Received October 9, 2019, accepted October 22, 2019, date of publication October 25, 2019, date of current version

November 6, 2019.
Digital Object Identifier 10.1109/ACCESS.2019.2949582

Optimal Operation of Hydropower System by

Improved Grey Wolf Optimizer Based on Elite
Mutation and Quasi-Oppositional Learning
ZHONG-KAI FENG 1 , SHUAI LIU1 , WEN-JING NIU2 , YI LIU1 ,
BIN LUO3 , SHU-MIN MIAO4 , AND SEN WANG5
1 School of Hydropower and Information Engineering, Huazhong University of Science and Technology, Wuhan 430074, China
2 Bureau of Hydrology, Changjiang Water Resources Commission, Wuhan 430010, China
3 Tsinghua Sichuan Energy Internet Research Institute, Chengdu 610200, China
4 State Grid Sichuan Electric Power Research Institute, Chengdu 610072, China
5 Key Laboratory of the Pearl River Estuarine Dynamics and Associated Process Regulation, Ministry of Water Resources, Guangzhou 510611, China

Corresponding author: Zhong-kai Feng ([email protected])

This work was supported in part by the Natural Science Foundation of Hubei Province under Grant 2018CFB573, in part by the National
Natural Science Foundation of China under Grant 51709119 and Grant 51909133, in part by the National Key Research and Development
Program of China under Grant 2016YFC0402210, and in part by the Fundamental Research Funds for the Central Universities (HUST)
under Grant 2017KFYXJJ192 and Grant 2017KFYXJJ193.

ABSTRACT As one of the most important renewable energy, hydropower is often asked to satisfy the load
demand of power system at peak periods. Thus, the optimal operation of hydropower system is modelled to
minimize the standard deviation of the residual load series obtained by subtracting the total power outputs
of all the involved hydropower plants from the original load curve. Hence, this paper develops an improved
grey wolf optimizer (IGWO) to effectively address the complex constrained optimization problem. In the
proposed method, the quasi-oppositional learning is used to enhance the convergence rate of the swarm;
the elite mutation operator is used to increase the probability of escaping from local optima; the elastic-ball
strategy and heuristic constraint handling method are used to help infeasible individuals rebound to feasible
space. Numerical experiments of 12 classical test functions demonstrate the feasibility of the IGWO method
in the global optimization problems. Then, the developed method is applied to solve the optimal operation
of two cascade hydropower systems. The results indicate that the proposed method outperforms several
traditional methods in smoothing the peak loads of power system. To sum up, an effective solution tool is
provided for the hydropower system operation optimization problem.

INDEX TERMS Hydropower system, grey wolf optimizer, quasi-oppositional learning, elite mutation,
elastic-ball strategy, constraint handling method.

I. INTRODUCTION guaranteeing the real-time dynamic balance between energy

Along with the growing load demand all over the world, supply and load demand [8]–[10]. However, due to the rapid
hydropower is attracting more and more attention due to expansion rate and system scale, the optimal operation of
the merits of low environmental pollution and high opera- hydropower system has become a challenging optimization
tional flexibility [1]–[4]. In recent years, many hydropower work for operator in power system [11].
bases are put into production and more hydropower reservoirs Generally, the goal of hydropower operation is to find
are in the pipeline [5]–[7]. In China, by the end of 2018, out the feasible scheduling decisions (like water level or
hydropower has provided about 20% power generation and total outflow) to achieve the best operation benefit (like
installed capacity to support the healthy and orderly devel- generation benefit or peak operation) while satisfying vari-
opment of economy. As one of the most important renew- ous kinds of physical constraints (like water balance equa-
able energies, hydropower is playing an irreplaceable role in tion or boundary constraints) [12]. Mathematically, the
optimal operation of hydropower system is a typical high-
The associate editor coordinating the review of this manuscript and dimensional, multiple coupling constraints, nonlinear opti-
approving it for publication was Yang Li . mization problem [13]. In the past decades, many famous

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see http://creativecommons.org/licenses/by/4.0/
VOLUME 7, 2019 155513
 Z.-K. Feng et al.: Optimal Operation of Hydropower System by IGWO Based on Elite Mutation and Quasi-Oppositional Learning

methods have been used to resolve this type of problem, like

linear programming [13], nonlinear programming [17]–[19],
dynamic programming [20]–[22], fuzzy methods [23]-[25],
network optimization [26]–[28], decomposition-coordination
method [29]–[31], and evolutionary methods [32], [33].
However, the above methods still suffer from different
levels of defects in practice, like high computation over-
head [33]–[35], dimensionality problem [37]–[39], duality
gap [40], parameter tuning [41]–[43] and premature conver-
gence [44]–[47]. Thus, it is of importance to further find
effective tools to address the studied problem. FIGURE 1. The leadership structure of grey wolf.
Grey wolf optimizer (GWO) is a new metaheuristic method
developed to solve the global optimization problem. The
biological foundation of the GWO method is the hunting II. IMPROVED GREY WOLF OPTIMIZER (IGWO)
behavior and coordination mechanism of grey wolves in A. STANDARD GREY WOLF OPTIMIZER (GWO)
nature [48]–[50]. In GWO, the population is divided into GWO is a novel heuristic method based on the leadership and
several components that oversee different tasks (like leader, hunting mechanism of the grey wolves [53]–[55]. In GWO,
adviser or soldier) during the evolutionary process, which it is believed that a wolf pack composed of several wolves
can help converge to satisfying solutions for the optimization obeys the strict social hierarchy (as showed in FIGURE 1),
problem at hand. With the advantages of less computational including ¬ the alpha wolves (α) that are in charge of
parameters and higher execution efficiency, GWO has been determining the social behaviors (like hunting, sleeping or
applied in a variety of engineering problems, like parameter waking); ­ the beta wolves (β) that collect information to
identification, machining learning and process control [51]. help the alpha wolves make the scientific decision; ® the
However, there are few reports about using GWO to address delta wolves (δ) that obey the alpha and beta wolves and
the hydropower operation problem by far. Thus, the authors are responsible for the detection, protection and feeding; ¯
try to refill the research gap, but it is found that the GWO the omegas wolves (ω) that belong to the lowest level in
method is easily tapped into local optima [52]. In order to swarm.
further enhance the GWO method, this research develops a For the target problem, the first three best individuals are
novel method called improved grey wolf optimizer (IGWO), respectively seen as the alpha, beta and delta wolves, while
where the quasi-oppositional learning strategy is employed to the rest of individuals are regarded as the omega wolves.
enhance the converge rate; the elite mutation operator is used For the sake of convenience, it is assumed that m wolves are
to increase the swarm diversity; the elastic-ball strategy and involved in the search process of the J -variable optimization
heuristic constraint handling method are adopted to modify problem. The GWO method attempts to encircle the prey by
infeasible solutions. The practicability of the IGWO method the following model:
is verified by the simulation results of numerical experiments
and engineering problems. D = C ◦ X kp − X k (1)
The main contributions of this paper are summarized as
X k+1 = X kp − λ · D (2)
below: (1) The standard GWO method is introduced to
solve the hydropower system operation optimization prob- λ = 2a ◦ r1 − a (3)
lem, which may be the first report in this research field by C = 2r2 (4)
far; (2) the IGWO method based on the elite mutation, quasi-
oppositional learning and elastic-ball strategy is proposed to where ◦ is the entry wise product of two vectors. |·| is the
alleviate the premature convergence of the GWO method; entry wise absolute value of the vector. X kp and X denotes
(3) the IGWO performances are superior to several conven- the position vectors of the prey and wolf. λ and C are two
tional evolutionary methods in both numerical experiments intermediate vectors. r1 , r2 ∈ [0, 1]J are two random vectors.
and two hydropower operation problems. Thus, an effective a is the coefficient vector to balance global exploitation and
method is provided for the complex engineering optimization local exploration, and its jth element (aj ) can be obtained by
problem. k
The rest of this paper is organized as below: the details aj = a0 · (1 − ) (5)
k
of the IGWO method is given in Section II; in Section III,
12 classical benchmark functions are used to testify the where a0 is the initial value. k is the maximum iteration.
IGWO performance; the detailed information of IGWO In the evolutionary process, the group search behaviors
for optimal operation of hydropower system is given in are mainly guided by three leader wolves (alpha, beta and
Section IV; the proposed method is used to address two delta) [56]–[58]. In other words, the wolves in the swarm
hydropower systems in Section V; and finally the conclusions update their positions by tracking the locations of the first
are given in the end. three leader wolves, which can be mathematically expressed

155514 VOLUME 7, 2019

Z.-K. Feng et al.: Optimal Operation of Hydropower System by IGWO Based on Elite Mutation and Quasi-Oppositional Learning

FIGURE 3. Sketch map of the elite mutation operator.

