Optimal Solution for Non-Convex Combined Heat and Power Dispatch Problems
using Differential Evolution
Nidul Sinha1, Senior Member IEEE,
ABSTRACT:
This paper investigates the performance of Differential Evolution
(DE) for solving combined heat and power dispatch (CHPD)
problems in power systems. Solution method based on Different
Evolution is developed and its performance is tested on a test case of
combined heat and power dispatch problem. The simulation results
demonstrate that the performance of DE based algorithm is far better
than the better known floating point Genetic Algorithm (FPGA) in
terms of success rate, convergence rate and solution quality.
Keywords: Genetic Algorithm, Differential Evolution, Economic
Load Dispatch, Combined Heat and Power Dispatch.
I. INTRODUCTION
With sharp rise in energy demand and resulting increased
pollution the issues of energy conservation and green power
gained much attention in 21st century. Cogeneration
technology proves to be promising with its greater conversion
efficiencies than traditional generation methods as it
harnesses heat that would otherwise be wasted. This can
result in up to more than a doubling of thermal efficiency or
higher heat values. Also, carbon dioxide emissions can be
substantially reduced. Furthermore, the heat by-product is
available for use without the need for the further burning of a
primary fuel. Cogeneration or combined heat and power
(CHP) is the simultaneous production of electricity and heat
using a single fuel such as natural gas, although a variety of
fuels can be used (refer to 'Cogeneration capacity by primary
fuel'). CHP or Cogeneration systems usually consist of more
than one gas turbine exhausting into a single waste heat
boiler, which recovers the exhaust heat by producing hot
water or steam for process and space heating. The CHP
dispatch (CHPD) problem is more complex than the
conventional ED problems.
Few works have been reported [1] [2] in the Combined Heat
and Power Economic dispatch area. In most of the reported
works on CHP, the fuel cost functions are considered to be of
quadratic nature. With the consideration of highly non-linear
cost curves because of incorporation non-linear features like
valve point loading, prohibited operating zones, combined
cycle units, the problem becomes a non-convex one and
reported methods fail to solve this type of problems. Solution
of these types of problems calls for modern heuristic search
methods which are very efficient to tackle highly non-linear
con-vex type optimization problems.
A lot of developments have taken place in the development of
heuristic search methods for solving economic dispatch
problems. Some of the best known are genetic algorithm
(GA) [3]-[5], evolutionary programming (EP) [7]-[10],
particle swarm optimization (PSO) [11], and differential
1
Authors are with the Department of Electrical Engineering, NIT, Silchar,
Assam, India-788010. Corresponding author: Nidul SInha:
(email:nidul@ieee.org).
Lalit Ch. Saikia1, Tanmoy Malakar1,
evolution (DE) [12],[13]. DE is reported [12][13] to be the
only algorithm, which consistently found the optimal
solution, and often with fewer function evaluations than the
other direct search methods on benchmark nonlinear
functions.
The highlights of Differential Evolution (DE) are:
• A population of solution vectors are successively updated
by addition, subtraction, and component swapping, until the
population converges, hopefully to the optimum.
• No derivatives are used.
• Very few parameters to set.
• A simple and apparently very reliable method.
However, little work has been reported in the literature on the
application of EA to the CHPD problem. Also, hardly any
work on CHPD with non-convex functions especially with
DE is reported in the literature.
The reported impressive performance of DE algorithm has
inspired the use of DE algorithm for solving the CHP
problem with non-convex fuel cost functions.
In view of the above, the main objectives of the present work
are:
(i) To develop a program based on DE algorithm for the
CHP problem.
(ii) To solve the same problem with best features of
floating point GA (FPGA) algorithm as reported in
[16].
(iii) To compare the performance of DE algorithm as
compared to FPGA.
In this work a DE based algorithm has been developed for
solving the CHPD problem considering quadratic function
together with valve point loadings for the electrical power
generating units. The CHPD problem is first formulated
mathematically. The GA-based algorithm is then described.
In the algorithm, methods for satisfying the heat and power
operation ranges of the units in the cogeneration systems and
the demand constraints are presented. The new algorithm is
used for solving a test system [4] containing two
cogeneration units, a boiler and a conventional generator. The
results are presented and the performance of the algorithms
assessed.
The rest of the paper is organized as follows: In section II the
mathematical problem formulation is briefly reviewed. DE
approach and its different features are described in section III.
Section IV presents the numerical test results on a test case.
Conclusions are drawn in section V.
 II. PROBLEM FORMULATION the minimum and maximum heat production of generator
In this section the optimization problem is formulated as a i. p i ,min ( hi ) , p i , m ax ( hi ) , hi ,min ( pi ) and hi ,max ( pi ) are
minimization of summation of fuel costs of thermal units,
cogeneration units and boiler units. the linear inequalities that define the FOR of the
cogeneration units. Fig.1 shows the FOR of a cogeneration
2.1 Mathematical Problem Formulation unit.
Mathematically, the problem is to minimize the following:
The objective function of the CHPD problem, which is to be
minimized, is:
α β n
F ( pG ) = ∑ Fti ( pi ) + ∑ F ci ( pi , hi ) + ∑β F bi ( hi ) (1)
i =1 α
i = +1 i = +1

where Fti is the fuel cost of conventional thermal unit, Fci is

the fuel cost of cogeneration unit and Fbi is the fuel cost of
boiler unit.
The power and heat demand constraints of the CHPD
problem can be written as:
α β

