You are on page 1of 42

Power System Reliability Analysis

with RES

Dr. Naran M. Pindoriya


Assistant Professor, EE Department
Indian Institute of Technology Gandhinagar

25/05/11

Talk outline

Introduction to Reliability

Power system Reliability

25/05/11

MCS-SVM Model for Composite


Reliability Assessment
Power System Reliability Analysis with
Renewable Energy Sources

What is Reliability ?
Basic Steps in System Reliability Analysis
Objective of the analysis

Component /system modeling

Performance function

Reliability Evaluation

Introduction to Reliability

25/05/11

What is Reliability?
Ability of a system to perform its intended function
Within a specified period of time
Under stated condition

Relate to the absence of failures, that due to random


phenomenon (e.g., Random failures, Uncertainties)
Define numerically as average or mean value
Can be treated as a parametric quantity
Can be traded off with other parameters such as cost
How to model uncertainty
?
25/05/11

How to model Uncertainty ?


Probability of failure
Chance that a component will fail
Probabilistic value with no unit
May be difficult to interpret

Frequency of failure (failure rate)


In terms of number of failure within specified time
Easier to predict from history
Express in per hour, per day, per year

How to quantify reliability ?


25/05/11

Example : Transmission Lines


100 MW

100 MW

Load
100 MW

Load 100
MW

100 MW
System B

System A

Given that each transmission lines has the following


level of reliability
System

Failure
Probability

Cost (million Rs)

0.01

100

0.1

25

Which system is more reliable?


Which system is more cost-effective?
25/05/11

Cost-Benefit Analysis
High reliability achieved with high cost
Is it worthwhile to have high reliability?

Source: http://www.eppo.go.th/power/ERI-study-E/ERI-EOCS-1-E.html
25/05/11

System Reliability Analysis


Objective: Interest to know the time-to-failure
distribution of a component/system
Helps to predict the failure probability at any point in time

= failure rate
q

Basic Steps:
q

Component/ System Modeling

Down
= repair rate

Describe state of each components in the system

Ex: a generator has two states (either up or down)

In terms of probability distribution

25/05/11

Up

Ex: a generator fails with probability of failure =


0.01.
8

System Reliability Analysis Contd


q

Performance Function

Need to define intended function.

Ex: Minimization of load curtailment

Reliability Evaluation

25/05/11

Each component described by random variables


Gen. states

Output (MW)

Probability

100

0.85

50

0.14

0.01

System states constructed from possible combinations of


component states

Objective of Reliability Analysis


Levels of Reliability Analysis
Power System Reliability Indexes and Criterion

Deterministic
Probabilistic

Power system Reliability

25/05/11

10

Uncertainties in Power Systems


Generation
Generating units with failure and repair rates
Generating capacity associated with probability

Transmission line capacity


Transmission line with failure and repair rates
Transmission line capacity associated with probability

System load
Vary with time
Construct load distribution from history

25/05/11

11

Objective
Reliability is a measure of the ability of the power system to
deliver electricity to all points of utilization within accepted
standards and in the amount desired, for the period of time
intended, under the operating conditions intended.
RELIABILITY

Adequacy
Adequacy : relates to the existence of
sufficient facilities within the system
to satisfy the consumer load demand
at all times; taking into account
scheduled/ unscheduled outages

assessed using the power flow


(AC/DC) solutions

25/05/11

Analyzed either on deterministic


or probabilistic basis

SECURITY
Security : ability of the electric
systems to respond to sudden
disturbances arising within that
system, such as electric short circuits

assessed using dynamic calculation


12

Areas of Power System Reliability Analysis


Generating capacity reliability
Concern with generation adequacy
All generators and loads are
connected to a single bus

Distribution system
reliability

LEVEL

Composite system reliability

Local network
connected to end-users
Interest to find out the
reliability level at load
point

Concern with generation and


transmission capability adequacy

25/05/11

13

Composite Power System Reliability


Basic intention:- to determine some probabilistic
measure of the undesirable events in power systems
Unit and System Models

State
Selection

Operating Strategies

Load
Curtailment
No

Yes

Success
State

Failure State

Classification of
system states in the
whole state space

Evaluation

Success
States

Failed
States

Reliability
Indices
Calculation

State Space
25/05/11

14

Power Systems Reliability Indexes


Probabilistic indexes

Deterministic indexes

Do not take into


account the
uncertainties that
affect reliability
Simple calculation and
require less data
Percentage reserve
Reserve margin as the
largest unit online