FIGURE 2. Sketch map of quasi-oppositional learning strategy.

as below:

Dα = |C1 ◦ X α − X| (6)
Dβ = C2 ◦ X β − X (7)
Dδ = |C3 ◦ X δ − X| (8)
0
X1 = X α − A1 ◦ Dα (9)
0
X 2 = X β − A2 ◦ Dβ (10)
0
X 3 = X δ − A3 ◦ Dδ (11) FIGURE 4. Sketch map of the elastic-ball strategy.
0 0 0
(X 1 + X 2 + X 3 )
X k+1 = (12)
3
2) ELITE MUTATION OPERATOR
where X α , X β and X δ are the position vectors of the alpha, Generally, the global best-known solution found by the
beta and delta wolves. swarm can provide multiple search directions for the agents,
while the mutation operator can increase the swarm diver-
B. THE PROPOSED IGWO METHOD sity [65]–[67]. Inspired by this case, the elite mutation opera-
1) QUASI-OPPOSITIONAL LEARNING tor in FIGURE 3 is proposed to improve the global search
For most evolutionary methods, all the individuals may fail ability of the population. Specially, the offspring swarms
to jump out of local optima and it is necessary to find some produced by the quasi-oppositional learning strategy and the
methods to enlarge the hunting range of the swarm [59]–[61]. GWO method are firstly merged to form a hybrid swarm with
In recent years, the opposition-based learning comparing the 2m wolves (m denotes the number of wolves for evolution);
performances of the current solution and its opposite solution secondary, all the wolves in the hybrid swarm will be sorted
at the same time is often used to improve the performance by their fitness values; finally, the first p(p < m) individuals
of evolutionary method [59]–[61]. For any one solution, with better performances will directly enter the next gener-
Equation (13) can be used to obtain the quasi-opposite ation, while the following m-p wolves are used to generate
position. As showed in FIGURE 2, the quasi-oppositional the mutant agents by Equation (15). In this way, the elite
learning forces the agent change to a new position in the individuals can be maintained while the distribution range of
problem space, which will increase the probability of find- solutions in the space can be enhanced.
ing better solutions. As a result, the global exploitation
Eki = List kp1 + r4 · (X ki − List kp1 ) (15)
and local exploration of the population can be effectively
enhanced.
where Eki is the ith mutant agent at the kth iteration. List kp1
denotes the element randomly selected from the set of X α ,
(
Cj + (Cj − Xijk ) · r3 if(Xijk < Cj )
X̃ijk = (13) X β and X δ . r4 ∈ [−1, 1]J is the random vector.
Cj − (Xijk − Cj ) · r3 if(Xijk ≥ Cj )
up
Cj = (Xj + Xjdown ) · 0.5 (14) 3) ELASTIC-BALL STRATEGY
up
After the mutation operation, it is possible that some newly
where Xj and Xjdown are the upper and lower limits of the obtained agents violate the boundary constraints. Generally,
ith dimension. X̃ijk is the quasi-oppositional position of the ith the method in Equation (16) is often used to adjust infeasible
wolf in the jth dimension at the kth iteration. r3 is the random agents. However, after this modification steps, many infea-
number uniformly distributed in the range of [0, 1]. sible agents may gather around the boundary region in the

VOLUME 7, 2019 155515

 Z.-K. Feng et al.: Optimal Operation of Hydropower System by IGWO Based on Elite Mutation and Quasi-Oppositional Learning

TABLE 1. Detailed information of 12 benchmark functions.

  
later evolution stage, which will harm the global exploration X k − X up if Xijk > Xj
up
ij j
ability of the swarm. valijk =   (18)
X down − X k
j ij if Xijk < Xjdown
  
X up if X k > X up
j  ij j
Xijk =  (16)
X down if X k < X down
j ij j
4) EXECUTION PROCEDURE OF IGWO
For alleviating this problem, this study introduces a new The main execution procedures of the IGWO approach are
elastic-ball strategy to modify infeasible agents [68]. Spe- given as below:
cially, the infeasible agents are firstly changed to feasible Step 1: Define the values of all the necessary computation
zones by Equation (17)∼(18); by this time, some elements parameters, like maximum iteration k̄ and swarm size m.
falling into the infeasible zone will be randomly generated Step 2: Set the counter k = 1, and then randomly generate
in the feasible problem space. As showed in FIGURE 4, the initial swarm in the problem space.
the elastic-ball strategy can effectively expand the distribu- Step 3: Evaluate the fitness values of all the wolves in the
tion diversity of the solutions and benefit the global explo- current swarm to obtain the new positions of three leader
ration of the swarm. wolves (alpha, beta and delta), and then update the positions
   of all the wolves by Equations (9)∼(12).
X up − val k if X k > X up Step 4: Use the quasi-opposition learning strategy in Equa-
j ij  ij i
Xijk =  (17) tions (13)∼(14) to increase the convergence rate of the
X down + val k if X k < X down
j ij ij j swarm.

155516 VOLUME 7, 2019

Z.-K. Feng et al.: Optimal Operation of Hydropower System by IGWO Based on Elite Mutation and Quasi-Oppositional Learning

FIGURE 5. Sketch map of 8 benchmark functions with 2 variables.

Step 5: Use the elite mutation strategy in Equation (15) to For the developed methods, the swarm size and maximum
improve the diversity of the swarm. iterations are respectively set as 30 and 500, while the other
Step 6: Use the elastic-ball strategy in Equations (17)∼(18) parameters are set as follows:
to modify the positions of infeasible individuals. DE: The scaling factor and crossover probability are set to
Step 7: Set k = k+1. If k ≤ k̄, go to Step 3 for the next 0.5 and 0.6, respectively.
cycle; otherwise, stop the iteration and the global best-known PSO: The inertia weight is linearly decreased from 0.9 to
wolf is treated as the final solution of the target problem. 0.3, and two learning factor c1 and c2 are set as 2.0.
SCA: The constant a is set to 2.0.
III. NUMERICAL EXPERIMENTS GSA: The initial gravitational constant G0 is set to 100, the
A. BENCHMARK FUNCTIONS attenuation factor is set to 20, respectively.
In this section, 12 famous test functions are chosen to test GWO: The coefficient a0 is set to 2.0.
the performance of the proposed method, including the uni- IGWO: The coefficient a0 is set as 2.0, the number of the
modal functions (F1 ∼ F8 ) with one global optimum and retained agents is set to 15.
multimodal functions (F9 ∼ F12 ) with multiple optimal
solutions. Generally, the unimodal functions and multimodal C. COMPARISON WITH OTHER ALGORITHMS IN
functions are used to test the global search and local explo- 30 DIMENSION PROBLEMS
ration performances of evolutionary methods, respectively.
1) RESULT COMPARISON ANALYSIS
As showed in TABLE 1, the global best objective values
To reduce the negative effect of random seeds, the developed
of almost all the functions (except for F8 ) are 0. FIGURE
methods are independently executed 30 times in the same
5 draws the 3-dimensional shape of 8 benchmark functions.
operating environment. TABLE 2 shows the average and stan-
It can be clearly seen that the selected functions have different
dard deviation (Std. for short) of the objective values obtained
distribution features (like concavity or convexity, unimodal
by 12 evolutionary methods. It can be found the IGWO
or multimodal) in the three-dimensional space, and then the
method has better performances than the control methods. For
performance of the developed methods can be fully tested and
instance, the proposed method can find the global optimal
compared.
objective values of several functions (F1 , F3 , F9 and F11 )
B. PARAMETER SETTINGS while the other methods fail to make it. Besides, the develop
To show the superiority of the proposed method, the results method is smaller than 1 second, demonstrating its high
of several methods are introduced for comparison, includ- execution efficiency. Thus, the proposed method can quickly
ing the firefly algorithm (FA) [69], whale optimal algo- find the near-optimal solutions for different functions.
rithm (WOA) [70], differential search algorithm (DSA) [71],
backtracking search algorithm (BSA) [72], harmony search 2) CONVERGENCE PROCESS ANALYSIS
algorithm (HSA) [73], modified cuckoo search (MCS) [74], FIGURE 6 shows the convergence trajectories of the devel-
differential evolution (DE) [75], particle swarm optimization oped methods. It can be found that the proposed method is
(PSO) [76], sine cosine algorithm (SCA) [76], [77], gravita- able to quickly seek out solutions with better performances
tional search algorithm (GSA) [79] and GWO [48]. than the control methods. Taking the four unimodal functions
To be mentioned, the results of 6 methods (DE, PSO, SCA, (F1 ∼F4 ) as the example, the IGWO method can converge
GSA, GWO and IGWO) are developed in JAVA language, to the global optimal solution as the iterations is about 50,
while the other methods are taken from previous literature. while the other methods take more iterations. Thus, three

VOLUME 7, 2019 155517

 Z.-K. Feng et al.: Optimal Operation of Hydropower System by IGWO Based on Elite Mutation and Quasi-Oppositional Learning

TABLE 2. Statistical results of different methods for 12 test functions with 30 variables.

