ISSN (Print) : 0974-6846

Indian Journal of Science and Technology, Vol 7(12), 2056–2063, December 2014 ISSN (Online) : 0974-5645

Unit Commitment in Power System by Combination

of Dynamic Programming, Genetic Algorithm and
Particle Swarm Optimization
Fazlollah Rouhi1 and Reza Effatnejad2*
Department of Electrical Engineering, Alborz Branch-University of Science and Research,
1

Islamic Azad University-Alborz-Iran; fazlolaruhi@yahoo.com

2
Electrical Engineering Department, Karaj Branch, Islamic Azad University, Alborz, Iran; Reza.efatnejad@kiau.ac.ir

Abstract
In this paper, a hybrid method is proposed for dynamic programming, genetic algorithm and particle swarm optimization
in order to solve the unit commitment of power systems considering system security constraints. The condition of the units
to be on or off and economic dispatch are determined and solved by considering system security constraints through previ-
ously mentioned combination method.
The simulation result is shown that hybrid system is a good methodology for unit commitment. The result is evaluated for
experimental case study considering all constraint in power system; the results obtained are acceptable.

Keywords: Combinatorial Solution Strategy, Dynamic Programming (DP), Genetic Algorithm (GA), SCUC Security
Constrained Unit Commitment, Particle Swarm Optimization (PSO)

1.  Introduction considered. Generation configuration is considered in

power system and security constraints are determined
Unit commitment (UC) in power system includes for UC operation function. Security constrained unit
­determination of planning the units to be on and off in ­commitment (SCUC) is evaluated for example limitation
order to supply the forecasted load in a period of time. of transmission line1.
Constraints such as load balance, system spinal reserve, The goal of SCUC is generation configuration plan-
unit generation limit, pollution etc. are included in this ning with the lowest cost and reliability2.
planning schedule can minimize the cost of power system Large modern units are equipped with multi-valves
operation. turbines and the unit output would have a waveform
There are two processes ahead in solving the unit curve which is called valve point effect3. By considering
commitment problem. First to determine the units to be this form of output the problem will be so complicated.
on or off. The numbers “1” and “0” represent the condi- The complexity of this problem makes us to get
tion of the units whether is on or off respectively. In the through dimensions and limitations of this prob-
other process the economic dispatch should be analyzed, lem and considering proposed a new method for
in the other words, how much power should be allocated ­optimization.
to each unit from whole generated power over a period of Thus, efforts are to be made to supply the load demand
time when the unit is on. of costumers in an economic and secure way.
But solving the UC problem alone will not solve the In this paper we attempt to present a suitable and
problem, since the network security constraints are not economic method to determine the unit commitment

