You are on page 1of 14

Expert Systems With Applications 152 (2020) 113370

Contents lists available at ScienceDirect

Expert Systems With Applications


journal homepage: www.elsevier.com/locate/eswa

An improved quantum particle swarm optimization algorithm for


environmental economic dispatch
Zhao Xin-gang a,b,∗, Liang Ji a,b, Meng Jin a,b, Zhou Ying a,b
a
School of Economics and Management, North China Electric Power University, Beijing, China
b
Beijing Key Laboratory of New Energy and Low-Carbon Development, Beijing, China

a r t i c l e i n f o a b s t r a c t

Article history: Consumption of traditional fossil energy has promoted rapid economic development and caused effects
Received 31 May 2019 such as climate warming and environmental degradation. In order to solve the problem of environmental
Revised 5 March 2020
economic dispatch (EED), this paper proposes a DE-CQPSO (Differential Evolution-Crossover Quantum Par-
Accepted 8 March 2020
ticle Swarm Optimization) algorithm based on the fast convergence of differential evolution algorithms
Available online 9 March 2020
and the particle diversity of crossover operators of genetic algorithms. In order to obtain better opti-
Keywords: mization results, a parameter adaptive control method is used to update the crossover probability. And
Environmental economic dispatch the problem of multi-objective optimization is solved by introducing a penalty factor. The experimental
Carbon emission reduction results show that: the evaluation index and convergence speed of the DE-CQPSO algorithm are better
Quantum particle swarm optimization than QPSO (Quantum Particle Swarm Optimization) and other algorithms, whether it is single-objective
Differential evolution operator optimization of fuel cost and emissions or multi-objective optimization considering both optimization
Crossover operator
objectives. A good compromise value is verified, which verifies the effectiveness and robustness of the
Adaptive control
DE-CQPSO algorithm in solving environmental economic dispatch problems. The study provides a new
research direction for solving environmental economic dispatch problems. At the same time, it provides
a reference for the reasonable output of the unit to a certain extent.
© 2020 Elsevier Ltd. All rights reserved.

1. Introduction In the past decades, many evolutionary algorithms and intel-


ligent algorithms used in multi-objective optimization have been
With the development of global economy and the increasing proposed to solve the EED problem, which has become an im-
demand for energy, people pay more and more attention to the co- portant subject of wide interest. The paper (Mahdi et al., 2018)
ordinated development between human and environment. Among are reviewed in this paper the application of all kinds of opti-
them, the pollution caused by thermal power generation accounts mization technology to deal with the problem of EED, this pa-
for the highest proportion, which causes more and more people per mainly introduces the genetic algorithm, particle swarm op-
to pay attention to the environmental pollution caused by pol- timization algorithm, the cuckoo search algorithm, the bat algo-
lutant emissions from thermal power plants. Environmental eco- rithm and some hybrid algorithms, also finds that the advanced
nomic dispatch (EED) problem of energy is an important optimiza- nature inspired method is the most suitable and the most success-
tion problem in power system operation. The ultimate goal of en- ful, and sums up the combined hybrid method is the most promis-
vironmental economic dispatch is to meet the system load require- ing way to solve the problem of EED. Genetic algorithm (GA) is
ments of all generator sets while meeting the requirements of all adopted to solve and analyze the environmental economic schedul-
unit and system loads with minimum operating costs and emission ing problem considering environmental pollution (Yasar & Ozyon,
costs. As for EED, it is necessary to satisfy both fuel cost and emis- 2012), and obtains the optimal solution of different weight val-
sion cost optimization while meeting unit operation constraints ues (pareto optimal solution) under the condition of power con-
(Basu, 2011). In addition, as EED is a multi-objective optimization, straint. Pareto optima obtained by different scaling methods are
it needs to reduce the emission cost while taking into account the compared. (Abido, 2003) used a niche pareto genetic algorithm
fuel cost, which makes EED optimization conflicted and requires a (NPGA) to find different pareto optimal solutions for solving EED
customized strategy to solve this conflict. problem. The optimal compromise solution is obtained on the bal-
ance curve between fuel cost and pollutant emission by fuzzy
mathematics. A reinforcement learning technique based on the pa-

Corresponding author. rameter self-tuning method, namely the reinforcement learning
E-mail address: rainman319@sina.com (Z. Xin-gang).

https://doi.org/10.1016/j.eswa.2020.113370
0957-4174/© 2020 Elsevier Ltd. All rights reserved.
2 Z. Xin-gang, L. Ji and M. Jin et al. / Expert Systems With Applications 152 (2020) 113370

