Journal of Cleaner Production 168 (2017) 1132e1142

Contents lists available at ScienceDirect

Journal of Cleaner Production

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

Optimization of energy management and conversion in the

multi-reservoir systems based on evolutionary algorithms
Mohammad Ehteram a, *, Hojat Karami a, Sayed-Farhad Mousavi a, Saeed Farzin a,
Ozgur Kisi b
a
Department of Water Engineering and Hydraulic Structures, Faculty of Civil Engineering, Semnan University, Semnan, Iran
b
School of Natural Sciences and Engineering, Ilia State University, Tbilisi, Georgia

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

Article history: Concerning water and energy crisis, energy management and correct utilization of resources are
Received 19 July 2017 important. In the present study, utilization of a multi-reservoir system was addressed with an approach
Received in revised form to improve production of hydroelectric energy. For this purpose, Monarch Butterfly Algorithm (MBA),
7 September 2017
which is a new evolutionary algorithm, was used. Three periods of dry (1963e64), wet (1951e52) and
Accepted 10 September 2017
Available online 13 September 2017
normal (1985e86) conditions were considered in the operation of a 4-reservoir system. Results showed
that MBA was capable of generating more energy as compared to Particle Swarm Optimization (PSO)
algorithm and Genetic Algorithm (GA). For instance, MBA improved the accuracy of GA and PSO in
Keywords:
Energy management
generating energy by 1.16 and 0.88 percent in the wet year, 1.28 and 1.2 percent in the dry year and 1.34
Energy conversion and 0.88 percent in the normal year, respectively. Moreover, quality of the responses obtained from the
Evolutionary algorithm MBA was better than those of the other two algorithms, because coefficients of variation of the responses
Monarch butterfly algorithm in MBA were less.
© 2017 Elsevier Ltd. All rights reserved.

1. Introduction
Human being has always been concerned with water and en-
ergy crisis. Regarding the global population growth and increased
demands, managing energy and resources is of a great importance
(Zou et al., 2017). One of the important energies used by the
humans is hydroelectric energy (Al-falahi et al., 2017). The hydro-
electric energy produced by power plants and the related turbines
are utilized based on water reception from upstream dams (Aali
et al., 2017). Therefore, the issue of exploitation of power plants
and electricity generation is associated with the amount of the
input water to the turbines (Jafarian et al., 2017; Dahunsi et al.,
2017). How the water is released from single-reservoir and multi-
reservoir systems of dams, such that the amount of power-plant
energy is adequate for supplying hydroelectric demands, have
significant importance (Bellos et al., 2017). The issue of energy
production is closely linked to the issue of exploitation of reservoirs
(Mathimani et al., 2017). Calculating the output of multi-reservoir
water systems, so that the energy production is maximal
* Corresponding author.
E-mail address: mohammdehteram@semnan.ac.ir (M. Ehteram).
regarding the input water to the turbines, is one of the objectives of
interest for researchers (Lv et al., 2017). During the recent years,
researchers have used various mathematical models in order to
calculate dams' water release, aiming at maximizing energy pro-
duction in the downstream power plants. One of the important
issues in this field is utilization of multi-reservoir water systems
with the purpose of increasing power generation in the down-
stream power plants. The evolutionary algorithms have recently
gained a good position in the field of solving the problems related
to the utilization of power plants and dams. Labadie (2004) used a
nonlinear approach in order to maximize energy production from a
dam's reservoir. This approach was properly able to perform water
management for energy production, in addition to the simplicity of
the calculations. Ballester and Carter (2017) used GA for utilizing a
multi-reservoir system in order to maximize energy production.
This method resulted in greater power production in the down-
stream power plant, as compared to the nonlinear method. Chang
et al. (2010) used GA to maximize energy production in a reser-
voir. Results indicated that this algorithm gave higher energy in the
downstream power plant, as compared to dynamic programming
method. Afshar (2012) used PSO algorithm to increase the profit
from hydroelectric energy production in multi-reservoir systems.
Results showed that PSO algorithm had high capability in energy

