000
6 11 3.010414e-003 0.000000e+000 0.000000e+000
8 12 2.574611e-002 0.000000e+000 0.000000e+000
8 13 0.000000e+000 0.000000e+000 0.000000e+000
9 14 0.000000e+000 2.692495e-002 4.018649e-002
9 15 0.000000e+000 4.689942e-002 6.999914e-002
9 16 1.257706e-002 0.000000e+000 0.000000e+000
10 17 6.288734e-002 4.689942e-002 6.999914e-002
12 18 0.000000e+000 0.000000e+000 0.000000e+000
12 19 2.574611e-002 0.000000e+000 0.000000e+000
12 20 0.000000e+000 0.000000e+000 0.000000e+000
13 21 0.000000e+000 0.000000e+000 0.000000e+000
16 22 7.546220e-003 0.000000e+000 0.000000e+000
16 23 5.030841e-003 0.000000e+000 0.000000e+000
16 24 0.000000e+000 0.000000e+000 0.000000e+000
17 25 5.049839e-003 1.852592e-002 2.765062e-002
17 26 5.783750e-002 4.689942e-002 6.999914e-002
18 27 0.000000e+000 0.000000e+000 0.000000e+000
19 28 2.574611e-002 4.689942e-002 6.999914e-002
22 29 7.546220e-003 2.692495e-002 4.018649e-002
23 30 5.030841e-003 2.692495e-002 4.018649e-002
25 31 5.049839e-003 0.000000e+000 0.000000e+000
26 32 5.783750e-002 0.000000e+000 0.000000e+000
28 33 3.023646e-003 1.852592e-002 2.765062e-002
28 34 2.272247e-002 4.689942e-002 6.999914e-002
32 35 1.744872e-002 0.000000e+000 0.000000e+000
32 36 4.038877e-002 0.000000e+000 0.000000e+000
95
33 37 3.023646e-003 0.000000e+000 0.000000e+000
34 38 5.044779e-003 1.852592e-002 2.765062e-002
34 39 1.767769e-002 4.689942e-002 6.999914e-002
35 40 1.744872e-002 2.692495e-002 4.018649e-002
36 41 4.038877e-002 4.689942e-002 6.999914e-002
38 42 5.044779e-003 0.000000e+000 0.000000e+000
39 43 1.767769e-002 4.689942e-002 6.999914e-002
39 44 0.000000e+000 2.692495e-002 4.018649e-002
40 45 8.344978e-003 2.692495e-002 4.018649e-002
40 46 9.103745e-003 2.692495e-002 4.018649e-002
41 47 1.518756e-003 1.852592e-002 2.765062e-002
41 48 5.088027e-003 1.852592e-002 2.765062e-002
41 49 3.378199e-002 4.689942e-002 6.999914e-002
43 50 1.767769e-002 4.689942e-002 6.999914e-002
43 51 0.000000e+000 0.000000e+000 0.000000e+000
43 52 0.000000e+000 0.000000e+000 0.000000e+000
44 53 0.000000e+000 0.000000e+000 0.000000e+000
45 54 7.585761e-004 1.852592e-002 2.765062e-002
45 55 7.586402e-003 2.692495e-002 4.018649e-002
46 56 9.103745e-003 0.000000e+000 0.000000e+000
47 57 0.000000e+000 0.000000e+000 0.000000e+000
48 58 0.000000e+000 0.000000e+000 0.000000e+000
49 59 3.378199e-002 0.000000e+000 0.000000e+000
50 60 1.010156e-002 1.852592e-002 2.765062e-002
50 61 7.576129e-003 0.000000e+000 0.000000e+000
54 62 0.000000e+000 0.000000e+000 0.000000e+000
55 63 7.586402e-003 0.000000e+000 0.000000e+000
56 64 0.000000e+000 0.000000e+000 0.000000e+000
56 65 0.000000e+000 0.000000e+000 0.000000e+000
59 66 3.378199e-002 4.689942e-002 6.999914e-002
60 67 1.010156e-002 0.000000e+000 0.000000e+000
96
61 68 0.000000e+000 0.000000e+000 0.000000e+000
66 69 5.073849e-003 1.852592e-002 2.765062e-002
66 70 2.870814e-002 4.689942e-002 6.999914e-002
67 71 0.000000e+000 0.000000e+000 0.000000e+000
69 72 5.073849e-003 0.000000e+000 0.000000e+000
70 73 5.087528e-003 2.692495e-002 4.018649e-002
70 74 7.615905e-004 4.689942e-002 6.999914e-002
70 75 2.285902e-002 4.689942e-002 6.999914e-002
73 76 0.000000e+000 0.000000e+000 0.000000e+000
74 77 0.000000e+000 4.689942e-002 6.999914e-002
74 78 7.615905e-004 1.852592e-002 2.765062e-002
75 79 2.285902e-002 0.000000e+000 0.000000e+000
77 80 0.000000e+000 2.692495e-002 4.018649e-002
78 81 0.000000e+000 0.000000e+000 0.000000e+000
79 82 2.285902e-002 0.000000e+000 0.000000e+000
82 83 2.285902e-002 4.689942e-002 6.999914e-002
83 84 1.015693e-002 4.689942e-002 6.999914e-002
83 85 1.270209e-002 4.689942e-002 6.999914e-002
84 86 1.015693e-002 0.000000e+000 0.000000e+000
85 87 5.080493e-003 2.692495e-002 4.018649e-002
85 88 7.621598e-003 4.689942e-002 6.999914e-002
87 89 5.080493e-003 0.000000e+000 0.000000e+000
88 90 7.621598e-003 1.852592e-002 2.765062e-002
89 91 0.000000e+000 0.000000e+000 0.000000e+000
89 92 0.000000e+000 0.000000e+000 0.000000e+000
90 93 7.621598e-003 0.000000e+000 0.000000e+000
TOTALES DEL CIRCUITO
***************************************************************
Perdidas activas totales (pu): 9.215657e-004
97
Perdidas reactivas totales (pu): 6.504554e-004
Cantidad de conductores violan el 67% capacidad: 28
Cantidad de conductores violan el 100% capacidad: 25
Total de nodos en este circuito: 93
Apndice D
CASO DE ESTUDIO: CIRCUITO DE 4 BARRAS
El sistema incluye 1 generador y 3 cargas equivalentes, conectadas a una red de distribu-
cin radial. Los datos de ramas y nodos del circuito se muestran en las tablas D.1 y D.2,
respectivamente.
Datos de lnea
El circuito est conformado por 3 lneas, descritas en la tabla D.1.
Tabla D.1: Datos de lnea (Ldat)
salida llegada r(p.u) x(p.u)
1 2 0,0265 0,0462
2 3 0,1005 0,0693
3 4 0,0670 0,0462
Datos de nodo
Los datos de los 4 nodos se encuentran en la tabla D.2.
98
99
Tabla D.2: Datos de nodos (Bdat)
N Tipo Pgen
0
(p.u) Qgen
0
(p.u) P
load
(p.u) Q
load
(p.u) V
0
fcap
1 1 0 0 0 0 1 1
2 3 0 0 0,1017 0,0635 1 1
3 3 0 0 0,0547 0,0342 1 1
4 3 0 0 0,0809 0,0596 1 1
Donde,
N. Es el nmero de la barra.
Tipo. Corresponde al tipo de barra: 1 barra slack, 2 barra PV y 3 barra PQ.
Pgen
0
. Generacin de potencia activa (valor inicial)(p.u).
Qgen
0
.Generacin de potencia reactiva (valor inicial)(p.u).
P
load
. Potencia activa consumida (p.u).
Q
load
. Potencia activa consumida (p.u).
V
0
. Tensin inicial de la barra (p.u).
fcap. Factor de capacidad de la barra.
Apndice E
REPORTE DEL MATPOWER PARA EL CASO DE 7 BARRAS
runpf(MATPOWERexample)
Converged in 0.01 seconds
================================================================================
| System Summary
================================================================================
How many? How much? P (MW) Q(MVAr)
------------- ------------------- ----------- ------------
Buses 7 Total Gen Capacity 250.0 -300.0 to 300.0
Generators 1 On-line Capacity 250.0 -300.0 to 300.0
Committed Gens 1 Generation (actual) 0.5 0.3
Loads 6 Load 0.5 0.3
Fixed 6 Fixed 0.5 0.3
Dispatchable 0 Dispatchable -0.0 of -0.0 -0.0
Shunts 0 Shunt (inj) -0.0 0.0
Branches 6 Losses (I^2*Z) 0.03 0.02
Transformers 0 Branch Charging(inj) - 0.0
Inter-ties 0 Total Inter-tie Flow 0.0 0.0
100
101
Areas 1
Minimum Maximum
------------------------- --------------------------------
Voltage Magnitude 0.917079 p.u. @ bus 7 1.000 p.u. @ bus 1
Voltage Angle -0.49 deg @ bus 6 0.00 deg @ bus 1
P Losses (I^2*R) - 0.02 MW @ line 1-3
Q Losses (I^2*X) - 0.02 MVAr @ line 1-3
================================================================================
| Bus Data
================================================================================
Bus Voltage Generation Load
# Mag(pu) Ang(deg) P (MW) Q (MVAr) P (MW) Q (MVAr)
----- ------- -------- -------- -------- -------- --------
1 1.000000 0.000000 0.50 0.34 - -
2 0.994335 -0.173777 - - 0.10 0.06
3 0.940742 -0.020595 - - 0.05 0.03
4 0.931971 -0.003891 - - 0.08 0.06
5 0.925992 -0.469903 - - 0.10 0.06
6 0.917416 -0.493752 - - 0.05 0.03
7 0.917079 -0.452656 - - 0.08 0.06
-------- -------- -------- --------
Total: 0.50 0.34 0.47 0.31
102
================================================================================
| Branch Data
================================================================================
Brnch From To From Bus Injection To Bus Injection Loss (I^2*Z)
# Bus Bus P (MW) Q (MVAr) P (MW) Q (MVAr) P (MW) Q (MVAr)
---- ---- ---- ------- ------ ------- -------- ------ -------
1 1 2 0.10 0.06 -0.10 -0.06 0.000 0.00
2 1 3 0.40 0.27 -0.38 -0.26 0.024 0.02
3 3 4 0.08 0.06 -0.08 -0.06 0.001 0.00
4 3 5 0.24 0.16 -0.24 -0.16 0.003 0.00
5 5 6 0.06 0.03 -0.05 -0.03 0.000 0.00
6 5 7 0.08 0.06 -0.08 -0.06 0.001 0.00
-------- --------
Total: 0.029 0.02
Apndice F
A COMPENSATION-BASED POWER FLOW METHOD FOR WEAKLY
MESHED DISTRIBUTION AND TRANSMISSION NETWORKS,
SHIRMOHAMMADI ET AL. 1988, [11]
103
753 IEEE Transacti ons on Power Systems, Vol. 3, No. 2, May 1988
A COMPENSATI ON- BASEDPOWER FLOWMETHdD
FOR WEAKLY MESHED DI STRI BUTI ONAND TRANSMI SSI ON NETWORKS
D. Shi rmoharmnadi H. W. Hong
Member Seni or Member
Syst ems Engi neeri ng Group
Paci f i c Gas and El ect ri c Company
San Franci sco, Cal i f orni a
ABSTRACT
Thi s paper descri bes a newpower f l ow met hod f or
sol vi ng weakl y meshed di st ri but i on and t ransmi ssi on
net works, usi ng a mul t i - por t compensat i on t echni que and
basi c f or mul at i ons of Ki rchhof f ' s l aws. Thi S met hod
has excel l ent conver gence charact eri st i cs and i s very
robust . A comput er program i mpl ement i ng t hi s power
f l owsol ut i on scheme was devel oped and successf ul l y
appl i ed t o sever al pract i cal di st ri but i on net wdrks wi t h
r adi al and weakl y meshed st ruct ure. Thi s program was
al so successf ul l y used f or sol vi ng r adi al and weakl y
meshed t ransmi ssi oh net works. The met hod can be
appl i ed t b t he sol ut i on of bot h t he t hr ee- phase
( unbal anced) and si ngl e- phase ( bal anced) represent at i on
of t he net bork. In t hi s paper , however , onl y t he
si ngl e phase represent at i on i s t reat ed i n det ai l .
I . I NTRODUCTI ON
Recent l y. at t he Paci f i c, Gas and El ect ri c Company
we devel oped a di st ki but i on net work opt i mi zat i on
sof t war e package. Thi s devel opment work cal l ed f or a
power f l ew sol ut i on al gori t hm wi t h t he f ol l owi ng
gener al charact eri st i cs:
1. Capabl e of Sol vi ng r adi al and weakl y meshed
di st ri but at i on net wor ks wi t h up t o several
t housand l i ne sect i ons ( branches) and nodes
( buses ) .
2. Robust and ef f i ci ent .
The ef f i ci ency of such a power f l owal gori t hmi s
of utni ost i mport ance as each opt i mi zat i on st udy
r equi r es numer ous power f l owruns. Fur t her mor e, t he
ext ensi on of t he appl i cat i on of t hi s power f l ow met hod
t o t hree phase net wor ks wi t h di st ri but ed l oads was al so
envi saged.
The Newt on Raphson and t he f ast decoupl ed power
f l ow sol ut i on t echni ques and a host of t hei r
deri vat i ves have ef f i ci ent l y sol ved "wel l - behaved"
power syst ems f or mor e t han t wo decades. Resear cher s,
however, have been awar e of t he shor t comi ngs of t hese
sol ut i on al gori t hms when t hey ar e "generi cal l y"
i mpl ement ed and appl i ed t o i l l - condi t i oned andl or
poorl y i ni t i al i bed power syst ems [ 1, 2, 3] . Hence,
conrneri cal power f l ow packages al ways modi f y t hese
This pap& was sponsoredby the IEEE Power Engineering Soci ety
for presentati onat theIEEE Power I ndustry Comput er Appl i ca-
tion Conf erence, Montreal , Canada, May 18-21,1987. Manuscri pt
was publ i shedinthe 1987 PICA Conference Record.
A. Seml yen G. X. Luo
Seni or Member
Depart ment of El ect r i cal Engi neeri ng
Uni versi t y of Tor ont o
Tor ont o, Ont ar i o, Canada
al gori t hms f or enhanced robust ness. The nat ur e of
modi f i cat i ons and t he degree of i mprovement obt ai ned
vari es f or di f f erent packages. The Gauss- Sei del power
f l owt echni que, anot her cl assi ci al power f l ow Et hod,
al t hough very r obust , has shown t o be ext remel y
i nef f i ci ent i n sol vi ng l arge power syst ems.
Di st r i but i on net works, due t o t hei r wi de rangi ng
r esi st ence and r eact ance val ues and r adi al st r uct ur e,
f al l i nt o t he cat egory of i l l - condi t i oned power syst ems
f or t he generi c Newt on- Raphson and f ast decoupl ed power
f l owal gori t hms. Our experi ence wi t h a basi c Newt on-
Raphson power f l ow program f or sol vi ng di st ri but i on
net works was most l y unsuccessf ul as i t di verged f or t he
maj ori t y of t he net works st udi ed. Lat er we
successf ul l y used t he Newt on- Raphson based West ern
Syst em Coordi nat i ng Counci l ( WSCC) power f l owprogram.
Thi s pr ogr am, whi ch i s commonl y used by t he WSCC member
ut i l i t i es, i ncl udes sever al enhancement s f or
i ncreasi ng i t s convergence capabi l i t i es. Al t hough t he
r obust ness of t he program was accept abl e, t he
comput at i on t i me was excessi ve. I n addi t i on, t he
ext ensi on of t he Newt on- Raphson al gori t hm t o t he
sol ut i on of t he t hree phase net wor ks, not even
consi deri ng di st ri but ed l oad, woul d resul t i n
subst ant i al det eri orat i on of t he numer i cal ef f i ci ency
of t he sol ut i on al gori t hm[ 41.
Ef f i ci ent power f l owal gori t hms f or sol vi ng si ngl e
and t hree phase r adi al di st ri but i on net works [ 5,6,71
have been ext ensi vel y used by di st ri but i on engi neers.
These al gori t hms ar e not , however , desi gned t o sol ve
meshed net works.
I n t hi s paper , we propose a new met hod f or t he
sol ut i on of weakl y meshed net works. I n t hi s met hod, we
f i rst break t he i nt erconnect ed gri d at a number of
poi nt s ( breakpoi nt s) i n order t o convert i t i nt o one
radi al net work. Each breakpoi nt wi l l open one si mpl e
l oop. The r adi al net work i s sol ved ef f i ci ent l y by t he
di rect appl i cat i on of Ki rchhof f ' s vol t age and current
l aws ( KVL and KCL). We t hen account f or t he f l ows at
t he breakpoi nt s by i nj ect i ng current s at t hei r t wo end
nodes. The breakpoi nt current s ar e cal cul at ed usi ng
t he mul t i - por t compensat i on met hod [ 8, 91. I n presence
of const ant P, Q l oads, t he net work i s nonl i near causi ng
t he compensat i on process t o become i t erat i ve. The
sol ut i on of t he r adi al net work wi t h t he addi t i onal
breakpoi nt current i nj ect i ons compl et es t he sol ut i on of
t he weakl y meshed net work.
Our st udi es have shown t hat , t ypi cal l y, onl y a f ew
i t erat i ons wer e requi red f or t he sol ut i on of
di st ri but i on net wor ks usi ng t he proposed power f l ow
sol ut i on t echni que. For t he weakl y meshed t ransmi ssi on
net wor ks t he number of i t erat i ons was hi gher, due t o
t he addi t i onal nonl i neari t i es i nt roduced by generat or
buses (PV nodes). I n al l t he cases st udi ed t he
proposed power f l owt echni que was si gni f i cant l y mor e
ef f i ci ent t han t he Newpn- Raphson power f l owal gori t hm
whi l e convergi ng t o t he same sol ut i on.
0885-8950/88/05oo-O753$01 .WO 1988 IEEE
754
The numer i cal ef f i ci ency of t he proposed
compensat i on- based power f l ow met hod, however ,
di mi ni shes as t he number of breakpoi nt s requi red t o
convert t he meshed net work t o a r adi al conf i gurat i on
i ncreases. Thi s r est r i ct s t he pract i cal appl i cat i on of
t he met hod t o weakl y meshed net works.
I n t hi s paper we emphasi ze t he appl i cat i on of t he
compensat i on- based power f l ow met hod t o t he
di st ri but i on net wor ks and provi de sever al pract i cal
exampl es. A compari son wi t h t he Newt on- Raphson based
WSCC power f l ow program i s al so present ed. We t hen
di scuss t he appl i cat i on of t he al gori t hm t o weakl y
meshed t ransmi ssi on net wor ks wher e agai n compari son
wi t h t he WSCC power f l owprogrami s provi ded. A bri ef
di scussi on of t he ext ensi on of t hi s met hod t o t hree
phase net works as wel l as t he t reat ment of di st ri but ed
l oads and ot her pract i cal consi derat i ons concl udes t he
paper.
11. SOLUTI ON OF A RADI AL DI STRI BUTI ONNETWORK
current s i n
f r om node L2 L=b, b- 1, ..., 1
JL(k) = - I Lt(k)+C(branches emanat i ng) (2)
where I L2(k) i s t he current i nj ect i on at node L2.
Thi s i s t he di rect appl i cat i on of t he KCL.
3. Forward sweep: Nodal vol t ages ar e updat ed i n
a f orward sweep st art i ng f rom branches i n t he
f i rst l ayer t oward t hose i n t he l ast . For
each branch, L, t he vol t age at node L2 i s
cal cul at ed usi ng t he updat ed vol t age at node L1
and t he branch cur r ent cal cul at ed i n t he
precedi ng backward sweep:
wher e z~ i s t he ser i es i mpedance of branch L.
Thi s i s t he di r ect appl i cat i on of t he KVL.
St eps 1, 2 and 3 ar e repeat ed unt i l conver gence i s
achi eved.
I n our al gor i t hm, r egar dl ess of i t s or i gi nal
t opol ogy, t he di st ri but i on net work i s f i r st convert ed
t o a r adi al net work. Hence, an ef f i ci ent al gor i t hmf or
t he sol ut i on of r adi al net works i s cr uci al t o t he
vi abi l i t y of t he over al l sol ut i on met hod.
The sol ut i on met hod used f or r adi al di st ri but i on
net works i s based on t he di rect appl i cat i on of t he KVL
and KCL. Si mi l ar t echni ques ar e al so descri bed i n
[5,6,7]. For our i mpl ement at i on, we devel oped a branch
ori ent ed approach usi ng an ef f i ci ent branch numberi ng
scheme t o enhance t he numer i cal perf ormance of t he
sol ut i on met hod. We f i rst descr i be t hi s branch
numberi ng scheme.
Br anch Number i ng
I n cont rast t o al l cl assi cal power f l owt echni ques
whi ch use nodal sol ut i on met hods f or t he net wor k, our
al gori t hmi s branch- ori ent ed. Fi gur e 1 shows a t ypi cal
r adi al di st ri but i on net work wi t h n nodes, b(=n-1)
branches and a si ngl e vol t age sour ce at t he root node.
I n t hi s t ree st ruct ure, t he node of a branch L cl osest
t o t he root not e i s denot ed by L1 and t he ot her node by
L2. We number t he branches i n l ayers away f r omt he root
node as shown i n Fi gur e 2. The numberi ng of branches
i n one l aver st ar t s onl v af t er al l t he branches i n t he
i mpl ement ed i n our power f l owprogram.
Sol ut i on Met hod
Gi ven t he vol t age at t he root node and assumi ng a
f l at prof i l e f or t he i ni t i al vol t ages at al l ot her
nodes, t he i t erat i ve sol ut i on al gori t hm consi st s of
t hree st eps:
1. Nodal current cal cul at i on: At i t erat i on k,
t he nodal current i nj ecti on, I i (k), at net work
node i i s cal cul at ed as,
I,@) = (si / vi (k- l ) ) *- Y~v~( ~- I ) i =1, 2, . . . , n (1)
wher e Vi (k-l ) i s t he vol t age at node i cal cul at ed
duri ng t he (k-l )th i t erat i on and Si i s t he
speci f i ed power i nj ect i on at node i . Yi i s t he
sumof al l t he shunt el ement s at t he node i .
LAYER 10
2. Backward sweep: At i t erat i on k, st ar t i ng
f r omt he branches i n t he l ast l ayer and movi ng Fi g. 2 Br anch numberi ng of t he r adi al di st ri but i on
t owar ds t he branches connect ed t o t he root node net work
t he current i n branch L, J,, i s cal cul at ed as:
755
MAXIMUM REAL
POWER MISMATCH
(W
Convergence Cri t eri on
~ ~~~~
MAXIMUM REACTIVE
POWER MISMATCH
(WAR)
We used t he maxi mum r eal and r eact i ve power
mi smat ches at t he net work nodes as our convergence
cri t eri on. As descri bed i n t he sol ut i on al gor i t hm, t he
nodal current i nj ect i ons, at i t erat i on k, ar e
cal cul at ed usi ng t he schedul ed nodal power i nj ect i ons
and node vol t ages f romt he previ ous i t erat i on ( equat i on
(1)). The node vol t ages at t he same i t erat i on ar e t hen
cal cul at ed usi ng t hese nodal current i nj ect i ons
( equat i ons (2) and (3)). He c , t he power i nj ect i on
f or node i at kt h i t erat i on, i s cal cul at ed as:
544 NODES
WXFIANGE
0.409-5.063
( 4 )
1
2
3
4
5
7
8
6
The r eal and r eact i ve power mi smat ches at bus i are
t hen cal cul at ed as:
mi ( k) =Re[Si (k) - Si ]
i =1, 2, . . . .n ( 4)
AQ~( ~) =I mI si (k) - si ]
Tabl e 1 shows t he val ues of t he maxi mumr eal and
react i ve power mi smat ches at t he vari ous i t erat i ons of
t hi s r adi al net work power f l owsol ut i on al gori t hmf or
t hree pract i cal di st ri but i on net works. An excel l ent
r at e of convergence can be observed f or al l t hree
net wor ks st udi ed. Thi s convergence behavi or can be
bri ef l y expl ai ned as f ol l ows. The error t hat i s
i ncurred i n est i mat i ng i ni t i al node vol t ages i s
propagat ed f i r st t o nodal and t hen t o branch current s
vi a equat i ons (1) and ( 2 ) . I n t he process of updat i ng
node vol t ages usi ng equat i on ( 3 1 , t he error i n branch
current s i s mul t i pl i ed by t he smal l l i ne i mpedance. ZL
()ZLI <<l ). and t hereby rapi dl y at t enuat ed.
DISTRIBUTION
l EEzRArn1
lTERATlON
NUMBER
244 NODES
WXFIANGE
0.245-5.065
1
2
3
4
5
6.134
0.301
0.008
0.000
0.000
13.092
0.567
0.024
0.001
0.000
I 1411 NODES I 1
I 0.000-5.480 1, ;
5.994
1.691
0.402
0.126
0.031
0.010
0.003
O.OO0
5.597
1.132
0.377
0.085
0.029
0.007
0.002
0.000
4.691 4.573
0.719
0.088 0.037
0.011 0.003
0.001
0.000
Tabl e 1 Conver gence charact eri st i cs of t he r adi al
net work sol ut i on al gori t hm
Fi gur e 3 shows t he f l owchar t of t he over al l power
f l owsol ut i on met hod f or r adi al net works.
111. SOLUTI ON OF WEAKLY MESHEDDI STRI BUTI ONNETWORKS
Fi gur e 4 shows an exampl e of a weakl y meshed
di st ri but i on net work cont ai ni ng f our si mpl e l oops. The
r adi al net wor k sol ut i on al gori t hmcan not be di rect l y
appl i ed t o t hi s net work. Never t hel ess, by sel ect i ng
f our br eakpoi nt s, t hi s net work can be convert ed t o a
r adi al conf i gurat i on. The branch cur r ent s i nt errupt ed
by t he cr eat i on of every breakpoi nt can be repl aced by
AND INITIAL CONDITION
SET ITERATON CWNT
CALCULATE MAXIMUM REAL AND
REACTIVE POWER MISMATCHES
EQUATION (4)
MAXlMUMlTERAT YES INT
Fi g. 3 Power f l ow sol ut i on al gori t hm f or t he r adi al
net wor ks
Fi g. 4 A weakl y meshed di st ri but i on net work
current i nj ect i ons at i t s t wo end nodes, wi t hout
af f ect i ng t he net work operat i ng condi t i on. Thi s
resul t i ng r adi al net wor k can now tp sol ved by t he
r adi al net wor k sol ut i on t echni que descri bed earl i er.
I n appl yi ng t he r adi al net work sol ut i on al gor i t hm,
t he current at breakpoi nt j , J j , must be i nj ect ed wi t h
opposi t e pol ari t y at t he t wo end nodes of t he
breakpoi nt . At i t erat i on k:
156
j = 1, 2, ..., p (5)
wher e j l and j 2 corr s ond t o h t wo end nodes of t he
breakpoi nt j , and I j i rkf and I j j ke become he nodal
current i nj ect i ons at t hese nodes, Tj(6 i s t he
breakpoi nt cur r ent and p t he t ot al number of
breakpoi nt s. I n t he presence of nodal cur r ent s at t he
t a poi nt no e due t o shunt el ement s andl or l oads,
$. ?kt and -3j h7' must be added t o t hese nodal current s.
r ocess i s c emat i cal l y shown i n Fi gur e 5. Once
$Tky and I -$? ar e updat ed, st eps 2 and 3 of t he
r adi al net wor i sol ut i on al gori t hm can be di rect l y
appl i ed.
BREAKPOINT
Fi g. 5 Breakpoi nt represent at i on usi ng nodal current
i nj ect i ons
Cal cul at i on of Breakpoi nt Current s Usi ng Compensat i on
Met hod
Breakpoi nt current s ar e cal cul at ed usi ng t he
mul t i - por t compensat i on met hod [ 81. Fi gur e 6
i l l ust rat es t he concept used i n t hi s approach. I n t hi s
f i gur e t he r adi al net work resul t i ng f romt he openi ng of
t he breakpoi nt s i s shown as a mul t i - por t ci r cui t wi t h
breakpoi nt nodes f ormi ng t he port s of t he ci rcui t . The
cal cul at i on of breakpoi nt current s r equi r es t hat t he
mul t i - por t equi val ent ci rcui t f or t he r adi al net work as
seen f romt he port s of t he breakpoi nt s be est abl i shed.
For a l i near net wor k, t hi s mul t i - por t equi val ent
ci r cui t can be t he Theveni n equi val ent ci rcui t of t he
radi al net work seen f romt he open port s creat ed b t he
breakpoi nt s. I n t hi s ci rcui t t he Theveni n vol t age f i s
t he (pxl ) vect or of open ci rcui t breakpoi nt vol t ages,
obt ai ned f romt he power f l ow sol ut i on of t he radi al
net work , [dl t he ( pxp) non- spar se mat ri x of t he
breakpoi nt i mpedances ( coef f i ci ent s rel at i ng breakpoi nt
current s and vol t ages) and 2 i s t he (pxl ) vect or of t he
desi red breakpoi nt current s ( Fi gure 7):
Fi g. 6
Mul t i - por t equi val ent of t he net work as seen
f romt he breakpoi nt port s
I n t he presence of const ant power l oads t he
di st r i but i on net work i s, however , nonl i near and equat i on
(6) cannot be di rect l y used. I nst ead, as we shal l
expl ai n, we cal cul at e breakpoi nt current s i t eri vel y
usi ng t he Theveni n equi val ent ci rcui t .
Pi g. 7 Theveni n equi val ent ci rcui t of t he net work as
seen f romt he breakpoi nt port s
Cal cul at i on of Breakpoi nt I mpedance Uat r i x
The breakpoi nt i mpedance mat ri x ( Theveni n
equi val ent i mpedance) can be det ermi ned usi ng t he
f ol l owi ng method:
Equat i on (6) can be wri t t en as,
A.
.. zpj ..... C '
Accordi ng t o equat i on ( 71, col umn j of t he
breakpoi nt i mpedance ma r i x wi l l be e ual t o vect or of
breakpoi nt vol t ages f or $j=l p. u. and j i =O, i =1, 2, . . . .p
and i # j . Thi s corresponds to t he appl i cat i on of
1 p. u. current at t he breakpoi nt j wi t h al l l oads and
t he sour ce at t he root node r emoved, whi ch i s, i n t ur n,
equi val ent t o t he i nj ect i on of 1 p. u. cur r ent s wi t h
opposi t e pol ari t y at t he t wo end nodes of t he
breakpoi nt j ( equat i on (5)). I n t he absence of l oads,
t he accur at e sol ut i on of t he power f l owf or t he r adi al
net work can be achi eved i n one i t erat i on. Each of t he
breakpoi nt vol t ages can be det ermi ned by subt ract i ng
t he vol t ages at t he t wo end nodes of t he breakpoi nt .
Thi s process must be repeat ed f or al l breakpoi nt s unt i l
al l t he col umns of t he breakpoi nt i mpedance mat ri x are
cal cul at ed.
I t er at i ve Compensat i on Pr ocess
The i t erat i ve compensat i on process f or cal cul at i ng
t he breakpoi nt cur r ent s, usi ng t he Theveni n equi val ent
ci rcui t of Fi gur e 7, i s descri bed i n t he f ol l owi ng:
1.
2.
Cal cul at e t he Theveni n equAval ent i mpedance
( breakpoi nt i mpedance mat ri x [Z] .of t he r adi al
net work) mai nt ai ni ng i t const ant t hr oughout t he
compensat i on process.
Cal cul at e t he Theveni n e ui val ent vol t age
( breakpoi nt vol t age vect or 1) of t he r adi al
net work usi ng t he radi al net work sol ut i on
al gor i t hm ( Fi gure 3) i ncl udi ng t he breakpoi nt
cur r ent s cal cul at ed f romt he previ ous i t erat i on
of t he compensat i on process. The i ni t i al val ues
of t he breakpoi nt cur r ent s ar e zero.
757
3. Calculate the incremental change in the
breakpoint currents using the Thevenin equivalent
circuit. At iteration m of the compensation
process:
4. Update the breakpoint currents.
At iteration m:
(8")
5. Repeat steps 2, 3 and 4 until convergence is
reached (the maximum breakpoint voltage
calculated at step 2 is within prescribed
limits)
This fixed tangent solution method is
schematically depicted in Figure 8 for a network having
a single breakpoint. Computationally there is no need
f r the inversion the breakpoint impedance matrix
1 is factorized once in the
beginning of the iterations and the forward fiand
backward substitution is then used to
in equation ( E' ) . Our test cases on practical
distribution networks showed that the number of
iterations required for the calculation of the
breakpoint currents was less than 5 in most cases.
[&. Complex matrix
calculate
Figure 9 shows the flow chart of the overall power
flow solution schome.
$( l ) $12)
Fig.8 Graphic representation of the iterative
compensation process
Selection of Breakpoints
Breakpoints are selected in order to convert
meshed network into a radial configuration. In
addition to this function breakpoints should be
selected in such a way as to ensure the convergence of
the overall solution algorithm. The latter requirement
for the selection of breakpoints usually is satisfied
by concentrating on the parts of the meshed network
where the power flows are low. On the other hand, the
power flows are the end product of the solution method
and are not known at the time of breakpoint selection.
In weakly meshed distribution networks breakpoint
selection does not affect the convergence performance
of the solution method in any noticeable manner.
Hence, we select them for the main purpose of opening
the network loops. Under these circumstances the
algorithm for identifying the breakpoints is very
simple and becomes part of the branch numbering scheme
described below:
1.
2.
3.
4.
5 ,
Examine all branches and select those
connected to the root node for the first branch
layer
Store the node number of the far node of the
branches in the branch layer just formed. For
all these nodes raise a flag indicating that they
have already been used
Examine all the remaining branches and select ,
those connected to any of far nodes of the
branches in the previous layer and place them in
a new branch layer
If the end node of a branch numbered in step 3
has been used before (flag identification of step
2) a loop has been formed and a breakpoint must
be created at this node
Repeat steps 2-4 ' until all branches are
processed.
SOLVE RADIAL NRWORK LOAD ROW ( W R E 3)
UPDATE THEVEMN VOLTAGE MISMATCH AT BREAKPOINTS
1
CALCULATE BREMPOhlT CURRENTS - ENATKIN (8)
ADD B E A P OM CURRENTS TO NODAL CURRENTS -
Fig.9 Compensation-based power flow method for the
weakly meshed networks
IV. RESULTS FOR THE DISTRIBUTION NETWORKS
We developed the program WNETPF (Weakly meshed
NETwork Power Flow) based on the proposed power flow
solution algorithm. This program successfully solved
several practical distribution networks with radial and
meshed structures. Table 2 shows the performance
results of this program alongside those of the
Newton-Raphson based WSCC power flow program. We also
used two other power flow programs using the generic
Newton-Raphson and the Gauss-Seidel solution algorithms
at this application stage. The generic Newton-Raphson
solution algorithm only converged for the smallest
network. The Gauss-Seidel power flow method converged
in all cases while requiring in excess of 20,000
iterations.
758
IISTRIWTION
IETwow(
ZNFWRATION
w NODES
WLOCPS
SLOOPS
For t he cases report ed i n Tabl e 2, t he proposed
al gor i t hm converged i n l ess t han 14 i t erat i ons i n 0. 21
t o 2. 1 CPU seconds on a mai nf rame comput er. Each
i t erat i on of t hi s programcorresponds t o one i t erat i on
i n t he r adi al net work sol ut i on al gori t hm pri or t o,
dur i ng, and af t er t he cal cul at i on of breakpoi nt
current s ( vari abl e k i n Fi gur e 3 ) . The number of out er
i t erat i ons f or cal cul at i ng t he breakpoi nt current s i n
t he compensat i on process ( vari abl e mi n Fi gur e 9) was
l ess t han 6 i n al l t hree cases. A f l at st ar t was used
i n al l cases wi t h t he t ol erance f or t he r eal and
react i ve power mi smat ches set t o 0. 05 kWand 0. 05 kvar.
Tabl e 2 i ndi cat es t hat t he proposed power f l ow
program i s si gni f i cant l y mor e ef f i ci ent t han t he
Newt on- Raphson power f l owmet hod when st udyi ng r adi al
and weakl y meshed di st ri but i on net works. Thi s
concl usi on i s part i cul arl y cr uci a1, f or : (a) on- l i ne
appl i cat i ons; (b) mul t i pl e power f l ow st udi es; (c)
mi cr o- and mi ni - comput er appl i cat i ons.
The WT P F program does not r equi r e doubl e
preci si on var i abl es and uses onl y one t wo di mensi onal
array ( breakpoi nt i mpedance matri x), hence, avoi ds
t axi ng comput er resources.
WSCCPOWERW -METHOD MAX MISMATCH
ml"S &!$!= ml"S
E,
REAL RUCTNE
5 0.97 3 021 O.CQ8 0.024
5 037 S 025 0.003 0.008
N3Loops 5 5.31 4 1.47 0011
36 LOOPS 5 527 10
1411 NOMS
Tabl e 2 Per f or mance resul t s f or di st r i but i on net wor ks
(I BM3090-200 mai nf rame comput er) - The t ot al
CPU t i me i n seconds i ncl udes t he t i me f or
i ni t i al processi ng of t he net wor k dat a and t he
i t erat i ve sol ut i on al gori t hm.
V. APPLI CATI ON TOWEAKLY MESHEDTRANSMI SSI ONNETWORKS
I n a weakl y meshed t ransmi ssi on net wor k, t he swi ng
bus i s assi gned as t he root node. Then t he branches
ar e numbered and breakpoi nt s sel ect ed i n exact l y t he
same manner used f or t he weakl y meshed di st ri but i on
net works. As a resul t a r adi al t ransmi ssi on net work i s
creat ed.
The sol ut i on al gori t hmf or a r adi al t ransmi ssi on
net work i s i dent i cal t o t hat of a r adi al di st ri but i on
net work except f or t he processi ng of t he generat or (PV)
nodes. For t he PV node i havi ng a speci f i ed power
i nj ect i on PT and vol t age magni t ude I VT I , we st ar t
t he i t erat i ons of t he r adi al net wor k sol ut i on al gori t hm
by assumi ng Vi (o) = l Vrl /O and =0. St eps 1
and 2 of t he r adi al net work sol ut i on met hod ar e t hen
perf ormed i n t he same manner as bef ore. St ep 3,
however , must be modi f i ed.
At i t erat i on k, t he vol t age magni t ude at t he
generat or node i , cal cul at ed usi ng equat i on ( 3) must
modi f i ed as,
Vi 2) i s t hen used f or cal cul at i ng t he vol t ages at t he
end nodes of t he branches i n t he next l ayer. The
react i ve power at t he generat or node i i s t hen updat ed
usi ng t he secant met hod as descri bed i n t he Appendi x.
I ncl usi on of generat or nodes usi ng t hi s appr oach
does not not i ceabl y det eri orat e t he convergence
propert i es of t he r adi al net work power f l owsol ut i on
al gori t hm. The ef f i ci ency of t he sol ut i on met hod f or
t he weakl y meshed net works wi l l , however, be af f ect ed
by t he i nt roduct i on of t he generat or nodes. The
nonl i neari t y i n t he t ransmi ssi on net wor k caused by t he
addi ti ori al generat or nodes i s more pronounced t han t hat
of t he di st r i but i on net wor ks havi ng const ant power
l oads al one. Thi s r esul t s i n an i ncreased number of
i t erat i ons f or cal cul at i ng t he breakpoi nt current s
usi ng equat i on ( 8) .
Tabl e 3 shows t he perf ormance r esul t s of appl yi ng
t he proposed al gori t hm and t he Newt on- Raphson based
WSCC power f l owprogramt o one r adi al and t hree weakl y
meshed t r ansmi ssi on net works. These net wor ks wer e
synt hesi zed f roma pract i cal 500kV t ransmi ssi on net work
wi t h a hi gh degr ee of ser i es compensat i on. The t abl e
shows t hat f or t he net wor ks wi t h l ow number of
breakpoi nt s, t he proposed l oad f l owt echhi ques i s more
ef f i ci ent t han t he Newt on- Raphson met hod. However , as
t he number of breakpoi nt s i ncreased, t he proposed
met hod requi red si gni f i cant l y hi gher number of
i t erat i ons whi l e t he Newt on- Raphson al gori t hmconverged
wi t h t he same number of i t erat i ons.
42 BUS, WOW
N3WNoDEs
NOLOOPS
1 LOOP
3 LOOPS
SLOOPS
42 BUS, 5W KV
4 w m
NO LOOPS
1 LOOP
3 LOOPS
5 LOOPS
Tabl e 3 Per f or mance r esul t s f or t ransmi ssi on net wor ks
(IBM 3090- 200 mai nf r ame comput er)
VI . PRACTI CAL CONSTDERATI ONS
Det ai l ed r epr esent at i on of a di st r i but i on net wor k
r equi r es [l o]:
1. The t hr ee phase represent at i on of t he net work to
account f or t he act ual l oad unbal ances.
2. The di st ri but ed l oad represent at i on al ong t he
di st r i but i on l i nes.
The r epr esent at i on of l oad t ap changer s, vol t age
r egul at or s, boost ers, etc.
3.
159
Requi rement 1 can be di rect l y i ncorporat ed i n t he
proposed power f l ow met hod by repl aci ng vol t age and
current scal ar s i n equat i ons (1) t hr u (l o), by ( 3x1)
vect ors of vol t ages and current s of t he t hr ee phases.
Under t hese ci r cumst ances, i n equat i ons (l), ( 31, (4)
and ( 71, t he admi t t ances and i mpedances shoul d be
represent ed by (3x3) mat ri ces.
Di st ri but ed l oads al ong di st ri but i on l i nes can be
approxi mat ed by l umped l oads at syst emnodes f or power
f l owcal cul at i ons purposes. Thi s, however , may requi re
t he addi t i on of pseudo- nodes al ong some of t he l i nes
wher e part of t he l i ne l oad i s l umped. The proposed
power f l owmet hod i s al so capabl e of di rect l y i ncl udi ng
di st ri but ed l oads. Thi s can be achi eved by t he
modi f i cat i on of equati ons(1) and (3).
Di st r i but i on net work equi pment ( regul at ors,
boost ers, ' etc. ) can be model ed i n t he proposed
al gori t hmwi t hout any rest ri ct i on.
Thi s met hod i s di rect l y appl i cabl e t o di st ri but i on
pl anni ng st udi es, wher e si ngl e phase r epr esent at i on of
t he net work wi t h l umped nodal l oads ar e consi dered
t o be adequat e. I n addi t i on we have successf ul l y
used t hi s power f l ow met hod f or our opt i mal
net work r econf i gur at i on st udi es.
I n t he case of weakl y meshed t ransmi ssi on
net works, a compl et e syst emrepresent at i on requi res t he
i ncl usi on of t ap changers and phase shi f t ers. The
proposed sol ut i on al gori t hm i s capabl e of i ncl udi ng
t hese component s di rect l y. Thi s r equi r es t hat t he KVL
and KCL be wr i t t en f or t he mat hemat i cal model of t he
t ap changers and phase shi f t ers.
VI I . CONCLUSI ON
Thi s paper present s a new, compensat i on- based
power f l ow met hod, f or t he sol ut i on of weakl y meshed
di st ri but i on and t ransmi ssi on net works. Thi s t echni que
i s si mpl e, st r ai ght f or war d, comput at i onal l y ef f i ci ent
and numeri cal l y robust . Ext ensi ve st udy of t he
perf ormance of t hi s compensat i on- based power f l ow
scheme shows t hat i t i s si gni f i cant l y more ef f i ci ent
t han t he Newt on- Raphson power f l owt echni que when used
f or sol vi ng r adi al and weakl y meshed di st r i but i on and
t ransmi ssi on net works.
ACKNOWLEDGEMENT
The aut hor s t hank Messrs R. L. Smi t h and J .
Monast eri o of PGhE' s El ect ri c Di st r i but i on Engi neeri ng
Depar t ment f or t hei r usef ul coment s on t he pract i cal
appl i cat i ons of t he proposed met hod. The Canadi an
aut hor s grat ef ul l y acknowl edge t he f i nanci al suppor t
f romt he Nat i onal Sci ences and Engi neeri ng Research
Counci l of Canada.
111 S. C. Tr i pat hy, G. Dur ga Pr asad, 0. P. Mal i k and
G. S. Hope, "Load Fl ow Sol ut i ons f or 111-
Condi t i oned Power Syst ems By a Newt on Li ke
Met hod", I EEE Trans. , PAS- 101, Oct ober 1982, pp.
3648- 3657.
[2] S. I wamot o, Y. Tamur a, "A Load Fl ow Cal cul at i on
Met hod f or I l l - Condi t i oned Power Syst ems", TEEE
Trans. , PAS- 100, Apri l 1981, pp. 1736- 1743.
[31 D. Raj i ci c, A. Bose, "A Modi f i cat i on t o t he Fast
Decoupl ed Power Fl owf or Net works wi t h Hi gh R/ X
Rat i os", PI CA ' 87 Conf er ence, Mont real , Canada.
K. A. Bi r t , J . J . Gr af f y, J . D. McDonal d,
A. H. El - Abi ad. "Three Phase Load Fl owProgram",
I EEE Trans. , PAS- 95, J anuaryI February 1976, pp.
59- 65.
W. H. Ker st i ng, D. L. Mendi ve, " An Appl i cat i on of
Ladder Net work Theor y t o t he Sol ut i on of Thr ee
Phase Radi al Load- Fl owProbl ems", I EEE PAS Wi nt er
Meet i ng, New Yor k, 1976, I EEE Paper No. A76 044- 8.
D. I . H. Sun, S. Abe, R. R. Shoul t z, M. S. Chen,
P. Ei chenber ger , D. Far r i s, "Cal cul at i on of Energy
Losses i n a Di st ri but i on Syst em", I EEE Trans. ,
PAS- 99, J ul y/ August 1980, pp. 1347- 1356.
R. Berg J r. , E. S. Hawki ns, W. W. Pl ei nes,
"Mechani sed Cal cul at i on of Unbal anced Load Fl ow
f or Radi al Di st ri but i on Ci rcui t s", I EEE Trans. ,
PAS- 86, Apr i l 1967, pp. 415- 421.
W. F. Ti nney, "Compensat i on Met hods f or Net work
Sol ut i ons by Tri angul ar Fact ori zat i on", Proc. of
PI CA Conf er ence, Bost on, Mass. , May 24- 26, 1971.
G. Gr oss, H. W. Hong, "A Two- st ep Compensat i on
Met hod f or Sol vi ng Shor t Ci rcui t Probl ems", I EEE
Trans. , PAS- 101, J une 1982, pp. 1322- 1331.
[ l o] N. Vempat i , R. R. Shoul t s, M. S. Chen,
L. Schwobel , "Si mpl i f i ed Feeder Model i ng f or Load
Fl owCal cul at i ons", 1EF.E Paper No. 86WM102- 8, PES
Wi nt er Power Meet i ng, New Yor k, 1986.
APPENDI X
At t he kt h i t erat i on, t he r eact i ve power i nj ect i on
requi red t o mai nt ai n t he vol t age at t he gener at or bus
i , can be cal cul at ed usi ng t he secant met hod:
wher e IVi ( k- l ) l and I Vi (k-2)l ar e t he vol t age
magni t udes at t he node i cal cul at ed i n t he previ ous
t wo i t erat i ons ( equat i on (3) i n st ep 3). The act ual
react i ve power i nj ect i on i s det ermi ned as:
Qic <Qimh
wher e Qimin and Qi- ar e t he r espect i ve mi ni m and
maxi mumr eact i ve power l i mi t s f or t he generat or node i .
R. P. Broadwater and A. Cbandrasekaran (Tennessee Technologkd
University, Cookeville, TN): The authors are to be complimented for
addressing an area that has received little attention, distribution power flow
anal ysi s.
Radial Distribution System Analysis
1) A suggested improvement to the radial power flow is to sumload
powers and power losses in the reverse trace (i.e., moving from the endiog
buses to the source bus) instead of summing load currents. This suggestion
has been tested on a four-line section systemas illustrated in Fig. D. 1. For a
nomi nal l y loaded case, both methods converged in four iterations.
However, for a very heavily loadedcase, the method of summing the
760
currents in thereversetracediverged, whereas themethod of summing the
powers converged.
A brief and heuristic explanation of this phenomenon is as follows.
Initially, when thecurrents are summed in the reversetrace, each current
will contain an error proportional to theinitially guessed voltage. If the
initially guessed voltages are maintained constant and a succession of power
flow problems are solved in which theloads &econtinually increased, the
errors that are proportional to theinitially guessed voltages wiU grow. For a
sufficiently heavily loaded system, theinitially guessed voltages fall outside
theregion of convergence, and thealgorithmwill diverge.
When thepowers are summedin thereversetrace, theerrors that exist
when thesourcebus is reached involveonly thepower losses, and not the
load powers. The power losses are always a small fraction of the load
powers. Hence, using thepower sum leads to good convergencefor even
heavily loaded systems.
2) Even though it is mentioned that themultiphaseunbalanced systems
are easily handled, theconvergencecharacteristics claimed for thesingle-
phase systemmay get severely impaired, sincetheerror involved in current
summation may become excessive.
3) The details of thedistribution networks given in Table1 do not include
theloading levels of thesystems. It would beinstructiveto know whether
thesystems are nominally loaded or lightly loaded. Further, thenumber of
nodes may not be a direct indication of the size of the systemsince
distributed loads can be modeled using any number of nodepoints.
Weakly Meshed Transmission Syst em
4) Theimpedancematrix of (7) of the paper appears to be the loop
impedancematrix of thesystemwith thebreakpoint currents chosen as the
loop currents. At theexact solution, thebreakpoii voltagevector must go
to zero. There appears to be a paradox here since for the constant
impedancematrix assumed thesolution for thecurrents is either trivial or
illfinite.
5) Thehandling of PV buses explained in section V does not mention
whether the alternating current directions in the traces would affect
convergenceif theP valueis higher than downstream lo&.
General
6) TheCPU timegiven inthetableis said to includetheinitial processing
timealso. This may show theWSCC Power Flow Programina bad light.
Exact CPU timerequired for theiterations aloneshould bea better index.
7) In Fig. 2, thenumbering of thedistribution network is laid out on a
grid in a very orderly fashion. With this scheme, it appears that choosing
buses at largeload centers may lead to conflicts. For instance, inLayer 2
developed for a load flow problem, thediscaqers believethat it is effective
to develop analgorithmwhich utiliis special features of theproblem; the
radial power systemis oneof such special feahres. In load flow problemof
a radial power system, thenumber of variables or computational burden can
be reduced extremely bytaking voltages of theends (or tips) of thebranches
as independent variables, in comparison with taking all nodevoltages as
variables. Themethod developed by theauthors al soutilizes such features.
Thediscussers have a question that it might be fast to make a rcduced
variableproblemas aboveand solveit directly by theNewton-Fbphson
method. Could theauthors comment on whether they haveever tried such a
reduced variable Newton-Raphson method, and if tried, how was the
comparison of computation timebetween thetwo methods?
N. Vempnti, B. K. Chea, and R. R. Shoolts (University of Texas at
Arlington, Arlington, TX): Theauthors havepresented an algorithmfor the
load flow solution of weakly meshed distribution and transmission
networks. Thesimplicity of thealgorithmmakes it a very interesting paper
to read. However, the practical limitations of the algorithmprevent its
general usagefor networks with multipleloops. It would be inadvisableto
havetwo programs, one to deal with strongly meshed networks and yet
another for weakly meshed ones.
Theresults werebased on tests performed on radial and weakly meshed
networks using thepositivesequencerepresentation. Theauthors attempt to
extrapolatetheresults to three-phase distribution and transmission networks
without a report of such an analysis. Until such results are presented, the
efficacy of such analgorithmis s t i l l in doubt.
Theauthors makean erroneous observation that power flow algorithms
for meshed distribution systems have not been developed. One of the
algorithms referred to [6] has the ability to analyzethree-phasenetworks,
irrespectiveof thecomplexity of themeshing. However, this algorithmwas
designed for distribution systems and thereforehas thelimitation of only
oneswing bus and no other voltagecontrolled (P- v) bus. Subsequently this
algorithmwas modified to accept numerous P- V buses, thereby enabling its
usagein the analysis of three-phasetransmission networks [A]. Wefeel that
the algorithm[6] based on an implicit Z-bus (i.e., bifactorcd Y-bus)
formulation is superior to theoneproposed by theauthors.
How do theauthors propose to model theopen-wyelopendelta trans-
formers in thedistribution systemanalysis? Our simulations haveshown
that theinjection currents dueto themodel are so largethat thecurrents due
