electronics

Article
Optimizing Economic Dispatch for Microgrid Clusters Using
Improved Grey Wolf Optimization
Xinchen Wang 1 , Shaorong Wang 1, * , Jiaxuan Ren 1 , Zhaoxia Song 2 , Shun Zhang 2 and Hupeng Feng 3

1 School of Electrical and Electronic Engineering, Huazhong University of Science and Technology,
Wuhan 430074, China; [email protected] (X.W.); [email protected] (J.R.)
2 Central Southern China Electric Power Design Institute Co., Ltd. of China Power Engineering Consulting
Group, Wuhan 430060, China; [email protected] (Z.S.); [email protected] (S.Z.)
3 Qiongzhong Power Supply Bureau of Hainan Power Grid Co., Ltd., Qiongzhong 572931, China;
[email protected]
* Correspondence: [email protected]

Abstract: With the rapid development of renewable energy generation in recent years, microgrid
technology has increasingly emerged as an effective means to facilitate the integration of renew-
able energy. To efficiently achieve optimal scheduling for microgrid cluster (MGC) systems while
guaranteeing the safe and stable operation of a power grid, this study, drawing on actual electricity-
consumption patterns and renewable energy generation in low-latitude coastal areas, proposes an
integrated multi-objective coordinated optimization strategy. The objective function includes not only
operational costs, environmental costs, and energy storage losses but also introduces penalty terms to
comprehensively reflect the operation of the MGC system. To further enhance the efficiency of solving
the economic dispatch model, this study combines chaotic mapping and dynamic opposition-based
learning with the traditional Grey Wolf Optimization (GWO) algorithm, using the improved GWO
(CDGWO) algorithm for optimization. Comparative experiments comprehensively validate the
significant advantages of the proposed optimization algorithm in terms of economic benefits and
scheduling efficiency. The results indicate that the proposed scheduling strategy, objective model, and
solution algorithm can efficiently and effectively achieve multi-objective coordinated optimization
scheduling for MGC systems, significantly enhancing the overall economic benefits of the MGC while
Citation: Wang, X.; Wang, S.; Ren, J.;
ensuring a reliable power supply.
Song, Z.; Zhang, S.; Feng, H.
Optimizing Economic Dispatch for Keywords: microgrid cluster (MGC); grey wolf optimization (GWO); optimal scheduling; multi-objective
Microgrid Clusters Using Improved optimization; economic dispatch
Grey Wolf Optimization. Electronics
2024, 13, 3139. https://doi.org/
10.3390/electronics13163139
1. Introduction
Academic Editor: Jen-Hao Teng
To address the current fossil energy crisis and increasingly severe climate issues, there
Received: 10 June 2024 is a global consensus on vigorously developing renewable energy sources. Over the past
Revised: 28 July 2024
few years, the application of microgrid technology has become increasingly widespread,
Accepted: 2 August 2024
aiming to promote the integration of renewable energy, enhance the flexibility and stability
Published: 8 August 2024
of power system structures, and ensure power supply reliability [1–5].
Distributed renewable energy generation often exhibits characteristics of decentralized
concentration, a proximity to users, and small-scale deployment [6–8]. This is especially true
Copyright: © 2024 by the authors.
for urban coastal areas with lower latitudes, where wind and solar resources are abundant,
Licensee MDPI, Basel, Switzerland. allowing for the dense placement of small-scale microgrids. Coordinated scheduling among
This article is an open access article multiple adjacent microgrids, often referred to as a microgrid cluster (MGC), can facilitate
distributed under the terms and the local integration of renewable energy sources. This coordination within a microgrid
conditions of the Creative Commons cluster is crucial for environmental protection and alleviating pressure on remote grid
Attribution (CC BY) license (https:// supplies [9–12].
creativecommons.org/licenses/by/ Given the rapid development of renewable energy and ongoing advancements in smart
4.0/). grid technologies, the coordinated operation and scheduling of MGCs have emerged as pivotal

Electronics 2024, 13, 3139. https://doi.org/10.3390/electronics13163139 https://www.mdpi.com/journal/electronics

Electronics 2024, 13, 3139 2 of 21

research avenues [1,12]. This issue spans multiple layers and dimensions, thereby presenting
considerable complexity. While various methods have been proposed and applied to address
practical challenges in this domain, significant hurdles remain to be overcome.
Identifying the key constraints in the power scheduling problem of the MGC system is
essential for maintaining the reliability and efficiency of the entire power system. Common
constraints include power balance constraints and equipment constraints. Power balance
constraints in the MGC system ensure that the total power generation matches the total
power demand at all times. By adhering to power balance constraints, the system can avoid
the extremes of power supply shortages and an oversupply of power [13–15]. Equipment
constraints ensure the safe and stable operation of the system while extending the lifespan
of various components. Common equipment constraints include the output power and
ramping constraints for conventional thermal power generation [13,16], as well as the
charging/discharging power, capacity, and State-of-Charge (SOC) constraints for energy
storage systems (ESSs) [16–18]. Understanding and managing these constraints is crucial
for optimizing MGC performance.
Constructing a reasonable objective function is central to the economic dispatch prob-
lem. The recent literature on MGC systems often constructs objective functions focusing on
operational and environmental costs [19–21]. The calculation of operational costs in MGC
systems resembles that in conventional microgrids and typically includes net costs after
transactions between each MG and the main grid, net costs after transactions between MGs,
maintenance costs for WT and PV generation, and generation costs for non-renewable en-
ergy sources. Environmental costs focus on treatment expenses for pollutants such as CO2 ,
SO2 , and NOX from thermal power generation. This paper attempts to enhance the com-
monly used objective functions by introducing more factors, providing a comprehensive
representation of the MGC system’s economic costs and power quality.
Finding the optimal solution method for the economic dispatch model of MGCs is a
key focus of many research papers in the related field, and numerous optimization algo-
rithms have been applied to this area. Although classical Linear Programming (LP) [22] and
Nonlinear Programming (NLP) [23] are simple and convenient, they are no longer suitable
for the increasingly complex structure of modern power systems [24–26]. In recent years,
Machine Learning has become increasingly prevalent in addressing scheduling challenges
within microgrids (MGs) and MGCs [27]. Deep learning algorithms like Recurrent Neural
Networks (RNNs), Long Short-Term Memory (LSTM) networks, and Deep Reinforcement
Learning (DRL) autonomously acquire operational strategies for MGs to adapt to environ-
mental shifts and market dynamics [27–29]. However, they demand substantial volumes of
training data, extensive training periods, and computational resources [30].
Alongside classical and Machine Learning algorithms, metaheuristic algorithms have
found expanded utility in MGC systems. For instance, the Genetic Algorithm (GA) proves
adept at tackling multi-objective optimization problems, commonly utilized for navigating
the trade-offs between minimizing costs and emissions in distribution network operations.
Although offering high flexibility, this algorithm demands precise parameter initializa-
tion and risks getting trapped in local optima. Particle Swarm Optimization (PSO), as
a prevalent classical intelligent algorithm, stands out for its straightforward implemen-
tation and rapid convergence. However, its pursuit of the optimal solution tends to be
singular, resulting in unstable performance in high-dimensional and complex problem
spaces [31–34]. The Firefly Algorithm (FA) is known for its robust global search capability,
making it suitable for various optimization problems. However, it can struggle with high-
dimensional problems and has longer iteration times. Similarly, the Whale Optimization
Algorithm (WOA) excels in global search and fast convergence with fewer parameters to
adjust, making it user-friendly. Nevertheless, it still has the risk of getting trapped in local
optima, and its performance can be highly influenced by the initial population quality and
specific parameters.
The Grey Wolf Optimization (GWO) algorithm is renowned for its simplicity, ease of
implementation, and strong global search capabilities. By effectively balancing exploration
Electronics 2024, 13, 3139 3 of 21

and exploitation phases, GWO is particularly suitable for complex optimization problems.
These qualities make GWO an excellent candidate for optimizing the economic dispatch
in MGCs. Consequently, this study enhances the GWO algorithm and applies it to solve
optimization problems in this context.
This study constructs the structure of an MGC system and comprehensively investi-
gates the constraints on the equipment, energy storage, and power balance within each
MG. To achieve an optimal economic dispatch while maintaining power quality and ex-
tending the lifespan of the MGC system’s equipment, penalty terms are introduced into the
objective function alongside various costs. This approach aims to balance the economic
efficiency and stability of the power system. Finally, the traditional GWO algorithm is
enhanced to improve its convergence speed and global search capabilities, making it more
suitable for optimizing the economic dispatch of MGCs.
The major contributions of this study can be summarized as follows:
1. This research develops the innovative integration of chaos mapping and a dynamic
opposition-based learning strategy with GWO (CDGWO), significantly improving
the performance of the optimization algorithm. Comparative experiments confirm
the notable advantages of the improved algorithm in solving the economic dispatch
problem of MGCs.
2. This study includes penalty terms in the objective function to account for power
exchange limits between the main grid and the MGC and for discrepancies in energy
levels between the start and end of the ESS operating cycle. This integration enhances
the reliability of the power supply in the MGC system and the longevity of the
microgrid components.