http://dx.doi.org/10.1016/j.jclepro.2017.09.099
0959-6526/© 2017 Elsevier Ltd. All rights reserved.
 M. Ehteram et al. / Journal of Cleaner Production 168 (2017) 1132e1142
generation, compared to GA. Afshar et al. (2015) used Ant Colony
Algorithm (ACA) for operation of multi-objective systems with the
aim of producing more energy in the downstream power plants.
Results revealed that a higher energy was obtained by applying
ACA, as compared to other evolutionary algorithms.
A successful algorithm that has recently been developed by
Wang et al. (2015) in the field of engineering optimization is the
Monarch Butterfly Algorithm (MBA). Wang et al. (2015) applied this
algorithm in optimizing complex mathematical functions. Also, this
algorithm has high efficiency in other engineering fields. It con-
verges fast and the obtained responses are close to the absolute
optimized answer.
In fact, there are different methods for operation of multi-
reservoir systems. Traditional methods include linear program-
ming, dynamic programming, and nonlinear methods. However,
each method has its own limitations. For instance, in the problems
of operating multi-reservoir systems, the use of nonlinear objective
functions makes the linear method non-applicable when there are
a lot of decision variables, or dynamic programming method re-
quires high computations time. Therefore, metaheuristic algo-
rithms are recommended for solving the important problems of
operating multi-reservoir systems (like water release rates, hy-
droelectric production, flood storage, etc.).
2. Problem statement and innovation
The meta-heuristic algorithms are of different problems. For
instance, some of them fall into the trap of local optimizations and
find the local optimal answer instead of finding the absolute
optimal answer. Moreover, there are some other algorithms that are
of low convergence speed. In addition, one another problem is the
lack of balance between exploration and exploitation capabilities.
The exploration capability is the ability of the evolutionary algo-
rithm in free search without any consideration of its findings during
the searching process. In contrary, the exploitation capability is
known as the amount of the algorithm attention paid to its
achievements during the searching process. Obviously, the more
the exploration capability in a searching algorithm, the more
random and unexpected behavior the algorithm will have. In
contrast, enhancement of the exploration in an algorithm causes it
to have a more cautious and counted behavior. The lack of balance
between the exploration and exploitation capabilities is observed
in a large number of the evolutionary algorithms (Amini et al.,
2017). The Monarch Butterfly algorithm can adjust a balance be-
tween the exploration and exploitation capabilities by using
migration and adjustment operators and applying parallel pro-
cessing. Also, unlike other evolutionary algorithms, the Monarch
Butterfly algorithm accelerates the calculation process by the help
of simpler operators compared with other evolutionary algorithms.
One of the important factors is the adjustment of random param-
eters of the evolutionary algorithms. By having less number of
parameters relative to other evolutionary algorithms, the Monarch
Butterfly algorithm is also of a better process in the determination
of random parameters. Therefore, the new Monarch Butterfly al-
gorithm is of more efficiency in the optimization process.
The present study aims to maximize energy production in
downstream power plants of a multi-reservoir water system. Water
level in the reservoirs is considered as decision variable. Moreover,
in order to solve this complex optimization problem with non-
convex, nonlinear objective function, the MBA is used. Outputs of
the above problem include the amount of produced hydroelectric
energy by the power plants, the hydraulic head, and the reservoirs'
water level. Hence, the novelty of this research is application of a
new algorithm in order to improve the efficiency of energy pro-
duction in a multi-reservoir system.
1133
3. Materials and methods
3.1. Monarch butterfly algorithm
One of the important species of butterflies in North America is
monarch butterfly. These butterflies are distinguished from other
butterflies by their colorful wings. Butterflies of the Southeast of
North Americas are known by their remarkable ability to migrate
from America and southern Canada to Mexico and travel thousands
of kilometers. Accordingly, the following assumptions are used in
simulating the behavior of monarch butterflies, and their migra-
tion, in solving the optimization problems:
1 Monarch butterflies include total population of monarch but-
terflies in lands 1 and 2. The monarch butterflies migrate from
land 1 to land 2 in April and return to land 1 in September.
2 The migration operator produces each child of the monarch
butterfly in lands 1 and 2.
3 To keep the population of monarch butterflies constant, older
ones are removed and younger butterflies are born.
4 Monarch butterflies with better objective function are auto-
matically transferred to the next generation. Thus, the superior
butterflies are never removed due to population regeneration.
3.2. Migration operator
To simplify the calculations, it is assumed that monarch but-
terflies live in land 1 from April to August and in land 2 from
September to March. Therefore, population of monarch butterflies
in land 1 is equal to NP1 ¼ rouðp  NPÞ and in land 2 is equal to
NP  NP1 , where rourounds p  NP to the nearest integer number,
NP is total number of monarch butterflies and p is the ratio of
monarch butterflies in land 1 to total population of monarch but-
terflies. The monarch butterflies in lands 1 and 2 are specified as
subpopulations 1 and 2, respectively. The migration is based on the
following equation:
xtþ1
i;k
¼ xtr1 ;k (1)
where, xtþ1
i;k
is the kth element of the generation xi in the generation
t þ 1, which represents the location of ith butterfly. Similarly, xtr1 ;k is
the kth element of xr1, which is the new location of butterfly r1. The
r1 monarch butterfly is selected from subpopulation 1. When r  p,
the kth element in the new generation of the butterflies is produced
based on equation (1) that r is calculated according to the following
equation:
r ¼ rand*peri (2)
where, peri denotes the migration period, usually taken as 1.2
(Wang et al., 2015) and rand is a random number of uniform dis-
tribution. On the other side, if r > p, the k element in the new
population is determined by the following relation:
xtþ1 ¼ xtr2;k (3)
i;k
where, xtr2;k is the kth element of xr2. Also, the monarch butterfly r2 is
from subpopulation 2. It is observed from the above analyses that
the direction of migration operator is adjusted by p rate. If the
mentioned parameter is large, then more butterflies are selected
from land 1, and subpopulation of land 1 plays an important role in
producing new butterflies. If the mentioned parameter is small,
then more butterflies are selected from land 2 and subpopulation of
1134 M. Ehteram et al. / Journal of Cleaner Production 168 (2017) 1132e1142

