Download as pdf or txt
Download as pdf or txt
You are on page 1of 15

Energy 171 (2019) 241e255

Contents lists available at ScienceDirect

Energy
journal homepage: www.elsevier.com/locate/energy

Multi-objective complementary scheduling of hydro-thermal-RE


power system via a multi-objective hybrid grey wolf optimizer
Chaoshun Li*, Wenxiao Wang, Deshu Chen
School of Hydropower and Information Engineering, Huazhong University of Science and Technology, Wuhan 430074, China

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

Article history: This paper presents a new short-term multi-objective complementary scheduling problem for hydro-
Received 6 July 2018 thermal-renewable power systems (HTRPSs). The economic/emission objectives, different from tradi-
Received in revised form tional economic/emission load dispatch problems, consider the on/off status of thermal power units as
28 November 2018
well as the dispatched load among engaged units as the optimization variables under various compli-
Accepted 28 December 2018
cated nonlinear constraints. To solve the model with hybrid optimization variables, a multi-objective
Available online 4 January 2019
hybrid grey wolf optimization algorithm is proposed, in which continuous and discrete optimization
variables are encoded and optimized synchronously. A daily scheduling simulation example of a hybrid
Keywords:
Hydro-thermal-RE power system
power system consisting of cascade hydropower stations, thermal power units, and renewable energy
Complementary scheduling (RE) power plants is studied to test the proposed model and algorithm. The results not only demonstrate
Multi-objective optimization that the proposed algorithm can achieve the best Pareto front for economic/emission bi-objectives
Pareto front compared to its competitors, but also confirm that the obtained scheduling schemes are completely
Unit commitment within the feasible domain. Moreover, the impact of RE capacity on the power system is analyzed. Results
Grey wolf optimizer indicate that the joint operation of RE and hydropower stations benefit both the economic and emission
objectives, and the operational costs and pollution emissions decrease by 11.9% and 17.4%, respectively,
when the RE capacity increases from 50% to 100% in the hybrid system.
© 2018 Elsevier Ltd. All rights reserved.

1. Introduction attracted extensive attention.


The economic dispatch (ED) [1e4] scheduling problem has great
With increasing power demand and growing concern over the significance in power system operations, and aims to achieve the
energy crisis as well as global warming, renewable energy (RE), minimum economic cost by optimally distributing the load among
mainly wind and solar power, have been developed rapidly with the units in a power system. Many researchers have contributed to
continuous increases in capacity in power systems around the solve the ED problem for power systems, especially those systems
world. The trend of large-scale RE integration has brought about consisting of thermal power and hydropower units [5e7]. With the
new challenges and has triggered a series of problems. To reduce unceasing concern over global warming and environmental
the adverse effects of RE uncertainty and randomness, hydropower pollution, it is important to consider minimizing pollution emis-
and pumped storage energy systems are used as effective tools for sions as another objective in the scheduling of power systems.
modulating and stabilizing a power system that contains inter- Therefore, economic emission dispatch (EED) has been proposed
mittent RE sources. Therefore, Hydro-Thermal-RE power systems [8e13], which is a complex multi-objective problem (MOP) with
(HTRPS) are built for high-efficiency utilization of clean and two conflicting objectives, i.e., minimizing fuel cost while mini-
renewable energy. For instance, in southwest China, large-scale mizing emissions [14]. As a typical MOP, EED problems may be
hydropower is connected to the electric power grid, providing efficiently solved by applying multi-objective heuristic optimiza-
significant complementary capacity to the grid to allow connection tion algorithms (MOHOAs). Although methods based on mathe-
of significant RE power sources. In this context, the operation and matical programming might also be effective in handling EED,
scheduling of HTRPS has become an important issue and has some challenges, such as computational efficiency, initial solution
selection, and non-convex objective functions, are difficult issues to
overcome.
* Corresponding author. MOHOAs based on swarm intelligence and Pareto theory have
E-mail address: csli@hust.edu.cn (C. Li). the ability to obtain a Pareto-optimal solution set in a single run

https://doi.org/10.1016/j.energy.2018.12.213
0360-5442/© 2018 Elsevier Ltd. All rights reserved.
242 C. Li et al. / Energy 171 (2019) 241e255

Nomenclature for CS-HTPRS model Hj;t output power of hydropower station j at time t (MW)
Vj;t reservoir storage volume of hydropower station j at
time t (m3)
a. Indices Qj;t water discharge of hydropower station j at time t
i index for thermal power units (m3)
j index for hydropower stations
P tem_min
i;t /P tem_max
i;t temporary minimum/maximum generation
t index for time period
l index for algorithm iteration of thermal power i at time t (MW)
PW V
t /P t generation output of wind/photovoltaic power at
b. Constants time t (MW)
N H /NI /NT number of hydropower stations/thermal power Rj;t /Sj;t inflow/spillage of hydropower station j at time t (m3)
units/time periods Leq
t equivalent load at time t (MW)
ath th th Lnet load taken by thermal power units at time t (MW)
i ,bi ,ci fuel cost coefficients of thermal power I ($/h) t
ai ,benv
env
i ,cenv
i environment emission coefficients of thermal
e. Functions
power i (ton/h)
F operational cost ($)
HSUi /CSUi hot/cold startup cost of unit i ($)
f th thermal fuel cost function ($)
T on down
i /T i minimum up/down time of unit i. (h)
E environment emission function (ton)
T cold
i cold start hours of unit i. (h) fF normalization value of operational cost
SRt spinning reserve at time t (MW) Fmin /Fmax minimum/maximum operational cost in Pareto set
Lt system demand at time t (MW) ($)
PW V
t /P t predictive generation output of wind/photovoltaic fE normalization value of environment emission
power at time t (MW) Emin /Emax minimum/maximum environment emission in Pareto
x1j /x2j /x3j /x4j /x5j /x6j power generation coefficients of the jth set (ton)
hydropower stations
ramp B. Nomenclature for algorithmsa. constants
Pi generation output ramping up/down rate of thermal
power i (MW/min) N P /N D number of populations/number of dimensions
P min /P max minimum/maximum generation of thermal power i NL number of maximum iterations
i i
(MW)
b. variables
H min /Hmax minimum/maximum generation of hydropower
j j ! !p
station j (MW) X/X position vector of a grey wolf/position vector of the
prey
Q min
j /Q max
j minimum/maximum water discharge of ! !
r 1/ r 2 random vectors in [0, 1]
hydropower station j (m3) ! !
a components of a are linearly decreased from 2 to
V min
j /V max
j minimum/maximum reservoir
0 over the course of iterations
storage volume of hydropower station j (m3)
begin
Vj /V End
j initial and terminal reservoir c. abbreviations
storage limits of hydropower station j (m3) HTRPS Hydro-Thermal-RE power systems
t time interval (h) CS-HTPRS Multi-objective Complementary Scheduling of
Hydro-thermal-RE power system
c. Binary variable NSGA-III: An Evolutionary Many-Objective Optimization
Ii;t state of unit i at time t Algorithm Using Reference-Point-Based Non-
c. Integer variables dominated Sorting Approach
T off on MOHOAs Multi-objective Heuristic Optimization Algorithms
i;t /T i;t continuously off/on time of thermal power i at time t
NPGA Niched Pareto Genetic Algorithm
(h)
NSGA Non-dominated Sorting Genetic Algorithm
SUi;t =SDi:t Startup/shutdown cost of thermal power i at time t
SPEA Strength Pareto Evolutionary Algorithm
($)
MODE Multi Objective Differential Evolution
MOGSA Multi Objective Gravitational Search Algorithm
d. Continuous variable
MOBC Multi Objective Big Crunch
Pi;t output power of thermal power i at time t (MW)