2. Structure and Economic Dispatch Model of the Microgrid Cluster System

2.1. Microgrid Cluster System Structure
The MGC analyzed in this study consists of three independent microgrids (MGs), each
equipped with key components such as wind turbines (WTs), Photovoltaic (PV) arrays,
microturbines (MTs) or Diesel Generators (DGs), battery energy storage systems (ESSs),
and Alternating Current (AC) loads for end-users [35,36].
The Energy Management Center (EMC) plays a crucial role in orchestrating the energy
exchange among these MGs, as well as handling the transactions for buying and selling elec-
tricity with the main grid. This centralized coordination ensures optimal energy distribution
and enhances the overall efficiency and reliability of the microgrid network [3,37–40].
Figure 1 illustrates the structure of the MGC system. The specific configuration of its
sub-microgrid MG2 is detailed in this figure. The structures of MG1 and MG3 are essentially
identical to MG2, except that the non-renewable energy generation equipment in these two
sub-microgrids is not microturbines (MTs) but diesel generators (DGs). This variation in
non-renewable energy generation equipment allows for a comparison of performance and
operational strategies across different energy source types within the MGC.

2.2. Constraints of Microgrid Cluster System

2.2.1. Power Balance Constraints
Mathematically, the power balance constraint can be expressed as (1).
N
∑ Pgen,i (t) + Pimport (t) = Pload (t) (1)
i =1

Pgen,i (t) represents the power output of the i-th generation unit at time t. Pimport (t)
denotes the power imported from the main grid at time t.
The MGi is as shown in (2).
Ploadi (t) = PWTi (t) + PPVi (t) + PMTi (t) + PESSi (t) + Pbuyi (t) − Pselli (t) + Pexi− j (t) + Pexi− j (t) (2)
The meanings of the symbols used in (2) are detailed in Table 1.
 13, 3139
Electronics 2024, 13, x FOR PEER REVIEW 4 4of
of 22
21

Figure 1. The structure of the microgrid cluster (MGC) system.

Table 1. Definition
2.2. Constraints of of symbols Cluster
Microgrid in the power balance constraint equation for each MG.
System
2.2.1. Power Balance Constraints
Symbol Definition
Mathematically,
Ploadi (t) the power balance constraint can be of
Total load demand expressed
the MG-i as (1). t.
at time
PWTi (t) N Wind turbine power output of the MG-i at time t.
PPVi (t) 
Pgen (t ) + Pimport power
Photovoltaic
i =Power
,i
(t ) = Pload (t ) of the MG-i at time t.
output (1)
PMTi (t) 1 output of the microturbine in the MG-i at time t.
Pgen,i (tP) DGi (t)
represents the power Poweroutputoutput of of thei−th
the diesel generator unit
generation in theatMG-i
timeat t.time t.
Pimport (t)
PESSi (t) Power output of the ESS in the MG-i at time t.
denotes thePbuyipower(t) imported fromPowerthe main grid atfrom
purchased timethet. main grid by MG-i at time t.
The MGi Pselliis(tas
) shown in (2). Power sold to the main grid by MG-i at time t.
Pexi− j (t) The power exchange between MG-i and MG-j at time t.
Ploadi (t) = PWTi (t) + PPVi (t) + PMTi (t) + PESSi (t) + Pbuyi (t) − Pselli (t) + Pexi − j (t) + Pexi − j (t) (2)
2.2.2.The meanings
Equipment of the symbols used in (2) are detailed in Table 1.
Self-Constrains
• Constraints on MT and DG
Table 1. Definition of symbols in the power balance constraint equation for each MG.
The Constraints on the output power for MTs and DGs are illustrated in (3).
Symbol  min Definition
(t) ≤load
PMTTotal max ( t )
(t) ≤ PMT
PMT demand
Ploadi (t) of the MG−i at time t. (3)
min max
PDG (t) ≤ PDG (t) ≤ PDG (t)
PWTi (t) Wind turbine power output of the MG−i at time t.
ThePramping
PVi
(t) constraints for MTs and DGs
Photovoltaic powerare illustrated in (4).
output of the MG−i at time t.
PMTi (t) PowerPMToutput
(t) − Pof the microturbine in the MG−i at time t.

MT ( t − 1) ≤ r MT
PDGi (t) (4)
Power PDG (t) − of
output PDGthe − 1) ≤generator
(t diesel r DG in the MG−i at time t.
PESSi (t) Power output of the ESS in the MG−i at time t.
• Constraints on ESS
Pbuyi (t ) Power purchased from the main grid by MG−i at time t.
The Charging/Discharging Power Constraints of the ESS are as illustrated in (5).
Pselli (t) Power sold to the main grid by MG−i at time t.
max

Pexi − j (t ) 0≤P
The power ch ( t ) ≤ Pbetween
exchange ch ( t ) MG−i and MG−j at time t. (5)
max ( t )
0 ≤ Pdis (t) ≤ Pdis
Electronics 2024, 13, 3139 5 of 21

max ( t )
where Pch (t) and Pdis (t) represent the charging and discharging power at time t, Pch
max
and Pdis (t) is the maximum charging and discharging power limit.
The Capacity Constraint of the ESS are as illustrated in (6).

Emin ≤ E(t) ≤ Emax (6)

where Emin and Emax set the upper and lower limits of the ESS capacity.
Effective SOC management strategies are employed to maintain the ESS within optimal
operating ranges and to extend the battery life, as illustrated in (7).

SOCmin ≤ SOC (t) ≤ SOCmax (7)

The minimum state of charge (SOCmin ) is often set at around 20–30% to ensure there
is always some reserve capacity in the battery, which helps in emergencies and prevents
deep discharging, while the maximum state of charge (SOCmax ) is typically set at 90–95%
to avoid overcharging, which can lead to overheating and battery degradation. The SOC of
the ESS in this study is required to be maintained between 30% and 90%.

2.3. Construction of the Objective Function

Most studies on microgrid economic planning focus on constructing objective func-
tions based on operational and environmental costs. To improve the accuracy and com-
prehensiveness of the analysis, this paper adds ESS Loss Costs to these considerations.
Besides these three costs, this study also incorporates two penalty terms into the objective
function to enhance the power quality of the MGC and extend the lifespan of the ESS. These
penalties are essential for maintaining system stability and efficiency.
Based on the above discussion, the objective function is constructed as shown in (8).
FMGC = COperation + CPollution + CESS + FMain− MGC + FESS (8)

where COperation , CPollution , and CESS denote the functions corresponding to Operational
Costs, Pollution Control Costs, and ESS Loss Costs. FMain− MGC and FESS are penalty
functions, representing the penalty for power-exchange-exceeding limits between the main
grid and the MGC and the penalty for discrepancies in energy levels between the start and
end of the ESS operating cycle.

2.3.1. Operational Costs

The operational costs of the MGC system mainly include the net expenditure for
purchasing and selling electricity between each sub-microgrid and the main grid, the net
expenditure for purchasing and selling electricity between sub-microgrids, the maintenance
costs for WT and PV generation, and the generation costs for non-renewable energy sources.
The specific mathematical expression is shown in (9) and (10).
FOperation = Cgrid1 + Cgrid2 + Cgrid3 + Cex1−2 + Cex2−3 + Cex3−1
3 3 3 (9)
+ ∑ CWTi + ∑ CPVi + ∑ Cnon−renew−i
i =1 i =1 i =1

Cgridi = Cbuyi − Cselli i = 1, 2, 3 (10)

2.3.2. Pollution Control Costs

The MGC system explored in this study has not achieved complete renewable energy
integration yet. Wind turbines and photovoltaics, crucial components of the energy gen-
eration process, are notable for their absence of emissions harmful to the environment.
Additionally, the environmental impact of ESSs remains minimal. This study consciously
overlooks any potential pollutants that the ESS may generate, focusing instead on micro
gas turbines and diesel generators, as depicted in (11).
Electronics 2024, 13, 3139 6 of 21

Z 24 M
FPollution =
0
∑ (cDG · λi · PDG (t) + c MT · λi · PMT (t))dt (11)
i =1

where c DG and c MT represent the unit costs of pollution control for MTs and DGs, respec-
tively, while λi denotes the emission coefficient of pollutants.

2.3.3. Energy Storage System Loss Costs

Accounting for ESS losses in cost calculations is essential for several reasons. The
depth and frequency of charge and discharge cycles directly affect the lifespan of the ESS.
In applications such as peak shaving and load leveling, frequent cycling can result in
substantial degradation. Therefore, considering the losses of supercapacitors is crucial for
the economic operation of MGCs. This consideration ensures a comprehensive evaluation
of the system’s long-term performance and cost-effectiveness, leading to more informed de-
cision making and optimized resource utilization [41–43]. The corresponding mathematical
model for ESS loss costs can be simplified to (12).