updated based on the following relation:

xtþ1
j;k
¼ xtbest;k (4)

where, xtþ1
j;k
is the kth element of generation xi, t þ 1 represents the
location of jth monarch butterfly and similarly xtbest;k is the kth
element of xbest, that is the best butterfly in lands 1 and 2. On the
other hand, if the random number is greater than p, the location is
updated by the help of the following equation:

xtþ1
j;k
¼ xtr3 ;k (5)

where, xtr3 ;k is the kth element of generation xr3, which is randomly

selected from land 2. With these conditions, if the value of rand is
more than Bar (monarch butterfly adjusting ratio), the location is
adjusted using the following equation:

xtþ1
j;k
¼ xtþ1
j;k
þ aðdxk  0:5Þ (6)

where, dx is the moving step of monarch butterfly that is calculated

based on Lvey flight. Levy flight is a random walk in which the step
lengths have a probability distribution that is heavily tailed. When
defined as a walk in a space of dimension greater than 1.0, the steps
Fig. 1. Details of the migration operator.
are in isotropic random directions (Wang et al., 2015).
In equation (6), a is weighting coefficient that is obtained from
the following equation:
land 2 plays an important role. In this study, p value is considered
5 (Wang et al., 2015). The migration operation is shown in
equal to 12 Smax
a¼ (7)
Fig. 1. t2

where, Smax is maximum step of the monarch butterfly and t is the

3.3. Adjusting operator
The adjusting operator along with the migration operator reg-
ulates the location of butterflies. For the jth element, if a random
number that is less that p is selected, then the butterfly location is
current generation number. Fig. 2 presents the steps of the
adjusting operator.
At first, the parameters of butterfly algorithm are set. Then, the
first generation is randomly selected and the objective function is
calculated for each butterfly. Thereupon, the butterflies' location is

Fig. 2. Adjusting operator.

 M. Ehteram et al. / Journal of Cleaner Production 168 (2017) 1132e1142 1135

updated and these steps are continued until convergence is 2 Continuity equations:
reached. Fig. 3 illustrates the monarch butterfly algorithm.  
Vm;tþ1 ¼ Vm;t þ h Im;t  Qm;t 
Vmþ1;tþ1 ¼ Vmþ1;t þ h Imþ1;t  Qmþ1;t (11)
4. Case study Imþ1;t ¼ IQmþ1;t þ Qm;ttm