to the loads are negligible. This affects theconvergencecharacteristics of
thealgorithmdirectly. ~ n y insight into themodeling and simulatioh of
thesetransformers would bewelcome.
References
[A] B. K. Chen, Transmission SystemUnbalance Analysis, Ph.D.
Diss., TheUniversity of Texas at Arlington, December 1986.
I
M. E. Baraa and F. F. Wu (University of California, Berkeley, CA): This
paper points out the reasons why a special power flow for distribution
systems is needed and provides a computationally attractivemethod. It is a
valuablecontribution.
There are a few points on which we would appreciatethe authors
claritlcation.
1) How does theconvergenceof themethod for radial networks depend
on the systempet er s , in particular, thelineresistances?
2) The amlication of the comuensation method for weaklv meshed
t
Fig. D. 1. Four-lie section test system.
supposethat a largeload exists at thecenter of branch 4, but thereis littleor
no load in the centers of branches 5 and 6. With theauthors proposed
scheme, it appears that additional buses would be inserted in branches 5 and
6 that would not be necessary for accurateload modeling.
8) It appears that if a component is inserted or added somewhereinthe
network, a completerenumbering of thenetwork is required. Is this true?
9) Even though the WSCC Power Flow Programis shown to converge
more slowly for the examples chosen in comparison to the proposed
method, it is a moot point whether a few extra seconds of CPU timealone
would besufficient to chooseany alternativemethod. TheNewtons method
has been so finely honed during the last two decades that negative
experiences must be very few to bedeemed almost nonexistent. Hence, the
virtues of the proposed method must be highlighted froma different
viewpoint.
K. Aolri, K. Nan, and T. Sat0 (HiroshimaUniversity, Higashihiroshima,
Japan): The authors have written an interesting paper on a load flow
problemof radial power system. Although many algorithms have been
networks isma very clever idea. ~e method uses an appro&te of the
linearized V-J function at thebreakpoiit. Wewonder if theconvergenceof
the method is always monotonic as implied in Fig. 8 or it exhibits
oscillatory behavior.
3) In calculating the breakpoiit impedancematrix, the authors have
observed that thecorresponding power flow solutions can beachieved in
oneiteration. Is this also true in cases wheretheshunts in thesystemare
significant and cannot beneglected?
Dr. Dromey: Theauthors haveinvestigated an aspect of load flow analysis
that has received littleattention in theliteratureto date. The widerangeof
R/X ratios ina distribution systemcan lead to difficulty with convergence
and this problemis aggravated by thepresenceof adjacent long and short
branches. In particular, the Newton-Raphson and decoupled algorithm
can, in some circumstances, fail to convergefor larger or illconditioned
systems. Themethod presented inthepaper produces anoptimal ordering
for solution and is very efficient, particularly wherethenumber of loops is
limited and thesystemis essentially radial in nature.
-*-
76 I
A number of questions arise in connection with theresults. Thepaper
infers that constant power loads are assumed. What is the effect on
convergencefor loads that havea mixtureof p/ u and q / u characteristics?
No mention is madeof theuse of convergencefactors. Wasthis studied and
are thereindications of optimal factors that can beconsidered generic for
theradial solution and for breakpoint injection currents?
Theauthors mention that the efficiency of thealgorithmdeteriorates with
the increasein the number of loops. In a large city with low voltage
downtown networks, the number of loops required to interconnect a large
number of radial sections can bein the hundreds. Thereis theadditional
problemof a significant number of very short cablesections used to balance
theflows between sections. What degreeof deterioration can beexpected in
the efficiency of solution for such a network wherethecoupling can bequite
strong? \
Thereis a significant advantagein being ableto solvethe networks
described aboveon a desktop microcomputer which will berestricted in the
memory availablefor processing. Would the authors like to suggest a
sensiblemethod of partitioning such a network to achievesolutions in an
acceptabletime? How would the compensation method be modified to
account for thepartitioning?
D. Shirmohammadi, H. W. Hong, A. Semlyen, and G. X. Luo: We
would like to thank the discussers for their interest in our paper and for
their questions anti comments. Many of theseconstitutecontributions to
the topic of the paper. We will give our answers to each discusser
separately.
' Messrs. ~roamVater and Chandrmehrun:
Webelievethat the discussers' suggestion of adding up powers
rather than currents in thebackward sweep is interesting and their
explanation quite plausible. Wesuspect however that the improved
convergenceis obtained with morecomputations. Moreover, we
havenot experienced any problems in dealing with heavily loaded
networks and low nodevoltages. In fact, in the 544 nodenetwork
exampleof thepaper, due to heavy loading and lack of VAR sup
port the voltages of someof the nodes a as low as 0.75 per unit.
This network was efficiently solved by theproposed algorithm.
The pmsS of the attenuation of theemrs described in thesection
of thepaper on "ConvergenceCriterion" is completely general and
would apply to multi-phaseunbalanced as well as single-phasenet-
works.
AU thm examples included in the paper repment peak load condi-
tions on the feeders. These were4 MW, 3 MW, and 8 MW loads
on the 244 node, 544 node, and 141 1 nodenetworks, respectively.
Wewould like to clarify the discussers' question related to equa-
tions (6) and (7): thevoltages thereare internal Thevenin voltages
whilethe breakpoint voltages arezero sincethe breakpoint ports
are shortcircuited. Wewould also like to point out that the actual
relationship between thebreak point voltages and currents is esta-
blished through equation (8).
Wehave not experienced any problemin theoverall convergence
of the proposed method in solving a variety of networks that
included P,V buses. Weagree, however, that theremay bemore
efficient ways of handling P,V buses in order to minimize the
impact on the convergencecharacteristics of themethod.
In contrast to the discussers assumption, theinclusion of initial pro-
cessing timeputs in m3refavorablelight the WSCC Power Flow
Program. For example, the timerequired for each iteration of the
proposed algorithmis around 0.005 seconds for the 244 node
exampleof the paper. At the sametime, every iterator of the
Newton-Raphson based WSCC Power Flow Programtook 0.08
seconds for the samenetwork (16 times more). Similar results
wereobtained for the 544 nodeand 141 1 nodenetworks.
In our distribution network model wehaveassumed that loads are
concentrated at network nodes. As a result, theintroduction of a
load at the center of branch 4 means the addition of a new nodeat
this location. This will also add a new branch to thenetwork. If
there are no loads at the end nodeof branch 5, this branch and
branch 10 can becombined and represented with a singlebranch.
Branch 6 must exist, becausetwo separatebranches emanatefrom
its end node.
Thereis no need for the renumbering of theentirenetwork as a
result of theinsertion or deletion of network components. Only
branches in the layers below the inserted or deleted component
must be renumbered.
(9) Wedo not agreewith thediscussers view on the merits of thepro-
posed algorithm. These are well documented throughout thepaper.
Furtliermore, wewould like to point out that many major advances
in the development of power system analysis techniques have
resulted frommethodologies that exploit thespecial structure of the
power system. An exampleis the Fast Decoupled Power Flow
which provides improved efficiency of "a few CPU-seconds" in the
solution of the transmission network by exploiting thelow RIX
ratios prevalent in thesenetworks. Yet, the impact of the Fast
Decoupled Power Flow in the field of the transmission network
analysis has been very significant.
Messrs. Aoki, Nara, and Sato:
As pointed out by thediscussers, special features of particular load
flow problems can be exploited to procluce more efficient solution
methods. In the approach of the paper, the transversal elements have
been lumped witH theloads (see our answers to Messrs. Vempati, Chen.
and Shoults): this has madethefactorization and the subsequent alge-
braic operations with the L and U matrices of a nodal approach
equivalent to using tree-branch voltages and currents. Thesematrix cal-
culations are implicit in the method and did not haveto beperformed
explicitly. This is the clue of the computational efficiency of branch-
oriented calculations in a radial network. However, at each nodewehave
had to enforce the power (P, Q) constraints. Therefore, all bus voltages
are used as variables. Because of thi s only little improvement of
efficiency can bederived fromthe radial network structurein a Newton-
type load flow.
Messrs. Vempati, chen, and Shoults:
It is true that the solution method of the paper becomes less
efficient as the number of loops increases. However, the ovenvhelming
majority of all distribution networks and also the transmission systems of
many countries are weakly meshed. Therefore, a special program, if it is
moreefficient than a general one, is celtainly of i nterest Very often
advances in any field of knowledgeare based on thespecial structureof a
palticular problem. Onecould cite innumerableexamples whereSpecial
programs are developed and used for particular situations.
Wehavenot yet attempted the application of our method to multi-
phasenetworks. However, as weexplained in our response to Messrs.
Broadwater and chandrasekaran wedo not foreseeany deterioration in
the convergencecharacteristics of this method when applied to multi-
phasenetworks.
With regard to our statement about thelack of power flow algo-
rithms for meshed distribution network, we note that we made two
erroneous remarks. First, as the discussers rightly argue, the2-bus solu-
tion algorithmof Ref. [ 6] is capableof solving meshed distribution net-
works. Second, westated that Ref. [6] proposes a popular algorithmfor
distribution network analysis. Based on further investigation, however,
wehavefound that this solution method is by no means popular among
distribution engineers.
The discussers raise the problemof the relative computational
efficiency of themethod of Ref.[6] compared to themethod of the paper.
It is easy to show that a nodal approach (for exampletheoneof Ref.[61)
requires more computation than the branch-oriented approach of the
paper. Wewill show that the WO differ computationally by two facts:
(a) Thecomputations in thebranch-oriented method are equivalent to
theforward andbackward substitutions of thenodal method but the
factorization of a matrix is not needed.
(b) The forward and backward substitutions in a nodal approach are
replaced in the branch-oriented method by additions and subtrac-
tions and no multiplications and divisions areneeded.
Consider, for example, the simpleradial network of Fig.D.l of the
discussion by Messrs. Broadwater and cbandrasekaran. Connect themot
node to ground via a voltage source with V=O but do not connect
impedancebranches to ground fromthe other nodes. Number thenodes
and branches moving outward fromthe mot. Wewill then havebranch
and bus voltages and currents. Wecan relatetheseto each other by the
incidencematrices AV and A,,
VbW=AvVb and l b =Ar l k (a)
762
It can be seen that theincidencematrices aresquare, lower triangular,
and consist of elements f l only. Their inverses could al so havebeen
formed directly by inspection of thenetwork graph. This reflects thefact
that equations (a) represent directly the two Kirchhoffs laws. They
correspond to theforward and backward sweeps used in thepaper.
Let Yb (diagonal) and Y& be theadmittancematrices of the net-
work. Wehave
Y,=AI YbwAV
Clearly, the two incidencematrices arethetransposeof each other.
Let us now solvethenodal problem
ybw v&=Ibur
Factorization of Y& yields
Comparing (d) with 0) weobtain
so that the solution of (c) becomes
Y & = L D L ~
L-'=AI, LTT=Av. and D=Ybr
vbur=L-T D-' L-' IbyI =AV (& (AI [&))=AV (Zbr It,,) =AV V, (r)
Equation (f) shows that the branch-oriented solution reproduces the
matrix operations of the nodal approach without theneed of preliminary
formulation and factorization of a bus admittancematrix. To achieve
this, it was essential to replaceall shunt branches connected to buses by
corresponding injections.
Wenote that reference[6] claims that theefficiency of its algo-
rithmis comparableto that of the Newton-Raphson Power Flow while
we have shown a substantial improvement over the Newton-Raphson
Power Fl ow method using our algorithm.
Weappreciatethe information provided by thediscussers on prob-
lems related to transformer modeling. We have not investigated this
topic.
Messrs. Bamn and Wu:
(1) Theconvergenceof the method for 4 radial network is linear and
dependent essentially on theline impedances IZ I and theapparent
powers IS I of theloads. This can be seen fromthe following
simplified convergenceanalysis, for a singleline of impedance
Z=R+j X and load S=P+j Q. For this, eqn.(l) becomes I=S*/V*
or, in incremental form,
N =- KAY' V" (g)
The resultant changein voltage is, taking (g) into account with
V=l,
AVNW=-ZN=& AV* (h)
This equation shows that the convergencerateof I AV 1 (not A I V I
!) is given by I z(i I =I Z I IS I . It depends only indirectly on line
resistances.
Fig. 8 of thepaper is used only to depict thebasic mechanismof
thefixed tangent solution algorithmused for calculating breakpoint
current injections according to equation (8). Nevefieless. our
experiencewith all distribution networks studied indicates a mono-
tone and rapid convergence in the calculation of bRakpoint
currents.
After a considerableamount of experimentation we found that a
better convergencein the calculation of breakpoiit cmnts can be
achieved when the breakpoint impedance matrix is calculated
neglecting the shunt components (mainly capacitors). Unfor-
tunately this important conclusion is not reflected in thepaper and
wewould like to thank the discussers for providing us with thi s
opportunity.
Dr. Dromey:
Themethod presented in the paper can handleany load characteris-
tics, sincethe load current is calculated at each step as a function of the
prevailing voltage. We did not use however any convergence
(accelerating) factors. Wefeel that such factors (not necessarily uniform
and real ) could improvetheconvergenceof themethod and appreciate
thesuggestion implied in the question.
Wehavedeveloped our methodology and theaccompanying pro-
grammainly for primary distribution networks. Primary distribution net-
works are, in almostall cases, either radial or weakly meshed. The pro-
gramis, however, capableof representing up to 5000 nodes and 300
loops which is adequatefor almost all practical cases including secon-
dary distribution networks in downtown metropolitan areas. Even with
such largedimensidns. the memory requirement for theprogramis less
than 500 Kbytes which makes it ideal for microcomputer applications.
Thelargest network studied using this programconsisted of 241 1 nodes
and 183 loops. It took theprograma total CPU timeof 12 seconds to
process theinput data, solvethepower flow and print theresults for this
network on an IBM 3090-200 computer, which is very reasonablecon-
sidering the size of thenetwork.
Partitioning of distribution networks, or even of transmission net-
works, involves network equivalencing. Reference[A] is pertinent to this
topic.
[AI F.F. Wu, A. Monticelli, "A Critical Review on External Network
Modelling for On-LineSecurity Analysis", Electrical Power and
Energy Systems, Vol. 5, October 1983, pp. 222-235.
Apndice G
A DIRECT APPROACH FOR DISTRIBUTION SYSTEM LOAD FLOW
SOLUTION, TENG. 2003, [19]
114
882 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 18, NO. 3, JULY 2003
A Direct Approach for Distribution System Load
Flow Solutions
Jen-Hao Teng, Member, IEEE
AbstractA direct approach for unbalanced three-phase
distribution load flow solutions is proposed in this paper. The
special topological characteristics of distribution networks have
been fully utilized to make the direct solution possible. Two
developed matricesthe bus-injection to branch-current matrix
and the branch-current to bus-voltage matrix and a simple
matrix multiplication are used to obtain load flow solutions. Due
to the distinctive solution techniques of the proposed method, the
time-consuming LU decomposition and forward/backward sub-
stitution of the Jacobian matrix or admittance matrix required
in the traditional load flow methods are no longer necessary.
Therefore, the proposed method is robust and time-efficient. Test
results demonstrate the validity of the proposed method. The
proposed method shows great potential to be used in distribution
automation applications.
Index TermsDistribution load flow, distribution automation
system, radial network, weakly meshed network.
I. INTRODUCTION
M
ANY programs of real-time applications in the area of
distribution automation (DA), such as network optimiza-
tion, Var. planning, switching, state estimation, and so forth, re-
quire a robust and efficient load flow method [1][3]. Such a
load flow method must be able to model the special features of
distribution systems in sufficient detail. The well-known char-
acteristics of an electric distribution system are
radial or weakly meshed structure;
multiphase and unbalanced operation;
unbalanced distributed load;
extremely large number of branches and nodes;
wide-ranging resistance and reactance values.
Those features cause the traditional load flow methods
used in transmission systems, such as the Gauss-Seidel and
Newton-Raphson techniques, to fail to meet the requirements
in both performance and robustness aspects in the distribution
system applications. In particular, the assumptions necessary
for the simplifications used in the standard fast-decoupled
Newton-Raphson method [4] often are not valid in distribution
systems. Therefore, a novel load flow algorithm for distribution
systems is desired. To qualify for a good distribution load flow
algorithm, all of the characteristics mentioned before need to
be considered.
Several load flow algorithms specially designed for distri-
bution systems have been proposed in the literature [5][13].
Manuscript received October 17, 2002. This work was sponsored by National
Science Council under research Grant NSC 88-2213-E-214-041.
The author is with the Department of Electrical Engineering, I-Shou Univer-
sity, Kaohsiung 840, Taiwan, R.O.C.
Digital Object Identifier 10.1109/TPWRD.2003.813818
Some of these methods were developed based on the general
meshed topology like transmission systems [5][9]. From those
methods, the Gauss implicit -matrix method [7] is one of the
most commonly used methods; however, this method does not
explicitly exploit the radial and weakly meshed network struc-
ture of distribution systems and, therefore, requires the solution
of a set of equations whose size is proportional to the number
of buses. Recent research proposed some new ideas on how
to deal with the special topological characteristics of distribu-
tion systems [10][13], but these ideas require new data format
or some data manipulations. In [10], the authors proposed a
compensation-based technique to solve distribution load flow
problems. Branch power flows rather than branch currents were
later used in the improved version and presented in [11]. Since
the forward/backward sweep technique was adopted in the so-
lution scheme of the compensation-based algorithm, new data
format and search procedure are necessary. Extension of the
method, which emphasized on modeling unbalanced loads and
dispersed generators, was proposed in [12]. In [13], the feeder-
lateral based model was adopted, which required the layer-lat-
eral based data format. One of the main disadvantages of the
compensation-based methods is that new databases have to be
built and maintained. In addition, no direct mathematical rela-
tionship between the system status and control variables can be
found, which makes the applications of the compensation-based
algorithm difficult.
The algorithm proposed in this paper is a novel but classic
technique. The only input data of this algorithm is the conven-
tional bus-branch oriented data used by most utilities. The goal
of this paper is to develop a formulation, which takes advan-
tages of the topological characteristics of distribution systems,
and solve the distribution load flow directly. It means that the
time-consuming LU decomposition and forward/backward sub-
stitution of the Jacobian matrix or the admittance matrix, re-
quired in the traditional Newton Raphson and Gauss implicit
matrix algorithms, are not necessary in the new develop-
ment. Two developed matrices, the bus-injection to branch-cur-
rent matrix and the branch-current to bus-voltage matrix, and a
simple matrix multiplication are utilized to obtain load flow so-
lutions. The treatments for weakly meshed distribution systems
are also included in this paper. The proposed method is robust
and very efficient compared to the conventional methods. Test
results demonstrate the feasibility and validity of the proposed
method.
II. UNBALANCED THREE-PHASE LINE MODEL
Fig. 1 shows a three-phase line section between bus and .
The line parameters can be obtained by the method developed by
0885-8977/03$17.00 2003 IEEE
TENG: A DIRECT APPROACH FOR DISTRIBUTION SYSTEM LOAD FLOW SOLUTIONS 883
Fig. 1. Three-phase line section model.
Carson and Lewis [2]. A 4 4 matrix, which takes into account
the self and mutual coupling effects of the unbalanced three-
phase line section, can be expressed as
(1)
After Krons reduction is applied, the effects of the neutral or
ground wire are still included in this model as shown in (2)
(2)
The relationship between bus voltages and branch currents in
Fig. 1 can be expressed as
(3)
For any phases failed to present, the corresponding row and
column in this matrix will contain null-entries.
III. ALGORITHM DEVELOPMENT
The proposed method is developed based on two derived ma-
trices, the bus-injection to branch-current matrix and the branch-
current to bus-voltage matrix, and equivalent current injections.
In this section, the development procedure will be described in
detail.
For distribution networks, the equivalent-current-injection-
based model is more practical [5][13]. For bus , the complex
load is expressed by
(4)
And the corresponding equivalent current injection at the -th
iteration of solution is
(5)
where and are the bus voltage and equivalent current
injection of bus at the -th iteration, respectively. and are
the real and imaginary parts of the equivalent current injection
of bus at the -th iteration, respectively.
Fig. 2. Simple distribution system.
A. Relationship Matrix Developments
A simple distribution system shown in Fig. 2 is used as an ex-
ample. The power injections can be converted to the equivalent
current injections by (5), and the relationship between the bus
current injections and branch currents can be obtained by ap-
plying Kirchhoffs Current Law (KCL) to the distribution net-
work. The branch currents can then be formulated as functions
of equivalent current injections. For example, the branch cur-
rents and , can be expressed by equivalent current
injections as
(6)
Therefore, the relationship between the bus current injections
and branch currents can be expressed as
(7a)
Equation (7a) can be expressed in general form as
(7b)
where BIBC is the bus-injection to branch-current (BIBC) ma-
trix.
The constant BIBC matrix is an upper triangular matrix and
contains values of 0 and only.
The relationship between branch currents and bus voltages as
shown in Fig. 2 can be obtained by (3). For example, the voltages
of bus 2, 3, and 4 are
(8a)
(8b)
(8c)
where is the voltage of bus , and is the line impedance
between bus and bus .
Substituting (8a) and (8b) into (8c), (8c) can be rewritten as
(9)
884 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 18, NO. 3, JULY 2003
From (9), it can be seen that the bus voltage can be expressed
as a function of branch currents, line parameters, and the sub-
station voltage. Similar procedures can be performed on other
buses; therefore, the relationship between branch currents and
bus voltages can be expressed as
(10a)
Equation (10a) can be rewritten in general form as
(10b)
where BCBV is the branch-current to bus-voltage (BCBV) ma-
trix.
B. Building Formulation Development
Observing (7), a building algorithm for BIBC matrix can be
developed as follows:
Step 1) For a distribution system with -branch section and
-bus, the dimension of the BIBC matrix is
.
Step 2) If a line section is located between bus and
bus , copy the column of the -th bus of the BIBC
matrix to the column of the -th bus and fill a to
the position of the -th row and the -th bus column.
Step 3) Repeat procedure (2) until all line sections are in-
cluded in the BIBC matrix. From (10), a building
algorithm for BCBV matrix can be developed as
follows.
Step 4) For a distribution system with -branch section and
-bus, the dimension of the BCBV matrix is
.
Step 5) If a line section is located between bus and
bus , copy the row of the -th bus of the BCBV
matrix to the row of the -th bus and fill the line
impedance to the position of the -th bus row
and the -th column.
Step 6) Repeat procedure (5) until all line sections are in-
cluded in the BCBV matrix.
The algorithm can easily be expanded to a multiphase line
section or bus. For example, if the line section between bus
and bus is a three-phase line section, the corresponding branch
current will be a 3 1 vector and the in the BIBC ma-
trix will be a 3 3 identity matrix. Similarly, if the line section
between bus and bus is a three-phase line section, the in
the BCBV matrix is a 3 impedance matrix as shown in (2).
It can also be seen that the building algorithms of the BIBC
and BCBV matrices are similar. In fact, these two matrices were
built in the same subroutine of our test program. Therefore,
the computation resources needed can be saved. In addition,
the building algorithms are developed based on the traditional
bus-branch oriented database; thus, the data preparation time
can be reduced and the proposed method can be easily integrated
into the existent DA.
C. Solution Technique Developments
The BIBC and BCBV matrices are developed based on the
topological structure of distribution systems. The BIBC ma-
trix represents the relationship between bus current injections
and branch currents. The corresponding variations at branch
currents, generated by the variations at bus current injections,
can be calculated directly by the BIBC matrix. The BCBV ma-
trix represents the relationship between branch currents and bus
voltages. The corresponding variations at bus voltages, gener-
ated by the variations at branch currents, can be calculated di-
rectly by the BCBV matrix. Combining (7b) and (10b), the re-
lationship between bus current injections and bus voltages can
be expressed as
(11)
And the solution for distribution load flow can be obtained by
solving (12) iteratively
(12a)
(12b)
(12c)
According to the research, the arithmetic operation number
of LU factorization is approximately proportional to . For a
large value of , the LU factorization will occupy a large por-
tion of the computational time. Therefore, if the LU factoriza-
tion can be avoided, the load flow method can save tremendous
computational resource. Fromthe solution techniques described
before, the LU decomposition and forward/backward substitu-
tion of the Jacobian matrix or the admittance matrix are no
longer necessary for the proposed method. Only the DLF ma-
trix is necessary in solving load flow problem. Therefore, the
proposed method can save considerable computation resources
and this feature makes the proposed method suitable for online
operation.
IV. TREATMENTS FOR WEAKLY MESHED NETWORKS
Some distribution feeders serving high-density load areas
contain loops created by closing normally open tie-switches.
The proposed method introduced before can be extended for
weakly-meshed distribution feeders.
A. Modification for BIBC Matrix
Existence of loops in the system does not affect the bus cur-
rent injections, but new branches will need to be added to the
system. Fig. 3 shows a simple case with one loop. Taking the
new branch current into account, the current injections of bus 5
and bus 6 will be
(13)
TENG: A DIRECT APPROACH FOR DISTRIBUTION SYSTEM LOAD FLOW SOLUTIONS 885
Fig. 3. Simple distribution system with one loop.
The BIBC matrix will be
(14a)
Equation (14a) can be rewritten as
(14b)
And the modified BIBC matrix can be obtained as
(15a)
The general form for the modified BIBC matrix is
(15b)
The building algorithm of Step 2) for BIBC matrix can be mod-
ified as follows:
Step 2a)If a new branch makes the system become
meshed (the new branch is between bus and ), copy the ele-
ments of the -th bus column to the -th column and minus the
elements of the -th bus column. Finally, fill a value to the
position of the -th row and the -th column.
B. Modification for BCBV Matrix
Considering the loop shown in Fig. 3, KVL for this loop can
be written as
(16)
Combining (16) and (10a), the new BCBV matrix is
(17a)
The general form for the modified BCBV matrix is
(17b)
The building algorithm of Step 5) for the BCBV matrix can be
modified as follows:
Step 5a)If a new branch makes the system become
meshed, adds a new row to the original BCBV matrix by KVL.
The general form of KVL for a loop can be expressed as
(18)
where is the number of branches in this loop, and is the
line impedance corresponding to the branch current .
C. Modification for Solution Techniques
Substituting (15) and (17) into (11), (11) can be rewritten as
(19)
Applying Krons Reduction to (19), the modified algorithm for
weakly meshed networks can be expressed as
(20)
Note that except for some modifications needed to be done for
the BIBC, BCBV, and DLF matrices, the proposed solution
techniques require no modification; therefore, the proposed
method can obtain the load flow solution for weakly meshed
distribution systems efficiently.
V. TEST RESULTS
The proposed three-phase load flow algorithm was im-
plemented using Borland C++ language and tested on a
Windows-98-based Pentium-II PC. Two methods are used for
tests and the convergence tolerance is set at 0.001 p.u.
Method 1: The Gauss implicit -matrix method [7].
Method 2: The proposed algorithm.
886 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 18, NO. 3, JULY 2003
Fig. 4. Eight-bus distribution system.
TABLE I
FINAL CONVERGED VOLTAGE SOLUTIONS
A. Accuracy Comparison
For any newmethod, it is important to make sure that the final
solution of the new method is the same as the existent method.
An eight-bus system(equivalent 13-node system), including the
three-phase, double-phase, and single-phase line sections and
buses as shown in Fig. 4 is used for comparisons. The final
voltage solutions of method 1 and method 2 are shown in Table I.
From Table I, the final converged voltage solutions of method
2 are very close to the solution of method 1. It means that the
accuracy of the proposed method is almost the same as the com-
monly used Gauss implicit -matrix method.
B. Performance Test
A main feeder trunk with 3 90-phase buses, which was ac-
quired from the Taiwan Power Company (TPC), is used for this
test. The single and double-phase laterals have been lumped to
form the unbalanced loads for testing purposes. This trunk is
then chopped into various sizes for tests as shown in Table II.
The substation is modeled as the slack bus.
1) Radial Network Test: Table III lists the number of itera-
tions and the normalized execution time for both methods. It can
be seen that method 2 is more efficient, especially when the net-
work size increases, since the time-consuming processes such as
LU factorization and forward/backward substitution of -ad-
mittance matrix are not necessary for method 2. For a 270-node
system, method 2 is almost 24 times faster than method 1.
TABLE II
TEST FEEDER
TABLE III
NUMBER OF ITERATION AND NORMALIZED EXECUTION TIME
TABLE IV
TEST RESULTS FOR THE WEAKLY MESHED FEEDERS
2) Weakly-Meshed Network Test: Some branches are con-
nected to the test feeder to make the system meshed. Table IV
shows the number of iterations and normalized execution time
for the weakly meshed network. Table IVshows that the number
of iterations of method 2 is stable. The normalized execution
time increases, since the meshed network increase the nonzero
terms of the BIBC and BCBV matrices, and extra procedure
needs to be done.
C. Robustness Test
One of the major reasons, which make the load flow program
diverge, is the ill-condition problem of the Jacobian matrix or
admittance matrix. It usually occurs when the system con-
tains some very short lines or very long lines. In order to prove
that the proposed method can be utilized in severe conditions,
IEEE 37-bus test feeder is used [14]. The test feeder is adjusted
by changing the length of eight line sections. Four of them are
multiplied by ten, and the other four are divided by ten. The test
result shows the number of iterations for this case is 4 and the
execution time is 0.0181 s. It means that the proposed method
is robust and very suitable for online use.
VI. DISCUSSION AND CONCLUSION
In this paper, a direct approach for distribution load flowsolu-
tion was proposed. Two matrices, which are developed from the
topological characteristics of distribution systems, are used to
TENG: A DIRECT APPROACH FOR DISTRIBUTION SYSTEM LOAD FLOW SOLUTIONS 887
solve load flow problem. The BIBC matrix represents the rela-
tionship between bus current injections and branch currents, and
the BCBV matrix represents the relationship between branch
currents and bus voltages. These two matrices are combined
to form a direct approach for solving load flow problems. The
time-consuming procedures, such as the LU factorization and
forward/backward substitution of the Jacobian matrix or ad-
mittance matrix, are not necessary in the proposed method. The
ill-conditioned problem that usually occurs during the LU fac-
torization of the Jacobian matrix or admittance matrix will
not occur in the proposed solution techniques. Therefore, the
proposed method is both robust and efficient. Test results show
that the proposed method is suitable for large-scale distribution
systems. Other issues involved in the distribution system op-
eration, such as multiphase operation with unbalanced and dis-
tributed loads, voltage regulators, and capacitors with automatic
tap controls, will be discussed in future work.
REFERENCES
[1] IEEE Tutorial Course on Distribution Automation.
[2] IEEE Tutorial Course on Power Distribution Planning.
[3] W. M. Lin and M. S. Chen, An overall distribution automation struc-
ture, Elect. Power Syst. Res., vol. 10, pp. 719, 1986.
[4] B. Stott and O. Alsac, Fast decoupled load flow, IEEE Trans. Power
Apparat. Syst., vol. 93, pp. 859869, May/June 1974.
[5] J. H. Teng and W. M. Lin, Current-based power flowsolutions for distri-
bution systems, in Proc. IEEE Int. Conf. Power Syst. Technol., Beijing,
China, 1994, pp. 414418.
[6] T. S. Chen, M. S. Chen, T. Inoue, and E. A. Chebli, Three-phase cogen-
erator and transformer models for distribution system analysis, IEEE
Trans. Power Delivery, vol. 6, pp. 16711681.2, Oct. 1991.
[7] T.-H. Chen, M.-S. Chen, K.-J. Hwang, P. Kotas, and E. A. Chebli, Dis-
tribution system power flow analysisA rigid approach, IEEE Trans.
Power Delivery, vol. 6, pp. 11461152, July 1991.
[8] T. H. Chen and J. D. Chang, Open wye-open delta and open delta-open
delta transformer models for rigorous distribution system analysis, in
Proc. Inst. Elect. Eng., vol. 139, 1992, pp. 227234.
[9] K. A. Birt, J. J. Graffy, J. D. McDonald, and A. H. El-Abiad, Three
phase load flow program, IEEE Trans. Power Apparat. Syst., vol.
PAS-95, pp. 5965, Jan./Feb. 1976.
[10] D. Shirmohammadi, H. W. Hong, A. Semlyen, and G. X. Luo, A com-
pensation-based power flowmethod for weakly meshed distribution and
transmission networks, IEEE Trans. Power Syst., vol. 3, pp. 753762,
May 1988.
[11] G. X. Luo and A. Semlyen, Efficient load flowfor large weakly meshed
networks, IEEE Trans. Power Syst., vol. 5, pp. 13091316, Nov. 1990.
[12] C. S. Cheng and D. Shirmohammadi, Athree-phase power flowmethod
for real-time distribution systemanalysis, IEEETrans. Power Syst., vol.
10, pp. 671679, May 1995.
[13] R. D. Zimmerman and H. D. Chiang, Fast decoupled power flow for
unbalanced radial distribution systems, IEEE Trans. Power Syst., vol.
10, pp. 20452052, Nov. 1995.
[14] W. M. Kersting and L. Willis, Radial Distribution Test Systems, IEEE
Trans. Power Syst.,, vol. 6, IEEE Distribution Planning Working Group
Rep., Aug. 1991.
Jen-Hao Teng (M99) was born in 1969 in Tainan, Taiwan, R.O.C. He received
the B.S., M.S., and Ph.D. degrees in electrical engineering from the National
Sun Yat-Sen University, Taiwan, R.O.C., in 1991, 1993, and 1996, respectively.
Currently, he is with I-Shou University, Taiwan, R.O.C., where he has been
since 1998. His current research interests include energy management system,
distribution automation system, and power system quality.
Apndice H
THE DISTRIBUTION TRX-POWER FLOW METHOD, DE OLIVEIRA.
2010, [20]
121
1217
The Distribution TRX-Power Flow Method
Paulo M. De Oliveira-De Jesus, Member, IEEE
Abstract
This paper proposes a new methodology to assess the power ow solution of distribution networks
by merging the classic Backward/Forward Sweep process in a unique state-of-the-system calculation
formula based on a global matrix denominated TRX. The TRX matrix is suitable to be stored in
memory in the context of Distribution Management Systems (DMS) environment and data exchange
framework supported in standard common information model. The TRX matrix reects the most
credible state of the network topology. The proposal is suitable to be applied in balanced and unbal-
anced aerial distribution systems, radially or weakly meshed operated with distributed generation.
This contribution is meaningful under real-time distribution system assessment and planning pur-
poses, since the state of the system is reached using present or historical system measurements as
well as present or future network topology arrangement. Proposed methodology has been applied in
a group of test systems showing better performances than other large-scale implementations like the
standard Newton Raphson with sparse matrix handling
Index Terms
Backward/forward sweep, load ow, power ow, distribution system analysis
I. Introduction
D
ISTRIBUTION system operation and planning require robust and reliable power ow analysis
techniques. Modern distribution management systems (DMS) need this support in order to
perform applications as distributed generation dispatch, service restoration, feeder reconguration,
phase balancing, volt/var control, optimal location of capacitors, etc. For a long time, traditional
and ecient methods as Newton Raphson (NR) [1] and fast decoupled power ow [2] have been
successfully applied in large power systems. Concerning distribution systems, early attempts to
develop power ow studies can be found in [3] [12].
Special features of distribution systems as radial or weakly meshed structure, low ratio between
reactance and resistance values, unbalancing and distributed generation have undesired results in
both performance and robustness of traditional power ow techniques. Weakly meshed distribution
networks may have thousands of busbars being considered ill-conditioned by causing numerical
problems for the conventional power ow algorithms [4], [5]. The degree of ill-conditioning is
evaluated by the number of iterations or the value maximum eigenvalues of the inverse Jacobian
matrix. These deciencies lead to the development of alternative power ow techniques specially
designed for distribution systems.
Several methods have been reported in Literature to solve distribution systems. They can be
divided into three categories: Jacobian-based methods, direct methods, and backward/forward
(BW/FW) sweep methods. The rst type of methods is based on modication of existing methods
such as Newton-Raphson [6], [7], [8], [9] where no node ordering is required.
Direct methods [10], [11] require the construction of an impedance matrix. These methods
usually present a heavy computational burden requiring a specic numbering scheme for nodes
and branches.
Finally, BW/FW sweep algorithms have been developed based on the assumption that nodes and
network branches are properly ordered. In 1967, Berg presented a paper which can be considered
P. M. De Oliveira is with the Department of Electrical Engineering, Simon Bolivar University, Caracas, Venezuela, Ap.89000.
Phone: +58 212 9063913, e-mail: pdeoliveira@usb.ve
1218 THE DISTRIBUTION TRX-POWER FLOW METHOD
as the source for the all variants of BW/FW sweep methods [12]. Later, a similar approach was
presented in [13] based on ladder network theory. In general, these algorithms can use the Kirch-
ho laws [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24], [26], [25], [27] or the
bi-quadratic equation [28], [29], [30], [31] in the iterative process. BW/FW sweep methods typi-
cally present a slow convergence rate but computationally ecient at each iteration. Using these
methods, power ow solution for a distribution network can be obtained without solving any set of
simultaneous equations.
The eectiveness of the BW/FW sweep algorithm has already been proven by comparing it to
the traditional NR methods [32] [34]. Recently, some comparative and convergence studies have
been presented by [33] [35].
All contributions mentioned above have been valuable tools to perform power distribution anal-
ysis. However, under modern DMS environment new considerations must be raised up about the
appropriateness of a given distribution power ow methodologies from system implementation point
of view:
1. The system data model: topology structure and nodal measurements can be stored in mem-
ory of DMS environment reecting the most credible state of the network at given time.
Standardization processes based on Extended Markup Language (XML) as Common Infor-
mation Model [36] and Open Data Model (ODM) [37], [46] have been carried out. In this
sense, power ow analysis in real time or o-line studies can be applied using the most
realistic data (present and historical) stored in memory or media with fast transfer rates.
Then, performance and robustness assessment of power ow methods implemented under
DMS must be made considering the impact of I/O data access.
2. Under DMS environment, power ow applications are object oriented [39] suitable to be
applied with distributed processing [38].
In this context, this paper proposes a new methodology to assess the power ow solution of
distribution networks by merging the standard BW/FW sweep steps into one unique state-of-
the-system calculation formula based upon a TRX matrix suitable to be stored in memory in the
context of a DMS environment. The TRXmatrix is formed by real numbers with three fundamental
elements: the triangular matrix T that relates nodal currents with branch currents, resistance R
and reactance X vectors that characterize the branches, lines or transformers, of the distribution
network. The T matrix and R and X vectors reect the present condition of the network being
constructed directly from the data exchange scheme adopted.
The proposal is simple and suitable to be applied in distribution systems, radially or weakly
meshed operated with distributed generation. The power ow is iteratively solved using only one
state-of-the system calculation step. This contribution is meaningful for actual assessment of the
distribution system and expansion system planning process, since the state of the system is reached
through multiplication and summation operations where no matrix inverse is required
Proposed methodology has been applied in a simple 4-node network for illustration purposes and
compared with a robust Newton Raphson solver [44] and a standard Backward/Forward Sweep
algorithm [24] in three test systems: a 12-bus [29], 33-bus [40] and 69-bus [41] networks. Finally,
a convergence study is performed using a uniformly loaded distribution network from 1000 to 3000
nodes.
This paper is organized as follows: Section II describes the standard BW/FW sweep method under
an appropriate notation for DMS implementation. Section III presents the proposed methodology
for balanced and unbalanced networks. Case studies are discussed in Section IV. Conclusions are
drawn in Section V. Nomenclature, list of symbols are provided in Appendix A. Proposed algorithm
THE DISTRIBUTION TRX-POWER FLOW METHOD 1219
Fig. 1. Branch and node numbering of a radial distribution network
coded in Matlab is presented in Appendix B.
II. Standard Backward/Forward Sweep Power Flow Method in Complex Form
This method was widely described by several authors [14], [24]. The input data of this algorithm
is given by node-branch oriented data used by most utilities. Basic data required is: nodal powers
and sending and receiving nodes of a given line impedance.
In the following, the standard BW/FW sweep power ow method is written in matricial notation
using complex variables. Branch impedances are stated as a vector Z corresponding to a distribution
line model containing a series impedance or transformer. Shunt impedances are not considered in
this rst approach. Fig. 1 shows a radial distribution network with n +1 nodes, and n branches
and a single voltage source at the root node 0. Branches are organized according to an appropriate
numbering scheme (list), which details are provided in [14].
Z =