*Author for correspondence

 Fazlollah Rouhi and Reza Effatnejad

arrangement considering system security constraints optimum location is found, the particle will move to that
which may lead to minimizing the cost and improving the location, otherwise the movement won’t be deviated.
security level of the network. Therefore the convergence velocity will remain standstill.
It is possible to use equation below to apply mutation
operator on particles6, 7.
2. Proposed Algorithm to Solve
the SCUC Problem ∆Χ = (varhi – Xi) ∗ (1 – rand(1 – (iter/maxiter))5)
– (Xi – varlo) ∗ (1 – rand(1 – (iter/maxiter))5)
In order to solve the SCUC problem using the combi-
nation of binary PSO algorithm and genetic algorithm
Where:
which are included with dynamic programming (DP)1, 4.
First the condition of the units to be on or off should ∆Χ: C hange in the position of members which
be determined one by one and then each unit genera- are mutated.
tion will be defined. Sometimes the PSO optimization varh: Maximum value of X
algorithm involves local optimization and would not varlo: Minimum value of X
be able to define the global optimization, by using the rand: R andom number generation function which
mutation and cutting operators of genetic algorithm produce numbers between 0 and 1 with
it would be possible to get close to local optimization. monotonous probability distribution.
Using these operators the convergence of data exchange iter: Loop Counter of algorithm.
is done better and response is searched to find the global maxiter: Maximum number of algorithm loop repeat.
optimization. Therefore, these operators improve the
functionality of PSO algorithm in getting away from The created mutation using the above equation causes
local optimal point5. some members to be out of the specified range of search.
In PSO algorithm, if all the group members ­considered So the locations which are out of the range of search
as neighboring of a single member then the best ­member space are investigated randomly. It should be noted that
of the group will be always the leader of the other the memory of PSO algorithm causes the particle swarm
­members. movement not to deviated in the case of the locations
Referring to the algorithm motion law and by which are obtained from mutation are not optimum.
­considering the equation pbest = gbest = xi for the best So by using mutation operator, convergence velocity
­member of the group, it can be understand that pbest = won’t decrease.
gbest has no leader and is moved based only on its velocity
The proposed algorithm is as follows:
­vector.
The algorithm is shown that suitable result.   1 - Form the initial population.
In this section, the operation of PSO algorithm is   2 - Evaluation of each particle current position.
improved using genetic algorithm operators “crossover”  3-D  etermine the best location ever experienced by any
and “mutation”. In the proposed method the crossover particle.
operator is used to transfer data between two optimal  4-D  etermine the best position experienced by the group
particles; and mutation operator is used to escape from of particles up to now.
the local optimum. For every iteration by choosing two   5 - Set the velocity and new location of each particle.
particles as parents randomly and using the crossover  6-A  pply crossover operators to the particles and adding
operator, a new member is added to set. By using the a new member.
mutation operator, fraction of particles is thrown to ran-   7 - Apply mutation operator to the percentage of ­particles
dom locations in order to replace the pervious “gbest” in randomly.
case of finding better points. Considering the fact that   8 - Form a new population.
particles do have memory, the point in using mutation  9-C  heck the stopping criterion for the new population,
and crossover operators in proposed algorithm, is that, go to step 10 if the criterion is met, go to step 2 if it is
the convergence velocity of PSO algorithm will remain at failed to fulfill the criteria.
its pervious value and if by using these operators a more 10 - Show new population.

Vol 7 (12) | December 2014 | www.indjst.org Indian Journal of Science and Technology 2057
 Unit Commitment in Power System by Combination of Dynamic Programming, Genetic Algorithm and Particle Swarm Optimization

3. Applying the Proposed t +1

Ti,on
t
+1 I =1 
Ti,on = t 
Algorithm to Solve the Ti,on I = 0

Problem” SCUC”
This section is to implement and apply the proposed
Where “I = 1” represents the unit is turned on and
algorithm. An IEEE 30 buses system, in which two hydro-
“I = 0” means it’s off and these values are obtained from
electric units are added to its buses number 15 and 20, is
“I” matrix and “Tit,on” is duration of time in which the
considered and will have the following steps 8, 9:
unit is turned on until the time period “T” of planning
schedule. Also, it is possible to define an initial activity
3.1  First Step duration for each unit which means how many times each
In this step the initial population is set. Since the power unit have been on at the beginning of planning schedule
system is consists of N generation units and also the and the duration of this condition can be shown by a posi-
time period which is considered in this study is a season tive number and the off condition is shown by a negative
which every month of it is divided into sections. And for number.
each section, three levels of load are considered includes
minimum, average and maximum. 18 time intervals are 3.3  Third Step
considered in planning so every group of population
After determining the initial population and satisfying
can be shown using a N ∗ 18 matrixes which is specified
the mentioned constraints listed in the previous step for
through following equation:
each of the categories, now, the production costs should
be determined for each category of the units status (on
 I11 I12  I1T  or off). To determine the production cost the “economic
 
 I 21 I 22  I 2T  dispatch” should be solved considering the units which
I =    are on and in order to do so the following steps should be
  taken into account:
  
