You are on page 1of 24

Journal of Ambient Intelligence and Humanized Computing (2021) 12:8561–8584

https://doi.org/10.1007/s12652-020-02589-5

ORIGINAL RESEARCH

An improved real‑coded genetic algorithm with random walk based


mutation for solving combined heat and power economic dispatch
A. Haghrah1 · M. A. Nekoui1 · M. Nazari‑Heris2   · B. Mohammadi‑ivatloo2,3

Received: 14 July 2019 / Accepted: 30 September 2020 / Published online: 15 October 2020
© Springer-Verlag GmbH Germany, part of Springer Nature 2020

Abstract
Combined heat and power economic dispatch (CHPED) is an energy management problem that minimizes the operation cost
of power and heat generation while a vast variety of operational constraints of the system should be met. The CHPED is a
complicated, non-convex and non-linear problem. In this study, a new real-coded genetic algorithm with random walk-based
mutation (RCGA-CRWM) is under study, which is effective in solving large-scale CHPED problem with minimum operation
cost. In the presented optimization method, a simple approach is introduced to combine the positive features of different
probabilistic distributions for the step size of random walk. Using the presented approach, while the genetic algorithm is
speeded up, the premature convergence is also avoided. After verifying the performance of the presented method on the
benchmark functions, two large-scale and two medium-scale case studies are used for determining the algorithm strength in
solving the CHPED problem. Despite the fact that the complexity of the CHPED rises dramatically by increasing its dimen-
sionality, the algorithm has solved the problems accurately. The application of RCGA-CRWM method improves the results of
the CHPED problem in terms of both operation cost and convergence speed in comparison with other optimization methods.

Keywords  Combined heat and power (CHP) · economic dispatch · real-coded genetic algorithm · random walk · complex
optimization problem

List of symbols 𝜆i and 𝜌i Coefficients of the valve-point influ-


i Index of the thermal units ence of the ith thermal plant.
j Index of the CHP units ak , bk , and ck Cost coefficients of the kth boiler
k Index of the boilers Ci Power production cost of the ith
aj , bj , cj , dj , and fj Cost coefficients of the jth CHP thermal unit
plant Cj Power production cost of the jth
𝛼i , 𝛽i , and 𝛾i Cost coefficients of the ith thermal CHP unit
plant Ck Power production cost of the kth
boiler
Pi Power generation of the ith conven-
p
* B. Mohammadi‑ivatloo
bmohammadi@tabrizu.ac.ir tional plant
A. Haghrah Pcj Power generation of the jth CHP
arslan.haghrah@email.kntu.ac.ir unit
M. A. Nekoui Hjc Heat generation of the jth CHP unit
manekoui@eetd.kntu.ac.ir Hkh Heat generation of the kth boiler
M. Nazari‑Heris Pd Power demand
m.nazari@ieee.org Hd Heat demand
Pi Minimum power generation of the
pmin
1
Faculty of Electrical Engineering, K. N. Toosi University
of Technology, Tehran, Iran ith conventional power unit
Pi Maximum power generation of the
pmax
2
Faculty of Electrical and Computer Engineering, University
of Tabriz, Tabriz, Iran ith conventional power unit
3 Pj Minimum power generation of the
cmin
Institute of Research and Development, Duy Tan University,
Da Nang 550000, Vietnam jth CHP unit

13
Vol.:(0123456789)

8562 A. Haghrah et al.

Pcmax
j
Maximum power generation of the the CHPED problem in the Rooijers and van Amerongen
jth CHP unit (1994), authors have used the separability property of the
Hjcmin
 Minimum heat generation of the jth cost function and problem constraints for simplification
CHP unit of the problem. The results show a twenty-fold increase in
Hjcmax Maximum heat generation of the jth speed of obtaining the solution compared with other sim-
ple quadratic programming based methods. In other early
CHP unit
work, CHPED is classified to two power dispatch and heat
Hkhmin Minimum heat generation of the kth
dispatch sub-problems (Guo et al. 1996). While subproblems
boiler unit
are almost independent, they are related to each other based
Hkhmax Maximum heat generation of the kth
on feasible operation region for CHP plants. The cost func-
boiler unit
tion used in the mentioned study is a simplified quadratic
Np Number of conventional power units
function, which is differentiable and convex. By introduction
Nc Number of CHP units
of heuristic algorithms, many studies have been done for
Nh Number of boiler units
solving engineering problems using these methods. These
Ploss Transmission loss
methods are used in a wide area of optimization problems
Bij , B0i , B00 Transmission loss coefficients
ranging from design of electrical motors (Shabanian et al.
2015) to DNA sequencing (Blum et al. 2008). They have
also shown a great capability to solve economic dispatch
1 Introduction
problems, for which CHPED problem can be also included
in such problems.
Although electric power consumption is going its growth
Recently, numerous optimization methods are introduced
trend, and climatic effects caused by pollutants entered
to deal with the CHPED problem (Mandal et  al. 2015;
into the atmosphere have made environmental concerns,
Naderi et al. 2017; Xiong and Shi 2018; Qin et al. 2017).
demand for efficient and clean energy production systems
In Nguyen et al. (2016), a Cuckoo search algorithm using
have arisen. The conventional combined cycle (CC) gen-
penalty function is proposed to solve complicate and difficult
eration units in the best condition experience efficiencies
CHPED problem. In this regard, the objective function of
between 50 and 60%. The largest share of energy waste in
the problem is assumed as the sum of the cost function and
CC units belongs to the heat power losses. By recovering
penalty function. The penalty function is considered such
this lost energy using combined heat and power (CHP)
a way that, more constraints violation would cause larger
generation plants, the achieved efficiency in power produc-
penalty values. So, as the objective function is reduced
tion increases up to 90% (Zou et al. 2019), while pollutant
in the optimization process, the population will be led to
emissions reduces about 13–18% (Nazari-Heris et al. 2018).
feasible and optimal space. A similar method is employed
Aforementioned characteristics for CHP generation have
for solving the CHPED problem in Mellal and Williams
made it a remarkable alternative for CC units. Recently, in
(2015), which has notable strengths in terms of few param-
Marseglia et al. (2019), experimental tests and modeling
eters required in optimization process, solution quality in
have been done for a combined heat and power biomass
terms of minimum operating cost and less calculation time.
plant, in which the reported results confirms the advantages
It should be noted that constraints handling in this paper is
of the CHP generation. The proper utilization of the CHP
likewise based on penalty function. Another research that
generation characteristics is subject to accurate and optimal
has used a method based on GA for solving combined heat
energy management of the co-generation units. CHP eco-
and power economic dispatch problem is Subbaraj et al.
nomic dispatch (CHPED) problem is a process to determine
(2009), where the self-adapted real-coded GA proposed
outputs of generation plants meeting constraints so that heat
in this reference is powered by tournament selection and
and power loads are satisfied by the minimum operational
simulated binary crossover. Harmony search (HS) method
costs (Li et al. 2018). Heat and power interdependence, non-
is utilized for solving CHPED problem (Vasebi et al. 2007),
smooth and non-convex objective function as well as taking
where several test systems are introduced for CHPED prob-
transmission losses into consideration make the CHPED a
lem. A new test system is introduced in this paper while the
complicated and hard optimization problem, which requires
reported results for previous test system demonstrate consid-
a strong optimization method to deal with such complex
erable improvements. A biogeography-based optimization
optimization problem (Nazari-Heris et al. 2017a; Majidi
algorithm is applied to solve CHPED problem in Roy and
et al. 2019).
Ghosh (2017). The proposed algorithm in the mentioned
The CHPED problem has attracted attention of many
paper is based on the geographic dispersion of organisms,
researchers in recent years for which notable efforts has been
in which, the search space is explored using two migration
made in the literature. As one of the earliest works done on
and mutation steps. In Amjady and Sharifzadeh (2010), a