[15,16], which is a significant advantage in solving complicated EED good performance in dealing with EED problems, premature
problems. The commonly used MOHOA techniques for solving EED convergence during the iteration process, as well as shortcomings
problems are summarized as the niched Pareto genetic algorithm in balancing global exploration and local exploitation, still exist in
[17], multi-objective adaptive clonal selection algorithm [18], multi these stochastic search algorithms. To enhance the performance of
objective evolutionary algorithm [19], strength Pareto evolutionary optimization, the latest efforts made by researchers are new
algorithm [20], adaptive group search algorithm [21], multi objec- MOHOAs for solving EED problems, including NSGA-III [24], the
tive differential evolution [22], and multi-objective particle-swarm multi-objective gravitational search algorithm (MOGSA) [25],
optimization (PSO) [23]. Although these algorithms have achieved multi-objective bee colony (MOBC) [26], differential harmony
C. Li et al. / Energy 171 (2019) 241e255 243

search [37], modified harmony search [38], floating search space can simultaneously optimize two different types of variables in the
[39], and enhanced moth-flame optimization [40]. It is widely multiple objective model. (3) The impact of RE capacity is analyzed
accepted that there remains great potential to improve the existing in terms of cost savings and emissions.
commonly used MOHOAs or introduce new MOHOAs to achieve
better performance in solving complicated EED problems.
2. Mathematical formulations for complementary scheduling
With the variation of power grid composition, especially by the
of HTRPS
integration of RE power, EED problems have become more and
more complicated [27]. The EED of HTPRSs is much more complex
By analyzing the operation rules of hydropower stations and
than for thermal power and hydro-thermal power systems. The
thermal power plants while considering the economic and envi-
effect of RE should be modeled and evaluated for the safe and stable
ronmental effects of the power system, a complementary sched-
operation of a power system. In Ref. [24], the researcher solved the
uling model of HTRPS is established. To meet the complex
hydro-thermal-wind scheduling problem and wind uncertainty
constraints of different system modules, the rational planning of
while considering the cost caused by wind uncertainty. In Ref. [25],
cascade hydropower stations and traditional TPUs allows the multi
a short-term hydrothermal-wind EED problem was presented and
energy complementary system to achieve the overall optimal so-
an improved MOGSA was proposed to solve this problem. Zhou
lution when considering economic and environmental aspects. In
et al. [26] proposed the scheduling problem of a short-term eco-
the model, hydropower stations and RE power plants have power
nomic/environmental hydro-thermal-wind complementary sys-
generation priority, and the remaining equivalent load is borne by
tem considering the uncertainty of wind power, as well as various
the TPUs. After determining the running states of the TPUs, the
complicated non-linear constraints. In Ref. [41], the roles of wind
equivalent load distribution is carried out in equal increment rate.
power and solar photovoltaic (PV) systems in reducing risks in a
The schematic diagram of the HTRPS is shown in Fig. 1.
Brazilian hydro-thermal power system were investigated, while the
In previous studies, similar economic/emission models for a
shares of renewable electricity production that can be maintained
hybrid system have been developed [26,29]. In these models, the
in the hydro-thermal-RE power system and the corresponding risks
on/off status of TPUs was rarely considered; therefore they involve
were quantifiably analyzed. In Ref. [42], the complementary
pure continuous optimization. In this paper, discrete optimization
scheduling problem of a hydro-thermal-wind power system with
variables will be integrated into the traditional economic/emission
electrical vehicles was studied, and an improved multi-objective
model to develop a new hybrid optimization model for not only
PSO algorithm was proposed to solve this model.
load dispatch of hydropower generation units (HGUs) and TPUs,
The above studies have similar modeling aspects as follows: (1)
but also for determining TPUs' on/off status.
there are typically two objectives concerning economic and envi-
ronmental targets; (2) wind/solar power is usually simulated by a
certain function to provide inputs for the scheduling model; (3) a 2.1. Objective function
MOHOA is always used to optimally dispatch the load to hydro and
thermal power units (TPUs) by solving the objective functions. 2.1.1. Economic objective
Other typical studies related to EED applied to HTPRSs contribute Because renewable energy and hydropower do not consume
mainly with algorithms. For example, a modified PSO [28] was fuel, their generating cost exists mainly in the construction stage.
presented for solving the EED problem of an HTRPS. In Ref. [29], the Thus, for short-term scheduling problems, the operational costs of
proposed algorithm combined differentiated-particle swarm opti- both RE and hydropower generation can be ignored. The opera-
mization with a virus-evolutionary technique and exhibited tional cost of the hybrid power system is mainly caused by the
notable performance in solving the EED of an HTPRS consisting of operation of TPUs, which includes the fuel cost and the startup cost.
ten TPUs, seven hydro units, and a wind energy system. Although
previous studies on EED problems involving HTRPSs are valuable X NI h
NT X     i
and instructive, the on/off status of TPUs is seldom considered. min F ¼ Ii;t f th Pi;t þ Ii;t 1  Ii;t1 SUi;t (1)
However, the operational cost and emissions are affected not only t¼1 i¼1

by the dispatch of the power load among units, but also by the on/ Because there is a number of TPUs in the HTRPS, the on/off
off states of units [30,31]. Thus, the complementary scheduling status of TPUs is decisive in determining the operational cost.
model for HTRPSs needs to improve, in which economic and Therefore, the on/off status, represented by discrete variables Ii,t, is
environmental goals are the objectives and the on/off status as well considered in the objective function.
as load dispatch are optimization variables. This new comple- The fuel cost of the f th thermal power unit is obtained by the
mentary scheduling of HTPRPSs is a type of hybrid optimization trial line method, and the coal consumption curve of a thermal
problem in which continuous and discrete optimization variables power unit can be approximated by a quadratic function [26],
should be solved synchronously. For solving this type of optimi- which is defined as:
zation problem, development of corresponding multi-objective
hybrid optimization algorithms is also needed.   th  2
f th Pi;t ¼ ath th
i þ bi Pi;t þ ci Pi;t (2)
Motivated by the above discussion, we have been inspired to
study the short-term complementary scheduling of HTRPS (CS- The start-up cost is composed of the hot start cost and the cold
HTRPS) to build an economic/emission scheduling model and start cost. The detailed formula is:
multi-objective optimization algorithm that consider the on/off 8
state of TPUs and the load dispatch scheme of thermal and hydro < HSU ; if T down  T off down
þ T cold
i i i;t  T i i
units with RE integration. The main contributions of this paper are SUi;t ¼ (3)
: CSUi ; if T off down
i;t > T i þ T cold
i
summarized as follows. (1) A new multi-objective complementary
scheduling model of HTRPS is proposed using hybrid optimization
variables for determining both the on/off state and dispatch load of where HSUi and CSUi represent the hot start cost and cold start cost,
units. The discrete variables and continuous variables need to be respectively. For the ith TPU at time t, the start cost is different,
optimized synchronously. (2) A two-stage multi-objective hybrid defined as either HSUi or CSUi according to the judgment between
grey wolf optimizer (MOHGWO) is designed to solve the model. It the continuous off time (T off
i;t ) and the combinations of minimum
244 C. Li et al. / Energy 171 (2019) 241e255

Load

Reservoir 1 Reservoir 2

Photovoltaic Wind

Reservoir 3

Hybrid
power
Reservoir 4
system

Thermal generators

Fig. 1. Schematic diagram of the HTRPS.

down time (T down


i ) and cold start hours (T cold
i ). PW V
t and P t are the power of wind and photovoltaic power sys-
tems, respectively. In this paper, the efficient method LHS-CD [36] is
2.1.2. Emission objective adopted to generate different scenarios within a given band at each
A power system based on thermal power is driven by burning time period to simulate wind and PV uncertainty. The 24 h power
coal, so a large amount of carbon dioxide and other exhaust gases series of wind and PV power generation systems can be obtained by
will be discharged. In this paper, the effect of carbon dioxide production and reduction scenarios [27].
emissions on the environment is considered. Hydropower and
renewable energy represent clean energy, so there will be negli- (2) The hydropower station power output limit constraints are:
gible pollution and emissions during their operation. The emission
of carbon dioxide in the process of coal-fired power generation is Hmin  Hj;t  Hmax (7)
j j
related to the power of the TPUs, and it can be approximately
described as a quadratic function [29]. The emissions objective
function for the hybrid system is defined as: (3) Reservoir storage volume and water discharge constraints
are:
X NI h
NT X   2 i
min E ¼ Ii;t , aenv þ benv Pi;t þ cenv Pi;t (4)
i i i (  
t¼1 i¼1
Q min
j  Qj;t þ Sj;t  Q max
j
(8)
V min
j  Vj;t  V max
j

Thus, the sum of the water discharge and spillage of a hydro-


2.2. Constraints
power station is constrained between its minimum and maximum
reservoir water discharge bounds. At the same time, the reservoir
(1) Power balance constraints:
storage volume of hydropower station j also has to satisfy the upper
and lower discharge bounds.
X
NI X
NH
Pi;t þ Hj;t þ P W V
t þ P t ¼ Lt (5)
i¼1 j¼1 (4) The water balance constraint is:

where
NJ 
X 
 2  2 Vj;t ¼ Vj;t1 þ Rj;t  Qj;t  Sj;t þ Qk;ttkj þ Sk;ttkj (9)
Hj;t ¼ x1j Vj;t þ x2j Qj;t þ x3j Vj;t Qj;t þ x4j Vj;t þ x5j Qj;t þ x6j k¼1
(6) For hydropower station j, the reservoir storage Vj;t is determined
The total power of the system should meet the load demand, by the reservoir storage volume (Vj;t1 ), inflow/spillage (Rj;t /Sj;t ),
where the sum of the power outputs of the TPUs, the HGUs, PV, and and water discharge (Qj;t ) of the hydropower station and the sum of
wind power systems is equal to the system load demand. Hj;t is the spillage and water discharge of the upstream stations.
output of a hydropower station, which is a quadratic function
related to the reservoir storage volume of hydropower station j (5) The initial and terminal reservoir storage volumes constraint
(Vj;t ) and water discharge of hydropower station j at time t (Qj;t ). is:
C. Li et al. / Energy 171 (2019) 241e255 245