Z
01
... Z
ij
... Z
mn

(1)
where,
Z
ij
= R
ij
+ jX
ij
i, j = 1, ..., n i = j (2)
Bus data is given by
S =

S
1
.
.
.
S
i
.
.
.
S
n

P
1
+jQ
1
.
.
.
P
i
+jQ
i
.
.
.
P
n
+jQ
n

(3)
where net nodal active and reactive powers are given by generated and demanded powers:
P
i
= P
Gi
P
Di
(4)
Q
i
= Q
Gi
Q
Di
(5)
The numbering of branches in one layer begins only after all the branches in the previous layer
have been numbered. Considering that initial voltages are known: voltage at substation V
k=0
and
an initial voltage vector given by:
V
0
=

V
0
1
... V
0
i
... V
0
n

(6)
The state of the system is reached solving two steps iteratively.
1220 THE DISTRIBUTION TRX-POWER FLOW METHOD
A. Step 1 - Backward Sweep
For each iteration k, branch currents are aggregated from loads to origin:
J
k
= T I
k
(7)
The relationship between nodal currents I
k
and branch currents J
k
is set through an upper
triangular matrix T accomplishing the Kirchho Current Laws (KCL) [18]. Each element I
k
i
of I
k
associated to node i is calculated as function of injected powers S
i
and its voltage prole V
k
i
as
shown below:
I
k
i
=
S

i
V
k
i
i = 1, ..., n (8)
B. Step 2 - Forward Sweep
Nodal voltage vector V is updated from the origin to loads according the Kirchho Voltage Laws
(KVL), using previously calculated branch currents vector J 7 branch impedances vector Z:
V
k+1
= V
0
T
T
D
Z
J
k
(9)
where V
0
is a n-elements vector with all entries set at voltage at origin (swing node) V
0
and
branch impedances D
Z
is the diagonal matrix of vector Z:
C. Convergence
Updated voltages are compared with previous voltages in order to perform convergence check in.
|V
k+1
i
V
k
i
| i = 1, ..., n (10)
III. The Proposed Methodology
This section describes the proposed power ow approach. Two versions are developed in detail:
a balanced and a three phase unbalanced power ow analysis. In both cases, the general algorithm
owchart is depicted in Fig. 2.
A. Balanced TRX Power Flow Algorithm
Similar to standard model presented in Section II, the proposed balanced power ow analysis
requires the apparent power injections vector S and distribution line or transformer model based on
series impedances provided from the data exchange model Z =R+jX. Given an initial condition
(k = 0), the nodal voltage n-elements vector V
0
are decomposed into real and imaginary parts:
V
0
x
=

V
0
x1
... V
0
xi
... V
0
xn

(11)
V
0
y
=

V
0
y1
... V
0
yi
... V
0
yn

(12)
Reference voltages V
0
x
and V
0
y
can be acquired from measurements of DMS support. Alternatively,
it can be used V
0
xi
= 1 and V
0
yi
= 0 for i = 1, ..., n.
The state of the system is updated at each iteration k by the following expression:
V
k+1
x
+jV
k+1
y
= V
0
x
+jV
0
y
T
T
(D
R
+jD
X
) (J
k
x
+jJ
k
y
) (13)
THE DISTRIBUTION TRX-POWER FLOW METHOD 1221
Fig. 2. TRX Load Method Flowchart
where,
D
R
= e{D
Z
} (14)
D
X
= Im{D
Z
} (15)
J
x
= T I
x
(16)
J
y
= T I
y
(17)
This expression correspond to Kirchho Voltage Laws (KVL) through all nodes and, where J
x
+
jJ
y
term corresponds to all nodal Kirchho Current Laws (KCL). After some algebra, real and
imaginary parts of 13 can be rewritten as:
V
0
x
= V
0
x
T
T
D
R
T I
x
+T
T
D
X
T I
y
(18)
V
0
y
= V
0
y
T
T
D
X
T I
x
T
T
D
R
T I
y
(19)
Eqs. 18 and 19 are settled in matricial form as:

V
k+1
x
V
k+1
y

Vx0
Vy0

T
T
DRT T
T
DXT
T
T
DXT T
T
DRT

I
k
x
I
k
y

(20)
where the proposed TRX matrix is given by:
TRX =

T
T
D
R
T T
T
D
X
T
T
T
D
X
T T
T
D
R
T

(21)
The dimension of the TRX matrix for balanced systems is 2nx2n. This matrix is not sparse but
suitable to be allocated using storage devices with high data transfer rate. However, it must be
pointed out that only T
T
D
R
T and T
T
D
X
T matrices should be allocated due to symmetry in the
TRX matrix. For typical, 1000 node distribution circuits, the TRX matrix will require al least
6MB using double-precision numbers
A general expression that relates each nodal voltage with the origin can be written as:
1222 THE DISTRIBUTION TRX-POWER FLOW METHOD
V = V
0
TRX I (22)
Note that, at each iteration, voltage drops V= V
0
V are directly obtained multiplying the
TRX matrix and the nodal current vector I.
The nodal currents I
k
xi
and I
k
yi
for i = 1, ..., n are given as function of present nodal power injections
(demands, capacitors or distributed generators) and operational voltages:
I
k
xi
= e{I
k
i
} =
P
i
V
k
xi
Q
i
V
k
yi
(V
k
xi
)
2
+(V
k
yi
)
2
(23)
I
k
yi
= Im{I
k
i
} =
Q
i
V
k
xi
P
i
V
k
yi
(V
k
xi
)
2
+(V
k
yi
)
2
(24)
For simplicity, this contribution only refers constant power P Q load models. Formulae for
constant current and constant impedance models will be provided in the future. In addition,
this model is suitable to be extended to weakly meshed networks using the compensation method
described in [14].
Convergence check: recent updated voltages are compared with previous voltages in all nodes in
order to perform convergence check in.
|V
k+1
i
V
k
i
| i = 1, ..., n (25)
B. Unbalanced TRX Power Flow Algorithm
Under the unbalanced approach, nodal power injections vector S are given per phase.
S =

S
p1
.
.
.
S
pi
.
.
.
S
pn

P
p1
+jQ
p1
.
.
.
P
pi
+jQ
pi
.
.
.
P
pn
+jQ
pn

p = a, b, c (26)
Branch impedances are given as a rectangular 3nx3 phase impedance matrix Z
abc
Z
abc
=

Z
01
... Z
ij
... Z
mn

(27)
where Z
ij
is the 3-phase matrix impedance corresponding to ij line section [42]:
Z
ij
=

Z
aaij
Z
abij
Z
acij
Z
baij
Z
bbij
Z
bcij
Z
caij
Z
cbij
Z
ccij

(28)
Given an initial condition (k = 0), nodal voltage 3xn-elements vector V
0
abc
are decomposed into
real and imaginary parts:
V
0
abcx
=

V
0
ax1
... V
0
axn
V
0
bxn
V
0
cxn

(29)
V
0
abcy
=

V
0
ay1
... V
0
ayn
V
0
byn
V
0
cyn

(30)
THE DISTRIBUTION TRX-POWER FLOW METHOD 1223
Initial values of three phase nodal apparent power injections vector S and three-phase voltage
proles are online scanned or estimated using DMS.
The state of the system is an n-elements voltage vector that is updated at each iteration k by
the following formula:

V
k+1
abcx
V
k+1
abcy

V
abcx0
V
abcy0

A B
B A

I
k
abcx
I
k
abcy

(31)
where;
A = T
abc
T
D
abc
R
T
abc
(32)
B = T
abc
T
D
abc
X
T
abc
(33)
T
abc
is the upper triangular matrix, with dimension 3nx3n, according current Kirchho laws
considering three phase connection. A general expression can be written as:
V
abc
= V
abc0
TRX
abc
I
abc
(34)
Note that, at each iteration, voltage drops per phase V
abc
=V
abc0
V
abc
are directly obtained
multiplying the TRX
abc
matrix and the nodal current vector I
abc
.
The dimension of the TRX
abc
matrix for unbalanced systems is 6nx6n. For instance, the cor-
responding matrix for a 1000 node-distribution circuits will require al least 56MB using double-
precision numbers.
Three phase currents I
k
pxi
and I
k
pyi
for i = 1, ..., n are given as function of present three phase
power injections (demands, capacitors or distributed generators) and three phase operational volt-
ages:
I
k
pxi
= e{I
k
pi
} =
P
pi
V
k
pxi
Q
pi
V
k
pyi
(V
k
pxi
)
2
+(V
k
pyi
)
2
(35)
I
k
pyi
= Im{I
k
pi
} =
Q
pi
V
k
pxi
P
pi
V
k
pyi
(V
k
pxi
)
2
+(V
k
pyi
)
2
(36)
Convergence check: recent updated voltages are compared with previous voltages in all nodes in
order to perform convergence check in.
|V
k+1
pi
V
k
pi
| i = 1, ..., n p = a, b, c (37)
IV. Testing
The proposed methodology was applied to a list of test systems:
A 4-node illustrative example
Comparison Analysis: a 12, 33 and 69 node network
Comparison Analysis: a uniformly distributed test system from 1000 to 3000 nodes
1224 THE DISTRIBUTION TRX-POWER FLOW METHOD
A. Illustrative Example: Simply 4-node Network
To illustrate the proposed balanced and unbalanced power ow methodology, it is used a 4-node
example shown in Fig. 3. An overhead 12.47kV three-phase distribution line is constructed as shown
in Fig. 4. Length of all sections is 1 mile. The phase conductors are 336,400 26/7 ACSR (Linnet),
and the neutral conductor is 4/0 6/1 ACSR. 336,400 26/7 ACSR: GMR = 0.0244 ft Resistance 0.306
/mile 4/0 6/1 ACSR: GMR = 0.00814 ft. Resistance = 0.5920 /mile. Load demand at nodes 2
and 3 are 2MW with cos = 1.0. The phase impedance matrix of the line is computed according [42]:
Zabc =

0.4576 +j1.0780 0.1560 +j0.5017 0.1535 +j0.3849

0.1560 +j0.5017 0.4666 +j1.0482 0.1580 +j0.4236
0.1535 +j0.3849 0.1580 +j0.4236 0.4615 +j1.0651

A.1 Balanced TRX Power Flow

Using the following bases S
B
=10MW and V
B
=12.47kV, data and results are given in per unit.
Loads are 0.2 in nodes 2 and 3. Reference voltage at node 0 is V
0
= 1 +j0 and initial voltages are
set V
0
x
=

1 1 1

and V
0
y
=

0 0 0

Branches are represented by:

Z = R+ jX =

.0296
.0296
.0296

+ j

.0683
.0683
.0683

(38)
Network topology is represented through a 3x3 upper triangular matrix T.
T =

1 1 1
0 1 0
0 0 1

(39)
Then, TRX matrix is built by aggregating T
T
D
R
T and T
T
D
X
T matrices as indicated in 21:
T
T
D
R
T =

.0296 .0296 .0296

.0296 .0592 .0296
.0296 .0296 .0592

(40)
T
T
D
X
T =

.0683 .0697 .0683

.0683 .1367 .0683
.0683 .0683 .1367

(41)
Using 22, solution reached at iteration 3 for = 10
4
and displayed in Table I. Results are
presented in per unit and degrees.
TABLE I
4 Node State of the System - balanced Approach
V
0

0
V
1

1
V
2

2
V
3

3
1.000 0.00 0.987 -1.59 0.981 -2.40 0.981 -2.40
THE DISTRIBUTION TRX-POWER FLOW METHOD 1225
Fig. 3. 4-Node Network Topology
Fig. 4. 4-Node Network Three-phase distribution line spacing
A.2 Unbalanced TRX Power Flow
Using S
B
=3.33MW and V
B
= 12.47/

3kV. Three phase data is given in p.u.:

S
a
= S
b
= S
c
=

0 0.2 0.2

(42)
Three phase impedance matrix is the same for all line sections:
Z
abc
= R
abc
+ jX
abc
(43)
R
abc
=

0.0294 0.0100 0.0098

0.0100 0.0299 0.0100
0.0098 0.0100 0.0296

(44)
X
abc
=

0.0692 0.0322 0.0247

0.0322 0.0672 0.0322
0.0247 0.0322 0.0683

(45)
As this system is modeled with three phase circuits in all section lines, three phase upper trian-
gular matrix is given by the following expression:
T
abc
=