13
An improved real‑coded genetic algorithm with random walk based mutation for solving combined… 8563

modified differential evolution (DE) algorithm is proposed, introduced optimization algorithm, WOA has its positive
where a novel mutation operator and selection process are and negative points which becomes apparent by compar-
motivated by GA, particle swarm optimization (PSO), and ing achieved results with other methods. In the mentioned
simulated annealing. The reported results prove high per- paper, a new large-scale test system is also introduced which
formance of the presented algorithm compared with former challenges the abilities of the optimization algorithm more
methods. Another bio-inspired algorithm used for solving potent. Due to the novelty of the introduced test systems,
the CHPED problem is invasive weed optimization (IWO) data is not adequately available for comparison of different
algorithm (Jayabarathi et al. 2014). This algorithm is based algorithms. Another algorithm, which has achieved remark-
on simulation of the invasive weed growth and proliferation able results is exchange market algorithm (EMA) (Ghorbani
behavior. In more recent researches squirrel search algorithm 2016), which is based on behavior of the shareholders in the
(Basu 2019), whale optimization algorithm (Nazari-Heris market fluctuations. Random walk with variable step size as
et al. 2019), bat optimization algorithm (Dinh et al. 2018), mutation operator of GA has improved capabilities of GA to
and cuckoo search algorithm (Nguyen et al. 2018a) have solve the CHPED in Haghrah et al. (2017). A new method
been used for solving CHPED problem. In addition to com- based on the wild goats life in high and steep mountains is
mon CHPED problem, other problems has also raised in this provided in Shefaei and Mohammadi-Ivatloo (2017), which
field and has attracted the researchers’ attention to itself in is utilized for solving the CHPED, and the obtained results
which price-based demand response (Alipour et al. 2019), prove capability of the method for dealing with complex
multi objective unit commitment of CHP units (Anand et al. problems. A more detailed study on recent heuristic meth-
2019), and accommodation of wind power for CHP systems ods applied for solving the CHPED is done in Nazari-Heris
(Luo et al. 2019) can be mentioned among them. et al. (2017c).
The application of heuristic methods to solve large-scale In this study, an improved real-coded GA random walk
CHPED problem can be hotly debated considering the based mutation (RCGA-CRWM) is presented for solving
requirement of a strong optimization method to deal with large scaled CHPED problems. GA has been widely used
such complexity level. PSO algorithm with time-varying for dealing with optimization problems in engineering sci-
acceleration coefficients is applied for solving the CHPED ences such as multi-level yard crane scheduling (Lei et al.
in Mohammadi-Ivatloo et al. (2013a). The reported results 2018), scheduling of power generation of hydrothermal
corresponding to all test systems and cases are improved in plants (Ganguly et al. 2019), optimal placement and sizing
this study. Also, a new test system is introduced in the men- of wind turbines in power systems (Mokryani et al. 2013),
tioned paper with larger dimensions, compared with prior optimal consumption of energy in home management sys-
test systems. Cuckoo optimization is employed for solving tems (Hussain et al. 2020), and forecasting objectives in
the CHPED in Mellal and Williams (2015) and Mehdine- stock market (Ding et al. 2020). CHPED problem cannot be
jad et al. (2017). Both small and large systems are investi- solved using traditional gradient based optimization tech-
gated in such papers, where the results show a significant niques due to its non-convex, non-smooth and non-linear
improvement compared with previously reported results. nature. As a result, many studies are aimed at solving the
Another algorithm based on GA is proposed in Haghrah problem using heuristic algorithms. These algorithms have
et al. (2016), which has improved simple GA using an inno- the property that they can solve the problem to a satisfactory
vative Muhlenbein mutation operator. Constraints handling level without the need of being linear, convex or smooth.
is like most of the previous algorithm is done by the penalty One of algorithms with such capability is the GA. Many
function. Small-size and medium-size test systems in terms studies have been done to improve the GA. A significant
of number of generation plants are studied in this paper. number of these studies have targeted the mutation opera-
The results show the capability of the proposed algorithm tor with a wide range of innovations. The mutation opera-
for solving complicated optimization problems. Civilized tor that is the strength of the presented method is based on
swarm optimization (CSO) integrated with Powels̀ pattern random walk adopted from Haghrah et al. (2017). Step size
search algorithm is introduced in Narang et al. (2017) to in the random walk can be achieved randomly using vari-
solve a medium-scale CHPED problem. The local and global ous probabilistic distributions. Nowadays, the use of vari-
search are up to Powells̀ pattern search and CSO respec- ous probabilistic distributions for parameter adjustment and
tively. Powells̀ pattern search does not need gradient or hes- improvement of metaheuristic optimization algorithms has
sian matrix so it is applicable to problems with non-differen- attracted many researchers. As an example in Ganesan et al.
tiable objective functions like CHPED. Whale optimization (2018) by using two extreme value distributions, two DE
algorithm (WOA) in Nazari-Heris et al. (2017b) is used for algorithm variants are developed. The proposed methods are
solving the problem, where the method is applied on a large- applied to solve many objective gas turbine system optimiza-
scale system. WOA is based on social treatment of hump- tion. Two important families of probabilistic distributions
back whales and their special hunting method. As a recently are exponential and heavy-tailed distributions. As reported

13

8564 A. Haghrah et al.

in Haghrah et al. (2017), exponential distributions avoid objective function of the CHPED is the summation of opera-
premature convergence, which causes better global search, tional costs for power only plants, CHP plants and heat only
while the heavy-tailed distributions experience faster con- plants, which can be stated as (1):
vergence. The mutation mechanism presented in this study Np
is looking to integrate the positive attributes of aforemen- ∑ p

Nc

Nh
min Ci (Pi ) + Cj (Pcj , Hjc ) + Ck (Hkh ) ($∕h) (1)
tioned two probabilistic distributions. Therefore, a procedure i=1 j=1 k=1
is proposed to deal with this task. According to the achieved
results, the proposed RCGA-CRWM procedure gathers the In the above formulation, C indicates the total production
fast convergence and escaping from the local minima both cost for the unit and, Np , Nc and Nh respectively are the num-
together in unique algorithm. In short, we can mention the ber of conventional plants, co-generation plants, and heat-
features of the proposed algorithm causing improvement in only plants. Power and heat generation output of units are
results, the powerful local search along with escape from the represented by H and P. i, j and k are indices corresponding
local minimums. These features are obtained by using the to above-mentioned unit types respectively. Production cost
potentials of both exponential and heavy-tailed probabilistic formula for conventional plants, CHP plants, and heat-only
distributions together. The mentioned features have made plants are as below:
this algorithm superior to other methods. p p p
Six benchmark functions and four test systems are under Ci (Pi ) = 𝛼i (Pi )2 + 𝛽i Pi + 𝛾i ($∕h) (2)
investigation in this study for evaluating the capability of the
RCGA-CRWM method. Using the benchmark functions, a Cj (Pcj , Hjc ) = aj (Pcj )2 + bj Pcj + cj + dj (Hjc )2 + ej Hjc
complete and detailed comparison has been made between the (3)
well-known methods (i.e., GA, PSO and GSO algorithms), + fj Hjc Pcj ($∕h)
RCGA-RWM and the proposed improved algorithm. RCGA-
RWM has been applied to solve small and medium-size test Ck (Hkh ) = ak (Hkh )2 + bk Hkh + ck ($∕h) (4)
systems but not large-scale ones (Haghrah et al. 2017). Two ( p)
test systems, studied in this paper, are large-scale cases which where Ci Pi is the cost function of conventional thermal
multiplies the complexity of the optimization problem several plant i over one hour period, producing power of Pi MW  . 𝛼i ,
p

times greater. The other two test systems are medium-scale and 𝛽i , and 𝛾i are the cost coefficients of ith conventional thermal
they are very useful in terms of comparability with other meth- unit. The cost function of conventional thermal plants are
ods. The results show a significant improvement in test systems modelled using quadratic function estimation (2) (Khorram
compared with other methods, which confirm the capability and Jaberipour 2011; Song et al. ( 1999; ) Wang and Singh
of the presented RCGA-CRWM method to solve complicated 2008; Esmaeeli et al. 2019). Cj Pj , Hj is operation cost of
c c

problems. One of the important issues in comparing the capa- CHP plant j, and aj , bj , cj , dj , ej and fj are the cost coefficients
bilities of the proposed methods is the correct time comparison of such plant. As it can be seen from (3), the cost function
of the methods. Some articles in the field of ED have made this of the co-generation plant is convex in both power supply Pc
comparison well (Nguyen et al. 2018b; Nguyen 2019). In this and heat supply H(c . The
study, inspired by the mentioned articles, the execution time of ) operation cost of heat-only plant k
is indicated by Ck Hkh supplying H h MWth heat. The cost
the proposed methods has been compared. coefficients of kth heat-only plant are defined by ak , bk , and
The remainder of this study has been organized as: Sect. 2 ck . Constraints which are involved in this optimization prob-
proposes the mathematical formulation of the CHPED. lem can be listed as below:
Section 3 presents a summarized description of GA and a
detailed description of the presented algorithm. Section 4 • Power supply and load balance
provides the application of the presented method to six
benchmark functions and four test instances as well as com- Np
∑ ∑
Nc
p
parison of the optimal results with the literature. The impor- Pi + Pcj = Pd (5)
tant conclusions are provided in Sect. 5. i=1 j=1

In which Pd indicates the power load of the system.


• Heat supply and load balance
2 Formulation of the CHPED problem

Nc

Nh

The major goal of the CHPED is minimizing operation cost Hjc + Hkh = Hd (6)
of units in line with satisfying the numerous constraints j=1 k=1

such as power and heat demands, capacity limits, etc. The Where thermal load of the system is represented by Hd.
schematic diagram of the problem is depicted in Fig. 1. The

13
An improved real‑coded genetic algorithm with random walk based mutation for solving combined… 8565

Fig. 1  Schematic diagram of the


CHPED problem

• Capacity limitations of conventional plants


pmin p pmax
Pi ≤ Pi ≤ Pi i = 1, … , Np (7)

• Capacity limits of CHP units

Pcmin
j (Hjc ) ≤Pcj ≤ Pcmax
j (Hjc ) j = 1, … , Nc (8)

Hjcmin (Pcj ) ≤Hjc ≤ Hjcmax (Pcj ) j = 1, … , Nc (9)

The minimum and maximum limitations of jth CHP


plant power supply are represented as functions of gener-
ated heat, Pcmin
j
(Hjc ) and Pcmax
j
(Hjc ) . Similarly, for heat
generation of unit, the limits are functions of generated
power in the form Hjcmin (Pcj ) and Hjcmax (Pcj ) . These two
limiting inequality form a feasible region which the Fig. 2  The heat-power feasible operating region for a CHP plant
power and heat generation of CHP unit must lie within
it. In practice, this region is modeled as a polygon, which
can be sometimes non-convex. An example of such zone
is demonstrated in Fig. 2.