( 3.2. Fundamental operators for multi-objective optimization


Vj;0 ¼ V begin
j
(10) The multi-objective minimization problem is used as the
Vj;NT ¼ V End
j
explanation model, which can be generally described as:

Minimize : FðXÞ ¼ ðf1 ðXÞ; f2 ðXÞ; :::; fo ðXÞÞ


(6) Spinning reserve constraints of TPUs:
s: t: gi ðXÞ  0; i ¼ 1; 2; :::; n
hj ðXÞ ¼ 0; j ¼ 1; 2; :::; m (16)
X
NI
up
P max
i ,Ii;t  Leq
t þ SRt (11) xlow
k  xk  xk ; k ¼ 1; 2; :::; s
i¼1
The core of the multi-objective optimization problem is the
contradiction between the objective functions. The improvement of
(7) Minimum up/down time constraints of TPUs: one goal will cause another goal to deteriorate. Therefore, in the
( multi-objective optimization problem, we cannot reach a solution
T on on
i;t  T i in which all the target functions can reach their respective optimal
(12)
T off down
i;t  T i
values, and we can only coordinate and search among multiple
targets to reach the best possible compromise. At the same time,
the final optimization results no longer represent one solution to
(8) Generation limit and ramping up/down capacity constraints the single objective optimization. They represent an “optimal front”
of TPUs: consisting of a series of solutions (the Pareto front). There are
several theories pertaining to the Pareto frontier:
Ii;t ,P tem_min
i;t
 Pi;t  Ii;t ,P tem_max
i;t
(13) Theory 1. If the two objective function values of individual A are
less than the function values of individual B, then individual A is
where determined to dominate individual B. Conversely, individual B is
h i determined to dominate individual A; in other cases, neither indi-
P tem_min
i;t ¼ max P min
i ; Pi;t1  P ramp
i
,t (14) vidual dominates.

h i Theory 2. In the search space, according to the dominating rela-


tem max ramp tionship determined by Theory 1, all the non-dominated in-
P i;t ¼ min P max
i ; Pi;t1 þ P i ,t (15)
dividuals in the group form an external archive set.
Theory 3. The search space is meshed according to the maximum
and minimum of different targets.

3. Multi-objective hybrid grey wolf optimizer To develop a multi-objective version of GWO, the concepts of
archive and leader selection are introduced into the optimizer
As described in Section 2, the CS-HTRPS problem is a hybrid mechanism [33].
optimization problem with discrete and continuous optimization Concept of Archive The archive can be regarded as a container
variables. In order to solve this problem, a powerful multi-objective for Pareto point sets, and a maximum number of members are
hybrid optimization algorithm should be designed. The grey wolf allowed in the archive. In the iterative process of the algorithm,
optimizer (GWO) [32], proposed by Mirjalili in 2014, has been non-dominated solutions obtained are compared with the archive
confirmed as a powerful meta-heuristic algorithm. In this section, a points set. To ensure that the number of archive points does not
multi-objective hybrid grey wolf optimizer is proposed based on exceed the maximum value, we may need to remove some points
the GWO. by calculating the degree of congestion to make space for new non-
dominating points.
Selection mechanism The selection mechanism of the leader
agent for the multi-objective algorithm is different from the single
3.1. Grey wolf optimizer
objective GWO. For multi-objective optimization, the agent cannot
be judged only by simple comparison of numerical values, but is
The grey wolf optimizer [32] is inspired by the social behaviors
based instead on the Pareto front concept. The leader selection
of grey wolves, which follow a very strict social dominant hierarchy.
component chooses the least crowded segments of the search
There are four levels of grey wolves, which are alpha, beta, delta,
space and offers one of its non-dominated solutions as an alpha,
and omega. Different hierarchies influence all wolves in their
beta, or delta wolf.
respective ways. The alpha is the top of the hierarchy; it controls
most of the behavior of the wolf pack, such as hunting, migrating
territory, and allocating food. Usually the strongest wolf is the 3.3. Multi-objective hybrid grey wolf optimizer (MOHGWO)
alpha. The second hierarchy level is the beta, and they are advisers
to the alpha and assist the alpha in making decisions. Moreover, In this paper, a multi-objective hybrid grey wolf optimizer is
when the alpha becomes old, they can also be candidates for the top proposed for the complementary scheduling of HTRPS on the basis
hierarchy. The omega level grey wolf plays the role of scapegoat. of MOGWO, in which continuous and discrete variables are coded
Omegas are dominated by other high-level wolves. This assists in for optimization. The basic concepts of GWO and multi-objective
maintaining the dominance structure. There are also delta level optimization theory have been introduced above. In this subsec-
wolves, which play the role of elders and caretakers in the pack. By tion, the principles and implementation of MOHGWO will be
analyzing the behavior of wolves in different levels, the GWO al- illustrated within the frame of GWO.
gorithm is constructed. In order to adapt to different problems, the The main improvement concerns the hybrid optimization
algorithm has been improved and expanded. The specific in- mechanism for different types of variables. The updating strategies
structions of GWO are detailed in Ref. [32]. of continuous variables and discrete variables are inconsistent.
246 C. Li et al. / Energy 171 (2019) 241e255

3.3.2. Movement strategies for discrete variables


For discrete variables, positions consist of a combination of 0s
and 1s. The simplest and most effective way is to use the sigmoid
function to judge the threshold. Then, depending on whether or not
the threshold is satisfied, the variable is equal to 1 or 0. The specific
formula is defined as follows:
8 1
>
!
>
<1 if ! ! !  > rand
X 1þ X 2þ X 3
X lþ1 ¼  (25)
>
> 1þe 3

:
0 otherwise

3.3.3. Operators for multi objective optimization


3.3.3.1. Operator for the archive. The archive is a set of Pareto points
Fig. 2. Encoding process for different variable types. with a limited number of members, which is accumulated by non-
dominated solutions obtained from iterations. By comparing the
current non-dominated solutions with members in the archive, a
When the population is updated, two types of variables can be new set of non-dominated solutions can be obtained. Once the
processed synchronously. The update mechanism for the different number exceeds the archive limit, a deletion operation should be
types of variables is shown in Fig. 2. conducted. The strategies for archive renewal are presented as
follows.
3.3.1. Movement strategy for continuous variables
The original GWO adopts continuous optimization variables, 1) The new member is dominated by at least one of the archive
and the movement strategy of the agents has been introduced in members. In this case, the solution should not be allowed into
Ref. [32]. The behavior of grey wolves can be abstracted by some the archive.
mathematical formulas. The position of an agent in the (lþ1) step in 2) The new solution dominates one or more solutions in the
the population is determined as follows: archive. In this case the dominated solution(s) in the archive
should be deleted and the new solution will enter the archive.
! !p ! ! 3) If neither the new solution nor archive members dominate each
X lþ1 ¼ X l  A , D (17)
other, the new solution should be added to the archive.
! ! !p !  4) If the archive is full, the grid mechanism should be run to re-
D ¼  C , X l  X l (18) arrange the segmentation of the objective space to find the
most crowded segment and delete one of its solutions. Then, the
! !p new solution should be inserted in the least crowded segment to
where X l is the position vector of a grey wolf and X l is the position
improve the diversity of the final approximated Pareto optimal
vector of the prey.
! ! front.
The vectors A and C are formulated as:
! ! ! !
A ¼ 2a,r 1  a (19) 3.3.3.2. Operator selection mechanism. The difference between the
selection mechanism of the leader agent for the multi-objective
! ! algorithm and the single objective GWO is that the agent in the
C ¼ 2, r 2 (20)
multi-objective hybrid optimization should be judged by the Pareto
front concept. The leader selection component chooses the least
! 2
a ¼ 2  l, L (21) crowded segments of the search space and offers one of its non-
N dominated solutions as an alpha, beta, or delta. The selection is
The hunting pattern of the wolves is directed by the alpha. performed by a roulette-wheel method with the following proba-
Meanwhile, beta and delta are also involved in hunting. Thus, the bility for each hypercube:
first three best solutions are chosen to be the hunting wolves, and
their current positions can update all wolves' positions. The cor- C
Pi ¼ (26)
responding formula is as follows: Ni

