Mod 9

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA
Decision support models in electric power systems
Generation reliability
Andrés Ramos
http://www.iit.comillas.edu/aramos/
Andres.Ramos@comillas.edu
Topic objectives
• To understand
– Why reliability must be studied
– Where does reliability appear
– How to measure or asses reliability
– What is a probabilistic production cost model by analytical
simulation
– How unit output is computed
– How to compute some generation reliability measures
• To see a true model

Decision support models in electric power systems – Generation reliability - 1
Electric Reliability—Planning in an Uncertain Future

References
• Billinton, R. and Allan, R.N. Reliability Evaluation of Power Systems, Springer,
1996.
• Billinton, R. and Li, W. Reliability Assessment of Electrical Power Systems Using
Monte Carlo Methods, Springer, 1994
• Billinton, R. and Allan, R.N. Reliability Evaluation of Engineering Systems Concepts
and Techniques, Springer, 1992
• Billinton, R. and Allan, R.N. Reliability Assessment of Large Electric Power Systems,
Springer, 1988.
– Good references about electric system reliability
– Scope and insight that exceeds this presentation
• [IAEA, 84] “Chapter 7. Generating System Reliability”, Expansion planning for
electrical generating systems, A guidebook. International Atomic Energy Agency
Technical report No 241. Vienna, Austria, 1984.
(http://www-pub.iaea.org/MTCD/publications/PDF/TRS1/TRS241_Web.pdf)
– Original source of part of this presentation
• Wood, J., and Wollenberg, B.F., Sheble, G.B. Power Generation Operation and
Control, 3rd edition, Wiley, 2013.
• Vardi, J. and Avi-Itzhak, B. Electric energy generation. Economics, reliability and
rates. The MIT Press, 1981.
Maintenance scheduling
• A. Froger, M. Gendreau, J.E. Mendoza, É. Pinsona, L-M. Rousseau

Maintenance scheduling in the electricity industry: A literature review
European Journal of Operational Research 251 695–706 2016
• C. Feng and X. Wang A Competitive Mechanism of Unit Maintenance
Scheduling in a Deregulated Environment IEEE Transactions on Power
Systems 25 (1): 351-359 Feb 2010
• L. Muñoz, A. Ramos Goal programming approach to maintenance
scheduling of generating units in large scale power systems IEEE
Transactions on Power Systems 14 (3): 1021-1027 Aug 1999

Contents
Introduction
• Reliability assessment
• Reliability indexes classification
– Deterministic
– Stochastic
• Probabilistic production cost model
• Statistical concepts
• Unit dispatch
• Unit convolution
• Reliability measures
• Discrete convolution
• Limited energy plants dispatch
• FLOP model
• Summary
What is reliability?
• The traditional objective of electric system planning is to
supply electricity demand at minimum cost and with an
acceptable long term reliability
– Supply continuity or existence of voltage supply
– Wave quality (value, shape, frequency, …)
• Reliability of electricity supply:
– Adequacy (long term)
• The ability of the electric system to supply the aggregate electrical demand
and energy requirements of the customers at all times, taking into account
scheduled and reasonably expected unscheduled outages of system elements
[ENTSO-E]
– Security (short term)
• The ability of the electric system to withstand sudden disturbances such as
electric short circuits or unanticipated loss of system elements [ENTSO-E]
Why nobody speaks about reliability?
• Currently, in electric energy systems nobody speaks about

reliability
• We talk about
– Capacity payments (price set centrally)
– Capacity markets (volume set centrally)
– Reliability options
– Long-term reserves contracting
• However, all these topics are very closely related to
reliability (adequacy)

How to study reliability?
• Historical way
– Statistics about failures and load shedding from the past
– Studies easy to be done
• Predictive way
– Assess the expected values of failures by a reliability model
– Complex studies
– Data are extracted from historical data

Contents
• Introduction
Reliability assessment
– Deterministic
– Stochastic
• Unit dispatch
• FLOP model
• Summary
Reliability assessment
• Reliability index
– Parameter that measures a particular aspect of reliability of an
electric energy system
• We are going to talk about generation
• The more relevant aspects to be assessed by reliability
indexes are:
– Number or frequency of failures
– Duration of failures
– Incidence of failures
• There is no index able to assess the total reliability
including all the aspects

