Вы находитесь на странице: 1из 10

IET Generation, Transmission &

Distribution

An Efficient Sensitivity Based Approach for AC Transmission


Expansion Planning

GTD-2019-1081 | Research Article

Submitted by: Naga Yasasvi Puvvada, Abheejeet Mohapatra, Suresh Chandra Srivastava

Keywords: TRANSMISSION EXPANSION PLANNING, OPF, DCOPF, REACTIVE POWER DISPATCH AND PLANNING PROBLEM

PDF auto-generated using ReView


from
Auto-generated PDF by ReView IET Generation, Transmission & Distribution

IET Research Journals

An Efficient Sensitivity Based Approach for ISSN 1751-8644


doi: 0000000000

AC Transmission Expansion Planning


www.ietdl.org

P. Naga Yasasvi1∗ , Abheejeet Mohapatra1 , S. C. Srivastava1


1
Department of Electrical Engineering, IIT Kanpur, Kanpur, India 208016
* E-mail: yasasvi@iitk.ac.in

Abstract: AC Transmission Expansion Planning (TEP) is a complex problem to solve due to it’s large scale, mixed integer, non-
linear and non-convex characteristics. This paper presents an efficient sensitivity based approach, employing the Primal-Dual
Interior Point Method (PDIPM) for tackling integer variables in AC TEP. PDIPM’s Hessian is used to evaluate the sensitivities of
the objective function with respect to the line candidacies and thereby, finding a feasible AC TEP solution. The proposed approach
solves this non-convex Mixed-Integer Non-Linear Programming (MINLP) problem, without the need of any approximations/ relax-
ations in the AC TEP model. Also, the need and superiority of AC TEP over the conventional DC TEP is established by comparing
both using the proposed approach. Combined AC TEP with reactive power planning is also solved by the proposed approach.
Computational efficiency and superiority of the solution quality from the approach are confirmed by comparing the same with that
of the conventional branch and bound/ cut approach, Constructive Heuristic Algorithm (CHA), Merit Function (MF) based approach
and Particle Swarm Optimization (PSO). The studies performed in MATLAB and CPLEX for various test cases and the Northern
Regional Power Grid (NRPG) network of India, prove the efficacy and computational efficiency of this approach.

Nomenclature 1 Introduction
Transmission sector forms the critical backbone of the electric power
system network and hence, it’s expansion has to keep pace with
Cost of a circuit to be installed between bus i the rising power demand and associated generation capacity, so as
cij
and bus j ($/year or INR/year) to avoid power transmission bottlenecks, which otherwise may not
Number of new parallel circuits to be installed allow re-routing of power flows from power surplus to deficit areas.
nij
between bus i and bus j TEP thus, aids in determining the number, location and the installa-
Cost coefficients of real power generation of tion time of new transmission lines required in the power network,
ai , bi , ci to economically meet the forecasted load while subject to various
ith generator
Number of existing circuits in the base physical and system operating constraints [1].
n0ij The first linear model of TEP was given by Garver [2] in 1970.
configuration between bus i and bus j
Since then, several works in the literature have studied the pros and
min/max Minimum/ maximum number of circuits cons of the various TEP models. [3] presents the formulation of the
nij
between bus i and bus j mixed integer TEP problem as the highly complex AC model and
Set of all transmission circuits ij, where bus i the simplified DC model. In TEP, an interesting as well as chal-
Ωs
is the sending end in the network topology lenging fact is that both DC TEP and AC TEP formulations have
Set of all transmission circuits ij, where bus i line candidacies as integer variables, which makes them nonlin-
Ωr ear and further non-convex. Several solution approaches have been
is the receiving end in the network topology
Ω Ωs ∪ Ωr developed to provide cost effective solutions for TEP [1]. These
θi Voltage angle of bus i (rad) can be grouped into mathematical optimization techniques, such as
PGi /PDi Real power generation/ demand at bus i (pu) linear [2], nonlinear [4], mixed integer programming [5], and meta-
heuristic approaches such as harmony search algorithm [6], PSO [7],
QGi /QDi Reactive power generation/ demand at bus i (pu)
genetic algorithm [8], artificial neural networks [9], etc., and their
Vi Voltage phasor of bus i (pu) variants [1]. Meta heuristic approaches, in general, suffer from their
ei /fi Real/ imaginary parts of Vi (pu) computational efficiency and multiple solution attributes [1].
Nb Number of buses DC model and its variants [10] have been popularly used for TEP.
NG Number of buses with generation units An attempt to extend DC TEP [11] to AC TEP [12] was first carried
PGi
min/max
Minimum/ maximum real power generation at bus i out using CHA. CHA is an iterative process which adds a single cir-
min/max cuit in each step based on a sensitivity. This sensitivity is based on
QGi Minimum/ maximum reactive power generation at bus i the power flows in the circuits. A line with the highest power flow
min/max
Vi Minimum/ maximum voltage at bus i is assumed to be the most attractive circuit and a new line is added
min/max
θi Minimum/ maximum voltage angle at bus i in the same corridor. Further, removal of lines is also done in one
Conductance, susceptance and half line charging by one fashion, in the decreasing order of the line cost. The feasi-
gij , bij , bsh
ij bility of addition/ removal of a line is checked by solving a relaxed
susceptance of line between bus i and bus j (pu)
AC TEP in every iteration of CHA. The solution of the relaxed AC
bsh
i Shunt susceptance at bus i (pu) TEP is also used for evaluation of sensitivity of each line. Since, line
Pij Real power flow between bus i and bus j (pu) addition/ removal is done in one by one fashion based on the power
max
Pij Maximum real power flow between bus i and bus j (pu) flow sensitivity, such an algorithm is expected to be computation-
Pi (∗) Real power injection at bus i (pu) ally burdensome and may end up giving infeasible TEP plans during
Qi (∗) Reactive power injection at bus i (pu) constrained reactive power situations. Also, the TEP plan by CHA