3 R 24
ESS = m ESS ∑ 0 | PSCi ( t )| · f ( SOCSCi ( t )) dt
 F


i =1 (12)
C
 m ESS = Investment


Q ESS

where PSC1 (t), PSC2 (t), and PSC3 (t) represent the power of the first and second supercapac-
itors, respectively. SOCSC1 (t), SOCSC2 (t), and SOCSC3 (t) denote the state of charge of the
first and second supercapacitors, respectively. f (SOCSCi (t)) represents the loss function
associated with the state of charge of the supercapacitor.
m ESS is the unit loss cost coefficient of the energy storage system. Mathematically, it is
defined as the ratio of the investment cost of the energy storage system to the total charge
and discharge energy over the entire lifecycle of the supercapacitor.

2.3.4. Penalty Function for Power-Exchange-Exceeding Limits

Introducing the penalty function for a power exchange between the main grid and the
MGC exceeding predefined limits serves a critical role in optimizing the objective function.
It ensures an equilibrium of the power exchange between these entities, which is vital
for sustaining the stability and efficiency of the overall system operation. By penalizing
deviations from the desired balance, the optimization process is steered toward solutions
that prioritize a smooth power transmission between the main grid and the microgrid
cluster. This approach enhances the system reliability and performance by promoting a
balanced and stable power exchange mechanism. The corresponding penalty function
expressions for the MG and MGC are given in (13) and (14), respectively.
24
FMain− MGi = δ ∑ ( PMG,i (t) − PMI,i (t)) i = 1, 2, 3 (13)
t =1
3
FMain− MGC = ∑ FMain− MGi (14)
i =1

where δ represents the penalty coefficient, while PMG,i (t) and PMI,i (t) denote the equiva-
lent generation power and equivalent load power transmitted between the MGi and the
distribution network at time t.

2.3.5. Penalty Function for Discrepancies in Energy Levels between Start and End of ESS
Operating Cycle
The penalty term, FESS , serves as a crucial mechanism to mitigate irregularities in the
charging and discharging patterns of the ESS. Penalizing deviations in ESS energy levels
encourages the optimization algorithm to prioritize balanced energy usage throughout
the operational cycle, enhancing the storage and distribution efficiency. This optimization
Electronics 2024, 13, 3139 7 of 21

boosts the microgrid cluster system’s overall performance while also extending the ESS’s
operational lifespan, ensuring sustained reliability within the microgrid ecosystem.

24 24
FESS = γ ∑ Pdis (t)ηdis + ∑ Pch (t)/ηch (15)
t,Pdis (t)>0 t,Pch (t)>0

In (15), γ represents the constraint penalty factor for the battery, while ηdis and ηch ,
respectively, denote the discharging and charging efficiencies of the ESS.

3. Model Solving Method

This study improves the traditional GWO algorithm to enhance its efficiency and
accuracy and applies it to solve the economic dispatch model for the MGC system described.

3.1. Traditional GWO

For the economic dispatch problem of microgrid systems, a comprehensive analysis
of existing metaheuristic algorithms shows that the GWO algorithm performs well in
terms of convergence speed and computational cost, demonstrating strong global search
capabilities [44–46]. Building upon these advantages, this paper chooses to refine and
enhance the GWO algorithm to facilitate the achievement of economic dispatch objectives
in the MGC system.
Proposed by Mirjalili et al., the GWO algorithm emulates the leadership hierarchy and
hunting mechanism of grey wolves in nature [44].
The GWO algorithm is founded on a fundamental mathematical framework that
mirrors the social hierarchy observed in grey wolf populations. By emulating the hunting
behaviors and population dynamics of grey wolves, it aims to achieve optimization 8objec-
Electronics 2024, 13, x FOR PEER REVIEW of 22
tives. The hierarchical structure is illustrated in Figure 2, with the roles and responsibilities
of wolves at each level outlined in Table 2.

Figure 2.
Figure 2. Hierarchy
Hierarchy structure
structure of
of wolf
wolf pack
pack in
in GWO
GWO algorithm.
algorithm.

Table 2.
Table 2. The
The hierarchy
hierarchy and
and corresponding
correspondingroles
rolesof
ofwolf
wolfpack
packin
inGWO
GWOalgorithm.
algorithm.

Level
Level Rank
Rank Roles
Rolesand
and Responsibilities
Responsibilities
The first level
The first level Alpha (α)wolf
Alpha (α) wolf Leader—dominates
Leader—dominates thethepack.
pack.
The secondlevel
The second level Beta
Beta(β)
(β)wolf
wolf Assists
Assistsalpha
alphaand
and helps managethe
helps manage the pack.
pack.
The third level Delta (δ) wolf Follows alpha
Follows alphaandandbeta
betaand ensures
and tasks
ensures are carried
tasks out.
are carried
Thefourth
The thirdlevel
level Omega
Delta (ω)
(δ) wolf
wolf Lowest rank—follows alpha, beta, and delta.
out.
Omega (ω)
The fourth level Lowest rank—follows alpha, beta, and delta.
In the hunting sequence
wolfof grey wolves, three distinct phases are discernible: searching
for prey, encircling prey, and attacking prey.
In the hunting sequence of grey wolves, three distinct phases are discernible: search-
ing for prey, encircling prey, and attacking prey.
Within the framework of the GWO algorithm, the act of grey wolves surrounding
prey can be conceptualized through the following position update (16).
 D = C  X ( t ) − X( t )
 prey
 (16)
X(t + 1) = Xprey (t ) − AD
Electronics 2024, 13, 3139 8 of 21

Within the framework of the GWO algorithm, the act of grey wolves surrounding prey
can be conceptualized through the following position update (16).

→ → → →
 D = C ∗ X prey (t) − X (t)

(16)
→→
 →
 →
X (t + 1) = X prey (t) − A D
→ →
where D represents the distance between individual grey wolves and the prey, X denotes
→
the position of each individual in the population, X prey stands for the position of the prey,
→ →
and t signifies the iteration count. Additionally, A and C represent the vector coefficients
determined by (17).
 →

→ → →
A = 2a · r1 − a
→ (17)
 C = 2·→ r 2
→ → →
where r 1 and r 2 are random vector values within the range of [0, 1], and a represents the
convergence factor, which governs the equilibrium between the “surrounding prey” and
“attacking prey” actions in the GWO algorithm. Typically, for this parameter, it is commonly
handled by linearly decreasing its value from 2 to 0 as the iteration count increases.
Following the natural hierarchy, the pursuit of prey in GWO is guided by the α wolf,
with other rank wolves cooperating to encircle, pursue, and attack the prey. Accordingly,
during the algorithm’s iteration, the α wolf is considered the optimal individual, followed
by β and then δ individuals. Following the outlined logic and rules, the position update
(18) and (19) for the grey wolf individual ω, distinct from α, β, and δ, is established.
→ → →

→ → →
 X 1 (t) = X α (t) − A1 · C 1 · X α (t) − X (t)





→ → →

 → → →
X 2 (t) = X β (t) − A2 · C 2 · X β (t) − X (t) (18)


→ → →


 → → →
 X 3 (t) = X δ (t) − A3 · C 3 · X δ (t) − X (t)



→ 1 3 →
3 i∑
X ω ( t + 1) = · X i (t) (19)
=1
→ → →
where X α (t), X β (t), and X δ (t) represent the positions of the α, β, and δ wolves, respectively,
→ → → → → → → →
at time t. Additionally, A1 , A2 , and A3 are similar to A, C 1 , and C 2 , and C 3 is similar to C,
which is described as (17).

3.2. The Imroved GWO Algorithm (CDGWO)

Traditional GWO is a nature-inspired algorithm celebrated for its simplicity and
effectiveness in solving optimization problems. A key advantage of GWO is its ability to
balance exploration and exploitation, making it robust across various types of optimization
tasks. However, traditional GWO has its limitations, including a propensity to get trapped
in local optima and slow convergence speeds, especially in complex, high-dimensional
problem spaces.
To address these shortcomings, this study introduces chaos optimization and a dy-
namic opposition-based learning strategy. Chaos optimization helps enhance the global
search ability and prevent premature convergence by introducing randomness and nonlin-
earity into the search process. The dynamic opposition-based learning strategy improves
the convergence speed and solution accuracy by considering opposite solutions and dy-
namically updating them throughout the optimization process. These enhancements aim
Electronics 2024, 13, 3139 9 of 21

to improve the overall performance of GWO, thereby enabling more efficient solutions for
the economic dispatch challenges in the MGC system.

3.2.1. Chaos Optimization

Chaos mappings can generate sequences characterized by high randomness and
nonlinearity, suitable for replacing conventional uniformly distributed random numbers
in intelligent optimization algorithms [47–49]. These mappings enhance the global search
efficiency and robustness of algorithms [47], reduce dependence on the selection of initial
solutions and parameter settings, and offer more flexible optimization strategies for MGC
scheduling problems. This paper discusses several common chaos mappings and selects
the most appropriate one to enhance the GWO algorithm, tailored to the context described.
Although Tent mapping exhibits piecewise linear properties, it may lack sufficient
flexibility in large-scale optimization scenarios. Sine mapping, when applied to high-
dimensional optimization problems, is prone to converging to local optima, with its perfor-
mance being highly dependent on the choice of parameters. In contrast, Logistic mapping
is straightforward to implement and possesses robust chaotic characteristics. Chebyshev
mapping maintains better uniformity and distribution within a certain range. Both map-
pings can significantly improve the search capabilities of the GWO algorithm, expand the
diversity of the search space, and elevate the efficiency of optimization. The expressions for
these mappings are presented in Table 3.