13

8566 A. Haghrah et al.

• Supply limitations of heat-only plants Np N p


∑ ∑ Np Nc
∑ ∑
p p
Ploss = Pi Bim Ppm + Pi Bij Pcj
Hkhmin ≤ Hkh ≤ Hkhmax k = 1, … , Nh (10) i=1 m=1 i=1 j=1
(12)
∑∑
Nc Nc
+ Pcj Bjn Pcn
2.1 Valve point impact j=1 n=1

Quadratic and cubic functions are used in most of the So (5), which represents power supply and load balance,
reported studies (Song et al. 1999; Su and Chiang 2004). must be modified as:
However, the wire drawing effects cause a ripple in genera- Np
tion cost when steam admission valve begins to open. This ∑ p

Nc
Pi + Pcj = Pd + Ploss (13)
impact is modeled by adding a sinusoid term to generation i=1 j=1
cost of the units (Basu 2015; Mohammadi-Ivatloo et al.
2013b). The sinusoid term, which is taken into considera-
tion in this study, makes the problem non-convex and non-
differentiable. Finally, cost function considering valve point 3 The proposed optimization method
effect is expressed as below:
So far, various algorithms have been introduced to improve
p p p pmin p
Ci (Pi ) = 𝛼i (Pi )2 + 𝛽i Pi + 𝛾i + |𝜆i sin(𝜌i (Pi − Pi ))| (11) the GA. The vast majority of these methods have targeted the
combination and mutation operators because of their great
Where, 𝜆i and 𝜌i are coefficients of valve-point impact. The impact on the algorithm characteristics and the results. One
cost by consideration of such case is depicted in Fig. 3 for of the recent efforts to improve GA has ended with new
𝛼i = 0.00028  , 𝛽i = 8.10  , 𝛾i = 550  , 𝜆i = 300  , 𝜌i = 0.035 , mutation operator based on the random walk with variable
= 0 and Pi = 680.
pmin pmax
Pi step size which is done in Haghrah et al. (2017). In the men-
tioned reference, the real-coded GA is improved by intro-
2.2 Transmission loss consideration ducing a new mutation operator based on the random walk
with variable step size. In this method, the step size used for
There are two approaches to calculate transmission loss con- the random walk is achieved using a random variable with
taining load flow method (Abdelaziz et al. 2008) and Kron’s specific probabilistic distribution. According to the results
loss formula that is known as B-matrix method (Victoire reported in Haghrah et al. (2017), which are achieved using
and Jeyakumar 2005). Regardless of the calculation method, two exponential and heavy-tailed family of probabilistic dis-
transmission loss is a function of power generation of all tributions for determining random walk step size, the expo-
plants. In this paper, Kron’s loss formula is used to study the nential distributions prevent premature convergence while
transmission loss of the system, which can be represented by the heavy-tailed distributions cause faster convergence. The
the following formula: main idea for improving the RCGA-RWM algorithm is to
earn both positive characteristics of exponential and heavy-
tailed distributions. The realization of this idea means faster
convergence along with avoiding local minima. Two prob-
ability distributions are under study for each exponential and
heavy-tailed families, which are listed as below:

• Normal distribution (RCGA-RWM1)

1 x2
f (x�𝜎 2 ) = √ e− 2𝜎2 (14)
2𝜋𝜎 2
• Exponential distribution (RCGA-RWM2)

f (x|𝜆) = 𝜆e−𝜆x (15)


• Lévy distribution (RCGA-RWM3)
√ c
c e− 2x
f (x|c) = (16)
2𝜋 x 32
Fig. 3  Consideration of valve-point impacts in operation cost of units

13
An improved real‑coded genetic algorithm with random walk based mutation for solving combined… 8567

• Burr distribution (RCGA-RWM4) the local minima allow us to expect good answers from this
algorithm.
xc−1 The strategy maintained during the execution of the
f (x|c, k) = ck (17)
(1 + xc )k+1 RCGA-CRWM algorithm in this paper is to make the heat
and power demand equality constraint be met using two sep-
3.1 Combined random walk mutation arated units, one for each generation type. Using this method
for initial population generation, crossover operator and
In order to have both benefits of the step sizes achieved from mutation, there will be no worries about the heat and power
exponential and heavy-tailed distributions, the step size in demand equality. The crossover operator used in this paper is
this paper is derived as below: simply weighted averaging with randomly generated weight
{ coefficients. Mutation operator as described, is based on the
𝜌 if 𝜓 > 𝜓,̂ random walk with variable step size, in which the step size
𝛾=
𝜁 if 𝜓 ≤ 𝜓.̂ (18) is a random variable. Constraints handling, selection and
handling the constraints when power loss is also taken into
In which, 𝜌 and 𝜁 are random variables with exponential account are adopted from Haghrah et al. (2016). The objec-
and heavy-tailed distributions and 𝜓 is a random variable tive function of the optimization problem is the summation
with a uniform distribution in the range of [0, 1) . Also, 𝜓̂ is of operation cost of units and penalty values dedicated to the
a threshold to determine the emphasis on each of random violation of constraints. The penalty functions are calculated
variables. as a quadratic function of constraint violation. In the case of
The main task of the random walk-based mutation opera- operating point of CHP units, the penalty function would be
tor is to converge the population to local minima. However, a function of shortest distance of the operating point from
the point differentiating this method from other common feasibility region of the unit. As the algorithm progresses, in
approaches is that diversity of the step size may cause escap- order to reduce the objective function the population move
ing of individuals from one local minimum’s vicinity to oth- toward the points in which the penalty value is less. This
ers. So, if both local convergence and experiencing other means moving from infeasible points of the problem space
localities is seeked, the step size should be chosen in such to feasible subspace. By summarizing the above, the objec-
a way that in spite of the majority of the step sizes near to tive function of the optimization problem can be formulated
zero, there would be also a possibility of longer step sizes. as follows:
As a result of experiencing more local minima, there would
be larger probability of reaching better results. Therefore,
Np
∑ ( p) ( p)
it can be claimed that the strength of the proposed RCGA- F= Ci Pi + Ξi Pi
CRWM algorithm and its innovation is providing mentioned
i=1


Nc ( ) ( )
conditions for generating step size in random walk.
+ Cj Pcj , Hjc + Ξj Pcj , Hjc (19)
j=1
3.2 Implementation of the proposed algorithm
to solve CHPED problem ∑
Nh
( ) ( )
+ Ck Hkh + Ξk Hkh
k=1
According to the explanations given in the Section 3.1 about
the characteristics of the algorithm, the proposed RCGA-
⎧ � �2 � �
CRWM is able to solve complex, non-linear and non-convex � p � ⎪ 𝜂i1 Ppmin − Pp + 𝜂i2 Ppmin − Pp + 𝜂i3 if Pp < Ppmin
Ξi Pi =⎨ � i i
�2 � p
i i i i
optimization problems as CHPED. Properly exploring the pmax �
⎪ 𝜂i1 Ppi − Ppmax + 𝜂 P − P + 𝜂 if P
p
> P
pmax
problem space and reducing the probability of falling into ⎩ i i2 i i i3 i i

(20)
{ ( ) ( ) ( )
( ) 𝜂j1 Δ2j Pcj , Hjc + 𝜂j2 Δj Pcj , Hjc + 𝜂j3 if Pcj , Hjc is outside FR
Ξj Pcj , Hjc = (21)
0 oth.

{ ( )2 ( )
( ) 𝜂k1 Hkhmin − Hkh + 𝜂k2 Hkhmin − Hkh + 𝜂k3 if Hkh < Hkhmin
Ξk Hkh = ( )2 ( ) (22)
𝜂k1 Hkh − Hkhmax + 𝜂k2 Hkh − Hkhmax + 𝜂k3 if Hkh > Hkhmax

13

8568 A. Haghrah et al.

In which Ξ s indicate the penalty functions, which are in a Due to the influence of the execution time of algorithms
quadratic form. The coefficients of the penalty functions are from multiple factors, it is hard to compare them. Articles
indicated using 𝜂 indices. Also, the function Δj calculates
( the
) like Mernik et al. (2015) and Črepinšek et al. (2014) have
shortest distance of the CHP unit j operating point, Pcj , Hjc  , addressed this issue. One of the best methods for compar-
from the sides of its polygon feasibility region. It must be ing the execution time of the algorithms is to consider the
noted that, in our implementations all the 𝜂i3 s and 𝜂k3 s are total fitness evaluations. The number of total fitness evalu-
set to zero. ations for the proposed algorithm and other members of
Mismatch error as a stopping condition is a useful the RCGA-RWM family can be calculated by the formula
method for ending the optimization process in proper itera- below:
tions. However, in the designing process of the algorithm ( )
Population
for solving the CHPED problem, the appropriate number of Iterations × + Number of Mutations (23)
2
iterations has been determined by numerous experiments. So
no need for using the mismatch error is felt. The flowchart In each iteration of the proposed algorithm, combination is
of the presented RCGA-CRWM algorithm proposed in this done Population times. In each combination operator, the fitness
2
paper is illustrated in Fig. 4. is evaluated one time for the new resulted individual. Also,