 I I 
 N ,1 N ,2 I N ,T 
1 - Generate the initial population.
2 - Supply power generation limit.
Where “T” represents the time intervals in the sched- 3 - Supplying “supply and demand balance” constraints.
ule and here is equal to 18. Also “Ix, y” is the position of 4 - Supplying maximum allowable pollution constraints.
the “x” unit in the time period of “y”. It should be noted 5 - Supplying generation limit in hydropower units.
that each element of matrix “I” is determined randomly 6 - Supplying transmission lines limit constraint.
and is equal to 1 or 0 and the previously mentioned matrix 7-D  etermine the value of each group, optimal level of
is only related to one group of population and in order to personal and global (pbest and gbest).
produce the whole population, several matrix should be 8 - S et the stop criterion for the economic dispatch
created like matrix “I” equal to number of existing groups ­algorithm.
in population.
There are various ways to determine the stop time
of the algorithm. But one of the most common ways is
3.2  Second Step
­considering a certain number of iterations. In this paper
After the initial population is generated spinning 1000 iterations is considered.
reserve constraints and periodic inspections and over-
hauls are checked and if any of these two constraints
are not meet various methods can be used to satisfy 3.4  Fourth Step
these constraints. In this study, a method based on After determining the generation cost for each of the
the “priority list” of the units is used to satisfy these categories, in this section, the best answer should be
­constraints. determined so it can be used for next steps. Therefore, the

2058 Vol 7 (12) | December 2014 | www.indjst.org Indian Journal of Science and Technology
 Fazlollah Rouhi and Reza Effatnejad

Table 1.  Proposed load pattern for test system

load Level load Level
Section mount Section mount
maximum Average minimum maximum Average minimum
283.14 198.2 133.81 1 242.88 170.02 111.72 1
7 1
277.48 194.24 131.14 2 240.55 168.39 110.655 2
246.77 172.74 126.08 1 240.137 168.01 127.47 1
8 2
248.66 174.06 127.04 2 243.26 170.28 129.19 2
251.47 176.03 138.31 1 292.26 240.58 1131.81 1
9 3
251.1 175.76 138.10 2 351.37 245.96 158.46 2
250.01 175.01 116.71 1 425.15 297.61 161.56 1
10 4
249.74 174.82 116.58 2 402.73 281.91 153.04 2
249.20 174.44 140.78 1 389.21 272.45 146.30 1
11 5
251.37 175.96 141.10 2 324.34 227.04 121.92 2
255.28 178.70 133.47 1 294.34 206.03 140.87 1
12 6
248.39 173.87 129.86 2 275.69 192.98 9131.95 2

best answer for each category and group is ­determined.

It should be noted, definition for best response is to
satisfy all the constraints and do have the minimum
­generation cost.

3.5  Fifth Step

New population is determined according to the best
obtained from the previous steps and the proposed
H-DP-PSO-GA hybrid algorithm.

3.6  Six Step

The algorithm introduced in the previous steps is applied
on the proposed test system and the result of the simula-
tion will be analyzed.

4.  Determining Load Pattern

Load pattern for a period of one year is specified in the
following table and graph. In this study, each Season is
divided to three Months and each month are divided into
two parts. And three load levels (minimum, average and
maximum) are considered for each part, so, 18 load levels
Figure 1.  IEEE 30 Buses Test System. is d
­ etermined for each season.

Vol 7 (12) | December 2014 | www.indjst.org Indian Journal of Science and Technology 2059
 Unit Commitment in Power System by Combination of Dynamic Programming, Genetic Algorithm and Particle Swarm Optimization

Figure 2.  Flowchart of the proposed algorithm.