Table 3. Expressions of several commonly used chaotic maps.

Example Expression
(
xi /a xi < a
Tent x i +1 =
(1 − x i ) / (1 − a ) x i ≥ a
a ∈ (0, 1)
Sine xi+1 = 4a sin(πxi )
a=4
Chebyshev xi+1 = cos( a · cos−1 ( xi ))
a=4
Logistic xi+1 = axi (1 − xi )
a ∈ (0, 4]

By qualitatively analyzing the characteristics of several common mappings, it can be

inferred that Logistic mapping and Chebyshev mapping are better suited for the problem
context outlined in this paper. To precisely select the chaotic mapping that enhances the
performance of GWO, this study will empirically compare and identify the most effective
chaotic mapping in Section 4.

3.2.2. Dynamic Opposition-Based Learning Strategy

The GWO algorithm employs a hierarchical strategy, with α, β, and δ wolves leading
the search process. This hierarchy helps direct subordinate agents to explore the most
promising areas for optimal solutions, effectively reducing the risk of premature conver-
gence to local optima. However, the strong exploitation capability inherent in this method
can limit the search diversity and can hinder the exploratory potential [48].
To overcome this limitation, this paper introduces the dynamic opposition-based
learning (DOBL) strategy to enhance the GWO algorithm. Traditional opposition-based
learning (OBL) [50,51] explores both current and opposite positions to improve the solution
quality, but DOBL advances this concept by incorporating a dynamic factor r that changes
nonlinearly with each iteration. This innovation allows for the generation of reverse
Electronics 2024, 13, 3139 10 of 21

Electronics 2024, 13, x FOR PEER REVIEW

solutions 11 of 22
to be more responsive to the search space’s evolving landscape, thereby boosting
the algorithm’s capacity to explore diverse regions and escape local optima.

r = sin(t/T )
Step 3: Establish the fitness function, which is identical to the objective function con- (20)
e i (t) = popmax + popmin − rXi (t),
X = 1, 2, · · ·as, nshown in (8). Eval-
structed for the microgrid cluster optimization schedulingi strategy,
uate the fitness value of the entire population.
Equation (20) illustrates the modified position update formula, which accounts for the
dynamic 4:
Step Identify
factor the three
r. Here, popmaxgrey
andwolves
popmin with
define the
thelowest
upperfitness values
and lower as the α wolf,
boundaries β
of the
wolf,
search and δ wolf.
space, These wolves
respectively. Xi (twill guide the
) indicates therest of theof
position population.
the i-th individual grey wolf at
Step 5: perform dynamic opposition−based learning
the t-th iteration. Xi (t) represents the corresponding dynamic
e by executing searchest denotes
reverse solution. at both
the current positions and their direct opposites to increase
the iteration time, and T represents the total number of iterations. the likelihood of finding supe-
rior solutions.
3.2.3.Step
The6:Specific
UpdateStepsindividual positions based
of the Improved GWOon the guidance of the α wolf, β wolf, and
(CDGWO)
δ wolf. Then, update the global optimum.
This paper presents an enhanced GWO algorithm for the economic dispatch of the
MGC Step 7: Assess
system, whether the
incorporating termination
chaotic mappingcriteria
and thehave been met.
dynamic If fulfilled, output
opposition-based the
learning
optimal fitness value. If not, recalculate the fitness value based on the updated
strategy. The detailed algorithmic procedure is illustrated in Figure 3. The specific steps of individual
positions
executionand are iterate
outlined again.
below.

Figure
Figure3.3.The
Theprocedure
procedurefor
forthe
theimproved
improvedGWO
GWOalgorithm.
algorithm.

3.3. The Solution

Step Process
1: initialize forparameters
the the Constructed MGC
of the greyEconomic Dispatch Model
wolf population.
Step 2: Use chaotic maps to generate sequences as initial
This paper begins by constructing the topology of the microgrid positions for the
cluster andwolf popu-
defining
lation.
the Theconstraints
system chaotic mapnecessary
formulasfor
aresafe
presented in Table
operation. 3, whichdispatch
An optimal includesobjective
expressions of
func-
several commonly used chaotic maps.
tion is then developed to balance economic efficiency and power quality, resulting in a
comprehensive multi−objective economic dispatch model. Subsequently, this study en-
hances the traditional GWO algorithm, resulting in the CDGWO algorithm, which fea-
tures improved global search capabilities and higher optimization efficiency and
 Electronics 2024, 13, 3139 11 of 21

Step 3: Establish the fitness function, which is identical to the objective function
constructed for the microgrid cluster optimization scheduling strategy, as shown in (8).
Evaluate the fitness value of the entire population.
Step 4: Identify the three grey wolves with the lowest fitness values as the α wolf, β
wolf, and δ wolf. These wolves will guide the rest of the population.
Step 5: perform dynamic opposition-based learning by executing searches at both the cur-
rent positions and their direct opposites to increase the likelihood of finding superior solutions.
Step 6: Update individual positions based on the guidance of the α wolf, β wolf, and δ
wolf. Then, update the global optimum.
Step 7: Assess whether the termination criteria have been met. If fulfilled, output the
optimal fitness value. If not, recalculate the fitness value based on the updated individual
positions and iterate again.

3.3. The Solution Process for the Constructed MGC Economic Dispatch Model
This paper begins by constructing the topology of the microgrid cluster and defining
the system constraints necessary for safe operation. An optimal dispatch objective function
13, x FOR PEER REVIEW is then developed to balance economic efficiency and power quality, resulting in a compre-
12 of 22
hensive multi-objective economic dispatch model. Subsequently, this study enhances the
traditional GWO algorithm, resulting in the CDGWO algorithm, which features improved
global search capabilities and higher optimization efficiency and accuracy. This enhanced
accuracy. This enhanced
algorithmalgorithm is solve
is applied to applied to solve
the model. the
The model.
specific The specificapproach
methodological methodo-is illustrated
logical approach is illustrated in the block diagram
in the block diagram shown in Figure 4. shown in Figure 4.

Figure 4. Construction and4. solution

Figure of the
Construction MGC
and economic
solution dispatch
of the MGC model.
economic dispatch model.

4. Simulations
4. Simulations and Results and Results
4.1. Parameters of the Numerical Example
4.1. Parameters of the Numerical Example
This study addresses the optimization scheduling of the MGC system on a daily
This study addresses
cycle, withthe optimization
each schedulinginterval,
hour as a scheduling of the MGC system
resulting in 24onintervals
a daily per
cy- day. The
cle, with each hour as a scheduling interval, resulting in 24 intervals per day. The meteor-
meteorological data for the region were sourced from the European Centre for Medium-
Range
ological data for the Weather
region wereForecasts
sourced(ECMWFs). Using real Centre
from the European data from foraMedium−Range
low-latitude coastal region,
Weather Forecastsincluding
(ECMWFs). meteorological
Using realparameters
data fromand historical wind
a low−latitude and solar
coastal power
region, output data, the
includ-
output for a typical day is forecasted. The load forecasting
ing meteorological parameters and historical wind and solar power output data, the out-data are based on historical data
from the local grid.
put for a typical day is forecasted. The load forecasting data are based on historical data
To calculate the operational revenues of the MGC system, this paper employs time-
from the local grid.of-use (TOU) pricing. TOU pricing is a dynamic strategy where electricity rates vary
To calculatethroughout
the operational
the day, revenues
with higherof theduring
rates MGCpeak system,
demand this paper
periods andemploys
lower rates during
time−of−use (TOU) pricing. TOU pricing is a dynamic strategy where electricity
off-peak hours. By incorporating TOU pricing, utilities can encourage rates vary to shift
consumers
throughout the day, with higher rates during peak demand periods and lower rates dur-
ing off−peak hours. By incorporating TOU pricing, utilities can encourage consumers to
shift their electricity usage to off−peak times, reducing grid strain during peak periods
and optimizing resource utilization. Figure 5 clearly illustrates the electricity purchase
and sale prices for each time period throughout the day in the region.
 Weather Forecasts (ECMWFs). Using real data from a low−latitude coastal region, includ-
ing meteorological parameters and historical wind and solar power output data, the out-
put for a typical day is forecasted. The load forecasting data are based on historical data
from the local grid.
To calculate the operational revenues of the MGC system, this paper employs
Electronics 2024, 13, 3139
time−of−use (TOU) pricing. TOU pricing is a dynamic strategy where electricity rates vary 12 of 21
throughout the day, with higher rates during peak demand periods and lower rates dur-
ing off−peak hours. By incorporating TOU pricing, utilities can encourage consumers to
theirtheir
shift electricity usage
electricity usagetotooff-peak
off−peaktimes,
times, reducing
reducing gridgrid strain
strainduring
during peak
peak periods and
periods
optimizing
and resource
optimizing utilization.
resource Figure
utilization. Figure5 clearly illustrates
5 clearly illustratesthe
theelectricity
electricitypurchase
purchaseand sale
and salefor
prices prices
eachfortime
eachperiod
time period throughout
throughout the day
the day in the in the region.
region.

