EE5712 Power System Reliability

:: Simulation Methods
Panida Jirutitijaroen
Basic steps in power system reliability analysis
Component modeling
System modeling
Levels of reliability analysis

REVIEW
 Reliability Assessment Steps

Unit & System Models

State Selection

Load Curtailment Evaluation

No Yes

Success Failure Reliability Indices

State State Calculation
 Component Modeling
• Analyze failure effects of each component
• Usually presented by Markov model
• Identify capacity of each state associated with
its probability and frequency
• Steady state probability and frequency
calculated from Markov equations or
frequency balance approach
 System Modeling
• Interaction among • Power flow analysis
components
– Series/parallel network

2
3
1 5 8
4
6 7
 Levels of Reliability Analysis
• Single-area reliability analysis • Composite system reliability
analysis
G1 G2 G3

Load

• Multi-area power systems

reliability analysis • Distribution system reliability
analysis
 Reliability Techniques
• Single-area • Composite system
– Conditional probability – Analytical: contingency
– Convolution enumeration and selection
– Continuous distribution – Simulation
approximation: moment – Hybrid
matching • Distribution
• Multi-area – Series/parallel network
– Analytical: contingency – Network reduction
enumeration and selection – Conditional probability
– Simulation approach
– Hybrid – Cut-set or tie-set approach
Monte Carlo simulation
Random sampling
Sequential sampling
Application to power system reliability analysis
Variance reduction techniques
Issues with renewable energies

OUTLINE
Basic concept
Random number generation
Classification
Estimate reliability measures

MONTE CARLO SIMULATION

 Basic Concepts
• Reliability indices can be estimated by collecting
data on the durations of failures and repairs.
• Monte Carlo method mimics the failure and
repair history of components and the system by
using the probability distributions of component
states.
• Indices are estimated by statistical inference.
• Random numbers are required for Monte Carlo
simulation
 Random Number Generator
• Basic requirement
– Each number should have equal probability of
occurrence
– Statistically independent of the other numbers in the
sequence
• Multiplicative congruential method:
Rn+1 = (a Rn ) (modulo m)
Where
Rn = n th random number
a,m = Positive integers, a < m
Rn+1 = Remainder when (aRn) is divided by m
 Monte Carlo Classification
• Two main approaches in Monte Carlo
simulation
1. Random sampling or Non-sequential simulation
2. Sequential simulation
• Fixed step method
• Next event method
 Estimate Reliability Measures
• Calculate from realizations
• Use statistical inference
• Possibly construct probability distribution for
some indexes, usually main parameters are mean
value
• For example,
– Unavailability (loss of load probability)
– Mean cycle time
– Steady state frequency
 Steady State Probability
• Steady state probability of being in success
states, P̂ 
• Calculate as a fraction of time spent in success
state
• T is total time, t+ is time spent in success state
in interval (0,T)

ˆ  t
P 
T
 Mean Cycle Time
• Average time spent in success state before
leaving to other states, T ˆ 

• ti is the time interval between (i-1) and ith

encounter of success state
• n is total number of success state encounter
n

t i
Tˆ   i 1
n
 Steady State Frequency
• Average frequency of encountering success
state, F̂ 
• Calculate as number of encountering success
state over total time spent in success state
ˆ  n 1
F  n  
ˆ
 ti
i 1
T
Basic steps
Component state sampling
System state sampling
Probability estimation
Convergence criterion

RANDOM SAMPLING
 Basic Steps
• Given probability distribution
• Draw a random number uniformly distributed
between 0 to 1 to determine a component
state
• Two basic approaches
– Proportional probability method for discrete
distribution
– Probability distribution method for continuous
distribution
 Component State Sampling
• Components with • We can assign the
following probability random numbers
distribution proportional to their
probability as follows:
State Probability Random Number State Sampled
(random variable) Drawn
1 0.1 0 to 0.1 1
2 0.2 0.1+ to 0.3 2
3 0.4 0.3+ to 0.7 3
4 0.2 0.7+ to 0.9 4
5 0.1 0.9+ to 1.0 5
 Probability Distribution Method
• Inverse of probability distribution
1.0 1.0
0.9 0.9
0.8 0.8
0.7 0.7

P(X= x)0.6 0.6

0.5 0.5 F(x)=P(X < x)
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
Random Observation
0 0
1 2 3 4 5 1 2 3 4 5
x x
 System State Sampling
• If a system consists of n independent
components, then to sample a system state, n
random numbers are needed to sample the
state of each component
• For a system of two components with the pdfs
shown above, sampling may proceed as
follows
 System State Sampling Example
RN for Component 1 RN for Component 2 System State

0.946 0.601 (5,3)

0.655 0.671 (3,3)
Random State
Number Drawn Sampled 0.791 0.333 (4,3)
0 to 0.1 1 0.345 0.532 (3,3)
0.1+ to 0.3 2 0.438 0.087 (3,1)
0.3+ to 0.7 3 0.311 0.693 (3,3)
0.7+ to 0.9 4 0.333 0.918 (3,5)
0.9+ to 1.0 5 0.998 0.209 (5,2)
0.923 0.883 (5,4)
0.851 0.135 (4,2)
0.651 0.034 (3,1)
0.316 0.525 (3,3)
0.965 0.427 (5,3)
0.839 0.434 (4,3)
 Probability Estimation
• Let’s say we want to estimate the probability
of state (3,3) RN for
Component 1
RN for
Component 2
System
State
0.946 0.601 (5,3)

P(3,3) = n/N 0.655

0.791
0.345
0.671
0.333
0.532
(3,3)
(4,3)
(3,3)
0.438 0.087 (3,1)
Where 0.311 0.693 (3,3)
0.333 0.918 (3,5)
n = number of times state sampled 0.998
0.923
0.209
0.883
(5,2)
(5,4)
0.851 0.135 (4,2)
N = total number of samples. 0.651 0.034 (3,1)
0.316 0.525 (3,3)

• From the table 0.965

0.839
0.427
0.434
(5,3)
(4,3)

P(3,3) = 4/14 = 0.286

• The actual probability of (3,3) is 0.4x0.4 = 0.16
• If this sampling and estimation are continued
the estimated value will approach 0.16.
• The indices obtained are estimates.
• One must have some criterion to decide
whether the indices have converged or not.
 Convergence Criterion
• Typically based on Coefficient of Variation
(COV)
COV = Standard deviation of the estimator
Estimator

• Thus,
1 VarF x 
COV 
Eˆ F x  Number of sample
• The sample sizes can be found from,
VarF x 
Number of sample 

COV  EF x 
ˆ 2

• Not affected by system size or complexity

• Affected by accuracy required (COV)
• Computational efforts depends on number of
sample and CPU time per sample
Basic steps
Fixed time interval method
Next event method
Estimation and convergence

SEQUENTIAL SIMULATION
 Basic Steps
• Generate artificial history over time
• Given mathematical model (Markov chain) of
each component when transition probabilities
over a time step are defined
• Start with initial state
• Sampling conditioned on a given state
• Determine time to move to the next state, two
methods, fixed time interval or next event
 Fixed Time Interval Method
• Synchronous timing method
• Simulated time is advanced by a fixed amount.
• Time step chosen based on operating
characteristic of the system
• At each time step, the system may remain the
same state or change to new state
 Fixed Time Interval Example
• Assume that there is a two state system and
the probability of transiting from state to state
over a single step is given by the following
matrix
Initial State Final State
0 1
0 0.3 0.7
1 0.4 0.6
• Starting in state 0, • Similarly starting in
select a random state 1,
number and the next
state is determined as
follows,
Digit Event Digit Event
0.0+ to 0.3 Stay in state 0 0.0+ to 0.4 Transit to state 0
0.3+ to 1.0 Transit to state 1 0.4+ to 1.0 Stay in state 1
• Construction of a realization for 10 steps,
Step RN State
1 0.947 0
2 0.601 1
Initial State Final State 3 0.655 1
0 1 4 0.671 1
0 0.3 0.7 5 0.791 1
1 0.4 0.6 6 0.333 1
7 0.345 0
8 0.531 1
9 0.478 1
10 0.087 1
 Next Event Method
• Asynchronous timing method
• Simulated time is advanced in variable
amount.
• Keep record of when system will change state
(when new event will occur).
• Advance time step accordingly
• Useful when the times in system states are
defined by continuous distribution functions
 Probability Integral Transform
• Consider a random variable X representing the
up time of a component and distribution,
F(x) = P(X x)
RN
F(x) = P(X < x)
Need to find the time to
change state, i.e. to find
value of random variable X
Value of RV
from random number RN
RV X
• Given probability distribution function of
up/down time of a component, F(x)
• Sample a random number, z, from a uniform
distribution
• Let z = F(x), compute x = F (z)
-1

• It can be shown that x is the random number

generated from distribution F(x).
 Exponential Distribution
• Probability distribution,
P X  x  1  e x
1/ρ is mean of the random variable x.

• Let z  P X  x  1  e x
, since z is uniformly
distributed between 0 and 1, it is equivalent
to z  e  x .
• The random observation with exponential
distribution is found from
ln( z )
x

 Next Event Method Example
• A two component example
Time Random Number Time to change Component
for Component State*

1 2 1 2 1 2
Component Failure rate, λ Repair rate, µ 0 0.946 0.601 5 101 U U
(failure per (repair per 5 0.655 - 0/4  96 D U
hour) hour) 9 0.670 - 0/40  92 U U
49 0.790 - 0/2  52 D U
1 0.01 0.1 51 0.332 - 0/110 50  U U
101 - 0.345 60 0/11  U D
2 0.005 0.1 112 - 0.531 49  0/127 U U
161 0.437 - 0/8  78 D U
169 0.087 - 0/244 70  U U
239 - 0.311 174 0/12  U D
251 - 0.693 162 0/73  U U
324 - 0.333 89 0/ U D

 Reliability Measure Estimation
• Sequential simulation gives yearly reliability
indexes such as number of hours of load loss,
amount of load loss.
• Let
– Ii = a value of reliability index obtained from simulation data for year i
– N = number of years of simulated data

• Estimate is found from,

N
ˆI  1 I i
N i 1
• The standard deviation of reliability indexes
from sequential simulation is
SD2
SDI 
Ny Ny
where SD2 
1
 I i  I 2

NY i 1
• The standard deviation of the estimate, Ī,
varies as 1 N y and will approach zero as Ny
approaches infinity.
Single-area reliability analysis
Composite system reliability analysis
Multi-area power systems reliability analysis
Distribution system reliability analysis

APPLICATION TO POWER SYSTEM

RELIABILITY ANALYSIS
 Reliability Assessment Steps

Unit & System Models

State Selection

Load Curtailment Evaluation

No Yes

Success Failure Reliability Indices

State State Calculation
 Single Area Reliability Analysis
• Single bus analysis
• Generators and loads
are within the same bus G1 G2 G3

• Each generators has

their own performance Load
indexes
• No transmission
capacity limits
Application of a Simulation Method

Sample a state
according to each Evaluate: Reserve = Generation - Load
component’s failure
and repair rate
Update
G1 G2 G3 statistical
State space
value of
reserve
Load

Keep sampling until the

convergence criteria is satisfied
 Composite System Reliability
• Interconnected single-
area
• Concern with generation
and transmission
capability adequacy
• High-voltage transmission
lines
• May include high-voltage
transformers, circuit
http://www.powerworld.com/images/7-bus%20Oneline.jpg
breakers
Application of a Simulation Method

Evaluate: Optimal power flow

Sample a state
• AC flow
according to each
• DC flow
component’s failure
and repair rate
Update
State space statistical
value of
Load loss

Keep sampling until the

convergence criteria is satisfied
 Multi-Area Reliability Analysis
• Consider thousands of
nodes then simplify the
system to small workable
nodes (areas), each of
them form a control area
• Generator and load from
different nodes within the
same area are grouped
into one.
• Tie-line capability
between areas
Application of a Simulation Method
Evaluate: Optimal power flow
Sample a state • AC flow
according to each • DC flow
component’s failure • Network flow
and repair rate
Update
statistical
State space value of
area load
loss

Keep sampling until the

convergence criteria is satisfied
Motivations
Some variance reduction techniques
Roles of artificial intelligences