25/05/11

Reflect uncertainties in the


system
Loss of load probability
(LOLP)

Probability that generation


will not meet demand in a
year
Loss of load frequency (LOLF)

How often does the system


fail in a year
Expected energy not supplied
(EENS)

15

Power Systems Reliability Criterion


Deterministic criteria
N-m contingency analysis
System with N components should be able to serve peak
load when loss m components
Sometimes called security analysis

Probabilistic criteria
Loss of load expectation, for example, 1 day in 10 years

25/05/11

16

Monte Carlo Simulation (MCS): Introduction


SVM and LSSVM classifier Tool
MCS-LSSVM Model for Reliability Assessment

Simulation results for different case studies

MCS-SVM Model for Composite Reliability


Assessment

25/05/11

17

Reliability Evaluation: Methods

Classical
Approaches
demands
strict
mathematical analysis

use some device to


circumvent the problem
of straightforward
enumeration such as

25/05/11

State space pruning,


Variance reduction
technique

Simulation

Select system states based on


their respective sampling
mechanism (e.g., sequential or
random sampling)

Monte Carlo Simulation (MCS)


AI based algorithm used in

State selection as an
alternative to MCS (for ex.
PSO, GA, etc)

Pattern classification
techniques for state
evaluation as an aid to
MCS

18

Simulation Methods: Remarks (1)


The most significant difference between MCS and AI
based search algorithm lies in their sampling
mechanism.
MCS Method: Some Remarks

system states are sampled based on their occurrence probability, and both success
and failure states sampled contribute to the estimation of reliability indices.
ability to model complex systems in more detail and accuracy than is possible in
analytical methods;
can not only calculate the expected value of reliability indices but also their
distributions
Even though the state is a repeated sample, is still count for index calculation
when MCS is used to deal with highly reliable systems, its efficiency may become
low since a large number of system states need to be sampled and evaluated. (e.g.
quite time-consuming)

25/05/11

19

Simulation Methods: Remarks (2)

AI based Search Method: Some Remarks

25/05/11

Unlike MCS, AI based search method is rather problem-dependent,


where system states with higher failure probabilities have higher
chances to be selected and evaluated.
the failure probability of system state is used to guide the search.
Also, unlike MCS, in AI based search method only the failure states
are useful in estimating reliability indices.

20

MCS for Composite Reliability Evaluation


MCS: the non-sequential and the sequential
MCS
the non-sequential approach samples the system states
randomly,

while in sequential approach the system states preserve the


Computational
steps for the of
non-sequential
chronological characteristics
the system

MCSSelect a state of the power system,


1.

2.

by random
sampling the states of all components and the load levels.
Characterize (or classify) the selected state, x, (success or failure)
through test function f(x) , by performing the adequacy analysis,
which usually involves optimal power flow (OPF) analysis.

3.

Update the estimate, E(f )

4.

If the stopping criterion is satisfied, stop; otherwise return to step 1.

25/05/11

21

IEEE-RTS-79 Test System


Unit 22
(1400 MW)

Unit 23
(1400 MW)

BUS 18

Unit 24~29
(650 MW)

BUS 17

BUS 21

BUS 22

BUS 23
Unit 21
(1155 MW)

BUS 16

BUS 19

Unit 30~31
(2155 MW)
Unit 32
(1350 MW)

BUS 20

BUS 14

Synch.
Cond.

BUS 15

Unit 12~14
(3197MW)

Unit 15~19
(512 MW)
Unit 20 (1155
MW)

BUS 13
(slack bus)

BUS 24

BUS 11

BUS 12

138 kV

BUS 9

BUS 10

cable

BUS 6

BUS 4

BUS 5

BUS 8

cable
BUS 1
Unit 1~2
(220 MW)
Unit 3~4
(276 MW)

25/05/11

BUS 2
Unit 5~6
(220 MW)
Unit 7~8
(276 MW)

BUS 7
Unit 9~11
(3100 MW)

230 kV (dominated by
generation, 2721 MW)
Ckts are, fully available at all
times
Load buses are considered to
the fully correlated with the
total system load

230 kV

BUS 3

24 buses (10 generation buses


and 17 load buses), 38 Ckts, 32
generating units, Total installed
capacity: 3405 MW and peak
load: 2850MW
Two well defined areas:

138 kV (dominated by load)

22

Load Profile of IEEE RTS-79


1
0.9

Load (pu)

0.8
0.7
0.6
0.5
0.4

1000

2000

3000

4000

5000

6000

7000

8000

8736

Hours

Case 1: Fixed peak load =


2850 MW
Case 2: Multiple load levels
Case 3: Time varying load

25/05/11

Load

Prob.