Wujiang River crosses the capital of Guizhou State, China. The where, Vm;t is mth reservoir storage at the beginning of time period
basin area is 87,900 km2 and the river length is 1037 km. Average t, Im;t is mth reservoir inflow at tth time, Qm;t is mth reservoir
annual rainfall is 900e1400 mm, which more than 80% of it occurs outflow at the tth time including the outflow from spillway and
in April to August. The river system is one of the 30 greatest power power plants, h is conversion coefficient for unit, IQmþ1;t is inflow to
plants in the world. Four reservoirs including HJD, DF, SFY and the m þ 1th reservoir, and tm is routed flow from reservoir m to
WJD are located on this river. Since the SYF reservoir has a little reservoir m þ 1.
storage capacity, the optimization model is considered for the
other three reservoirs. The HJD reservoir is located at the up- 3 Water level
stream of the Wujiang River. The reservoir's normal level is
1140 m and its dead-storage level is 1076 m amsl. Maximum ca- Z m;t  Zm;t  Z m;t (12)
pacity of the power plant is 600 MW. In order to provide water
demands in dry years, the water level must not get lower than where, Z m;t and Z m;t are upper and lower limits of water level,
1095 m. The main task of the DF reservoir, located at 65 km respectively.
downstream of the HJD reservoir, is generation of electricity. The
normal and dead-storage levels of the DF reservoir are equal to 4 The outflow
970 m and 936 m, respectively. The power plant's installed ca-
pacity is 695 MW. The WJD reservoir is located at 74.9 km Q m;t  Qm;t  Q m;t (13)
downstream of the SYF reservoir. The reservoir's normal water
level is 760 m and its dead-storage level is 720 m. An important where, Q m;t and Q m;t are lower and upper limits of the reservoir
item about WJD reservoir is its outflow. In order to ensure safe
outflow, respectively.
navigation, the outflow from WJD reservoir should be equal to or
more than 100 m3/s. The system's total output electricity must not
5 Boundary conditions
be less than 680 MW.
The goal is to maximize the energy produced in this multi-
Zm;1 ¼ Zm;b ; Zm;Tþ1 ¼ Zm;e (14)
reservoir water system. The reservoirs' water level is considered
as decision variable. Because the reservoir storage is reflected by
where, Zm;b and Zm;Tþ1 are water level in each reservoir at the
the water level, the water level fluctuations show the outflow
beginning and end of the time period, respectively.
variations. When the upstream water level is specified, the outflow
In any constrained problem, it must be noticed that in the case of
is determined based on the continuity equation:
noncompliance of the constraints, penalty functions are added to
the objective function:
X
T X
M
F ¼ max Nm;t DT 0 !2 1
(8)
t¼1 m¼1 X
T X
M X
M
Nm;t ¼ km qm hm;t F¼ @ Nm;t  d N t  Nm;t ADt (15)
t¼1 m¼1 m¼1
where, F is total energy production for all the power plants (kWh),
Nm;t is the electricity output from reservoirs (MWh), Dt is time where,
interval (hr), km is a coefficient for each power plant, qm is the
2
X
M 3
amount of water released and passed through turbines, and hm;t is
6 1; Nm;t  N t7
hydraulic head.
Concerning the objective function, there are different d¼6
6 m¼1
7
7 (16)
4 X M 5
constraints: 0; Nm;t  Nt
m¼1
1 The constraint of electricity generation,
The mth reservoir outflow in each time interval is calculated as
follows:
For each power plant:
2    3
N m;t  Nm;t  N m;t (9) p1 qm;t  qm;t ; qm;t 2  ∞; qm;t
6   7
6 7
Nm;t ¼ 6 km qm;t hm;t ; qm;t 2 qm;t ; qm;t 7 (17)
4  7
and for the entire system:    5
p2 qm; t  qm;t ; qm;t 2 qm;t ; þ∞
N t  Nm;t  N t (10)
where, p1 and p2 are penalty coefficients.
where, N m;t is lower limit of electricity generation, Nm;t is upper In order to investigate the multi-reservoir system (Fig. 4), a wet
limit of electricity generation for the mth power plant and N and N year (1951e1952), normal year (1985e1986), and dry year
are lower and upper limits of electricity generation for the entire (1963e1964) are considered. The reservoirs' data is presented in
system, respectively. Table 1.
1136 M. Ehteram et al. / Journal of Cleaner Production 168 (2017) 1132e1142

Fig. 3. a: Monarch Butterfly pseudo code and b:Monarchbutterfly algorithm, c:Genetic algorithm, d: Particle swarm algorithm, f: a schematic overview of paper.
 M. Ehteram et al. / Journal of Cleaner Production 168 (2017) 1132e1142 1137

Fig. 3. (continued).