!a ! !a ! !b ! !b ! !d ! !d ! Here, c is a constant number greater than one and N is the number
D ¼  C , X  X ; D ¼  C , X  X ; D ¼  C , X  X 
of obtained Pareto optimal solutions in the ith segment.
(22) It should be noted from Eq. (23) that less-crowded hypercubes
have higher probability of suggesting new leaders, but there might
! !a ! !a ! !b ! !b ! !d ! !d be some special cases, as we have to choose three leaders:
X 1 ¼ X  A 1$ D ; X 2 ¼ X  A 1$ D ; X 3 ¼ X  A 1$ D
(23) 1) If there are three solutions in the least crowded segment, then
they are randomly assigned as alpha, beta, or delta solutions.
! ! ! 2) If there are less than three solutions in the least crowded hy-
! X þ X2 þ X3
X lþ1 ¼ 1 (24) percube, the second least crowded hypercube is also used to
3 choose other leaders. This scenario is repeated if the second
least crowded hypercube has only one solution, so the delta
C. Li et al. / Energy 171 (2019) 241e255 247

leader should be chosen from the third least-crowded 4.2. Initialization of individuals
hypercube.
The continuous and discrete optimization variables are encoded
With this method, we prevent MOHGWO from picking similar as position vectors of agents in the population. Each dimension of a
leaders for alpha, beta, or delta. Consequently, the search always position vector is generated randomly within the feasible domain.
moves toward the unexplored/unexposed areas of the search space The initialization method is expressed as:
because the leader selection mechanism favors the least crowded (
hypercubes and offers leaders from different segments if there are Qj;t ¼ Q min
j þ ðQ max
j  Q min
j Þ,rand
(28)
not enough leaders (less than three) in the least-crowded segment. Ii;t ¼ 0 or 1
Based on the discussion above, the MOHGWO is designed and
the pseudo code of the MOHGWO algorithm is shown in Fig. 3. We select the appropriate initialization method according to the
type of variable. For the discharge of hydropower stations, a nu-
merical value is generated randomly between the maximum and
minimum discharge values. For the running state of TPUs, a random
4. Implementation of the solution to the CS-HTRPS problem
selection of 0 or 1 sets the initial state of a TPU.

In this section, the implementation of the proposed MOHGWO


4.3. Constraints handling
algorithm for solving the CS-HTRPS problem, which considers
complex multiple constraints, is introduced. In particular, the cor-
After population initialization and iterative updating, there is a
responding repair strategies for the constraints are described in
significant possibility that the constraints of the model may not be
detail.
satisfied. Therefore, constraints are needed to restrict the popula-
tion to the scope of the feasible domain. The solution process for the
CS-HTRPS is divided into two parts, calculation of the hydropower
4.1. Structure of individuals station sub-system, and TPU system calculation. Each sub-system
has its own related constraints. Then the hybrid system is
For the CS-HTRPS problem, the discharge capacity of the hy- coupled by power balancing constraints.
dropower station(s) and the running state of the TPUs are taken as
the optimization variables. The expression for the combined opti- 4.3.1. Constraints of hydropower stations
mized variable is formulated as follows: In the hydropower sub-system, hydraulic constraints are the
most important. There are not only the discharge constraints of the
2 3
Q1;1 Q1;2 / Q1;m I1;1 I1;2 / I1;n hydropower stations, but also the constraints of reservoir capacity.
6 Q2;1 Q2;2 / Q2;m I2;1 I2;2 / I2;n 7 Handling the relationship between the two constraints is the
X¼6
4 «
7 (27)
« 1 « « « 1 « 5 principal requirement. The steps for handling the constraints of
QNP ;1 QNP ;2 / QNP ;m INP ;1 INP ;2 / INP ;n hydropower stations are described as follows:

The dimensions of the variable are n þ m. Q represents the Step 1: Calculate the violation degree of the initial and terminal
continuous variables, which represent the discharge of the cascade reservoir storage volumes DV ¼ V end  V begin
j j
hydropower stations; I represents the binary variables, which Step 2: Set a moderate error threshold D1 , and judge whether
represent the running state of each TPU. the jDVj > D1 condition is satisfied. If it is satisfied, go to step 3;
otherwise go to step 7.
Step 3: Calculate the average violation degree avgV ¼ jDTVj
Step 4: For each time period, update the discharge using Qj;t ¼
Qj;t þ avgV. If Qj;t > Q max
j , limit the discharge to the upper
boundary Qj;t ¼ Q max
j ; if Q min
j;t < Q j , limit the discharge to the
min
lower boundary Qj;t ¼ Q j .
Step 5: Calculate the corresponding capacity according to the
new discharge.
Step 6: Calculate the violation degree DV ¼ V end j  V begin
j
. Go
back to step 2.
Step 7: Set a very small error threshold D2 , and judge whether
the jDVj > D2 condition is satisfied. If it is satisfied, go to step 8;
otherwise go to step 11.
Step 8: Select a period of time t at random and update the
discharge Qj;t ¼ Qj;t þ DV. If Qj;t > Q max j , limit the discharge to
the upper boundary Qj;t ¼ Q max j ; if Qj;t < Q minj , limit the
discharge to the lower boundary Qj;t ¼ Q min j .
Step 9: Calculate the corresponding capacity Vj;t according to
the new discharge.
begin
Step 10: Calculate the violation degree DV ¼ V end j  Vj . Go
back to step 7.
Step 11: Return the discharge of the hydropower stations.

4.3.2. Constraints of TPUs


Fig. 3. Pseudo code of the MOHGWO algorithm. Calculation of the TPUs is a unit combination problem, which
248 C. Li et al. / Energy 171 (2019) 241e255

includes an internally- and externally-optimized problem. Some Step 4: Update the value of l. If diff t is positive, then
important procedures with respect to constraints handling are as lmin ¼ l and l ¼ ðlmax þ lmin Þ=2. If diff t is negative, then
follows: lmax ¼ l and l ¼ ðlmax þ lmin Þ=2.
Step 5: Go back to step 2 and repeat steps 2 through 5
(1) For the external sub-problem, the on/off status of the units is until the load dispatching of this interval is complete.
the optimal variable. Optimal variables can be adjusted by
the constraints discussed in Section 2.2. Consideration of
spinning reserve constraints and minimum up/down time
constraints must be comprehensive. 4.4. Procedure
(2) For the internal sub-problem, the dynamic l-iteration syn-
thesizing dichotomy is applied to satisfy the power balance Based on the above discussion, the process of CS-HTRPS prob-
and the ramping up/down capacity constraints. Dichotomy lem solving with MOHGWO is described as follows, and the flow-
can provide a good solution for the problem of the death chart is shown in Fig. 4.
cycle of the l-iteration:
Step 1: Initialize the equal differential increment rate l of Step 1: Set up the corresponding basic operation parameters of
each interval and set the range values ½lmin ; lmax  of l. hydropower station(s) and thermal power plant(s); the initial-
Step 2: For each interval t, calculate Pi;t using the equal ization algorithm parameters are population size N P , population
differential increment rate principle to handle the con- dimension ND , and maximum iterations NL . Set the number of
straints of the power output. If the power output is current iterations as l ¼ 1.
Pi;t  Ii;t P tem_min
i;t
, then Pi;t ¼ P tem_min
i;t
. If Pi;t  Ii;t P tem_max
i;t
, Step 2: Initialize populations based on the related constraints.
tem_max
then Pi;t ¼ P i;t . Here, P tem_min
i;t ¼ max½P min
i ; For the hydropower stations, the model fully considers factors
ramp
Pi;t1  P i ,t and P tem i;t
max
¼ min½P max i ; Pi;t1 þ such as hydraulic connections. For TPUs, the most basic unit
ramp
Pi ,t. commitment constraints are also considered.
P I
Step 3: Calculate the deviation, diff t ¼ N net
i¼1 Pi;t  ðLt  Step 3: Calculate the economic cost and emission of each pop-
PNH
H
j¼1 j;t Þ. If diff t converges, then the load dispatching of ulation and determine the non-dominated ranking of pop-
this interval is complete; otherwise, go to step 4). ulations. Update the set of non-dominated solutions using the
theories presented in Section 3.2.