U U U
0 U 0
0 0 U

where U =

1 0 0
0 1 0
0 0 1

(46)
Using 31 ,three phase TRX
abc
matrix is formed by T
abc
T
D
abc
R
T
abc
and T
abc
T
D
abc
X
T
abc
shown
in Tables II and IV, respectively.
Reference three phase voltages at slack node 0 and initial values in nodes 1,2 and 3 are given by:
1226 THE DISTRIBUTION TRX-POWER FLOW METHOD
TABLE II
T
abc
T
D
abc
R
T
abc
Matrix
.0294 .0100 .0098 .0294 .0100 .0098 .0294 .0100 .0098
.0100 .0299 .0101 .0100 .0299 .0101 .0100 .0299 .0101
.0098 .0101 .0296 .0098 .0101 .0296 .0098 .0101 .0296
.0294 .0100 .0098 .0587 .0200 .0197 .0294 .0100 .0098
.0100 .0299 .0101 .0200 .0599 .0203 .0100 .0299 .0101
.0098 .0101 .0296 .0197 .0203 .0592 .0098 .0101 .0296
.0294 .0100 .0098 .0294 .0100 .0098 .0587 .0200 .0197
.0100 .0299 .0101 .0100 .0299 .0101 .0200 .0599 .0203
.0098 .0101 .0296 .0098 .0101 .0296 .0197 .0203 .0592
TABLE III
T
abc
T
D
abc
X
T
abc
Matrix
.0692 .0322 .0247 .0692 .0322 .0247 .0692 .0322 .0247
.0322 .0672 .0272 .0322 .0672 .0272 .0322 .0672 .0272
.0247 .0272 .0683 .0247 .0272 .0683 .0247 .0272 .0683
.0692 .0322 .0247 .1383 .0644 .0494 .0692 .0322 .0247
.0322 .0672 .0272 .0644 .1345 .0544 .0322 .0672 .0272
.0247 .0272 .0683 .0494 .0544 .1367 .0247 .0272 .0683
.0692 .0322 .0247 .0692 .0322 .0247 .1383 .0644 .0494
.0322 .0672 .0272 .0322 .0672 .0272 .0644 .1345 .0544
.0247 .0272 .0683 .0247 .0272 .0683 .0494 .0544 .1367
V
abc0
= V
0
abc1
= V
0
abc2
= V
0
abc3
=

0
1

120
o
1

120
o

(47)
Using 34, solution reached at iteration 3 for = 10
4
and displayed in Table IV. Results are
presented in per unit and degrees.
TABLE IV
4 Node State of the System - balanced Approach
p V
0