Fig. 4  Flowchart of the pro-


posed RCGA-CRWM algorithm

13
An improved real‑coded genetic algorithm with random walk based mutation for solving combined… 8569

the number of mutations in each iteration is an algorithm before, in general, this parameter acts as a threshold value.
parameter. In each mutation operator, the fitness is evalu- For example when 𝜓̂ = 0.5 , it means that half of mutation
ated just one time. Finally it must be added that number of operators are RWM2, and the other half are RWM4. But the
the individuals in the population, number of the iterations, important point is that by using the proposed method the
number of the applied mutation operators in each iteration, order of application of these operators is random, regard-
and determined top percent of the population, which the less of their type. These parameters are determined after
mutation is applied on them are 4000, 2000, 4000, 0.5%, some tests for different values. The mentioned values have
respectively. Considering this and knowing that for all the resulted the best convergence characteristics for CHPED
algorithms with different random walk based mutations problem. The 𝜓 parameter used in this table is a uniformly
these parameters are same, total fitness function evaluation distributed probabilistic variable in the range [0, 1) . In all
would be twelve million times for all five types of the algo- cases of benchmark functions and test systems, results are
rithm, similarly. obtained by running the method for 100 times. The running
time is average running time of all 100 runs. It is notable
that the reported results for the CHPED problem are rounded
4 Case studies up to four decimal digits. The simulations are done using
Python 3.6 and NumPy (Walt et al. 2011) over Intel Core i7
The presented RCGA-CRWM is applied to solve six bench- 2.80 GHz CPU and the figures are depicted using Matplotlib
mark functions and four test systems. Convergence speed of (Hunter 2007).
the presented algorithm is compared with simple RCGA-
RWM and other recently proposed algorithms. In the case 4.1 Benchmark functions
of step size with combined probabilistic distribution, Nor-
mal and Burr distributions are used. Algorithm parameters In this section, six benchmark functions are studied for com-
are mentioned in Table 1. There is no significant differ- paring the performance of the presented RCGA-CRWM
ence in choosing the parameters for the basic algorithms algorithm with other well-known methods. Benchmark func-
and the proposed algorithm. In the proposed algorithm, tions are different functions to be minimized for measuring
the 𝜓̂ is introduced for combining two mutation methods the capabilities of optimization algorithms. Many of these
of RCGA-RWM2 and RCGA-RWM4. As it is described functions are introduced in Yao et al. (1999) for the first time
as a benchmark. Data of such functions are adapted from He
et al. (2009) that are provided in Table 2. Mean and standard
deviation of the results obtained by applying RCGA-CRWM
Table 1  Algorithm parameters to benchmark functions are provided in Table 3. The results
Method Parameters show the ability and superiority of the presented algorithm
in solving the optimization problems and compared with
RCGA-RWM1 𝜎2 = 𝜓 other well-known optimization algorithm methods like GA,
RCGA-RWM2 𝜆=𝜓 PSO and GSO algorithms.
RCGA-RWM3 p = 1.5 × 𝜓
RCGA-RWM4 c = 5 × 𝜓, k = 𝜓 4.2 Test system I
RCGA-CRWM 𝜆 = 𝜓  ,
c = 5 × 𝜓, k =  ,𝜓 The first under study test system contains 52conventional
𝜓̂ = 0.5
units, 24 cogeneration plants, and 20 heat-only units. Test

Table 2  Data of benchmark Benchmark functions n Search space Global minimum


functions
∑n
f1 (x) = 2 2
i=1 (100(xi+1 − xi ) + (xi − 1))
2 30 [− 30, 30]n 0
∑n
f2 (x) = 2
i=1 (xi − 10cos(2𝜋xi ) + 10)
2 30 [− 5.12, 5.12]n 0
∑n �∑i �2 30 [−100, 100] n
0
f3 (x) = i=1 j=1 xj

f4 (x) = 4x12 − 2.1x14 + 13 x16 + x1 x2 − 4x22 + 4x24 2 [− 5, 5]n − 1.0316285


{ }
f5 (x) = maxi |xi |, 1 ≤ i ≤ n 30 [− 100, 100]n 0
� �
1 ∑n 30 0
n
f6 (x) = 4000 (x − 100)2 − Πni=1 cos i √
x −100
+1 [− 600, 600]
i=1 i i

13

8570 A. Haghrah et al.

Table 3  Comparison of different algorithms for benchmark functions


Method f1 f2 f3 f4 f5 f6

GA (He et al. 2009) Mean 338.5516 0.6509 9749.9145 − 1.0298 7.9610 1.0038
Std. 361.497 0.3594 2594.9593 3.1314 × 10−3 1.5063 6.7545 × 10−2
PSO (He et al. 2009) Mean 37.3582 20.7863 1.1979 × 10 −2 − 1.0160 0.4123 0.2323
Std. 32.1436 5.94 2.1109 × 10−3 1.2786 × 10−2 0.2500 0.4434
GSO (He et al. 2009) Mean 49.8359 1.0179 5.7829 − 1.031628 0.1078 3.0792 × 10−2
Std. 30.1771 0.9009 3.6813 0 3.9981 × 10 −2
3.0867 × 10−2
RCGA-RWM1 Mean 0.1847 1.7951 × 10 −10 1.3973 × 10 −3 −1.0316284535 9.4839 × 10 −5 3.294 × 10−4
Std. 0.3965 8.7545 × 10−10 1.0654 × 10−3 2.2204 × 10−16 5.9860 × 10−5 1.8658 × 10−3
RCGA-RWM2 Mean 5.0528 5.0991 × 10 −24
4.8756 × 10 −4 − 1.0316284535 2.9658 × 10 −4
9.8573 × 10−5
Std. 5.2829 5.1574 × 10 −24
2.8674 × 10 −4
2.2204 × 10 −16
6.9127 × 10 −5 9.8079 × 10−4
RCGA-RWM3 Mean 26.1954 5.2520 × 10−5 1.1756 − 1.0316284535 0.0924 0.0060
Std. 0.2930 5.7212 × 10−5 0.4755 2.2204 × 10−16 0.0166 0.0075
RCGA-RWM4 Mean 3.1480 0.0 3.3851 × 10 −8 −1.0316284535 1.7656 × 10−8
2.2188 × 10−4
Std. 5.4024 0.0 3.0861 × 10 −8
2.2204 × 10 −16 7.1230 × 10 −9
1.2617 × 10−3
RCGA-CRWM Mean 2.0601 0.0 2.0701 × 10−8 −1.0316284535 1.5374 × 10−8 4.1903 × 10−4
Std. 4.4067 0.0 1.4722 × 10 −8
2.2204 × 10 −16 6.1329 × 10 −9
1.8463 × 10−3

system data, in this case, is obtained from Meng et al. applied to solve the first test system are real-coded GA with
(2015), which is also represented in Table 8 in the “Appen- improved Muhlenbein mutation (RCGA-IMM), PSO with
dix” section. It should be mentioned that valve-point effects time varying acceleration coefficients (TVAC-PSO), WOA
are investigated in this test system. Power and heat loads and Crisscross optimization (CSO). Obviously, the results
are respectively considered as 9400MW and 5000MWth . of proposed algorithm in this paper is superior to all other
The results achieved by proposed RCGA-CRWM are algorithms. The computational burden of the proposed
mentioned in Table 9 in the “Appendix” section. Also, the algorithm is greater than other algorithms, but it can be
results and execution time of the algorithms are compared justified due to the significant enhancement of the results
in Table 4. The scheduled heat and power generations are beside the fact that CHPED problem is solved offline.
plotted on the feasible region of the CHP units in test sys- The minimum operation cost using proposed algorithm is
tem I, for showing the feasibility of the results. Figure 5. $231,623.4800 , which saves $2039.793 hourly with respect
shows the feasible region and operation point for 24 CHP to best prior result. In other words, an annual cost saving
units. It must be noted that in each plot, it is zoomed to of $744,524.445 can be attained by applying the proposed
the interval [Pc − 0.1, Pc + 0.1] and [H c − 0.1, H c + 0.1] RCGA-CRWM method with respect to the best reported
around the operaion point for better visibility. The Fig. 5 result in the literature. Convergence characteristics of the
confirms that for all CHP units the heat and power genera- proposed algorithm compared with ordinary RCGA-RWM,
tion is feasible. which is demonstrated in Fig. 6, shows acceptable conver-
The sum of power supply by conventional plants and gence rate along with escaping local minimums. For all
CHP units reported in Table 9 is equal to 9400MW , which 100 runs, the results achieved using the proposed algorithm
means the power load can be provided by sum of the ther- are feasible. For showing this, the Figure 7 is plotted. As it
mal units and CHP plants. In addition, the sum of heat can be seen from this figure, the average of penalty value
generation by heat-only plants and CHP units reported over 100 runs converges to zero, which approves the fea-
in Table 9 is equal to 5000MWth , which means the heat siblity of the solutions.
load can be supplied by sum of the heat-only units and
CHP plants. The methods in the table are sorted by the 4.3 Test system II
minimum cost function in a descending order. As it can be
seen in comparisons, the results indicate superiority of the The second test system used for proving the functionality
proposed algorithm versus others. The other algorithms of the presented method to solve the CHPED is composed

13
An improved real‑coded genetic algorithm with random walk based mutation for solving combined… 8571

Fig. 5  Heat and power genration plot for test system I

of 40 conventional thermal units, 24 cogeneration units and solving this test system and obtained results are reported
20 heat only units. This recently introduced test system is in Table 11 in the “Appendix” section. The reported results
taken from Nazari-Heris et al. (2017b) that is classified as a for this case study in Table 11 shows that the sum of power
large-scale test system. The data for the second test system generation by thermal units and CHP plants is equal to
is also represented in Table 10 in the “Appendix” section. the power demand of 12, 700 MW . In addition, the heat
Total power and heat demand in the under investigation demand 5000 MWth is satisfied by sum of heat production
test system is respectively 12, 700 MW and 5000 MWth . of CHP units and heat-only plants as the results provided
The proposed RCGA-CRWM algorithm is employed for in Table 11. To show the feasibility of the scheduling done

13

8572 A. Haghrah et al.

Fig. 6  Convergence of the proposed algorithm results for test system I compared with simple RCGA-RWM

(a) All iterations for all three values. (b) Last 250 iterations for penalty value.