Since each month is divided into two parts in the the steam valve in order to be close to the real model, also
­ roposed test system so, the horizontal axis consists of
p the amount of water in the reservoir of dam is consid-
24 portions instead of 12 in the figure below. ered as a constraint for the hydroelectric unit, therefore,
these units do have limitations to produce electric energy.
5. Simulation and Analysis of The spinning reserve constraint is also considered in this
scenario. Also, in addition to security constraint, ­periodic
Simulation Results controls and overhauls of the units are considered as other
In order to demonstrate the capabilities of the proposed constraints of the system.
algorithm, the simulation is carried out as follows: Periodic controls and overhauls constraints mean
System contains hydro and thermal units. Thermal that every single unit is removed from the network after
units in the system are equipped with the effect of ­opening specified period of time to be controlled to see whether

2060 Vol 7 (12) | December 2014 | www.indjst.org Indian Journal of Science and Technology
 Fazlollah Rouhi and Reza Effatnejad

it needs any overhaul and maintenance process or not. so it is clear that the proposed algorithm is more capable
The activity time is considered 18 intervals for hydroelec- than PSO algorithm reaching optimal value. There would
tric units and 15 generators for thermal units. Security not any significant change after about 200 iterations by
constraint put a limit on the power transmission lines in using PSO algorithm but the proposed algorithm reach
a way that if the power which is carried through the lines the cost 12447 $ after about 200 iterations which is less
exceeded from an acceptable value then the answer will than the corresponding number of iterations in PSO
not be accepted. In order to consider these constraints algorithm and by increasing the number of iterations the
lines 7, 12, and 20 are selected. The maximum ­permissible proposed algorithm will reach better results values and
power that is considered to cross the lines is 260, 100 and that is because of the algorithm does not trapped by local
80 MW respectively. Tables 2 and 3 represent the results optimal which is caused by the mutations of genetic algo-
of implementing the proposed algorithm on the test rithm and that is why the proposed algorithm is more
­system. And the total cost is equal to 11,992 $. In table capable.
2, “0” and “1”, represent whether the unit is turned on or
off, ­respectively.
To illustrate the convergence of the proposed 450
­algorithm, the convergence of the algorithm is shown Min
in Figure 4, which shows a proper convergence of the 400
Ave
Max
­algorithm.
350

6. Comparison Between Proposed 300

Method and the PSO Algorithm
250
If the PSO algorithm is applied to the system output
will be most likely to what is shown in Figure 5. If the 200
­proposed algorithm (DP and PSO and GA) is applied the
result would be similar to what is shown in figure 4. AS 150
it is obvious the proposed algorithm is more efficient and
also the cost is much lower Compare to PSO algorithm 100
0 5 10 15 20 25
alone. By using PSO algorithm the cost is 14581 $ and if
the proposed algorithm is used the cost would be 11992 $ Figure 3.  Model of proposed load test system.

Table 2.  Results of the simulation

Period 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

Unit 1 1 1 1 1 1 1 1 1 0 1 1 1 1 1 1 1 1 1

Unit 2 0 1 1 1 1 1 1 1 1 1 1 1 0 1 1 1 1 1

Unit 3 1 1 1 1 1 1 1 0 1 1 1 1 1 1 1 1 1 1

Unit 4 0 1 1 1 1 1 1 1 1 1 1 1 0 1 1 1 1 1

Unit 5 0 1 1 1 1 1 1 1 1 1 1 1 1 0 1 1 1 1

Unit 6 1 1 1 1 1 1 1 1 1 0 1 1 1 1 1 1 1 1

Unit 7 1 1 1 1 1 1 1 1 0 1 1 1 1 1 1 1 1 1

Unit 8 1 1 1 1 1 1 1 1 0 1 1 1 1 1 1 1 1 1

Vol 7 (12) | December 2014 | www.indjst.org Indian Journal of Science and Technology 2061
 Unit Commitment in Power System by Combination of Dynamic Programming, Genetic Algorithm and Particle Swarm Optimization