IET Research Journals, pp. 1–9


© The Institution of Engineering and Technology 2015 1

ACTEPs ensitivitya nalysis.pdf M ainDocument IET Review Copy Only 2


Auto-generated PDF by ReView IET Generation, Transmission & Distribution

for large and complex networks is usually non-optimal [12]. More approach is highly efficient and is capable of giving TEP plans for
DC TEP solution approaches can be seen in [1, 13]. Although, DC cases when CHA and MF based approach may fail. For the cases
TEP is simple and easy to implement, it has the following disadvan- where solutions exist, the proposed approach gives solutions, which
tages: (i) requires line reinforcement stage for the implementation of are not inferior, in much lesser CPU times for all the considered test
DC TEP plan in AC network, (ii) requires implementation of sep- cases in this paper. The capability of the approach to give feasible
arate reactive power allocation problem, and (iii) system losses are TEP plans is expected due to it’s similarity to the predictor-corrector
ignored. Thus, it is safe to conclude that DC TEP plan when imple- philosophy.
mented for AC network may lead to infeasible or non-optimal system The rest of the paper is as follows. Section 2 discusses the AC TEP
operation [5]. and DC TEP models. The proposed approach for solution to AC TEP
In comparison to DC TEP, obtaining AC TEP plan is a her- is discussed in section 3. Test results for a 3 bus, Garver 6 bus with
culean task due to the associated nonlinearity and non-convexity. and without initial network, IEEE RTS 24 bus, IEEE 118 bus and
Despite this, the use of AC TEP over DC TEP is justified [5, 14–16] the NRPG networks are given in section 4. Section 5 concludes the
due to the following advantages: (i) gives practically implementable paper.
plan for the AC network, (ii) enables the planner to do TEP along
with reactive power allocation in the presence of reactive resources
while considering voltage stability effects, and (iii) aids the plan- 2 Transmission Expansion Planning (TEP)
ner to perform line reinforcement when DC TEP plan is used for
AC network. Due to these benefits, various approaches in the lit- TEP is solved to determine the number and location of new trans-
erature for AC TEP have focused on handling the nonlinearity and mission lines to be added to the power system network so as to meet
non-convexity through relaxations and binary variable search space the requirement of balance between future demand and generation
or approximations. These include AC TEP with linear relaxations capacities. A short-term AC TEP can be formulated [5] as
or approximations [16–21], conic programming [5, 22] and AC NG 
based differential evolution TEP [14]. [17] claims of a new linear X X 
2
relaxation for AC TEP being better than existing linear DC mod- min F = cij nij + ai PGi + bi PGi + ci (1)
els. In [18], an improved relaxed AC model has been considered in ij∈Ωs i=1
which linear expressions of network losses and reactive powers are s.t. Pi (e, f , n) − PGi + PDi = 0, ∀ i = 1, . . . , Nb (2)
retained. Further, a linearized AC TEP formulation has been pro-
posed in [19] via Taylor series and piecewise linear approximations. Qi (e, f , n) − QGi + QDi = 0, ∀ i = 1, . . . , Nb (3)
A special ordered set approach to linearize the AC TEP has been con- min
PGi max
≤ PGi ≤ PGi , ∀ i ∈ 1, . . . , NG (4)
sidered in [20]. [21] presents MILP formulation for AC TEP through
piecewise approximation. The flaw with these approaches is that fur- Qmin
≤ QGi ≤
Gi Qmax
∀ i ∈ 1, . . . , NG
Gi , (5)
ther reinforcement is required in [16–21] to yield feasible solutions. 2
min 2 2 max 2
Also, although obtaining AC TEP plans with relaxations makes the Vi ≤ ei + fi ≤ Vi , ∀ i = 1, . . . , Nb (6)
solution process easy due to reduced complexity, but problems such
max
as high optimality gap and computational burden still need to be −(nij + n0ij )P ij ≤ (nij + n0ij )Pij , ∀ ij ∈ Ωs (7)
addressed. In [22], a bilevel conic AC TEP formulation has been
proposed which however, does not ensure zero optimality gap. Also, (nij + n0ij )Pij ≤ (nij + n0ij )Pij
max
, ∀ ij ∈ Ωs (8)
the formulation is inexact for mesh networks as cyclic constraints are
ignored. In [5], the mixed-integer conic programming problem has nmin
ij ≤ nij ≤ nmax
ij , ∀ ij ∈ Ωs (9)
been solved using a branch and cut method to obtain AC TEP plans.
The objective function F in (1) consists of line investment cost
Usually, branch and bound/ cutting plane approaches have been used
and generation cost. Line investment cost is the cost associated with
to solve MILP problems, and such approaches become computa-
all new lines to be installed. The generation cost is the cost of real
tionally prohibitive as the network size increases [23]. Besides, the
power generation of all NG generators. Let the added circuit or line
approach in [5] tends to be computationally burdensome due to the
vector be n = {nij } ∀ ij ∈ Ωs and nij ∈ Z. (2) and (3) represent
binary variables used to eliminate nonlinearity.
the real and reactive power balance equations, respectively, for each
Among the existing generic DC and AC TEP models in litera-
bus. The injection equations are given as
ture, solving AC TEP model is difficult due to its high degree of
nonlinearity and non-convexity. In order to get AC TEP solution,
Pi (e, f , n) = (e2i + fi2 )Gii (n) + (10)
approximations or relaxations have been used and further solved X
using existing solvers. However, as stated earlier, use of approxima- {(ei ej + fi fj )Gij (n) − (ei fj − ej fi )Bij (n)}
tions and/ or relaxations in AC TEP need not always yield feasible ij∈Ω
plans. Thus, further research effort is required to find feasible and (11)
practical solution to AC TEP with reduced computational burden.
Hence, the primary aim of this paper is to efficiently find the fea- Qi (e, f , n) = −(e2i + fi2 )Bii (n) −
sible solution of the exact mixed integer nonlinear, non-convex AC X
TEP, without the use of any relaxations/ approximations. To do so, {(ei ej + fi fj )Bij (n) + (ei fj − ej fi )Gij (n)}
an efficient and novel PDIPM sensitivity based approach has been ij∈Ω
proposed here for solving AC TEP. The sensitivities are obtained (12)
from the Hessian of PDIPM, which play a major role in directing the
integer transmission line candidacies to a feasible solution. The sen- where, e = {ei } ∀i = 1, . . . , Nb and f = {fi } ∀i = 1, . . . , Nb .
sitivities provide the effect of movement of integer variables in the The search space of AC TEP is highly nonlinear and non-convex
feasible region on the objective function. Based on the sign of sen- due to the presence of third order power balance equations. The bus
sitivities, the integer variables are adjusted to their adjacent lower admittance matrix elements vary with installation of new lines and
or upper settings, and further re-optimization is done if required, to are given as
find TEP plans. The secondary aim of this paper is to justify the   X 
need of using AC TEP instead of DC TEP by comparing the plans Gij (n) = − nij gij + n0ij gij 0
, Gii (n) = nij gij + n0ij gij
0

obtained for both from the proposed approach. Further, the proposed j6=i
 
approach has also been used to obtain solution of AC TEP combined
Bij (n) = − nij bij + n0ij b0ij (13)
with reactive power planning. Comparisons of results by the pro-
posed approach for several test cases have been done with the same X     0 
obtained from existing branch and bound/ cut approach, CHA [12], Bii (n) = bsh
i + nij b ij + b sh
ij + n0
ij b 0
ij + bsh
ij
MF based approach [24] and PSO [7]. This proves that the proposed j6=i