Start
Start handling
thermal uints
constrains
Start handling Initialize the population and
thermal uints parameters of algorithm; set l=1
constrains Handle the up spinning reserve constraints
NI

Deal with constraints U t ,s Pi max I i ,t , s Pt ,Ws Pt ,Vs PGt , s PPt , s / Lt SRt , s


The discharge volume Q is i 1
generated randomly and the
corresponding storage capacity V is
calculated by hydraulic connection. No Start idle unit and
Calculate the economic cost and
emission of each population Ut,s 0 handle minimum up
time constrain
Whether the reservoir
capacity constraints at the beginning and the
end are satisfied
Yes Update set of non-dominated Handle the down spinning reserve constraints
No solutions NI
Dt , s Pi min I i ,t , s Pt ,Ws Pt ,Vs PGt , s PPt , s / ( Lt SRt , s )
end ini i 1
V V
avgV
NT shut down the excess
No
units and handle
Population evolution upgrading Dt,s 0
minimum down time
Yes constrain
V V avgV

Yes No calculate the deviation


Are variables continuous? Load dispatching by
Calculate the corresponding storage NI
interation
diff t , s Pi ,t , s Lnet
capacity V is calculated by i 1
t,s

hydraulic connection. Processing Processing


by formula x by formula x No
difft , s 0.001 Adjust the value of

End
Yes
l<Nl
Is network No
constraint is
No satisfied ?

End
End

Fig. 4. Flow chart of solution for CS-HTRPS problem.


C. Li et al. / Energy 171 (2019) 241e255 249

Step 4: Population evolution upgrading. For different types of Table 2


variables, use the corresponding update method. Meanwhile, Emission coefficients of TPUs.

related constraint processing needs to be performed. Thermal power aðkg=hÞ bðkg=MWhÞ gðkg=ðMWÞ2 hÞ
(1) For hydropower stations, the discharge volume Q is the
1 130 2.86 0.022
optimization variable. Take the initial and terminal reservoir 2 132 2.72 0.02
storage volumes constraint as the outermost control 3 137.7 2.94 0.044
constraint. Then, handle reservoir storage volume and water 4 130 2.35 0.058
discharge constraints based on time intervals. 5 125 2.36 0.065
6 110 2.28 0.08
(2) For TPUs, the problem can be divided into two layers, in-
7 135 2.36 0.075
ternal and external. For the external sub-problem, the on/off 8 157 1.29 0.082
status of TPUs is the variable to optimize. Consideration of 9 160 1.14 0.09
spinning reserve constraints and minimum up/down time 10 137.7 2.14 0.084
constraints must be comprehensive. For the internal sub-
problem, the dynamic-iteration synthesizing dichotomy is
applied to satisfy the power balance and the ramping up/ 105
1.69
down capacity constraints. NSGA-3
Step 5: if l < NL , then l ¼ l þ 1, and go back to step 3. Else the 1.68 MOHPSO
procedure is complete, output the Pareto set of the EED MOHGWO
problem. 1.67

Emission objective(kg)
1.66
4.5. Scheme selection method
1.65
The scheme selection is important to the results of multi-
1.64
objective problems, because a set of non-dominated schemes, in
accordance to the Pareto front, are obtained after the solution. To 1.63
select a scheme that achieves useful results for CS-HTRPS, a
detailed analysis should be carried out. Considering that the 1.62
operating costs and environmental emissions of TPUs are different
dimensional data, it is reasonable to normalize the objective values 1.61
to select a suitable scheme. The normalization method is as follows:
1.6
F  Fmin E  Emin 5.3 5.32 5.34 5.36 5.38 5.4 5.42
fF ¼ ;f ¼ (29)
Fmax  Fmin E Emax  Emin Economic objective($) 105
The sum of the two different types of normalized objective Fig. 5. Pareto fronts obtained by three different algorithms.
values is used as a measure of the selection criteria, which con-
siders the economic and environmental aspects, where the point Table 3
corresponding to the minimum value is selected. The formula is Optimal Pareto schemes obtained by the different algorithms.
expressed as follows:
scheme MOHPSO NSGA-3 MOHGWO

minðfF þ fE Þ (30) F ($) E (kg) F ($) E (kg) F ($) E (kg)

1 533918 166521 533312 166636 532610 166325


2 534577 166373 533662 166487 532615 166291
3 534681 166299 534163 165780 532641 166273
5. Case studies 4 534805 166280 534185 165755 532643 166008
5 535086 165571 534410 165403 532666 165966
6 535420 165169 534427 165360 532682 165960
To validate the effectiveness of the proposed model and algo- 7 535959 164731 534483 165353 532688 165946
rithm for CS-HTRPS problems, numerical experiments based on a 8 536949 164540 534490 165332 533058 165553
typical HTRPS are designed. This test system contains four cascade 9 537059 163828 534537 165287 533178 165082
hydropower stations, ten conventional TPUs, a wind generator, and 10 537372 163666 535092 165208 533532 164686
11 537662 163469 535788 164367 533558 164679
a photovoltaic generator. The coefficients for the TPUs are taken
12 538099 163376 535895 164019 533569 164678
from Ref. [35] and are presented in Tables A1 and A2 in the Ap- 13 538442 162834 535895 164019 534005 164545
pendix. The detailed coefficient data of the hydropower stations are 14 539079 162687 536248 163471 534728 163510
taken from Ref. [34] and are listed in Table A3 in the Appendix. The 15 539168 162598 537113 163004 535134 163065
loads of the complemental system are shown in Table 1, and the 16 539242 162515 537200 162971 535273 162586
17 540266 162370 538055 162943 535611 162173
emission coefficients of the TPUs are shown in Table 2. The 18 540424 162234 538055 162943 535630 162169
scheduling time horizon is one day, comprising 24 t intervals of 1 h 19 540610 162162 538097 162328 536126 162011
each. The coefficients of MOHGWO are as follows: the population 20 541583 161951 538581 162170 537579 161733

Table 1
Load demand.

time(h) 1 2 3 4 5 6 7 8 9 10 11 12

Load (MW) 1200 1248 1120 1040 1072 1280 1520 1616 1744 1728 1760 1840
time(h) 13 14 15 16 17 18 19 20 21 22 23 24
Load (MW) 1776 1648 1616 1696 1680 1792 1712 1680 1456 1376 1360 1280
250 C. Li et al. / Energy 171 (2019) 241e255

10 5 10 5
5.34 1.625
Economic objective($)
5.338 Emission objective(kg) 1.624

5.336 1.623

5.334 1.622

5.332 1.621

5.33 1.62

5.328 1.619

5.326 1.618

5.324 1.617

5.322 1.616

5.32 1.615
MOHPSO NSGA-3 MOHGWO

Fig. 6. Optimal trends for each algorithm.

size is N P ¼ 500, population dimension is ND ¼ 336, and maximum to the true Pareto front, so it can be concluded from the figure that
iterations are NL ¼ 500. MOHGWO can deal with the hydro-thermal complemental system
In the first part of the experiments, the renewable power model well.
sources are not taken into account, and the HTRPS simplifies to a The detailed values of the optimal Pareto schemes are shown in
traditional hydro-thermal complemental system for algorithm Table 3. It is known from these data that the change trend of the
comparison. Three different multi-objective optimizers are economic objective value is inversely proportional to the emissions
compared. In addition, the detailed optimization strategy obtained objective value. The corresponding optimal value appears in bold-
by the best methodology is displayed and correlation constraints face in Table 3. The two-objective value change trend of the most
are verified. optimal scheme is shown in Fig. 6. Meanwhile, the most economical
In the second part of the experiments, wind power and solar scheme and the most environment-friendly scheme of the different
power sources are considered and the system is a complete HTRPS, algorithms are compared. Under the same model condition, the
for which the scheduling results are provided and the effect of RE MOHGWO algorithm can obtain the optimal value in different
capacity is studied based on MOHGWO. objective dimensions, indicating that the proposed MOHGWO has a
much better performance in searching for the optimal value to
5.1. Methodology comparison for hydro-thermal system

To verify the proposed model and algorithm, the hydro-thermal 1


power system consisting of four cascade hydropower stations and
10 TPUs, is evaluated by numerical experiments. Hydropower sta- 0.95
tions have power generation priority, and the remaining equivalent
Sum of normalization

load is borne by the TPUs. To verify the effectiveness of the 0.9


MOHGWO algorithm, the same model is evaluated with other
excellent algorithms, such as NSGA-3 and MOHPSO. To avoid the 0.85
impact of parameter setting, the corresponding population size and
maximum iterations are set to be consistent with the MOHGWO 0.8
algorithm. The parameters of NSGA-3 are as follows: population
size is 100, max iterations are 500, crossover probability is 0.5,
0.75
mutation probability is 0.01, crossover index is 30, and mutation
index is 20. The parameters of MOHPSO are as follows: population
0.7
size is 100, max iterations are 500, the value of acceleration co-
efficients c1 and c2 is 2, and inertia weight u is 1. The Pareto fronts
obtained by the different algorithms are shown in Fig. 5. By 0.65
0 5 10 15 20
analyzing the distribution of the Pareto fronts, there are obvious
Scheme
stratifications among the different algorithms. Generally, the Pareto
front appearing near the lower left side indicates closer proximity Fig. 7. MOHGWO scheme selection.
C. Li et al. / Energy 171 (2019) 241e255 251

solve a hydro-thermal complemental problem than NSGA-3 and as the target. By analyzing the bar graph in Fig. 7, the 17th scheme is
MOHPSO. selected as the representative for a detailed discussion.
After obtaining the Pareto front, the results of MOHGWO are The detailed results of the selected scheme are presented in
analyzed in detail. Because the dimensions of the economic Table 4, and the power outputs of the hydropower stations and
objective and emissions objective are different, the MOHGWO TPUs, and the discharges of the hydropower stations are exhibited.
schemes are normalized by the method discussed in Section 4.5 to The optimization variables, including the discharge of hydropower
select a specific scheduling strategy. Under the comprehensive stations and on/off status of the TPUs are also listed. For TPUs, a
consideration of the two objective factors, the value of ðfF þ fE Þ for zero power output indicates the off status. From the results in
each scheme is calculated, and the results are shown in Fig. 7. The Table 4, the constraint conditions of the model can be checked. The
scheme that corresponds to a minimum value of ðfF þ fE Þ is selected power output limit constraints and water discharge constraints are

Table 4
Detailed parameters of the power system based on MOHGWO results.