Table 3.  The value of each unit of energy produced

Period 1 2 3 4 5 6 7 8 9

Unit 1 13.5 13.5 13.5 13.5 13.5 13.5 11.5 11.5 0

Unit 2 0 13.5 13.5 13.5 13.5 13.5 13.5 13.5 13.5

Unit 3 5 5 5 5 5 20.218 5 0 50

Unit 4 0 35.14 16.82 5 5 5 5 5 57.54

Unit 5 0 49.56 5 5 5 98.23 5 72.7 27.42

Unit 6 88.21 48.30 36.20 64.58 73.13 89.30 90.77 61.74 110.7

Unit 7 5 5 100 5 5 5 5 5 0

Unit 8 5 5 60 5 54.69 5 5 5 0

More Table 3.  Amount of each unit of energy produced

Period 10 11 12 13 14 15 16 17 18

Unit 1 11.5 11.5 11.5 11.5 11.5 11.5 11.5 11.5 11.5

Unit 2 13.5 13.5 13.5 0 11.5 11.5 11.5 11.5 11.5

Unit 3 5 5 5 5 50 50 5 5 5

Unit 4 60 5 5 0 60 5 60 60 5

Unit 5 1169 100 100 100 0 100 5 5 100

Unit 6 0 30.96 51.37 6.96 35.70 67.28 26.86 70.87 5039

Unit 7 35.31 5 5 5 5 5 5 5 5

Unit 8 5 5 60 5 5 5 5 5 60

Figure 4.  Convergence curves of the proposed Algorithm

(H DP PSO GA). Figure 5.  Convergence curve of PSO algorithm.

2062 Vol 7 (12) | December 2014 | www.indjst.org Indian Journal of Science and Technology
 Fazlollah Rouhi and Reza Effatnejad

7.  Conclusion 3. Fu Y, Shahidehpour M, Li Z. Security-constrained unit

commitment with AC constraints. IEEE Trans Power Syst.
In our proposed method, SCUC problem has both binary 2005; 20(2):1001–13.
variables (on/off condition of the units in form of 1/0) and 4. Rong A, Hakonen H, Lahdelma R. A dynamic regroup-
real variables (real power generation) and the challenge in ing based sequential dynamic programming algorithm
solving such problems is to establish a logical connection for unit commitment of combined heat and power
between these two groups of variables. In this research the ­systems. Int J Energy Conversion & Management. 2009;
50:1108–15.
relationship between these two variable established using
5. Park JB, Lee KS, Shin JR, Lee KY. A particle swarm
PSO and binary algorithms together, then in order to
­optimization for economic dispatch with no smooth cost
prevent the PSO algorithm from involving in/ trapped in functions. IEEE Trans Power Syst. 2005; 20(1):34–42.
local optimal points combined it with GA algorithm. By 6. Padhy NP. Unit commitment- a bibliographical survey.
considering DP in solving process, repetitive calculation IEEE Trans Power Syst. 2004; 19(2).
is avoided and Optimal response is achieved. Solving the 7. Walters DC, Sheble GB. Genetic algorithm solution of
problem by using proposed algorithm resulted in more ­economic dispatch with valve point loading. IEEE Trans
optimal responses and supplying costumers economically Power Syst. 1993; 8(3):1325–32.
as it is shown in convergence curves. 8. Lin CE, Viviani GL. Hierarchical economic dispatch for
piecewise quadratic cost functions. IEEE Trans Power Syst.
PAS-103. 1984; 1170–5.
8.  References 9. Wood AJ, Wolenberg BF. Power Generation, Operation,
and Control. 2nd ed. NewYork: Wiley; 1996. p. 131–70.
1. Ma H, Shahidehpour M. Unit commitment with transmis-
sion security and voltage constraints. IEEE Trans Power
Syst. 1999; 14(2).
2. Shahidehpour M, Yamin H, Li Z. Market Operations in
Electric Power Systems: Forecasting, Scheduling and Risk
Management. New York: Wiley; 2002.

Vol 7 (12) | December 2014 | www.indjst.org Indian Journal of Science and Technology 2063