best population was equal to 30, because energy generation had

Table 1
maximum value among the considered populations, and the best
Information for the studied reservoirs.
value of Smax was equal to 0.7. Moreover, the best population of
Reservoir item Unit HJD DF WJD SYF MBA in dry year was equal to 30. Furthermore, Wang et al. (2015)
Normal water level m 150 345 502 385 reported a value of 125 for the parameter BAR.
Dead-storage water level m 1076 936 720 822
Table 4 illustrates the energy values for normal, wet, and dry
Total storage Billion m3 4.5 0.9 2.1 0.2
Regulation storage e Multi year seasonal Seasonal daily years. In the normal year, the mean produced energy, based on
Power coefficient e 8.5 8.35 8.17 8.5 MBA, is equal to 105.23  108 kWh, which has increased by 1.2 and
0.96 percent as compared to GA and PSO, respectively. In addition,
coefficient of variation of MBA is smaller than those of the GA and
PSO by 670 and 276 times, indicating the high quality of MBA. The
5. Results and discussion mean produced energy by MBA approach in the wet year is equal to
124.12  108 kwh, which is about 1.34 and 0.88 percent greater
Table 2 presents the sensitivity analysis of the MBA parameters. than those of the GA and PSO, respectively. The energy produced by
Since the MBA's nature is random, the mentioned parameters are MBA approach in the dry year is equal to 101.45  108 kwh, which is
determined using the results of sensitivity analysis. According to about 2.8 and 1.2 percent greater than those of the GA and PSO,
Table 3, the best population of MBA for a normal year was equal to respectively. The MBA approach has smaller coefficient of variation,
30, because the energy generation reached its peak value and also too.
the best value of Smax was equal to 0.8. In the wet year scenario, the Lingo is optimization and engineering software (Lingo User's
Guide, 1999). The solution of a problem by Lingo is considered as
global solution. The global solution of our case study in this article
is 105.67 (108 kwh), which is very close to the average solution of
MBA method. The average solution of MBA method is 99% of global
solution in normal year. The global solution for wet year is 124.34
(108 kwh) and the average solution of MBA method is 124.12
(108 kwh). Also, the global solution for dry year is 101.47 (108 kwh)
and the average solution of MBA method is about 99% of the global
solution. Thus, results reveal that MBA has high potential for
computing accurate solutions of the complex problems.
Fig. 5 shows how the three algorithms converge. It is seen that
MBA needs less number of iterations for convergence for the three
years of wet, normal and dry conditions, compared to GA and PSO.
Fig. 4. Multi-reservoir system. For example, the MBA algorithm converges after 400 iterations in
wet year (Fig. 5a), while the GA and PSO converge after 600
1138 M. Ehteram et al. / Journal of Cleaner Production 168 (2017) 1132e1142

Table 2
Inflow information for the studied reservoirs.

interval Wet year (m3/s) Normal year (m3/s) Dry year (ms/s)

HJD H-D D-S S-W HJD H-D D-S S-W HJD H-D D-S S-W

1 140 179 89 50 88 159 111 57 143 331 93 137

2 291 393 113 196 482 538 240 230 123 256 61 82
3 438 602 150 250 444 616 180 320 330 720 290 210
4 467 643 60 190 154 149 53 25 171 230 35 60
5 198 182 34 45 206 204 40 63 86 146 18 23
6 186 191 106 65 84 86 14 16 143 136 16 72
7 94 72 37 79 74 75 20 37 81 123 62 33
8 76 59 30 51 46 47 11 39 58 100 46 39
9 58 37 18 26 37 39 10 6 47 44 19 39
10 45 30 13 15 41 34 9 34 52 44 19 43
11 39 25 11 11 42 31 5 12 46 39 14 48
12 51 27 10 12 48 44 9 24 114 136 102 135
Average 173 203 56 82 146 168 58 72 116 192 65 77

Notes: H-D represents the inter zone flow between HJD reservoir and DF reservoir, D-S represents the inter zone flow between DF reservoir and SYF reservoir and S-W
represents the inter zone flow between SYF and WJD.