Load

Prob.

Load

Prob.

1681.5

0.08333

2109.0

0.04167

2679.0

0.04167

1710.0

0.08333

2365.5

0.04167

2707.5

0.04167

1795.5

0.08333

2451.0

0.04167

2736.0

0.12500

1909.5

0.04167

2593.5

0.04167

2821.5

0.04167

2080.5

0.04167

2650.5

0.04167

2850.0

0.08333

23

MCS-LSSVM: Flowchart
Testing patterns obtained
by random states
sampling (MCS
computation-step 1)

Input/output training
data set obtained by
MCS procedure

Extract the training


patterns through Kmeans clustering

LSSVM classifier modeling


and supervised training
(10-fold cross validation)

Once LSSVM is trained

Identify most
relevant input
variables

Power system state


space pre-classification
by the trained LSSVM
model instead of OPF

Classifier accuracy
assessment and
calculate reliability
indices by analyzing
only failure states
classified by LSSVM

# Naran M. Pindoriya, Panida Jirutitijaroen, Dipti Srinivasan, and Chanan Singh, Composite reliability
evaluation using MCS and least squares support vector classifier, IEEE Transactions on Power Systems, Feb.
2011 (Accepted and available for early access).
25/05/11

24

Introduction to SVM
SVM provides an approach to the two-category (operating or
failed) classification problem with clear connections to the
underlying statistical learning theory
Let, the problem of separating the set of training
vectors (N data points) belongs to two separate
classes:
D=

{ ( x , y ) ,..., ( x
1

with a hyperplane:

, y N ) , x n , y { 1,1}
H : y = w
xb = 0

w (weight vector) and b (bias) are the


parameters that control the function.
the
is
b /the
w perpendicular distance to the origin.

Linear
separation
25/05/11

Optimization problem:
1
Min wT w , s.t. yi ( w
x b ) 1
w ,b 2

LR augmented optimization function


25

Non-linear SVM: If the surface separating the two classes is not


linear, the data points can be transformed to another high
dimensional feature space where the problem is linearly separable
( ) the
Let, the transformation be
then
lagrangian function in the high dimensional
feature space is:
LD = i
i

1
i j yi y j 1( 4xi 2) 4( x3j )
2 ij
(

k xi , x j

Mapping the input space to the feature space,


where linear classification is possible

25/05/11

26

LSSVM
In contrast to the standard SVM,
the LSSVM uses a least squares cost function and involves
equality constraints instead of inequalities in the problem
formulation.
As a result, the solution is obtained by solving a set of
linear equations instead of QP and hence, LSSVM can
reduce the computational complexity.
Kernel functions

25/05/11

27

Input Data Projection


Input data projection (Case 2)
1800

Input data projection (Case 3)


2500

Success state

Success state

1500

1150

800

450

Failure state

Generation Reserve (MW)

Generation Reserve (MW)

Failure state

25/05/11

200
400
600
800
1000
Unavailable Generation Capacity (MW)

1200

2000

1500

1000

500

300
600
900
1200
Unavailable Generation Capacity (MW)

1500

28

Case 1 (Fixed peak load = 2850 MW)


Training patterns are generated
by
MCS

Algorithm
runs until coefficient of

0.5

250

0.4

200

No. of total samples = 873

No. of success states = 783

No. of failure states = 90

LOLP = 0.1031, EPNS = 187.19 MW

Execution time for MCS= 17.78


Sec

SVM training data patterns

Inputs : [ unavlbe_gen, res_gen ]


Training samples = 270 [failure
state (1): success states(2)]
Obtained using K-means
clustering algo.