Electronics 2024, 13, x FOR PEER REVIEW 13 of 22

Time-of-use
Figure5.5.Time−of−use
Figure (TOU)
(TOU) electricity
electricity pricing
pricing for microgrid
for microgrid purchase
purchase and sale.
and sale.
4.2. A Quantitative Analysis of the Effects of Different Chaotic Mappings
4.2. A Quantitative Analysis of the Effects of Different Chaotic Mappings
InInSection
Section 3.2.1,
3.2.1, a qualitative
a qualitative analysis
analysis was was conducted
conducted on theon the compatibility
compatibility of various
of various
chaotic maps with the research context described in this paper. It
chaotic maps with the research context described in this paper. It was concluded that Che- was concluded that
Chebyshev mapping and Logistic mapping are relatively more
byshev mapping and Logistic mapping are relatively more suitable for optimizing the suitable for optimizing the
GWO algorithm. To determine the most effective type of chaotic
GWO algorithm. To determine the most effective type of chaotic mapping for the eco- mapping for the economic
dispatch
nomic optimization
dispatch of the
optimization of MGC,
the MGC,thisthis
study integrates
study the
integrates thefour
fourcommonly
commonly used
used chaotic
maps with
chaotic mapsGWO.with GWO.The resulting fourfour
The resulting improved
improved chaotic
chaoticGWO
GWO(CGWO) algorithms, along
(CGWO) algorithms,
along
with with the unmodified
the unmodified traditional
traditional GWO,are
GWO, areapplied
applied to the
the economic
economicdispatch
dispatchopti-
optimiza-
mization of the MGC system. Their optimization effects are compared,
tion of the MGC system. Their optimization effects are compared, and the comparative and the compara-
tive convergence
convergence characteristiccurves
characteristic curvesareareillustrated
illustrated in
in Figure
Figure6.6.

Figure
Figure6.6.Comparison
Comparisonofofconvergence
convergencecharacteristic
characteristiccurves
curvesforforfour
fourCGWO
CGWOvariants and
variants traditional
and traditional GWO.
GWO.
It can be observed that different chaotic mappings significantly influence the con-
It can be
vergence observed
speed thatoptimization
of the different chaotic mappings
algorithm insignificantly influence the
its early iterations. conver-the four
Among
gence speed of the optimization algorithm in its early iterations.
commonly used chaotic mappings discussed in this paper, CGWO improved Among the four com-
with Sine
monly used chaotic mappings discussed in this paper, CGWO improved with Sine map-
mapping demonstrated the poorest performance, initially exhibiting the slowest conver-
ping demonstrated the poorest performance, initially exhibiting the slowest convergence
gence speed. Tent mapping also showed suboptimal optimization results, with both the
speed. Tent mapping also showed suboptimal optimization results, with both the conver-
convergence speed and precision falling short. In contrast, the GWO algorithm optimized
gence speed and precision falling short. In contrast, the GWO algorithm optimized with
with Chebyshev mapping produced solutions closest to the actual optimum. However,
Chebyshev mapping produced solutions closest to the actual optimum. However, Logistic
mapping achieved the fastest convergence, with its acceleration effect being particularly
notable within the first 20 iterations. To further compare the optimization effects of these
two chaotic mappings, a detailed analysis is provided in Table 4.

Table 4. Comparison of the effects of various chaotic mappings.

Electronics 2024, 13, 3139 13 of 21

Logistic mapping achieved the fastest convergence, with its acceleration effect being partic-
ularly notable within the first 20 iterations. To further compare the optimization effects of
these two chaotic mappings, a detailed analysis is provided in Table 4.

Table 4. Comparison of the effects of various chaotic mappings.

Type of Chaotic Mapping Optimal Fitness Value Runtime/s

Tent 2.0920 × 103 9.1720009
Sine 2.0597 × 103 10.496083
Chebyshev 1.6895 × 103 11.616561
Logistic 1.7150 × 103 7.9097413
Traditional GWO 1.9030 × 103 10.534876

Upon analyzing Table 4, it is evident that while the GWO algorithm optimized with
Chebyshev
Electronics 2024, 13, x FOR PEER REVIEW mapping achieves results closer to the actual optimum, the difference 14 of 22compared
to the GWO algorithm optimized with Logistic mapping is negligible. Considering that
Chebyshev mapping requires significantly more computational time than Logistic mapping,
and with and
mapping, the aim
withto thestrike
aim toa balance between
strike a balance algorithm
between efficiency
algorithm and and
efficiency precision, the superior
precision,
optimization effect of Logistic mapping is asserted.
the superior optimization effect of Logistic mapping is asserted.
Buildingononthe
Building the comprehensive
comprehensive analysis
analysis thatthat includes
includes both both qualitative
qualitative and quantitative
and quantita-
evaluations,
tive it isitestablished
evaluations, is establishedthatthat
Logistic mapping
Logistic mappingexhibits superiority
exhibits superiorityover
overother
othermappings
in enhancing
mappings the performance
in enhancing of theofGWO
the performance the GWO algorithm.
algorithm.Consequently, thisstudy
Consequently, this study adopts
adopts Logistic
Logistic mappingmapping to optimize
to optimize the the
GWOGWO algorithm.
algorithm.

4.3.
4.3.Analysis
Analysisof of
Simulation Results
Simulation Results
4.3.1.An
4.3.1. AnAnalysis
Analysisthethe Optimization
Optimization Effect
Effect of theofProposed
the Proposed
CDGWOCDGWO
In
InSection
Section4.2, experiments
4.2, experimentswerewere
conducted to compare
conducted the optimization
to compare effects ofeffects of
the optimization
various
variouschaotic
chaoticmappings
mappings on the GWO
on the GWOalgorithm, ultimately
algorithm, selecting
ultimately Logistic Logistic
selecting mappingmapping
as the most effective overall. However, the convergence characteristic
as the most effective overall. However, the convergence characteristic curves curves of CGWOof CGWO
indicate that while chaotic mapping significantly accelerates convergence in the early
indicate that while chaotic mapping significantly accelerates convergence in the early
stages, its impact diminishes in the later stages. Therefore, this study also incorporates the
stages, its impact diminishes in the later stages. Therefore, this study also incorporates the
dynamic opposition−based learning strategy to further enhance the convergence speed
dynamic opposition-based learning strategy to further enhance the convergence speed and
and global search capability of the GWO algorithm.
global
To search
verify capability of the GWO
the optimization algorithm.
performance of the improved CDGWO algorithm,
To verify the optimization performance
CDGWO, traditional GWO, and three other commonly of the improved CDGWO
used intelligent algorithm,
optimization al-CDGWO,
traditional GWO, and three other commonly used intelligent optimization
gorithms—the Firefly Algorithm (FA), Particle Swarm Optimization (PSO), and Whale algorithms—the
Firefly Algorithm
Optimization (FA),
Algorithm Particle Swarm
(WOA)—were appliedOptimization
to the economic (PSO),
dispatchand Whale
of the MGCOptimization
sys-
Algorithm
tem. (WOA)—were
The optimization resultsapplied to compared
were then the economic dispatch through
and analyzed of the MGC
charts system.
and The
graphs. Figureresults
optimization 7 illustrates
werethe
thenconvergence
compared characteristic
and analyzedcurves of five
through intelligent
charts opti- Figure 7
and graphs.
mization
illustratesalgorithms
the convergence characteristic curves of five intelligent optimization algorithms.

Comparison
Figure7.7.Comparison
Figure of improved
of improved GWOGWO (CDGWO)
(CDGWO) with common
with common intelligent
intelligent optimization
optimization algorithms.algorithms.

For a more comprehensive comparison of algorithm performance, Table 5 provides

precise numerical indicators, including the Optimal Fitness Value, Runtime, Number of
Iterations at Convergence, and convergence variance.
Convergence variance reflects the average squared differences among optimization
Electronics 2024, 13, 3139 14 of 21

For a more comprehensive comparison of algorithm performance, Table 5 provides

precise numerical indicators, including the Optimal Fitness Value, Runtime, Number of
Iterations at Convergence, and convergence variance.

Table 5. Comparison of improved GWO (CDGWO) with common intelligent optimization algorithms.