based on non-dominated sorting genetic algorithm (NSGA - RL) thereby reducing the cost of environmental emissions. Tan and
(Bora & Mariani, 2019) is used to solve the problem of EED, con- Yu (2012) showed that the dual-objective optimization model can
sidering the quality of the resulting solution, the proposed algo- better take into accounting the unification of coal consumption and
rithm has the advantage of no time spent on parameter adjust- emission targets, and provided more accurate quantitative analysis
ment, and the experimental results show that the method for dis- basis for the formulation of power saving and emission reduction
tribution of good pareto frontier goal conflict has a better satis- policies.
faction. An improved partheno genetic algorithm (IPGA) based on Sun proposed the quantum particle swarm optimization al-
non - uniform mutation operator is proposed (Wang, Huang, Ma gorithm from the perspective of quantum mechanics (Clerc &
& Chen, 2015). IPGA introduces the idea of greedy selection into Kennedy, 2002). Subsequently, the algorithm entered a stage of
the non-uniform mutation operator to search the solution space rapid development and was applied in various fields. Zhang Z
uniformly in the initial stage and locally in the later stage, avoid- (2010) applied the quantum particle swarm optimization algorithm
ing random blind jump and staying in the promising solution re- to the economic load scheduling of power system, which effec-
gion. Pattanaik, Basu and Dash (2018) proposes an improved real- tively solved the economic load scheduling problem. However, this
coded genetic algorithm (IRCGA) to solve the dynamic economic algorithm also has some shortcomings, such as too fast decline
scheduling problem. The test function verifies the effectiveness of of diversity, reducing the performance of the algorithm to solve
the algorithm from the aspects of convergence speed and solu- complex multi-peak optimization problems, and insufficient abil-
tion quality. Shang et al. (2017) proposed a solution method of ity of the algorithm to jump out of local optimization at the later
economic load distribution problem (ELDP) based on genetic algo- stage. Davoodi, Hagh and Zadeh (2014) proposed a hybrid algo-
rithm (GA). The improved genetic algorithm can significantly re- rithm based on improved quantum particle swarm optimization
duce the inefficiencies of the current genetic algorithm and avoid (IQPSO) and simplex algorithm, and proposed a hybrid power flow
the operation of turbines in cavitation/vibration region, thus ensur- algorithm considering the reactive power limit of generating bus.
ing the safety of turbines in the process of power generation. Par- Simulation results showed that IQPSOS have better robustness and
ticle swarm optimization (PSO) algorithm is an important kind of convergence under normal and critical conditions when conven-
swarm intelligence (Eberhart & Kennedy, 1995; Kennedy & Eber- tional power flow method fails. Lu, Sun and Lu (2010) proposed
hart, 1995) random search algorithm, because of its simple con- an improved quantum behavior particle swarm optimization al-
cept, easy to implement (Eberhart & Shi, 2001; Hu, Shi & Eberhart, gorithm (QPSO) for short-term combined economic emission hy-
2004; Li & Engelbrecht, 2007), much attention has been paid to the drothermal scheduling (CEES) of hydrothermal power systems with
PSO algorithm. However, PSO algorithm is prone to precocity and multiple equality and inequality constraints. The improved algo-
cannot fully guarantee global convergence (Van den Bergh, 2002). rithm was compared with the algorithm with the same heuristic
Based on the above problems, hybrid particle swarm optimization strategy in terms of solving quality, robustness and convergence.
(HPSO) (Angeline, 1998) appears, and many related algorithm im- Simulation results showed that compared with other optimiza-
provements are applied to the optimization of EED problem. De- tion algorithms, this method could solve short-term hydrothermal
veloping efficient multi-objective evolutionary algorithms (MOEAs) scheduling problems stably and efficiently. The paper (Liu, Jiao, Ma,
provide an effective tool for solving the environmental economic Ma & Shang, 2016) introduced the cultural evolution mechanism
dispatch (EED) problem (Xiao, Zhou, Zhai & Wei, 2016). Wu, Wang, into the quantum particle swarm optimization (QPSO) to solve the
Yuan and Zhou (2010) applied the particle swarm optimization al- multi-objective environment/economic scheduling (EED) problem,
gorithm (PSO) to the dual-objective optimization problem of mini- and its effectiveness was obtained by comparing with several ad-
mizing fuel costs and pollutant emissions. And the well-distributed vanced algorithms. In other applications, the paper (Coelho, 2010)
pareto front is obtained. (Abido, 2009) proposed a new multi- applied the method of combining QPSO with gaussian probabil-
objective particle swarm optimization algorithm (MOPSO) to solve ity distribution to the continuous optimization of engineering de-
the problem of competitive and non-commensurable costs and the sign, and the results showed that the gaussian QPSO method can
speed of emission targets. Simulation results show that the algo- deal with this kind of problems in terms of the precision and the
rithm is superior in diversity and quality of pareto optimal re- convergence. In addition, chaos search and a collaborative quan-
sults. Cai, Ma, Li, Li and Peng (2009) proposed a multi-objective tum particle swarm optimization algorithm with learning behav-
chaotic particle swarm optimization (MOCPSO) method to solve ior were also proposed (Dong, Lu & Gong, 2014; Feng et al., 2015).
the EED problem. Compared with the traditional MOPSO with min- Simulation results showed that the improved algorithm had better
imal fuel cost and pollutant emission, this method has achieved stability and robustness.
good results in practical application. Hadji, Mahdad, Srairi and Pandit, Tripathi and Tapaswi (2012) used fuzzy sequencing
Mancer (2015) proposed a time varying acceleration based on par- to determine the best compromise between cost and emission
ticle swarm optimization. The performance of the standard particle targets, and proposed an improved bacterial foraging algorithm
swarm optimization algorithm was improved by dynamically ad- (IBFA), which adopted parameter automation strategy and cross
justing the acceleration coefficient during the process search and operation to improve the computational efficiency of the algo-
balancing the development and exploration capabilities. The exper- rithm. Elattar (2019) proposed a new shuffle frog-jump algo-
imental results showed that the proposed method was effective rithm (SFLA). Based on the original SFLA, the inertia equation
and robust in solving environmental/economic scheduling prob- of particle swarm optimization was introduced to improve the
lems. Zuo, Li, Li and Kong (2017) proposed a new global parti- local search mechanism. At the same time, the crossover op-
cle swarm optimization algorithm (NGPSO) to solve the problem erator and mutation operator of genetic algorithm are used to
of carbon emissions and cost optimization, and through the single improve the global search mechanism. Experimental results
objective optimization and multi-objective optimization (Agrawal, demonstrated the effectiveness and superiority of the proposed
Panigrahi & Tiwari, 2008) the experiment results proved that the algorithm. Firefly algorithm and bat algorithm are the latest al-
method had the strong convergence and stability, and to be able gorithm; They have proved their effectiveness in several research
to balance the need to minimize fuel costs and emissions and the fields (Gherbi, Bouzeboudja & Gherbi, 2016). This paper will use
need for power balance and generation constraints. Arandian and these two methods to optimize the production cost and NOx
Ardehail (2017) optimized the combination of renewable and non- emission, among which the transmission loss is variable and the
renewable energy CHP (Combined heat and power) technology by load is different. Then the two meta-heuristic algorithms are
introducing a mutation operator in the particle swarm algorithm, hybridized.
Z. Xin-gang, L. Ji and M. Jin et al. / Expert Systems With Applications 152 (2020) 113370 3

In order to achieve the goal of energy conservation, emission contributed to the existing knowledge system, because the im-
reduction and environmental protection, in addition to the above- proved quantum particle swarm optimization and parameter con-
mentioned EED research, domestic and foreign scholars also con- trol methods can provide better diversity and convergence, which
ducted research from the following three directions: provides a new research direction for solving EDD problems.
In this paper, I aim to use an improved quantum particle swarm
(1) Adopted the combination of renewable energy and tradi- optimization algorithm to solve the multi-objective optimization
tional thermal power generation. Current studies show that problem of EED. In view of the deficiency of the original quantum
wind power and coal-fired power generation (Huang et al., particle swarm algorithm in global search ability, this paper intro-
2017) and solar power and coal-fired power generation duces differential evolution operator to improve the global search
[Zhai, Liu, Li, Zhao & Yang, 2016; Zhu, Zhai, Qi, Yang & Reyes- ability of quantum particle swarm algorithm. At the same time, the
Belmonte, 2017) can effectively reduce carbon emissions. crossover operator of genetic algorithm is introduced in this pa-
(2) Use traditional carbon capture and storage technologies. per to ensure the diversity of QPSO particle population and main-
The paper (Han, Zhao, Zhang, Yu & Liao, 2017) discusses tain the overall vitality of particles, and at the same time to over-
the coal-fired power plant carbon capture technology (post- come the instability of QPSO algorithm convergence and the con-
combustion capture, former catcher, oxygen-enriched com- tingency of falling into local optimal under special circumstances.
bustion) of domestic and foreign research progress. There The value of crossover probability p has an important influence
are some papers (Chang, Wang, & Meng, 2012; Wang, Du, on the search ability and convergence speed of the algorithm. The
Zhao & Li, 2016; Zhang, Li & Wei, 2010) analyze the tra- smaller crossover probability enables the individuals in the group
ditional carbon capture and sequestration technologies of to retain more of their own information and maintain a higher
main technical links and problems, and a systematic method diversity of the group, which is conducive to the overall explo-
is used to quantify the changes of carbon capture and stor- ration of the algorithm. On the contrary, large crossover probability
age cost and uncertainty (Hu & Zhai, 2017), and shows that urges individuals to learn more empirical knowledge in the group,
when dealing with uncertainty, real options method is supe- thus accelerating the convergence speed of the algorithm. In order
rior to the traditional net present value method (Wang & Du, to get the appropriate p value, this paper proposes an improved
2016). Although the "carbon capture and storage" method crossover probability repair method based on the adaptive method.
can effectively reduce carbon emissions, it will also increase Both the contract-expansion factor is reduced linearly with the in-
costs (Singh, Rao & A., 2016; Yu, Zhang & Cheng, 2016). crease of the number of iterations.
(3) Through the allocation of carbon emission task, the optimal This paper is divided into five parts. In addition to this part,
power generation model based on coal consumption is es- the second part lists the EED’s objective optimization function,
tablished. The papers (Chen, Cheng, Song & Wu, 2016; Zhao, constraint conditions and the multi-objective function introduc-
Min, Geng & He, 2017) show that carbon distribution has a ing penalty factors. The third part describes the improvement of
significant impact on the pressure of policy makers, and in QPSO in detail. In the fourth part, through the data of the two
the carbon distribution, the production and distribution of cases, the experimental results of single-objective optimization and
power and heat should account for the largest proportion, multi-objective optimization are compared, and the convergence
and puts forward policy suggestions to improve its emission and effectiveness of the algorithm are analyzed. The fifth part will
reduction efficiency. summarize the conclusion of this paper. The specific architecture
is shown in Fig. 1.
In general, the research on the improvement of each algorithm
is not enough. Existing research focuses on only one aspect of the
algorithm; They do not provide a comprehensive evaluation. In
2. EED calculation model
a word, the previous studies are deficient in the following three
aspects. First, the evaluation index is not comprehensive enough
The goal of environmental economic dispatch is to minimize
to estimate the time efficiency, stability and accuracy of the al-
both fuel cost and emission cost, in this process, the optimization
gorithm. Secondly, the experimental conditions were inconsistent.
process needs to be constrained, the constraint conditions includ-
The performance step size, coding method, algorithm structure,
ing power balance constraints, unit operation up and down limit,
parameter selection and algorithm improvement operation of the
loading valve point effect, ramp rate limits and forbidden zone
algorithm are different. Different algorithm forms and calculation
and other constraints. First, the cost and emission are optimized
parameters will lead to different conclusions. In the end, most of
according to the single objective optimization formula given in
the multi-objective optimization in EED problem avoids operating
Section 2.1 and Section 2.2 respectively. Then, the penalty factor is
in the restricted area by means of penalty fitness function, but for
used to convert the two multi-objective optimizations of fuel cost
some systems, neither a quantitative penalty factor nor a variable
and emission into a single-objective optimization, and the optimal
penalty factor can be found to solve this problem, leading to the
compromise value between fuel cost and emission is obtained.
advanced convergence of the improved algorithm. In addition, in-
complete constraint conditions will also lead to errors in the re-
sults. This study aims to solve the following three problems: 2.1. The objective function
(1) Solve the problem of EED multi-objective optimization;
2.1.1. Fuel cost function
(2) How to improve QPSO to solve the problem of premature
The goal is to minimize the total fuel cost, which can be ex-
convergence and low diversity;
pressed as:
(3) Provide an improved adaptive method to control the update
of crossover probability. 
N 
N
FT = Fi (Pi ) = ai + bi Pi + ci Pi2 (1)
The literature review systematically describes various methods
i=1 i=1
for solving the EED problem. It can enable readers to understand
the working principles and improvement directions of different in- Where, Fi is the fuel cost of the unit, ai ,bi ,ci is the fuel cost co-
telligent evolution algorithms, which may bring some inspiration efficient, and Pi is the active power output of the unit i. FT is the
to the solution of the EED problem. In addition, this research has total fuel cost of N generator sets.
4 Z. Xin-gang, L. Ji and M. Jin et al. / Expert Systems With Applications 152 (2020) 113370