Time(h) Hydro output (MW) Hydro water discharge Thermal output (MW)

H1 H2 H3 H4 Q1 Q2 Q3 Q4 U1 U2 U3 U4 U5 U6 U7 U8 U9 U10

1 80 66 29 134 9 9 23 6 455s 346 0 0 90 0 0 0 0 0


2 71 51 21 130 7 6 24 6 455 435 60 0 25 0 0 0 0 0
3 74 52 26 126 8 6 22 6 455 305 57 0 25 0 0 0 0 0
4 71 58 37 121 7 7 19 6 455 175 97 0 25 0 0 0 0 0
5 73 56 42 117 8 7 17 6 455 174 130 0 25 0 0 0 0 0
6 75 62 29 133 8 7 20 6 455 304 130 0 91 0 0 0 0 0
7 72 67 42 165 8 8 17 7 455 434 130 60 94 0 0 0 0 0
8 74 71 35 229 8 9 19 12 455 455 130 120 47 0 0 0 0 0
9 74 63 42 282 8 8 16 17 455 455 130 130 88 0 25 0 0 0
10 85 59 27 298 10 7 20 19 455 455 130 130 64 0 25 0 0 0
11 73 66 40 285 8 8 17 17 455 455 130 130 81 20 25 0 0 0
12 87 74 42 299 10 10 16 19 455 455 130 130 122 20 25 0 0 0
13 92 67 44 294 10 9 16 18 455 455 130 130 89 20 0 0 0 0
14 87 72 39 294 10 10 18 18 455 416 130 130 25 0 0 0 0 0
15 79 68 48 287 8 9 14 17 455 393 130 130 25 0 0 0 0 0
16 71 60 44 277 7 7 18 16 455 455 130 130 49 0 25 0 0 0
17 68 63 38 272 7 8 20 15 455 455 130 130 43 0 25 0 0 0
18 72 74 49 293 7 10 16 18 455 455 130 130 89 20 25 0 0 0
19 64 58 50 283 6 7 16 17 455 455 130 130 67 20 0 0 0 0
20 89 74 53 299 9 11 11 19 455 405 130 130 25 20 0 0 0 0
21 72 58 53 298 7 8 11 19 455 365 130 0 25 0 0 0 0 0
22 73 73 55 301 7 11 11 20 455 394 0 0 25 0 0 0 0 0
23 95 73 58 297 11 11 13 20 455 357 0 0 25 0 0 0 0 0
24 90 66 58 290 10 10 15 19 455 320 0 0 0 0 0 0 0 0

Hydro station 1 Hydro station 2 Hydro station 3 Hydro station 4


180

160
Reservoir volume(104m 3)

140

120

100

80

60
0 5 10 15 20 25
Time(h)
Fig. 8. Water-level fluctuation in detail for each hydro station.
252 C. Li et al. / Energy 171 (2019) 241e255

5
10
1.5
50% RE output
75% RE output
1.45 100% RE output

1.4

Emission objective(kg)
1.35

1.3

1.25

1.2

1.15

Fig. 9. Cumulative power generation results. 1.1


3.8 4 4.2 4.4 4.6 4.8 5
Economic objective($) 5
10
checked and it is confirmed that they have not been violated. In
Fig. 10. Pareto fronts obtained for different RE capacities.
addition, the other related variables of the hydropower station are
also within feasible regions. Fig. 8 shows the water-level fluctuation
at different periods of the hydropower stations, for which the thus plays a positive role in both economic and environmental
maximum reservoir volume is 180  104 m3 and the minimum is aspects. Three typical schemes are selected from the corresponding
60  104 m3, and the reservoir volume of different hydro stations is solution sets through the normalization method, and the objective
satisfied within this confining range. Water level is also controlled values are shown in Table 6. From this table, it can be seen that the
within the confining range. For TPUs, the minimum up/down time operational costs and pollution emissions are reduced as RE ca-
constraints, ramping up/down capacity constraints, and generation pacity increases. The effect of RE on economic and emissions ob-
limit constraints are also satisfied. Fig. 9 clearly shows the load jectives is positive and significant.
balancing constraints of the system and the proportion of the Moreover, the detailed scheduling scheme of the CS-HTRPS
output from the different units. The total hydro output and thermal model with 100% RE capacity is shown in Table 7. The power
output meet the demand of load. In addition, hydropower station 4, output limit constraints and water discharge constraints are
thermal power 1, and thermal power 2 take up most of the load of checked and confirmed and the other related variables of the hy-
the system. All results verify the feasibility of the proposed thermal dropower station and TPUs also fall within the scope of the feasible
complemental model. region.
Because the priority of RE generation is higher than the thermal
5.2. Effects evaluation on RE capacity power generation, RE power plants will take part of the load,
allowing some TPUs to shut down to reduce both the cost of the
In this section, the CS-HTRPS problem is studied by applying operation and the emission of pollutants. The total running period
MOHGWO to solve the problem. An HTRPS is built based on the of the TPUs is compared when RE output changes from 0% to 100%,
hydro-thermal system by considering wind power and photovoltaic and we can see that the overall running time trend of the thermal
power. The forecast output power of RE sources is shown in Table 5. power units is reduced in Fig. 11. Because the 1st unit and 2nd unit
For simulating different RE capacity, the forecast RE values are are large-capacity TPUs with better efficiency and economic per-
varied by multiplying certain proportions with the basis values. To formance, they remain in the running state for longer periods of
illustrate the impact of different RE capacity on the hybrid system,
three cases: 50%, 75%, and 100% RE capacities are analyzed. The CS-
HTRPS model with different RE capacity levels is solved by Table 6
MOHGWO and the results are compared. Operational costs and pollutant emissions for different RE capacities.

The Pareto fronts at different capacity levels are shown in Fig. 10. 50% capacity 75% capacity 100% capacity
From the results, the Pareto front gradually moves to the lower left F ($) E (kg) F ($) E (kg) F ($) E (kg)
as the RE capacity grows, meaning that the solutions with high RE
470562 140073 441349 128857 414479 115635
inputs dominate those with lower RE inputs. The RE integration

Table 5
Generated output of renewable energy generators.

time(h) 1 2 3 4 5 6 7 8 9 10 11 12

PW
t (MW)
152 240 264 288 280 296 352 368 280 200 336 304
PV
t (MW)
0 0 0 0 0 0 4 12 36 64 80 88
time(h) 13 14 15 16 17 18 19 20 21 22 23 24
PW
t (MW)
312 272 256 96 8 32 40 16 4 200 280 192
PV
t (MW)
96 88 80 64 36 12 4 0 0 0 0 0
C. Li et al. / Energy 171 (2019) 241e255 253

Table 7
Detailed parameters considering RE.