Intelligent Optimization Number of Iterations Convergence

Optimal Fitness Value Runtime/s
Algorithm at Convergence Variance
FA 1.425 × 103 1100.305307 70 132.656897
PSO 2.147 × 103 27.753490 90 213.926534
WOA 3.045 × 103 26.265349 350 587.452367
GWO 1.903 × 103 10.534876 93 196.567398
GA 1.576 × 103 103.635457 56 206.875623
SA 2.803 × 103 52.786543 130 670.246676
CDGWO 1.044 × 103 6.906439 65 48.678354

Convergence variance reflects the average squared differences among optimization

results from multiple runs. A lower convergence variance indicates more consistent results
across runs, suggesting algorithm stability. Conversely, a higher convergence variance
suggests greater variability, possibly due to algorithm randomness or instability. Analyzing
the convergence variance enables an assessment of the optimization algorithm’s stability
and consistency.
1 N
2
2
SCon = ∑ Fi − F (21)
N i =1
where N denotes the number of runs, Fi represents optimal fitness value of the i-th run, and
F denotes the mean of all optimization results.
Figure 7 should be analyzed in conjunction with Table 5. The Whale Optimization
Algorithm (WOA) exhibits the slowest convergence speed, needing about 350 iterations to
reach the global optimum. This slow convergence is due to its simulation of whale foraging
behaviors, which introduces uncertainty in the search direction [52]. Additionally, the
WOA has a high convergence variance of 587.45, indicating instability.
In contrast, the Firefly Algorithm (FA) converges more quickly, reaching the global
optimum in about 70 iterations, with a lower convergence variance below 150, indicating
better stability. However, its computational complexity (O(n2 )), due to interactions between
fireflies, results in significantly longer computation times compared to other algorithms [53,54].
The Genetic Algorithm (GA) achieves a similar solution accuracy to the FA but is less
computationally efficient, with a runtime of 103 s. This is due to the resource-intensive
nature of genetic operations, which require significant computational resources.
Simulated Annealing (SA) has a lower solution accuracy and a runtime of 53 s. Its high
convergence variance of 670.25 reflects significant instability and a tendency to converge to
local optima due to its probabilistic approach.
Both Particle Swarm Optimization (PSO) and traditional Grey Wolf Optimization (GWO)
show good stability with a convergence variance of around 200. However, PSO often gets
trapped in local optima due to rapid early convergence and limited disturbance mechanisms.
Among the mentioned algorithms, GWO performs reasonably well. However, com-
pared to traditional GWO, CDGWO stands out with improved efficiency and enhanced
search capabilities. By the 65th iteration, CDGWO converges to the optimal fitness value,
achieving results closest to the actual global optimum among the five algorithms, with
the highest precision. Additionally, CDGWO’s computation time is 3.6 s shorter than
traditional GWO, marking a significant efficiency boost. Moreover, CDGWO demonstrates
a notably increased stability, with a convergence variance of only 48.678354, surpassing
other algorithms.
In conclusion, the improved GWO algorithm achieves optimization in iteration ef-
ficiency, precision, and stability. Among the commonly used intelligent optimization
Electronics 2024, 13, 3139 15 of 21

algorithms discussed in this paper, the CDGWO algorithm demonstrates the best optimiza-
tion performance. It is worth noting that the inclusion of penalty terms in the objective
function means that the fitness value and the actual daily cost of the MGC system, as
obtained by various optimization algorithms, are not numerically equivalent. Table A1
presents the actual daily costs obtained by each algorithm. Remarkably, CDGWO achieves
the lowest cost, further affirming the superiority of this enhanced algorithm.

4.3.2. Analysis of Economic Dispatch Results Based on CDGWO

In this section, the previously proposed CDGWO algorithm is applied to solve the
MGC scheduling model described in Section 4.1. To further analyze the effectiveness
of the economic dispatch and verify the stability of the proposed optimization method,
comparative experiments were conducted. Initially, forecast data based on actual renewable
energy generation and load profile data from a low-latitude coastal region were used as
inputs under normal conditions. Additionally, a ±10% random disturbance was introduced
to MG1’s wind power output, MG2’s photovoltaic output, and MG3’s load forecast to
simulate uncertainties.
In Figures 8–10, (a) represents the optimal scheduling results for each sub-microgrid
under normal conditions, and (b) represents the power balance scheduling results for each
sub-microgrid after introducing random disturbances. Appendix A includes Tables A2
and A3, which present the power purchase and sale situation of each MG under normal
conditions and after introducing random disturbances, respectively. Additionally, Tables A4
and A5 show the various costs of the MGC system under normal conditions and after
Electronics 2024, 13, x FOR PEER REVIEW 17 of 22
introducing random disturbances. These figures and tables allow for an analysis from
multiple perspectives.

(a) Normal Condition (b) With random disturbance

Figure
Figure 8. Resultsofofpower
8. Results powerbalance
balance scheduling
schedulingofofMG1.
MG1.

(a) Normal Condition (b) With random disturbance

Figure 9. Results of power balance scheduling of MG2.

 (a) Normal Condition (b) With random disturbance
(a) Normal Condition (b) With random disturbance

Electronics 2024, 13, 3139 16 of 21

Figure 8. Results of power balance scheduling of MG1.
Figure 8. Results of power balance scheduling of MG1.

(a)
(a)Normal
NormalCondition
Condition (b) With
(b) Withrandom
randomdisturbance
disturbance

Figure
9. 9.
Figure
Figure 9.Resultsofof
Results
Results power
ofpower balance scheduling
powerbalance scheduling of
schedulingofofMG2.
MG2.
MG2.

(a)Normal
(a) Normalconditions
conditions (b)
(b) With
Withrandom
randomdisturbance
disturbance

Figure 10. Results of power balance scheduling of MG3.

1. ESS. From the standpoint of energy storage, the ESS effectively plays a vital regulatory
role, achieving peak shaving and valley filling through flexible charging and discharg-
ing. Moreover, the ESS aims to maintain energy balance throughout its operational
cycle, which helps extend its lifespan.
2. Economic Efficiency. Based on the time-of-use pricing shown in Figure 5, an economic
analysis reveals that the three MGs prefer to purchase electricity from the main grid
during periods of low prices and sell electricity back to the grid during high-demand
periods (at around 1 PM and 8 PM) when prices are higher. This strategy, combined
with the ESS discharging and energy exchange between MGs, ensures internal power
balance and cost efficiency.
3. Environmental Protection and Power Quality. In terms of environmental protection
and power quality, all three MGs prioritize utilizing their internal renewable energy
sources. When renewable energy is insufficient, traditional thermal power generation
is employed for peak shaving and valley filling, ensuring a balanced approach to
energy management.
4. Robustness. A comparison of the optimization results under normal conditions and
with disturbances revealed that the three microgrids maintain perfect power balance,
ensuring the power quality of the system, as shown in Figures 8–10. According to
Tables A2–A5, the operational costs, pollution control costs, and energy storage sys-
tem loss costs all increase under disturbances, with the overall costs rising by 7.80%.
Electronics 2024, 13, 3139 17 of 21

The variation is not significant. These results indicate that although the optimization
results change slightly after introducing disturbances, the system performance re-
mains stable and ensures excellent economic benefits, demonstrating the reliability
and robustness of the method in practical applications.

5. Conclusions
Based on real wind and solar power outputs and load data from a low-latitude
coastal region, this paper conducts a comprehensive study on the economic dispatch
optimization of microgrid cluster (MGC) systems. This study begins by presenting the
topology and equipment configuration of the MGC. Within this framework, it provides a
detailed discussion of the self-constraints of each microgrid (MG), the constraints of the
energy storage system (ESS), and the power balance constraints within the MGC system.
This ensures that the optimization results for the economic dispatch of MGCs are feasible
in principle, economically rational, and reliable in system operations. Accordingly, a more
optimal multi-objective economic dispatch model is constructed and solved using the
improved CDGWO algorithm. Finally, through comparative experiments, the superiority
of the proposed CDGWO is comprehensively and specifically analyzed in graphical and
tabular form. The robustness of the model is also validated by analyzing the economic
dispatch results before and after introducing random disturbances. The main contributions
of this study are as follows:
1. Incorporation of Penalty Terms. In addition to economic indicators, two penalty terms
are introduced into the objective function of the multi-objective economic dispatch
model: the penalty for power-exchange-exceeding limits between the main grid and
the MGC and the penalty for discrepancies in energy levels between the start and
end of the ESS operating cycle. This integration minimizes costs while incorporating
critical factors such as power quality and equipment lifespan, promoting a more
reliable, efficient, and sustainable operation of the MGC.
2. Proposed Improved GWO (CDGWO). The proposed CDGWO combines chaotic map-
ping and the dynamic opposition-based learning strategy with traditional GWO.
Through experiments, the most suitable type of chaotic mapping for the research
context, Logistic mapping, was identified. The improved GWO algorithm achieves
significant enhancements in optimization performance.
Overall, the integration of CDGWO with the comprehensive multi-objective opti-
mization model presents a robust solution for the economic dispatch of MGCs, offering
both economic benefits and enhanced system reliability. Future research should explore
further refinements to this optimization algorithm and its objective function, as well as its
application to larger and more complex microgrid systems.