Reliability criterion in generation expansion planning
• Reliability is a key aspect of long term generation expansion planning
studies
– The more installed capacity the more reliability
• Reliability criterion
– Reliability standard or maximum value allowed for a reliability index for any
expansion plan of an electric energy system
• Constraint in the optimization problem
• Cost associated to a reliability index in the design process of an electric
energy system
– Included in objective function as the cost of the not served electricity
• In multiobjective optimization reliability indexes are potential
objective functions

Reliability criterion in generation expansion planning
Minimize:
∑ Operation costs + Investment costs
Subject to:
• Load supply
• Reliability index < maximum reliability index
Minimize:
∑ Operation costs + Investment costs
+ Costs associated to Not served energy
Subject to:
• Load supply
Minimize:
Objective function #1: ∑ Operation costs + Investment costs
Objective function #2: Reliability index
Subject to:
• Load supply
Contents
• Introduction
Reliability indexes classification
Deterministic
Stochastic
• Unit dispatch
• FLOP model
• Summary
Reliability indexes classification (i)
• Deterministic
– Reflect the average behavior of the supply continuity
– Do not consider stochasticity in the operation
– Very frequently used:
• Intuitive
• Simple to determine
• Require few data to compute them
– Easy comparison among systems
• Probabilistic
– Do consider stochasticity in the operation
• Unit failures
• Load variation
• Hydro inflows
– Offer more information and of better quality than the deterministic
indexes
Contents
• Introduction
Deterministic
– Stochastic
• Unit dispatch
• FLOP model
• Summary
Deterministic reliability indexes (i)
• Reserve Margin (RM):

– Excess of generation capacity available to satisfy yearly load
demand
RM ( MW ) = Available generation - Maximum demand
Available generation - Maximum demand

RM ( pu ) =
Maximum demand
– Main characteristic is simplicity:
• Intuitive, easy to compute
• Limited because does not consider water reserves, sizes,
technologies, or outage rates

Reserve margin of the Spanish electric system
Source: REE

SOAF (Scenario Outlook and system Adequacy
Forecast)
ARM: Adequacy Reference Margin
RC: Remaining Capacity
Source: ENTSO-e

Deterministic reliability indexes (ii)
• Largest Unit (LU):
– Potential unavailability of the largest unit
– Beat the RM because considers unit sizes
RM ( MW )
LU ( pu ) =
Largest unit power
LU>1 → we can loss the largest unit
LU<1 → if we loss the largest unit → not served energy
• Dry year:
– Used in hydro dominated systems
– Criterion more than index
• In the driest year (or drier year series) demand has to be satisfied (deterministically)

Deterministic reliability indexes (iii)
• Generation system and load

– Installed thermal power: 10000 MW
– Installed hydro power: 3000 MW
– Maximum forecasted load: 11500 MW
– Largest unit: 1000 MW
– Maximum hydro power on dry year: 1300 MW
• Deterministic indexes
RM(GW)=(10+3-11.5)=1.5 GW → excess of 1.5 GW
RM(pu)=(10+3-11.5)/11.5=0.13 → excess of 13 % of maximum load
LU(pu)=(10+3-11.5)/1=1.5 → excess of 1.5 largest unit

Deterministic reliability indexes (iv)
• Deterministic reliability indexes on a dry year

RM(pu)=(10+1.3-11.5)/11.5=-0.017 → lack of -1.7 % of maximum load
LU(pu)=(10+1.3-11.5)/1=-0.2 → lack of 20 % of the largest unit
→ In a hydrothermal system things change on a dry year
• Power needed for a reliability criterion RM > 2 GW

RM(MW)=(10+3+PNEW-11.5)=2 GW → PNEW= 0.5 GW
• Power needed for a reliability criterion RM > 2 GW

on a dry year
RM(MW)=(10+1.3+PNEW-11.5)=2 GW → PNEW= 2.2 GW

Contents
• Introduction
– Deterministic
Stochastic
• Unit dispatch
• FLOP model
• Summary
Probabilistic reliability indexes:
LOLP and LOLE (i)
• Loss Of Load Probability (LOLP):
– According to the name, should be defined as probability of being unable to satisfy
all the power with the available generation
– However, usually is measured in number of hours or days in a year with
insufficient generation. For example, 0.1 day/year
• To avoid this double and ambiguous definition, Billinton defines the
Loss Of Load Expectation (LOLE):
– Number of days (hours) in a year that we expect not to satisfy the demand with
the available generation
LOLE
LOLP =
365 days o 8760 hours