Table 3 HJD reservoir has a water level of 1100 m in the beginning. Then,
Sensitivity analysis for Monarch Butterfly Algorithm. the water level reaches 1076 m. Thereafter, it reaches 1130 m
Population Energy production Smax Energy production during the periods of 2 to 10 and the water level gets a descending
(108 kwh) (108 kwh) trend from period 10 onwards. The reason for the descending
Normal year trend of the water level is the greater amount of water outflow
10 104.58 0.5 104.67 from reservoir HJD, which is in the direction of increasing inflow
20 104.79 0.6 104.78 to other reservoirs in order to generate the minimal electricity at
30 105.23 0.7 105.12
the end of the period that is equal to 680 MW. The water level in
40 104.79 0.8 105.23
50 104.65 0.9 104.79 reservoirs DF and WJD begins from normal value and remains
Wet year constant during the periods 2 to 12, and then gets a descending
10 123.54 0.5 123.11 trend, because the outflow from these reservoirs reaches its
20 123.78 0.6 123.87
maximum value at the final period and the water level gets its
30 124.12 0.7 124.56
40 124.01 0.8 123.98
lowest possible value.
50 123.98 0.9 123.67 The following points could be deducted from Table 5:
Dry year
10 100.98 0.5 100.98 1) The outflow from HJD reservoir is considered as inflow to other
20 101.12 0.6 101.43
reservoirs. When the inflow to other reservoirs increases, the
30 101.45 0.7 101.45
40 100.78 0.8 100.98 water levels in downstream reservoirs increase to normal level
50 100.68 0.9 100.68 and get fixed (970 m for DF reservoir and 760 m for WJD
reservoir) and the hydraulic head in these reservoir changes.
The effect of this phenomenon can be seen in the variations of
hydraulic head in Table 5. It is clear that HJD reservoir has an
iterations. Thus, the number of iterations has decreased by 33% in
important role in this multi-reservoir system, especially in dry
MBA method. Also, the MBA method converged after 400 iterations
periods. The storage capacity of HJD reservoir can be used for
in normal and dry years, while the iteration numbers for the GA and
other reservoirs. So that its storage capacity is decreased little by
PSO are 20% and 40% more than the MBA, respectively. Thus, the
little to compensate the deficits of other reservoirs in dry pe-
MBA approach has considerably decreased the computational time
riods. However, the water levels of other downstream reservoirs
for solving the hydroelectric generation problem.
will not descend until the end of the operation periods.
2) The outflow of WJD reservoir should be greater than 100 m3/s,
5.1. Rationality of the results for multi-reservoir operation to ensure safe navigation. This point is clear in Table 5 that
outflow from WJD reservoir is equal or more than 100 m3/s at all
The computed results in the previous section need accurate time intervals.
analysis to verify the obtained results. Thus, normal year was 3) Total amount of generated power is more than 680.3 MW (final
selected as a sample among other years to be discussed. First, column in Table 5).
Fig. 6 and HJD reservoir were considered. The inflow to HJD 4) The final point in Table 5 is related to negligible amount of
reservoir is 88 m3/s for the first time interval (Table 2). Thus, water generated power by HJD reservoir at time interval 2 (0.0 MW)
level in the first period drops down (Fig. 6). The multi-reservoir and time interval 3 (0.5 MW). This is due to the abundant inflow
system has an important constraint: The guaranteed power to this reservoir at time intervals 2 (482 m3/s) and 3 (444 m3/s).
output of the system should be equal or more than 680 MW. Thus, In fact, these abundant inflow values are stored for critical pe-
the outflow of HJD reservoir should compensate water deficits in riods. Thus, the outflow of HJD reservoir decreases in these
other reservoirs. Therefore, the outflow of the HJD reservoir periods and the flow is used for the next periods. This causes
should be increased so that the total generated power can meet higher hydraulic head and power output for the next periods
680 MW. When the outflow of HJD reservoir increases, the water (Table 5).
level in this reservoir drops down (Fig. 6). Fig. 6 illustrates that the
 M. Ehteram et al. / Journal of Cleaner Production 168 (2017) 1132e1142 1139

Table 4
Energy production for: a) normal year, b) wet year and c) dry year.

Run Energy production (108 kwh)

Monarch Butterfly Algorithm Genetic Algorithm (Ming et al., 2015) Particle Swarm Optimization Algorithm (Ming et al., 2015)

1 105.22 104.15 104.16

2 105.22 104.15 104.16
3 105.23 104.07 104.25
4 105.23 104.15 104.18
5 105.23 103.58 104.22
6 105.23 103.87 104.24
7 105.23 104.71 104.07
8 105.23 103.11 104.34
9 105.23 103.99 104.32
10 105.23 104.16 104.12
Worst solution 105.22 104.16 104.07
Best solution 105.23 103.58 104.32
Average 105.23 103.98 104.22
Coefficient of variation 0.0003 0.201 0.083
Global solution 105.67

Run Energy production (108 kwh)

Monarch Butterfly Algorithm Genetic Algorithm (Ming et al., 2015) Particle Swarm Optimization Algorithm (Ming et al., 2015)