IET Research Journals, pp. 1–9


2 © The Institution of Engineering and Technology 2015

ACTEPs ensitivitya nalysis.pdf M ainDocument IET Review Copy Only 3


Auto-generated PDF by ReView IET Generation, Transmission & Distribution

where, i, j ∈ 1, 2, . . . , Nb . The bus admittance matrix elements in change in an integer variable di on the feasibility and optimality of
(13) include admittances of existing lines (base case represented by the KKT conditions of TEP formulation [26], which may be posi-
superscript 0) and new transmission lines. Limits on real and reac- tive, negative or zero. A zero indirect sensitivity would indicate that
tive power generations and bus voltage magnitudes are given by (4), the feasibility and optimality of the KKT conditions are ensured for
(5) and (6), respectively. Limits on real power flows of lines (exist- changes in the local neighborhood of di . A vice-versa inference can
ing and new) are given by (7) and (8) where, Pij = Re{Vi∗ (Vi − be made for a non-zero indirect sensitivity. It is also to be observed
F
Vj )(gij + ĵbij )} and Vi = ei + ĵfi , where ĵ is the complex oper- that the absolute value of the indirect sensitivity Sd2 (i) will be quite
ator. Maximum permitted lines in each corridor is specified by (9). F
small as compared to the inherently positive direct sensitivity Sd1 (i)
The large non-convex search space of AC TEP is defined by (2)- as transmission investment cost is generally very high as compared
(9), from which optimal line installations are to be determined. It to the generation cost [12]. Thus, SdF (i) for the AC TEP will be pos-
consists of line candidacies as integer variables while real and imag- itive. The same shall also be seen from the results obtained in the
inary parts of bus voltages, real and reactive power generations as next section.
continuous variables. The DC TEP can be formulated as Further, the sensitivities in (19) are acceptable only for close
proximities near the current solution d [24]. Hence, the maximum
min (1) s.t. (4), (9) (14) allowable integer variable movement at any iteration of this approach
has been limited to either the adjacent upper or lower integer. Fur-
Pi (θ, n) − PGi + PDi = 0, ∀ i = 1, . . . , Nb (15) ther, any integer solution will always be costly as compared to the
−(nij + n0ij )Pij
max
≤ Pij (θ, n), ∀ ij ∈ Ωs (16) cost obtained with all variables assumed as continuous. Hence, the
motivation behind using the sensitivities from the converged Hes-
Pij (θ, n) ≤ (nij + n0ij )Pij
max
, ∀ ij ∈ Ωs (17) sian of PDIPM is to obtain an integer solution. While doing so,
the change in objective function F , with respect to the integer vari-
θimin ≤ θi ≤ θimax , ∀ i = 1, . . . , Nb (18) ables obtained in the beginning of this approach (with all variables
assumed as continuous), be as minimum as possible. Through the use
where, real power balance equation Pfor each bus is given by (15). of these sensitivities, feasibility and optimality of all KKT conditions
Real power injection Pi (θ, n) = ij∈Ω (θi − θj )(nij + n0ij )bij [26] are also ensured.
is a nonlinear function of new transmission lines nij and polar Thus, at latest d, for each di , adjacent lower and upper integer val-
bus voltage angles θ = {θi }, ∀i = 1, . . . , Nb . Limits on real power ues are chosen. If di is non-integer, then adjacent lower and upper
flows are imposed using (16) and (17), where Pij (θ, n) = (θi − integer values are (dji , dj+1 ) so that dji ≤ di ≤ dj+1 . If di is inte-
i i
θj )(nij + n0ij )bij . Bounds on voltage angles are given by (18). j
ger, i.e. di = di , in any iteration of this approach, then associated
Variable vector of DC TEP consists of line candidacies as inte-
integer values are (dij−1 , dj+1 ), where j refers to the j th integer
ger variables and bus voltage angles, real power generations as i
value.
continuous variables.
Intuitively, if SdF (i) in (19) is negative, then in order to have least
It can be observed that the above AC and DC TEP formula-
change in F , the integer variable di should be set at its upper adja-
tions consider new transmission line candidacies as integer vari-
ables, instead of binary variables as in [5]. This consideration is cent value. If SdF (i) is positive, then the integer variable should be
an advantage in PDIPM, as in PDIPM, the computational effort set at its lower adjacent value. However, as discussed earlier, SdF (i)
involved in the factorization of the Hessian matrix increases with is generally positive. Thus, for any integer variable di , the associated
the number of variables [25]. However, it should be noted that the discretization shall always be set at its adjacent lower integer value,
proposed approach can also be used for getting TEP plans with line which is not desirable and may lead to infeasibility in the interme-
candidacies considered as binary variables. diate iterations of the proposed approach (the same has also been
shown in the next section). Hence, the dominant positive first term
F
Sd1 (i) in SdF (i) is neglected during the discretization. Only Sd2 F
(i)
3 Sensitivity Based Approach for AC TEP is used for the discretization of the integer line candidacy variables
in AC TEP. The impact of the neglected dominant first sensitivity is
The process of getting AC TEP plan by the proposed sensitivity inherently considered in the proposed approach, which is discussed
based approach is discussed here. The advantage of this approach later.
is that no conic or linear relaxations/ DC model based approxima- A peculiar characteristic has been observed in the AC TEP (as
tions are required in AC TEP to get the associated plan. It begins by well as in DC TEP) sensitivities obtained from Hessian of PDIPM,
solving (1)-(9) by PDIPM (details in Appendix) while assuming all where zero or near zero valued sensitivities are also present. Such
integer variables to be continuous. It is to be noted that this assump- sensitivities imply that the positive or negative movements of the
tion is not a relaxation/ approximation of the exact AC TEP. This is integer variable in the local neighborhood of the current solution
a requirement only in the initial phase of the proposed approach and does not change the objective function while also ensuring opti-
the effect of such an assumption gets nullified in the later phase of mality and feasibility around the local neighborhood of the current
the approach. solution. Hence, if the corresponding sensitivity is zero, the associ-
At the converged continuous solution set, the sensitivities of F ated line candidacy integer variable is given a higher priority and it
with respect to the integer line candidacy variables d from the con- is discretized to adjacent upper integer value (as for zero sensitiv-
verged factorized Hessian of PDIPM, in vector form, can be stated ity, both lower and upper movements are possible) keeping in view
as the feasibility in every iteration of the proposed approach. The dis-
 
∂F ∂F ∂x ∂ r˜y ∂x ∂rλ cretization process in case of positive and/ or negative sensitivities
SFd = + + (19) remains the same as discussed earlier.
∂d ∂x ∂ r˜y ∂d ∂rλ ∂d
Having discretized the line candidacies, a fresh AC TEP is solved
where, ∂x/∂ r˜y and ∂x/∂rλ are the appropriate sub-matrices cho- by PDIPM, where the base topology for a given corridor ij ∈ Ωs
sen from K11 and K12 in (21), respectively. now includes the latest topology before discretization and the recent
It can be seen from (19) that SdF (i), the sensitivity of F with lines, which are to be added after discretization. The new line can-
respect to an integer variable di consists of two terms. The first term didacies are still assumed to be continuous. Also, the lower limit
F
Sd1 (i) is the direct sensitivity of F with respect to di , which is noth- of the line candidacies for every corridor in (9) is now assigned as
ing but the marginal cost of constructing a line on a given corridor negative of nij where, nij is the latest integer value after discretiza-
(as can be seen from (1) where di refers to the line candidacy on a tion instead of 0. This is necessary so as to give the new AC TEP
given corridor). This marginal cost will always be positive, as cost (to be solved by PDIPM) with new base topology the freedom to
of constructing a line cannot be negative or zero. The second term remove any costlier lines which are added in the current iteration of
F
Sd2 (i) in (19) is an indirect sensitivity and is derived from the con- the approach and replace them with low cost lines in the subsequent
verged Hessian of PDIPM. This essentially represents the impact of

IET Research Journals, pp. 1–9


© The Institution of Engineering and Technology 2015 3

ACTEPs ensitivitya nalysis.pdf M ainDocument IET Review Copy Only 4


Auto-generated PDF by ReView IET Generation, Transmission & Distribution

iterations. It is to be noted that by solving a fresh AC TEP (as dis- and 7 are associated with the correction phase. As will be seen in the
cussed above), the effect of direct sensitivities are inherently taken results, this approach has been successful in terms of the predictor-
care as the objective in AC TEP is to minimize the total cost F . The corrector philosophy as the predictor step coupled with the corrector
overall algorithm for this approach is thus as follows step helps in tracking the feasible and superior TEP plans for all test
cases, as compared to existing approaches. Associated results from
1. Read system data. the proposed approach are discussed next.
2. Let n0ij be assigned with base topology and lower limit of each
new line candidacy be zero as in (9).
3. Solve AC TEP (1)-(9) using PDIPM with all variables assumed 4 Numerical Results
as continuous.
4. Evaluate PDIPM based sensitivities (19) of objective function The proposed PDIPM sensitivity based approach for solving AC
with respect to integer variables at the converged solution. TEP (and DC TEP) has been implemented and the results have
5. Assign an integer value to each variable di based on the sign of been compared with those obtained from the conventional branch
F
Sd2 (i) and accordingly for the zero sensitivities. Let nij be vector and bound/ cut method for solving MILP/ MINLP problems, CHA
of new lines to be installed. [12], MF based approach [24] and PSO [7] on five test systems: a
6. Reassign n0ij as n0ij of previous iteration and nij obtained in step 3 bus, Garver 6 bus, green field expansion of Garver 6 bus (6-gf),
5. Also, lower limit of line candidacies are set to be −nij . IEEE RTS 24 bus and IEEE 118 bus systems. The test systems’
7. Solve AC TEP using PDIPM to find optimum values of y and F , data are given in [28]. Maximum and minimum voltage limits for
with d set to latest n0ij . all test systems (except IEEE 118 bus system) are 1.1pu and 0.9pu,
8. Stop, if no new lines come up and change in objective is less than respectively. The proposed approach has also been validated on the
tolerance (ξ), else repeat from step 4. Indian NRPG system. The computing platforms used are MATLAB
and CPLEX API for MATLAB [29] on an Intel Core I5-7400 CPU
at 3.00 GHz and hwith 8 GB RAM.i PDIPM’s convergence tolerance
Start is of the order of 10−6 , 10−3 .