0
V
1

1
V
2

2
V
3

3
a 1 0 0.989 -0.95 0.984 -1.43 0.984 -1.43
b 1 -120 0.994 -120.86 0.991 -121.30 0.991 -121.30
c 1 120 0.993 119.02 0.989 118.52 0.989 118.52
These illustrative examples were solved using Microsoft Excel, and spreadsheets can be requested
to the author.
THE DISTRIBUTION TRX-POWER FLOW METHOD 1227
B. Comparative Analysis - Three Test Networks
The proposed methodology was applied in the three (3) distribution test networks widely used
in literature (12-bus [29], 33-bus [40] and 69-bus [41]).
The performance and robustness of the method was assessed under Matlab platform [45] and
compared with two alternative methods. First, the standard BW/FW sweep power ow approach
presented in [24] and, second the Matpowers Newton-Raphson solver [44]. The BW/FW sweep
method has been coded with complex variables using the theoretical basis presented in Section II.
Matpower-NR solver is based on a standard Newtons method [1] using a full Jacobian, updated
at each iteration. Matpower performance is excellent even on very large-scale test cases, since the
algorithms and implementation take advantage of Matlabs built-in sparse matrix handling. In
Appendix B, it is included the the codication of the proposed method. It is important to note
that overall iteration process only require ten programming lines demonstrating the simplicity of
the proposal.
Convergence and robustness of the algorithms are analyzed through the number of iterations
needed to reach a solution and the CPU time spent in the iterative process. Due to small size of
the problems, CPU time registered includes input-output (I/O) access time.
Stop criteria is 10
6
in all nodal voltages. A Macbook Intel Core 2 Duo T8300@2.4GHz with
2GB RAM under OSX Leopard 10.5.5 has been used for all simulations.
Results are presented in Table V
TABLE V
Comparison between TRX, Standard BW/FW sweep and NR Power Flow
Number of Iterations
12-bus 33-bus 69-bus
Complex Back/Forward Sweep 5 6 7
NR (Matpower) 4 4 4
TRX Method 5 6 7
CPU Time (10
3
Seconds)
12-bus 33-bus 69-bus
Complex Back/Forward Sweep 0.78 0.94 4.38
NR (Matpower) 8.17 10.44 12.15
TRX Method 0.64 0.82 1.25
Results show that NR method has better convergence behavior than TRX and Backward/Forward
Sweep. However, despite TRX method requires the same number of iterations than BW/FW sweep
method, it has the better CPU time. Regarding NR and BW/FW sweep methods, these results
conrm the conclusions reported by Eminoglu in broad comparative convergence analysis among
several BW/FW sweep based methods [33]. In this paper, it was used the same test systems in
order to compare with results reported in this study. Proposed TRX method presents the slowest
convergence rate but with the highest computationally eciency. Note that in these small examples
CPU time corresponding to NR results is extremely dierent than other methods. This is due to
required I/O data exchange time in sparse matrix handling.
1228 THE DISTRIBUTION TRX-POWER FLOW METHOD
Fig. 5. The n-Node Network
Fig. 6. CPU time required to convergence
C. Comparative Analysis - Uniformly Loaded Distribution Feeder
The proposed method has also applied in a large-scale test network and its performance compared
with the BW/FW sweep and the Matpowers Newton-Raphson. Same computer conditions than
previous comparison analysis are used. The test case is a 12.47kV n-node radial distribution network
with a load of 4MW uniformly distributed through L=1mile. Each branch is 336,400 26/7 ACSR
(Linnet) conductors with impedance given in pu (S
B
=10MW):
Z
01
= Z
12
= .. = Z
n1n
=
0.0293 + j0.697
n
(48)
To illustrate the impact of the number of nodes in the CPU time required by each power ow
method, line section depicted in Fig. 48 has been divided in n branches connected in radial form
and total load has been fractionated in n nodal loads.
It is well known that analytical solution of this type of system is given by [43]:
%V
0n
=
V
0
V
n
V
0
=
S
3
5V
2
0
(Rcos + X sin ) (49)
where S
3
= 0.4, the load factor is equal to 1. For V
0
= 1, 49 solution is V
n
is 0.9940pu. In order
to verify this solution, an exact power ow solution is obtained using NR for n=1000: at the end
of the feeder, voltage is V
n
=0.99398 and angle
n
=-0.016 radians.
The n-node network has been solved using the proposed TRX method and two alternative meth-
ods (NR, Back/Forward Sweep) varying n parameter from 1000 to 3000 nodes. All simulations lead
to the same solution.
CPU process time results are depicted in Fig. 6. Convergence time spent in the iterative process
does not consider I/O data exchange time.
When I/O data exchange is not considered, and CPU time is only related to the while loop (see
Appendix B), results show that TRX method has better CPU time behavior respect to NR and
THE DISTRIBUTION TRX-POWER FLOW METHOD 1229
BW/FW sweep method. It must be pointed out that BW/FW sweep method developed in [24] is
not direct, as indicated in paper title but iterative like all sweep-based methods. Not considering
I/O data exchange is justied by the fact that under DSM environment, a sentinel program should
maintain all TRX matrices in memory with the most realistic topology.
As a result, the proposed TRX method presents the slower convergence rate but with the higher
computationally eciency. The overall process is based on summation and multiplication of xed
real numbers numbers previously allocated in memory being highly competitive respect to NR with
LU factorization and sparse handling. The TRX matrix does not require update at each iteration
like Jacobian in Full NR method. In this case, CPU time corresponding to NR results is the second
best. The worst behavior correspond to the standard BW/FW sweep in complex form which is
extremely aected by operations required by summation and multiplication of complex numbers.
The proposed method is a valuable tool to analyze aerial distribution systems, since they can
be modeled as a single series impedance. Research eorts are focused in the generalization of
the method including shunt impedances associated to cables. Future research is also oriented in
compare the proposed methodology with other distribution-oriented power ow methodologies, in
particular decoupled versions of NR and BW/FW sweep methods based on biquadratic formula.
Ill-conditioned systems will be also considered.
V. Conclusion
This paper proposes a new methodology to assess the power ow solution of distribution networks
by merging the standard Backward/Forward Sweep process in a unique state-of-the-system calcula-
tion formula. The proposal is suitable to be applied in balanced and unbalanced aerial distribution
systems, radially or weakly meshed operated with distributed generation. Proposed methodology
has been applied in test-case systems showing better performances than other large-scale methods.
Appendices
A. Nomencalture
TRX
abc
Three Phase TRX matrix
TRX TRX matrix
D
Z
Diagonal matrix of branch impedance vector Z
D
R
Diagonal matrix of branch resistance vector R
D
Z
Diagonal matrix of branch reactance vector Z
Convergence criteria
I Current vector
I
i
Current at node i
I
x
Real part of Current vector I
I
y
Imaginary part of Current vector I
I
abc
Three Phase Current vector
I
abcx
Real part of Three Phase Current vector I
abc
I
abcy
Imaginary part of Three Phase Current vector I
abc
J Branch Current vector I
J
ij
Branch current between node i and node j
J
x
Real part of Branch Current vector I
J
y
Imaginary part of Branch Current vector I
n Number of nodes, excluding origin
P Active Power Injected vector
1230 THE DISTRIBUTION TRX-POWER FLOW METHOD
Q Reactive Power Injected vector
P
i
Active Power Injected at node i
Q
i
Reactive Power Injected at node i
P
Di
Active Power Demanded at node i
Q
Di
Reactive Power Demanded at node i
P
Gi
Active Power Generated at node i
Q
Gi
Active Power Generated at node i
R Branch resistance vector
R
abc
Three Phase Branch resistance vector
R
ij
Resistance between node i and node j
S Apparent Power Injected vector
S
abc
Three Phase Apparent Power Injected vector
S
i
Apparent Power Injected at node i
S
Di
Apparent Power Demanded at node i
S
Gi
Apparent Power Generated at node i
T Triangular matrix
T
abc
Three Phase Triangular matrix
V Voltage vector
V
i
Voltage at node i
V
x
Real part of Voltage vector V
V
y
Imaginary part of Voltage vector V
V
abc
Three Phase Voltage vector
V
abcx
Real part of Three Phase Voltage vector V
abc
V
abcy
Imaginary part of Three Phase Voltage vector V
abc
X Branch reactance vector
X
abc
Three Phase Branch reactance vector
X
ij
Reactance between node i and node j
Z Branch Impedance vector Z
Z
abc
Three Phase Branch Impedance matrix Z
Z
ij
Branch Impedance between node i and node j
Z
ij
Impedance matrix between node i and node j
Z
aaij
Self impedance phase a between node i and node j
Z
abij
Mutual impedance phase ab between node i and node j
B. TRX Algorithm Coded in MATLAB
%-----Calculations Under DSM environment-------------
T %Triangular matrix (topology) is known
r; x; %List of branch impedances are known
P; Q; % List of Active and reactive powers are known
Vo; %Slack voltage is known
A=T*diag(r)*T; B=T*diag(x)*T;
TRX=vertcat(horzcat(A,-B),horzcat(B,A));
%-----TRX Power Flow Analysis------------------------
Vx=[ones];Vy=[zeros];V=vercat(Vx,Vy)%Initial voltage profile
starttime=cputime;
while max(abs(delta))> 0.0001 %Stop Criteria
for j=1:n %Number of line sections and nodes
I(j)=(-P(j)*V(j)-Q(j)*V(j+n))/(V(j)^2+V(j+n)^2);
I(j+n)=(Q(j)*V(j)-P(j)*V(j+n))/(V(j)^2+V(j+n)^2); end
V2= Vo-TRX*I.; %Voltage Update
delta=V-V2; %Voltage MISMATCH
V=V2;
THE DISTRIBUTION TRX-POWER FLOW METHOD 1231
end %end while
processtime=cputime-starttime;
References
[1] W. F. Tinney and C. E. Hart, Power Flow Solution by Newtons Method, IEEE Transactions
on Power Apparatus and Systems, Vol. PAS-86, No. 11, pp. 1449-1460, Nov. 1967
[2] B. Stott and O. Alsac, Fast decoupled load ow, IEEE Transactions on Power Apparatus and
Systems Vol. PAS-93, No. 3, pp. 859-869, 1974
[3] R.D. Hoestenbach, Load Flow Analysis of Electrical Distribution Systems Using the Digital
Computer, IEEE Transactions on Industry Applications, Volume IA-12, No. 2, pp. 152 - 160,
March 1976
[4] S.C. Tripathy, G.D. Prasad, O.P. Malik, O. P. and G.S. Hope, Load ow solutions for illcondi-
tioned power systems by a Newton-like method, IEEE Transactions on Power Apparatus and
Systems, Vol. PAS-101, No. 10, pp. 3648-3657, 1982.
[5] M.M.M. El-Arini, Decoupled power ow solution method for well-conditioned and ill-
conditioned power systems, IEE Proceedings Generation, Transmission and Distribution, Vol.
140, No.1,pp.7-10, Jan. 1993
[6] H. Chiang, and Hsiao, A decoupled load ow method for distribution power networks: algo-
rithms, analysis and convergence study, Electrical Power & Energy Systems, Vol.13, No.3, pp.
130-138, 1991
[7] R. D. Zimmerman and H. D. Chiang, Fast decoupled power ow for unbalanced radial distri-
bution systems, IEEE Transactions on Power Systems, vol. 10, pp. 2045-2052, Nov. 1995.
[8] F. Zhang and C. Cheng, A modied newton method for radial distribution system power ow
analysis, IEEE Transactions on Power Systems vol. 12, pp. 389-397, 1997
[9] P. Aravindhababu, R. Ashokkumar A Fast Decoupled Power Flow for Distribution Systems,
Electric Power Components and Systems, Vol. 36, No.9, pp. 932-940, September 2008
[10] S.K. Goswami and S.K. Basu, Direct solution of distribution systems IEE Proceedings C
Generation, Transmission and Distribution Vol. 138, No. 1 pp.78-88, Jan 1991
[11] T.Chen K. Hwang, P. Kotas and E. Chebli, Distribution system power ow analysis - a rigid
approach, IEEE Transactions on Power Delivery vol. 6, pp 1146-1152, 1991
[12] R. Berg, E.S. Hawkins, and W.W. Pleines, Mechanized Calculation of Unbalanced Load Flow
on Radial Distribution Circuits, IEEE Transactions on Power Apparatus and Systems, Volume
PAS-86, No. 4, pp.415-421, Apr 1967
[13] W. H. Kersting and D. L. Mendive, An application of ladder network theory to the solution
of three phase radial load ow problem, in Proc. IEEE PES Winter Meeting, 1976, New York,
paper A76044-8 (IEEE, New York, 1976).
[14] D. Shirmohammadi, H.W. Hong, A. Semlyen, and G.X. Luo, A compensation-based power
ow method for weakly meshed distribution and transmission networks, IEEE Transactions on
Power Delivery, Vol. 3, No. 2, pp.753-762, 1988.
[15] S. Ghosh and D. Das, Method for load-ow solution of radial distribution networks, IEE
Proc. Generat. Transm. Distrib., Vol. 146, No. 6, pp. 641-648, 1996.
[16] Y Fukuyama, Y Nakanishi, H-D Chiang Fast distribution power ow using multi-processors,
Electrical Power & Energy Systems, Vol. 18, No. 5, pp. 331-337, 1996
[17] D. Thukaram, H.M.W.Banda, and J. Jerome, A robust three-phase power ow algorithm for
radial distribution systems, Electric Power Systems Research, Vol. 50, No. 3, pp. 227-236, 1999.
1232 THE DISTRIBUTION TRX-POWER FLOW METHOD
[18] S. Jovanovic and F. Milicevic, Triangular distribution load ow, IEEE Power Engineering
Review, pp. 60-62, 2000.
[19] D. Rajicic, R. Taleski, Two novel methods for radial and weakly meshed network analysis,
Electric Power Systems Research, Vol. 48, No.2, pp. 79-87, Dec. 1998
[20] Aravindhabuba, P., Ganapathy, S., and Nayar, K. R., A novel technique for the analysis of
radial distribution systems, Electric Power Systems Research, Vol. 23, No. 3, pp. 167-171, 2001.
[21] Y. Zhu and K. Tomsovic, Adaptive power ow method for distribution systems with dispersed
generation, IEEE Transactions on Power Delivery, Vol. 17, No. 3, pp. 822-827, 2002.
[22] J. Liu, M.M.A. Salama and R.R. Mansour, An ecient power ow algorithm for distribution
systems with polynomial load, Int. J. Elect. Eng. Educat., Vol. 39, No. 4, pp. 371-386, 2002.
[23] M. Afsari, S. P. Singh G. S. Raju G. K. Rao A Fast Power Flow Solution of Radial Distribution
Networks, Electric Power Components and Systems, 30:1065-1074, 2002
[24] J.H. Teng,A direct approach for distribution system load ow solutions, IEEE Transactions
on Power Delivery, Vol. 18, No. 3, pp. 882-887, 2003.
[25] K, Prasad, N.C. Sahoo, A. Chaturvedi and R. Ranjan. A simple approach for branch current
computation in load ow analysis of radial distribution systems, Int. J. Elect. Eng. Educat. Vol.
44 No. 1 pp. 49-63, Jan. 2007
[26] G.W. Chang, S.Y. Chu and H.L. Wang An Improved Backward/Forward Sweep Load Flow
Algorithm for Radial Distribution Systems, IEEE Transactions on Power Systems, Vol. 22, No.
2, pp.882-884, May 2007
[27] W.C. Wu and B.M. Zhang, A three-phase power ow algorithm for distribution system power
ow based on loop-analysis method, Electrical Power & Energy Systems, pp. 8-15, No. 30 (2008)
[28] R. Cespedes, New method for the analysis of distribution networks, IEEE Transactions on
Power Delivery, Vol. 5, No. 1, pp. 391-396, 1990
[29] D. Das,H.S. Nagi and D.P. Kothari, Novel method for solving radial distribution networks,
IEE Proc. Generat. Transm. Distrib., Vol. 141, No. 4, pp. 291-298, 1994.
[30] A. Gomez Exposito and E. Romero Reliable Load Flow Technique for Radial Distribution
Network, IEEE Transactions on Power Systems, Vol. 14. 1999, pp 1063-1069.
[31] U. Eminoglu and M.H. Hocaoglu, A new power ow method for radial distribution systems
including voltage dependent load models, Electric Power Systems Research, Vol. 76, No. 1-3,
pp. 106-114, 2005
[32] L.R.Araujo, D.R.R. Penido, S.Carneiro, J.L.R.Pereira and P.A.N. Garcia, A Comparative
Study on the Performance of TCIM Full Newton versus Backward-Forward Power Flow Methods
for Large Distribution Systems, IEEE Power Systems Conference and Exposition, 2006 pp.522-
526, Nov. 2006
[33] U. Eminoglu amd M.H. Hocaoglu Distribution Systems Forward/Backward Sweep-based
Power Flow Algorithms: A Review and Comparison Study, Electric Power Components and
Systems Vol. 37 No.1 pp.91-110, 2009
[34] C.E.Lin, Y.W. Huang and C.L. Huang. Distribution system load-ow calculation with micro-
computer implementation,. Electric Power Systems Research,13,pp.139-45,1987
[35] E. Bompard, E. Carpaneto, G. Chicco, R. Napoli Convergence of the backward/forward sweep
method for the load-ow analysis of radial distribution systems, Electrical Power & Energy
Systems No. 22 pp. 521-530, 2000
[36] X. Wang, N. N. Schulz, and S. Neumann, CIM extensions to electrical distribution and CIM
XML for the IEEE radial test feeders, IEEE Transactions on Power Systems, vol. 18, pp.
1021-1028, Aug. 2003.
THE DISTRIBUTION TRX-POWER FLOW METHOD 1233
[37] F. Milano, M. Zhou, and H. Guan Ji, Open model for exchanging power system data, IEEE
Power & Energy Society General Meeting, 2009. PES 09. pp.1-7 26-30 July 2009
[38] M. Kleinberg,K. Miu, K.and C. Nwankpa, A Study of Distribution Power Flow Analysis Us-
ing Physically Distributed Processors, IEEE Power Systems Conference and Exposition, 2006,
pp.518-521, Nov.2006
[39] A. Losi and M. Russo Object-oriented load ow for radial and weakly meshed distribution
networks, IEEE Transactions on Power Systems Vol. 18, No. 4, pp.1265-1274, Nov. 2003
[40] M.E. Baran and F.F. Wu, Optimal sizing of capacitors placed on a radial distribution system,
IEEE Transactions on Power Delivery, Vol. 4, No. 1, pp. 735-742, 1989
[41] M.E. Baran and F.F. Wu, Network reconguration in distribution systems for loss reduction
and load balancing, IEEE Transactions on Power Delivery, Vol. 4, No. 2, pp. 1401-1407, 1989.
[42] W.H. Kersting, Distribution System Modeling and Analysis, Boca Raton, FL: CRC Press, 2002.
[43] T. Gonen, Electric Power Distribution System Engineering, McGraw-Hill Book Company, 1986
[44] MATPOWER 3.2, Cornell University, [Online]. Available: http://www.pserc.cornell.edu/matpower
[45] Mathworks, Matlab 7.2 Users manual 2006
[46] IEEE Task Force on Open Source Sofware fo Power Systems [Online]. Available:
http://ewh.ieee.org/cmte/psace/CAMS taskforce/format.htm
Apndice I
OPTIMAL SIZING OF CAPACITORS PLACED ON RADIAL
DISTRIBUTION SYSTEM, BARAN, M. E. ET AL. 1989, [31]
139
IEEE Transactions on Power Delivery, Vol. 4, No. 1, J anuary 1989
OPTIMAL CAPACITOR PLACEMENT ON RADIAL DISTRIBUTION SYSTEMS
725
Mesut E. Baran Felix F. Wu
Department of Electrical Engineering and Computer Sciences
University of Califomia, Berkeley
Berkeley, CA 94720
Abstract - Capacitor placement problemon radial distribution systems is
formulated and a solution algorithmis proposed. The location, type, and
size of capacitors, voltageconstraints, and load variations are considered
in theprobIem. The objectiveof capacitor placement is peak power and
energy loss reduction by taking into account the cost of capacitors. The
problemis formulated as a mixed integer programming problem. The
power flows in thesystemare explicitly represented and thevoltagecon-
straints are incorporated. The proposed solution methodology decom-
poses theprobleminto a master problemand a slaveproblem. The mas-
ter problemis used to determine the location of the capacitors. The
slaveproblemis used by the master problemto determinethe type and
size of the capacitors placed on the system. In solving theslaveprob-
lem, an efficient phaseI - phaseII algorithmis used. Proposed solution
methodology has been implemented and the test results are included in
this paper.
I. INTRODUCTION
The general capacitor placement problemconsists of determining
thelocation, type, and the sizeof capacitors to beinstalled in thenodes
of a radial distribution systemsuch that the economic benefits due to
peak power and energy loss reduction be weighted against the cost of
installment of such capacitors while keeping the voltage profile of the
systemwithin defined limits.
The optimal capacitor placement problem as defined above has
many parameters, such as; thelocation, type, and cost of capacitors, vol-
tage constraints, and load variations on the system. These parameters
determinethecomplexity of theproblem.
Conventionally, the problemhas been formulated by using a vol-
tage independent reactivecurrent model and solved by fixing some of
the parameters and using analytical methods [l-31. Recently there has
been some studies to solvethe problemin its general form. There are
basically three approaches. The first one is the dynamic programming
type approach by treating the sizes of capacitors as discrete variables,
[4-51. The second approach is to combine the conventional analytical
methods with heuristics 16-71. Third approach, pioneered by Grainger
et. al., is to formulatetheproblemas a nonlinear progmming problem
by treating the capacitor sizes and the locations as continuous variables
[8-131. The application of this approach to general problemwith the
voltageregulator problemis given in [ 131.
In this paper, a formulation for the general capacitor placement
problem as a mixed integer programming problemwill be given first.
The formulation considers all theparameters of theproblemstated above
and also thevoltageconstraints. Furthermore, theac power flow formu-
lation is used to represent thepower flows and thevoltageprofilein the
radial distribution systems.
A solution methodology for thegeneral problemis proposed in this
88 WM 064-8 A paper recommended and approved
by the IEEE Transmission and Distribution Committee
of the IEEE Power Engineering Society for presentat-
ion at the IEEE/PES 1988 Winter Meeting, New York,
New York, January 31 - February 5, 1988. Manuscript
submitted September 1, 1987; made available for
printing November 13, 1987.
paper. The solution is based on the decomposition of theprobleminto
hierarchical levels. The problemat thetop level, called themuster prob-
lem, is an integer programming problemand is used to place the capaci-
tors ( i.e. U) determinethenumber and thelocation of the capacitors).
A search algorithmhas been developed for the master problem. The
problemat the bottomlevel is called theslave problem and is used by
themaster problemto determinethetypes and thesettings of thecapaci-
tors placed. Decomposition schemes are also used to further decompose
the slave problems into base problems. Base problems are solved by
using the efficient solution algorithmdeveloped for a special capacitor
placement problemcalled the sizing problem. The sizing problemand
theassociated solution algorithmarepresented in another paper [141.
This paper consists of seven sections. In section 11, the general
capacitor placement problemis formulated and its complexity is dis-
cussed. h section 111, it is shown that theproblemcan bedecomposed
into a muster problem and a slave problem. Section IV and V are
devoted to the development of solution methodologies for theslaveand
master problems respectively. Section VI contains thetest studies of the
solution method applied to two different systems. Conclusions aregiven
in section VII.
IL. FORMULATION OF THE PROBLEM
We consider thegeneral capacitor placement problemas determin-
ing theplaces (number and location), types, and settings (capacities) of
the capacitors to be placed on a radial distribution system. Theobjec-
tives are to reducethe power and energy losses on the systemand to
maintain thevoltageregulation whilekeeping thecost of capacitor addi-
tion at a minimum.
Since weareinterested in energy loss in thesystem, it is necessary
to take into account the load variations for a given period of time, T.
We assumethat the load variations can beapproximated in discretelev-
els. Furthermore, the loads are assumed to vary in a conforming way
(i.e., all theloads enjoy the samepattemof variations). Welet S(T) be
thecommon Loud Duration Curve as shown in Fig.1. Then a load, say
load QL , can berepresented as
Where, Q2 represents the peak value.
QL@) =Q ~ W ) (1)
S"t
: F+>; ~
y 4;" qT I-
Figure 1 : Load Duration Curve
Under theseassumptions, thetimeperiod, T can bedivided into intervals
during which the load profileof the systemis assumed to be constant.
Let therebent such loud levels (loud projiles) .
Then for each load level, we have: (i) power flow equations, (ii)
voltage constraints as bounds on themagnitudeof thesystemnodevol-
tages. (iii) capacity and control constraints on the control variables
(capacitors).
We will represent the constraints imposed by power flows on a
radial distribution systemby a new set of ac power flow equations,
called DistFkw equations. They substitute for the conventional ac
power flow equations. To summarizetheidea, consider a 3-0, balanced
radial distribution feeder with n branches/nodes, I laterals, andnc shunt
0885-8977/89/01oo-O725$01 .WO 1989 IEEE
Authorized licensed use limited to: UNIVERSIDADE DO PORTO. Downloaded on October 28, 2009 at 23:22 from IEEE Xplore. Restrictions apply.
726
capacitors placed at the nodes of the system. In Fig.2, one-line diagram
of such a network is shown.
Fig.2 : Oneline diagramof a Distribution Feeder
It can be shown that power flow through each branch in the lateral can
bedescribed by thefollowing recursive equations.
Where,
Pk ,Qk : real and reactive power flows into the receiving end of
branch k+l connecting nodek and nodek+l,
v k : bus voltage magnitude at node k ,
Qck : reactive power injection fromcapacitor at node k.
Eq.(2), called the brnnchjbw equation, has the following form
4 + 1 =f k+l (xk &k +d (3.i)
Note that if there is no capacitor at node k, then uk does not appear in
Eq(3.i). By abusing notation, wewill simply use U as an nc dimensional
vector containing the nc capacitors i.e.,
Where, Xk =[ p k Qk v?lT and4 + 1 = e&+, .
UT =[U1 . . . U,] =[ Qc n, . . . Q-]
In addition to the branch flow equations of (3.i). there are terminal
conditions to be satisfied for each lateral (counting the main feeder as the
O'th lateral). For example, for lateral k shown in the figure, wehave the
following terminal conditions:
(i) at the branching node k where the lateral is connected to the main
feeder, wedefine a dummy variable VkOand let
XkO, =v& =v l =XOk, (3.ii)
(ii) at the end of the lateral, there is no power sent to the other branches,
i.e.,
x h , = P h = O ; x h r = Q h = O (3.iii)
Hence, for the general feeder considered, there are 3(n+l+1) Dist-
Flow equations corresponding to Eq.(2) and Eq.(3). They will be
represented by thefollowing equations.
Where, x = [ x f . . . x [ $l T , xk =[ x & . . . x ; l T.
DistFlow equations can be used to determine the operating point, x
of the systemfor a given load profile, PL; , Q L ~ i =1 , . . . , n and the
capacitor settings, U . Weprefer to use DistFlow equations over conven-
tional ac power flow equations because the special structure of the Dist-
Flow equations can be utilized to develop a computationally efficient and
numerically rbbust solution algorithm. The details of such a solution algo-
rithm are presented in [ 141.
For the capacitor placement problemsince there are nt different load
profiles to beconsidered, the overall DistFlow equations are
G(x,u) =0 (4)
Gi ( $, ui ) =O i =0,1...nr ( 5)
Where, xi , ui represent the state and the contml variables corresponding
to the load profiie i respctively.
The voltage constraints can be taken into account by specifying
upper and lower bounds on themagnitude of the nodevoltages as follows,
v m2< v; ' =Y; ' (x' ) 1v-2 j =l . . . n i - - 0, 1. . . nt (6)
These bounds constitute a set of functional inequality constraints of the
form,
H'(x')IO i =0, 1, . . . .nt (7)
We will consider two different types of capacitors and represent
themas follows:
i) Fixed Capacitors : They will be treated as reactive power sources with
the constant magnitude at al load levels, i.e..
" o =u l = ... = u M (8.i)
ii) Switched Capacitors: It will be assumed that thes et t i n g s (capacities)
of a switched capacitor, U: can bechanged/contmlled at every load level
qnsidered. Therefore, for each capacitor, there are nt+l settings,
U: i =0,l. . . . , nt to be determined. Wewill also assume that thesetting
of a capacitor for the peak load, U: will be bigger than theones for other
lower load levels U:. Hence, the sizes (nominal capacities) of capacitols
will be determined by uo and the relationship between thesize and and the
settings of a capacitor will beas follows.
O I U i S U , o (8.ii)
The objective terms. namely the real power and energy loss and the
capacitor cost. can be formulated ,as follows. For each load level i , let the
power loss in the systembepi (x'). Then thetotal cost of energy loss can
be wrimn as
nl
ke x = T i P i ( X ' ) (9)
i 4
where, Ti is the duration of the load for load level i and the constant k, is
the energy cost per uni t. The cost for thereal power loss at peak load level
can be added to this sum by modifying To accordingly.
The capacitor cost hcti on is usually step like as shown in Fig.3
since in practice capacitors are grouped in banks of standard discrete capa-
cities (usually 300 kvar sizes at 23 kV level).
4)
"0
cvb
Y X
Figure 3 : Capacitor cost functions
Such a function is not easy to handle within t hi s formulation framework;
therefore, it will be approximated by a linear function with a fixed charge
as shown in Fig3 by a dotted line. This function can beformulated by
using a decision variable e E (0.1) as
f ( uo) =c. e +rc. uo OSu o Su".e (10)
Where, uo and r, represent thesize and themarginal cost of the capacitor
respectively. Note that e =0 corresponds to thedecision that thecapacitor
not to beplaced.
To summarize, let the types and places of nc "candidate" capacitors,
initially considered for installment, begiven and let the sets C 1 , C2 con-
tain the switched and the fixed capacitors respectively. Then wecan write
the general capacitor placement problemas a standard optimization prob-
lemas follows.
s.t. G' (xi ,d) =0
H'(x')SO i=O,1, ..., nt
0 I uo I u-.e
0 I U; I U:
U: =U :
k EC1 = {W. cap.)
k E C z =l f i xed c ap . }
Authorized licensed use limited to: UNIVERSIDADE DO PORTO. Downloaded on October 28, 2009 at 23:22 from IEEE Xplore. Restrictions apply.
127
base problems can besolved by the efficient solution method developed
for thesizing problem, PS in [14]. Finally, in the last subsection. wewill
come back to the general slave problemand consider thetypes as well as
the settings of the capacitors.
4.1 Fixed Capacitor Problem
becomes
When all the capacitors to beplaced are of fixed type, theproblem
nl
Pfx mill f , =ke CTiPi(Xi) + C rckuk
i=O k EC2
s.t. Gi ( xi , u) =O i=O,l,. .. ,nt
H'(x') 5 0
OI U I U -
This general formof fixed capacitor problem, Pfx is a base problem
because it is essentially the same as the sizing problem, PS. Here, because
of the consideration of more load levels, the objective function comprises
more power loss terms and the constraint set is bigger due to extra Dist-
Flow equations and the corresponding voltage regulation constraints. But
this does not change the structure of the problemvery much. Therefore,
the solution methodology developed for the sizing problemcan readily be
applied for this problemwith small modification.
Note that for thi s general case, existence of different load levels
makes it possible to have both lower bound voltage violation at peak load
level and upper bound voltage violation at light load level for a given con-
trol U . This is especially true when the load levels are diverse. Such
cases are handled in the Phase I - Phase I1 type solution algorithmof the
sizing problemby considering both the lower andthe upper voltage con-
straints in calculating the search direction (for details, seethe solution
algorithmfor the base problemin [ 141).
This is a non-linear, mixed integer programming problem. Decision
variables, e =[ e, . . . e,IT are to be used to choose the capacitors among
initially designated ones and continuous variables ui , i =0.1.. . . .nt will
be used to determine theoptimal settings of the capacitors.
Voltage regulators are not explicitly represented in the capacitor
placement problempresented here. However, the formulation and the
solution algorithmintroduced in t hi s paper can be generalized to include
the voltage regulators in the following fashion:
Voltage regulators. such as regulating transformers, can be
represented by their equivalent circuits in the DistFlow equations.
The solution algorithmfor sizing problem, proposed in [14] for
determining theoptimal sizes of capacitors placed on the system, can
begeneralized to obtain the optimal settings of thevoltage regulators
placed on thesystem,
Voltage regulation for a given set of capacitors and voltage regula-
tors can beobtained again by using the solution algorithmdeveloped
for the sizing problem. This is demonstrated in I141by using only
the capacitors.
It seems appropriate to put the voltage regulator placement problem
(to find the locations for the voltage regulators) at the top level of
hieravhical decomposition scheme proposed in t hi s paper in solving
the overall problem.
Further investigation is required to complete the generalization.
III. DECOMPOSITION OF THE PROBLEM
The problem formulated in the previous section is a non-liiear,
e E ; U E U ] (1 1)
where, e and U correspond to the decision vector, and the control vector,
respectively, and E and U represent the constraint sets imposed on these
vectors. We adapt a general solution approach which decomposes the
problemby making use of the following property of the optimization
mixed integer programming problemof the following form,
min {f,(e.u) I
e.u
min f, (e& =min {i f f , (e&] (12)
WE. UEU W E UE
assuming that for each decision, e the problemin braces, called the slave
problem,
has a solution and it is "easy" to find it. Then the main problembecomes,
min g(e) (14)
( EE
and is called the musterproblem. These problems can becharacterized as,
Master Problem (MP) : Integer Programming Problem
Slave Problem (SP) : Non-linear Differentiable Optimization Problem
The solution for thi s decomposed problemrequires an efficient solu-
tion algorithmfor the slave problemand a search procedure over E, the set
defined by all the possible decisions, for the master problem. In the next
two sections. we'll discuss and develop such solution schemes.
IV. SLAVE PROBLEM
As indicated in theprevious section, the slave problemassumes that
the capacitors are placed and it is a special case of the capacitor placement
problem. The problemis a non-linear differentiable optimization problem
with quite a large number of equality and inequality constraints. It is st i l l
not easy to solve the slave problem. Our aimhere is to study the special
features and the structure of the problcmand to develop an efficient solu-
tion methodology by exploiting these features.
For the slave problem, the type of capacitors (fixed or switched), and
the number and diversity of load levels are important parameters that
determine the structure and the size of the problem. Therefore, we first
relax the parameter "type" by assuming that they are given. Then the
problembecomes a sizing problem(i.e., determining thecapacitor settings
) and weneed to consider two cases; one with fixed type capacitors and
one with switched type capacitors. In thefollowing first two subsections,
wewill show that for these two cases the problcmis either of the base type
problem or it can be decomposed into the basetype subproblems. The
4.2. Switched Capacitor Problem
slave problemcan be re-written as follows.
When all the capacitors to beplaced are of the switched type. the
(17) U' -U'SO i =I , . . . ,nr
Note that the two constraint sets (15) and (16) are coupled through (17), to
indicate that the capacitor sizes, U ' are the upper bund constraints for
capacitor settings at the other load levels, U' . This weak coupling
between U ' and ui's can be exploited to decompose the probleminto
smaller subproblems. In appendix, it is shown that Psw can be decom-
posed into thefdlowing nt +1 base problems.
The main problem, SW,
Authorized licensed use limited to: UNIVERSIDADE DO PORTO. Downloaded on October 28, 2009 at 23:22 from IEEE Xplore. Restrictions apply.
728
43. Mixed Type Capacitor Problem
In t hi s section, wego back to consider thegeneral slaveproblemin
which it is not known a priori which capacitor is of fixed and which one is
of switched typeand onehas to determinethetypes of capacitors in addi-
tion to their settings. Notethat, in general, the switching capacitors are
moreexpensivein both fixed cost, c and the marginal cost, r, than the
fixed capacitors. Therefore, weproposethefollowing heuristic selection
scheme by using the solution methodology developed for the switched
capacitor problem.
Step 0 : Assumeall thecapacitors are of theswitched type.
Step 1 : Solve theproblemconsidering only theswitched capa-
citors and keeping thefixed capacitor power injections as con-
stant at their nominal settings, i.e., as loads.
Step 2 : Using the results of step 1, check the settings of
switched capacitors at light load level (level nr ). For the ones
with nonzero light load settings, (i.e., u t #0 ) assign that por-
tion of capacitor as fixed type.
Step 3 : If any hxi g occurred, then go to step 1; otherwise,
stop.
Subproblems, SWi i =1 . . . nt
SWi fi (U? =min k, Tipi (xi)
S. t C'(x',u') =0
H'(xi) S 0
OIUU' s uo
The subproblems, SWi involve only oneload level and therefore,
they can besolved by thealgorithmdeveloped for thesizing problemfor
given capacitor sizes, U ' . The solutions will correspond to the oprimal
capacitor senings , 8' for theoff-peak load 'eye's considered.
The main problem, SW,, is also a sizing problem; but computation-
ally it is not of theeasy type due to existenceof fi (114's - theextra terms
coupled to the subproblems - in theobjectivefunction. Updating these
terms at each iteration during the solution of SWo requires solution of the
subproblems. However, wecan use a simple, heuristic schemeto update
fi(uO) and Vfi(u? in themain problemSWo. To begin with, let thesolu-
tions of the subproblems SWi for a given U ' be denoted (ui)* . As we
moveU ' fromiteration to iteration in SW, , f i (u4 andVfi (U') should be
calculated by solving SWi with the new i? for a new (U'): , say (iii)* .
Weprppose, however, instead of solving SWi for a new 6")' , to usethe
old (U')' unless theconstraint (17) is violated, i.e., weset thek'th com-
ponent of 0'
ii; if ii;S(u;)*
a; = (18)
(U;)* otherwise
(19)
I
Wesimply use0' in evaluating fi @) and Vfi @) as follows:
f; (if) =k, Tipi (2' ,tii)
,