LOLP and LOLE (ii)
• The most frequently used reliability indexes
• No information regarding:
– Duration and frequency of failures
– Incidence of loss of load
• LOLP or LOLE can be calculated with:
– Not to serve the 365 daily load peaks in a year
– Not to serve the 8760 hourly loads in a year
• For the same system LOLP values obtained are greater for the first
case
– Example: loss of load in just the maximum demand hour
LOLE 1 1
LOLPN = LOLP365 = > LOLP8760 =
N 365 8760

LOLP and LOLE (iii)
– Unit 1:
• Output: 1000 MW
• Equivalent Forced Outage Rate (EFOR): 0.05 (5%)
– Unit 2:
• Output : 900 MW
• Equivalent Forced Outage Rate (EFOR): 0.04 (4%)
– Forecast maximum load:
• Case A: 1100 MW
• Case B: 800 MW
• LOLP case A: there will be loss of load for any unit failure
LOLPA = 0.05+0.04-0.04·0.05=0.088 (8.8%)

• LOLP case B: there will be loss of load only when both units fail
LOLPB = 0.05 ⋅ 0.04=0.002 (0.2%)

• LOLP ≠ 0 although RM is positive because there is always a probability, and it can
be very small, of failing all the units

Probabilistic reliability indexes :
LOEE and LOEP (i)
• Loss Of Energy Expectation (LOEE) widely known as Expected Energy

Not Served (EENS) (Expected Unserved Energy EUE):
– It is defined as energy expected not to supply in a year by generation unavailability
or by lack of primary energy
• Loss Of Energy Probability (LOEP):
– It is defined as probability of not supplying a kWh with the available generation
– As it is expressed in per unit, it allows to compare systems of different size
EENS
LOEP =
Total load

LOEE and LOEP (ii)
• Frequently used indexes, specially in systems with limited

primary energy as hydrothermal ones
• Better than LOLP and LOLE because they consider
incidence of loss of load as not served energy
– LOLP measures the probability of not serving demand
– However, LOLP does not say anything about how much energy
can be supplied

POPM and XLOL
• eXpected Loss Of Load (XLOL) in MW:

– Power we expect to supply once the failure has occurred
– It is also called eXpected Load Not Supplied (XLNS)
EENS (MWh) EENS (MWh)

XLOL= =
LOLE (hours) LOLP ⋅ 8760 hours
• Probability Of Positive Margin (POPM):

– It is defined as the probability of satisfying the demand in the maximum
yearly demand hour with the available generation
– It is a probability of success
LOLPN ≤ 1 − POPM = LOLP1

LOEE, LOEP and XLOL (example 1)
• Generation system and demand
– Peak: 2000 hours at 9000 MW (18 TWh)
– Shoulder: 4000 hours at 6000 MW (24 TWh)
– Off-peak: 2760 hours at 4000 MW (11.04 TWh)
– Thermal capacity: 5000 MW (EFOR=0)
– Hydro capacity: 5000 MW (EFOR=0)
– Hydro energy: 10000 GWh
Not served energy =
= (9 – 8) GW · 2 kh = 2 TWh
Hydro energy available for peaking = Shoulder hydro output =

= 10 – 4 = 6 TWh 9000 MW = (6 – 5) GW · 4 kh = 4 TWh
8000 MW
6000 MW
Maximum hydro output = 4000 MW
= 6 TWh/2 kh = 3000 MW
5000 MW
2000 h 4000 h 2760 h

LOEE, LOEP and XLOL (example 2)
• Reliability indexes
RM(MW)=(5+5-9)=1 GW → excess of 1 GW
LOLE(hours)= 2000 hours → lack of generation during 2000 hours
LOLP(pu)=2000/8760=0.22 → rationing 22 % of the time
EENS(GWh)= 2000 GWh → lack of primary energy
LOEP(pu)=2 TWh/(2·9+4·6+2.76·4)TWh=0.0377 → LOEP = 3.77 %
XLOL(GW)= 2 TWh/2 kh = 1 GW → lack of 1 GW when rationing
Not served power due to lack of Not served energy =

primary energy = 9000 MW = (9 – 8) GW · 2 kh = 2 TWh
= 9 – 8 = 1 GW
8000 MW
6000 MW
Maximum hydro output = 4000 MW
= 6 TWh/2 kh = 3000 MW
5000 MW
2000 h 4000 h 2760 h

Other probabilistic reliability indexes: EOPE and F&D
• Emergency Operating Procedure Expectation (EOPE):