1 124.12 122.58 123.11

2 124.11 122.58 123.11
3 124.11 122.20 123.11
4 124.12 122.61 123.11
5 124.12 122.57 122.67
6 124.12 122.18 123.11
7 124.12 122.46 122.95
8 124.12 122.69 122.91
9 124.12 122.65 123.11
10 124.12 122.53 123.11
Worst solution 124.11 122.18 122.67
Best solution 124.12 122.69 123.011
Average 124.12 122.47 123.03
Coefficient of variation 0.00003 0.195 0.147
Global solution 124.34

run Energy production (108 kwh)

Monarch Butterfly Algorithm Genetic Algorithm (Ming et al., 2015) Particle Swarm Optimization Algorithm (Ming et al., 2015)

1 101.45 97.84 100.20

2 101.44 98.45 100.24
3 101.45 98.72 100.18
4 101.45 98.37 100.26
5 101.45 98.97 100.23
6 101.45 99.19 100.23
7 101.45 99.29 100.33
8 101.45 99.04 100.26
9 101.45 99.10 100.23
10 101.45 99.07 100.21
Worst solution 101.44 97.84 100.18
Best solution 101.45 99.29 100.33
Average 101.45 98.60 100.18
Coefficient of variation 0.00003 0.716 0.065
Global solution 101.47

6. Conclusions
In this study, operation of a multi-reservoir system was inves-
tigated in order to increase the energy production. The Monarch
Butterfly Algorithm was applied for operation of a 4-reservoir
system. Three years of normal, wet and dry conditions were
selected in the study period and the results were compared with
the GA and PSO. Results showed that the accuracy of GA and PSO in
generating energy was improved by 1.16 and 0.88 percent in the
wet year, 1.28 and 1.2 percent in the dry year and 1.2 and 0.96
percent in the normal year using the MBA. Furthermore, the coef-
ficient of variation obtained in the Monarch Butterfly Algorithm
was less than the other two algorithms. It is concluded from the
case study of 4-reservoir system that the HJD reservoir plays a vital
role in the energy production because it stores water in the dry
periods, and releases during the periods that other downstream
reservoirs need it for electricity generation. The following out-
comes can be drawn through the present study:
1 The convergence speed for the Monarch Butterfly algorithm in
achieving the absolute optimal answer is more than the genetic
and particle swarm algorithms.
2 The Monarch Butterfly algorithm leads to generate greater en-
ergy in the power plants of four-reservoir system in such a way
that the use of this algorithm could improve the energy man-
agement in a more logical form.
3 The HJD reservoir in the 4-reservoir system plays a key role in
energy supplying. In a way that, the water outflow from the
1140 M. Ehteram et al. / Journal of Cleaner Production 168 (2017) 1132e1142

Fig. 5. Convergence for a: wet year, b: normal year and c: dry year.
 M. Ehteram et al. / Journal of Cleaner Production 168 (2017) 1132e1142 1141

Fig. 6. Trajectories of the water level optimized by different algorithms in normal year for: a) HJD reservoir, b) DF reservoir and c) WJD reservoir.
1142 M. Ehteram et al. / Journal of Cleaner Production 168 (2017) 1132e1142

Table 5
Outflow, hydraulic head and power output, optimized by MBA, for the studied reservoirs.

Period Outflow (m3/s) Hydraulic head (m) Power output (MW) Total power (MW)

HJD DF SFY WJD HJD DF SFY WJD HJD DF SFY WJD e

1 119.6 278.9 388.4 99.12 120.5 101.3 69 110.2 122.4 236.3 227.6 89.4 725.7
2 0.0 350.4 591.2 678.2 130.3 112.22 69 125.4 0.0 331.4 345.3 700.4 1377.2
3 0.40 614.1 794.3 1086.2 149.7 125.46 69 130.7 0.5 651.4 465.8 1162.3 2280.0
4 110.0 255.2 311.1 337.1 160.2 128.23 69 128.2 151.4 278.8 182.3 368.7 981.2
5 202.1 407.9 445.6 510.4 159.3 127.22 69 133.1 277.34 435.9 260.4 554.3 1517.94
6 108.5 194.3 208.3 225.1 160.3 128.31 69 133.7 148.2 210 122.3 243.6 724.10
7 147.23 221.3 242.1 278.11 158.4 128.30 69 132.2 200.2 220.7 140.1 305.4 866.40
8 206.54 252.1 263.2 304.3 153.22 128.25 69 130.1 265.3 263.4 153.2 330.22 1012.12
9 228.12 266.3 275.2 277.2 147.21 128.12 69 129.4 278.3 287.3 160.2 308.14 1033.94
10 245.67 280.1 288.1 321.2 138.76 128.10 69 128.4 291.1 302.1 167.3 350.12 1110.62
11 302.11 332.1 338.2 350.1 127.1 127.4 69 127.2 302.1 357.6 198.3 380.23 1238.23
12 61.28 292.2 302.1 842.2 123.2 123.4 69 117.2 65.4 279.3 178.3 808.6 1331.60