Notethat if U ' and i? do not differ very much, theapproximation will be
good. To assurethis, westart the procedureby first solving the subprob-
lems with thecapacitor sizes set to their maximum, U'"=.
The ovcrall iterativealgorithmis shown in Fig.4. In thealgorithm,
an iteration comprises thesolution of thesubproblems, SWi first and then
themain problem, WO. Convergencecheck at end of an iteration involves
checking if thereis a st at us changein theconstraint set of Eq.(17). (i.e., a
non-binding constraint becomes binding or viceversa ). If thereis such a
status change, then wego back and updatetheslave problems; otherwise
westop iterating sincethesolution is converged.
s = s + l
fori =1, ..., nt
SolveSWo
UseEq.(19) and Eq.(20) to calculatef,O and Vfi&) i =1. . . . ,nl
11converged
Figure4 : Block diagramof SW Capacitor Algorithm
V. MASTER PROBLEM
Wefollow a general approach in solving themaster problem, which
is an integer programming problem, and first construct a decision graph
and then develop an efficient search schemeto placethecapacitors (i.e., to
determinetheir numbers and places).
Let there be nc initially given candidatecapacitors. Then, all possi-
bledecisions about choosing the capacitors for placement among the can-
didate capacitors can bemg ed as a decision graph assuming one deci-
sion is madeat a time. Such a graph is shown in Fig5 for nc =3.
Figure 5 : Decision graph for 3 variablecase
In the figure, each node represents a particular decision,
e=[el e2 e3 IT ; ei E { OJ J . Where, ei =0 means that thecapacitor is
not chosen, and ei =1 means it is still a candidate. A branch f" a node
to another indicates how thetransition can beachieved: taking out the can-
didatecapacitor whosenumber is shown on thebranch. Such a relation-
ship is indicated in graph theory by calling a node and all the nodes
emanating fromit. aparent and its children, respectively.
Search starts fromtheroot which corresponds to thc casewhereall
candidatecapacitors are chosen to beplaced at thedesignated nodes of the
system. Then thereare two possiblesearch techniques, depth-frrst search
and breudth-first search, that can be employed to get a local optimum
[17]. Here, rather than employing thesegeneral search techniques directly,
thechildren of a given nodebarent) aresorted first according to their con-
tribution to theobjective. Weproposea sorting procedure which works
sort of likea "steepest descent" approach in discretecase. The procedure
is
- given a nodewhich is identified by its decision vector, eand thesolution
of the corresponding Slave Problem, ( i.e., the control vector
ii =[ z i 1 . . .
- for all existing capacitors, k =1, . . . , nc s.t. ek z 0
and the objectivef, (ii) ).
- construct U by removing thecapacitor k and keeping therest, i.e.,
0 i f j =k
Authorized licensed use limited to: UNIVERSIDADE DO PORTO. Downloaded on October 28, 2009 at 23:22 from IEEE Xplore. Restrictions apply.
729
rev 3 800 910
AE 225 630
itr. 9 175
V,,,i,, .90 160
- calculate the new objective, f:(u) (Thi s requires only DistFlow
- let the contribution of the capacitor k to the objective be
solutions).
Af : =f," -f 0 .
- sort the existing capacitors by using the Af ,"s
The A f ts can used to choose the child to branch out (i.e., capacitor to
take out) since the child with the smallest Af ,"is more liiely the one who
contributes to theobjective the least.
The first procedure, depth-first search, visits the children of a parent
node (i.e., solves the associated slave problems) according to the order
determined by the sorting procedure and branches out on the, first one
which gives a lower objective than that of the parent. This search is of
order nc and therefore computationally attractive. However thebranching
criteria is appropriate only for unconstrainted case. When the voltage con-
straints are active, a better criteria would be to check how well the
deficiency created by a capacitor removal be compensated by the rest of
the capacitors by Nnning the corresponding slave problems. This
corresponds to the breath-& type search which is conducted by visiting
all the children first and then branching out on theone with the smallest
objective. This search is of order nc2 and requires at most nc(nc+l)/2
slave problemsolutions. However, the search can bemade much faster by
noting that the objectives, f," calculated by sorting procedure will be
bigger than the ones calculated by the search. Therefore, in practice, the
search can beconducted only on the children with negative Af ,"s.
6
5
1
2
VI. TESTSTUDIES
The proposed solution methodology has been implemented in For-
tran 77 on both VAX llh'50 and IBM AT. The programuses the algo-
rithmdeveloped for the sizing problemin [14] as a subroutine to solve the
base problems. We present the test results of two systems in thi s section
to illustrate the performance of the proposed solution scheme.
The first test system, TS1 is a 9-branch main feeder test system
developed by Grainger et. al., [9]. The second test system, TS2 is a 69-
branch, %lateral test systemderived froma portion of the PG&E distribu-
tion system. The network data of this systemis given in Table 1.
We adopted the following cost figures : energy cost,
k, =0.06 $/kwh ; capacitor fixed cost, c =lOOO$ ; capacitor marginal
cost, r, =3 $/bur. The load duration data assumed for the systems is
given in Table 2. It is also assumed that substations have regulating
transformers which are tapped to +5% during peak load and set to nominal
otherwise. The substation voltage without tap is taken as the base kV and
the lower and upper voltage limits are assumed to be 0.9 and 1.1 p.u.
respectively.
2 J
4700 9?0 4550
225 700 226
.90 - .90
4 3 - 270
Br. Sd. Rv. - Br . Par. - Rv. Nd. Load
No Nd. Nd. r(ohm) x(ohm) P(KW) Q(KVRR)
1 0 1 0.0005 0.0012 0. 0.
2 ; 2
0.0005 0.0012 0. 0.
2e 0. 0. 0. 0.
1 1 6
960 3700 -
660 191 1160
360 .901 210
5 200
4 2e 3 0.0015 0.0036 0. 0.
5 3 4 0.0251 0.0294 0. 0.
6 4 5 0.3660 0,1864 2.60 2.20
7 5 6 0.3811 0.1941 40.40 30.00
8 6 7 0.0922 0.0470 75.00 54.00
9 7 8 0.0493 0.0251 30.00 22.00
10 8 9 0.8190 0.2707 28.00 19.00
11 9 10 0.1872 0.0619 145.00 104.00
12 10 11 0.7114 0.2351 145.00 104.00
13 11 12 1.0300 0.3400 8.00 5.50
14 12 13 1.0440 0.3450 8.00 5.50
15 13 14 1.0580 0.3496 0. 0.
16 14 15 0.1966 0.0650 45.50 30.00
17 15 16 0.3744 0.1238 60.00 35.00
18 16 17 0.0047 0.0016 60.00 35.00
19 17 18 0.3276 0.1083 0. 0.
20 18 19 0.2106 0.0696 1.00 0.60
21 19 20 0.3416 0.1129 114.00 81.00
22 20 21 0.0140 0.0046 5.30 3.50
23 21 22 0.1591 0.0526 0. 0.
24 22 23 0.3463 0.1145 28.00 20.00
25 23 24 0.7488 0.2475 0. 0.
26 24 25 0.3089 0.1021 14.00 10.00
27 25 26 0.1732 0.0572 14.00 10.00
System So S, S2
TS1. 1.1 0.6 0.3
TS2 1.8 1. 0.5
To TI T 2
1OOO. 6760. 1OOO.
1OOO. 6760. 1OOO.
Table 3: Test runresults for TS1 - fixed capacitor placement
Table 1 : Network Dataof TS2
Br. Sd. Rv. - Br. Par. -
No Nd. Nd. r(ohm) x(ohm) P(Kw) Q(KVAR) No Nd. Nd. r(ohm) X ( 0 h m P(KW) Q(")
Rv. Nd. Load Br. Sd. Rv. - Br. Par.
RV. Nd. Load
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
2 27
27 28
28 29
29 30
30 31
31 32
32 33
33 34
2e 27e
27e 28e
28e 65
65 66
66 67
67 68
68 69
69 70
70 88
88 89
89 90
3 35
35 36
36 37
37 38
0.0044
0.0640
0.3978
0.0702
0.3510
0.8390
1.7080
1.4740
0.0044
0.0640
0.1053
0.0304
0.0018
0.7283
0.3100
0.0410
0.0092
0.1089
0.0009
0.0034
0 .0851
0.2898
0.0822
0.0108
0.1565
0.1315
0.0232
0.1160
0.2816
0.5646
0.4873
0.0108
0.1565
0.1230
0.0355
0.0021
0 .E509
0.3623
0.0478
0.0116
0.1373
0.0012
0.0084
0.2083
0.7091
0.2011
26.00
26.00
0.
0.
0.
14.00
19.50
6.00
26.00
26.00
0.
24.00
24.00
1.20
0.
6.00
0.
39.22
39.22
0.
79.00
384.70
384.70
18.60
18.60
0.
0.
0.
10.00
14.00
4.00
18.55
18.55
17.00
17.00
1.00
0.
4.30
0.
26.30
26.30
0.
56.40
274.50
274.50
0.
51 7 40 0.0928
52 40 41 0.3319
53 8 42 0.1740
54 42 43 0.2030
55 43 44 0.2842
56 44 45 0.2813
57 45 46 1.5900
58 46 47 0.7837
59 47 48 0.3042
60 48 49 0.3861
61 49 50 0.5075
62 50 51 0.0974
63 51 52 0.1450
64 52 53 0.7105
65 53 54 1.0410
66 10 55 0.2012
67 55 56 0.0047
68 11 57 0.7394
69 57 58 0.0047
Substati on Vol tage
Base KVA
Base Vol tage (kV)
0.0473
0.1114
0.0886
0 .l o34
0.1447
0.1433
0.5337
0.2630
0.1006
0.1172
0.2585
0.0496
0.0738
0.3619
0.5302
0.0611
0.0014
0,2444
0.0016
(kv) -
-
-
40.50 28.30
3.60 2.70
4.35 3.50
26.40 19.00
24.00 17.20
0. 0.
0. 0 ..
0. 0.
100.00 72.00
0. 0.
1244.00 888.00
32.00 23.00
0. 0.
227.00 162.00
59.00 42.00
18.00 13.00
18.00 13.00
28.00 20.00
28.00 20.00
12.66
10.000
12.660
t ot al load: P(KW) - 3802.19 , Q(KVAR) - 2694.60
Authorized licensed use limited to: UNIVERSIDADE DO PORTO. Downloaded on October 28, 2009 at 23:22 from IEEE Xplore. Restrictions apply.
730
AE
vmi,
Test run for TSl, summarized in Table 3, starts with 4 capacitors.
The solution of the slave problemfor the mot (initial one) gives the
optimal settings for these 4 capacitors. Then the capacitor's contributions
to the objective, A j t s are calculated by using the sorting procedure. The
results indicate that three of the capacitors - Q,, . e,,, QC6 - have
Af: e 0; which implies that they are not economically feasible ( i.e., their
economic contribution due to energy loss reduction is less than their cost).
Therefore, the search is conducted only on these three capacitors at the
second search level. As a result, QC2 is found to bethe least economical
and hence is taken out. This corresponds to thebranching out on the first
child of the mot in the table. Then a new search process resumes from t hi s
new node; first sorting out its children by using thesorting procedure and
then visiting the ones that are economically infeasible. The solution is
obtained at thethird level of search when the evaluation of capacitors Qc6
and Q,, in the last node by the sorting procedure indicated that they are
economically feasible.
The total run time for t hi s test on VAX is about 45 sec. of CPU and 8
sec. of vo.
785 620 1240 1700 52
.go9
Test System2
Test run for TS2 is summarized in Table 4. The test starts with 5
capacitors. After visiting the root. the capacitor with the zero setting, Qcas
is taken out and the other two Q, and are found to be economically
infeasible by the sorting procedure. The search therefore is conducted
only on these two capacitors and as a result the last node containing the
capacitors QCl8 and Qc52 is identified as the solution node. The total run
time for this test on VAX is recorded as 165 sec. of CPU and 8 sec. of VO.
137.;; 1 3-y 1
1 200
Table 4: Test run results for TS2 - Fixed Capacitor placement
We have the followmg comments/observations about these test
RSults.
1. Search Procedure :
0 The order of search is about nc (number of capacitors placed);
which ismuch less than theworst case bound of nc '.
The search converges to the global optimal point in both tests;
although in test 1, for example, there are some other suboptimal
solutions with revenues close to each other.
0 As search goes further down to higher levels, morecapacitors are
taken out to increase the revenues. This causes less energy loss
reduction and lower voltage profile. This also shows the sensitivity
of theoptimal point with respect to the cost figuresk, , r, , c.
0 The convergence of slave problemat theroot (the initial one) gets
slower as the number of load levels increase (the number of itera-
tions is in theorder of 2-3 nc ). But theother slave problems con-
verge much faster (in the order of 1-2 nc ); mainly because it is
easier to find a good initial point for themby simply using theresults
of theparent node.
2. The slave problem- fixed capacitor problem:
Although there are regulators at thesubstations, voltage profiles of
both systems are below lower voltage limits at thepeak load level
before the capacitor placement. The solution for TSl indicates that
the capacitors are needed to beused for raising thevoltage profile of
the systemat the peak load level ( Vmh =0.9 P.u.) as well as for loss
reduction. The solution for TS2 corresponds to the unconstrainted
optimal point ( Vmh=0.9O7 P.u.); indicating that maximumloss
reduction is achieved.
2. Switched Capacitor Case
Wenow present the test runs for the switched capacitor problemon
the same test systems. The optimal places obtained fromthe general fixed
capacitor problemtests are used in these tests also to avoid the search.
Test System1
The first step of the test runs for TS1 is the solution of the
corresponding slave problems SWi , i =1,2 to get the optimal settings for
the off-peak load levels assuming no limit on thesize of the capacitors.
Each of such solution is obtained in about less than 2 nc iterations by cal-
ling the sizing problemSubroutine. In the second step of thetest, the main
problem, SW, is solved in about 3 nc iterations by using the solution pro-
cedure described in section 5. In the third step, convergence checks indi-
cate that the setting of QC5 for the first load level is binding, i.e.,
=Q,"S. This constitutes the end of the first iteration. Another iteration
is performed to see if the binding status of the capacitors will change. The
convergence is obtained at the second iteration when the capacitor set-
tings, Q2 , Q:, are updated and it is found that there is no binding status
change.
The solution is summarized in Table 5. The box in the table is simi-
lar to that of fixed capacitor case; except here, in addition to thecapacitor
sizes, Q : , capacitor settings at the off-peak load levels, d: , d:, are given
also. The total run takes about 12 sec. of CPU and 2 sec. of VO on VAX.
Table 5 : Test runresults for TSl - Switched Capacitor case.
QZ Q: Q," bus
rev.
AE
Vmin
Test System2
Test runfor TS2 is similar to that of TSl. The solution is obtained in two
iterations and found out that only the setting of Qc18 for the first load level
is binding, i.e., QJ18 = The solution is summarized in Table 6. The
total run takes about 82 sec. of CPU and 4 sec. of VO on VAX.
Table 6 : Test run results for TS2 - Switched Capacitor case
QZ Q,' Q," bus
rev. I 39180 I 197 330 330 I 18
Wehave the following comments about the test results:
0 The performed test runs indicate very good convergence charac-
teristics; the number of iterations between the main problem, SW,
and the subproblems, SWi is usually one or two.
0 When the fixed capacitor and the switched capacitor test results are
compared, it is Seenthat: (i) switched capacitor placement yields
Egher revenues and higher capacitor sizes, especially when the cost
data is the same for both cases and the load variations are diverse.
(ii) the voltage profile is higher ( Vmh is higher) and voltage regula-
tion is better for the switched capacitor case. This is because of the
fact that for the switched capacitor case better compensation is
achieved by adjusting the value of capacitors as the load changes.
The final point to be noted about theoverall test results is about the
effect of the regulators on the solution; (i) feasibility becomes less of a
problem, (ii) no upper limit voltage violation has been observed for the
given test systems, although the load levels werequite diverse. (iii) as
exemplified here by test run for TSl, capacitors can beused together with
voltage regulators to keep the voltage profile of thesystemwithin defined
limits.
Authorized licensed use limited to: UNIVERSIDADE DO PORTO. Downloaded on October 28, 2009 at 23:22 from IEEE Xplore. Restrictions apply.
73 I
Analysis of the Results
Consider the starting point of TS2 in Table 4, wherethefixed cost of
capacitors m not included. The solution of capacitor sizing gives 1040
kvar and 210 kvar on thenodes 52 and 47 of a lateral respectively, and 170
kvar and 230 kvar on the nodes 11 and 18 of the main feeder respectively.
In the system, theloads in the laterals are moreconcentrated whereas the
loads in the main feeder are more evenly distributed. The result is that the
size of the capacitors are also more concentrated in the lateral and more
evenly distributed in the main feeder. This further reaffirms the fact that
the nature of reactive power compensation is rather local.
W. CONCLUSIONS
In this paper, a general formulation and an efficient solution metho-
dology have been developed for general capacitor placement problemon
radial distribution systems.
The general capacitor placement problemconsists of placing the
capacitors (determining their number and thelocations) and determining
their types and sizes. The objective is peak power and energy loss reduc-
tion while keeping thecost of capacitors at a minimum.
The proposed formulation is comprehensive in the sense that: (i) it
considers all the variables of the problemstated above, (ii) it uses the ac
power flow equations to represent the system, (iii) voltage constraints are
taken into account.
A solution method has been developed for this general problemby
decomposing the probleminto two hierarchical levels. The top level prob-
lem. called the master problem, is an integer programming problemand is
used to place the capacitors (determine their number and locations). An
efficient search scheme has been developed for the master problem. The
second level problem, called the slave problem, is used by the master
problemas a subroutine. This problemis further decomposed into two
levels: at the top level, the problemconsists of determining the type of
capacitors and at the bottomlevel, the problemis to determine the capaci-
tor sizes once the capacitors are placed with their types assigned. These
slave problems, called the fixed and switched capacitor problems, are
shown to beeither the base type or can be decomposed into base t ype
problems. The base type problemis a capacitor sizing problemand is
solved by using an efficient phase I - Phase I1 type solution method
presented in [ 141.
Test results are presented for theproposed solution scheme. They
indicate. that the method is computationally efficient and the decomposi-
tion scheme performs well.
Although not implemented, it is also shown that theformulation and
the solution methodology presented in this paper can be generalized to
include the voltage regulators in theproblem.
Acknowledgements
We thank Mr. Wayne Hong and Dr. Dariush Shirmohammadi of
Pacific Gas and Electric for their helpful discussions. This research is sup-
ported by TUBITAK-TURKEY and by National Science Foundation
under grant ECS-8715132.
REFERENCES
[l ]
[2]
Distribution Systems, East Pittsburgh, PA. : Westinghouse Electric
Corp., 1965
R. F. Cook, "Optimizing the Application of Shunt Capacitors for
Reactive Volt-Ampere Control and Loss Reduction" , AIEE Trans.,
Y. G. Bae, "Analytical Method of Capacitor Application on Distribu-
tion Primary Feeders", IEEE Trans. on Power Apparatus and Sys-
tems, vol. 97, pp. 1232-1237, July/Aug. 1978.
H. Duran, "OptimumNumber, Location, and Size of Shunt Capaci-
tors in Radial Distribution Feeders: A Dynamic Programming
Approach, IEEE Trans. on Power Apparatus and Systems, vol. 87,
pp. 1769-1774, Sept. 1968.
vol. 80, pp. 430-444, August 1961.
[3]
[4]
M. Ponnavaikko and K. S. Prakasa Rao, "Optimal Choice of Fixed
and Switched Shunt Capacitors on Radial Distribution Feeders by
the Method of Local Variations", IEEE Trans. on Power Apparatus
andsyst em, vol. 102, pp.1607-1614, June 1983.
T. H. Fawzi. S. M. El-Sobki, and M. A. Abdel-Halim, "A New
Approach for the Application of Shunt Capacitors to the Primary
Distribution feeders", IEEE Trans. on Power Apparatus and Sys-
tems, vol. 102, pp.10-13, Jan. 1983.
M. Kaplan, "Optimization of Number, Location, Size, Control Type,
and Control Setting of Shunt Capacitors on Radial Distribution
Feeders", IEEE Trans. on PAS, vol. 103, pp.2659-2665, Sept. 1984.
J . J. Grainger, and S. H. Lee, "OptimumSize and Location of Shunt
Capacitors for Reduction of Losses on Distribution Feeders", IEEE
Trans. on Paver Apparatus and Systems, vol. 100, pp. 1105-1 118,
March 1981.
J. J. Grainger, and S. H. Lee, "Capacity Release by Shunt Capacitor
PhxtwU on Distribution Feeders: a New Voltage Dependent
Model", IEEE Trans. on Power Apparatus and Systems. vol. 101.
pp. 1236-1244, May 1982.
J. J. Grainger, S. H. Lee, and A. A. El-Kib, "Design of a Real-Time
Switching Control Scheme for Capacitive Compensation of Distribu-
tion Feeders", IEEE Trans. on Power Apparatus and Systems, vol.
J. J. Grainger, S. Civanlar, and K. N. Clinard, L. J. Gale, "Optimal
Voltage Dependent Continuous Time Control of Reactive Power on
Primary Distribution Feeders", IEEE Trans. on Power Apparatus
andsyst em, vol. 103, pp. 2714-2723, Sept. 1984.
A. A. El-Kib, J. J. Grainger, and K. N. Clinard, L. J. Gale, "Place-
ment of Fixed and/or Non-Simultaneously Switched Capacitors on
Unbalanced Three-phase Feeders Involving Laterals", IEEE Trans.
on Paver Apparatus and Systems, vol. 104, pp. 3298-3305, Nov.
1985.
S. Civanlar and J. J. Grainger, "Volt/Var Control on Distribution
Systems wilh Lateral Branches Using Shunt Capacitors and Voltage
Regulators: Part I, Part 11, Part III", IEEE Trans. on Power
Apparatus andsystems , vol. 104, pp. 3278-3297, Nov. 1985.
M. E. Baran and F. F. Wu, "Optimal Sizing of Capacitors Placed on
a Radial Distribution System". submitted to IEEE PES winter
meeting, 1988.
A. M. Geoffrion, "Elements of Large Scale Mathematical Program-
ming, Part I: Concepts", Management Science, vol. 10, pp. 652-675.
July 1970.
A. M. Geoffrion, "Generalized Benders Decomposition", JOTA, vol.
E. M. Reingold and W. J. Hansen. Data Structures. Little Brown and
Comp., 1983.
101, pp. 2420-2428, August 1982.
10, pp. 237-260, April 1972.
Appendix: Decomposition of the Switched Capacitor Problem
Weshall apply decomposition techniques to the switched capacitor
problemintroduced in section 4.2. We present first a general scheme,
which leads to the Benders Decomposition, and then a simpler, heuristic
based decomposition scheme for the special switched capacitor problem.
Decomposition
A general decomposition scheme is given in [15]. To adopt the
derivation for the switched capacitor problem, Psw , wefirst re-group the
variables and constraints as follows,
nf T
U 1
Uo= {uo I Eq.(15) is satisfied ) U ={ U I Eq.(16) is satisfied )
;
Similarly, wepartition the objective function as,
k r C I i =l
Authorized licensed use limited to: UNIVERSIDADE DO PORTO. Downloaded on October 28, 2009 at 23:22 from IEEE Xplore. Restrictions apply.
132
Then Psw can bere-written as
min if, =f,(u?+fr(u) I UOEU, ; U E U ; F(U',U)L;O l(a.1)
u9u
Where, F(uo,u) comsponds to the coupling constraints of Eq.(17).
onto space of uo alone as follows.
Westart the decomposition by projecting (purtitioning) the problem
(a.2)
Weassume that the subproblem in thebraces,
f s (U? =:$i f r (u)-F(uo*u) 5 0 1
can be evaluated for a given U ' as an optimization problemwith respect to
the variable U. Then the main problembecomes
min f0 ('4 =fm ( ' 9 +fJ ('4 6.3)
U W O
W O
To assure that the above assumption holds, wemust avoid thevalues of U '
such that f,@) does not have a feasible solution. For this, wedefine a
new set. V as follows.
V ={U' I 5 UEU s.t. F(u0,b) I O }
The set V can bethought of as the projection of theconstraint set defined
by F(uo,u) and U onto the space defined by U ' alone as illustrated in
Fig.a.1.
Figure a. 1 : Set hOjeCtiOn
Now wegeneralize the projection by rewriting the problem(a.3) as fol-
lows.
min {f,,,(u") +f,(u") I U'E n v 1 (a.4)
U0
m0
For the switched capacitor problem, wehave a conjecture that
U, c v.
Justification of t hi s conjecture will be discussed later.
Assuming that this conjecture holds for the general case, wecan use
the partitioned problem(a.3) rather than (a.4) for the main problem, SW, .
The subproblemf, (U? can further be decomposed into nt subproblems of
the following form.
The solution algorithminvolves the iterative solution of the main
problemsw& and the subproblems, SW;. After each iteration, a new set
of constraints of thetype(a.5.i) is added to Swab until the solution con-
verges.
It is noted that the Benders decomposition has the following features.
(i) The main problem, SW, is not a base type problem. Additional con-
straints of (a.5.9, called the cuts, are difficult to handle with the solution
algorithmused by the bask problem; one has to identify which one these
constraints will bebinding during the solution of web.
(ii) The contribution of the subproblems to theobjective of themain prob-
lem, fJi(u? is approximated by linearizing this termaround the previ-
ously calculated points, uoJ . To see this, let thebinding constraint for the
subproblemi be the k'th one in Eq.(aS.i). Then the solution for the
correspondingyi will be
Therefore, the approximation is good only if theactual solution point, U ' is
close to the calculated point U* in solving Sw&.
A Heuristic Based Decomposition Scheme
Note that the main problemof (a.3) need not be transcribed into
Benders form, SW, if One can estimate thebinding constraints in thesets
U' - U ' L; 0. Wedevelop a solution algorithmbased on this principle. The
solution scheme uses the sizing algorithmto solve both the main problem
and the subproblems andit also uses a better estimate for thecontribution
of the subproblems to the objective of the main problem. The details of
the algorithmis given in section 4.2.
J ustification of the Conjecture
U, c v
The idea behind this conjecture is as follows. Let uo E U,. This
means that U ' amount of reactive power compensation fromthe capacitors
is enough to satisfy the voltage constraints for thepeak load. But this
amount of compensbtion must suffice to have a feasible point for lower
load levels too because the lower the load the higher the voltage profile
will be. This observation is due to the strong coupling between the reac-
tive power flow and the voltage profile of the system.
Note that this conjecture was also the underlining idea behind the
assumption made when formulating the problemin section 2; where, it
was assumed that the capacitor size will bethe capacitor setting for the
peak load level, uo and capacitor settings for all the other lower load lev-
els, U' will be smaller than U ' .
f s i (U" =ic[fri (U')
Mesut E. Baran received his B.S. :indM.S. fromMiddle East Technical
University, Turkey. Heis currently a Ph.D. student at theUniversity of
California, Berkeley.
Felix F. Wu received his B.S. fromNational Taiwan University, M.S.
fromtheUniversity of Pittsburgh, and Ph.D. fromtheUniversity of Cal-
ifornia, Berkeley. Heis a professor of Electrical Engineering and Com-
S.t . ui -U010
U' EUi
Where, Ui ={ui I Eq.(I6) }and f, (u? =us, (U?.
The explicit
SWi i =1. . . . , nt are given in "4.2.
Benders Decomposition
Assuming that the duality conditions holds for theswitched capacitor
problemof (a.2) [16], themain problemcan betranscribed into thefollow-
ing form.
m o b
of the main problem Swo and the subproblems puter Sciences at the University of Califomia, Berkeley.
min fo (u'.Y) =f d (U? +Cyi
s.t. yi + 3 c " r ~ u o - u o ~ ~ ~ ~ , i ~ u o ~ ~ j =I , . . . . p (a.5.i)
uo E U, i =1.. . . .nt (a.5.ii)
Where, U" and I.'' corresponds to the solution of thesubproblemSW, at
iteration j.
Authorized licensed use limited to: UNIVERSIDADE DO PORTO. Downloaded on October 28, 2009 at 23:22 from IEEE Xplore. Restrictions apply.
133
the optimal location of the one capacitor bank needed at minimum
load.
2) Increment the load in steps. At each load level ask thequestion, Is
more capacitance needed? If so, then determine optimal switched
capacitor locations and sizes based primarily on loss and voltage
profile improvement.
3) Once peak load is reached, decrement theload to determine switch-off
load levels.
This very simplistic method of locating and sizing capacitors incremen-
tally tends to distribute smaller size banks over the feeder where they are
needed and when they are needed. Generally, no more than one capacitor is
added at a load level and the search for optimumlocation is trivial. My
simulations of the feeder over a years time considering daily and seasonal
load cycles indicate that it is difficult to significantly improve upon this
method economically. One can always find a solution that appears to bea
few percentage points better, but I do not think that the basic data are known
with sufficient accuracy to quibble over a small difference in an off-line
analysis. It would require on-line control to take advantage of the small
gains possible. Wehave implemented this method in an interactive program
that uses the above algorithmto get close to the most economical solution.
Then the user can tweek the solution interactively taking into account
practical considerations. The whole process takes but a few minutes using a
personal computer (E).
Differing approaches to the problemof capacitor size and placement will
yield differing optimal solutions. 1 suspect that noneare truly optimal
and most approaches that consider load variation are generally adequate
(methods that optimize only peak load sometimes give poor results when the
entire load cycle is considered). I would hesitate to defend oneapproach
over another too strongly becausefeeder load varies somewhat randomly
and it would bedifficult to prove which is more optimal. However, I think
that methods like I have described, which are based on how the feeder
operates, are apt to bemore optimal more of the time. They are also simple
to programand the programs execute quickly. Therefore, I question the
practicality of abandoning the simpler approach in favor of a more
sophisticated method like the authors have presented.
Reference
[l] T. Frantz et al., Load behavior observed in LILCO and RG&E
systems, ZEEE Trans. Power App. Syst., April 1984, pp. 819-831.
Manuscript receivedFebruary 19, 1988
Discussion
Roger C. Dugan (McGraw-Edison Power Systems, Cooper Industries
Inc., Canonsburg, PA): The reader will discover in a few moments that I
have several objections to themethods presented in this paper. However, 1
do not wish for my objections to reflect poorly upon the efforts of the
authors. It would appear that a great deal of good work has been done and I
suspect that this paper was intended to emphasize the application of more
sophisticated techniques to the problem. Therefore, it may be unfair to
expect theauthors to respond to all of myobjections because they are based
on more practical aspects and also apply to previous investigators in this
field who havemadesimilar assumptions.
First, one would hopethat a method employing nonlinear programming
techniques would be faster than simple exhaustive searches. It is not
apparent fromthepaper that this would bethe case, and the times quoted
during the papers presentation and discussion lead meto believe otherwise.
The number of discrete solutions to this problemare not necessarily large
due to the practical constraints that I will mention below. Therefore, an
intelligent exhaustive search method that automatically discards many cases
due to its knowledgeof the way the feeder operates can be reasonably fast. I
haveinvestigated a number of search techniques that give very good, but
perhaps not optimal, solutions, and the execution times grow approximately
linearly with problemsize rather than geometrically.
The method proposed in the paper is based on the assumption that the
feeder is discrete and the capacitor size is continuous. Perhaps, a more
realistic choice would have been the opposite. There are many nodes in a
circuit, but utilities generally wish to consider only two or three different
capacitor sizes, for example600- and 1200-kvar banks. I have generally
approached the problemby assuming that both the feeder and capacitor
sizes are discrete. When considering such a s d number of sizes, a simple
search can often be done quickly.
Another assumption I would question is that the capacitor is a source of
reactive power (Q). (I assume that this implies a constant source because I
amnot able to ascertain otherwise fromthe paper.) Of course, this will
introduce inaccuracies becausea capacitor is a constant impedance element.
Sincecapacitor placement on a feeder significantly affects the voltage, this
assumption weakens any claimthat the method results in an optimal
location. It would seemto be a simple matter to correctly represent
capacitors and avoid t hi s difficulty.
My examination of this problemhas also indicated that line regulator tap
position and control characteristics affect the optimal solution. It is not
clear how the method presented in the paper properly accounts for these
effects.
The constant P-Q load model employed in theproposed method also leads
to inaccuracies. The P-Q load model is a peculiar bias of transmission
analysts, and it should not be employed on distribution systems without
question. Frantz et al. [I] have clearly shown that distribution systemloads
are sensitive to voltage. My experimentation with different load models has
shown that one will usually get a different optimal solution for each load
model. The P-Q load model is best employed to establish the base case
voltages fromknown load conditions. Then one should switch to a more
realistic model when studying capacitor additions. In theabsence of better
information, I will typically use a load model in which the P varies linearly
with voltage and the Q varies by the square of the voltage. Lacking this
capability, I suspect that a simple constant impedance load model would be
better than a constant P-Q model.
Differing economic evaluation criteria among utilities require different
approaches to theoptimization problem. Utilities using very high values for
released substation and generation capacity savings may achieve a more
economical solution by optimizing the peak load condition first, although I
must admit to being skeptical of this. Then the load is decremented in steps
to determine when switched banks should beturned off. For utilities where
the costs of losses is more important, a more economical solution can
generally be achieved by fist optimizing the location of fixed banks at
minimumload and then incrementing the load in steps to determine the
location, size, and switching levels of the switched banks.
I believe that the latter approach is more practical for most utilities. One
reason is that most feeders operate near minimumor average load levels
much morethan they operate near peak loads. Another reason is that I
question whether values for substation and generation costs are based on
assumptions compatible with assumptions made for capacitor economic
evaluations. This approach can be easily programmed using an intuitive
algorithmthat recognizes how a feeder typically operates. It yields a near-
optimal solution that is difficult to improve upon significantly. I will state
thealgorithmin words, giving the reader the freedomof choice in selecting
techniques for solving the load flow and making decisions.
1) Select the fixed capacitors. This frequently is simply theselection of
M. E. Baran and F. F. Wu : Wewould like to thank Mr. Dugan for his
interest in the paper and his insightful questions about the capacitor place-
ment problem.
The algorithmoutlined by Mr. Dugan is a special case of switched
capacitor problemintroduced in this paper. In Sec. 4.2 it is shown that
assuming all the capacitors are of switched type, and for a given set of
capacitors placed on the system, the problemcan be decomposed into
smaller subproblems each of which corresponds to minimization of losses
at each of the load levels considered. However, these subproblems are
coupled to each other due to the cost of capacitors. It is easy to show that
when the cost of capacitors are neglected the subproblems become
independent and hence the optimization for each load level can be carried
out independently. Mr. Dugans algorithmincorporates both placement
and sizing problems into this decomposition scheme. It is indeed a good
idea especially since he considers the capacitor sizes as discrete. There-
fore, his algorithmwi l l work, as hepoints out, when theobjective function
is power loss minimization or voltage regulation.
However, for thegeneral case, wherethecost of capacitor is impor-
tant and it effects the number of capacitors to beplaced, the method may
not give good answers because of thecoupling between the subproblems.
Hence, in this case, it is not easy to answer the question how capacitors
should be added as the load level increases. It seems that thebest aid to
answer this question would be the use of switched capacitor problem.
Starting froma candidate set of capacitors and assuming themall switch-
able the problemcan besolved by using the switched capacitor algorithm
introduced in. this paper. The solution will give thecapacities of thecapa-
citors, U and their settings at other load levels, ui , i =1, . . . , nt assuming
the capacities are continuous variables. These results then can beused in
answering thequestion mentioned above and hence thecapacitors can be
placed by the method proposed by h4r. Dugan. This way thesearch intro-
Authorized licensed use limited to: UNIVERSIDADE DO PORTO. Downloaded on October 28, 2009 at 23:22 from IEEE Xplore. Restrictions apply.
134
duced in thi s paper may not need to becani d out any further thanthe
mot.
Another point raised by Mr. Dugan is the assumption made in
modeling the capacitors. It involves approximating the reactive power
injected by a capacitor as constant, independent of the voltage. This
assumption is justified based on the fact that V, =1 pa. and sizes of capa-
citors determined by the solution need to bemunded off to get the practi-
cal size of capacitors. However, theexact model can beincorporated in
the method if needed. This is explained in the closure of [14]. Note that
thi s approximation will most likely affect the sizes, not the location, of
capacitors.
The power Rowmodel used in thi s paper (DistFlow equations) can
handle voltage dependent loads and the solution algorithmcan begeneral-
ized to take into account such loads, as explained again in the closure of
~41.
Manuscript received May 2, 1988.
Authorized licensed use limited to: UNIVERSIDADE DO PORTO. Downloaded on October 28, 2009 at 23:22 from IEEE Xplore. Restrictions apply.
Apndice J
SIMPLE AND EFFICIENT COMPUTER ALGORITHM TO SOLVE
RADIAL DISTRIBUTION NETWORKS, RANJAN, R. ET AL. 2003, [27]
150