Time Hydro output (MW) Hydro water discharge Thermal output (MW) RE (MW)

H1 H2 H3 H4 Q1 Q2 Q3 Q4 U1 U2 U3 U4 U5 U6 U7 U8 U9 U10 PW PV

1 75 65 19 134 8 9 25 6 455 299 0 0 0 0 0 0 0 0 152 0


2 64 51 18 130 6 6 24 6 455 291 0 0 0 0 0 0 0 0 240 0
3 85 55 45 127 10 7 17 6 382 161 0 0 0 0 0 0 0 0 264 0
4 75 55 27 123 8 7 21 6 322 150 0 0 0 0 0 0 0 0 288 0
5 90 62 36 120 11 8 19 6 333 150 0 0 0 0 0 0 0 0 280 0
6 63 56 37 134 6 6 19 6 455 240 0 0 0 0 0 0 0 0 296 0
7 76 57 42 161 8 7 17 7 455 313 60 0 0 0 0 0 0 0 352 4
8 68 56 45 236 7 6 16 13 455 256 120 0 0 0 0 0 0 0 368 12
9 90 77 40 294 11 10 18 19 455 342 130 0 0 0 0 0 0 0 280 36
10 89 72 40 292 11 9 18 18 455 362 130 0 25 0 0 0 0 0 200 64
11 74 57 34 276 8 7 19 16 455 293 130 0 25 0 0 0 0 0 336 80
12 76 74 36 279 8 9 18 16 455 374 130 0 25 0 0 0 0 0 304 88
13 88 68 36 293 10 9 19 18 455 273 130 0 25 0 0 0 0 0 312 96
14 80 58 38 287 8 7 18 17 455 215 130 0 25 0 0 0 0 0 272 88
15 87 54 41 285 9 6 18 17 455 204 130 0 25 0 0 0 0 0 256 80
16 73 76 39 299 7 10 19 19 455 334 130 60 71 0 0 0 0 0 96 64
17 65 66 50 296 6 8 14 19 455 429 130 120 25 0 0 0 0 0 8 36
18 81 68 48 299 8 8 16 19 455 455 130 130 57 0 25 0 0 0 32 12
19 82 71 52 299 8 9 13 19 455 400 130 130 25 0 25 0 0 0 40 4
20 80 74 51 289 8 10 15 18 455 405 130 130 25 0 25 0 0 0 16 0
21 68 80 54 298 7 12 13 19 455 455 0 0 43 0 0 0 0 0 4 0
22 60 80 55 298 6 12 12 20 358 325 0 0 0 0 0 0 0 0 200 0
23 79 73 56 292 8 11 11 19 386 195 0 0 0 0 0 0 0 0 280 0
24 73 67 58 284 7 10 12 19 455 152 0 0 0 0 0 0 0 0 192 0

25
without RE
50% RE
75% RE
20 100% RE
Total running time(h)

15

10

0
1 2 3 4 5 6 7 8 9 10
Thermal unit
Fig. 11. Total running time of the units under various scenarios.

time. The 8th to the 10th units are less economical TPUs, and 6. Conclusions
remain in the off state because the load can be shared by other,
more efficient power sources. For most TPUs, the running periods In this paper, the problem of complementary scheduling of
will decrease with rising RE capacity. hydro-thermal-renewable power systems has been studied. A new
To evaluate the influence of different capacities on the operation economic/emission model for CS-HTRPS is constructed, in which
of the hydropower stations, the discharge volumes of the four hy- discrete variables and continuous variables are optimized. In order
dropower stations during the scheduling periods are determined, to solve the hybrid multi-objective problem, the MOHGWO algo-
and the sum and the variance of the discharge flow are compared. rithm, which embeds a hybrid optimization mechanism, is pro-
From Table 8, the variance of the discharge volume of the hydro- posed. MOHGWO can simultaneously optimize two different types
power stations increases with the increase of RE capacity, It means of variables in the multiple objective model. To verify the feasibility
that the discharge volume of hydropower stations needs to be and effectiveness of the proposed CS-HTRPS model and the
adjusted in wider zone, in order to adapt to the RE fluctuation. The MOHGWO algorithm, numerical experiments were conducted on a
total discharge flow of all hydropower stations in all periods show typical hybrid power system with four cascade hydropower sta-
an increasing trend as the rising of RE capacity. tions, ten TPUs, and two types of RE power sources. Based on the
experimental results, the following conclusions can be drawn:
254 C. Li et al. / Energy 171 (2019) 241e255

Table 8
Hydropower station discharges for different RE capacities.

time 50% capacity 75% capacity 100% capacity

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

1 8.52 7.72 23.67 6.13 8.67 7.46 22.71 6.01 8.05 8.61 24.61 6.15
2 9.50 7.59 21.13 6.11 8.12 6.50 25.27 6.05 6.29 6.27 23.93 6.06
3 7.37 6.70 23.69 6.09 9.85 6.86 19.10 6.06 9.57 6.80 17.14 6.12
4 10.26 7.46 17.52 6.19 8.61 7.44 18.78 6.03 7.77 6.59 20.97 6.14
5 5.68 6.89 20.53 6.06 9.04 7.51 18.08 6.09 10.72 7.51 18.58 6.35
6 7.86 6.39 17.23 6.16 8.48 7.38 17.65 6.04 6.38 6.45 18.54 6.10
7 7.93 7.29 17.80 6.05 8.82 8.25 18.08 6.25 8.30 6.54 16.98 7.01
8 8.49 7.86 18.42 13.98 7.92 7.28 17.58 10.97 7.03 6.49 16.41 12.69
9 7.59 8.14 18.81 17.09 7.41 7.64 17.92 17.42 11.05 10.14 18.01 18.58
10 7.95 9.86 18.08 17.42 7.97 7.65 17.06 17.60 10.71 9.36 17.79 18.29
11 7.56 7.03 17.09 17.25 6.91 7.44 19.06 17.97 7.82 6.84 18.85 16.12
12 10.13 7.75 18.62 18.31 7.53 7.24 16.53 17.57 7.92 9.46 18.35 16.37
13 8.32 9.38 16.00 18.76 9.08 8.66 20.48 18.02 9.97 8.56 18.82 18.22
14 8.87 8.48 18.51 18.06 6.55 9.02 18.57 17.06 8.45 6.95 18.47 17.49
15 6.94 7.53 17.11 17.69 8.50 8.49 18.15 18.95 9.49 6.20 17.61 17.12
16 7.19 9.89 18.06 18.12 8.17 8.73 18.32 17.43 7.26 9.50 18.78 18.85
17 8.93 9.53 16.48 18.24 8.26 8.37 16.18 19.58 6.16 7.97 14.22 18.55
18 10.27 9.93 16.66 17.76 8.49 8.82 15.91 19.35 8.23 8.40 16.39 18.93
19 9.85 8.58 14.39 18.47 7.39 9.63 11.57 19.35 8.39 9.04 12.57 19.07
20 6.68 9.02 14.07 18.89 8.09 10.55 13.68 18.37 8.15 9.88 15.01 17.76
21 5.93 10.27 12.24 19.22 6.27 9.98 12.16 17.47 6.56 11.56 13.37 18.88
22 6.52 11.03 11.40 19.39 8.65 11.66 10.31 19.48 5.61 12.15 12.42 19.62
23 8.44 9.14 14.09 19.79 8.02 10.72 14.19 19.37 7.96 10.80 11.20 19.27
24 8.19 8.53 12.37 19.42 8.20 8.72 14.65 18.99 7.14 9.95 12.28 19.08
VAR 1.61 1.51 9.64 30.90 0.64 1.66 10.75 31.75 2.16 3.03 11.00 29.99
SUM 43.67 44.79 46.18

(1) The comparison results show that the Pareto front obtained Table A2
by MOHGWO is superior to two competitors, which indicates Parameters of operational cost functions
that a strong and significant multi-objective optimization a b c CST HST Tcold
performance is achieved by the CS-HTRPS model.
1 1000 16.19 0.00048 9000 4500 5
(2) CS-HTRPS can be well solved by MOHGWO, where an 2 970 17.26 0.00031 10000 5000 5
optimal balance between economic and emission objectives 3 700 16.6 0.002 1100 550 4
can be reached and the obtained scheduling schemes fall 4 680 16.5 0.00211 1120 560 4
completely within the feasible domain. 5 450 19.7 0.00398 1800 900 4
6 370 22.26 0.00712 340 170 2
(3) The impact of RE capacity on the CS-HTRPS problem was 7 480 27.74 0.00079 520 260 2
analyzed, and the results show that the operational costs and 8 660 25.92 0.00413 60 30 0
pollution emissions are reduced as RE capacity increases. In 9 665 27.27 0.00222 60 30 0
addition, the discharge volume of hydropower stations needs 10 670 27.79 0.00173 60 30 0
to be adjusted over a wider zone to adapt to the RE
fluctuation.
Table A3
Acknowledgment Hydro power generation coefficients

Plant x1 x2 x3 x4 x5 x6
This work was supported by the National Natural Science
Foundation of China (51879111, 51679095) and the Applied 1 0.0042 0.42 0.03 0.9 10 50
2 0.004 0.3 0.15 0.14 9.5 70
Fundamental Frontier Project of Wuhan Science and Technology
3 0.0016 0.3 0.014 0.55 5.5 40
Bureau (2018010401011269). 4 0.003 0.31 0.0027 1.44 14 90