FIGURE 6. Convergence trajectories of six developed algorithms for 12 test functions with 30 variables.

modified strategies (like quasi-oppositional learning and elite 3) BOX AND WHISKER ANALYSIS
mutation) can effectively improve the convergence rate of the The Box and whisker plot is widely adopted to compare
standard GWO method. the distribution of different methods because it can provide

155518 VOLUME 7, 2019

Z.-K. Feng et al.: Optimal Operation of Hydropower System by IGWO Based on Elite Mutation and Quasi-Oppositional Learning

FIGURE 7. Box plot obtained by GWO and IGWO for 12 test functions with 30 variables.

TABLE 3. The results of Wilcoxon test for single problem at alpha=0.05 for 12 test functions with 30 variables.

TABLE 4. The results of Wilcoxon signed rank test at alpha=0.05 for 12 test functions with 30 variables.

abundant information of the studied data samples, including found that the proposed method has better performances due
show the minimum, second quartile, median, third quartile, to the large number of the Win symbol as well as the smaller
maximum. FIGURE 7 draws the Box and whisker of the values of p and R− values. Therefore, it can be concluded
solutions obtained by GWO and IGWO for 12 functions with that the IGWO method is statistically superior to the other
30 variables. It can be found that the IGWO method has a methods.
smaller scale distribution than GWO in all the test functions,
demonstrating its superior performance. D. COMPARISON WITH OTHER ALGORITHMS IN HIGH
DIMENSION PROBLEMS
4) WILCOXON NONPARAMETRIC TEST 1) RESULT COMPARISON ANALYSIS
The Wilcoxon nonparametric test is chosen to statis- To further test the performance of the proposed method,
tically compare the performances of different methods. 5 methods (DE, PSO, SCA, GSA and GWO) are employed to
TABLE 3∼4 show the detailed results between the IGWO resolve 12 functions with 100 and 500 variables, respectively.
method and other methods, where the average objective val- TABLE 5 shows the statistical results of 6 methods obtained
ues are chosen as the samples of the target methods. It can be in 30 independent runs. It can be clearly seen that the average

VOLUME 7, 2019 155519

 Z.-K. Feng et al.: Optimal Operation of Hydropower System by IGWO Based on Elite Mutation and Quasi-Oppositional Learning

TABLE 5. Statistical results of 6 developed heuristic algorithms for 12 test functions with 100 and 500 variables.

TABLE 6. The results of Wilcoxon signed rank test at alpha=0.05 for the IGWO method always wins its competitor because the
12 test functions with 100 and 500 variables.
number of win symbol is equal to the number of functions;
the IGWO method obtains higher R+ value than R− value in
all the comparisons, while all the p values are smaller than
0.05. Thus, it can be concluded that the developed method
can generate better solutions than the other methods in all the
benchmark functions.

IV. IGWO FOR THE OPTIMAL OPERATION OF

HYDROPOWER SYSTEM
A. MATHEMATICAL MODEL
1) OBJECTIVE FUNCTION
In this research, the goal is chosen to minimize the standard
deviation of the residual load series obtained by subtracting
the total power outputs of all the hydropower reservoirs from
the original load curve, which can be expressed as:
v
u #2
u1 T N
u X "
X
and standard deviation of the proposed method are obviously mininize f = t Lt − Pn,t (19)
smaller than that of the other methods; the IGWO time is 2
t=1 n=1
smaller than GSA and slightly larger than the other methods,
proving the practicability of the proposed modified strategies. 2) PHYSICAL CONSTRAINTS
1) Inflow balance equations
2) WILCOXON NONPARAMETRIC TEST Nn
X
TABLE 6 gives the statistical analysis results obtained by In,t = qn,t + (Qi,t + Si,t ) (20)
the Wilcoxon signed-rank test method. It can be found that i=1

155520 VOLUME 7, 2019

Z.-K. Feng et al.: Optimal Operation of Hydropower System by IGWO Based on Elite Mutation and Quasi-Oppositional Learning

2) Outflow balance equations C. CONSTRAINT HANDLING METHOD

As mentioned above, a series of complicated inequality and
On,t = Qn,t + Sn,t (21) equality constraints are involved in the hydropower operation
3) Water balance equations problem, and a slight adjustment of the decision variable in
the upstream reservoir may bring about significant changes
Vn,t+1 = Vn,t + In,t − On,t · 1t
 
(22) of downstream reservoirs. Meanwhile, some newly generated
wolves may violate one or more constraints or even the swarm
4) Water head equations fails to find feasible solutions in the end. Thus, a practical

constraint handling method is proposed to address this prob-
Hn,t = 0.5 · Zn,t−1 + Zn,t − dn,t (23)
lem, and the detailed procedures for the individual X ki are
5) Storage volume limits given as below:
up
Step 1: Set the reservoir index n =1 and start the constraint
down
Vn,t ≤ Vn,t ≤ Vn,t (24) handling process.
6) Power output limits Step 2: Set the counter l =1.
Step 3: Set l = l+1. Calculate the storage volumes of the
up
Pdown
n,t ≤ Pn,t ≤ Pn,t (25) nth reservoir during the scheduling horizon by Equation (33)
and then obtain the violation value φn of the final storage by
7) Turbine discharge limits Equation (34).
up n
Qdown
n,t ≤ Qn,t ≤ Qn,t (26) down
, min Vn,t−1 + 1t−1

Vn,t = max Vn,t
8) Total outflow limits  up o
× In,t−1 − On,t−1 , Vn,t

(33)
up
Odown
 
n,t ≤ On,t ≤ On,t (27) φn = Vn,t − fi1 Znend (34)
9) Total output limits Step 4: If |φn | is smaller than the stopping precision ε or l is
N larger than the maximum iteration L, go to Step 5; otherwise,
up
X
Rdown
t ≤ Pn,t ≤ Rt (28) modify the total outflows of the nth reservoir by Equation (35)
n=1 and then go to Step 3.
n o
10) Initial and final water levels limits up
On,t =max Odown , φ 1
 
n,t min O n,t + n [T × t ], On,t
(
beg (35)
Zn,0 = Zn
(29)
Zn,T = Znend Step 5: Set n = n+1. If n > N , go to Step 6; otherwise,
turn to Step 2 for the next cycle.
11) Nonlinear characteristic curves  k
Step 6: Obtain  objective function f X i and constraint
 kthe
violation viol X i to calculate the fitness value F X ki for
  
1


Vn,t = fi Zn,t


d = fi2 On,t (30) comparing the performances of different wolves, which can
 n,t be expressed as below:
Pn,t = fi3 Qn,t , Hn,t


h i h i h i
F X ki = f X ki + viol X ki (36)
B. WOLF STRUCTURE AND INITIALIZATION
A
Generally, the agent structure has an important effect on the
h i X n h i o
viol X ki = ca · max ga X ki , 0
execution efficiency of evolutionary algorithms. After deep
a=1
consideration, the total outflow of each hydroplant is chosen B h h ii2
as the decision variable, and then a wolf is composed of the +
X
cb · eb X ki (37)
outflow processes of all the reservoirs during the scheduling
b=1
horizon. The ith wolf in the kth iteration can be expressed as:
where ga X ki and ca are the value
 