PLEASE SCROLL DOWN FOR ARTICLE
This article was downloaded by: [B-on Consortium - 2007]
On: 29 October 2009
Access details: Access Details: [subscription number 908038079]
Publisher Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,
37-41 Mortimer Street, London W1T 3JH, UK
Electric Power Components and Systems
Publication details, including instructions for authors and subscription information:
http://www.informaworld.com/smpp/title~content=t713399721
Simple and Efficient Computer Algorithm to Solve Radial Distribution Networks
Rakesh Ranjan
a
; D. Das
b
a
Multi Media University, Ayer Keroh Lama, Melaka, Malaysia.
b
Electrical Engineering Department, Indian
Institute of Technology, Kharagpur, India.
Online Publication Date: 01 January 2003
To cite this Article Ranjan, Rakesh and Das, D.(2003)'Simple and Efficient Computer Algorithm to Solve Radial Distribution
Networks',Electric Power Components and Systems,31:1,95 107
To link to this Article: DOI: 10.1080/15325000390112099
URL: http://dx.doi.org/10.1080/15325000390112099
Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf
This article may be used for research, teaching and private study purposes. Any substantial or
systematic reproduction, re-distribution, re-selling, loan or sub-licensing, systematic supply or
distribution in any form to anyone is expressly forbidden.
The publisher does not give any warranty express or implied or make any representation that the contents
will be complete or accurate or up to date. The accuracy of any instructions, formulae and drug doses
should be independently verified with primary sources. The publisher shall not be liable for any loss,
actions, claims, proceedings, demand or costs or damages whatsoever or howsoever caused arising directly
or indirectly in connection with or arising out of the use of this material.
EMP 31(1) #2689
Electric Power Components and Systems, 31:95107, 2003
Copyright c 2003 Taylor & Francis
1532-5008/03 $12.00 + .00
DOI: 10.1080/15325000390112099
Simple and Ecient Computer Algorithm to
Solve Radial Distribution Networks
RAKESH RANJAN
Multi Media University
Ayer Keroh Lama
Melaka, Malaysia
D. DAS
Electrical Engineering Department
Indian Institute of Technology
Kharagpur, India
A simple and ecient algorithm is presented to solve radial distribution net-
works (RDN). It solves the simple algebraic recursive expression of voltage
magnitude and all the data are stored in vector form. The algorithm uses the
basic principle of circuit theory and can be easily understood. The proposed
algorithm has been tested with several distribution networks and results are
compared with two other existing methods. The eectiveness of the proposed
algorithm is demonstrated through two examples.
Keywords radial distribution networks, load ow, circuit model
Nomenclature
NB total number of the load
LN1 total number of the branch (LN1 = NB 1)
PL(i) real power load of ith node
QL(i) reactive power load of ith node
|V (i)| voltage magnitude of ith node
R(jj) resistance of the branchjj
X(jj) reactance of the branchjj
Z(jj) impedance of the branchjj
I(jj) current owing through branchjj
P(m2) total reactive power load fed through node m2
Q(m2) total reactive power load fed through node m2
(m2) voltage angle of the node m2
LP(jj) reactive power loss of branchjj
LQ(jj) reactive power loss of branchjj
Manuscript received in nal form on 16 January 2002.
Address correspondence to R. Ranjan. E-mail: rakesh.ranjan@mmu.edu.my
95
D
o
w
n
l
o
a
d
e
d