Author Contributions: Conceptualization and methodology, X.W. and S.W.; software, X.W. and J.R.;
validation, X.W., S.W., J.R., Z.S. and S.Z.; formal analysis, X.W. and H.F.; investigation and resources,
H.F.; data curation, X.W. and J.R.; writing—original draft preparation, X.W.; writing—review and
editing, X.W., S.W., J.R., Z.S. and S.Z.; visualization, H.F.; supervision, S.W. and H.F.; project adminis-
tration and funding acquisition, S.W. All authors have read and agreed to the published version of
the manuscript.
Funding: This research was supported by Hainan Power Grid Co., Ltd. (research on key tech-
nologies for the power supply security of mountainous power grids based on microgrid clusters,
070000KK52210030).
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: The original contributions presented in the study are included in the
article, further inquiries can be directed to the corresponding author.
Conflicts of Interest: Authors Zhaoxia Song and Shun Zhang were employed by the company Central
Southern China Electric Power Design Institute Co., Ltd. of China Power Engineering Consulting
Electronics 2024, 13, 3139 18 of 21

Group. Author Hupeng Feng was employed by the company Qiongzhong Power Supply Bureau of
Hainan Power Grid Co., Ltd. The remaining authors declare that the research was conducted in the
absence of any commercial or financial relationships that could be construed as a potential conflict of
interest. The authors declare that this study received funding from Hainan Power Grid Co., Ltd. The
funder was not involved in the study design, collection, analysis, interpretation of data, the writing
of this article or the decision to submit it for publication.

Appendix A

Table A1. Actual daily cost of MGC solved by various intelligent optimization algorithms.

Intelligent Optimization Algorithm Daily Cost/CNY

FA 898.42
PSO 1003.21
WOA 1343.77
GWO 937.86
CDGWO 780.46

Table A2. Power purchase and sale situation of each MG under normal conditions.

MG1 Power/kW MG2 Power/kW MG3 Power/kW

Total power Total power Total power
purchased from 414.8024 purchased from 53.5579 purchased from 396.2862
the main grid the main grid the main grid
Total power sold Total power sold Total power sold
88.9414 297.6848 113.1721
to the main grid to the main grid to the main grid
Net power Net power Net power
exchange between 15.4243 exchange between 75.4065 exchange between 108.5003
MG1 and MG2 MG2 and MG3 MG3 and MG1
Net power Net power Net power
exchange between 108.5003 exchange between 15.4243 exchange between 75.4065
MG3 and MG1 MG1 and MG2 MG2 and MG3

Table A3. Power purchase and sale situation of each MG after introducing random disturbances.

MG1 Power/kW MG2 Power/kW MG3 Power/kW

Total power Total power Total power
purchased from 437.9258 purchased from 96.2172 purchased from 442.5331
the main grid the main grid the main grid
Total power sold Total power sold Total power sold
79.5980 274.2019 169.3209
to the main grid to the main grid to the main grid
Net power Net power Net power
exchange between 15.8488 exchange between 105.5076 exchange between 100.7911
MG1 and MG2 MG2 and MG3 MG3 and MG1
Net power Net power Net power
exchange between 100.7911 exchange between 15.8488 exchange between 105.5076
MG3 and MG1 MG1 and MG2 MG2 and MG3

Table A4. Various costs of the MGC system under normal conditions.

Operational Pollution
CNY CNY ESS Lose Cost CNY
Cost Control Cost
MG1 146.3235 MG1 44.1442 MG1 0.0022
MG2 94.7552 MG2 90.4444 MG2 0.0030
MG3 403.1297 MG3 1.6524 MG3 0.0033
Total 644.2084 Total 136.2411 Total 0.0085
Total cost CNY 780.46
Electronics 2024, 13, 3139 19 of 21

Table A5. Various costs of the MGC system after introducing random disturbances.

Pollution
Operational Cost CNY Yuan ESS Lose Cost CNY
Control Cost
MG1 182.8861 MG1 44.6929 MG1 0.0002
MG2 105.6780 MG2 115.3801 MG2 0.0017
MG3 388.0392 MG3 4.6675 MG3 0.0031
Total 676.6033 Total 164.7405 Total 0.0049
Total cost CNY 841.35

References
1. Chen, Z.; Fan, F.; Tai, N.; Hu, Y.; Li, R. Coordinated Energy Dispatch and Flexibility Support for Microgrid Cluster Using
Rule-Based Stackelberg Gaming Approach. IEEE Trans. Ind. Appl. 2024, 60, 1564–1575. [CrossRef]
2. Wu, Z.; Zou, Y.; Zheng, F.; Liang, N. Research on Optimal Scheduling Strategy of Microgrid Considering Electric Vehicle Access.
Symmetry 2023, 15, 1993. [CrossRef]
3. Tan, S.; Xie, P.; Guerrero, J.M.; Vasquez, J.C.; Alcala, J.M.; Carreño, J.E.M.; Zapata, M.G. Lyapunov-Based Resilient Cooperative
Control for DC Microgrid Clusters Against False Data Injection Cyber-Attacks. IEEE Trans. Smart Grid 2024, 15, 3208–3222.
[CrossRef]
4. Chen, W.; Zhang, Y.; Chen, J.; Xu, B. Pricing Mechanism and Trading Strategy Optimization for Microgrid Cluster Based on CVaR
Theory. Electronics 2023, 12, 4327. [CrossRef]
5. Zaery, M.; Wang, P.; Wang, W.; Xu, D. A Novel Optimal Power Allocation Control System with High Convergence Rate for DC
Microgrids Cluster. Energies 2022, 15, 3994. [CrossRef]
6. Radisavljevic-Gajic, V.; Karagiannis, D.; Gajic, Z. Linear, Nonlinear, and Distributed-Parameter Observers Used for (Renewable)
Energy Processes and Systems—An Overview. Energies 2024, 17, 2700. [CrossRef]
7. Jain, H.; Mather, B.; Jain, A.K.; Baldwin, S.F. Grid-Supportive Loads—A New Approach to Increasing Renewable Energy in Power
Systems. IEEE Trans. Smart Grid 2022, 13, 2959–2972. [CrossRef]
8. Gómez-Restrepo, A.M.; González-Ruiz, J.D.; Botero Botero, S. Financial Investment Valuation Models for Photovoltaic and Energy
Storage Projects: Trends and Challenges. Energies 2024, 17, 2653. [CrossRef]
9. Abdolrasol, M.G.M.; Mohamed, R.; Hannan, M.A.; Al-Shetwi, A.Q.; Mansor, M.; Blaabjerg, F. Artificial Neural Network Based
Particle Swarm Optimization for Microgrid Optimal Energy Scheduling. IEEE Trans. Power Electron. 2021, 36, 12151–12157.
[CrossRef]
10. Xu, H.; Sun, J.; Huang, J.; Lin, X.; Ma, C. Distributed Optimization of Islanded Microgrids Integrating Multi-Type VSG Frequency
Regulation and Integrated Economic Dispatch. Energies 2024, 17, 1618. [CrossRef]
11. Zhu, X.; Wang, Y.; Liu, C.; Hou, N.; Li, Y.; Chen, Z. A Dual-Level Optimal Control Strategy for Offshore Microgrid Considering
Efficiency and Operation Cost in Wide Load Range. IEEE Trans. Power Electron. 2024, 39, 6734–6744. [CrossRef]
12. Zhang, Y.; Ni, M. Fully Distributed Optimal Economic Dispatch for Microgrids under Directed Communication Networks
Considering Time Delays. Energies 2023, 16, 7898. [CrossRef]
13. Zhang, Y.; Li, H. Research on Economic Load Dispatch Problem of Microgrid Based on an Improved Pelican Optimization
Algorithm. Biomimetics 2024, 9, 277. [CrossRef]
14. Mahmoodi, M.; Shamsi, P.; Fahimi, B. Economic Dispatch of a Hybrid Microgrid with Distributed Energy Storage. IEEE Trans.
Smart Grid 2015, 6, 2607–2614. [CrossRef]
15. Lin, F.J.; Liao, J.C.; Zhang, Y.M.; Huang, Y.C. Optimal Economic Dispatch and Power Generation for Microgrid Using Novel
Lagrange Multipliers-Based Method with HIL Verification. IEEE Syst. J. 2023, 17, 4533–4544. [CrossRef]
16. Hussain, A.; Bui, V.-H.; Kim, H.-M. A Proactive and Survivability-Constrained Operation Strategy for Enhancing Resilience of
Microgrids Using Energy Storage System. IEEE Access 2018, 6, 75495–75507. [CrossRef]
17. Xu, Y.; Liu, J.; Cui, Z.; Liu, Z.; Dai, C.; Zang, X.; Ji, Z. Economic Scheduling Model of an Active Distribution Network Based on
Chaotic Particle Swarm Optimization. Information 2024, 15, 225. [CrossRef]
18. Bin, L.; Shahzad, M.; Javed, H.; Muqeet, H.A.; Akhter, M.N.; Liaqat, R.; Hussain, M.M. Scheduling and Sizing of Campus
Microgrid Considering Demand Response and Economic Analysis. Sensors 2022, 22, 6150. [CrossRef]
19. Mu, C.; Shi, Y.; Xu, N.; Wang, X.; Tang, Z.; Jia, H.; Geng, H. Multi-Objective Interval Optimization Dispatch of Microgrid via Deep
Reinforcement Learning. IEEE Trans. Smart Grid 2024, 15, 2957–2970. [CrossRef]
20. Zhang, Y.; Lv, Y.; Zhou, Y. Research on Economic Optimal Dispatching of Microgrid Based on an Improved Bacteria Foraging
Optimization. Biomimetics 2023, 8, 150. [CrossRef]
21. Behnamfar, M.R.; Barati, H.; Karami, M. Stochastic Multi-objective Short-term Hydro-thermal Self-scheduling in Joint Energy and
Reserve Markets Considering Wind-Photovoltaic Uncertainty and Small Hydro Units. J. Electr. Eng. Technol. 2021, 16, 1327–1347.
[CrossRef]
22. Tan, Y.; Li, Y.; Cao, Y.; Shahidehpour, M. Cyber-Attack on Overloading Multiple Lines: A Bilevel Mixed-Integer Linear Program-
ming Model. IEEE Trans. Smart Grid 2018, 9, 1534–1536. [CrossRef]
Electronics 2024, 13, 3139 20 of 21