X ki = O1,1 , · · · , O1,T , · · · , On,t , · · · , ON ,1 , · · · , ON ,T
   and penalty coefficient of
the ath inequality constraint. eb X ki and cb are the value and
(31) penalty coefficient of the bth equality constraint. A and B are
the number of inequality and equality constraints.
During the initialization phase, the element will be randomly
generated in the feasible search range by
D. DETAILED EXECUTE PROCEDURE
h i
up The pseudocode of the IGWO method for the hydropower
On,t = Odown
n,t + r6 × On,t − Odown
n,t (32)
system peak operation problem is given as below:
where r6 is the random number uniformly distributed in the //initialization
range of [0, 1]. For each individual i = 1:m //m and i are set and index of

VOLUME 7, 2019 155521

 Z.-K. Feng et al.: Optimal Operation of Hydropower System by IGWO Based on Elite Mutation and Quasi-Oppositional Learning

wolf TABLE 7. Final storage volumes of all the reservoirs in four cases.
Randomly determine the position of each wolf by
Eq. (32)
Evaluate the fitness value of the target wolf by
Eq. (36)
End For
While (Stopping criteria not met) do
//GWO optimizer
Update the positions of three wolves (alpha, beta and
delta)
For each individual i = 1:m //Obtain the original wolf
set S1
X ki ← Update the position of each wolf by Eq. (12)
//Elastic-ball strategy
up
If the jth value in X ki exceeds the upper limit Xi operator. For a m-wolf swarm, the number of calculations
up
valijk = Xijk − Xj is (2m-p) at k̄ iterations. Considering the condition that p is
up
Xijk = Xj − valijk smaller than m, the time
complexity of the IGWO method
End if is approximate to O mk̄ as k̄ and m go to infinity. Besides,
If the jth value in X ki exceeds the lower limit Xjdown for a N -plant and T -period hydropower system, each wolf in
IGWO needs a vector of size NT to store it. Thus, the space
valijk = Xjdown − Xijk
complexity of IGWO with m wolves is O(NTm). From the
Xijk = Xjdown + valijk
above analysis, it can be found that the computation cost
End if
of the IGWO method shows a polynomial growth with the
End For
increase of swarm size, iteration and problem size. Thus,
//Quasi-opposition learning strategy
the proposed approach is computationally efficient to be used
For each individual i =1:m //Obtain the quasi wolf
in the real system to be deployed.
set S2
X̃ik ← Obtain the quasi-opposition wolf by Eq. (13)
V. CASE STUDIES
Evaluate the fitness value of the target wolf by
In this section, two hydropower systems are chosen to ver-
Eq. (36)
ify the IGWO performance. In the first hydropower system,
End For
to reduce the complexity, it is assumed that the outflow is
//Elite mutation operator
used for generation (no spillage); the power output is linearly
Form a temporary set S = S1 ∪ S2
obtained by turbine discharge; no units are in the deduced
Wolves in S are sorted in ascending order by fitness
variables. In the second hydropower system, all the nonlinear
values
characteristic curves and physical constraints are considered
For each individual i = 1:m
in the modeling process.
If i ≤ P
X ki ← The ith wolf in S
A. CASE STUDY 1
End if
In this section, the IGWO method is applied to a mature
Else
k hydropower system consisted of 10 coupled reservoirs (Res.
X̃ i ← Obtain the mutant wolf by Eq. (15)
for short). Four cases are designed to testify the feasibil-
Modify the infeasible wolf by the elastic-ball
ity of the proposed method: the scheduling horizon has
strategy
20 periods with 1 hour per period; the initial storage volumes
Evaluate the fitness value of the target wolf by
of 10 hydropower reservoirs are set as [6, 6, 3, 8, 8, 7, 15,
Eq. (36)
6, 5, 15]T ; four different kinds of final storages are given
End else
in TABLE 7; the other computational parameters are not
End For
given here to save space, which can be found in [48]. Then,
End While
five evolutionary methods are introduced for comparison,
The best wolf found is seen as the final scheduling scheme
including DE, PSO, SCA, GSA and GWO. Five algorithms
are independently executed in 30 times, while in each exper-
E. COMPUTATIONAL COST OF IGWO iment, the number of maximum iteration and individuals are
Due to the complexity of engineering matter, it is assumed set to be 30 and 200, respectively.
that the time spent on calculating the fitness value of
each wolf is much larger than other calculating operations. 1) ROBUSTNESS TESTING IN DIFFERENT CASES
In IGWO, the swarm should calculate all the wolves in GWO To demonstrate the robustness of the proposed method,
optimizer and m-p newly-obtained wolves in the mutation the statistical results of all the involved methods are given in

155522 VOLUME 7, 2019

Z.-K. Feng et al.: Optimal Operation of Hydropower System by IGWO Based on Elite Mutation and Quasi-Oppositional Learning

TABLE 8. Detailed results of different algorithms in 4 cases.

FIGURE 9. Topological structure of the Wu hydropower system.

FIGURE 10. The box plot of six heuristic algorithms in four cases.

2) DETAILED SCHEDULING RESULTS

TABLE 9 shows the detailed outputs of all the hydropower
reservoirs obtained by the IGWO method in case 1. It can be
found that all the hydroplants can increase the output at peak
periods and reduce the generation at valley periods, which
will help produce a smooth residual load series without peak-
valley differences. Thus, the IGWO method can smooth the
FIGURE 8. The box plot of two methods for mature cascade system.
load demand by reasonably allocating the hydropower output,
providing a valid tool for optimal operation of hydropower
system.
TABLE 8, including the best, average and standard deviation
of the objective values and the average time. As showed in B. CASE STUDY 2
TABLE 8, it can be clearly found that the variation ranges of In this section, the proposed method is applied to address
the objective values by the IGWO method in four cases are the peak operation of the Wu hydropower system in south-
obviously smaller than the control methods; while there are west China. As one of the largest hydropower bases, the Wu
no obvious differences in the time of different methods. For hydropower system has effectively promoted the socialist
instance, in case 1, the IGWO method makes about 58.9%, modernization construction of the central and western region
49.5%, 70.8%, 29.7% and 0.5% improvements in the average of China. FIGURE 9 shows the characteristic information and
objective values compared with DE, PSO, SCA, GSA and topological structure of 5 reservoirs, including HJD, DF, SFY,
GWO, respectively. WJD and GPT. As mentioned in the above section, the same
FIGURE 8 illustrates the box plots of GWO and IGWO five methods are chosen for comparisons.
for the mature hydropower system in different cases. It can
be found that the variation ranges of the objective values 1) ROBUSTNESS TESTING OF DIFFERENT METHODS
obtained by the IGWO method is much smaller than GWO TABLE 10 shows the statistical results of the objective values
method. Thus, it can be concluded that three modified search obtained by 6 methods. It shows that the IGWO method needs
strategies can effectively enhance the GWO performances in about 18 seconds to find the solutions with the best perfor-
the changing environments, while the IGWO method is able mances in terms of all the performance indicators. In four
to produce stable solutions that can satisfy the load change of cases, the range of the objective values by the IGWO method
electrical power system. are reduced by about 99.89%, 99.91%, 99.93% and 99.82%

VOLUME 7, 2019 155523

 Z.-K. Feng et al.: Optimal Operation of Hydropower System by IGWO Based on Elite Mutation and Quasi-Oppositional Learning

TABLE 9. The detailed outputs obtained by IGWO in case 1.

as compared with DE, respectively. FIGURE 10 draws the TABLE 10. Statistical results of various methods in 10 independent runs
for 4 cases.
distributions of objective values obtained by six methods.
It can be found that although the results of five control meth-
ods are relatively stable, it is still possible for these methods
to generate outliers; besides, the proposed method has the
smallest undulations of the objective value, demonstrating
its strong search performance. Thus, the IGWO method can
quickly find stable scheduling schemes for the complicated
multi-constrained hydropower operation problem regardless
of the initial random seeds.

2) DETAILED SCHEDULING RESULTS

TABLE 11 shows the statistical results of the best schedul-
ing scheme obtained by 6 methods in Case 1, including
peak load, valley load, peak-valley difference (PVD), and
standard deviation. It can be found that the IGWO method
has better performances than the control methods in terms
of all the indexes. For instance, the reductions in standard
deviation and peak-valley difference are obviously larger than
other methods, demonstrating the effectiveness of the IGWO
method. Hence, compared with the comparison methods,
the developed method can generate better schemes to support
the optimal operation of hydropower system.