Fig. 1. Article architecture diagram.

2.1.2. Emission cost function In this case, PD is the total actual load demand, and PL is the
The goal is to minimize the total amount of pollution gas pro- system loss. In order to calculate the system loss, the method
duced in the process of energy use. These air pollutants mainly based on penalty factor and constant loss formula coefficient or
include sulfides, nitrides and carbon dioxide. The total amount of b coefficient (Feng, Liao, Niu, Shen, Cheng, & Li, 2015) is adopted.
air pollutants can be expressed as: The transmission loss of the system can be expressed as:

N  
N 
N 
N

N     PL = Pi Bi j Pj + Boi Pi + Boo (4)
ET = Ei (Pi ) = −2
10 × αi + βi Pi + γ i Pi
2
+ ξi exp (λi Pi )
i=1 j=1 i=1
i=1 i=1
(2) Here, Bij ,Boi and Boo are the loss coefficient of the generators.
(2) Unit output constraint
Where, Ei is the pollutant discharge amount of the unit i.
α i ,β i ,γ i , and ξ i are the pollutant emission coefficient of the unit Pimin ≤ Pi ≤ Pimax f or ∀i = 1 , 2 , . . . , N (5)
i, and Pi is the active output of the unit i.ET is the total emissions
generated by N generator sets. Here Pimin and Pimax respectively represent the minimum and
maximum output limits of the unit i.
2.2. Constraint condition (3) Unit climbing speed constraint

In the process of environmental and economic dispatch, due When the unit’s climbing speed limit is taken into account dur-
to the attributes of the unit itself, the problems in the opera- ing operation, the constraint can be expressed as:
tion process and in order to achieve the best optimization effect,
   
Max Pimin , Pio − DRi ≤ Pi ≤ Min Pimax , Pio − U Ri (6)
the optimization objectives need to be constrained. These con-
straints mainly include system power balance, unit output limit, DRi is the rate limit for the upper slope of the unit i;URi is the
unit climbing speed, valve point loading effect and other con- rate limit for the lower slope of the unit i;
straints. (4) Valve point loading effect
(1) Power balance constraint The objective function is discontinuous, non - convex and multi
- modal due to the valve-point effect. In this paper, a correction

N
Pi − (PD + PL ) = 0 (3) sine function is added into the cost function (1), which can be ex-
pressed as [3]:
i=1
Z. Xin-gang, L. Ji and M. Jin et al. / Expert Systems With Applications 152 (2020) 113370 5


N 
N     random, parallel optimization algorithm. The original PSO mathe-
FT = Fi (Pi ) = ai + bi Pi + ci Pi2 + ei × sin fi × Pimin − Pi  matical model is as follows:
i=1 i=1
(7) Vid (t + 1 ) = Vid (t ) + c1 r 1id (t )(Pid (t ) − Xid (t ) )
+ c2 r 2id (t )(Gd (t ) − Xid (t ) )Xid (t + 1 )
Here, ai ,bi and ci are the fuel cost coefficients, ei and fi are the
fuel cost coefficients of the unit i with valve-point effect. = Xid (t ) + Vid (t + 1 ) (13)

2.3. Introducing multi-objective optimization of penalty factors Where, Xid and Vid respectively represent the position and ve-
locity components of the particle i and the dimension d, |Vid | ≤
As for EED problem, it is a multi-objective optimization that Vmax ; Pid is the historical best position component experienced by
needs to optimize both fuel cost and emission cost, so it will the group; r1 and r2 are the random numbers with independent
be difficult. At present, domestic and foreign scholars usually and uniform distribution [0,1] and are called random factors. c1
use price penalty factor or fuzzy scheme to solve the multi- and c2 are the acceleration coefficients (Angeline, 1998).
objective optimization problem, and this paper will adopt the for- During the evolution process, particles in the population will
mer method. The price penalty factor attempts to assign the same take the weighted average position Sti = {sti1 , sti2 , . . . , stiD } of their
importance to the cost of fuel as to the cost of emissions. The price own historical optimal position Pit and the historical optimal po-
penalty factor can be defined as the ratio between the fuel cost sition of the population Pgt as the attraction point, and gradually
and emission cost of the unit i when the unit outputs the max- approach to this point (Zhang, 2010). The calculation formula of sti
imum power (Pimax ). In order to define the price penalty factor is as follows:
PF under the total load demand, the following steps are required c1 r1t ,id Pidt + c2 r2t ,id Pgd
t

(Feng, Liao, Niu, Shen, Cheng, & Li, 2015): stid = 1≤d≤D (14)
c1 r1t ,id + c2 r2t ,id
(1) The fuel cost and emission cost when the unit outputs the
maximum power; Because the settings of Vmax in the original PSO model are too
      dependent on the problem itself and there is no empirical rule,
Fi Pi max
= ai + bi Pimax + bi Pimax 2
+ ei × sin fi × Pimin − Pimax  when the settings of Vmax are not correct, the particles will lose
(8) the ability to transform from breadth search to depth search, so
that the trajectory of particles will often fail to converge. For the
     
Ei Pi max
= 10−2 × αi + βi Pimax + γi Pi max 2
+ ξi exp λi Pimax improvement of performance of particle swarm optimization algo-
rithm, Sun and others from the perspective of quantum mechanics,
(9)
proposed a quantum behaved particle swarm optimization [24].
(2) Calculate the price penalty factor of the unit i; Based on the above analysis of the motion of particles in the PSO
Fi (Pimax ) algorithm, QPSO algorithms assume that the particle swarm sys-
P Fi =
Ei (Pimax ) tem is a particle system satisfying the basic hypothesis of quantum
  mechanics, the Particle i moves in the δ potential well centered
ai + bi Pimax + ci Pimax 2 + ei × sin( fi × (Pimin − Pimax )) at point Stid in the d-th dimension with quantum basic behavior
=    
10−2 × αi + βi Pimax + γi Pimax 2 + ξi exp λi Pimax characteristics [26,27], and its state can be described by the wave
function  , namely
(10)
 