Fig. 7  Penalty value, cost function of problem, and operation cost of system convergence diagrams for test system I

Table 4  Comparison of the result achieved using proposed algorithm with other methods for test system I
Method Mean cost($) Maximum cost ($) Minimum cost ($) Time(s) Population Iteration Fitness evaluation

RCGA-IMM (Haghrah et al. 2016) – – 239, 896.4083 280.47 3000 1000 NA


TVAC-PSO (Mohammadi-Ivatloo – – 239, 139.5018 198.25 500 300 NA
et al. 2013a)
WOA (Nazari-Heris et al. 2017b) 237, 431.4678 238, 877.0492 236, 699.1501 227.35266 NA 1000 NA
NCSO (Meng et al. 2015) 233, 742.6482 233, 863.5784 233, 663.2730 201.14 30 1000 NA
RCGA-RWM1 234, 448.1330 238, 889.2605 231, 816.3893 276.5426 4000 2000 12 × 106
RCGA-RWM2 234, 718.4552 239, 942.8586 231, 711.3487 284.9564 4000 2000 12 × 106
RCGA-RWM3 233, 538.5850 236, 559.9750 232, 155.2907 291.4512 4000 2000 12 × 106
RCGA-RWM4 233, 738.2287 240, 174.5000 231, 792.5158 279.1526 4000 2000 12 × 106
RCGA-CRWM 233, 691.7439 238, 715.8580 231, 623.4800 311.2147 4000 2000 12 × 106

13
An improved real‑coded genetic algorithm with random walk based mutation for solving combined… 8573

Fig. 8  Heat and power genration plot for test system II

using the proposed algorithm, heat and power generation of A comparison of achieved results with previously
the CHP units are also plotted on the feasible region of the reported ones is represented in Table 5. The methods in the
units, demonstrated in Fig. 8. For a better demonstration of table are sorted by the minimum cost function in a descend-
the operation point position, it is zoomed to the region in ing order. The results indicate superiority of the proposed
such a way that only the interval [Pc − 0.1, Pc + 0.1] and algorithm versus others also in second test system. The pro-
[H c − 0.1, H c + 0.1] is visible. As it can be seen all the heat posed RCGA-CRWM algorithm saves $2156.7317 hourly
and power generations are inside the feasible region of the for the second test system, which is a significant amount.
CHP units. In other words, an annual cost saving of $787,202.071 can

13

8574 A. Haghrah et al.

Fig. 9  Convergence of the proposed algorithm results for test system II compared with simple RCGA-RWM

(a) All iterations for all three values.


(b) Last 250 iterations for penalty value.

Fig. 10  Penalty value, cost function of problem, and operation cost of system convergence diagrams for test system II

Table 5  Comparison of the results achieved using proposed algorithm with other methods for test system II
Method Mean cost($) Maximum cost ($) Minimum cost ($) Time (s) Population Iteration Fitness evaluation

RCGA-IMM (Haghrah et al. 2016) – – 298,417.1870 140.91 3000 1000 NA


TVAC-PSO (Mohammadi-Ivatloo – – 295,680.9138 90.21 500 300 NA
et al. 2013a)
WOA (Nazari-Heris et al. 2017b) 290,947.9406 291,347.9406 290,123.9742 158.18 NA 1000 NA
RCGA-RWM1 292,377.1560 301,516.7098 288,840.5973 170.2875 4000 2000 12 × 106
RCGA-RWM2 292,259.3548 300,721.9103 288,611.9276 175.5821 4000 2000 12 × 106
RCGA-RWM3 289,887.6796 293,496.6437 288,529.4466 181.7934 4000 2000 12 × 106
RCGA-RWM4 292,081.7548 299,813.3320 288,001.5549 172.8762 4000 2000 12 × 106
RCGA-CRWM 291,733.4830 297,748.0864 287,967.2425 203.2781 4000 2000 12 × 106

13
An improved real‑coded genetic algorithm with random walk based mutation for solving combined… 8575

Table 6  Comparison of the Method Mean cost($) Maximum cost ($) Minimum cost ($)
results achieved using proposed
algorithm with other methods CPSO (Mohammadi-Ivatloo et al. 2013a) 59,853.478 60,076.6903 59,736.2635
for 24-unit test system
TVAC-PSO (Mohammadi-Ivatloo et al. 2013a) 58,359.552 58,498.3106 58,122.7460
GSO (Hagh et al. 2014) 58,295.9243 58,318.8792 58,225.7450
IGSO (Hagh et al. 2014) 58,156.5192 58,219.1413 58,049.0197
RCGA-IMM (Haghrah et al. 2016) 58, 066.6354 58,301.9013 57,927.6919
EMA (Ghorbani 2016) 57,832.7361 57,841.1469 57, 825.4792
NCSO (Meng et al. 2015) 57,908.3154 57,911.9533 57,907.3154
CSO (Narang et al. 2017) 62,102 62,119 62,068
CSO-PPS (Narang et al. 2017) 57837 57,842 57,834
MCH (Narang et al. 2017) 57,838 57,844 57,835
RCGA-RWM1 58,556.0065 59,429.7148 57,979.4517
RCGA-RWM2 58,003.1647 58,305.3323 57,842.1977
RCGA-RWM3 57,921.9415 58,001.8286 57,847.5085
RCGA-RWM4 57,922.0647 58,000.7138 57,49.4261
RCGA-CRWM 57,877.0069 57,899.9327 57,825.8690

Table 7  Comparison of the Method Mean cost($) Maximum cost ($) Minimum cost ($)
results achieved using proposed
algorithm with other methods TVAC-PSO (Mohammadi-Ivatloo – – 117,824.8956
for 48-unit test system et al. 2013a)
GSA (Beigvand et al. 2016) – – 117,266.6810
COA (Mehdinejad et al. 2017) 116,835.5491 117,068.2693 116,789.91535
NCSO (Meng et al. 2015) 115,995.8842 116,047.2154 115,967.7205
CSO (Narang et al. 2017) 128,992 129,067 128,882
CSO-PPS (Narang et al. 2017) 115,658 115,661 115,656
MCH (Narang et al. 2017) 115,659 115,661 115,658
RCGA-RWM1 116,436.0793 119,532.6516 115,869.5310
RCGA-RWM2 116,217.5015 117,714.2621 115,800.9527
RCGA-RWM3 116,041.4993 116,781.3671 115,810.7142
RCGA-RWM4 115,908.6170 117,244.2918 115,778.7941
RCGA-CRWM 115,689.2890 115,888.7385 115,653.9039

be attained by using the proposed RCGA-CRWM method comparison shows that the proposed RCGA-CRWM has the
with respect to the best result reported in the literature in best performance among other optimization techniques be
terms of operation cost. Also, the computational burden of obtaining the optimal solution with less iteration numbers.
the algorithm is greater than other algorithms in this case, Figure 10 shows the average of penalty value, cost func-
but it can be justified due to the significant enhancement of tion of the problem, and operation cost of the system over
the results beside the fact that CHPED problem is solved 100 runs. As it can be seen the penalty value converges to
offline. Figure 9 represents convergence diagram of the pro- zero, which shows all the runs are terminated with feasible
posed algorithm in comparison with other methods. The solutions.

13

8576 A. Haghrah et al.

4.4 Medium‑scale test systems minimum is increased. These two properties are integrated
in the proposed method, which enhance the performance of
In addition to the two large-scale test systems mentioned the genetic algorithm.
above, there are also two medium-scale test systems, which In order to indicate the abilities of the proposed method,
have been the subject of many proposed optimization algo- it is implemented to solve six benchmark functions and four
rithms. In this section without going into the details of these test systems of CHPED problem. The results achieved by
24 and 48-unit test systems, the results of the implementa- implementation of the presented method on benchmark
tion of the proposed algorithm on them are reported. These functions, confirm the power of the algorithm compared
test systems data are accessible over literature (Majidi with well-known optimization algorithms GA, PSO, and
et al. 2019; Vasebi et al. 2007). The results achieved by the GSO. Moreover, significant results are achieved by imple-
RCGA-CRWM algorithm for 24-unit test system is pre- mentations for solving large-scale test systems. In the first
sented in Table 12 in the “Appendix” section and they are case, the scheduling cost is reduced by $2039.793 compared
compared with other methods in Table 6. Also for 48-unit with previous works. Also, in the second case, the saved cost
test system the results are demonstrated in Table 13 in the is equal to $2156.7317 . The reported results in this study
“Appendix” section and they are compared with well-known indicate the potential ability of the presented algorithm to
algorithms in Table 7. As the results approve, the proposed solve optimization problems. Considering the large-scale
algorithm overcome to previously proposed algorithms in test systems studied in this paper, it can be claimed that,
medium-scale cases, too. this algorithm maintains its capability in dealing with the
problems having large dimension. It must be noticed that
the minimum, maximum and mean costs and run time of
5 Conclusions the algorithms are achieved by running the algorithm for
100 times, which indicates the robustness of the proposed
In this study, the real-coded GA with random walk mutation algorithm. Also, the constraints handling method used in
is used to solve the non-linear and non-convex CHPED prob- the proposed algorithm yields 100% feasible results, which
lem. The step size in random walk algorithm is achieved by shows its effectiveness.
random variables with exponential family and heavy-tailed The future trends include the application of the proposed
family probabilistic distributions. Given that each proba- optimization method to solve the optimal scheduling of com-
bilistic distribution has a specific local and global impact bined heat and power systems considering a daily time inter-
on the convergence of the presented method, mutual use val with uncertainty of system parameters such as power
of probabilistic distributions from both families improves and heat demands uncertainty. Also, the proposed optimi-
the performance of the method, which is done in this paper. zation method can be applied to large scale combined heat
Most of the recent approaches that have outperformed and power economic dispatch problem with high number of
other methods have two capabilities in parallel; the first power and heat generation plants, where we have a complex
is a strong local search, and the second is avoiding local optimization problem.
minimums. Since random processes with exponential proba-
bilistic distributions occur more near the mean value, they
can be very effective for converging the minimum in a local Appendix
search task. In contrast, heavy-tailed probabilistic distribu-
tions that make occurring the values far than the mean value Detailed data for test systems I and II, and the results
more probable compared with exponential one, increases the achieved by the proposed algorithm are demonstrated in
likelihood of experiencing more local minimums. In other this section.
words, in this case the probability of reaching the global See Tables 8, 9, 10, 11, 12, 13.