3) PERFORMANCE OF THE PROPOSED METHOD IN THE

LONG RUN
TABLE 12 gives the statistical indicators of the IGWO
method in 12 scenarios with different operational conditions, scheduling results obtained by the IGWO method. From
including the peak load, peak-valley difference, standard TABLE 12, it can be clearly found that in different cases,
deviation and load rate (LR) of the objective value as well the IGWO method only needs about 18 seconds to produce
as the computation time. FIGURE 11 shows the detailed satisfying scheduling results with obvious improvements in 5

155524 VOLUME 7, 2019

Z.-K. Feng et al.: Optimal Operation of Hydropower System by IGWO Based on Elite Mutation and Quasi-Oppositional Learning

TABLE 11. Detailed results obtained by six algorithms in case 1.

TABLE 12. The detailed results obtained by IGWO in 12 scenarios.

FIGURE 11. Scheduling results obtained by IGWO in 12 scenarios.

evaluation indexes. For instance, the peak load, peak-valley VI. CONCLUSION
difference, standard deviation and load rate are improved by With the rapid economic development in recent years,
about 30.4%, 86.1%, 21.4%, 86.4% and 12.9% on average, the load demand grows sharply and more attention is paid to
respectively. From FIGURE 11, it can be observed that the the peak operation of hydropower system. As a novel method
IGWO method can collect hydropower outputs to respond inspired by the leadership and hunting behaviors of wolves,
the load demand and produce smoother residual load curves the grey wolf optimizer (GWO) has not been reported to
compare with the original load curves. Thus, this case fully solve the target problem by far. In practice, it is found that
proves that even though the external environments are under the standard GWO method still suffers from the premature
dynamic change, the proposed method can effectively solve convergence problem. Thus, this paper develops an improved
the complicated hydropower operation problem. grey wolf optimizer (IGWO) to enhance the performance of

VOLUME 7, 2019 155525

 Z.-K. Feng et al.: Optimal Operation of Hydropower System by IGWO Based on Elite Mutation and Quasi-Oppositional Learning

the standard GWO method. The developed method is applied

to several benchmark functions and two hydropower systems. up
On,t Maximum total outflow of the nth reservoir
The significant conclusions obtained from the results include: at period t in m3 /s.
(1) even for the same optimization problem, the results of Qdown Minimum turbine discharge of the nth reser-
n,t
various methods often change with the adopted evolutionary voir at period t in m3 /s.
mechanism, demonstrating the importance of search strategy. up
Qn,t Maximum turbine discharge of the nth reser-
For instance, in multiple independent experiments for the voir at period t in m3 /s.
mature hydropower system, the IGWO method bettered DE Pdown Minimum power outputs of the nth reservoir
n,t
with about 99.9% lower values of the range of the objective at period t in MW.
values, respectively. up
Pn,t Maximum power outputs of the nth reservoir
(2) due to the loss of swarm diversity, the traditional GWO at period t in MW.
method falls into local optima with a high probability, while Rdown Minimum total output of the hydropower
n,t
the proposed method with three modified strategies (quasi- system at period t in MW.
oppositional learning, elite mutation operator and elastic-ball up
Rn,t Maximum total output of the hydropower
strategy) can effectively improve the swarm diversity, search system at period t in MW.
rate and individual feasibility. For instance, in the experiment
for the real-world hydropower system, the IGWO method
VARIABLES
makes about 10.38% and 7.83% reductions in the peak-valley f Value of objective function.
difference and standard deviation compared with the GWO Pn,t Power output of the nth hydroplant at period
algorithm, respectively. t in MW.
(3) in the peak operation of power system, the hydroplant In,t Total inflow of the nth reservoir at period t
should increase the generations at peak periods and reduce in m3 /s.
the output at valley periods to produce satisfying residual load Qn,t turbine discharge of the nth reservoir at
series left for other power plants. period t in m3 /s.
Meanwhile, the future directions in this field can be deep- Sn,t Spillage of the nth reservoir at period t in
ened from the following aspects: one is to develop more effec- m3 /s.
tive strategies to enhance the GWO performance; the second On,t Total outflow of the nth reservoir at period t
is to parallelize the proposed method to take full advantage in m3 /s.
of the abundant computing resources in normal computers; Vn,t Storage volume of the nth reservoir at period
the last is to verify the feasibility of the developed method t in m3 .
in similar engineering problems (like coordinative operation Zn,t Forebay water level of the nth reservoir at
between hydropower and other energies) under uncertainty. period t in m.
NOMENCLATURE dn,t Upstream water level of the nth reservoir at
The notations used in the mathematic model for hydropower period t in m.
operation are given as below: Hn,t Water head of the nth reservoir at period t in
m.
SETS
N Set of hydroplants.
T Set of periods. FUNCTIONS
Nn Set of upstream reservoirs directly con- fi1 (·) Stage-storage curve of the nth reservoir.
nected to the nth reservoir. fi2 (·) Stage-discharge curve of the nth reservoir.
fi3 (·) Nonlinear function between turbine dis-
charge, water head and power output of the
CONSTANTS
1t Number of seconds at the tth period. nth reservoir.
beg
Zn Initial water levels of the nth reservoir in m.
Znend Final water levels of the nth reservoir in m. ACKNOWLEDGMENT
Lt Load demand of electric system at period t The writers would like to thank editors and reviewers for their
in MW. valuable comments and suggestions.
qn,t Local inflow of the nth reservoir at period t REFERENCES
in m3 /s. [1] Z.-K. Feng, W.-J. Niu, R. Zhang, S. Wang, and C.-T. Cheng,
down
Vn,t Minimum storage volumes of the nth reser- ‘‘Operation rule derivation of hydropower reservoir by k-means
clustering method and extreme learning machine based on particle
voir at period t in m3 . swarm optimization,’’ J. Hydrol., vol. 576, pp. 229–238, Sep. 2019,
up
Vn,t Maximum storage volumes of the nth reser- doi: 10.1016/j.jhydrol.2019.06.045.
voir at period t in m3 . [2] X.-Y. Guo, C. Ma, and Z.-B. Tang, ‘‘Multi-timescale joint optimal dispatch
model considering environmental flow requirements for a Dewatered river
Odown
n,t Minimum total outflow of the nth reservoir channel: Case study of Jinping cascade hydropower stations,’’ J. Water
at period t in m3 /s. Resour. Planning Manage., vol. 144, no. 10, Oct. 2018, Art. no. 05018014.

155526 VOLUME 7, 2019

Z.-K. Feng et al.: Optimal Operation of Hydropower System by IGWO Based on Elite Mutation and Quasi-Oppositional Learning