(3) Arrange the obtained values in ascending order and add   1 xt+1 − st 
their corresponding values successively until the total load ψ xt+1 =  ∗ exp − id id
(15)
id
Ltid Ltid
demand is met. The process can be expressed as:

N Where L is the characteristic length of the potential well. The
Pimax ≥ PD (11) probability density function of the particle i in dimension d is:
i=1  

(4) When the above conditions are met, the PF value associated   1 2xt+1
id
− stid 
Q xt+1 = t exp − (16)
with the last unit is the price penalty factor for a given to- id Lid t
Lid
tal load demand. After the price penalty factor is obtained,
multi-objective optimization can be converted into single- In order to obtain the position of the particle, the state of the
objective optimization, that is, the optimization of fuel cost particle needs to be collapsed from the quantum state to the clas-
and emission cost can be converted into the optimization of sical state, and the Monte Carlo method is used to measure the
the minimum value of the total load demand, which can be position of the particle [25], so as to obtain the updated equation
expressed as: of the particle position as follows:
N 
    
Ltid
min F (PD ) = ai + bi Pi + ci Pi2 + ei × sin fi × Pimin − Pi  xt+1 = stid ± ln
1
(17)
i=1
id 2 utid
N 
     
+P F (PD ) 10−2 × αi + βi Pi + γi Pi2 + ξi exp(λi Pi ) Ltid = 2α Cdt − xtid  (18)
i=1
(12) utid is a random number satisfying uniform distribution be-
tween (0,1); Where α is the contract-expansion coefficient; Ct =
3. DE-CQPSO based on quantum particle swarm optimization {C1t , C2t , . . . , CDt } is the average optimal position, represents the av-
erage value of the optimal position in the individual history of all
3.1. Quantum particle swarm optimization particles, and the expression is:
In 1995, Kennedy and Eberhart proposed Particle Swarm op- N
i=1 ptid
timization algorithm (PSO) (Kennedy, & Eberhart, 1995). PSO is a Cdt = (19)
N
6 Z. Xin-gang, L. Ji and M. Jin et al. / Expert Systems With Applications 152 (2020) 113370

3.2. Introduce differential evolution operator [0,1] to the second term gbestdt , and get the new evolution equation
as follows:

Particle attraction point in the QPSO algorithm, each particle   1
takes the weighted average position of the individual historical op- xt+1 = φ  δd + (1 − φ )  gbestdt ± α mbestd − xtid   ln (25)
id utid
timal position and the group historical optimal position as its at-
traction point. Guided by the analysis results of particle trajecto- From the above equation, it can be seen that the introduction of
ries (Gherbi, Bouzeboudja, & Gherbi, 2016), this calculation method differential evolution operator avoids the group moving in a small
has the advantage of simple calculation, but it also has two ob- range and falling into local optimization, which is conducive to im-
vious problems:  1 in addition to learning their own experience, proving the ability of global search.
each particle only takes the historical optimal position of the group
as the guide, which leads to the rapid decline of the diversity 3.3. Introduce crossover operator
of large groups and reduces the performance of the algorithm to
solve complex multi-peak optimization problems.  2 according to In quantum particle swarm optimizer, in order to increase the
attract point got from the type (14) will limit in D dimensional diversity of the population, improve the performance of algorithm
hyperrectangle Taking individual historical optimal position and for solving complex multi-peak optimization problems, Liu, Jiao,
the group historical optimal position as the apex, with the algo- Ma, Ma and Shang (2016) introduces chaos search to enhance the
rithm of evolution process, each particle attraction point’s possi- quality of initial population, through weighted update population
ble distribution space is gradually decreasing, making the particle optimal position center to improve population evolution model
attraction point gradually tends to group historical optimal posi- and improve convergence speed, uses neighborhood search varia-
tion, eventually leading to the ability of the algorithm to jump tion to increase species diversity to avoid premature convergence.
out of local optimization insufficient in the later stage. In order to Coelho (2010) proposes a collaborative quantum particle swarm
improve the ability of particle global search, QPSO algorithm can optimizer algorithm with learning behavior, by introducing multi-
be combined with differential mutation operation (Davoodi, Hagh, subgroup cooperative search strategy, the global search ability of
& Zadeh, 2014). Cultural evolution mechanism can also be intro- the population was improved, so that the diversity of the popula-
duced into the quantum behavioral particle swarm optimization al- tion is maintained in the later evolutionary stage. Secondly, parti-
gorithm, in which a particle will be measured many times in each cle learning behavior is given to improve the local search ability of
iteration to enhance its global search ability (Lu, Sun, & Lu, 2010). the population. In this paper, crossover operator is introduced into
In order to solve this problem, differential evolution operator is in- QPSO algorithm, which is derived from genetic algorithm (Wang,
troduced in this paper. Differential evolutionary algorithm is a kind Du, Zhao, & Li, 2016). Through cross-operation the information ex-
of evolutionary algorithm put forward by Rainer Storn and Kenneth change between individuals in a group can be carried out, and
price in 1995 based on population differences, initially it tried to those excellent genes can be retained gradually along with the
use vector of individual difference vector populations of mixing, in continuation of the evolutionary process, so that the group can
order to solve the problem of Chebyshev polynomial fitness, the evolve in a good direction.
principle of the algorithm is by competition and cooperation be- According to Eqs. (14), (18), (19) and (25), the measurement
tween individuals in the population to solve the optimization prob- position Xit+1 corresponding to particle i is generated; Then the
lem. The operators in the evolutionary process of this algorithm measurement position Xit+1 and the optimal position of the in-
include: mutation, crossover and selection. The initial population dividual position Pit are separated to generate the test position
of the differential evolutionary algorithm is generated randomly in Zit = {zti1 , zti2 , . . . , ztiD }, the cross formula is:
the search space, and is generally generated by using the uniformly 
distributed random function. xt+1
id
randd (0, 1 )< p or d = drand
zt+1
id
= (26)
The update formula of QPSO algorithm contains the following ptid other
three equations:
Where, randd (0,1) is a random number satisfying uniform distri-
Stid = φ  pbestid
t
+ (1 − φ )  gbestdt (20) bution between [0,1]; drand is the integer generated randomly and
uniformly on [1,D]; p is the crossover probability.
1
N Then, the optimal position of the particle’s individual history is
mbestd = pbestid (21) updated according to the following formula:
N
i=1     
zt+1
id
f zt+1
id
< f ptid
  pt+1 = (27)
1 id
xt+1 = stid ± α mbestd − xtid   ln (22) ptid other
id utid
Where, f(∗ ) is the adaptive value function.
φ is a random number between (0,1); sid is a random position The value of crossover probability p has an important influence
between pbest and gbest; Combine (14) and (16) to get: on the search ability and convergence speed of the algorithm. A
smaller crossover probability can enable individuals in a group to
    1
− gbestdt + gbestdt ± α mbestd − xtid   ln
retain more of their own information and maintain a higher diver-
xt+1 = φ  pbestid
t
id utid sity of the group, which is conducive to the global exploration of
the algorithm. On the contrary, higher crossover probability urges
(23)
individuals to learn more empirical knowledge in the group, thus
For any particle i in the current particle swarm, randomly select accelerating the convergence speed of the algorithm.
two particles different from particle i from the swarm, respectively
k and m, and i = k = m, the difference of position between parti- 3.4. Parameter settings
cles k and m can be obtained.
Different optimization problems and the same optimization
δ = xm − xk (24)
problem need to set different optimal parameters in different
Substitute Eq. (24) for ( pbestid
t − gbest t ) from Eq. (23), and in
d
stages of evolution. Therefore, it is difficult to select a p value suit-
order to increase the randomness, add a random number between able for all problems. Referring to the parameter adaptive method
Z. Xin-gang, L. Ji and M. Jin et al. / Expert Systems With Applications 152 (2020) 113370 7