LOLP/Coefficient of variation

LOLP
Coeff. of variation
EPNS

0.3

150

0.2

100

0.1

50

100

200

300 400
500 600
Number of iterations

700

800

0
900

EPNS(MW)

variation () converge to 10%

Case 1 (Fixed peak load)


Case-1 : Simulation
results
LSSVM Classifier Performance (Case 1)
MCS-LSSVM

MCS
(benchmark)

Linear kernel

RBF kernel

# success states

24269

23967

24269

# failure states

2285

2587

2285

Sensitivity (%)

NA

98.76

100

Specificity (%)

NA

100

100

g-mean (%)

NA

99.38

100

Composite Reliability Indices Comparison (Case 1)


LOLP
Esti.
index

Error
(%)

Esti.
index

Error
(%)

Total
Comp.
time
(sec)

0.08609

--

174.05

--

603

Linear
kernel

0.08609

0.000

174.05

0.000

70

RBF
kernel

0.08609

0.000

174.05

0.000

64

MCS
(benchmark)
# Naran M. Pindoriya, Panida Jirutitijaroen,
Dipti Srinivasan, and Chanan Singh, Composite
reliability evaluation using MCS and least
squares support vector classifier, IEEE
Transactions on Power Systems, Feb. 2011
(Accepted and available for early access).

MCSLSSVM

EPNS (MW)

Case 2 (Multiple load levels)


0.1

200

100

Coefficient of variation

0.05

EPNS(MW)

LOLP

0.8

0.6

0.4

0.2

500

1000

1500 2000 2500


No. of samples

3000

3500

0
4000

500

1000

1500 2000 2500


No. of samples

3000

3500

Training patterns are generated by MCS

Algorithm runs until coefficient of variation () converge to


0.15

No. of total samples = 3701

Execution time for MCS= 180.76 Sec

4000

Case 2 (Multiple load levels)


INPUT DATA PROJECTION
1800

1600

1600
Generation reserve (MW)

Generation reserve (MW)

INPUT DATA PROJECTION


1800

1400

1200

1000

800

600

400

1400

1200

1000

800

600

200
400
600
800
1000 1200
Unavailable generation capacity (MW)

1400

No. of success states =


3603
No. of failure states = 98
LOLP = 0.0265
EPNS = 144.28 MW

400

200
400
600
800
1000 1200
Unavailable generation capacity (MW)

1400

SVM training data patterns

Inputs : [ unavlbe_gen, res_gen ]


Training samples = 294 [failure
state (1): success states(2)]
Obtained using K-means
clustering algo.

Case 2 (Multiple load levels)


LSSVM Classifier Performance (Case 2)
MCS
(benchmark)

MCS-LSSVM
Linear
kernel

RBF
kernel

# success states

84079

83119

83487

# failure states

2430

3391

3023

Sensitivity (%)

NA

98.86

99.29

Specificity (%)

NA

100

99.60

g-mean (%)

NA

99.43

99.44

Composite Reliability Indices Comparison (Case 2)


LOLP

25/05/11

Comp.
time
(sec)

Esti.
index

Error
(%)

Esti.
index

Error
(%)

0.02811

--

156.76

--

2639

Linear
kernel

0.02811

0.00

156.76

0.00

100

RBF
kernel

0.02800

-0.4032

155.18

-1.010

99

MCS
(benchmark)

MCSLSSVM

EPNS (MW)

33

Case 3 (Time varying load)


-3

x 10

120

1
0.9

80

1.5

60

40

0.5

20

5000

10000
No. of samples

15000

Coefficient of variation ( )

100

EPNS(MW)

LOLP

2.5

0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1

2000 4000

6000 8000 10000 12000 14000 16000


No. of samples

Training patterns are generated by MCS

Algorithm runs until coefficient of variation () converge to


0.2

No. of total samples = 16518

Execution time for MCS= 350.78 Sec

Case 3 (Time varying load)


TRAINING SAMPLES

INPUT DATA PROJECTION


2500

2400

success state

2200

failure state

2000

Generation reserve (MW)

Generation reserve (MW)

2000

1500

1000

1800
1600
1400
1200
1000
800
600

500

500
1000
Unavailable generation capacity (MW)

1500

No. of success states = 16493

No. of failure states = 25

LOLP = 0.0015

EPNS = 98.54 MW

400

500
1000
Unavailable generation capacity (MW)

1500

SVM training data


patterns
Inputs : [ unavlbe_gen, res_gen ]
Training samples = 175 [failure
state (1): success states(6)]
Obtained using K-means
clustering algo.