23. Patil, R.; Boit, S.V.; Gudivada; Nandigam, J. A Survey of Text Representation and Embedding Techniques in NLP. IEEE Access
2023, 11, 36120–36146. [CrossRef]
24. Zhang, H.; Song, Y.; Yang, M.; Jia, Q. Modeling and Optimization of LoRa Networks under Multiple Constraints. Sensors 2023, 23,
7783. [CrossRef] [PubMed]
25. Nayak, A.; Rathinam, S. Heuristics and Learning Models for Dubins MinMax Traveling Salesman Problem. Sensors 2023, 23, 6432.
[CrossRef] [PubMed]
26. Qiu, H.; Gooi, H.B. A Unified MILP Solution Framework for Adaptive Robust Scheduling Problems with Mixed-Integer Recourse
Objective. IEEE Trans. Power Syst. 2023, 38, 952–955. [CrossRef]
27. Zou, L.; Wang, K.; Wang, X.; Zhang, J.; Li, R.; Wu, Z. Automatic Recognition Reading Method of Pointer Meter Based on
YOLOv5-MR Model. Sensors 2023, 23, 6644. [CrossRef]
28. Liu, J.; Liang, Y.; Chen, Z.; Li, H.; Zhang, W.; Sun, J. A Double-Layer Vehicle Speed Prediction Based on BPNN-LSTM for Off-Road
Vehicles. Sensors 2023, 23, 6385. [CrossRef]
29. Pulatov, I.; Oteniyazov, R.; Makhmudov, F.; Cho, Y.-I. Enhancing Speech Emotion Recognition Using Dual Feature Extraction
Encoders. Sensors 2023, 23, 6640. [CrossRef]
30. Wang, C.; Chen, Y.; Liu, F.; Elliott, M.; Kwok, C.F.; Pena-Solorzano, C.; Frazer, H.; McCarthy, D.J.; Carneiro, G. An Interpretable
and Accurate Deep-Learning Diagnosis Framework Modeled with Fully and Semi-Supervised Reciprocal Learning. IEEE Trans.
Med. Imaging 2024, 43, 392–404. [CrossRef]
31. Manbachi, M.; Ordonez, M. Intelligent Agent-Based Energy Management System for Islanded AC–DC Microgrids. IEEE Trans.
Ind. Inform. 2020, 16, 4603–4614. [CrossRef]
32. Keypour, R.; Adineh, B.; Khooban, M.H.; Blaabjerg, F. A New Population-Based Optimization Method for Online Minimization of
Voltage Harmonics in Islanded Microgrids. IEEE Trans. Circuits Syst. II Express Briefs 2020, 67, 1084–1088. [CrossRef]
33. Alotaibi, M.M.; Alturki, A.A. Optimizing Renewable Energy Integration for Sustainable Fuel Production: A Techno-Economic
Assessment of Dimethyl Ether Synthesis via a Hybrid Microgrid-Hydrogen System. Fuels 2024, 5, 176–209. [CrossRef]
34. Amiri, F.; Eskandari, M.; Moradi, M.H. Virtual Inertia Control in Autonomous Microgrids via a Cascaded Controller for Battery
Energy Storage Optimized by Firefly Algorithm and a Comparison Study with GA, PSO, ABC, and GWO. Energies 2023, 16, 6611.
[CrossRef]
35. Hussein, H.M.; Rafin, S.M.S.H.; Abdelrahman, M.S.; Mohammed, O.A. Hardware Implementation of a Resilient Energy Manage-
ment System for Networked Microgrids. World Electr. Veh. J. 2024, 15, 209. [CrossRef]
36. de la Cruz, J.; Wu, Y.; Candelo-Becerra, J.E.; Vásquez, J.C.; Guerrero, J.M. Review of Networked Microgrid Protection: Architec-
tures, Challenges, Solutions, and Future Trends. CSEE J. Power Energy Syst. 2024, 10, 448–467. [CrossRef]
37. Gopalan, S.A.; Sreeram, V.; Iu, H.H.C. A review of coordination strategies and protection schemes for microgrids. Renew. Sustain.
Energy Rev. 2014, 32, 222–228. [CrossRef]
38. Guan, Y.; Wei, B.; Guerrero, J.M.; Vasquez, J.C.; Gui, Y. An overview of the operation architectures and energy management
system for multiple microgrid clusters. iEnergy 2022, 1, 306–314. [CrossRef]
39. Meng, L.; Shafiee, Q.; Trecate, G.F.; Karimi, H.; Fulwani, D.; Lu, X.; Guerrero, J.M. Review on Control of DC Microgrids and
Multiple Microgrid Clusters. IEEE J. Emerg. Sel. Top. Power Electron. 2017, 5, 928–948. [CrossRef]
40. Bui, V.-H.; Hussain, A.; Kim, H.-M. A Multiagent-Based Hierarchical Energy Management Strategy for Multi-Microgrids
Considering Adjustable Power and Demand Response. IEEE Trans. Smart Grid 2018, 9, 1323–1333. [CrossRef]
41. KHOLER Company. Energy Storage. Available online: https://www.clarke-energy.com/us/energy-storage/ (accessed on 15
January 2024).
42. Ibrahim, H.; Ilinca, A.; Perron, J. Energy storage systems—Characteristics and comparisons. Renew. Sustain. Energy Rev. 2008, 12,
1221–1250. [CrossRef]
43. Diaz-Gonzalez, F.; Bianchi, F.D.; Sumper, A.; Gomis-Bellmunt, O. Control of a Flywheel Energy Storage System for Power
Smoothing in Wind Power Plants. IEEE Trans. Energy Convers. 2013, 29, 204–214. [CrossRef]
44. Mirjalili, S.; Mirjalili, S.M.; Lewis, A. grey wolf optimizer. Adv. Eng. Softw. 2014, 69, 46–61. [CrossRef]
45. Zhang, Y. Modified grey wolf optimization algorithm for global optimization problems. J. Univ. Shanghai Sci. Techol. 2021, 43,
73–82. [CrossRef]
46. Alzubi, Q.M.; Anbar, M.; Alqattan, Z.N.; Al-Betar, M.A.; Abdullah, R. Intrusion detection system based on a modified binary grey
wolf optimisation. Neural Comput. Appl. 2020, 32, 6125–6137. [CrossRef]
47. Sayed, G.I.; Tharwat, A.; Hassanien, A.E. Chaotic dragonfly algorithm: An improved metaheuristic algorithm for feature selection.
Appl. Intell. 2019, 49, 188–205. [CrossRef]
48. Long, W.; Jiao, J.; Liang, X.; Cai, S.; Xu, M. A Random Opposition-Based Learning Grey Wolf Optimizer. IEEE Access 2019, 7,
113810–113825. [CrossRef]
49. Varol Altay, E.; Alatas, B. Bird swarm algorithms with chaotic mapping. Artif. Intell. Rev. 2020, 53, 1373–1414. [CrossRef]
50. Ramadan, A.; Kamel, S.; Hassan, M.H.; Ahmed, E.M.; Hasanien, H.M. Accurate Photovoltaic Models Based on an Adaptive
Opposition Artificial Hummingbird Algorithm. Electronics 2022, 11, 318. [CrossRef]
51. Bao, X.; Jia, H.; Lang, C. Dragonfly Algorithm with Opposition-Based Learning for Multilevel Thresholding Color Image
Segmentation. Symmetry 2019, 11, 716. [CrossRef]
Electronics 2024, 13, 3139 21 of 21

52. Sun, W.Z.; Wang, J.S. Elman Neural Network Soft-Sensor Model of Conversion Velocity in Polymerization Process Optimized by
Chaos Whale Optimization Algorithm. IEEE Access 2017, 5, 13062–13076. [CrossRef]
53. Bui, D.K.; Nguyen, T.N.; Ngo, T.D.; Nguyen-Xuan, H. An artificial neural network (ANN) expert system enhanced with the
electromagnetism-based firefly algorithm (EFA) for predicting the energy consumption in buildings. Energy 2020, 190, 116370.
[CrossRef]
54. Almomani, O. A Feature Selection Model for Network Intrusion Detection System Based on PSO, GWO, FFA and GA Algorithms.
Symmetry 2020, 12, 1046. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.