– Number of days per year that will activate emergency procedures (see page 258) to avoid
loss of load
• Load curtailment
• Voltage reduction
• Frequency and Duration of failures (F&D):
– Number of expected failures in a year and their duration
LOLP = Frequency ⋅ Duration

• EOPE and F&D are more realistic and contain more information than LOLP or
LOLE although they are barely used in planning:
– By the computational complexity (hourly models)
– And their lower influence in planning decisions with respect to other indexes

Resilience
• Ability of the electric system to cope with extremely adverse
conditions associated to climate and a combination of system
contingencies whose probability of occurrence is above a
certain threshold level.
• For example, increases in temperatures up to extreme values
(hot in summer and cold in winter), decreasing water
availability, storm events, low wind speeds and cloudy
weather over large areas, etc.

Source: http://astro.ukho.gov.uk/eclipse/0112015/ 31 January 2011
Contents
• Introduction
– Deterministic
– Stochastic
Probabilistic production cost model
• Unit dispatch
• FLOP model
• Summary
Generation capacity adequacy assessment techniques
1. State Table
– Exact
– Impossible to use for large-scale systems
2. Monte Carlo Simulation

– Allows to compute not only mean values but their distributions
– Sequential or non-sequential
3. Probabilistic Production Cost Model

– Very quick analytical technique

Reliability measures by a state table

– Unit 1:
• Output: 1000 MW
• EFOR: 0.05 (5%)
– Unit 2:
• Output : 900 MW
• EFOR: 0.04 (4%)
– Forecast maximum load:
• Case A: 1100 MW State Available Probability PNS
capacity
• Reliability measures 11 1900 0.912 0

– LOLP = 0.088 10 1000 0.038 100
– EENS = 15.6 MW
01 900 0.048 200
00 0 0.002 1100

Reliability measures by Monte Carlo simulation
• For each sample

– Obtain generator availability
– Dispatch available units to supply demand
– Determine if this sample has PNS and its quantity
• Compute LOLP and EENS

# of samples with PNS
LOLP =
# of samples
Total PNS
EENS =
# of samples

Estimation
n
X (n ) = µˆ =
∑ i =1
xi
• Sample mean n
n
S 2 (n ) = σˆ2 =
∑ i =1
(x i
− X (n ))2
• Sample variance n −1
n
var X (n ) =

S (n )
=
2
∑ i =1
(x i
− X (n ))2
• Variance of the mean   n n(n − 1)

 S 2 (n ) S 2 (n ) 
X (n ) − t
• Confidence interval of the mean  n −1,α 2 , X (n ) + tn −1,α 2 
 n n 
• Coefficient of variation S 2 (n )
cv = n
X (n )
t = 2.581 t = 1.962
1000−1,1% 1000−1,5%
2 2

Small example
• An electric system has 2 generators with a failure

probability of 0.1
• The system fails if at least one of the units is failed
• Determine the system reliability by Monte Carlo simulation
• Determine number of samples needed to achieve a half-
width of 2 % with a confidence level of 95 %

Exact computation of the LOLP
• Case p=0.9
– Mean: 0.81
– Variance of the r.v.: 0.1539
– Coefficient of variation of the r.v.: 0.484322
• Case p=0.95
– Mean: 0.9025
– Variance of the r.v.: 0.08799375
– Coefficient of variation of the r.v.: 0.328684

LOLP by Monte Carlo simulation
• 2000 samples
p=0.9 p=0.95
media 0.802000 media 0.905500
varianza 0.158875 varianza 0.085613
var media 0.000079 var media 0.000043
coef var 0.496997 coef var 0.323132
• 20000 samples
p=0.9 p=0.95
media 0.809200 media 0.904950
varianza 0.154403 varianza 0.086020
var media 0.000008 var media 0.000004
coef var 0.485593 coef var 0.324097

Probabilistic production cost model
•Unit expected output

•System reliability
Probabilistic
•Electric demand and measures
Production Cost •Loss of load
•Unit available capacity
Model by probability LOLP
are independent random •Expected energy not
Analytical
variables served EENS
Simulation •Loss of energy
probability LOEP

Contents
• Introduction
– Deterministic
– Stochastic
Statistical concepts
• Unit dispatch
• FLOP model
• Summary
Statistical basic concepts
• Let X a random variable

• Probability density function (pdf) f X ( x)
– Probability that the variable belongs to an infinitesimal interval
f X ( x) = P[ x < X < x + dx]/ dx
• Cumulative distribution function (cdf) FX ( x )