Case 3 (Time varying load)


INPUT DATA PROJECTION (TESTING SET)
2500

MCS (Benchmark)

Testing samples through MCS until it


reaches ( = 0.05) = 2,97,139
Execution time for MCS= 20753.21 Sec.
Success states = 2,96,739 and Failure
states = 400
LOLP = 0.001346, EPNS = 120.516 MW

Generation reserve (MW)

2000

1500

1000

500

MCS-LSSVM

200

400
600
800 1000 1200 1400
Unavailable generation capacity (MW)

LOLP

EPNS (MW)

Kernel
type

# Success
states

# Failure
stats

Sensitivity
(%)

Specificity
(%)

g-mean

Esti.
index

Error
(%)

Esti.
index

Error
(%)

LSSVM+
OPF
time (sec)
[A]

Lin. kernel
RBF kernel
Poly. kernel

295890
295702
296175

1249
1437
964

99.71
99.65
99.81

99.75
100
100

99.73
99.83
99.91

0.001343
0.001346
0.001346

-0.25
0.00
0.00

120.768
120.516
120.516

0.21
0.00
0.00

26.35
33.25
20.74

1600

1800

Total
(MCS for
tra. patt.
=350.8+A)
377.15
384.05
371.54

Case 3 (Time varying load)


LSSVM Classifier
Performance
# Naran M. Pindoriya, Panida Jirutitijaroen,
Dipti Srinivasan, and Chanan Singh, Composite
reliability evaluation using MCS and least
squares support vector classifier, IEEE
Transactions on Power Systems, Feb. 2011
(Accepted and available for early access).

LOLP

Linear
kernel

RBF kernel

Polynomial
kernel

# success states

305362

303581

304520

304790

# failure states

400

1242

1468

972

Sensitivity (%)

NA

99.72

99.65

99.81

Specificity (%)

NA

99.60

99.88

100

g-mean (%)

NA

99.66

99.76

99.90

Esti.
index

Error
(%)

Esti.
index

Error
(%)

Com.
time (sec)
[A]

0.00131

--

126.41

--

21080

NA

Lin. kernel

0.00130

-0.400

126.86

0.355

26.02

376.75

RBF kernel

0.00131

-0.125

126.53

0.1026

33.33

383.79

Poly. kernel

0.00131

0.000

126.41

0.000

20.74

371.97

MCS
(benchmark)

MCSLSSVM

EPNS (MW)

MCS-LSSVM

MCS
(benchmark)

Total [MCS
for tra. patt.
(=350.8) +A]

Composite Reliability
Indices Comparison

Concluding Remarks
LSSVM classifier takes the equality constraints in place of
the inequality counterparts with SVM, and the solution
follows from solving a set of linear equations, instead of
quadratic optimization problem for SVM.
Because the LSSVM is fast and effective nonlinear
classifier in compare to ANN classifiers, it has used to preclassify the entire system operating states into success or
failure, so then only failure states are fully evaluated for
adequacy analysis to calculate composite reliability
indices.
MCS LSSVM allows to avoid the adequacy analysis of
success states (which are usually much greater than the
number of failure states in power systems) and hence it
provides significant reductions in the computational cost
required while evaluating composite reliability.
25/05/11

38

Case Studies

ERCOT System with wind energy


Augmented IEEE RTS with PV generation

Power System Reliability Analysis with


Renewable energy sources

25/05/11

39

Objective
Reliability analysis of power system including RES,
with an emphasis of bus loads and intermittent
behavior of RES such as wind and solar power
q

ERCOT System with wind energy

Zhen Shu and Panida Jirutitijaroen, Latin Hypercube Sampling Techniques for Power Systems Reliability
Analysis With Renewable Energy Sources, IEEE Transactions on Power Systems, Jan. 2011 (Accepted and
available for early access).
25/05/11

40

Augmented IEEE RTS with PV


generation

Weekly load and PV curves in IEEE RTS


case
PV power generated from MIT Weather Station in
2009, Available: http://pvbase.mit.edu/cgibin/index.py.
Zhen Shu and Panida Jirutitijaroen, Latin Hypercube Sampling Techniques for Power Systems Reliability
Analysis With Renewable Energy Sources, IEEE Transactions on Power Systems, Jan. 2011 (Accepted and
available for early access).
25/05/11

41

Thank you for your attention !!!

Questions ???

25/05/11

42

You might also like