B
y
:

[
B
-
o
n

C
o
n
s
o
r
t
i
u
m

-

2
0
0
7
]

A
t
:

0
3
:
0
4

2
9

O
c
t
o
b
e
r

2
0
0
9
96 R. Ranjan and D. Das
IS(jj) sending end node of branchjj
IR(jj) receiving end node of branchjj
PLOSS total reactive power loss
QLOSS total reactive power loss
1. Introduction
In the past few years, the developments of automated distribution systems solu-
tions have increased considerably. With the development of the microcomputer,
the requirement of distribution substation-owned computer programs has become
a necessity. However, the choice of a solution method for the practical application is
dicult. It requires a careful analysis of comparative advantages and disadvantages
of those methods available in respect to storage, computation speed, and conver-
gence criterion. Generally, radial distribution networks has a high R/X ratio. Due
to this, conventional Newton Raphson [1] and fast decoupled load ow [2] methods
fail to converge. Many other researchers [35] have suggested modied versions of
conventional load ow methods with a high R/X ratio.
Kersting and Mendive [6] and Kersting [7] have developed load-ow techniques
based on ladder theory and Stevens et al. [8] modied it and proved faster than
earlier methods. However, it fails to converge in ve out of twelve case studies.
Baran and Wu [9] have developed a load-ow method based on the Newton-Raphson
method but it requires a Jacobian matrix, a series of matrix multiplications, and
at least one matrix inversion. Hence, it is not computationally ecient. Chiang [10]
has developed decoupled and fast decoupled load-ow methods based on a method
suggested by Baran and Wu [9]. The very fast decoupled method is impressive
because it does not require any Jacobian matrix. Many other researchers [115] have
proposed generalized methods of modeling and analysis of distribution systems.
However, the diculty arises from the fact that no method posseses all the desirable
features.
In this article, a simple algorithm that is based on basic systems analysis
methods and circuit theory is developed. The purpose of this article is to develop a
new calculation model that requires less computer memory and is computationally
fast for radial distribution networks. The proposed method involves only recursive
algebraic equations to be solved to get the following information:
1. Status of the feeder line, overloading of the conductor and feeder line cur-
rents;
2. Whether the system can maintain adequate voltage level for the remote
loads;
3. The line losses in each segment;
4. Suggestion of the necessity of rerouting or network reconguration for the
existing distribution networks.
The proposed method is compared with those of Das et al. [12] and Baran and
Wu [9]. It is observed that the proposed algorithm is computationally very e-
cient. Several distribution networks have been tested with this algorithm with the
consideration that all loads are constant power. However, the algorithm can easily
accommodate composite load modeling, if the composition of the load is known.
The algorithm has a good convergence property for practical radial distribution
networks.
D
o
w
n
l
o
a
d
e
d

B
y
:

[
B
-
o
n

C
o
n
s
o
r
t
i
u
m

-

2
0
0
7
]

A
t
:

0
3
:
0
4

2
9

O
c
t
o
b
e
r

2
0
0
9
Algorithm for Radial Distribution Networks 97
Figure 1. Sample radial distribution net-
works.
2. Circuit Model
In this section, a circuit model of a radial distribution network (RDN) is presented.
It is assumed that a three-phase RDN is balanced and can be represented by
equivalent single line diagram. Line shunt capacitance at distribution voltage level
is negligibly small. Figure 1 shows single line diagram of a sample radial distribution
network.
The electrical equivalent of Figure 1 is shown in Figure 2.
3. Mathematical Model of Radial Distribution Networks
A mathematical model of radial distribution networks can easily be derived from
Figure 2.
I(jj) =
|V (m1)|(m1) |V (m2)|(m2)
Z(jj)
(1)
and
P(m2) jQ(m2) = V

(m2) I(jj) (2)

where Z(jj) = R(jj)+X(jj), m1, and m2 are the sending and receiving end nodes,
respectively, {m1 = ISS(jj) and m2 = IRR(jj)}.
P(m2) = sum of the real power loads of all the nodes beyond node m2 plus the
real power load of the node m2 itself plus the sum of the real power losses of
all the branches beyond node m2.
Q(m2) = sum of the reactive power loads of all the nodes beyond node m2 plus the
reactive power load of the node m2 itself plus the sum of the reactive power
losses of all the branches beyond node m2.
Figure 2. Electrical equivalent of one branch of Figure 1.
D
o
w
n
l
o
a
d
e
d

B
y
:

[
B
-
o
n

C
o
n
s
o
r
t
i
u
m

-

2
0
0
7
]

A
t
:

0
3
:
0
4

2
9

O
c
t
o
b
e
r

2
0
0
9
98 R. Ranjan and D. Das
Table 1
Branch number, sending end node, and receiving end
nodes of Figure 1
Branch no. Sending end node Receiving end node
(jj) ISS(jj) IRR(jj)
1 1 2
2 2 3
3 3 4
4 4 5
5 2 8
6 8 9
7 9 10
8 3 6
9 6 7
10 9 11
From equations (1) and (2) we get
|V (m2)| =

{B(jj) A(jj)} (3)

where
A(jj) = P(m2) R(jj) +Q(m2) X(jj) 0.5 |V (m1)|
2
(4)
B(jj) =

{A
2
(jj) {Z
2
(jj) (P
2
(m2) +Q
2
(m2))} (5)
Since the substation voltage magnitude |V (1)| is known, for jj = 1, (see
Table 1), m1 = ISS(jj) = ISS(1) = 1 and m2 = IRR(jj) = IRR(1) = 2.
A(jj) = A(1) and B(jj) = B(1) can be computed using equations (4) and (5), if we
know P(m2) = P(2) and Q(m2) = Q(2). After that |V (2)| can easily be computed
from equation (3). Similarly for jj = 2, (see Table 1), m1 = ISS(jj) = ISS(2) = 2,
m2 = IRR(jj) = IRR(2) = 3. A(jj) = A(2), B(jj) = B(2) can be computed
using equations (4) and (5), if we know P(m2) = P(3) and Q(m2) = Q(3) and
hence |V (3)| can easily be computed by using equation (3). In general, for jj =
1, 2, . . . , LN1, voltage magnitude of all the nodes can easily be computed by using
equations (4), (5), and (3), if we know P(m2) and Q(m2) for m2 = 2, 3, . . . , NB.
Computation of P(m2) and Q(m2) (m2 = 2, 3, . . . , NB) is explained in section 4.
Real and reactive power losses in the branch jj are
LP(jj) =
R(jj) (P
2
(m2) +Q
2
(m2))
|V (m2)|
2
(6)
LQ(jj) =
X(jj) (P
2
(m2) +Q
2
(m2))
|V (m2)|
2
(7)
4. Computation of P(m2) and Q(m2)
A computer logic is presented in this section for automatic computation of P(m2)
and Q(m2). For the requirements of the computer logic presented, in Table 1, branch
D
o
w
n
l
o
a
d
e
d

B
y
:

[
B
-
o
n

C
o
n
s
o
r
t
i
u
m

-

2
0
0
7
]

A
t
:

0
3
:
0
4

2
9

O
c
t
o
b
e
r

2
0
0
9
Algorithm for Radial Distribution Networks 99
number, sending end, and receiving end nodes of the feeder shown in Figure 1 are
given.
First set IR(jj) = IRR(jj) and IS(jj) = ISS(jj) for jj = 1, 2, . . . , LN1.
Now for jj = 1 (rst branch of Figure 1 and Table 1), IR(jj) = IR(1) = 2,
and set P(2) = PL(2), Q(2) = QL(2). Now computer logic will check whether
IR(1) = IS(i) or not for i = 1, 2, . . . , LN1. It is seen that (Table 1) for i = 2
(Branch 2), IR(1) = IS(2) = 2 and for i = 5 (Branch 5), IR(1) = IS(5) = 2,
and corresponding receiving end nodes are IR(2) = 3 and IR(5) = 8. Now it will
compute
P(2) = P(2) + PL(3) + PL(8) + LP(2) + LP(5)
= PL(2) + PL(3) + PL(8) + LP(2) + LP(5)
and
Q(2) = Q(2) + QL(3) + QL(8) + LQ(2) + LQ(5)
= QL(2) + QL(3) + QL(8) + LQ(2) + LQ(5)
Proposed computer logic will again check whether nodes 3 and 8 are connected with
the other nodes. It is seen that node 3 is connected with node 4 (branch 3) and
node 6 (branch 8). Node 8 is connected with the node 9 (branch 6). Now computer
will compute
P(2) = P(2) + PL(4) + PL(6) + LP(3) + LP(8) + PL(9) + LP(6)
= PL(2) + PL(3) + PL(8) + LP(2) + LP(5) + PL(4)
+ PL(6) + LP(3) + LP(8) + PL(9) + LP(6)
and
Q(2) = QL(2) + QL(3) + QL(8) + LQ(2) + LQ(5) + QL(4)
+ QL(6) + LQ(3) + LQ(8) + QL(9) + LQ(6)
Similarly, computer logic will check whether nodes 4, 6, and 9 are connected
with the other nodes. It is seen that 4 is connected with node 5 (branch 4), node 6
is connected to node 7 (branch 9) and node 9 is connected with node 10 (branch 7)
and node 11 (branch 10). Therefore,
P(2) = P(2) + PL(5) + LP(4) + PL(7) + LP(9) + PL(10) + LP(7)
+ PL(11) + LP(10)
P(2) = PL(2) + PL(3) + PL(8) + LP(2) + LP(5) + PL(4) + PL(6)
+ LP(3) + LP(8) + PL(9) + LP(6) + PL(5) + LP(4) + PL(7)
+ LP(9) + PL(10) + LP(7) + PL(11) + LP(10)
For jj = 2 (branch 2), IR(jj) = IR(2) = 3 (from Table 1), set P(3) = PL(3)
and Q(3) = QL(3). Now computer logic will check whether IR(2) = IS(i) for
i = 1, 2, . . . , LN1. It is seen that (Table 1) for i = 3 (branch 3), IR(2) = IS(3) = 3
D
o
w
n
l
o
a
d
e
d

B
y
:

[
B
-
o
n

C
o
n
s
o
r
t
i
u
m

-

2
0
0
7
]

A
t
:

0
3
:
0
4

2
9

O
c
t
o
b
e
r

2
0
0
9
100 R. Ranjan and D. Das
and for i = 8 (branch 8), IR(2) = IS(8) = 3 and correspondingly receiving end
nodes are IR(3) = 4 and IR(8) = 6. Now it will compute
P(3) = P(3) + PL(4) + PL(6) + LP(3) + LP(8)
= PL(3) + PL(4) + PL(6) + LP(3) + LP(8)
Similarly,
Q(3) = Q(3) + QL(4) + QL(6) + LQ(3) + LQ(8)
= QL(3) + QL(4) + QL(6) + LQ(3) + LQ(8)
Proposed computer logic will again check whether nodes 4 and 6 are connected
with any other node. It is seen that (Table 1) node 4 is connected with node 5 and
node 6 is connected with node 7, i.e., i = 4, IR(4) = 5 and i = 9, IR(9) = 7.
Therefore,
P(3) = P(3) + PL(5) + PL(7) + LP(4) + LP(9)
= PL(3) + PL(4) + PL(6) + LP(3) + LP(8)
+ PL(5) + PL(7) + LP(4) + LP(9)
and
Q(3) = Q(3) + QL(5) + QL(7) + LQ(4) + LQ(9)
= QL(3) + QL(4) + QL(6) + LQ(3) + LQ(8)
+ QL(5) + QL(7) + LQ(4) + LQ(9)
Exact computation of P(3) and Q(3) is complete because nodes 5 and 7 are not
connected with any other node. Similarly, computer logic will compute exact load
fed through each node. It is to be noted here that if the receiving end node of any
branch is an end node, then total load fed through that node is the load of node
itself. For example, consider node 5 of Figure 1, which is an end node. Therefore,
P(5) = PL(5) and Q(5) = QL(5). The proposed algorithm will also identify the
end nodes. It is worth reporting that node numbering of the feeder is arbitrary and
the algorithm will automatically compute the exact load fed through all the nodes.
Initially, if LP(jj) and LQ(jj) are set to zero for all jj, then the initial estimate
of P(m2) and Q(m2) (m2 = 2, 3, . . . , NB) will be the sum of the loads of all nodes
beyond node m2 plus the load of m2 itself.
5. Load Flow Computation of RDN
Once P(m2) and Q(m2) are computed by the above logic (section 4), voltage
magnitudes of all the nodes can be easily computed using equation (3). Further real
and reactive power losses are obtained using equations (6) and (7). The complete
load ow algorithm is shown in the form of a ow chart in Figure 3.
The convergence criterion of the algorithm is that if, in successive iteration,
the dierence of real and reactive power delivered from the substation is less than
0.1kW and 0.1kVAr, respectively, the solution has converged.
D
o
w
n
l
o
a
d
e
d

B
y
:

[
B
-
o
n

C
o
n
s
o
r
t
i
u
m

-

2
0
0
7
]

A
t
:

0
3
:
0
4

2
9

O
c
t
o
b
e
r

2
0
0
9
Algorithm for Radial Distribution Networks 101
(a)
Figure 3. (a) Flow chart of load ow technique (continues).
D
o
w
n
l
o
a
d
e
d

B
y
:

[
B
-
o
n

C
o
n
s
o
r
t
i
u
m

-

2
0
0
7
]

A
t
:

0
3
:
0
4

2
9

O
c
t
o
b
e
r

2
0
0
9
102 R. Ranjan and D. Das
(b)
Figure 3. (b) Flow chart of load ow technique.
D
o
w
n
l
o
a
d
e
d

B
y
:

[
B
-
o
n

C
o
n
s
o
r
t
i
u
m

-

2
0
0
7
]

A
t
:

0
3
:
0
4

2
9

O
c
t
o
b
e
r

2
0
0
9
Algorithm for Radial Distribution Networks 103
6. Example
To demonstrate the eectiveness of the proposed algorithm, two examples are
selected. The rst example is a 69-node radial distribution network [9] as shown in
Figure 4 in the appendix. The system data is available in [9]. Results of the load
ow study are tabulated in Table 2.
For the load ow study, we have considered the following:
substation voltage = 12.66 kV,
base kVA = 10 and base voltage = 12.66 kV.
The second example is a 33-node radial distribution network [14]. Data are
given in Table 5 in the appendix. Load ow results of the system are given in
Table 3.
Comparison of relative speed and memory requirements is given in Table 4
below.
Table 2
Load ow result of 69-node radial distribution network
Voltage Voltage
Node magnitude Node magnitude
number (p.u.) number (p.u.)
1 1.00000 36 0.99992
2 0.99997 37 0.99975
3 0.99993 38 0.99959
4 0.99984 39 0.99954
5 0.99902 40 0.99884
6 0.99009 41 0.99884
7 0.98079 42 0.99855
8 0.97858 43 0.99850
9 0.97745 44 0.99850
10 0.97245 45 0.99841
11 0.971135 46 0.99840
12 0.96819 47 0.99979
13 0.96526 48 0.99854
14 0.96237 49 0.99470
15 0.995950 50 0.99415
16 0.95897 51 0.97854
17 0.95809 52 0.97853
18 0.95808 53 0.97466
19 0.95761 54 0.97142
20 0.95732 55 0.96694
21 0.95683 56 0.96257
22 0.95683 57 0.94010
23 0.95676 58 0.92904
24 0.95660 59 0.92476
25 0.95643 60 0.91974
(continued)
D
o
w
n
l
o
a
d
e
d

B
y
:

[
B
-
o
n

C
o
n
s
o
r
t
i
u
m

-

2
0
0
7
]

A
t
:

0
3
:
0
4

2
9

O
c
t
o
b
e
r

2
0
0
9
104 R. Ranjan and D. Das
Table 2
(Continued)
Voltage Voltage
Node magnitude Node magnitude
number (p.u.) number (p.u.)
26 0.95636 61 0.91234
27 0.95634 62 0.91205
28 0.99993 63 0.91166
29 0.99973 64 0.90976
30 0.99985 65 0.90919
31 0.99971 66 0.97129
32 0.99961 67 0.97129
33 0.99935 68 0.96786
34 0.99901 69 0.96786
35 0.99895
Total real power loss = 224.9606 kW; total reactive power
loss = 102.147 kVAr; minimum voltage observed at node 65,
|V65| = 0.90919 p.u.
Table 3
Load ow result of 33-node radial distribution network
Voltage Voltage
Node magnitude Node magnitude
number (p.u.) number (p.u.)
1 1.00000 18 0.90377
2 0.99703 19 0.99650
3 0.98289 20 0.99292
4 0.97538 21 0.99221
5 0.96796 22 0.99158
6 0.94948 23 0.97931
7 0.94595 24 0.97264
8 0.93230 25 0.96931
9 0.92597 26 0.94755
10 0.92009 27 0.94499
11 0.91922 28 0.93354
12 0.91771 29 0.92532
13 0.91153 30 0.92177
14 0.90924 31 0.91760
15 0.90782 32 0.91669
16 0.90643 33 0.91640
17 0.90439
Total real power loss = 210.998 kW; total reactive power
loss = 143.032 kVAr; minimum voltage observed at node 18,
|V18| = 0.90377 p.u.
D
o
w
n
l
o
a
d
e
d

B
y
:

[
B
-
o
n

C
o
n
s
o
r
t
i
u
m

-

2
0
0
7
]

A
t
:

0
3
:
0
4

2
9

O
c
t
o
b
e
r

2
0
0
9
Algorithm for Radial Distribution Networks 105
Table 4
Comparison of relative speed and memory
Method Relative memory Relative CPU time
Proposed method 1 1.0
D. Das et al. [12] 4 1.5
Baran and Wu [9] 6 2.9
It is reported in the literature that authors have tried to solve above systems
by NR (Newton-Raphson) and GS (Gauss-Siedel) methods but for both the cases
NR and GS did not converge.
7. Conclusions
In this paper a simple and ecient computer algorithm has been presented to solve
radial distribution networks. The proposed method has a good convergence property
for any practical distribution networks with practical R/X ratio. Computationally,
this method is extremely ecient, as compared to Baran and Wu [9] and Das
et al. [12], as it solves a simple algebraic recursive equation for voltage magnitude.
Another advantage of the proposed method is that all the data are stored in
vector form, thus saving an enormous amount of computer memory. The method is
successfully implemented on PIII with several realistic distribution networks. The
proposed algorithm can be used eectively with SCADA (supervisory control and
data acquisition) and DAC (distribution automation and control) as the algorithm
quickly solves the system and even suggests rerouting or network reconguration
for ecient operation of the system.
References
[1] W. F. Tinny and C. E. Hart, 1967, Power Flow Solution of the Newton Method,
IEEE Trans. PAS, Vol. PAS-86, No. 11.
[2] B. Stott and O. Alsac, 1974, Fast Decoupled Load Flow, IEEE Trans., Vol. PAS-93,
pp. 859869.
[3] B. Stott, 1984, Review of Load Flow Calculation Methods, Proc. IEEE, Vol. 62,
No. 7.
[4] D. Rajicic and Y. Tamura, 1988, A Modication to Fast Decoupled Power Flow for
Network with High R/X Ratios, IEEE Trans., Vol. PWRS-3, pp. 743746.
[5] S. C. Tripathy, D. Prasad, O. P. Malik, and G. S. Hope, 1982, Load Flow Solution for
Ill Conditioned Power Systems by Newton Like Method, IEEE Trans., Vol. PAS-101,
pp. 36843657.
[6] W. H. Kersting and D. L. Mendive, 1976, An Application of Ladder Network Theory
to the Solution of Three Phase Radial Load Flow Problem, IEEE PES Winter
Meeting.
[7] W. H. Kersting, 1984, A Method to Design and Operation of Distribution System,
IEEE Trans., Vol. PAS-103, pp. 19451952.
[8] R. A. Stevens, D. T. Rizy, and S. L. Puruker, 1986, Performance of Conventional
Power Flow Routines for Real Time Distribution Automation Applications, Proc.
of 18th Southeastern Symposium on Systems Theory, (IEEE), pp. 196200.
D
o
w
n
l
o
a
d
e
d

B
y
:

[
B
-
o
n

C
o
n
s
o
r
t
i
u
m

-

2
0
0
7
]

A
t
:

0
3
:
0
4

2
9

O
c
t
o
b
e
r

2
0
0
9
106 R. Ranjan and D. Das
[9] M. E. Baran and F. F. Wu, 1989, Optimal Sizing of Capacitor Placed on Radial
Distribution Systems, IEEE Trans., Vol. PWRD-2, pp. 735743.
[10] H. D. Chiang, 1991, A Decoupled Load Flow Method for the Distribution Power
Network Algorithm: Analysis and Convergence Study, Electrical Power and Energy
Systems, Vol. 13, No. 3, pp. 130138.
[11] D. Das, H. S. Nagi, and D. P. Kothari, 1994, Novel Method for Solving Radial
Distribution Networks, IEE Proc. C, Vol. 141, No. 4, pp. 291298.
[12] D. Das, D. P. Kothari, and A. Kalam, 1995, Simple and Ecient Method for Load
Flow Solution of Radial Distribution Networks, Electrical Power & Energy Systems,
Vol. 17, No. 5, pp. 335346.
[13] S. Bhowimik, S. K. Goswami, and P. K. Bhattacherjee, 2000, A New Power Distri-
bution System Planning through Reliability Evaluation Technique, Electrical Power
Systems Research, Vol. 24, pp. 169179.
[14] M. A. Kashem, V. Ganapathy, G. B. Jasmon, and M. I. Buhari, 2000, A Novel
Method for Loss Minimization in Distribution Networks, Proc. of International Con-
ference on Electric Utility Deregulation and Restructuring and Power Technologies,
pp. 251255.
[15] T. Gonen, 1986, Electric Power Distribution Systems Engineering, McGraw-Hill.
Appendix
A.1. 69-Node Radial Distribution Networks
Figure 4. 69-node RDN [9].
D
o
w
n
l
o
a
d
e
d

B
y
:

[
B
-
o
n

C
o
n
s
o
r
t
i
u
m

-

2
0
0
7
]

A
t
:

0
3
:
0
4

2
9

O
c
t
o
b
e
r

2
0
0
9
Algorithm for Radial Distribution Networks 107
A.2. Data for 33-Node Test Systems [14]
Table 5
Br. Send. Rec. Resis. Reac. Real load Reac. load
no. node node (ohm) (ohm) (kW) (kVAr)
1 1 2 0.0922 0.0477 100.0 60.0
2 2 3 0.4930 0.2511 90.0 40.0
3 3 4 0.3660 0.1864 120.0 80.0
4 4 5 0.3811 0.1941 60.0 30.0
5 5 6 0.8190 0.7070 60.0 20.0
6 6 7 0.1872 0.6188 200.0 100.0
7 7 8 1.7114 1.2351 200.0 100.0
8 8 9 1.0300 0.7400 60.0 20.0
9 9 10 1.0400 0.7400 60.0 20.0
10 10 11 0.1966 0.0650 45.0 30.0
11 11 12 0.3744 0.1238 60.0 35.0
12 12 13 1.4680 1.1550 60.0 35.0
13 13 14 0.5416 0.7129 120.0 80.0
14 14 15 0.5910 0.5260 60.0 10.0
15 15 16 0.7463 0.5450 60.0 20.0
16 16 17 1.2890 1.7210 60.0 20.0
17 17 18 0.7320 0.5740 90.0 40.0
18 2 19 0.1640 0.1565 90.0 40.0
19 19 20 1.5042 1.3554 90.0 40.0
20 20 21 0.4095 0.4784 90.0 40.0
21 21 22 0.7089 0.9373 90.0 40.0
22 3 23 0.4512 0.3083 90.0 50.0
23 23 24 0.8980 0.7091 420.0 200.0
24 24 25 0.8960 0.7011 420.0 200.0
25 6 26 0.2030 0.1034 60.0 25.0
26 26 27 0.2842 0.1447 60.0 25.0
27 27 28 1.0590 0.9337 60.0 20.0
28 28 29 0.8042 0.7006 120.0 70.0
29 29 30 0.5075 0.2585 200.0 600.0
30 30 31 0.9744 0.9630 150.0 70.0
31 31 32 0.3105 0.3619 210.0 100.0
32 32 33 0.3410 0.5302 60.0 40.0
D
o
w
n
l
o
a
d
e
d

B
y
:

[
B
-
o
n

C
o
n
s
o
r
t
i
u
m

-

2
0
0
7
]

A
t
:

0
3
:
0
4

2
9

O
c
t
o
b
e
r

2
0
0
9