– Probability that the variable is lower or equal to a certain value
x
FX ( x ) = P[ X ≤ x ] = ∫ f X ( x )dx
−∞

Continuous and discrete random variables
f X ( x) FX ( x )
1
0
D Dx D D x
fY ( y ) FY ( y )
1
0
0 y 0 y
P P

Weekly demand (i)
Semana 22-28 NOV
35,000
30,000
25,000
Sábado
Demanda [MW]
20,000 Domingo
Lunes
Martes
Miércoles
15,000 Jueves
Viernes
10,000
5,000
-
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Hora

Weekly demand (ii)
Demanda 22-28 NOV
35000
30000
25000
20000
Demanda [MW]
15000
10000
3966989 MWh
5000
0
1
5
9
13
17
21
25
29
33
37
41
45
49
53
57
61
65
69
73
77
81
85
89
93
97
101
105
109
113
117
121
125
129
133
137
141
145
149
153
157
161
165
Horas

Load-duration curve
– Decreasing ordering of the 168 demand values

– Area = total weekly energy
Monótona decreciente semana 22-28 NOV
35000
30467 MW 30000
25000
20000
Demanda [MW]
15766 MW 15000
3966989 MWh
10000
5000
0
1
5
9
13
17
21
25
29
33
37
41
45
49
53
57
61
65
69
73
77
81
85
89
93
97
101
105
109
113
117
121
125
129
133
137
141
145
149
153
157
161
165
Hora
Demand probability density function
– Probability of the demand is equal to a certain value f X ( x)
– Obtained from load forecasts for this week
Función de densidad f(x)
Probability density
0.045
function approximated
in a “deterministic” way
0.040
0.035
0.030
Probabilidad
0.025
0.020
0.015
0.010
0.005
0.000
00
00
00
00
00
00
00
00
00
0
0
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
10
20
30
40
50
60
70
80
90
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
Demanda [MW]
Demand cumulative distribution function
– Probability of the demand is lower or equal to a certain value FX ( x )
– Obtained accumulating the probability density function f X ( x)
Función de distribución F(x)
1.000
0.900
0.800
0.700
0.600
Probabilidad
0.500
0.400
0.300
0.200
0.100
0.000
00
00
00
00
00
00
00
00
00
0
0
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
10
20
30
40
50
60
70
80
90
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
Demanda [MW]

Complementary cumulative distribution function (ccdf) (also
called Load-Duration Curve LDC)
– Probability of the demand is greater or equal to a certain value
– Total weekly energy = area x week duration
G X ( x ) = 1 − FX ( x )
Complementaria de la función de distribución G(x)=1-F(x)
1.000
Probability of the
0.900 demand is >= D =
15766 MW = 1
0.800
0.700
0.600
Probabilidad
0.500
23615 MW
0.400
0.300
0.200
Probability of the
0.100 demand is >= D =
30467 MW = 0
0.000
0
0
00
00
00
00
00
00
00
00
00
0
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
00
10
20
30
40
50
60
70
80
90
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
Demanda [MW]

Probability density function of the unit available
capacity
• Random variable with Bernoulli distribution
– A unit can only be into two possible states (available or
unavailable)
– q unit failure probability. Available capacity = 0
– p = 1− qunit operation probability. Available capacity = rated
capacity
Potencia disponible
1.00
0.90
p = 1− q 0.80
0.70
0.60
Probabilidad [p.u.]
0.50
0.40
0.30
0.20
q 0.10
0.00
0 350
Potencia [MW]

Contents
• Introduction
– Deterministic
– Stochastic
Unit dispatch
• FLOP model
• Summary
Unit dispatch
• Thermal units are dispatched in single loading order by

increasing variable cost:
– Nuclear, brown lignite, black lignite, imported coal, domestic coal, gas,
oil, etc.
• It does NOT consider minimum load neither start-ups of the
units
• Hydroelectric units are dispatched where they produce their
available energy (unregulated flows)
• Unit failures are supposed to be independent among them
and independent of the demand uncertainty
• It does NOT exist inter-period relations