13
Table 8  Cost function parameters of test system I
Unit a ( $∕ MW2) b ( $∕MW) c ( $) d ( $) e (rad/ pmin (MW) pmax (MW)
MW)

Power-only units
 1, 14, 27, 40 0.00028 8.1 550 300 0.035 0 680
 2, 3, 15, 16, 28, 29, 41, 42 0.00056 8.1 309 200 0.042 0 360
 4, 5, 6, 7, 8, 9, 17, 18, 19, 20, 0.00324 7.74 240 150 0.063 60 180
 21, 22, 30, 31, 32, 33, 34, 35,
43,
 44, 45, 46, 47, 48
 10, 11, 23, 24, 36, 37, 49, 50 0.00284 8.6 126 100 0.084 40 120
 12, 13, 25, 26, 38, 39, 51, 52 0.00284 8.6 126 100 0.084 55 120
Unit a ( $∕MW2) b ( $∕MW) c ( $) d ( $∕MWh2 e ( $∕MWh) $∕MW MWh Feasible region coordinates [ Pc , H c]
)

CHP units
 53, 55, 59, 61, 0.0345 14.5 2650 0.030 4.200 0.031 [98.8, 0], [81, 104.8], [215, 180], 247, 0]
 65, 67, 71, 73
 54, 56, 60, 62, 0.0435 36.0 1250 0.027 0.600 0.011 [44, 0], [44, 15.9], [40, 75], [110.2, 135.6],
 66, 68, 72, 74 [125.8, 32.4], [125.8, 0]
 57, 63, 69, 75 0.1035 34.5 2650 0.025 2.203 0.051 [20, 0], [10, 40], [45, 55], [60, 0]
An improved real‑coded genetic algorithm with random walk based mutation for solving combined…

 58, 64, 70, 76 0.072 20.0 1565 0.020 2.340 0.040 [35, 0], [35, 20], [90, 45], [90, 25], [105, 0]
h h
Unit a ( $∕MWh2) b ( $∕MWh) c ( $) Hmin (MWh) Hmax (MWh)

Heat-only units
 77, 82, 87, 92 0.038 2.0109 950 0 2695.20
 78, 79, 93, 94, 88, 0.038 2.0109 950 0 60
89, 93, 94
 80, 81, 85, 86, 90, 0.052 3.0651 480 0 120
91, 95 ,96

13
8577

8578 A. Haghrah et al.

Table 9  The results of Output RCGA-CRWM Output RCGA-CRWM Output RCGA-CRWM


the proposed algorithm
implemented for solving test P1 538.6071 P41 224.6733 H57 39.9997
system I
P2 224.8561 P42 224.5798 H58 20.0041
P3 299.7651 P43 110.0325 H59 104.8574
P4 109.9418 P44 109.886 H60 75.0058
P5 109.8726 P45 109.8813 H61 104.8562
P6 159.7508 P46 159.7813 H62 74.9928
P7 109.8884 P47 109.8657 H63 40.0052
P8 159.7662 P48 109.8701 H64 20.0938
P9 109.9021 P49 77.4064 H65 104.7797
P10 77.7472 P50 77.4023 H66 75.0185
P11 77.4692 P51 92.4132 H67 104.8102
P12 55.0473 P52 92.4093 H68 74.9964
P13 92.4387 P53 81.0547 H69 40.0026
P14 628.3168 P54 40.0265 H70 20.0012
P15 224.7251 P55 81.012 H71 104.801
P16 299.6662 P56 40.0721 H72 75.0249
P17 109.8952 P57 10.0051 H73 104.8376
P18 109.9061 P58 35.0399 H74 75.0191
P19 109.8553 P59 81.1214 H75 40.0042
P20 109.9497 P60 40.0078 H76 20.0002
P21 109.8679 P61 81.1079 H77 471.7645
P22 109.8815 P62 40.006 H78 59.9952
P23 77.4601 P63 10.0181 H79 59.9982
P24 40.3116 P64 35.2063 H80 119.9823
P25 92.812 P65 81.025 H81 119.9983
P26 55.0347 P66 40.0221 H82 455.0804
P27 448.8196 P67 81.0196 H83 59.9951
P28 300.8788 P68 40.0005 H84 59.9949
P29 224.6908 P69 10.0064 H85 119.9787
P30 109.9723 P70 35.0114 H86 119.9359
P31 109.9242 P71 81.02 H87 490.2248
P32 110.1147 P72 40.0366 H88 59.9982
P33 109.9348 P73 81.0886 H89 59.9948
P34 109.9262 P74 40.0222 H90 119.9817
P35 110.0015 P75 10.0135 H91 119.9509
P36 77.5056 P76 35.0057 H92 464.3765
P37 114.8053 H53 104.8119 H93 59.9965
P38 92.4331 H54 75.0128 H94 59.9977
P39 92.4942 H55 104.7983 H95 119.9787
P40 538.6133 H56 75.0611 H96 119.9821
Total Power 9400.0000
Total Heat 5000.0000

13
Table 10  Cost function parameters of test system II
Unit a ( $∕MW2) b ( $∕MW) c ( $) d ( $) e (rad/MW) pmin (MW) pmax (MW)

Power-only units
 1 36 114 0.0069 6.73 94.705 100 0.084
 2 36 114 0.0069 6.73 94.705 100 0.084
 3 60 120 0.02028 7.07 309.54 100 0.084
 4 80 190 0.00942 8.18 369.03 150 0.063
 5 47 97 0.0114 5.35 148.89 120 0.077
 6 68 140 0.01142 8.05 222.33 100 0.084
 7 110 300 0.00357 8.03 287.71 200 0.042
 8 135 300 0.00492 6.99 391.98 200 0.042
 9 135 300 0.00573 6.6 455.76 200 0.042
 10 130 300 0.00605 12.9 722.82 200 0.042
 11 94 375 0.00515 12.9 635.2 200 0.042
 12 94 375 0.00569 12.8 654.69 200 0.042
 13 125 500 0.00421 12.5 913.4 300 0.035
 14 125 500 0.00752 8.84 1760.4 300 0.035
 15 125 500 0.00708 9.15 1728.3 300 0.035
 16 125 500 0.00708 9.15 1728.3 300 0.035
 17 220 500 0.00313 7.97 647.85 300 0.035
 18 220 500 0.00313 7.95 649.69 300 0.035
 19 242 550 0.00313 7.97 647.83 300 0.035
 20 242 550 0.00313 7.97 647.81 300 0.035
 21 254 550 0.00298 6.63 785.96 300 0.035
 22 254 550 0.00298 6.63 785.96 300 0.035
An improved real‑coded genetic algorithm with random walk based mutation for solving combined…

 23 254 550 0.00284 6.66 794.53 300 0.035


 24 254 550 0.00284 6.66 794.53 300 0.035
 25 254 550 0.00277 7.1 801.32 300 0.035
 26 254 550 0.00277 7.1 801.32 300 0.035
 27 10 150 0.52124 3.33 1055.1 120 0.077
 28 10 150 0.52124 3.33 1055.1 120 0.077
 29 10 150 0.52124 3.33 1055.1 120 0.077
 30 47 97 0.0114 5.35 148.89 120 0.077
 31 60 190 0.0016 6.43 222.92 150 0.063
 32 60 190 0.0016 6.43 222.92 150 0.063
 33 60 190 0.0016 6.43 222.92 150 0.063
 34 90 200 0.0001 8.95 107.87 200 0.042
 35 90 200 0.0001 8.62 116.58 200 0.042

13
8579

 36 90 200 0.0001 8.62 116.58 200 0.042


Table 10  (continued)
8580

Unit a ( $∕MW2) b ( $∕MW) c ( $) d ( $) e (rad/MW) pmin (MW) pmax (MW)