∑
i =1
pi + ∑α
i= +1
p i = p int + p util + L (2)

β n

∑
α
h + ∑h
i = +1 β
i
i = +1
i =H (3) Fig.1 FOR of a cogeneration unit.

III. Differential Evolution (DE)

where Pint is the on-site power demand (MWe), Putil is the The DE algorithm is a population based algorithm like GAs
host utilities power demand (MWe), L is the active power using the similar operators; crossover, mutation and selection.
loss in transmission host network (MWe), and H is the heat The main difference between GA and DE is that GAs rely
demand (MWth) . The operation ranges of boilers and mostly on crossover while DE relies on mutation operation.
conventional The algorithm uses mutation operation as a search
mechanism and selection operation to direct the search
The operation ranges of conventional thermal units and toward the prospective regions in the search space. Mutation
boilers are expressed as in DE uses differences of randomly sampled pairs of
solutions in the population and greediness may be embedded
in it. For detail theory [12] may be referred.
Let xi =[X1, X2, …..Xn] be a trial vector denoting the
p i ,m in ≤ p i ≤ p i ,m a x individual of a population to be evolved. The elements of the
i ∈ 1 , 2 , . . .α (4) xi are the real power outputs and heat powers of the
committed n co-generating units which are subjected to their
respective capacity constraints. To satisfy the equality
constraint for electrical power, a unit (conventional unit) and
hi ,min ≤ h j ≤ hi ,max
(5) for heat power balance, a boiler unit (only thermal) are
i ∈ β + 1, β + 2,..., n considered. Let Xd be the power output of the dependent unit,
then Xd is calculated by
n
The operation ranges for the cogeneration units, the heat and
X d = Dp − ∑ X j (8)
power are given by equation (6) and (7): j =1
≠d

p i ,m in ( hi ) ≤ p i ≤ p i ,m ax ( hi ) Where Dp is the power demand.

i ∈ α + 1, α + 2, ..., β
hi ,min ( pi ) ≤ hi ≤ hi ,max( pi )
i ∈ α + 1, α + 2,...β
where pi ,min and
power generation of generator i .
In this work the power loss is ignored.
(6)
3.1.1 Initialization:
An initial population of Np individuals having the form:
XiG = [X1,i,G, X2,i,G, …….Xn,i,G], i = 1, 2, …Np is generated.
(7)
Where Np, G, and n are the number of individuals, generation
number and committed cogeneration units respectively. Each
individual is determined by setting the jth component Pj ~
p i ,max are the minimum and maximum U(Xjmin, Xjmax), for j=1,2,…n. U(Xjmin, Xjmax) denotes a
uniform random variable ranging over [Xjmin, Xjmax].
hi ,min and hi ,m ax are
3.1.2 Creation of offspring
New solutions (offspring) are created based on both
mutation and crossover as given below:
3.1.2.1 by mutation
Though there are number of variant mutation schemes in
DE, the most promising one is considered here. In each
generation, for individual pi,G, the donor vector vi is obtained
using the mutation operation through
vi = xi,G + λ.(xbest,G – xi,G) + F.(xr2,G – xr3,G)
Where the mutation factor F is a constant from [0, 2] and the
vectors xr2,G, and xr3,G are selected randomly such that the
indices i, r2 and r3 are distinct from (1, 2, … Np).
The idea behind the introduction of the additional control
variable λ is to provide a means to enhance the greediness of
the scheme by incorporating the current best vector pbest,G.
3.1.2.2 by crossover
Secondly, the trial vector ui is obtained by crossover
through
 v j ,i ,G +1 if rand j ,i ≤ C R or j = I rand
u j , i , G +1 = 
 X j , i , G +1 if rand j , i > C R and j ≠ I rand
contains one conventional electricity unit, one boiler unit
and two cogeneration units. The operating range and cost
characteristics of the conventional units are given in eqs.
(12) and (15) respectively.
A program based on DE is developed in Matlab command
line for solving the problem. The best reported [22] floating
point GA (FPGA) program with heuristic crossover and
multi-non-uniform mutation using the GA toolbox, GAOT, in
Matlab as proposed by C.R. Houck, J. Joines, and M. Kay [8]
(9) is used after minor modification for solving the problem for
comparing the performance of the DE method. The programs
were run on a 2.67 GHz, Pentium, with 2 GB RAM PC.
The feasible operating region (FOR) diagrams of the
cogeneration units are given in Fig. 2 and Fig.3. And the
cost characteristics of cogeneration units are given in eqs.
(13) and (14) respectively. The power and heat demands of
the test case are 200 MWe and 115 MWth, respectively.
(10)