Common cause failure
• P. Fairley Nuclear Shutdowns Put Belgians and Britons on

Blackout Alert IEEE Spectrum Sep 2014

Dispatch WHITOUT thermal units’ failure (i)
• Thermal units are dispatched from bottom up (in a
pile) under the load curve
• Energy output = curve area
35000
30000
25000
20000
Demanda [MW]
15000
10000
Energy of unit 3
5000
Energy of unit 2
Energy of unit 1
0
1
5
9
13
17
21
25
29
33
37
41
45
49
53
57
61
65
69
73
77
81
85
89
93
97
101
105
109
113
117
121
125
129
133
137
141
145
149
153
157
161
165
Horas

Dispatch WHITOUT thermal units’ failure (ii)
• Thermal units are dispatched from left to right under the
complementary cumulative distribution function curve
• Energy output = area x period duration
Power of unit 1
0.9
Power of unit 2 0.8
0.7
Power of unit 3
0.6
Probabilidad [p.u.]
0.5
0.4
0.3
0.2
0.1
0
0 2500 5000 7500 10000 12500 15000 17500 20000 22500 25000 27500 30000 32500 35000
Demanda [MW]

Dispatch WHIT thermal unit failure (i)
Unit does NOT fail, with probability p = 1− q
Probability
1
Expected energy output of the unit =

0
Demand
area x duration x probability
Unit fails, with probability q
Probability
1
0
Demand

Dispatch WHIT thermal unit failure (ii)
Unit does NOT fail, with probability p = 1− q
Probability
1
Energy not served if the unit
does NOT fail
EENS1
Expected energy output of the unit =
0
EENS before unit dispatch –
Demand
EENS after unit dispatch
Unit fails, with probability q
Probability
1
Energy not served if the
unit fails
EENS2
EENS after unit dispatch =

0 (EENS1 . p + EENS2 . q) x duration
Demand

Contents
• Introduction
– Deterministic
– Stochastic
• Unit dispatch
Unit convolution
• FLOP model
• Summary
Thermal unit convolution (i)
EENS if the unit i does NOT fail
Probability
1 Gi ( x − Pi )
EENS after unit i dispatch
EENS1 × pi
Probability
1 Equivalent Load-
0 Duration Curve
Demand ELDC
EENS if the unit i fails

Probability
1 Gi ( x )
0
Demand
EENS2 ×qi Gi +1 ( x ) = pi Gi ( x − Pi ) + qi Gi ( x )
0
Demand

Thermal unit convolution (ii)
• Complementary cumulative distribution function

G1 ( x ) = G X ( x )
• Unit convolution i=1,...,N
Gi +1 ( x ) = pi Gi ( x − Pi ) + qi Gi ( x )
• Expected energy output for each thermal unit i
Ei = T ∫ [Gi ( x ) − Gi +1 ( x )] dx
D
being T period duration and D the maximum demand

• Commitment hours of the unit
H i = T ⋅ Gi (0)

Dispatch of two units versus a single aggregated unit
• The dispatching result is NOT the same:

– two units with a certain failure probability that dispatching
– a single unit with capacity equal to the sum of both and equivalent
probability

Contents
• Introduction
– Deterministic
– Stochastic
• Unit dispatch
Reliability measures
• FLOP model
• Summary
Reliability measures (i)
• EENS Expected energy not served

D
EENS = E N +1 = T ∫ GN +1 ( x )dx
0
• LOLP Loss of load probability

LOLP = GN +1 (0)
• LOEP Loss of energy probability
EENS
LOEP = D
∫ G1 ( x )dx
0

Unit contribution to system reliability: ELCC and FCE
(i)
• Effective load carrying capability (ELCC)
– Increment in maximum demand that can be covered by the
system when adding a given unit keeping constant a certain
reliability index
– Depends on the unit characteristics
• Maximum output
• Forced outage rate (EFOR), scheduled maintenance
• Primary energy limits
– One year scope
– Algorithm (for decrement in maximum demand)
• All the units of the system are dispatched
• LOLP is determined
• Deconvolution of a unit
• Unit ELCC is determined

(ii)
LOLE= days
year
Original System
Original System +
1 unit with PMAX
PMAX
0.1 ELCC Reliability criterion de 0.1

days per year
Maximum yearly
load
D1 D2
(iii)
• Firm Capacity Equivalent (FCE)
– Alike ELCC, it is computed considering only the demand of a given hour
– Firm capacity is defined as the ideal unit always available
• No outages, no maintenance
– Given a generation system and a load, FCE is the power capacity with
failure probability equal to 0 such as if substituted by the unit gives
exactly the same system reliability measure
• FCE and ELCC