(Chen, Cheng, Song, & Wu, 2016), the parameter p is directly en- Where, λti is a random number satisfying uniform distribution
coded into each particle in this paper to achieve adaptive control. between [0.9,1].
After extended encoding particle i in the population can be de- For the contract-expansion coefficient α , the method adopted
scribed as: in this paper is as follows: as the number of iterations increases, α
  decreases linearly, and the expression is:
Xit = xti,1 , xti,2 , . . . , xti,D , pti (28)
t
For each particle in the population, use the following rules to α = αmax − ∗ (αmax − αmin ) (32)
T
update the crossover probability:
Where, T is the maximum number of stated iterations.

rand (0, 1 ) rand (0, 1 ) < τ
pt+1
i
= (29)
pti other 3.5. Algorithm process

Where, τ is the update probability of the parameter p. 1. Set t = 0, initialize the current position Xi0 of each particle in
This paper proposes an improved crossover probability repair the particle swarm in the problem space, and collocate Pi0 =Xi0 ;
method. For ease of description (Zhao, Min, Geng, & He, 2017), Set other relevant parameters;
particle i is associated with an additional binary vector Mt+1
i
= 2. According to formula (21), calculate the average optimal posi-
{mt+1
i1
, mt+1
i2
, . . . , mt+1
iD
}, and tion of particle swarm;
 3. For each particle i (1≤i≤N) in the particle group, perform steps
1 randd (0, 1 ) < pt+1 or d = drand 4–7;
mt+1
id
= i
(30)
0 other 4. Introduce differential evolution operator, according to
D Eq. (25) to update particle position;
Make Yit+1 = 1
D i=1 mt+1
id
, if the influence of drand is ignored, 5. Introduce the crossover operator, obtain the measurement po-
Yit+1 follows the binomial distribution of parameters D and pt+1
i
, sition according to the updated position, and generate the test
namely Yit+1 ∼ B(D, pt+1
i
). Therefore, Yit+1 is the maximum likeli- position from Eq. (26);
hood estimate of the parameter pt+1 . The following formula is used 6. Evaluate the adaptive value of the dimension of each particle at
i
to evolve the value of the parameter p: the test position, and update the crossover probability accord-
       ing to Eq. (31);
λti Yit+1 + 1 − λti pt+1 f Zit+1 < f Pit 7. According to formula (27), update the individual optimal posi-
pt+1
i
= i
(31)
pti other tion of particles;

Fig. 2. DE-CQPSO algorithm flowchart.


8 Z. Xin-gang, L. Ji and M. Jin et al. / Expert Systems With Applications 152 (2020) 113370

Table 1
The unit operating parameters.

Cases Unit i Generation limits Fuel cost coefficients Emission coefficients

Pimin Pimax ai bi ci ei fi αi β γi ξ λi
The standard IEEE 1 0.05 0.5 10 200 100 – – 4.091 5.554 6.490 2.0e-4 2.857
30-bus system 2 0.05 0.6 10 150 120 – – 2.543 6.047 5.638 5.0e-4 3.333
with six generators 3 0.05 1.0 20 180 40 – – 4.258 5.094 4.586 1.0e-6 8.000
(PD = 2.834 p.u.) 4 0.05 1.2 10 100 60 – – 5.326 3.550 3.380 2.0e-3 2.000
5 0.05 1.0 20 180 40 – – 4.258 5.094 4.586 1.0e-6 8.000
6 0.05 0.6 10 150 100 – – 6.131 5.555 5.151 1.0e-5 6.667
The ten-unit 1 10 55 1000.403 40.5407 0.12951 33 0.0174 360.0012 3.9864 0.04702 0.25475 0.01234
system (PD = 2000 2 20 80 950.606 39.5804 0.10908 25 0.0178 350.0056 3.9524 0.04652 0.25475 0.01234
MW) 3 47 120 900.705 36.5104 0.12511 32 0.0162 330.0056 3.9023 0.04652 0.25163 0.01215
4 20 130 800.705 39.5104 0.12111 30 0.0168 330.0056 3.9023 0.04652 0.25163 0.01215
5 50 160 756.799 38.5390 0.15247 30 0.0148 13.8593 0.3277 0.00420 0.24970 0.01200
6 70 240 451.32 46.1592 0.10587 20 0.0163 13.8593 0.3277 0.00420 0.24970 0.01200
7 60 300 1243.531 38.3055 0.03546 20 0.0152 40.2669 0.5455 0.00680 0.24800 0.01290
8 70 340 1049.998 40.3965 0.02803 30 0.0128 40.2669 0.5455 0.00680 0.24990 0.01203
9 135 470 1658.569 36.3278 0.02111 60 0.0136 42.8955 0.5112 0.00460 0.25470 0.01234
10 150 470 1356.659 38.2704 0.01799 40 0.0141 42.8955 0.5112 0.00460 0.25470 0.01234

Table 2
Transmission loss coefficients.

Transmission loss coefficients

The standard IEEE B6 × 6 1 2 3 4 5 6 – – – –


30-bus system 1 0.1382 0.0299 0.0044 0.0022 0.001 0.0008 – – – –
with six-generators 2 0.0299 0.0487 0.0025 0.0004 0.0016 0.0041 – – – –
(PD = 2.834p.u.) 3 0.0044 0.0025 0.0182 0.007 0.0066 0.0066 – – – –
4 0.0022 0.0004 0.007 0.0137 0.005 0.0033 – – – –
5 0.001 0.0016 0.0066 0.005 0.0109 0.0005 – – – –
6 0.0008 0.0041 0.0066 0.0033 0.0005 0.0244 – – – –
B0 0.0107 0.0060 0.0017 0.0009 0.0002 0.0030 – – – –
B00 9.8573e-4 – – – – – – – – –
The-ten-unit B10 × 10 1 2 3 4 5 6 7 8 9 10
system 1 4.9e–5 1.4e–5 1.5e–5 1.5e-5 1.6e-5 1.7e–5 1.7e–5 1.8e–5 1.9e–5 2.0e–5
(PD = 2000MW) 2 1.4e–5 4.5e–5 1.6e–5 1.6e-5 1.7e-5 1.5e-5 1.5e-5 1.6e–5 1.8e–5 1.8e–5
3 1.5e–5 1.6e–5 3.9e–5 1.0e–5 1.2e–5 1.2e–5 1.4e–5 1.4e–5 1.6e–5 1.6e–5
4 1.5e–5 1.6e–5 1.0e–5 4.0e–5 1.4e–5 1.0e–5 1.1e–5 1.2e–5 1.4e–5 1.5e–5
5 1.6e–5 1.7e–5 1.2e–5 1.4e–5 3.5e–5 1.1e–5 1.3e–5 1.3e–5 1.5e–5 1.6e–5
6 1.7e–5 1.5e–5 1.2e–5 1.0e–5 1.1e–5 3.6e–5 1.2e–5 1.2e–5 1.4e–5 1.5e–5
7 1.7e–5 1.5e–5 1.4e–5 1.1e–5 1.3e–5 1.2e–5 3.8e–5 1.6e–5 1.6e–5 1.8e–5
8 1.8e–5 1.6e–5 1.4e–5 1.2e–5 1.3e–5 1.2e–5 1.6e–5 4.0e–5 1.5e–5 1.6e–5
9 1.9e–5 1.8e–5 1.6e–5 1.4e–5 1.5e–5 1.4e–5 1.6e–5 1.5e–5 4.2e–5 1.9e–5
10 2.0e–5 1.8e–5 1.6e–5 1.5e–5 1.6e–5 1.5e–5 1.8e–5 1.6e–5 1.9e–5 4.4e–5
B0 0 0 0 0 0 0 0 0 0 0
B00 0 – – – – – – – – –