reservoir for more electricity generation increases with reduc-
tion of the reservoir water level in order to generate the minimal
electricity in the system.
4 Based on all the three evolutionary algorithms used in this
study, the energy generation for 4-reservoir system in wet year
is greater than dry and normal years.
5 The quality of the answers related to the Monarch Butterfly al-
gorithm is higher than the other evolutionary algorithms by
having smaller variation factor in 10 random runs.
In the future researches, the combination of Monarch Butterfly
algorithm with fuzzy methods could be used in order to enhance
the algorithm and obtain more accurate results. Additionally, the
mentioned algorithm could be utilized for systems with more
reservoirs as well as larger number of objectives in order to better
evaluate the capability of this algorithm.
References
Aali, A., Pourmahmoud, N., Zare, V., 2017. Exergoeconomic analysis and multi-
objective optimization of a novel combined flash-binary cycle for Sabalan
geothermal power plant. Energy Convers. Manag. 143, 377e390.
Afshar, M.H., 2012. Large scale reservoir operation by constrained particle swarm
optimization algorithms. J. Hydro-Environ. Res. 6 (1), 75e87.
~ o, M.A., 2015. State of the art review of ant
Afshar, A., Massoumi, F., Afshar, A., Marin
colony optimization applications in water resource management. Water Resour.
Manag. 29 (11), 3891e3904.
Al-falahi, M.D.A., Jayasinghe, S.D.G., Enshaei, H., 2017. A review on recent size
optimization methodologies for standalone solar and wind hybrid renewable
energy system. Energy Convers. Manag. 143, 252e274.
Amini, M.H., Moghaddam, M.P., Karabasoglu, O., 2017. Simultaneous allocation of
electric vehicles' parking lots and distributed renewable resources in smart
power distribution networks. Sustain. Cities Soc. 28, 332e342.
Ballester, P.J., Carter, J.N., 2017. Model calibration of a real petroleum reservoir using
a parallel real-coded genetic algorithm. In: Proceedings of IEEE Congress on
Evolutionary Computation, Singapore, pp. 4313e4320.
Bellos, E., Christos, T., Kimon, A., 2017. Parametric analysis and optimization of a
solar assisted gas turbine. Energy Convers. Manag. 139, 151e166.
Chang, L.C., Chang, F.J., Wang, K.W., Dai, S.Y., 2010. Constrained genetic algorithms
for optimizing multi-use reservoir operation. J. Hydrol. 390 (1e2), 66e74.
Dahunsi, S., Oranusi, S., Efeovbokhan, V.E., 2017. Optimization of pretreatment,
process information, mass and energy balance. Energy Convers. Manag. 139,
260e275.
Jafarian, H., Sayyaadi, H., Torabi, F., 2017. Modeling and optimization of dew-point
evaporative coolers based-on a developed GMDH-type neural network. En-
ergy Convers. Manag. 43, 49e65.
Labadie, J.W., 2004. Optimal operation of multireservoir systems: State-of-the-art
review. J Water Resour. Plann. Manage 130 (2), 93e111.
Lingo (Computer software), 1999. Lingo (Computer Software). LINDO System,
Chicago.
Lv, M., Li, J., Du, H., Zhu, W., Meng, J., 2017. Solar array layout optimization for
stratospheric airships using numerical method. Energy Convers. Manag. 135,
160e169.
Mathimani, T., Uma, L., Prabaharan, D., 2017. Optimization of Direct Solvent Lipid
Extraction Kinetics on Marine Trebouxiophycean Alga by Central Composite
Design- Bioenergy Perspective, vol. 142, pp. 334e346.
Ming, B., Chang, J., Huang, Q., Wang, Y., Huang, S., 2015. Optimal operation of multi-
reservoir system based-on cuckoo search algorithm. Water Resour. Manag. 29,
5671e5687.
Wang, G.G., Deb, S., Cui, Z., 2015. Monarch butterfly optimization. Neural comput.
Appl. 112, 24e36.
Zou, D., Li, S., Li, Z., Kong, X., 2017. A new global particle swarm optimization for the
economic emission dispatch with or without transmission losses. Energy
Convers. Manag. 139, 45e70.