Input Data: Network topology 4.1 Test systems’ data


Network parameters
Line costs, iter = 0 4.1.1 3 bus system: This system as in [28] has 3 existing lines.
It is assumed to be expanded for a future condition where load is
to be increased by 50%. 3 line corridors are considered with max-
Let, nij0 (iter ) is base topology imum of 3 lines per corridor. Associated costs are - branch L1−2
nijmin (iter ) = 0, " ij Î W s - 6800$/year, branch L1−3 - 6000$/year and branch L2−3 -
4000$/year (also shown in Table 1).
Solve ACTEP (1)-(9) continuous relaxation,
using PDIPM and evaluate objective function F (iter )
Table 1 Transmission corridors of test systems
iter = 1 Line Potential Existing Investment
Bus
corridors circuits circuits cost range ($/year)
Calculate PDIPM based sensitivity SdF (19) 3 3 9 3 [4, 6.8] × 103
6 15 75 6 [20, 68] × 103
Evaluate nij (iter ) by assigning integer value to 6-gf 15 75 0 [20, 68] × 103
each integer variable based on the sign of S dF2
24 41 123 38 [3, 134] × 106
118 186 930 193 [0.911, 61.725] × 106
Solve ACTEP (1)-(9)
iter = iter + 1 continuous relaxation using PDIPM with,
nij0 (iter ) = nij0 (iter - 1) + nij (iter )
nijmin (iter ) = -nij (iter ) 4.1.2 Garver 6 bus system: Garver system [12] has 15 right-
of-ways with total demand of (7.6 + j1.52)pu. Maximum lines in
each corridor are assumed to be 5 (and hence, 75 potential circuits
in Table 1). In the base topology, a generating unit is placed at
bus 6, which is not connected to the rest of the system. Associated
if (F (iter ) - F (iter - 1) <x )
No
generation cost characteristics for this system are given in [28].
&&
(nij (iter ) = 0) Another test system is the green field expansion of 6 bus system
(6-gf) [10], which is an unconnected system and for such systems,
Yes
obtaining TEP plan is generally computationally difficult. Potential
and existing circuits with associated investment costs are given in
Stop
Table 1.
Fig. 1: Flowchart of the proposed approach for AC TEP 4.1.3 IEEE RTS 24 bus system [28]: This system has been
considered for a future condition with generations and loads being
3 times of actual values. Total demand in the system is (85.5 +
Fig. 1 shows the flowchart of the proposed PDIPM sensitivity based j17.4)pu. Potential circuits and associated costs are given in Table
approach to solve AC TEP. It is to be noted that this can also be used 1. Only one type of line is considered and it is assumed that up to
for solving DC TEP (14)-(18). 3 new lines can be installed per corridor. Associated generation cost
The design of the proposed approach has been inspired from the characteristics are from [28].
concept of the predictor-corrector philosophy [27]. The basic idea
of this approach is to extrapolate the desired quantity in predic- 4.1.4 IEEE 118 bus system [28], [30]: This system has also
tor step and refine the prior step prediction using another means in been considered for a future condition with generations and loads
corrector step so that feasibility is ensured. Referring to the afore- being 3 times of actual values. Maximum and minimum voltage
mentioned algorithm of the proposed approach, steps 4 and 5 come limits are 1.06pu and 0.94pu, respectively. Potential circuits and
under the prediction step of the line reinforcements while steps 6 associated costs are given in Table 1. The investment cost can be

IET Research Journals, pp. 1–9


4 © The Institution of Engineering and Technology 2015

ACTEPs ensitivitya nalysis.pdf M ainDocument IET Review Copy Only 5


Auto-generated PDF by ReView IET Generation, Transmission & Distribution

estimated by parameters like cost per mile, total length and cost mul- For the Graver 6 bus system, a significant difference in the final
tipliers [31]. Assuming that all the transmission lines are 230 kV AC and DC TEP plans can be observed, as shown in Table 5. Ini-
double circuit lines, the investment cost of a transmission line is tially, the continuous variable solution y for Graver 6 bus system is
determined as obtained by solving AC TEP using PDIPM. At the converged solu-
tion, the sensitivities SF
d calculated using (19) and associated second
Cost of Line = P × M × (Length of Line) (20) term sensitivities SF F
d2 in (19) are given in Table 4. Sd evaluated
using (19) is always positive for all lines due to the inherent nature
For 230 kV double circuit lines [31], P = 1.5 × 106 $/mile has of TEP. Hence, integer line candidacies if set to their adjacent lower
been considered as cost per mile and M is the cost multiplier of settings, lead to an infeasible situation in the subsequent iteration
a line based on its length. It is to be noted that (20) gives a rough of the proposed approach. Later, based on the sensitivity analysis
estimate of the line investment cost in dollars, while other factors using SF d2 and by resolving AC TEP by the proposed approach, the
considered in [31] may also be included to obtain a better estimate. planning process results in a line investment of 160 × 103 $/year
and generation cost of 1587.224$/year. The proposed approach for
4.2 AC TEP vs DC TEP by proposed approach the Garver 6 bus system converges in 2 iterations with total CPU
time being less than 0.3s for AC TEP. On similar lines, the DC
Results for AC and DC TEP for the above five test systems have TEP plan for the Garver 6 bus system and the AC TEP and DC TEP
been obtained using the proposed approach in MATLAB. For the 3 plans for the 6-gf bus system have been obtained using the proposed
bus system, the continuous variable AC TEP solution y obtained by approach. It can be seen from Table 5 that the solution of DC TEP
solving it with PDIPM is given in Table 2. At this solution, volt- gives 31.04% and 35.79% reduction in the total cost with respect to
age magnitude at bus 1 is at its maximum limit of 1.1pu. Also, real the solution of AC TEP for the Garver 6 bus system with and with-
power generation at bus 2 is at its maximum limit. Objective F at out an initial network topology, respectively. CPU time for DC TEP
this solution is 8.8226 × 103 $/year. plan of 6 bus system is less than 0.15s. As green field expansion
studies are computationally more challenging to solve, AC TEP for
Garver 6 bus system with green field expansion takes CPU time of
Table 2 Continuous variable solution for 3 bus system
0.55 − 0.6s while the same in DC TEP is 0.3 − 0.35s.
y Value (pu) y Value (pu)
e1 1.1000 Pg2 1.5 Table 4 Sensitivities of F for Garver 6 bus system
e2 1.0653 Qg1 0.400
e3 0.9606 Qg2 0.6139 Line SF
d SF
d2 Line SF
d SF
d2
f1 0 L1−2 3.16e-12 L1−2 40006.82 6.821 L2−6 29999.99 -1.816e-9
f2 -0.0642 L1−3 0.2566 L1−3 38016.99 16.996 L3−4 58855.09 -144.90
f3 -0.1549 L2−3 0.5915 L1−4 60016.20 16.207 L3−5 20000 5.178e-11
Pg1 1.7420 L1−5 19940.77 -59.226 L3−6 48246.23 246.227
L1−6 68161.81 161.811 L4−5 62948.53 -51.472
L2−3 20269.72 269.723 L4−6 29999.99 -1.77e-10
The sensitivities of F with respect to the line candidacies at this L2−4 40003.81 3.816 L5−6 61114.27 114.274
converged solution using (19) and the associated second term SF d2 L2−5 30948.50 -51.493
in (19) (i.e. excluding the first direct sensitivity term SF F
d1 in Sd )
F
are shown in Table 3. The sensitivity vector Sd evaluated using (19)