[3] J. Lian, X. Sun, and C. Ma, ‘‘Multi-year optimal operation strategy of [23] S. Wen, T. Huang, X. Yu, Z. Q. M. Chen, and Z. Zeng, ‘‘Aperiodic sampled-
Danjiangkou reservoir after dam heightening for the middle route of the data sliding-mode control of fuzzy systems with communication delays
south-north water transfer project,’’ Water Sci. Technol. Water Supply, via the event-triggered method,’’ IEEE Trans. Fuzzy Syst., vol. 24, no. 5,
vol. 16, no. 4, pp. 961–970, 2016. pp. 1048–1057, Oct. 2016, doi: 10.1109/TFUZZ.2015.2501412.
[4] D. Liu, S. Guo, Q. Shao, P. Liu, L. Xiong, L. Wang, X. Hong, Y. Xu, and [24] S. Wen, Z. Q. M. Chen, X. Yu, Z. Zeng, and T. Huang, ‘‘Fuzzy con-
Z. Wang, ‘‘Assessing the effects of adaptation measures on optimal water trol for uncertain vehicle active suspension systems via dynamic sliding-
resources allocation under varied water availability conditions,’’ J. Hydrol., mode approach,’’ IEEE Trans. Syst., Man, Cybern., Syst., vol. 47, no. 1,
vol. 556, pp. 759–774, Jan. 2018. pp. 24–32, Jan. 2017, doi: 10.1109/TSMC.2016.2564930.
[5] B. Ming, P. Liu, T. Bai, R. Tang, and M. Feng, ‘‘Improving optimization [25] S. Wang, Y. Cao, T. Huang, and S. Wen, ‘‘Passivity and passifi-
efficiency for reservoir operation using a search space reduction method,’’ cation of memristive neural networks with leakage term and time-
Water Resour. Manage., vol. 31, no. 4, pp. 1173–1190, 2017. varying delays,’’ Appl. Math. Comput., vol. 361, pp. 294–310, Nov. 2019,
[6] B. Ming, P. Liu, J. Chang, Y. Wang, and Q. Huang, ‘‘Deriving operating doi: 10.1016/j.amc.2019.05.040.
rules of pumped water storage using multiobjective optimization: Case [26] X. Xie, S. Wen, Z. Zeng, and T. Huang, ‘‘Memristor-based cir-
study of the Han to Wei Interbasin water transfer project, China,’’ J. Water cuit implementation of pulse-coupled neural network with dynamical
Resour. Planning Manage., vol. 143, no. 10, 2017, Art. no. 5017012. threshold generators,’’ Neurocomputing, vol. 284, pp. 10–16, Apr. 2018,
[7] B. Ming, P. Liu, L. Cheng, Y. Zhou, and X. Wang, ‘‘Optimal daily genera- doi: 10.1016/j.neucom.2018.01.024.
tion scheduling of large hydro–Photovoltaic hybrid power plants,’’ Energy [27] S. Wen, X. Xie, Z. Yan, T. Huang, and Z. Zeng, ‘‘General memristor
Convers. Manage., vol. 171, pp. 528–540, Sep. 2018. with applications in multilayer neural networks,’’ Neural Netw., vol. 103,
[8] Y. Li, T. Zhao, P. Wang, H. B. Gooi, L. Wu, Y. Liu, and J. Ye, ‘‘Optimal pp. 142–149, Jul. 2018, doi: 10.1016/j.neunet.2018.03.015.
operation of multimicrogrids via cooperative energy and reserve schedul- [28] X. Zeng, S. Wen, Z. Zeng, and T. Huang, ‘‘Design of memristor-based
ing,’’ IEEE Trans. Ind. Informat., vol. 14, no. 8, pp. 3459–3468, Aug. 2018, image convolution calculation in convolutional neural network,’’ Neural
doi: 10.1109/TII.2018.2792441. Comput. Appl., vol. 30, no. 2, pp. 503–508, 2018, doi: 10.1007/s00521-
[9] Y. Liu, Y. Li, H. B. Gooi, Y. Jian, H. Xin, X. Jiang, and J. Pan, ‘‘Dis- 016-2700-2.
tributed robust energy management of a multimicrogrid system in the [29] F. Zheng and A. Zecchin, ‘‘An efficient decomposition and dual-stage
real-time energy market,’’ IEEE Trans. Sustain. Energy, vol. 10, no. 1, multi-objective optimization method for water distribution systems with
pp. 396–406, Jan. 2019, doi: 10.1109/TSTE.2017.2779827. multiple supply sources,’’ Environ. Model. Softw., vol. 55, pp. 143–155,
[10] Z.-K. Feng, W.-J. Niu, S. Wang, C.-T. Cheng, Z.-Q. Jiang, H. Qin, May 2014.
and Y. Liu, ‘‘Developing a successive linear programming model for [30] B. Jia, P. Zhong, X. Wan, B. Xu, and J. Chen, ‘‘Decomposition-
head-sensitive hydropower system operation considering power shortage coordination model of reservoir group and flood storage basin for real-time
aspect,’’ Energy, vol. 155, pp. 252–261, Jul. 2018. flood control operation,’’ Hydrol. Res., vol. 46, no. 1, pp. 11–25, 2015.
[11] C. Ma, J. Lian, and J. Wang, ‘‘Short-term optimal operation of Three- [31] F. Zheng, A. R. Simpson, A. C. Zecchin, and J. W. Deuerlein,
gorge and Gezhouba cascade hydropower stations in non-flood season with ‘‘A graph decomposition-based approach for water distribution net-
operation rules from data mining,’’ Energy Convers. Manage., vol. 65, work optimization,’’ Water Resour. Res., vol. 49, no. 4, pp. 2093–2109,
pp. 616–627, Jan. 2013. 2013.
[12] X. W. Zhuang, Y. P. Li, G. H. Huang, and X. T. Zeng, ‘‘An inexact joint- [32] X. Yuan, Y. Zhang, and Y. Yuan, ‘‘Improved Self-Adaptive Chaotic Genetic
probabilistic programming method for risk assessment in water resources Algorithm for Hydrogeneration Scheduling,’’ J. Water Resour. Planning
allocation,’’ Stochastic Environ. Res. Risk Assessment, vol. 29, no. 5, Manage., vol. 134, no. 4, pp. 319–325, 2008.
pp. 1287–1301, 2015. [33] X. Yuan, B. Ji, Y. Yuan, R. M. Ikram, X. Zhang, and Y. Huang, ‘‘An efficient
[13] T. Yang, X. Gao, S. L. Sellars, and S. Sorooshian, ‘‘Improving the chaos embedded hybrid approach for hydro-thermal unit commitment
multi-objective evolutionary optimization algorithm for hydropower reser- problem,’’ Energy Convers. Manage., vol. 91, pp. 225–237, Feb. 2015.
voir operations in the California Oroville-Thermalito complex,’’ Environ. [34] J. P. S. Catalao, S. J. P. S. Mariano, V. M. F. Mendes, and
Model. Softw., vol. 69, pp. 262–279, Jul. 2015. L. A. F. M. Ferreira, ‘‘Scheduling of head-sensitive cascaded hydro sys-
[14] Z. Feng, W. Niu, S. Wang, C. Cheng, and Z. Song, ‘‘Mixed integer linear tems: A nonlinear approach,’’ IEEE Trans. Power Syst., vol. 24, no. 1,
programming model for peak operation of gas-fired generating units with pp. 337–346, Feb. 2009.
disjoint-prohibited operating zones,’’ Energies, vol. 12, no. 11, p. 2179, [35] J. P. S. Catalao and V. M. F. Mendes, ‘‘Influence of environmental con-
2019, doi: 10.3390/en12112179. straints on profit-based short-term thermal scheduling,’’ IEEE Trans. Sus-
[15] Z.-K. Feng, W.-J. Niu, W.-C. Wang, J.-Z. Zhou, and C.-T. Cheng, ‘‘A mixed tain. Energy, vol. 2, no. 2, pp. 131–138, Apr. 2011.
integer linear programming model for unit commitment of thermal plants [36] J. P. S. Catalão, H. M. I. Pousinho, and J. Contreras, ‘‘Optimal hydro
with peak shaving operation aspect in regional power grid lack of flexible scheduling and offering strategies considering price uncertainty and risk
hydropower energy,’’ Energy, vol. 175, pp. 618–629, May 2019. management,’’ Energy, vol. 37, no. 1, pp. 237–244, 2012.
[16] Z.-K. Feng, W.-J. Niu, C.-T. Cheng, and J.-Z. Zhou, ‘‘Peak shav- [37] Z. Jiang, H. Qin, C. Ji, Z. Feng, and J. Zhou, ‘‘Two dimension reduction
ing operation of hydro-thermal-nuclear plants serving multiple power methods for multi-dimensional dynamic programming and its application
grids by linear programming,’’ Energy, vol. 135, no. 3, pp. 210–219, in cascade reservoirs operation optimization,’’ Water, vol. 9, no. 9, p. 634,
Sep. 2017. 2017, doi: 10.3390/w9090634.
[17] K. Madani, ‘‘Game theory and water resources,’’ J. Hydrol., vol. 381, [38] T. Zhao, J. Zhao, and D. Yang, ‘‘Improved dynamic programming for
nos. 3–4, pp. 225–238, 2010. hydropower reservoir operation,’’ J. Water Resour. Planning Manage.,
[18] K. Madani and M. Hooshyar, ‘‘A game theory-reinforcement learning vol. 140, pp. 365–374, Mar. 2012.
(GT-RL) method to develop optimal operation policies for multi-operator [39] T. Zhao and J. Zhao, ‘‘Improved multiple-objective dynamic programming
reservoir systems,’’ J. Hydrol., vol. 519, no. 27, pp. 732–742, 2014. model for reservoir operation optimization,’’ J. HydroInform., vol. 16,
[19] K. Madani and A. Dinar, ‘‘Cooperative institutions for sustainable common no. 5, pp. 1142–1157, Sep. 2014.
pool resource management: Application to groundwater,’’ Water Resour. [40] J. Chen, P.-A. Zhong, Y. Zhang, D. Navar, and W. W.-G. Yeh,
Res., vol. 48, no. 9, Sep. 2012, Art. no. W09553. ‘‘A decomposition-integration risk analysis method for real-time operation
[20] Z.-K. Feng, W.-J. Niu, C.-T. Cheng, and X.-Y. Wu, ‘‘Optimization of of a complex flood control system,’’ Water Resour. Res., vol. 53, no. 3,
hydropower system operation by uniform dynamic programming for pp. 2490–2506, 2017.
dimensionality reduction,’’ Energy, vol. 134, pp. 718–730, Sep. 2017. [41] S. Xiao, X. Xie, S. Wen, Z. Zeng, T. Huang, and J. Jiang, ‘‘GST-memristor-
[21] Z.-K. Feng, W.-J. Niu, and C.-T. Cheng, ‘‘Optimizing electrical power based online learning neural networks,’’ Neurocomputing, vol. 272,
production of hydropower system by uniform progressive optimality pp. 677–682, Jan. 2017, doi: 10.1016/j.neucom.2017.08.014.
algorithm based on two-stage search mechanism and uniform design,’’ [42] G. Ren, Y. Cao, S. Wen, Z. Zeng, and T. Huang, ‘‘A modified Elman neural
J. Cleaner Prod., vol. 190, pp. 432–442, Jul. 2018. network with a new learning rate scheme,’’ Neurocomputing, vol. 286,
[22] Z. Jiang, C. Ji, H. Qin, and Z. Feng, ‘‘Multi-stage progressive optimality pp. 11–18, Apr. 2018.
algorithm and its application in energy storage operation chart optimiza- [43] J.-X. Chang, T. Bai, Q. Huang, and D.-W. Yang, ‘‘Optimization of water
tion of cascade reservoirs,’’ Energy, vol. 148, pp. 309–323, Apr. 2018, resources utilization by PSO-GA,’’ Water Resour. Manage., vol. 27, no. 10,
doi: 10.1016/j.energy.2018.01.176. pp. 3525–3540, 2013.