13
 37 25 110 0.0161 5.88 307.45 80 0.098
 38 25 110 0.0161 5.88 307.45 80 0.098
 39 25 110 0.0161 5.88 307.45 80 0.098
 40 242 550 0.00313 7.97 647.83 300 0.035
Unit a ( $∕MW2) b ( $∕MW) c ( $) d ( $∕MWh2 e ( $∕MWh) $∕MW MWh Feasible region coordinates [ Pc , H c]
)

CHP units
 41, 42, 43, 0.0345 14.5 2650 0.030 4.200 0.031 [98.8, 0], [81, 104.8], [215, 180], [247, 0]
44
 45, 46, 47, 0.0435 36.0 1250 0.027 0.600 0.011 [44, 0], [44, 15.9], [40, 75], [110.2, 135.6],
48
[125.8, 32.4], [125.8, 0]
 49, 50, 51, 0.0345 14.5 2650 0.030 4.200 0.031 [98.8, 0], [81, 104.8], [215, 180], [247, 0]
52
 53, 54, 55, 0.0435 36.0 1250 0.027 0.600 0.011 [44, 0], [44, 15.9], [40, 75], [110.2, 135.6],
56
[125.8, 32.4], [125.8, 0]
 57, 58, 59, 0.1035 34.5 2650 0.025 2.203 0.051 [20, 0], [10, 40], [45, 55], [60, 0]
60
 61, 62, 63, 0.072 20.0 1565 0.020 2.340 0.040 [35, 0], [35, 20], [90, 45], [90, 25], [105, 0]
64
h h
Unit a ( $∕MWh2) b ( $∕MWh) c ( $) Hmin (MWh) Hmax (MWh)

Heat-only units
 65, 66, 67, 68 0.038 2.0109 950 0 2695.20
 69, 70, 71, 72 0.038 2.0109 950 0 60
 73, 74, 75, 76 0.038 2.0109 950 0 60
 77, 78, 79, 80 0.052 3.0651 480 0 120
 81, 82, 83, 84 0.052 3.0651 480 0 120
A. Haghrah et al.
An improved real‑coded genetic algorithm with random walk based mutation for solving combined… 8581

Table 11  The results of Output RCGA-CRWM Output RCGA-CRWM Output RCGA-CRWM


the proposed algorithm
implemented for solving test P1 110.9634 P37 109.9897 H49 104.803
system II
P2 111.9544 P38 109.99 H50 105.7204
P3 117.7255 P39 109.9923 H51 104.8235
P4 179.7987 P40 511.2924 H52 104.8497
P5 88.0273 P41 81.3348 H53 76.1398
P6 139.9989 P42 81.3256 H54 75.0001
P7 281.1083 P43 81.3134 H55 81.6051
P8 289.3265 P44 81.0346 H56 75.2431
P9 288.6283 P45 40.0091 H57 40.2382
P10 279.6026 P46 52.921 H58 40.012
P11 243.6035 P47 40.1385 H59 40.0019
P12 243.6397 P48 41.6106 H60 40.0326
P13 484.0535 P49 81.0323 H61 25.1907
P14 484.0405 P50 82.6595 H62 22.8918
P15 484.0431 P51 81.0544 H63 24.8141
P16 394.2814 P52 81.1229 H64 22.0413
P17 489.2851 P53 41.3211 H65 444.8423
P18 489.3814 P54 40.0003 H66 506.2119
P19 511.2905 P55 47.6517 H67 444.6573
P20 511.2972 P56 40.2817 H68 448.7913
P21 523.446 P57 10.5559 H69 59.9838
P22 523.5186 P58 10.0282 H70 59.9905
P23 523.3627 P59 10.0068 H71 59.9993
P24 523.2837 P60 10.077 H72 59.9877
P25 523.286 P61 46.4254 H73 60.0
P26 523.4204 P62 41.3663 H74 59.9971
P27 10.0118 P63 45.5927 H75 59.9983
P28 10.0045 P64 39.5035 H76 59.9983
P29 10.001 H41 104.9834 H77 119.9983
P30 88.0064 H42 104.9826 H78 119.9757
P31 189.9974 H43 104.9576 H79 119.9674
P32 189.9951 H44 104.814 H80 119.9985
P33 189.9982 H45 74.9947 H81 119.8727
P34 199.9976 H46 86.1506 H82 119.9922
P35 199.9923 H47 75.1051 H83 119.9715
P36 199.9967 H48 76.387 H84 119.9836
Total power 12700.0000
Total heat 5000.0000

13

8582 A. Haghrah et al.

Table 12  The results of Output RCGA-CRWM Output RCGA-CRWM Output RCGA-CRWM


the proposed algorithm
implemented for solving 24-unit P1 628.2956 P11 40.0002 H15 75.0001
test system
P2 299.1787 P12 55.0001 H16 104.8008
P3 298.8267 P13 55.0000 H17 75.0001
P4 109.8666 P14 81.0000 H18 40.0002
P5 109.8550 P15 40.0000 H19 20.0000
P6 109.86298 P16 81.0015 H20 470.4006
P7 109.8498 P17 40.0002 H21 60.0000
P8 109.8623 P18 10.0005 H22 60.0000
P9 60.0000 P19 35.0000 H23 119.9983
P10 77.3999 H14 104.8000 H24 119.9999
Total power 2350.0000
Total heat 1250.0000

Table 13  The results of Output RCGA-CRWM Output RCGA-CRWM Output RCGA-CRWM


the proposed algorithm
implemented for solving 48-unit P1 628.3089 P21 109.8315 H29 104.8005
test system
P2 299.1887 P22 60.0015 H30 75.0000
P3 298.7806 P23 77.4009 H31 40.0001
P4 109.8672 P24 40.0059 H32 19.9999
P5 109.8666 P25 55.0010 H33 104.7992
P6 109.8509 P26 55.0006 H34 75.0123
P7 109.8683 P27 81.0084 H35 104.8032
P8 109.8550 P28 40.0001 H36 75.0053
P9 60.0001 P29 81.0010 H37 40.0001
P10 77.3480 P30 40.0002 H38 20.0000
P11 40.0079 P31 10.0004 H39 471.2813
P12 55.0001 P32 35.0011 H40 59.9999
P13 55.0100 P33 81.0010 H41 59.9988
P14 628.2834 P34 40.0143 H42 119.9842
P15 299.1986 P35 81.0098 H43 119.9930
P16 298.9570 P36 40.0063 H44 469.5370
P17 109.7390 P37 10.0006 H45 59.9990
P18 109.8668 P38 35.0006 H46 59.9998
P19 109.8669 H27 104.8038 H47 119.9849
P20 109.8509 H28 75.0000 H48 119.9975
Total power 4700.0000
Total heat 2500.0000

References Anand H, Narang N, Dhillon J (2019) Multi-objective combined heat


and power unit commitment using particle swarm optimization.
Energy 172:794–807
Abdelaziz A, Kamh M, Mekhamer S, Badr M (2008) A hybrid hnn-qp
Basu M (2015) Combined heat and power economic dispatch using
approach for dynamic economic dispatch problem. Electr Power
opposition-based group search optimization. Int J Electr Power
Syst Res 78(10):1784–1788
Energy Syst 73:819–829
Alipour M, Zare K, Seyedi H, Jalali M (2019) Real-time price-based
Basu M (2019) Squirrel search algorithm for multi-region combined
demand response model for combined heat and power systems.
heat and power economic dispatch incorporating renewable
Energy 168:1119–1127
energy sources. Energy 182:296–305
Amjady N, Sharifzadeh H (2010) Solution of non-convex economic
Beigvand SD, Abdi H, La Scala M (2016) Combined heat and power
dispatch problem considering valve loading effect by a new modi-
economic dispatch problem using gravitational search algorithm.
fied differential evolution algorithm. Int J Electr Power Energy
Electr Power Syst Res 133:160–172
Syst 32(8):893–903