i= 1, 2, …..Np, j= 1, 2, …..n
randj,i ∼U[0,1], Iraand is a random integer from [1, 2, …. n], CR
is user defined number between [0,1] which is called crossing
factor.

3.1.3 Selection
All solutions in the population have the same chance of
being selected as parents without dependence of their fitness
values. The child produced, after the mutation and crossover
operations, is evaluated. Then, the performance of the child
vector and its parent is compared and the better one is
selected. If the parent is still better, it is retained in the
population. The target vector, xi,G+1 is selected after Fig.2 FOR of cogeneration unit-1
comparison of parents xi,G with their offspring ui,G+1 as given
below:

u i ,G +1 if f (u i ,G +1 ) ≤ f ( xi ,G ) (11)
X i ,G +1 =  i = 1, 2,...N p
 xi ,G otherwise

Mutation, crossing and selection continue until the stopping

criterion is satisfied.

3.1.4 Stopping Rule

The iterative procedure of generating new trials by
selecting those with minimum function values from the
competing pool is terminated when there is no significant
improvement in the solution. It can also be terminated when a
given maximum number of generations (iterations) are
reached. In the present work the latter method is employed.
Fig. 3 FOR of cogeneration unit-2.
IV Numerical Test Results and Discussion:
The performance of the proposed algorithms are
tested on a test case as in Ref. [21] but modified to include
valve point loading effects. The cogeneration system
 FC1 ( p1 ) = 1055.1 + 3.33 p1 + 0.5214( p12 + Table 1: Results of the different proposed algorithms with
(12) 50 different initial trial populations on the test case.
abs (120.0 sin( 0.077(0.0 − p1 )))) Cost in Rs.
0<p1 < 150 Method
Average Maximum Minimum
2
FC2 ( p2, h2 ) = 2650+14.5 p2 + 0.0345p + 4.2h2 +
2 FPGA 10935.24 11253.67 10842.32
(13) DE 10614.43 11226.17 10203.67
2
0.03h + 0.031p2h2
2

FC3 ( p3 , h3 ) = 1250 + 34.5 p3 + 0.0435 p32 + Table 2: Solution times of the algorithms in solving the test
(14) case.
2
0.6h + 0.011 p3h3
3 Method Solution time
FPGA 30.8 msec.
F c 4 ( h 4 ) = 2 3 .4 h 4 (15) DE 5.17 msec.
0 ≤ h 4 ≤ 2 6 9 5 .2

Tuned parameters of FPGA are: V. CONCLUSION:

Maximum number of generation=100 Matlab program based on DE algorithm is developed for
Population size = 20 solving the CHPD problems. The best Floating point GA
Penalty factor = 1000 program as reported in [22] using modified GAOT toolbox is
Crossover probability = 0.9 devbeloped to solve the same test problem. The performance
mutation probabilities = 0.01 of the developed programs is tested on a test case of CHPD
Addl. tuned parameters for DE are: problem in power system. It has been observed that the DE
Mutation factor, F = 0.4 based algorithm outperforms the best FPGA algorithm in
Crossing factor, CR = 0.7 finding faster and better quality solutions for the test case.
Greediness, λ = 0.6 The success rate, convergence rate and quality of solutions
achieved with DE algorithm is far better than that with FPGA
algorithm. Hence, DE algorithm may prove to be a very
promising solution technique for highly non-convex CHPD
problems in power system.