VOLUME 7, 2019 155527

 Z.-K. Feng et al.: Optimal Operation of Hydropower System by IGWO Based on Elite Mutation and Quasi-Oppositional Learning

[44] Z.-K. Feng, W.-J. Niu, and C.-T. Cheng, ‘‘Multi-objective quantum- [64] B. Mandal and P. K. Roy, ‘‘Multi-objective optimal power flow using quasi-
behaved particle swarm optimization for economic environmental oppositional teaching learning based optimization,’’ Appl. Soft Comput.,
hydrothermal energy system scheduling,’’ Energy, vol. 131, pp. 165–178, vol. 21, pp. 590–606, Aug. 2014, doi: 10.1016/j.asoc.2014.04.010.
Jul. 2017. [65] R. Zhang, J. Zhou, H. Zhang, X. Liao, and X. Wang, ‘‘Optimal operation
[45] F. Zheng, A. C. Zecchin, J. P. Newman, H. R. Maier, and G. C. Dandy, of large-scale cascaded hydropower systems in the upper reaches of the
‘‘An adaptive convergence-trajectory controlled ant colony optimization Yangtze River, China,’’ J. Water Resour. Planning Manage., vol. 140, no. 4,
algorithm with application to water distribution system design problems,’’ pp. 480–495, 2014.
IEEE Trans. Evol. Comput., vol. 21, no. 5, pp. 773–791, Oct. 2017. [66] B. Ming, J.-X. Chang, Q. Huang, Y.-M. Wang, and S.-Z. Huang, ‘‘Optimal
[46] C. L. Wu and K. W. Chau, ‘‘Prediction of rainfall time series using operation of multi-reservoir system based-on cuckoo search algorithm,’’
modular soft computingmethods,’’ Eng. Appl. Artif. Intell., vol. 26, no. 3, Water Resour. Manage., vol. 29, no. 15, pp. 5671–5687, 2015.
pp. 997–1007, 2013. [67] T. Yang, A. A. Asanjan, M. Faridzad, N. Hayatbini, X. Gao, and
S. Sorooshian, ‘‘An enhanced artificial neural network with a shuffled com-
[47] X. Lei, J. Zhang, H. Wang, M. Wang, S.-T. Khu, Z. Li, and Q. Tan,
plex evolutionary global optimization with principal component analysis,’’
‘‘Deriving mixed reservoir operating rules for flood control based on
Inf. Sci., vols. 418–419, pp. 302–316, Dec. 2017.
weighted non-dominated sorting genetic algorithm II,’’ J. Hydrol., vol. 564,
[68] Z. Chen, X. Yuan, H. Tian, and B. Ji, ‘‘Improved gravitational search
pp. 967–983, Sep. 2018.
algorithm for parameter identification of water turbine regulation system,’’
[48] S. Mirjalili, S. M. Mirjalili, and A. Lewis, ‘‘Grey wolf Energy Convers. Manage., vol. 78, pp. 306–315, Feb. 2014.
optimizer,’’ Adv. Eng. Softw., vol. 69, pp. 46–61, Mar. 2014, [69] X. S. Yang, Firefly Algorithm In Engineering Optimization, Book Firefly
doi: 10.1016/j.advengsoft.2013.12.007. Algorithm Engineering Optimization, Series Firefly Algorithm In Engi-
[49] S. Arora, H. Singh, M. Sharma, S. Sharma, and P. Anand, ‘‘A new hybrid neering Optimization. Hoboken, NJ, USA: Wiley, 2010.
algorithm based on Grey wolf optimization and crow search algorithm for [70] S. Mirjalili and A. Lewis, ‘‘The whale optimization algorithm,’’ Adv. Eng.
unconstrained function optimization and feature selection,’’ IEEE Access, Softw., vol. 95, pp. 51–67, May 2016.
vol. 7, pp. 26343–26361, 2019, doi: 10.1109/ACCESS.2019.2897325. [71] P. Civicioglu, ‘‘Transforming geocentric Cartesian coordinates to geodetic
[50] P. Hu, S. Chen, H. Huang, G. Zhang, and L. Liu, ‘‘Improved alpha-guided coordinates by using differential search algorithm,’’ Comput. Geosci.,
Grey wolf optimizer,’’ IEEE Access, vol. 7, pp. 5421–5437, 2019, doi: vol. 46, pp. 229–247, Sep. 2012.
10.1109/ACCESS.2018.2889816. [72] P. Civicioglu, ‘‘Backtracking Search Optimization Algorithm for numer-
[51] M. H. Qais, H. M. Hasanien, and S. Alghuwainem, ‘‘Augmented grey ical optimization problems,’’ Appl. Math. Comput., vol. 219, no. 15,
wolf optimizer for grid-connected PMSG-based wind energy conver- pp. 8121–8144, 2013.
sion systems,’’ Appl. Soft Comput., vol. 69, pp. 504–515, Aug. 2018, [73] Z. W. Geem, J. H. Kim, and G. V. Loganathan, ‘‘A new heuristic opti-
doi: 10.1016/j.asoc.2018.05.006. mization algorithm: Harmony search,’’ J. Simul., vol. 76, no. 2, pp. 60–68,
[52] K. Luo, ‘‘Enhanced grey wolf optimizer with a model for dynamically esti- Feb. 2001.
mating the location of the prey,’’ Appl. Soft Comput., vol. 77, pp. 225–235, [74] S. Walton, O. Hassan, K. Morgan, and M. Brown, ‘‘Modified cuckoo
Apr. 2019, doi: 10.1016/j.asoc.2019.01.025. search: A new gradient free optimisation algorithm,’’ Chaos, Solitons
[53] B. Martin, J. Marot, and S. Bourennane, ‘‘Mixed grey wolf optimizer for Fractals, vol. 44, no. 9, pp. 710–718, 2011.
the joint denoising and unmixing of multispectral images,’’ Appl. Soft Com- [75] Z.-K. Feng, W.-J. Niu, J.-Z. Zhou, C.-T. Cheng, and Y.-C. Zhang,
put., vol. 74, pp. 385–410, Jan. 2019, doi: 10.1016/j.asoc.2018.10.019. ‘‘Scheduling of short-term hydrothermal energy system by parallel multi-
objective differential evolution,’’ Appl. Soft Comput., vol. 61, pp. 58–71,
[54] C. Lu, L. Gao, Q. Pan, X. Li, and J. Zheng, ‘‘A multi-objective cellular
Dec. 2017.
grey wolf optimizer for hybrid flowshop scheduling problem considering
[76] W.-J. Niu, Z.-K. Feng, C.-T. Cheng, and X.-Y. Wu, ‘‘A parallel multi-
noise pollution,’’ Appl. Soft Comput., vol. 75, pp. 728–749, Feb. 2019, doi:
objective particle swarm optimization for cascade hydropower reservoir
10.1016/j.asoc.2018.11.043.
operation in southwest China,’’ Appl. Soft Comput., vol. 70, pp. 562–575,
[55] M. H. M. QaisH Hasanien and S. Alghuwainem, ‘‘A grey wolf optimizer Sep. 2018.
for optimum parameters of multiple PI controllers of a grid-connected [77] W. Fu, J. Tan, C. Li, Z. Zou, Q. Li, and T. Chen, ‘‘A hybrid fault diagnosis
PMSG driven by variable speed wind turbine,’’ IEEE Access, vol. 6, approach for rotating machinery with the fusion of entropy-based feature
pp. 44120–44128, 2018, doi: 10.1109/ACCESS.2018.2864303. extraction and SVM optimized by a chaos quantum sine cosine algorithm,’’
[56] W. Long, T. Wu, S. Cai, X. Liang, J. Jiao, and M. Xu, ‘‘A novel grey Entropy, vol. 20, no. 9, p. 626, 2018.
wolf optimizer algorithm with refraction learning,’’ IEEE Access, vol. 7, [78] W. Fu, K. Wang, C. Li, X. Li, Y. Li, and H. Zhong, ‘‘Vibration trend
pp. 57805–57819, 2019, doi: 10.1109/ACCESS.2019.2910813. measurement for a hydropower generator based on optimal variational
[57] K. Albina and S. G. Lee, ‘‘Hybrid stochastic exploration using mode decomposition and an LSSVM improved with chaotic sine cosine
grey wolf optimizer and coordinated multi-robot exploration algorithm optimization,’’ Meas. Sci. Technol., vol. 30, no. 1, 2019,
algorithms,’’ IEEE Access, vol. 7, pp. 14246–14255, 2019, doi: Art. no. 015012.
10.1109/ACCESS.2019.2894524. [79] W.-J. Niu, Z.-K. Feng, M. Zeng, B.-F. Feng, Y.-W. Min, C.-T. Cheng, and
[58] S. K. Goudos, T. V. Yioultsis, A. D. Boursianis, K. E. Psannis, and J.-Z. Zhou, ‘‘Forecasting reservoir monthly runoff via ensemble empiri-
K. Siakavara, ‘‘Application of new hybrid Jaya grey wolf optimizer to cal mode decomposition and extreme learning machine optimized by an
antenna design for 5G communications systems,’’ IEEE Access, vol. 7, improved gravitational search algorithm,’’ Appl. Soft Comput., vol. 82,
pp. 71061–71071, 2019, doi: 10.1109/ACCESS.2019.2919116. Sep. 2019, Art. no. 105589, doi: 10.1016/j.asoc.2019.105589.
[59] Z.-K. Feng, S. Liu, W.-J. Niu, Z.-Q. Jiang, B. Luo, and S.-M. Miao, ‘‘Multi- [80] S. Liu, Z.-K. Feng, W.-J. Niu, H.-R. Zhang, and Z.-G. Song, ‘‘Peak opera-
objective operation of cascade hydropower reservoirs using TOPSIS and tion problem solving for hydropower reservoirs by elite-guide sine cosine
gravitational search algorithm with opposition learning and mutation,’’ algorithm with Gaussian local search and random mutation,’’ Energies,
Water, vol. 11, no. 10, p. 2040, 2019, doi: 10.3390/w11102040. vol. 12, no. 11, p. 2189, 2019, doi: 10.3390/en12112189.
[60] X. Yuan, B. Zhang, P. Wang, J. Liang, Y. Yuan, Y. Huang, and X. Lei,
ZHONG-KAI FENG was born in Heze, Shandong,
‘‘Multi-objective optimal power flow based on improved strength Pareto
evolutionary algorithm,’’ Energy, vol. 122, pp. 70–82, Mar. 2017. China, in 1988. He received the B.S. and Ph.D.
degrees in hydraulic and hydropower engineering
[61] T. Bai, Y.-B. Kan, J.-X. Chang, Q. Huang, and F.-J. Chang, ‘‘Fusing
feasible search space into PSO for multi-objective cascade reservoir opti- from the Dalian University of Technology, Dalian,
mization,’’ Appl. Soft Comput., vol. 51, pp. 328–340, Feb. 2017. Liaoning, China, in 2011 and 2016, respectively.
[62] Y. Gao and W. Liu, ‘‘Cloud estimation of distribution algorithm with quasi- Since 2016, he has been the Lecturer with the
oppositional learning and preference order ranking for multi-objective School of Hydropower and Information Engineer-
optimisation,’’ Int. J. Grid Utility Comput., vol. 7, no. 3, pp. 200–207, ing, Huazhong University of Science and Tech-
2016, doi: 10.1504/IJGUC.2016.080188. nology, Wuhan, Hubei, China. He is the author
[63] P. K. Roy, A. Sur, and D. K. Pradhan, ‘‘Optimal short-term hydro-thermal of about 30 inventions and 50 articles in peer-
scheduling using quasi-oppositional teaching learning based optimiza- reviewed journals. His current research interests include hydropower and
tion,’’ Eng. Appl. Artif. Intell., vol. 26, no. 10, pp. 2516–2524, 2013, renewable energy operation optimization, machine learning, and decision
doi: 10.1016/j.engappai.2013.08.002. support system development.