8. For particle i, compare the adaptive value of Pit with the adap- Table 3
The unit ramp rate limit.
tive value of the best global location, if the adaptive value is
superior to Gt-1 , namely f [ pti ] < f [Gt−1 ], then set Gt = Pit ; Oth- Unit i DR. UR
erwise, Gt = Gt-1 ; The standard IEEE 1 0.08 0.08
9. If the termination condition of the algorithm is not satisfied, 30-bus system 2 0.11 0.11
set t = t + 1 and return to step 2; Otherwise the algorithm with six-generators 3 0.15 0.15
(PD = 2.834p.u.) 4 0.18 0.18
ends.
5 0.15 0.15
6 0.18 0.18
The detailed flowchart of DE-CQPSO algorithm is shown in
The-ten-unit 1 30 30
Fig. 2. system 2 30 30
(PD = 2000MW) 3 30 30
4. Simulation results 4 30 30
5 50 50
6 50 50
4.1. The experiment case 7 50 50
8 80 80
This paper will use two cases to verify the effectiveness of DE- 9 80 80
CQPSO algorithm, as follows: 10 80 80

(1) standard IEEE 30 bus system, with 6 generators


(PD = 2.834p.u.) (Agrawal, Panigrahi, & Tiwari, 2008); 4.2. Compare different approaches to solve EED’s problem
(2) 10-unit system (PD = 20 0 0MW) (Mahdi, Vasant, Kallimani,
Watada, Siew Fai, & Abdullah-Al-Wadud, 2018). To verify the validity of the DE-CQPSO algorithm, this paper se-
lect the original QPSO algorithm, GQPSO (Coelho, 2010) and NGPSO
Relevant data of the two cases are shown in Tables 1–3. (Zuo, Li, Li, & Kong, 2017) algorithm to carry out single-objective
Z. Xin-gang, L. Ji and M. Jin et al. / Expert Systems With Applications 152 (2020) 113370 9

Table 4
Parameter table.

Algorithm Different parameters Population size Iterations times Running times

QPSO α max = 1; α min = 0.05; D = 6; 40 200 50


GQPSO [30] α max = 0.7; α min = 0.01; 40 200 50
NGPSO [21] μ = 3; pr = 0.01 40 200 50
DE-CQPSO α max = 0.7; α min = 0.1; p = 0.9; 40 200 50

Table 5
Comparison among three improved PSOs for the Fuel cost minimizations of two EED problems.

Problems Algorithms Min Max Mean Std

Case 1: The standard IEEE QPSO 614.12941352 619.75370853 617.07361762 1.86


30-bus system with GQPSO 606.38037047 611.86970507 609.49864973 1.18
six-generators NGPSO 605.99836962 605.99836862 605.99836962 0
(PD = 2.834p.u.) DE-CQPSO 605.91695121 605.97170162 605.93671317 2.36e-5
Case 2: The ten-unit QPSO 119,005.30386535 122,144.84545171 121,621.75567906 3.72e + 2
system (PD = 2000MW) GQPSO 112,429.74442183 113,327.0680907 113,102.462755 2.56e + 2
NGPSO 111,497.63081044 111,497.63081097 111,497.63081047 1e-7
DE-CQPSO 110,931.85331921 111,159.46175027 111,231.9603513 2.56e-8

Fig. 3. Comparison of minimum and average values of several algorithms for the fuel cost optimization.

optimization comparison of coal burning cost and emission cost 4.2.1. Parameter setting and running environment
for the data of case 1 and case 2, introduces the penalty factor The main parameters of the four algorithms are set as shown
of multi-objective optimization, and will use the optimal value, in Table 4. For detailed settings of each algorithm, please refer to
the mean, square error three indicators to measure the optimiza- references.
tion effect; In the second part, the convergence of the four al- Where, α max and α min respectively represent the maximum
gorithms is compared to further prove the effect of DE-CQPSO contraction expansion coefficient and the minimum contraction
algorithm. expansion coefficient; For NGPSO, μ and pr respectively represent
10 Z. Xin-gang, L. Ji and M. Jin et al. / Expert Systems With Applications 152 (2020) 113370

Fig. 4. Comparison of minimum and average values of several algorithms for emission optimization.

Table 6
Comparison among three improved PSOs for the emission minimizations of two EED problems.

Problems Algorithms Min Max Mean Std

Case 1: The standard IEEE QPSO 0.19684221 0.19731539 0.19714182 1.32e−4


30-bus system with GQPSO 0.19422159 0.19460562 0.19445065 8.7e−5
six-generators NGPSO 0.19417851 0.19417851 0.19417851 0
(PD = 2.834p.u.) DE-CQPSO 0.19418201 0.19420138 0.19419167 1e−7
Case 2: The ten-unit QPSO 4032.38753214 4058.36159124 4041.91714513 8.06
system (PD = 2000MW) GQPSO 4011.92441621 4042.18786944 4032.93203734 7.55
NGPSO 3932.24326915 3932.2432704 3932.24326924 2.1e−7
DE-CQPSO 3931.52924107 3931.7157089 3931.57612561 1.83e−7

the normal number and the randomization probability; p is the ini- rithm. In case 1, the standard deviation of NGPSO algorithm is bet-
tial probability of the crossover probability. ter than the other three algorithms, indicating that it is the most
The experimental test environment is set as follows: the oper- stable, while the stability of DE-CQPSO algorithm is second, bet-
ating system is Windows10 (64-bit), the processor is Intel (R) i7- ter than the stability of QPSO algorithm and GQPSO algorithm. In
6700HQ, the memory is 8GB, the programming language is Matlab, case 2, the optimization effect and stability of DE-CQPSO algorithm
and the compiler is Matlab 2018b. are better than the other three algorithms, and through the com-
parison of standard deviations, it can be concluded that DE-CQPSO
4.2.2. Single objective optimization contrast algorithm has better stability than the other three algorithms.
In this paper, the single objective optimization comparison It can be concluded from Table 6 and the Fig. 4 that, the op-
of coal burning cost and emission cost of the four algorithms timization effect of DE-CQPSO algorithm is better than the other
is compared, and the optimization effect is measured and com-
pared by three indexes of optimization maximum value, minimum
value, mean value and standard deviation respectively, as shown in Table 7
The best compromise solutions for the fuel cost and emission minimizations of the
Table 5.
standard IEEE 30-bus system with six-generators.
Table 5 compares the optimization results of the original QPSO
algorithm and the three improved algorithms for fuel cost and The evaluation index QPSO MOPSO NGPSO DE-CQPSO
emissions with the goal of minimizing,and the minimum, mean, C 637.48 626.10 623.87 619.53
and standard deviation are compared more specifically in Fig. 3. E 0.2418 0.2106 0.1969727 0.1964
Through the comparison of fuel cost optimization, it can be con- ρ 0.941057 0.979918 0.997527 –
θ% −0.21592 −0.07792 −0.00986 –
cluded that the optimization effect of three improved algorithms F 1546.6338 1551.4974 1489.3882 1496.1093
including DE-CQPSO algorithm is better than that of QPSO algo-
Z. Xin-gang, L. Ji and M. Jin et al. / Expert Systems With Applications 152 (2020) 113370 11

Fig. 5. The best compromise solutions for the fuel cost and emission in case 1.

fect and stability of DE-CQPSO algorithm are better than the other
three algorithms through the comparison of several indicators. The
best compromise solutions for the fuel cost

4.2.3. Multi-objective optimization for EED problem


Table 7 provides the best compromise solution of several al-
gorithms for multi-objective optimization in Case 1, and makes a
more specific comparison through Figs. 5 and 6. It can be seen
from the figure that DE-CQPSO is completely superior to QPSO,
MOPSO and NGPSO, and its fuel cost and emission are lower than
other methods. Fuel costs are 2.8% lower than QPSO, and emis-
sions are 18.77% lower than QPSO. Moreover, the multi-objective
optimization value is better than QPSO and MOPSO, and slightly
worse than NGPSO, which proves the effectiveness of DE-CQPSO.
Fig. 6. The optimal value of multi-objective optimization in case 1.
However, as the target value of multi-objective optimization is
larger than one algorithm and smaller than another algorithm, it
is difficult to directly determine whether DE-CQPSO is better than
three algorithms in case 1. On the other hand, the standard devia- other methods. Therefore, we introduce two indicators in the text
tion of DE-CQPSO is higher than that of NGPSO algorithm, its sta- to evaluate the advantages and disadvantages of the algorithm,
bility is not as good as that of NGPSO algorithm, but better than which are respectively standardized terms ρ and total increment
that of QPSO and GQPSO algorithm. In case 2, the optimization ef- percentage θ . More specifically, ρ and θ are expressed as follows

Fig. 7. The best compromise solutions for the fuel cost and emission in case 2.