Appendix

Table A1
Parameters of thermal units
References

Unit P i ðMWÞ P i ðMWÞ Pi;ramp ðMW=hÞ Ti;on ðhÞ Ti;down ðhÞ Orig: stateðhÞ [1] Moradi-Dalvand M, Mohammadi-Ivatloo B, Najafi A, Rabiee A. Continuous
quick group search optimizer for solving non-convex economic dispatch
1 455 150 130 8 8 8 problems. Electr Power Syst Res 2012;93:93e105.
2 455 150 130 8 8 8 [2] Sajjadi SM, Sadeghi Yazdankhah A, Ferdowsi F. A new gumption approach for
3 130 20 60 5 5 5 economic dispatch problem with losses effect based on valve-point active
4 130 20 60 5 5 5 power. Electr Power Syst Res 2012;92:81e6.
5 162 25 90 6 6 6 [3] Nemati M, Braun M, Tenbohlen S. Optimization of unit commitment and
6 80 20 40 3 3 3 economic dispatch in microgrids based on genetic algorithm and mixed
7 85 25 40 3 3 3 integer linear programming. Appl Energy 2018;210:944e63.
8 55 10 40 1 1 1 [4] Zou D, Li S, Wang G-G, Li Z, Ouyang H. An improved differential evolution
9 55 10 40 1 1 1 algorithm for the economic load dispatch problems with or without valve-
point effects. Appl Energy 2016;181:375e90.
10 55 10 40 1 1 1
[5] Cavazzini G, Pavesi G, Ardizzon G. A novel two-swarm based PSO search
C. Li et al. / Energy 171 (2019) 241e255 255

strategy for optimal short-term hydro-thermal generation scheduling. Energy based on a non-dominated sorting genetic approach for hydro-thermal-wind
Convers Manag 2018;164:460e81. economic emission dispatching. Energy 2017;121:276e91.
[6] Chen Y, Wei W, Liu F, Mei S. Distributionally robust hydro-thermal-wind [26] Zhou J, Lu P, Li Y, Wang C, Yuan L, Mo L. Short-term hydro-thermal-wind
economic dispatch. Appl Energy 2016;173:511e9. complementary scheduling considering uncertainty of wind power using an
[7] Espinosa S, Cazco DA, Salcedo MY. Economic dispatch hydrothermal system enhanced multi-objective bee colony optimization algorithm. Energy Convers
with CO2 emissions constraints. IEEE Lat Am Trans 2017;15(11):2090e6. Manag 2016;123:116e29.
[8] Liang H, Liu Y, Li F, Shen Y. A multiobjective hybrid bat algorithm for com- [27] Wang W, Li C, Liao X, Qin H. Study on unit commitment problem considering
bined economic/emission dispatch. Int J Elec Power 2018;101:103e15. pumped storage and renewable energy via a novel binary artificial sheep al-
[9] Shaabani Ya, Seifi AR, Kouhanjani MJ. Stochastic multi-objective optimization gorithm. Appl Energy 2017;187:612e26.
of combined heat and power economic/emission dispatch. Energy 2017;141: [28] Damodaran SK, Kumar TKS, Sciubba E. Hydro-thermal-wind generation
1892e904. scheduling considering economic and environmental factors using heuristic
[10] Zhu Y, Wang J, Qu B. Multi-objective economic emission dispatch considering algorithms. Energies 2018;11:353.
wind power using evolutionary algorithm based on decomposition. Int J Elec [29] Liang RH, Liau YS, Chen YT. A virus-evolutionary differentiated-PSO approach
Power 2014;63:434e45. for short-term generation scheduling with uncertainties. Int Trans Electr
[11] Mason K, Duggan J, Howley E. A multi-objective neural network trained with Energy 2016;26:2288e307.
differential evolution for dynamic economic emission dispatch. Int J Elec [30] Wang B, Wang S, Zhou X, Watada J. Multi-objective unit commitment with
Power 2018;100:201e21. wind penetration and emission concerns under stochastic and fuzzy un-
[12] Ma H, Yang Z, You P, Fei M. Multi-objective biogeography-based optimization certainties. Energy 2016;111:18e31.
for dynamic economic emission load dispatch considering plug-in electric [31] Wang B, Wang S, Zhou XZ, Watada J. Two-stage multi-objective unit
vehicles charging. Energy 2017;135:101e11. commitment optimization under hybrid uncertainties. In: International con-
[13] Modiri-Delshad M, Rahim NA. Multi-objective backtracking search algorithm ference on genetic & evolutionary computing; 2016. p. 128e31.
for economic emission dispatch problem. Appl Soft Comput 2016;40:479e94. [32] Mirjalili S, Mirjalili SM, Lewis A. Grey wolf optimizer. Adv Eng Softw
[14] Zhou J, Wang C, Li Y, Wang P, Li C, Lu P, et al. A multi-objective multi- 2014;69(3):46e61.
population ant colony optimization for economic emission dispatch consid- [33] Mirjalili S, Saremi S, Mirjalili SM, et al. Multi-objective grey wolf optimizer: a
ering power system security. Appl Math Model 2017;45:684e704. novel algorithm for multi-criterion optimization. Expert Syst Appl 2016;47:
[15] Lai X, Li C, Zhang N, Zhou J. A multi-objective artificial sheep algorithm. 2018. 106e19.
https://doi.org/10.1007/s00521-018-3348-x. [34] Basu M. An interactive fuzzy satisfying method based on evolutionary pro-
[16] Wang Z, Li C, Lai X, et al. An integrated start-up method for pumped storage gramming technique for multiobjective short-term hydrothermal scheduling.
units based on a novel artificial sheep algorithm. Energies 2018;11(1):151. Electr Power Syst Res 2004;69:277e85.
[17] Abido MA. A niched Pareto genetic algorithm for multiobjective environ- [35] Logenthiran T, Srinivasan D. Particle Swarm Optimization for unit commit-
mental/economic dispatch. Int J Elec Power 2003;25(2):97e105. ment problem. In: IEEE international conference on probabilistic methods
[18] Rao BS, Vaisakh K. Multi-objective adaptive Clonal selection algorithm for applied to power systems; 2010. p. 642e7.
solving environmental/economic dispatch and OPF problems with load un- [36] Shukla A, Singh SN. Clustering based unit commitment with wind power
certainty. Int J Elec Power 2013;53(1):390e408. uncertainty. Energy Convers Manag 2016;111:89e102.
[19] Abido MA. Multiobjective evolutionary algorithms for electric power dispatch [37] Arul R, Velusami S, Ravi G. A new algorithm for combined dynamic economic
problem. IEEE Trans Evol Comput 2006;10:315e29. emission dispatch with security constraints. Energy 2015;79:496e511.
[20] Abido MA. Environmental/economic power dispatch using multiobjective [38] Elattar EE. Modified harmony search algorithm for combined economic
evolutionary algorithms. IEEE Trans Power Syst 2003;18:1529e37. emission dispatch of microgrid incorporating renewable sources. Energy
[21] Daryani N, Hagh MT, Teimourzadeh S. Adaptive group search optimization 2018;159:496e507.
algorithm for multi-objective optimal power flow problem. Appl Soft Comput [39] Amiri M, Khanmohammadi S, Badamchizadeh MA. Floating search space: a
2016;38:1012e24. new idea for efficient solving the Economic and emission dispatch problem.
[22] Lu Y, Zhou J, Qin H, Wang Y, Zhang Y. A hybrid multi-objective cultural al- Energy 2018;158:564e79.
gorithm for short-term environmental/economic hydrothermal scheduling. [40] Elsakaan AA, El-Sehiemy RA, Kaddah SS, Elsaid MI. An enhanced moth-flame
Energy Convers Manag 2011;52:2121e34. optimizer for solving non-smooth economic dispatch problems with emis-
[23] Lokeshgupta B, Sivasubramani S. Multi-objective dynamic economic and sions. Energy 2018;157:1063e78.
emission dispatch with demand side management. Int J Elec Power 2018;97: [41] Schmidt J, Cancella R, Pereira Jr O. The role of wind power and solar PV in
334e43. reducing risks in the Brazilian hydro-thermal power system. Energy
[24] Yuan X, Tian H, Yuan Y, Huang Y, Ikram RM. An extended NSGA-III for solution 2016;115:1748e57.
multi-objective hydro-thermal-wind scheduling considering wind power [42] Zhang Y, Le J, Liao X, et al. Multi-objective hydro-thermal-wind coordination
cost. Energy Convers Manag 2015;96:568e78. scheduling integrated with large-scale electric vehicles using IMOPSO. Renew
[25] Chen F, Zhou J, Wang C, Li C, Lu P. A modified gravitational search algorithm Energy 2018;128:91e107.

You might also like