– Define the (incremental or marginal) contribution to system
reliability that each unit is reasonably able to provide
– Define the maximum capacity that each generating units is
allowed to offer or to be remunerated for.
– Must be used to determine capacity payments by the
contribution to system reliability.
(iv)
LOLP
Original System
Original System
- PMAX
1 unit with PMAX Same reliability
LOLP B C measure
FCE
Decrement in firm capacity
=
Load increment
Load
D Study -∆D D Study
Capacity credit of wind generation
• Contribution of wind power to system
adequacy.
• It is estimated by determining the
capacity of conventional plants displaced
by wind power, whilst maintaining the
same degree of system reliability, in
other words an unchanged probability of
failure to meet the reliability criteria for
the system.
• Alternatively, it is estimated by
determining the additional load that the
system can carry when wind power is
added, maintaining the same reliability
level.
• M. Amelin Comparison of Capacity Credit Calculation Methods for Conventional Power Plants and Wind Power IEEE
Transaction on Power Systems, 24(2): 685-691 May 2009 10.1109/TPWRS.2009.2016493
• NERC Methods to Model and Calculate Capacity Contributions of Variable Generation for Resource Adequacy Planning March
2011 http://www.nerc.com/files/ivgtf1-2.pdf
Contents
• Introduction
– Deterministic
– Stochastic
• Unit dispatch
Discrete convolution
• FLOP model
• Summary
Discrete convolution method
• An interval for specifying the complementary cumulative

distribution function (for example, 10 MW)
• Thermal unit capacities have to be multiple of this interval
• Shifts of the load-duration curve are an exact number of
points

Small case (i)
Units’ capacity have
to be multiple of this
number
EENS before unit

dispatch
LOLP before unit

dispatch
EENS = 53 MW
LOLP = 0.595
Complementary
cumulative
distribution function

Piecewise linear
approximation
Small case (ii) to the CCDF
1.00
0.90 Unit 1 Expected

Output
10 MW
0.80
Unit 2 Expected
0.70 Output
90.85 MW
0.60
Probability [p.u.]
DEMAND Unit 3 Expected

0.50
300 MW Output
34.64 MW
0.40
G1 ( x )
LOLP = 0.595
0.30
G6 ( x) G2 ( x )
0.20
G3 ( x )
EENS = 53 MW
0.10
LOEP
0.00 = 0.065
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41
Demand [MW]

Real case (i)
• Interval size: 250 MW

• Maximum demand 30467 MW, minimum 15766 MW
• Installed capacity of thermal units 32000 MW
Nº of units Capacity EFOR

[MW]
1 (~Run-Of-the-River) 2500 0
9 1000 0.05
20 500 0.05
12 250 0.05
1 (~Storage Hydro) 7500 0

Real case (ii)
1.00
0.90
RUN-OF-THE-RIVER
HYDRO
0.80
EXPECTED OUTPUT
2500 MW
0.70
0.60
Probabilidad [p.u.]
DEMAND
0.50
23615 MW
STORAGE HYDRO
0.40
EXPECTED
OUTPUT
1830 MW 0.30
0.20
LOLP = 0.012 0.10
0.00
EENS = 7 MW
00
50
00
50
00
50
00
50
00
50
00
50
0
0
75
50
25
00
75
50
25
00
75
50
25
00
75
50
25
00
75
50
25
00
75
50
25
00
75
50
25
00
15
22
30
37
45
52
60
67
75
82
90
97
10
11
12
12
13
14
15
15
16
17
18
18
19
20
21
21
22
23
24
24
25
26
27
27
28
29
30
Demanda [MW]
LOEP = 0.0003
Contents
• Introduction
– Deterministic
– Stochastic
• Unit dispatch
Limited energy plants dispatch
• FLOP model
• Summary
Dispatch of limited energy plants
• LEP (limited energy plants)
– Thermal units with maximum fuel available (take-or-pay contracts)
– Hydro units with limited unregulated flows
• If they are dispatched first, because their variable cost is 0, they will
produce more energy than available
• Unregulated flows are supposed to be deterministic (expected value)
• They are dispatched at maximum capacity to minimize total variable costs
(but, which is their maximum capacity?)
• Algorithm
– We try to dispatch the unit at every step until the energy produced by
the LEP unit is lower to its available energy
– Convolution invariance property: the sum of the energy produced by
the thermal unit and the LEP unit is the same regardless the order with
which they are dispatched
– Energy produced by the thermal unit is recalculated
Contents
• Introduction
– Deterministic
– Stochastic
• Unit dispatch
FLOP model
• Summary
FLOP model (i) (www.iit.comillas.edu/aramos/FLOP.htm)