Table 8
The best compromise solutions for the fuel cost and emission minimizations of the ten-
unit system.

The evaluation index QPSO MODE [1] NGPSO [21] DE-CQPSO

C 1.1946e+5 1.1348e+5 116,179.6487 1.1621e+5


E 4032.4625 4124.9 3939.2278 3936.3721
ρ 0.987076 0.994249 0.999884 –
θ% −0.05103 −0.02165 −0.00046 –
F 221,097.63 218,181.5 216,170.54 217,361.18
12 Z. Xin-gang, L. Ji and M. Jin et al. / Expert Systems With Applications 152 (2020) 113370

Fig. 8. The optimal value of multi-objective optimization in case 2.

Fig. 10. Comparison of convergence in case 2.

through DE-CQPSO optimization; FOther andEOther respectively repre-


sent the fuel cost and emission of the best compromise solution
obtained by other algorithms. As can be seen from Table 7, all ρ
values are less than 1. According to the normalized equation, DE-
CQPSO is the optimal method. In addition, θ being negative values
indicates that the DE-CQPSO solution achieves the decrease in the
overall goal of other solutions. Therefore, in terms of total incre-
mental percentage, DE-CQPSO performs best in reducing both fuel
costs and emissions in Case 1.
The same result can be obtained from Table 8 and Fig. 8, DE-
CQPSO is completely superior to QPSO, MOPSO and NGPSO. all ρ
values are less than 1. According to the normalized equation, DE-
CQPSO is the optimal method. In addition, θ being negative values
indicates that the DE-CQPSO solution achieves the decrease in the
overall goal of other solutions. Therefore, in terms of total incre-
Fig. 9. Comparison of convergence in case 1. mental percentage, DE-CQPSO performs best in reducing both fuel
costs and emissions in Case 2.

(Zuo, Li, Li, & Kong 2017):


4.3. Convergence efficiency analysis
FDE−CQPSO EDE−CQPSO
ρ= + (33)
(FDE−CQPSO + FOther ) (EDE−CQPSO + EOther ) This part mainly compares the convergence of four algorithms,
namely QPSO, GQPSO, NGPSO and DE-CQPSO, to further prove the
(FDE−CQPSO − FOther ) (EDE−CQPSO − EOther )
θ= + (34) effect of DE-CQPSO algorithm.
FOther EOther
The FDE − CQPSO andEDE − CQPSO here respectively represent the Contrastive analysis of convergence in Case 1:
fuel cost and emission of the optimal compromise value obtained Contrastive analysis of convergence in Case 2:
Z. Xin-gang, L. Ji and M. Jin et al. / Expert Systems With Applications 152 (2020) 113370 13

According to the analysis in Figs. 9 and 10, except that the con- References
vergence rate of the emission optimization in Case 1 is slightly
lower than that of NGPSO. In other cases, the convergence trend of Abido, M. A. (2003). A niched pareto genetic algorithm multi-objective environmen-
tal / economic dispatch. International Journal Electrical. Power & Energy System,
DE-CQPSO is faster and more stable than that of QPSO and GQPSO, 25(2), 97–105.
and the final optimization result is better. Abido, M. A. (2009). Multi-objective particle swarm optimization for environmental
From the evaluation of optimization results and the analysis / economic dispatch problem. Electric Power System Resrarch, 79(7), 1105–1113.
Agrawal, S., Panigrahi, B. K., & Tiwari, M. K. (2008). Multi-objective particle swarm
and comparison of convergence, it can be concluded that the DE- algorithm with fuzzy clustering for electrical power dispatch. IEEE Transactions
CQPSO algorithm has high accuracy, good stability and strong ro- on Evolution Computation, 12(5), 529–541.
bustness, thus confirming the practicability of DE-CQPSO algo- Angeline, P. J. (1998). Using selection to improve-particle-swarm optimization.
IEEE-International conference-on evolutionary computation.
rithm.
Arandian, B., & Ardehail, M. (2017). Effects of environmental emissions on optimal
combination and allocation of renewable-and non-renewable-CHP technologies
in heat and electricity distribution networks based on improved particle-swarm
5. Conclusion optimization algorithm. Energy, 140, 466–480.
Basu, M. (2011). Economic environmental dispatch using multi-objective differential
This paper studies the performance of DE-CQPSO algorithm, and evolution. Applied Soft Computer, 11, 2845–2853.
Bora, T. C., & Mariani, V. C. (2019). Multi-objective-optimization of the-en-
improve the comprehensive index of solving EED problem from vironmental-economic dispatch with reinforcement learning based
fuel cost and pollution emission cost. This paper adds differential on non-dominated sorting genetic algorithm. Applied Thermal Engineering,
evolution operator and crossover operator on the basis of QPSO al- 146, 688–700.
Cai, J., Ma, X., Li, Q., Li, L., & Peng, H. (2009). A multi-objective chaotic parti-
gorithm to improve the global search ability and particle diversity cle swarm optimization for environmental/economic dispatch. Energy Conversion
of QPSO algorithm, enhances the ability of particles to jump out of and Management, 50(5), 1318–1325.
local optimization, and overcomes the instability of convergence of Chang, K., Wang, W. H., & Meng, Z. (2012). Investment evaluation of carbon cap-
ture and storage technology in coal-fired power plants based on compound real
QPSO algorithm under special circumstances and the contingency options. Scientific Management Research, 32(24), 39–42.
of falling into local optimization. Meanwhile, the conflict between Chen, J. D., Cheng, S. L., Song, M. L., & Wu, Y. Y. (2016). A carbon emissions reduc-
cost minimization and emission minimization is alleviated by in- tion index, integrating the-volume-and allocation of regional emissions. Applied
Energy, 184, 1154–1164.
troducing price penalty factor. The simulation results of two cases
Clerc, M., & Kennedy, J. (2002). The-particle-swarm-explosion, stability, and conver-
show that: both single objective optimization on costs, emissions gence in a multidimensional complex space. IEEE-Transactions on Evolutionary
and introducing the penalty factor of multi-objective optimization Computation, 6(1), 58–73.
Coelho, L. S. (2010). Gaussian quantum-behaved particle swarm optimization ap-
dispatch, the optimal compromise result and convergence speed
proaches for constrained engineering design problems. Expert System Applied,
of DE-CQPSO algorithm are better than those of QPSO algorithm 37(2), 1676–1683.
and other algorithms, proves that the improved quantum particle Davoodi, E., Hagh, M. T., & Zadeh, S. G. (2014). A hybrid improved quantum-behaved
swarm optimization algorithm of pareto optimal front has good particle-swarm optimization–simplex method (IQPSOS) to solve-power system
load flow problems. Applied Soft Computing, 21, 171–179.
precision and distribution. In conclusion, DE-CQPSO algorithm is Dong, H. S., Lu, P., & Gong, S. R. (2014). Cooperative quantum particle swarm opti-
an effective method to solve the environmental economic dispatch mization algorithm with learning behavior. Computer Application Research, 31(9),
problem. Due to the strong convergence and stability of DE-CQPSO, 2588–2591.
Eberhart, R. C., & Kennedy, J. (1995). A new optimizer using particle swarm theory.
this algorithm is expected to be effectively applied to other eco- In Proceedings of the sixth international symposium on micro machine and human
nomic dispatch problems, such as the combined scheduling of re- science (pp. 39–43).
newable energy and thermal power, hot spot co-generation, wind Eberhart, R. C., & Shi, Y. H. (2001). Particle swarm optimization: Developments, ap-
plications and resources. In Proceedings of the 2001 Congress on Evolutionary
power and carbon capture power plant, etc. Computation (pp. 81–86).
In addition, due to the large span of demand data in the study Elattar, E. (2019). Environmental economic dispatch with heat optimization in
of periodic dispatch problem and the volatility of data, this pa- the-presence-of renewable-energy based on modified shuffle-frog leaping algo-
rithm. Energy, 171, 256–269.
per carried out experiments under the condition of single load de-
Feng, Z. K., Liao, S. L., Niu, W. J., Shen, J. J., Cheng, C. T., & Li, Z. H. (2015). Appli-
mand, and did not study the periodic generation load dispatch of cation of improved quantum particle swarm optimization algorithm in optimal
unit output. The next research direction is not only how to better operation of hydropower stations. Progress in Water Science, 26(3), 413–421.
Gherbi, Y. A., Bouzeboudja, H., & Gherbi, F. Z. (2016). The-combined economic envi-
adjust the parameters of the algorithm, but also the periodic envi-
ronmental dispatch using new hybrid metaheuristic. Energy, 115, 468–477.
ronmental economic dispatch problem. Hadji, B., Mahdad, B., Srairi, K., & Mancer, N. (2015). Multi-objective-PSO-TVAC for
environmental/economic dispatch problem. Energy Procedia, 74, 102–111.
Han, T., Zhao, R., Zhang, S., Yu, X. H., & Liao, H. Y. (2017). Research and application of
Declaration of Competing Interests carbon dioxide capture technology in coal-fired power plants. Coal Engineering,
49(5), 24–28.
Hu, B., & Zhai, H. (2017). The-cost of carbon capture-and storage-for coal-fired
The authors declared that they have no conflicts of interest to power plants in China. International Journal of Greenhouse-Gas Control, 65,
this work. 23–31.
Hu, X. H., Shi, Y. H., & Eberhart, R. C. (2004). Recent advances in particle swarm. In
Proceedings of the congress on evolutionary computation (pp. 90–97).
Credit authorship contribution statement Huang, X. Z., Sun, Y., Xu, Z. F., Xue, T. L., Wang, Z., & Cai, H. Y. (2017). Techno-e-
conomic performance of wind and coal-fired power with CCS joint planning.
Energy Procedia, 114, 6677–6684.
Zhao Xin-gang: Conceptualization, Methodology, Validation, Kennedy, J., & Eberhart, R. (1995). Particle-Swarm optimization. In Proceeding of
Writing - original draft, Writing - review & editing. Liang Ji: IEEE-international conference-on neural network (pp. 1942–1948).
Li, X. D., & Engelbrecht, A. P. (2007). Particle swarm optimization: An introduction
Methodology, Validation, Formal analysis, Data curation, Validation.
and its recent developments. In Proceedings of the 9th annual conference com-
Meng Jin: Software, Validation, Formal analysis, Visualization, Data panion on genetic and evolutionary computation (pp. 3391–3414).
curation. Zhou Ying: Formal analysis, Data curation. Liu, T., Jiao, L., Ma, W. P., Ma, J. J., & Shang, R. H. (2016). Cultural quantum-behaved
particle-swarm optimization for environmental/economic dispatch. Applied Soft
Computing, 48, 597–611.
Acknowledgement Lu, S., Sun, C., & Lu, Z. (2010). An improved quantum-behaved particle-swarm op-
timization method for short-term combined economic emission hydrothermal
scheduling. Energy Conversion and Management, 51, 561–571.
This paper is supported by the Beijing Municipal Social Science Mahdi, F. P., Vasant, P., Kallimani, V., Watada, J., Siew Fai, P. Y., & Abdul-
Foundation (No. 16JDYJB031), and 2017 Special Project of Cultiva- lah-Al-Wadud, M. (2018). A holistic review on optimization strategies for com-
bined economic emission dispatch problem. Renewable-and Sustainable-Energy
tion and Development of Innovation Base (No. Z17110 0 0 02217024). Reviews, 81, 3006–3020.
14 Z. Xin-gang, L. Ji and M. Jin et al. / Expert Systems With Applications 152 (2020) 113370