Similarly, for the IEEE RTS 24 bus and the IEEE 118 bus systems,
Table 3 Sensitivities of F for 3 bus system the associated TEP plans are shown in Table 5. DC TEP plans have
Sensitivity L1−2 L1−3 L2−3 71.76% and 62.80% reduction in total cost as compared to the AC
TEP plans, respectively with significant difference in the final line
SF
d 6.8072 × 103 6.000 × 103 4.000 × 103 candidacies. As the system size increases, the computational time of
SF
d2 7.1855 −6.926 × 10−14 4.359 × 10−9 AC TEP increases, i.e. for the IEEE RTS 24 bus system and IEEE
118 bus system, CPU time for obtaining AC TEP plan is more than 5
times and 25 times than the same for the DC TEP plan, respectively.
is positive for all lines. If these sensitivities are considered for dis- However, it has been observed in all the test cases (except the 3
cretization of line candidacies, then no new lines should come up bus system) that the converged DC TEP plan when implemented for
as the continuous line candidacy solution for all lines lies between the actual AC network, leads to infeasibile system operation. Infea-
0 and 1, as can be observed from Table 2. Since no new lines are sibility and/ or non-optimality has been checked by evaluating the
added, the solution in the next iteration of the proposed approach system constraints after obtaining the power flow solution of actual
is infeasible. However, based on SF d2 (second term) sensitivities, AC network with DC TEP plans. The same has also been checked by
a line each is added in corridors L1−3 and L2−3 at an invest- solving AC Optimal Power Flow (OPF) with fixed TEP plans from
ment cost of 10, 000$/year. Thereafter, AC TEP with the proposed DC TEP for evaluation of generation cost. Also, low CPU time of
approach and SF d2 sensitivities for the 3 bus system converges in DC TEP over AC TEP, at the expense of infeasible and non-optimal
2 iterations as no new lines come up and there is no change in the plans, has no significance. The proposed approach for AC TEP is,
objective function value. Finally, the optimized cost (investment cost thus, a good alternative. Also, the results for various test systems in
and generation cost) obtained to meet the given future condition is the following sections prove that the proposed approach gives feasi-
14.917 × 103 $/year as shown in Table 5. Total CPU time involved ble and superior AC TEP plans with lesser CPU times as compared to
in attaining the AC TEP plan for the 3 bus system by the proposed existing branch and bound/ cut approach, CHA, MF based approach
approach is less than 0.15s. and PSO. This, hence, justifies the need of solving AC TEP instead
A similar study using the proposed approach has been performed of DC TEP.
for DC TEP of the 3 bus system and the associated plan is given in Convergence characteristics of the proposed approach for solving
Table 5. It can be observed that there is a small change in the gener- AC TEP for the various test systems, is shown in Fig. 2. F c is the
ation cost with no effect on the TEP plan and associated investment objective value obtained by solving AC TEP with continuous line
cost. This is because the real power loss and reactive power effect in candidacies. On this continuous solution set, PDIPM Hessian based
AC model as compared to DC model for the 3 bus system is negligi- sensitivities of the objective with respect to line candidacies are used
ble. The total CPU time involved to obtain the DC TEP plan for the for discretization of line candidacies. The final objective function
3 bus system by the proposed approach is less than 0.1s. F , thus obtained, is always more than F c. However, the sensitivities

IET Research Journals, pp. 1–9


© The Institution of Engineering and Technology 2015 5

ACTEPs ensitivitya nalysis.pdf M ainDocument IET Review Copy Only 6


Auto-generated PDF by ReView IET Generation, Transmission & Distribution

Table 5 AC TEP solutions vs DC TEP solutions by proposed approach


Number of
System Model Final line candidacy Line cost ($/year) Total cost ($/year) Time T (s)
variables
AC n1−3 =1, n2−3 =1 10 × 103 14.917 × 103 12 0.14-0.15
3
DC n1−3 =1, n2−3 =1 10 × 103 14.380 × 103 7 0.07-0.08

AC n2−6 =2, n3−5 =2, n4−6 =2 1.6 × 105 1.6158 × 105 32 0.25-0.30
6
DC n3−5 =1, n4−6 =3 1.1 × 105 1.1142 × 105 23 0.14-0.15

AC n1−5 =1, n2−3 =1, n2−6 =3, n3−5 =2, n4−6 =2 2.5 × 105 2.5154 × 105 32 0.55-0.60
6-gf
DC n2−6 =2, n3−5 =2, n4−6 =2 1.6 × 105 1.6151 × 105 23 0.30-0.35

AC n3−24 =1, n6−10 =1, n7−8 =2, n15−24 =1 1.70 × 108 1.70002687 × 108 108 2.8-3.0
24
DC n6−10 =1, n7−8 =2 4.8 × 107 4.8002675 × 107 74 0.4-0.6

n1−3 =1, n3−5 =1, n4−5 =1, n5−6 =1,


AC n8−5 =1, n5−11 =1, n45−46 =1, 1.2632 × 108 1.263361 × 108 529 200-250
118 n77−78 =1, n85−86 =1, n87−86 =1
DC n3−5 =1, n5−6 =1, n8−5 =1, n5−11 =1, n77−78 =1 0.4699 × 108 0.470029 × 108 357 6-7

always aim to provide a discrete AC TEP solution, so that the change resources in the system for which TEP is to be done, indicating that
in F with respect to F c is as minimum as possible. the desired solution does not exist and this has nothing to do with the
proposed approach. This is also evidenced by the discussions in the
following sections.
3.5

4.3 AC TEP with reactive power planning


3 IEEE 24 Bus The proposed approach has also been tested to find solution to
Garver 6 Bus
Garver 6-gf Bus
AC TEP combined with reactive power planning for 3 bus, Garver
3 Bus 6 bus with and without initial topology, IEEE 24 bus and IEEE
2.5 IEEE 118 Bus 118 bus systems, by using the combined TEP and reactive source
optimization model as given in [5].
F/Fc

2
Table 7 Combined AC TEP and reactive power planning solutions vs DC TEP
solutions by proposed approach
1.5 System Final line Total cost % increase
candidacy ($/year) w.r.t DC TEP
1 3 n1−3 =1, n2−3 =1 14.917 × 103 0
0 1 2 5 10 20 30
Number of Iterations
6 n2−6 =1, n3−5 =1, n4−6 =2 1.2885 × 105 15.64
Fig. 2: Convergence characteristics of AC TEP model for various
test systems n1−5 =1, n2−3 =1,
6-gf
n2−6 =1, n3−5 =2, n4−6 =2 2.0614 × 105 27.63
Table 6 Optimized continuous state cost Fc and discrete state cost F
System Fc ($/year) F ($/year) 24 n7−8 =1, n16−19 =1, 5.2685 × 107 9.76
3 3
3 8.8226 × 10 14.917 × 10 118 n77−78 =1,n85−86 =1, 0.5317 × 10 8
13.12
6 1.4659 × 105 1.6158 × 105 n86−87 =1
6-gf 2.0269 × 105 2.5154 × 105
24 1.3206 × 108 1.70002687 × 108 It is to be noted that combined AC TEP with reactive power plan-
118 0.3679 × 108 1.263361 × 108 ning is still a MINLP problem and hence, the proposed approach
can be easily used to solve the problem, on lines of discussion as
given in section 3 and without any additional changes. The advantage
The change in F with respect to F c is significant as the costs of the combined AC TEP with reactive power planning is obvi-
associated with integer variables are very high as shown in Table 6. ous, as it helps in attaining a TEP plan with significantly lesser
From Fig. 2, it can be observed that the proposed approach gives AC increase in investment cost with respect to the DC TEP plan. The
TEP plans in 2 iterations for the various test systems which indicates reactive power planning coefficients considered here are c0i =$100
that the proposed approach does not cycle and has smooth conver- and c1i =0.3$/kVAr for 3 bus and Garver 6 bus systems, c0i =$1000
gence. This is further evidenced by the fact that the approach is based and c1i =3$/kVAr for IEEE 24 bus and IEEE 118 bus systems. The
on the predictor-corrector philosophy. In case of non-convergence of reactive resource limits considered for 3 bus and Garver 6 bus sys-
TEP, the proposed approach does not provide distance to optimality. tems are -0.1 p.u and 0.5 p.u. For IEEE 24 bus and IEEE 118 bus
Albeit, any difficulty in convergence of the approach for a test sys- systems, reactive resources with unlimited capacities are assumed.
tem can be identified through ill-conditioning of Hessian in PDIPM Table 7 shows the obtained plans of combined AC TEP with reac-
or divergence of PDIPM in case of constraints’ violation. The non- tive power planning for the various test systems. It can be observed
convergence of TEP, in this case, will only be due to unavailability of from Table 5 and Table 7 that the proposed approach is capable of

IET Research Journals, pp. 1–9


6 © The Institution of Engineering and Technology 2015

ACTEPs ensitivitya nalysis.pdf M ainDocument IET Review Copy Only 7


Auto-generated PDF by ReView IET Generation, Transmission & Distribution

giving plans for combined AC TEP with reactive power planning, and total demand of (10 + j1.52)pu. The new real power demands
with convergence characteristic being similar to as shown in Fig. 2. of bus 4 and bus 5 are 3.0pu and 3.4pu, respectively. The new reac-
Also, the associated increase in cost of the plan as compared to DC tive power generation maximum limits at buses 1, 3 and 6 are 0.4pu,
TEP plan is much lesser than the same in AC TEP plan as com- 1.0pu and 1.56pu, respectively. It is assumed that the maximum
pared to DC TEP plan, for all the test systems. This is due to the lines in each corridor are 5. AC TEP for this system, when solved
presence of locally installed reactive resources in combined AC TEP by the proposed approach, gives the final TEP plan as: n1−4 = 1,
with reactive power planning. n1−5 = 1, n2−3 = 1, n2−6 = 5, n3−5 = 5 and n4−6 = 5. The
feasibility of the obtained plan has been cross checked by verifying
4.4 Comparison with existing approaches that no constraint violations exist after doing power flow analysis
and ACOPF. AC TEP for the same system has also been solved
In here, the efficiency of the proposed approach and the resulting by CHA. New lines get added based on the sensitivity index, until
TEP plans have been compared with the same from the conventional n = 0. The initial few sequences of line additions in CHA are:
branch and bound/ cut method for solving MILP/ MINLP problems, n4−6 , n4−6 , n4−6 , n2−6 , n3−5 , n4−6 , n2−6 , n3−5 , n4−6 , n2−6 ,
CHA [12], MF based approach [24] and PSO [7]. n3−5 , n4−6 , with n 6= 0, indicating that there is still scope for line
Integer line candidacy based formulations of AC TEP (1)-(9) and additions. However, it is to be noted that number of lines in n4−6
DC TEP (14)-(18) are MINLP problems and hence, solvers available has reached its maximum line corridor limit but still the approach
in the optimization toolbox of MATLAB, CPLEX API for MATLAB demands for a new line in the same corridor. If CHA is stopped at
and Gurobi API for MATLAB fail to solve these. Except for the DC this point, the solution attained is infeasible. But, if CHA is further
model with binary line candidacy, all other models (DC model with continued, the final solution is: n1−2 = 1, n1−5 = 1, n2−3 = 2,
integer line candidacy, AC model with binary and integer line candi- n2−6 = 5, n3−5 = 5, n4−6 = 6, which is clearly infeasible. The
dacy) have nonlinear constraints. Solvers from MATLAB, CPLEX reason for this non-optimality or infeasibility is that, the line addi-
API for MATLAB and Gurobi API for MATLAB do not include tions are purely based on power flows and the sensitivity index in
formats for solving formulations which include nonlinear equality CHA does not account for feasibility and optimality. On the other
constraints. An attempt has also been made to solve the MINLP hand, the sensitivities evaluated in the proposed approach take care
F
binary variables based AC TEP, using CPLEX. The equality con- of feasibility and optimality via the second sensitivity term Sd2 (i)
straints are each converted into two inequality constraints in AC TEP in (19).
for execution of CPLEX. However, it still fails to give a solution due A comparison of the proposed approach with the MF based
to internal error in the solver. In order to have a comparative anal- approach [24] for AC TEP has also been done. In [24], MF based
ysis in terms of computational efficiency of the proposed approach approach has been used to deal with integer/ discrete variables in
over branch and bound/ cut approach, the effect of nonlinearity and AC OPF based on the sensitivities obtained from the power flow
non-convexity in DC TEP has been nullified by considering binary solution. The same approach can also be extended for obtaining AC
line candidacy based formulation of DC TEP [32]. Thus, the binary TEP plans. The associated TEP plans are given in Table 9. It can be
line candidacy formulation of DC TEP [32], which is a MILP prob- observed that the MF based approach, gives a discrete line candidacy
lem, has been implemented and solved by intlinprog, a MILP solver solution after the first iteration of the approach. However, this is not
available in MATLAB and by cplexmilp, a MILP solver available a feasible plan and hence, further iterations of the approach continue.
in CPLEX API for MATLAB. These solvers inherently makes use For the 3 bus and 6-gf systems, the final solution, convergence char-
of branch and bound/ cut technique to solve MILP problems [33]. acteristics and CPU times of MF based approach are same as that
Table 8 shows the range of CPU times of intlinprog and cplexmilp, of the proposed approach as shown in Table 5 and Fig. 2. However,
for obtaining the same optimal DC TEP plans, for various test cases for the Garver 6 bus, IEEE RTS 24 bus and IEEE 118 bus systems,
(as shown in Table 5). Computational efficiency of the proposed sensitivities in the MF based approach do not yield feasible AC TEP
approach is confirmed as it takes comparatively less CPU time. It plans. Such infeasibility issues are possible in MINLP problems [24]
can also be expected from the given CPU times that this efficiency when solved by power flow sensitivity based approach. However, no
shall increase with the increase in system size and shall also be true such issues exist in our proposed approach as the sensitivities used
for AC TEP. for discretization ensure feasibility and optimality of the associated
KKT conditions.
Table 8 Line candidacy and CPU times for DC TEP by Branch and Bound/ Cut Table 9 AC TEP plans by MF based approach [24]
approach
Discrete line candidacy Line
MATLAB CPLEX System (at the end of first Final line investment
System Final line candidacy solver’s time solver’s time iteration candidacy cost($/year)
Tint (s) Tcpl (s)
3 n1−3 =1, n2−3 =1 n1−3 =1, n2−3 =1 10 × 103
3 n1−3 =1, n2−3 =1 0.06-0.12 0.07-0.08
6 n3−5 =1, n4−6 =3 0.25-0.40 0.26-0.28 n2−6 =2, n3−5 =1 AC TEP encounters
6-gf n2−6 =2,n3−5 =2, n4−6 =2 0.67-0.80 0.26-0.28 6 -
n4−6 =1 infeasiblility
24 n6−10 =1, n7−8 =2 1.01-1.82 3.20-3.40
n3−5 =1, n5−6 =1, n8−5 =1, n1−5 =1, n2−3 =1 n1−5 =1, n2−3 =1
118 21.12-30 49.0-50.0
n5−11 =1, n77−78 =1 6-gf n2−6 =3, n3−5 =2 n2−6 =3, n3−5 =2 2.5 × 105
n4−6 =2 n4−6 =2
A comparative study of the proposed approach with CHA [12], to
solve AC TEP has also been done. AC TEP for the various test sys- n3−24 =1, AC TEP encounters
24 -
tems are solved using CHA. The associated AC TEP plans are the n7−8 =1, n15−24 =1 infeasiblility
same as obtained by the proposed approach in Table 5. However, the
CPU times required by CHA for AC TEP of 3 bus, 6 bus, 6-gf, 24 AC TEP encounters
118 n8−5 =1 -
bus and 118 bus systems are in the range of 0.6 − 0.7s, 3.3 − 3.5s, infeasiblility
5.3 − 5.5s, 29 − 31s and 6100 − 6300s, respectively. This is very
high as compared to that by the proposed approach. In order to fur-
ther showcase the superiority of the proposed approach over CHA, a Meta-heuristic approaches may provide sufficiently good and
stressed system condition of the 6-gf system [12] has been chosen as acceptable solution for an optimization problem. Hence, a compara-
the test system. Maximum and minimum voltage limits are 1.05pu tive analysis of the proposed PDIPM sensitivity based approach with
and 0.95pu, respectively for the stressed condition. The stressed 6- a popular meta-heuristic approach - PSO [7] to solve AC TEP has
gf system considers 15 right-of-ways with the same investment cost been done. Table 10 shows the comparison between the AC TEP