FLOP model (ii)

FLOP model (iii)

FLOP model (iv)

Contents
• Introduction
– Deterministic
– Stochastic
• Unit dispatch
• FLOP model
Summary
Summary of probabilistic production cost model
• Advantages:
– Demand and generation are independent random variables
– Computation of units’ output
– Computation of reliability measures
– Computation speed
• Disadvantages:
– Single loading order (heuristically obtained)
– No minimum load, no startup or shutdown
– No extensions for electricity markets

Generation reliability homework. Assignment A
Prob
1
• Determine the expected unit output, LOLP
and EENS by using
1. Monte Carlo simulation
Demand
• For this electric system 200 400
– Load varies linearly (see figure) from 200 to
Capacity EFOR Variable cost
400 MW
[MW] [p.u.] [€/MWh]
– Period duration 1000 h
200 0.10 25
– Thermal generation of the table dispatched
by increasing variable cost 150 0.15 40
120 0.05 30
100 0.05 50
• This assignment is done individually
100 0.20 65
• Due date: Wednesday, December 9
50 0.15 75
• Discussion of homework: Thursday,
December 17
Generation reliability homework. Assignment A
• Questions to address in the report (<3 pages, <1000 words)

– State table and probabilistic production cost model give exactly the
same results of LOLP and EENS
– Estimate the number of samples of Monte Carlo simulation needed to
achieve approximately a 10 % confidence interval in the LOLP with a
confidence level of 95 % and determine the LOLP and EENS for this
Monte Carlo simulation
– Compute the expected energy produced by each generator with each
method
– Compute the expected total variable costs of the system
– Substitute the biggest unit by two smaller ones with the same total
capacity and the same EFOR and evaluate the effect on the reliability
measures

Generation reliability homework. Assignment B
• Choose one of these
– Wind generation at Enel Green Power
– Thermal unit maintenance in the Canary Islands
• Write a technical report of 4 pages (2000 words ± 10 %,
excess or deficit will be penalized) explaining the
assumptions, how the results have been obtained and the
results
• Name the report as AuthorsFamiliyName.pdf. Upload it on to
SIFO
• Work is done in groups of 4 people. It is convenient to have
an Spanish member
• Due date: Wednesday, December 9
• Discussion of homework: Thursday, January 17
Modelos de ayuda a la decisión en el sector eléctrico

Decision support models in electric power systems
Generation reliability
Andrés Ramos
http://www.iit.comillas.edu/aramos/
Andres.Ramos@comillas.edu

Mod 9

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Mod 9

Загружено:

Авторское право:

Доступные форматы

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA

Decision support models in electric power systems

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA

• A. Froger, M. Gendreau, J.E. Mendoza, É. Pinsona, L-M. Rousseau

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA

• Currently, in electric energy systems nobody speaks about

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA

• Reserve Margin (RM):

Available generation - Maximum demand

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA

• Generation system and load

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA

• Deterministic reliability indexes on a dry year

• Power needed for a reliability criterion RM > 2 GW

• Power needed for a reliability criterion RM > 2 GW

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA

LOLPA = 0.05+0.04-0.04·0.05=0.088 (8.8%)

LOLPB = 0.05 ⋅ 0.04=0.002 (0.2%)

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA

• Loss Of Energy Expectation (LOEE) widely known as Expected Energy

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA

• Frequently used indexes, specially in systems with limited

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA

• eXpected Loss Of Load (XLOL) in MW:

EENS (MWh) EENS (MWh)

• Probability Of Positive Margin (POPM):

LOLPN ≤ 1 − POPM = LOLP1

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA

Hydro energy available for peaking = Shoulder hydro output =

2000 h 4000 h 2760 h

Not served power due to lack of Not served energy =

2000 h 4000 h 2760 h

• Emergency Operating Procedure Expectation (EOPE):

LOLP = Frequency ⋅ Duration

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA

2. Monte Carlo Simulation

3. Probabilistic Production Cost Model

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA

• Generation system and load

• Reliability measures 11 1900 0.912 0

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA

• For each sample

• Compute LOLP and EENS

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA

var X (n ) =

• Variance of the mean   n n(n − 1)

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA

• An electric system has 2 generators with a failure

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA

•Unit expected output

ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA