International Conference on Computational Intelligence and Multimedia Applications 2007

Thermal Unit Commitment Problem by using Genetic Algorithm,

Fuzzy Logic and Priority List Method

D. P. Kadam S. S. Wagh P. M. Patil

Electrical Engineering Department
K. K. Wagh Institute of Engineering Education and Research, Nasik (MS) India
[email protected] [email protected] [email protected]

Abstract
In electrical power industries saving in fuel costs can be achieved by optimized commitment
of available generating units. This paper describes the application of genetic algorithm and
fuzzy logic for determining short-term commitment of thermal units in electrical power
generation. Feasibility of these methods is examined and preliminary results to determine near
optimal commitment order of thermal units in studied power system over short term are
reported. The results obtained from genetic algorithm and fuzzy logic based approach are
compared with the priority list method solution to unit commitment problem. The comparison
proves that genetic algorithm and fuzzy logic based approach are powerful tools for solving
such highly non-linear, multi constrained optimization problems in electrical power systems.

1. Introduction
The Unit Commitment Problem (UCP) is to decide when to turn-on and turn-off available
electrical power generating units to meet the load requirements satisfying all operational
constraints at minimum operating cost. The production cost includes fuel, turn on, turn off and
no load costs. The operational constraints include demanded power, reserve requirements,
generation limits, minimum up time, minimum down time, crew constraints and ramp up limits.
Several optimization and mathematical programming techniques have been applied to solve
thermal UCP. These include methods like priority list based method, dynamic programming,
branch and bound method, lagrangian relaxation, expert systems and simulated annealing [3].
The more commonly used method being simple and fast by electricity utilities is the priority
list method. This method is used to rank generating units in a heuristic with increasing
operation cost. Priority listing of generating units can also take into account utilization factor
and average full load cost (AFLC) associated with generating units. The calculation time for
this method is less due to reduced combinations of units at each time period even for large
systems. This makes it more suitable for our purpose [1].
Genetic Algorithm (GA) provides a solution to UCP by working with a population of
individuals each representing a possible solution. Together with a set of the main genetic
operators of crossover and mutation this method provides a powerful global search mechanism,
whose computation code is simple [5].
Fuzzy logic (FL) is useful in reducing the need for complex mathematical models in problem
solving. Fuzziness is used to describe uncertainty, which is applicable to the UCP. Loading of
generators, start up cost, incremental cost and production cost are considered to be fuzzy
variables with the UCP [9].

0-7695-3050-8/07 $25.00 © 2007 IEEE 468

DOI 10.1109/ICCIMA.2007.338
2. Thermal unit commitment problem formulation
UCP is a large-scale non-linear mathematical optimization problem in electrical power
system involving large number of variables, whose accurate solution can result in large savings
for electricity utilities. The objective function for n generating units (i = 1, 2,….., n ) over
scheduling time horizon t ( j = 1, 2, ….., t) is to minimize total operating costs (OC) satisfying
operational constraints (8),
t n
Minimize OC = ∑ ∑ Sij (FCi) Pij + SUCi Sij (1 – Si,j-1) + SDCi Si, j-1 (1- Si,j)
j=1 i=1

where, Sij = status of unit i at hour j ( 1 if unit is on and 0 if unit is off)

FCi = fuel cost associated with unit i
Pij = power output of unit i at hour j
SUCi and SDCi is start up cost and shut down cost of unit i respectively.
FCi (Pij) = Ai + Bi Pij + Ci Pij2 where Ai, Bi and Ci are cost function constants for unit i.
SUCi = b1i (1- exp (- b 3i X i)) + b 2i where b 1i, b 2i and b3i are start up cost parameters
and Xi is the number of consecutive hours for which unit i has been down.
Subjected to following constraints,
i. Total power generated should meet the load requirement and system losses,
n
∑ Sij Pij = PDj j = 1, 2, ……, t where PDj is power demanded at hour j.
i=1

ii. Spinning reserve at each hour must be satisfied to cover any shortfall in generation,
n
PDj + SRj = ∑ Sij Pimax , j j = 1, 2, ……, t where SR j is spinning reserve at hour j.
i=1

iii. Each generator must operate within its minimum and maximum power output limits,
Pimin ≤ Pij ≤ Pimax i = 1, 2, ……, n

iv. The consecutive number of hours for which a generating unit must remain on (minimum up
time, MUT) or off (minimum down time, MDT) should not get violated,
Uij = 1 for Tion < MUTi and Uij = 0 for Tioff < MDTi
Where, Tion and Tioff is the consecutive number of hours for which the unit is on and off till the
end of last hour respectively.

3. Genetic algorithm based short-term thermal unit commitment

Genetic algorithms are search algorithms based on the mechanics of natural selection and
natural genetics. They combine Charles Darwin’s survival of fittest theory with a structured and
random information exchange. The resulting algorithm has innovative flair of human search [6].
Genetic algorithm works with a population of chromosomes. A chromosome is a string of
bits 0 or 1, whose length is the product of number of generators and scheduling period. Binary
encoding in which zero denotes the off state and one represents the on state is suitable for UCP
formulation. The vector initialization with random selection of bits is then repeated for the
number of chromosomes in the population. The fitness of each member of the population is

469
calculated by an evaluation function to measure performance using the forecasted load demand
at each period and operating constraints. It is repeated for fixed number of generations. The
chromosomes that are highly fitted get selected and those showing poor performance are
offered with penalty in terms of cost. Using genetic operators like selection, crossover, mutation
and accepting new offspring create the new generation.
At each time period, a number of possible commitment strategies with smaller overall costs
are saved. If for a particular period no feasible solution obtained, the steps are repeated to find
out at least one feasible combination. Some strategies are used to improve the performance of
search mechanism for UCP. Fitness proportional scaling is used to increase diversity in
population. In order to avoid loss of best solutions during run, elitist strategy to replace an equal
number of randomly selected population members in the next generation with best performing
members is used. It is also possible to change control parameters during run. The optimal path
for minimum cost strategy is determined to calculate complete unit commitment schedule
with total cost at the end of scheduled time period [7].

4. Test System
In order to prove effectiveness of GA for solving UCP, it is applied to test system of eight
units (n = 8) over time period of eight hours (t = 8) [1,9].
Table 1. Load demand and reserve requirement of test system.
Hour (t) 01 02 03 04 05 06 07 08
PDj ( MW ) 400 470 520 510 360 240 240 450
PDj+Rj ( MW) 450 530 600 540 400 280 290 500
Table 2. Generating unit characteristics of test system.

Unit n Pimax MUTi MTDi IC AFLC / SDCi SUCi

MWh b1i b1i b2i
1 080 3 2 -1 20.88 0 0350 158.00 0.400
2 250 2 1 -2 18.00 0 0400 162.62 0.900
3 300 4 2 1 17.46 0 1100 421.18 0.480
4 060 2 3 -4 23.80 0 0.02 000.02 0.018
where IC: Initial condition of unit ( - sign indicates that unit is down for hours otherwise +).

5. Simulation results with GA base approach to test system.

The programs are developed using MATLAB software and GA parameters used are,
Crossover probability: 0.65, Mutation probability: 0.01, Number of generations: 50, Elitism: 3.
Table 3. Result for test system with GA based approach

Hour j Sij Pj MW OCj

i=1 i=2 i=3 i=4
01 0 1 1 0 550 09966
02 0 1 1 0 550 09738
03 0 1 1 1 610 11166
04 0 1 1 1 610 11166
05 1 0 1 1 440 08797
06 1 1 0 0 330 06570
07 1 1 0 0 330 06170
08 0 1 1 0 550 10838
Cumulative cost in Indian Rupees (INR), OC 74411

470
6. Fuzzy logic (FL) based short-term thermal unit commitment.
For UCP formulation generator loading (GL), incremental cost (ICR), start up cost (SUC),
operating cost (OC) are selected as fuzzy variables. The operational constraints like MUP,
MDT and generation limits are treated as crisp variables. Fuzzy sets defining these variables are
selected and normalized between 0 and 1. This normalized value can be multiplied by a selected
scale factor to accommodate any desired variable. The fuzzy sets for selected variables are,
GL = [low, below average, above average, high], ICR = [zero, small, large],
SUC = [low, medium, high], OC = [low, below average, average, above average, high]
For fuzzy variables triangular membership function is selected and a series of fuzzy rules are
built up using if-then statements. In fuzzy set notation, OC = GL ∩ ICR ∩ SUC.
Using center of gravity defuzzification method, operating cost is formulated as follows [9],
n n
OC = [ Σ µ (OC)i (OC)i] / [ Σ µ (OC)i ]
i=1 i=1

where, µ (OC)i = the membership value of the clipped output

OCi = the quantitative value of the clipped output
n = the number of the points corresponding to quantitative value of the output.
Table 4. Result for test system with FL based approach
Hour j Pj MW P1j (MW) P2j (MW) P3j (MW) P4j (MW) OCj
01 450 300 150 0 0 09426.7
02 530 300 230 0 0 10927.0
03 600 300 250 50 0 13121.0
04 540 300 240 0 0 10927.0
05 400 300 100 0 0 08488.6
06 280 280 0 0 0 05417.3
07 290 290 0 0 0 05800.0
08 500 300 200 0 0 10576.0
Cumulative cost in Indian Rupees (INR), OC 74683.6
7. Priority list based short-term thermal unit commitment.
Priority list could be obtained after an exhaustive enumeration of all unit combinations at
each load level. The priority list could be obtained in a much simpler manner by noting the
AFLC for generating units of the selection pool. This method arranges the generating units in a
start-up heuristic ordering by operating cost combined with transition cost. The predetermined
order is then used to commit the units such that all operating constraints are satisfied.
Table 5. Result for test system with priority list
Hour j Sij Pj MW OCj
i=1 i=2 i=3 i=4
01 0 1 1 0 550 09966
02 0 1 1 0 550 09738
03 0 1 1 1 610 11166
04 0 1 1 1 610 11166
05 0 1 1 0 550 09738
06 0 0 1 0 300 05238
07 0 0 1 0 300 05238
08 0 1 1 0 550 09966
Cumulative cost in Indian Rupees (INR), OC 72216

471
8. Cost Comparison
Table 6. Cost comparison of GA and priority list based
approach for UCP.
GA Priority List
74411 INR 72216 INR
It is to note that GA and priority list based method uses power output limits at their
maximum value for selected generating units. Fuzzy logic based approach to UCP operates
each generating unit within its minimum and maximum power operating limits and resulted the
operating cost in INR as 74683.6.
9. Conclusion
Feasibility of applying Genetic algorithm, Fuzzy Logic and Priority List Method in solving
short-term thermal unit commitment is proved. These methods guarantees the production of
solution that do not violet system or unit constraints; so long as there are generators available in
the selection pool to meet the require load demand. Though the global optimality is desirable,
but in most practical cases near optimal solutions is generally sufficient. This paper attempts to
find the best schedule from a set of good feasible commitment decisions. The result shows that
it is possible to achieve improvements using these methods.
Fuzzy logic method allows a qualitative description of the behavior of a system, the system’s
characteristics and response without the need for exact mathematical formulations. Genetic
algorithm and priority list method usually suffer from large computational times and excessive
memory requirements as the problem size increases. The result shows that fuzzy logic method
is a powerful tool that can be used to solve short-term thermal unit commitment problem. An
advantage of fuzzy logic lies in its ability to handle any type of unit characteristics data.

10. References
[1] A. J. Wood and B. F. Wollenberg, ‘Power generation operation and control’, John Wiley and sons (Asia), second
edition, 2006.
[2] A. Perlin and P. Sandrin, ‘A new method for unit commitment at Electricite De France’, IEEE Transaction on
Power apparatus and system, Vol. PAS-102, No. 5, May 1983, pp 1218-1225.
[3] B. Gerald, Sheble and George N. Fahd, ‘Unit Commitment Literature Synopsis’ IEEE Transactions on Power
System, Vol.9, No.1, Feb 1993, pp 128-133.
[4] Chuan-Ping Cheng, Chih-Wen Liu & Chun-Chang Liu, ‘Unit commitment by Lagrangian relaxation & Genetic
Algorithms’, IEEE Transaction on Power system, Vol.15, No.2, May 2000, pp 707-714.
[5] D. Dasgupta, and D.R. McGregor, ‘Thermal Unit Commitment Using Genetic Algorithms’ IEEE proceedings on
Generation, transmission and distribution, vol 141, no 5, 1994, pp 459-465.
[6] David E. Goldberg, ‘Genetic algorithms in search, optimization and machine learning’, Pearson Education Asia,
2001.
[7] Dipankar Dasgupta, ‘Unit Commitment in Thermal Power Generation Using Genetic Algorithm’, In proceeding
of the Sixth International Conference on Industrial and Engineering application of Artificial Intelligence and expert
system (IEA/AIE-98), Edinburgh, UK, June 1-4-1993, pp 374-383.
[8] S. Baskar, Dr. P. Subbaraj & (Ms) P. Chidambaram, ‘Application of Genetic Algorithm to unit commitment
problem’, Journal of Institute of Engineering (India), Vol 81, March 2001, pp 195-199.
[9] S. Saneifard, N. R. Prasad, H. A. Smolleck, ‘A Fuzzy Logic Approach to Unit Commitment’, IEEE Transactions
on Power System, Vol.12, No.2, May 1997, pp 988-995.

472