155528 VOLUME 7, 2019

Z.-K. Feng et al.: Optimal Operation of Hydropower System by IGWO Based on Elite Mutation and Quasi-Oppositional Learning

SHUAI LIU was born in Baoding, Hebei, China, BIN LUO received the B.S. degree in water
in 1994. He received the B.S. degree in hydraulic resources and hydropower engineering and the
and hydropower engineering from the Hebei Uni- Ph.D. degree in power system and automation
versity of Engineering, Hebei, Handan, China, from the Dalian University of Technology, China,
in 2018. in 2010 and 2017, respectively.
Since 2018, he has been the Graduate Student He is currently with the Tsinghua Sichuan
with the School of Hydropower and Informa- Energy Internet Research Institute, Sichuan,
tion Engineering, Huazhong University of Science China. His research interests include optimal oper-
and Technology, Wuhan, Hubei, China. He is the ation of largescale hydropower systems, opera-
author of about five inventions and one article tion mode optimization and coordination, power
published in peer-reviewed journals. His current research interest includes system risk assessment, and control.
hydropower system operation optimization.

SHU-MIN MIAO received the B.S. degree in

WEN-JING NIU was born in Chifeng, Inner energy and power engineering from the North
Mongolia, China, in 1989. She received the B.S. China University of Water Resources and Elec-
and Ph.D. degrees in hydraulic and hydropower tric Power, China, in 2012, and the Ph.D. degree
engineering from the Dalian University of Tech- in water resources and hydropower engineering
nology, Dalian, Liaoning, China, in 2012 and from the Dalian University of Technology, China,
2017, respectively. in 2018.
Since 2017, she has been the Engineer with She is currently with the State Grid Sichuan
the Bureau of Hydrology, Changjiang Water Electric Power Research Institute, Sichuan, China.
Resources Commission, Wuhan, China. She is the Her research interests include hydropower system
author of about 40 inventions and 40 articles pub- operation and optimization, and electricity market for renewable energy.
lished in peer-reviewed journals. Her current research interests include runoff
forecasting, watershed management, and cascade reservoirs operation opti-
mization.

SEN WANG was born Yichun, Jiangxi, China,

in 1986. He received the Ph.D. degree in hydraulic
YI LIU was born in Honghu, Hubei, China, and hydropower engineering from the Dalian Uni-
in 1984. He received the Ph.D. degrees in system versity of Technology, Dalian, Liaoning, China,
analysis and integration from the Huazhong Uni- in 2014.
versity of Science and Technology, Wuhan, Hubei, He has been a Senior Engineer with the Key
in 2013. Laboratory of the Pearl River Estuarine Dynamics
Since 2018, he has been the Associate Pro- and Associated Process Regulation, Ministry of
fessor with the Huazhong University of Science Water Resources, Guangzhou, Guangdong, China.
and Technology, Wuhan. He is the author of His current research interests include coordinated
about 20 articles published in peer-reviewed jour- optimization of large-scale cascade reservoirs, watershed management and
nals. His current research interest includes water planning, and water resources information.
resource research.

VOLUME 7, 2019 155529