Pandit, N., Tripathi, A., & Tapaswi, S. (2012). An improved bacterial foraging algo- Xiao, J. M., Zhou, Q., Zhai, B. Y., & Wei, X. H. (2016). Multi-objective evolutionary al-
rithm for combined static/dynamic environmental economic dispatch. Applied gorithm and its application in power environmental and economic dispatching.
Soft Computing, 12(11), 3500–3513. Journal of Zhengzhou University (Engineering Science Edition), 37(2), 1–8.
Pattanaik, J. K., Basu, M., & Dash, D. P. (2018). Improved real coded genetic algo- Yasar, C., & Ozyon, S. (2012). Solution to scalarized environmental economic power
rithm for dynamic economic dispatch. Journal of Electrical Systems and Informa- dispatch problem by using genetic algorithm. International Journal of Electrical
tion Technology, 5, 349–362. Power & Energy Systems, 38(1), 54–62.
Shang, Y. Z., Lu, S. B., Gong, J. G., Liu, R. H., Li, X., & Fan, Q. X. (2017). Improved ge- Yu, S., Zhang, J., & Cheng, J. (2016). Carbon reduction cost estimating of Chinese–
netic algorithm for economic load dispatch in hydropower plants and compre- coal-fired power generation units, a perspective-from national energy consump-
hensive-performance-comparison with dynamic programming method. Journal tion standard. Journal of Clean Production, 139, 612–621.
of Hydrology, 554, 306–316. Zhai, R., Liu, H., Li, C., Zhao, M. M., & Yang, Y. P. (2016). Analysis of a solar-aided
Singh, U., & Rao, B. A. (2016). Techno-economic assessment of carbon mitigation op- coal-fired power generation system based on thermo-economic structural the-
tions for existing coal-fired power plants in India. Energy Procedia, 90, 326–335. ory. Energy, 102, 375–387.
Tan, Z. F., & Yu, C. (2012). Fuzzy optimization model for power distribution of coal– Zhang, H. X., Li, X. C., & Wei, N. (2010). Analysis of main technical links and prob-
fired units under the target of energy conservation and emission reduction. lems in carbon dioxide capture and storage. Progress in Earth Science, 25(3),
Power Grid Technology, 36(1), 219–223. 335–340.
Van den Bergh, F. (2002). An analysis of particle swarm optimizers. Hatfield, South Zhang, Z. (2010). Quantum-behaved particle-swarm optimization algorithm for eco-
Africa: University of Pretoria. nomic load dispatch of power system. Expert Systems with Applications, 37,
Wang, J., Huang, W., Ma, G., & Chen, S. J. (2015). An improved partheno genetic algo- 1800–1803.
rithm for multi-objective economic dispatch in cascaded hydropower systems. Zhao, R., Min, N., Geng, Y., & He, Y. L. (2017). Allocation of carbon emissions among
International Journal of Electrical Power & Energy Systems, 67, 591–597. industries/sectors, an emissions intensity reduction constrained approach. Jour-
Wang, X. P., Du, L., Zhao, S. J., & Li, Z. W. (2016). Research on carbon capture and nal of Cleaner Production, 142, 3083–3094.
storage investment decision of coal-fired power plants based on compound real Zhu, Y., Zhai, R., Qi, J. W., Yang, Y. P., & Reyes-Belmonte, M. A. (2017). Annual per-
option. China Power, 49(7), 179–184. formance of solar tower aided coal-fired power generation system annual per-
Wang, X., & Du, L. (2016). Study on carbon capture and storage (CCS) investment de- formance of solar tower aided coal-fired power generation system. Energy, 119,
cision making based on real options for China’s coal-fired power plants. Journal 662–674.
of Clean Production, 112, 4123–4131. Zuo, D., Li, S., Li, Z. Y., & Kong, X. Y. (2017). A new global particle-swarm optimiza-
Wu, L. H., Wang, Y. N., Yuan, X. F., & Zhou, S. W. (2010). Environmental/economic tion for the-economic emission dispatch with or without transmission losses.
power dispatch problem using multi-objective differential evolution algorithm. Energy Conversion and Management, 139, 45–70.
Electric Power Systems Research, 80(9), 1171–1181.

You might also like