13
An improved real‑coded genetic algorithm with random walk based mutation for solving combined… 8583

Blum C, Vallès MY, Blesa MJ (2008) An ant colony optimization Majidi M, Mohammadi-Ivatloo B, Anvari-Moghaddam A (2019) Opti-
algorithm for dna sequencing by hybridization. Comput Oper mal robust operation of combined heat and power systems with
Res 35(11):3620–3635 demand response programs. Appl Therm Eng 149:1359–1369
Črepinšek M, Liu SH, Mernik M (2014) Replication and comparison Mandal K, Mandal S, Bhattacharya B, Chakraborty N (2015) Non-
of computational experiments in applied evolutionary computing: convex emission constrained economic dispatch using a new
common pitfalls and guidelines to avoid them. Appl Soft Comput self-adaptive particle swarm optimization technique. Appl Soft
19:161–170 Comput 28:188–195
Ding S, Cui T, Xiong X, Bai R (2020) Forecasting stock market return Marseglia G, Medaglia CM, Petrozzi A, Nicolini A, Cotana F, Sormani
with nonlinearity: a genetic programming approach. J Ambient F (2019) Experimental tests and modeling on a combined heat and
Intell Hum Comput 2020:1–13 power biomass plant. Energies 12(13):2615
Dinh B, Nguyen T, Quynh N, Van Dai L et al (2018) A novel method Mehdinejad M, Mohammadi-Ivatloo B, Dadashzadeh-Bonab R (2017)
for economic dispatch of combined heat and power generation. Energy production cost minimization in a combined heat and
Energies 11(11):3113 power generation systems using cuckoo optimization algorithm.
Esmaeeli M, Golshannavaz S, Siano P (2019) Determination of optimal Energ Effi 10(1):81–96
reserve contribution of thermal units to afford the wind power Mellal MA, Williams EJ (2015) Cuckoo optimization algorithm with
uncertainty. J Ambient Intell Hum Comput 2019:1–12 penalty function for combined heat and power economic dispatch
Ganesan T, Aris MS, Vasant P (2018) Extreme value metaheuristics problem. Energy 93:1711–1718
for optimizing a many-objective gas turbine system. Int J Energy Meng A, Mei P, Yin H, Peng X, Guo Z (2015) Crisscross optimization
Optim Eng 7(2):76–96 algorithm for solving combined heat and power economic dispatch
Ganguly D, Das S, Hazra A, Laddha A, Basu M (2019) Improved real problem. Energy Convers Manage 105:1303–1317
coded genetic algorithm-based short-term hydrothermal genera- Mernik M, Liu SH, Karaboga D, Črepinšek M (2015) On clarify-
tion planning. Int J Hybrid Intell 1(2–3):118–146 ing misconceptions when comparing variants of the artificial
Ghorbani N (2016) Combined heat and power economic dispatch bee colony algorithm by offering a new implementation. Inf Sci
using exchange market algorithm. Int J Electr Power Energy Syst 291:115–127
82:58–66 Mohammadi-Ivatloo B, Moradi-Dalvand M, Rabiee A (2013a) Com-
Guo T, Henwood MI, Van Ooijen M (1996) An algorithm for com- bined heat and power economic dispatch problem solution using
bined heat and power economic dispatch. IEEE Trans Power Syst particle swarm optimization with time varying acceleration coef-
11(4):1778–1784 ficients. Electr Power Syst Res 95:9–18
Hagh MT, Teimourzadeh S, Alipour M, Aliasghary P (2014) Improved Mohammadi-Ivatloo B, Rabiee A, Soroudi A (2013b) Nonconvex
group search optimization method for solving chped in large scale dynamic economic power dispatch problems solution using hybrid
power systems. Energy Convers Manage 80:446–456 immune-genetic algorithm. IEEE Syst J 7(4):777–785
Haghrah A, Nazari-Heris M, Mohammadi-ivatloo B (2016) Solving Mokryani G, Siano P, Piccolo A (2013) Optimal allocation of wind tur-
combined heat and power economic dispatch problem using real bines in microgrids by using genetic algorithm. J Ambient Intell
coded genetic algorithm with improved mühlenbein mutation. Humaniz Comput 4(6):613–619
Appl Therm Eng 99:465–475 Naderi E, Azizivahed A, Narimani H, Fathi M, Narimani MR (2017)
Haghrah A, Nekoui M, Mohammadi-Ivatloo B (2017) Random walk A comprehensive study of practical economic dispatch prob-
with variable step size as mutation operator of genetic algorithm lems by a new hybrid evolutionary algorithm. Appl Soft Comput
for solving combined heat and power economic dispatch problem. 61:1186–1206
In: Electrical and electronics engineering (ELECO), 2017 10th Narang N, Sharma E, Dhillon J (2017) Combined heat and power eco-
International Conference on IEEE, pp 57–61 nomic dispatch using integrated civilized swarm optimization and
He S, Wu QH, Saunders J (2009) Group search optimizer: an opti- powell’s pattern search method. Appl Soft Comput 52:190–202
mization algorithm inspired by animal searching behavior. Evol Nazari-Heris M, Abapour S, Mohammadi-Ivatloo B (2017a) Optimal
Comput IEEE Trans 13(5):973–990 economic dispatch of fc-chp based heat and power micro-grids.
Hunter JD (2007) Matplotlib: a 2d graphics environment. Comput Sci Appl Therm Eng 114:756–769
Eng 9(3):90–95 Nazari-Heris M, Mehdinejad M, Mohammadi-Ivatloo B, Babamalek-
Hussain I, Ullah M, Ullah I, Bibi A, Naeem M, Singh M et al (2020) Gharehpetian G (2017b) Combined heat and power economic dis-
Optimizing energy consumption in the home energy management patch problem solution by implementation of whale optimization
system via a bio-inspired dragonfly algorithm and the genetic method. Neural Comput Appl 2017:1–16
algorithm. Electronics 9(3):406 Nazari-Heris M, Mohammadi-Ivatloo B, Gharehpetian G (2017c)
Jayabarathi T, Yazdani A, Ramesh V, Raghunathan T (2014) Combined A comprehensive review of heuristic optimization algorithms
heat and power economic dispatch problem using the invasive for optimal combined heat and power dispatch from economic
weed optimization algorithm. Front Energy 8(1):25–30 and environmental perspectives. Renew Sustain Energy Rev
Khorram E, Jaberipour M (2011) Harmony search algorithm for 81:2128–2143
solving combined heat and power economic dispatch problems. Nazari-Heris M, Mohammadi-Ivatloo B, Gharehpetian GB, Shahi-
Energy Convers Manage 52(2):1550–1554 dehpour M (2018) Robust short-term scheduling of integrated
Lei D, Zhang P, Zhang Y, Xia Y, Zhao S (2018) Research on opti- heat and power microgrids. IEEE Syst J 99:1–9
mization of multi stage yard crane scheduling based on genetic Nazari-Heris M, Mehdinejad M, Mohammadi-Ivatloo B, Babamalek-
algorithm. J Ambient Intell Hum Comput 2018:1–12 Gharehpetian G (2019) Combined heat and power economic dis-
Li Y, Wang J, Zhao D, Li G, Chen C (2018) A two-stage approach for patch problem solution by implementation of whale optimization
combined heat and power economic emission dispatch: Combin- method. Neural Comput Appl 31(2):421–436
ing multi-objective optimization with integrated decision making. Nguyen TT (2019) A high performance social spider optimization
Energy 162:237–254 algorithm for optimal power flow solution with single objective
Luo Y, Yin Z, Yang D, Zhou B (2019) A new wind power accom- optimization. Energy 171:218–240
modation strategy for combined heat and power system based on Nguyen TT, Vo DN, Dinh BH (2016) Cuckoo search algorithm for
bi-directional conversion. Energies 12(13):2458 combined heat and power economic dispatch. Int J Electr Power
Energy Syst 81:204–214

13

8584 A. Haghrah et al.

Nguyen TT, Nguyen TT, Vo DN (2018a) An effective cuckoo search Vasebi A, Fesanghary M, Bathaee S (2007) Combined heat and power
algorithm for large-scale combined heat and power economic dis- economic dispatch by harmony search algorithm. Int J Electr
patch problem. Neural Comput Appl 30(11):3545–3564 Power Energy Syst 29(10):713–719
Nguyen TT, Vo DN, Dinh BH (2018b) An effectively adaptive selective Victoire TAA, Jeyakumar AE (2005) Reserve constrained dynamic
cuckoo search algorithm for solving three complicated short-term dispatch of units with valve-point effects. Power Syst IEEE Trans
hydrothermal scheduling problems. Energy 155:930–956 20(3):1273–1282
Qin Q, Cheng S, Chu X, Lei X, Shi Y (2017) Solving non-convex/non- Walt SVD, Colbert SC, Varoquaux G (2011) The numpy array: a
smooth economic load dispatch problems via an enhanced particle structure for efficient numerical computation. Comput Sci Eng
swarm optimization. Appl Soft Comput 59:229–242 13(2):22–30
Rooijers FJ, van Amerongen RA (1994) Static economic dispatch for Wang L, Singh C (2008) Stochastic combined heat and power dispatch
co-generation systems. IEEE Trans Power Syst 9(3):1392–1398 based on multi-objective particle swarm optimization. Int J Electr
Roy PK, Ghosh M (2017) Combined heat and power dispatch using Power Energy Syst 30(3):226–234
hybrid genetic algorithm and biogeography-based optimization. Xiong G, Shi D (2018) Orthogonal learning competitive swarm opti-
Int J Energy Optim Eng 6(1):49–65 mizer for economic dispatch problems. Appl Soft Comput
Shabanian A, Tousiwas AAP, Pourmandi M, Khormali A, Ataei A Yao X, Liu Y, Lin G (1999) Evolutionary programming made faster.
(2015) Optimization of brushless direct current motor design IEEE Trans Evol Comput 3(2):82–102
using an intelligent technique. ISA Trans 57:311–321 Zou D, Li S, Kong X, Ouyang H, Li Z (2019) Solving the combined
Shefaei A, Mohammadi-Ivatloo B (2017) Wild goats algorithm: an evo- heat and power economic dispatch problems by an improved
lutionary algorithm to solve the real-world optimization problems. genetic algorithm and a new constraint handling strategy. Appl
IEEE Trans Ind Inf 14:2951–2961 Energy 237:646–670
Song Y, Chou C, Stonham T (1999) Combined heat and power eco-
nomic dispatch by improved ant colony search algorithm. Electr Publisher’s Note Springer Nature remains neutral with regard to
Power Syst Res 52(2):115–121 jurisdictional claims in published maps and institutional affiliations.
Su CT, Chiang CL (2004) An incorporated algorithm for combined
heat and power economic dispatch. Electr Power Syst Res
69(2):187–195
Subbaraj P, Rengaraj R, Salivahanan S (2009) Enhancement of com-
bined heat and power economic dispatch using self adaptive real-
coded genetic algorithm. Appl Energy 86(6):915–921

13

You might also like