Fig. 4 C0onvergence nature of the proposed DE algorithm
and FPGA on the test case.
Fig. 4 shows the convergence natures of different proposed
DE algorithm and best FPGA as reported in [22] in solving
the CHPD problem. It can easily be observed that observed
that DE algorithm converges at a faster rate as compared to
the best reported FPGA algorithm.
To investigate the effects of initial trial solutions all the
algorithms were run with 50 different initial trial solutions.
The results are reported in Table 1. Table 2 shows the
solution times of the algorithms in solving the test case.
Investigation of the results in both the tables shows superior
performance of the DE algorithm as compared to FPGA.
VI REFERENCES
1. F.J. Rooijers, R.A.M. Van Amerongen, “Static economic dispatch for
co-generation systems”, IEEE Trans. on Power Systems, Vol. 9, No.3,
1994, pp.1392-1398.
2. Tao Guo, M. I. Henwood, M. Van Ooijen, “An algorithm for heat and
power dispatch”, IEEE Trans. on Power Systems, Vol. 11, No. 4, 1996,
pp.1778-1784.
3. Wong, K.P. and Wong, Y.W., “Thermal generator scheduling using
hybrid genetic/simulated annealing approach,” IEE Proc. Part-C,
Vol.142, No.4, July 1995, pp.372-380.
4. Walter, D.C. and Sheble, G.B., “Genetic algorithm solution of
economic dispatch with valve point loading,” IEEE Trans. on Power
Systems, Vol.8, No.3, August, 1993, pp.1325-1332.
5. Bakirtzis, A. Petridis, V. and Kazarlis, S., “Genetic algorithm solution
to the economic dispatch problem,” IEE Proc. Part-D, Vol. 141, No. 4,
July 1994, pp. 377–382.
6. Sheble, G.B. and Brittig, K., “Refined genetic algorithm — Economic
dispatch example,” IEEE Trans. on. Power Systems, Vol. 10, Feb.
1995, pp. 117–124.
7. Chen, P.H. and Chang, H.C., “Large-scale economic dispatch by
genetic algorithm,” IEEE Trans. on Power Systems, Vol.10, No.4,
November 1995, pp.1919-1926.
8. Houck, C.R., Joines, J.A., and Kay, M.G., “A genetic algorithm for
function optimization: A Matlab implementation,” Technical Report
NCSU-IE TR 95-09, North Carolina State University, 1995.
9. Bäck, Th. and Schwefel, H.P., “An overview of evolutionary
algorithms for parameter optimization,” Evolutionary Computation,
Vol.1, No.1, 1993, pp.1-23.
10. Yang, H.T., Yang, P.C. and Huang, C.L., “Evolutionary programming
based economic dispatch for units with non-smooth fuel cost
functions,” IEEE Trans. on Power Systems, Vol.11, No.1, February
1996, pp.112-118.
11. Yao, X. , Liu, Y. and Lin, G., “Evolutionary programming made
faster,” IEEE Trans. Evolutionary Computation, Vol. 3, July 1999, pp.
82–102.
12. Sinha, Nidul, Chakrabarti, R. and Chattopadhyay, P.K., “Evolutionary
programming techniques for economic load dispatch,” IEEE Trans. On
Evolutionary Computation, Vol.7, No.1, February, 2003, pp.83-94.
13. Sinha, Nidul, Chakrabarti, R. and Chattopadhyay, P.K., “Fast
Evolutionary programming techniques for short-term hydrothermal
scheduling,” IEEE Trans. on Power Systems, Vol.18, No.1, February
2003, pp.214-220.
14. Kennedy, J., and Eberhart, R., “Particle Swarm Optimization,”
Proceedings of IEEE International Conference on Neural Networks,
Vol. IV, Perth, Australia, 1995, pp. 1942-1948.
15. Yoshida, H., Kawata, K., Fukuyama, Y., Takayama, S., and Nakanishi,
Y., “A particle swarm optimization for reactive power and voltage
control considering voltage security assessment,” IEEE Trans. On
Power Systems, Vol. 15, Nov. 2000, pp. 1232–1239.
16. Victory, T. A. A. and Jeyakumar, A. E., “Hybrid PSO-SQP for
economic dispatch with valve-point effect,” Elect. Power Syst. Res.,
Vol. 71, 2004, pp. 51-59.
17. Park, J.-B., Lee, K.-S., Shin, J.-R., and Lee, K.Y., "A particle swarm
optimization for economic dispatch with non-smooth cost Functions,”
IEEE Trans. on Power Systems, Vol. 20, No. 1, February 2005, pp. 34-
42.
18. Storn, R., “System design by constraint adaptation and differential
evolution,” IEEE Trans. on Evolutionary Computation, Vol. 3, No. 1,
1999, pp. 22-34.
19. Storn, R. and Price K., “Minimizing the real functions of the ICEC’96
contest by differential evolution,” Int. Conf. Evolutionary Computation,
IEEE, 1996, pp.842-844.
20. Song, Y. H., Xuan, Q.Y, “Combined heat and power economic dispatch
using genetic algorithm based penalty function method”, Electric
Machines and Power Systems, 26, No. 4 (1998).
21. Guo, T., Henwood, M.I., Ooijen, M., “An algorithm for combined heat
and power economic dispatch”, IEEE Trans. on Power Systems, Vol.
11, No. 4, November 1996, pp.192–195.
22. N. Sinha and T. Bhattacharya, “Genetic Algorithms for Combined Heat
and Power Dispatch Problems” Proceedings of IEEE int. conference
TENCON 2008, Hyderabad Univrsity, Hyderabad.