IET Research Journals, pp. 1–9


© The Institution of Engineering and Technology 2015 7

ACTEPs ensitivitya nalysis.pdf M ainDocument IET Review Copy Only 8


Auto-generated PDF by ReView IET Generation, Transmission & Distribution

plans obtained by the proposed approach and PSO. Solutions by Table 11 shows the obtained TEP plans for both the DC TEP and
PSO for AC TEP, shown in Table 10, are the ones which are the AC TEP formulations using the proposed approach for the Indian
best obtained from 50 individual runs of PSO in each test case (as NRPG system. DC TEP plan is 81.3% less costlier than the same of
a single run of PSO does not yield this solution). It can be seen AC TEP for this system. CPU times of DC TEP and AC TEP are in
that the AC TEP plan by PSO is costlier than the solution from the the range of 1900 − 2200s and 3200 − 3300s, respectively. How-
proposed approach for all the test cases. The associated percentage ever, the DC TEP plan when implemented to the actual AC system,
increase in cost by PSO with respect to the proposed approach is also leads to infeasible and non-optimal system operation. This again,
shown in Table 10. AC TEP, solved using PSO takes CPU time of proves the superiority and need of solving AC TEP over DC TEP
20 − 25s and 50 − 60s for the Garver 6 bus system with and with- while also proving the capability of the proposed approach to obtain
out initial topology (6-gf), respectively. For IEEE RTS 24 bus and TEP plans for practical systems. The AC TEP for Indian NRPG sys-
IEEE 118 bus systems, the CPU times of PSO to solve AC TEP are tem is also solved using PSO. The NRPG system when solved with
140 − 150s and 450 − 550s, respectively. Similarly, as the system PSO results in no solution, i.e. AC TEP for NRPG system did not
size increases, the time taken by PSO for AC TEP is significantly return an output value, even after allowing it to run for more than 24
high with respect to the proposed approach. One of the drawbacks hours. This may be due to the fact that the randomly generated pop-
with the meta-heuristic techniques is that size of the swarm and the ulation does not have the appropriate particle with proper variables
number of iterations are determined for each system by trial and combination.
error process. This further confirms the superiority of the proposed
approach over PSO for AC TEP.
5 Conclusion
Table 10 Comparison of AC TEP solutions by proposed approach and PSO
This paper presents an efficient and novel sensitivity based approach
PSO % increase to solve the short-term AC TEP. The approach has used the sensi-
System Final line Total cost in tivities of the objective function with respect to the line candida-
candidacy ($/year) Total cost cies considered as integer variables. These sensitivities have been
obtained from the converged factorized Hessian of PDIPM through
3 n1−3 =1, n2−3 =1 14.917 × 103 0 which the AC TEP has been solved. The crucial element in the pro-
posed technique is that the sensitivities are always positive due to the
6 n2−6 =2, n3−5 =3, n4−6 =2 1.7995 × 105 11.37 dominant investment costs over the cost involved in power genera-
tion. The zero or near zero sensitivities also play a vital role in the
n1−5 =1, n2−3 =1, decision making process for finding TEP plans. Results on a 3 bus
6-gf
n2−6 =4, n3−5 =2, n4−6 =2 2.9541 × 105 17.44
system have been discussed in detail and clearly indicate the impor-
tance of the sensitivity based approach. Also, the proposed approach
n1−8 =3, n6−10 =1, has been tested on the Garver 6 bus system with and without an ini-
24
n7−8 =1, n20−23 =3 2.1092 × 108 24.07
tial network, IEEE RTS 24 bus system, IEEE 118 bus system and
n1−3 =1, n3−5 =1, the practical Indian NRPG system. The results obtained for DC TEP
n4−5 =1, n8−5 =3, and AC TEP formulations using the proposed approach, recommend
the usage of AC TEP formulation as it represents a practical scenario
118 n30−38 =1,n45−46 =1, 1.5311 × 108 21.19 and implementation of DC TEP plans on AC networks lead to infea-
n69−70 =1, n87−86 =1, sible and non-optimal system operation. The proposed approach is
n99−100 =1, n114−115 =1 also capable of obtaining solution for combined AC TEP with reac-
tive power planning. It proves to be computationally efficient with
respect to the conventional branch and bound/ cut approach for
4.5 Validation of proposed approach on Indian NRPG solving MILP/ MINLP problems. This computational efficiency is
System expected to increase with the increase in system size. The proposed
approach also provides the same solution for DC TEP as that pro-
A practical Indian NRPG system [28], [34] has been considered vided by the branch and bound/ cut approach. For stressed system
for comparing the AC TEP and DC TEP plans by the proposed conditions, the proposed approach provides superior and feasible
approach. This system covers nine states with 30% geographical AC TEP plans, with much lesser CPU times as compared to CHA.
area and 28% population of India. A reduced representation of the The proposed approach also provides feasible AC TEP plans for
NRPG system has been considered, which consists of 246 buses, 376 Garver 6 bus, IEEE-RTS 24 bus and IEEE 118 bus systems, while
branches, 42 generating units and 40 shunt reactors. This system is the MF based approach suffers from infeasibility. It also offers bet-
assumed to be expanded for a future condition where load and gener- ter AC TEP plan with much lesser investment cost and CPU times in
ation are increased by 50%. Maximum and minimum voltage limits comparison to the PSO. These extensive numeric comparisons thus
are 1.05 pu and 0.95 pu, respectively. The options for expansion indicate that the proposed approach is capable of finding feasible
plans have been so considered that extra lines can be constructed on TEP plans which are not inferior to those already reported in the lit-
all 376 line corridors. Two different types of lines are considered and erature. This is also justified by the fact that the proposed technique
it is assumed that, up to 5 new lines can be installed per line corridor. is based on the predictor-corrector philosophy.
The investment cost for line installation has been estimated in rupees
by using (20).
6 References
Table 11 DC TEP and AC TEP solutions by proposed approach for Indian NRPG 1 Hemmati, R., Hooshmand, R.A., Khodabakhshian, A.: ‘State-of-the-art of trans-
Indian system mission expansion planning: Comprehensive review’, Ren Sustainable Energy
Reviews, 2013, 23, pp. 312–319
NRPG DC TEP AC TEP 2 Garver, L.L.: ‘Transmission network estimation using linear programming’, IEEE
Trans Power App Syst, 1970, PAS-89, (7), pp. 1688–1697
n116−229 =1 n116−229 =1, n17−229 =1, n133−18 =1, 3 Bent, R., Coffrin, C., Gumucio, R.R., Van.Hentenryck, P. ‘Transmission network
Final Line n154−155 =1 n18−134 =1, n154−240 =1, n181−158 =1, expansion planning: Bridging the gap between ac heuristics and dc approx-
candidacy n154−240 =2 n156−158 =1, n233−235 =1, n235−240 =2, imations’. In: 2014 Power Systems Computation Conference. IEEE, 2014.
n235−240 =3 n213−194 =1, n233−229 =2 pp. 1–8
4 Al.Hamouz, Z.M., Al.Faraj, A.S.: ‘Transmission-expansion planning based on a
non-linear programming algorithm’, Appl Energy, 2003, 76, (1), pp. 169–177
Total Cost 5 Jabr, R.A.: ‘Optimization of AC transmission system planning’, IEEE Trans Power
6.156 × 106 3.292 × 107
(INR/year) Syst, 2013, 28, (3), pp. 2779–2787