VARIANCE REDUCTION TECHNIQUES

 Motivations
• Precision of estimates increase with larger
sample size
• Repeated runs or single run with large
samples
• Ways to reduce variance of the estimates
• Try to extract information from small amount
of simulation
 Available Techniques
• Importance sampling
• Control variates sampling
• Antithetic variates sampling
• Stratified sampling (proportional sampling)
• Latin Hypercube sampling
 Importance Sampling
• Modify distribution functions to increase loss of
load occurrence
• Manipulate state space to yield more loss of load
states
• For example
– State space pruning
– Try to prune for non of loss of load state
– Perform Monte Carlo simulation only to the rest of
system states
• Need to map back to the original state space
Control Variates & Antithetic Variables
• Modify distribution of a random variable of
interest
– For example, loss of load
• Introduce new random variable
• Yield the same expected value as the random
variable of interest
• Yield lower variance
• Use only to estimate mean value of index
 Control Variates Sampling
• Z be a random variable that is strongly correlated with F,
Y = F – α(Z – E(Z))
• This means that
E(Y) = E(F)
• Variance of the new random variable,
V(Y) = V(F) + α²V(Z) - 2αC(F,Z)
• If 2αC(F,Z) > α²V(Z), then V(Y) < V(F)
• Example
– F = system loss of load
– Z = system loss of load due to generation only
– E(Z) can be computed without transmission line consideration
 Antithetic Variables
• Let Fα = ½(F’ + F’’)
• If E(F’) = E(F’’) = E(F) then,
E(Fα) = E(F)
• And,
V(Fα) = ¼*V(F’) + V(F’’)+ + ½C(F’,F’’)
• If C(F’,F’’) < 0, F’ and F’’ are negatively
correlated, then
V(Fα) < V(F)
Stratification or Proportional Sampling
• Divide state spaces into independent groups (stratum).
• Each group (strata) is mutually exclusive i.e. each state
belongs to only one group.
• Sampling on each group proportional to the probability of
occurrences of that group
• Each sample, however, has the same probability of
occurrence with in the group
• For example
– A state space is divided into 4 groups with the following
probabilities, 0.1, 0.25, 0.4, 0.35
– 100 samples
– Relative size of the samples are 10 from 1st group, 25 from 2nd
group, 40 from 3rd group, and 35 from 4th group.
 Lain Hypercube Sampling
• Mixed between random sampling and stratified
sampling
• Predetermined sample size
• Divide the distribution function into intervals of
same probability according to the sample size
• Perform random sampling from each subinterval
according to the distribution within that interval
• One sample per subinterval
• Estimate both mean value and distribution of
index
 LHS Example
• Given distribution
function
• 10 sample size
• Divide intervals with
equal probability of 0.1
• Draw randomly one
sample from each
subintervals
 LHS Application
• Most of the units described by failure and repair rates
• Need to construct generation probability distribution of
each bus
– May use unit addition algorithm
• Load distribution is modeled
• System state is found from random permutation of each
random variables in the problem
– G: generation distribution
– L: load distribution
– T: transmission capacity distribution
– LHS sampling G’, L’, and T’ separately
– System state (G,L,T) is found from randomly select without
replacement from G’, L’, and T’
Renewable energy sources
Challenges
Modeling
Reliability evaluation techniques
Simulation methods

ISSUES WITH RENEWABLE ENERGY

 Renewable Energy Sources
• Generated from natural resources
– Wind power
– Solar power
– Micro hydro
– Geothermal
– Ocean energy
• Focus mainly on wind and solar
 Challenges
• Generation depends on the weather
– Solar depends on sun intensity, duration.
– Wind depends on wind speed
• Load depends on the weather
• Correlations between generation and load
• Need to construct the multi-variable distribution
of both generation and load
• Or, use other simulation methods to incorporate
this correlation.
 Modeling
• Statistical analysis of
generation and load Generation
• To compute the level of Time
correlations
• Can be modeled as
multi-variable Load Time
distribution
– For example, area load Two random variables become
a two-dimensional random
modeling
variable
 Reliability Evaluation Techniques
Analytical Simulation
• Contingency enumeration • Monte Carlo simulation
• Generation and load • Need to incorporate
distribution are no longer correlations in two
independent approaches
• Need to consider the – Explicit multi-dimension
correlations in the analysis random variables and the
sampling process is the same
• Ranking may be changed – Or, remain separate random
variables and change the
sampling process
 Simulation Methods
• Monte Carlo simulation
– Sequential simulation
– Include all correlations in time sequence
• Latin Hypercube simulation
– Use permutation matrix to yield desired level of
correlations
– Involve optimization
 Reliability Assessment Steps

Unit & System Models

State Selection

Load Curtailment Evaluation

No Yes

Success Failure Reliability Indices

State State Calculation
 Summary
• Large state space, impossible to enumerate
• Monte Carlo simulation
– Random sampling
– Sequential sampling
• Variance reduction techniques
• Issues with renewable energies
– Incorporate correlations between random
variables
 Next Lecture
• Another analytical method to classify state
space to determine loss of load states
• Decomposition technique

Success Failure
states State space states
(Loss of load
states)
 References
• Electrical Power System Reliability Course notes,
C. Singh, part 7, 8, and 11.
– http://www.ee.tamu.edu/People/bios/singh/courseno
tes/
• “Risk Assessment of Power Systems”, W. Li,
chapter 4 and 5.
• “System Reliability Modeling and Evaluation”, C.
Singh, chapter 7
– http://www.ece.tamu.edu/People/bios/singh/sysrelia
bility/ch7.pdf