IET Research Journals, pp. 1–9


8 © The Institution of Engineering and Technology 2015

ACTEPs ensitivitya nalysis.pdf M ainDocument IET Review Copy Only 9


Auto-generated PDF by ReView IET Generation, Transmission & Distribution

6 Verma, A., Panigrahi, B., Bijwe, P.: ‘Harmony search algorithm for transmission 13, (4), pp. 1211–1218
network expansion planning’, IET Gener Transm Distrib, 2010, 4, (6), pp. 663–673 26 Mohapatra, A., Bijwe, P., Panigrahi, B. ’An OPF sensitivity based approach
7 Torres, S.P., Castro, C.A., Pringles, R.M., Guaman, W. ’Comparison of parti- for handling discrete variables’. In: IEEE PES General Meeting Conference &
cle swarm based meta-heuristics for the electric transmission network expansion Exposition. IEEE, 2014. pp. 1–5
planning problem’. In: IEEE Power Energy Soc. Gen. Meeting., 2011. pp. 1–7 27 Diethelm, K., Ford, N.J., Freed, A.D.: ’A predictor-corrector approach for the
8 Da.Silva, E.L., Gil, H.A., Areiza, J.M.: ‘Transmission network expansion planning numerical solution of fractional differential equations’, Nonlinear Dynamics, 2002,
under an improved genetic algorithm’, IEEE Transactions on Power Systems, 2000, 29, (1-4), pp. 3–22
15, (3), pp. 1168–1174 28 [Online]. Available: https://drive.google.com/open?id=
9 Al.Saba, T., El.Amin, I.: ’The application of artificial intelligent tools to the trans- 1VyeHNAbLJfu2wV4ZWudyYr-tklFCiCw8
mission expansion problem’, Electr Power Syst Res, 2002, 62, (2), pp. 117–126 29 ’IBM ILOG CPLEX V12.1 User’s Manual for CPLEX’, CPLEX Division, Incline
10 Romero, R., Monticelli, A., Garcia, A., Haffner, S.: ‘Test systems and mathemat- Village, NV, USA, 2009
ical models for transmission network expansion planning’, IEE Proc Gen Transm 30 [Online]. Available: http://motor.ece.iit.edu/data/Data\_118\_Bus.pdf
Distrib, 2002, 149, (1), pp. 27–36 31 Pletka, R., Khangura, J., Rawlins, A., Waldren, E., Wilson, D.: ‘Capital costs for
11 Romero, R., Rocha, C., Mantovani, J., Sanchez, I.: ’Constructive heuristic transmission and substations: updated recommendations for WECC transmission
algorithm for the dc model in network transmission expansion planning’, IEEE expansion planning’, Black and Veatch PROJECT, 2014, (181374)
Proc Gener Transm Distrib, 2005, 152, (2), pp. 277–282 32 Jabr, R.: ’Robust transmission network expansion planning with uncertain renew-
12 Rider, M., Garcia, A., Romero, R.: ‘Power system transmission network expansion able generation and loads’, IEEE Trans Power Syst, 2013, 28, (4), pp. 4558–4567
planning using AC model’, IET Gen Transm Distrib, 2007, 1, (5), pp. 731–742 33 [Online]. Available: http://in.mathworks.com/help/optim/ug/
13 Mahdavi, M., Sabillon Antunez, C., Ajalli, M., Romero, R.: ’Transmission expan- mixed-integer-linear-programming-algorithms.html
sion planning: Literature review and classification’, IEEE Syst J, 2018, pp. 1–12 34 ’NRPG 246 - Bus system’: https://www.iitk.ac.in/eeold/facilities/Research\_labs/
14 Alhamrouni, I., Khairuddin, A., Ferdavani, A.K., Salem, M.: ’Transmission expan- Power\_System/NRPG-DATA.pdf
sion planning using AC-based differential evolution algorithm’, IET Gener Transm
Distrib, 2014, 8, (10), pp. 1637–1644
15 Torres, S.P., Castro, C.A.: ’Specialized differential evolution technique to solve the 7 Appendices
alternating current model based transmission expansion planning problem’, Int J
Electr Power Energy Syst, 2015, 68, pp. 243–251
16 Akbari, T., Bina, M.T.: ’Approximated milp model for AC transmission expansion 7.1 PDIPM [25]
planning: global solutions versus local solutions’, IET Gener Transm Distrib, 2016,
10, (7), pp. 1563–1569 PDIPM is an iterative process, in which the Karush-Kuhn-Tucker
17 Taylor, J.A., Hover, F.S.: ’Linear relaxations for transmission system planning’, (KKT) conditions of (1)-(9) and (14)-(18) are solved using Newton’s
IEEE Trans Power Syst, 2011, 26, (4), pp. 2533–2538
18 Zhang, H., Heydt, G.T., Vittal, V., Quintero, J.: ’An improved network model for
method. The following update equation is solved in every iteration
transmission expansion planning considering reactive power and network losses’, of PDIPM
IEEE Trans Power Syst, 2013, 28, (3), pp. 3471–3479
19 Akbari, T., Bina, M.T.: ’Linear approximated formulation of AC optimal power       
∆y r̃y K11 K12 r̃y
flow using binary discretisation’, IET Gener Transm Distrib, 2016, 10, (5), = H−1 =
pp. 1117–1123 ∆λ rλ K21 K22 rλ
20 Arabpour, A., Besmi, M.R., Maghouli, P.: ’Transmission expansion planning with
linearized AC load flow by special ordered set method’, J Energy Eng, 2018, 144, where, y = [x d]T , x is vector of continuous variables and d is
(2), pp. 04018008 vector of integer variables in AC TEP and DC TEP. λ is the vector
21 Camponogara, E., de Almeida, K.C., Junior, R.H.: ‘Piecewise-linear approxi- of Lagrange multipliers for equality constraints. H is the reduced
mations for a non-linear transmission expansion planning problem’, IET Gener
Transm Distrib, 2015, 9, (12), pp. 1235–1244
Hessian and r˜y and rλ are mismatch vectors [26]. K11 , K12 ,
22 Haghighat, H., Zeng, B.: ’Bilevel conic transmission expansion planning’, IEEE K21 and K22 are sub-matrices, derived by factorizing H. K11 and
Trans Power Syst, 2018, 33, (4), pp. 4640–4642 K12 can be inferred as sensitivities of y with respect to r˜y and
23 Caccetta, L., Hill, S.: ’Branch and cut methods for network optimization’, Math rλ , respectively. Similar inference can be made for K21 and K22 .
Comput Modell, 2001, 33, (4), pp. 517–532
24 Capitanescu, F., Wehenkel, L.: ’Sensitivity-based approaches for handling discrete
Alternatively,
variables in optimal power flow computations’, IEEE Trans Power Syst, 2010, 25,
(4), pp. 1780–1789 ∂y ∂y ∂λ ∂λ
25 Torres, G.L., Quintana, V.H.: ’An interior-point method for nonlinear optimal K11 ≈ , K12 ≈ , K21 ≈ , K22 ≈ (21)
power flow using voltage rectangular coordinates’, IEEE Trans Power Syst, 1998, ∂ r˜y ∂rλ ∂ r˜y ∂rλ

IET Research Journals, pp. 1–9


© The Institution of Engineering and Technology 2015 9

ACTEPs ensitivitya nalysis.pdf M ainDocument IET Review Copy Only 10

Вам также может понравиться