This article has been accepted for publication in IEEE Access.

This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3351710

Date of publication xxxx 00, 0000, date of current version xxxx 00, 0000.
Digital Object Identifier 10.1109/ACCESS.2017.Doi Number

Multi-objective Security Constrained Unit

Commitment Via Hybrid Evolutionary
Algorithms
AAMIR ALI1, ARSLAN SHAH1, M. U. KEERIO1, NOOR HUSSAIN MUGHERI1, GHULAM
ABBAS2,E. TOUTI 3, MOHAMMED HATATAH 4, AMR YOUSEF5,6 , MOUNIR BOUZGUENDA7
1
Department of Electrical Engineering, Quaid-e-Awam University of Engineering Science and Technology Nawabshah 67450, Sindh, Pakistan
2
School of Electrical Engineering, Southeast University, Nanjing 210096, China
3
Department of Electrical Engineering, College of Engineering, Northern Border University, Arar 73222, Saudi Arabia
4
Department of Electrical Engineering, Al-Baha University, Alaqiq 65779-7738, Saudi Arabia
5
Department of Electrical Engineering, College of Engineering, University of Business and Technology, Jeddah 9 21589, Saudi Arabia
6
Engineering Mathematics Department, Faculty of Engineering, Alexandria University, Egypt
7
Department of Electrical Engineering King Faisal University Alahsa Saudi Arabia, PC 31982

Corresponding author: lashariabbas@seu.edu.cn

ABSTRACT
This paper addresses the challenging problem of Unit Commitment (UC), which involves the optimal
scheduling of power generation units while adhering to numerous network operational constraints called
security-constrained UC (SCUC). SCUC problem aims to minimize costs subject to turning on
economically efficient generators and turning off expensive ones. These operational constraints include
load balancing, voltage level at buses, minimum up and down time requirements, spinning reserve, and
ramp up and down constraints. The SCUC problem, subject to these operational constraints, is a complex
mixed-integer nonlinear problem (MINLP). There has been a growing interest in using evolutionary
algorithms (EAs) to tackle large-scale multi-objective MINLP problems in recent two decades. This
paper introduces a novel approach to address the SCUC problem, which is further complicated by
including network constraints. They are pioneering the integration of single and multi-objective EAs to
solve the SCUC problem while incorporating AC network constraints through hybrid binary and real
coded operators. The development of an ensemble algorithm that combines mixed real and binary coded
operators, extended by a bidirectional coevolutionary algorithm to tackle multi-objective SCUC
problems. The paper implements a new formulation based on three conflicting objective functions: cost
of energy supplied, startup and shutdown costs of generators, energy loss, and voltage deviation to solve
the SCUC problem. Implementing a new formulation also addresses the solution of single and multi-
objective SCUC problems using a combination of proposed technical and economic objective functions.
The proposed algorithm is rigorously tested on a 10-unit IEEE RTS system and a 6-unit IEEE 30-bus
test system, both with and without security constraints, addressing week-ahead and day-ahead SCUC
scenarios. Simulation results show that the proposed algorithm finds near-global optimal solutions
compared to other state-of-the-art EAs. Additionally, the research demonstrates the effectiveness of the
proposed search operator by integrating it with a multi-objective coevolutionary algorithm driven by
both feasible and infeasible solutions, showcasing superior performance in solving multi-objective
SCUC problems. These results are compared with various recently implemented Multi-Objective
Evolutionary Algorithms (MOEAs), demonstrating the superiority of the proposed algorithm in terms
of convergence and diversity. A comparison of simulation results demonstrates that the proposed
algorithm finds better convergence and diversity than state-of-the-art MOEAs.

INDEX TERMS Security Constrained Unit commitment; Evolutionary algorithms; Optimal Power
Flow; Constraint Handling Techniques; Multi-objective optimization

VOLUME XX, 2017 1

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3351710

NOMENCLATURE
Abbreviations Indices/ Variables/ Parameters
UC Unit Commitment 𝑓1 Cost of active power generation and Startup cost
SCUC Security Constrained Unit Commitment 𝑓2 Cost of Energy loss
MILP Mixed Integer Linear Problem 𝑓3 Voltage Deviation
MINLP Mixed Integer Nonlinear Problem 𝐶(𝑃𝑔 ) Total operating cost of thermal generators
𝑡
OPF Optimal Power Flow 𝑃𝑔𝑖 Active power generation of unit i at time t
RTS Reliability Test System 𝑆𝑈𝑖 The start-up Cost of ith thermal generator
EAs Evolutionary Algorithms 𝑆𝐷𝑖 Shut Down cost of ith thermal generator
DE Differential Evolution 𝑃𝐺𝑖 and 𝑄𝐺𝑖 Actual active and reactive power injection of ith unit
MOEAs Multi-objective Evolutionary Algorithms 𝐹𝑖 (𝑃𝑖 (𝑡)) Quadratic Cost of ith Thermal Generators
ANN Artificial Neural network 𝑃𝐷 and 𝑄𝐷 Active and reactive power demand
LR Lagrangian Relaxation 𝑉𝐺𝑖 The voltage set point of ith generator
BiCo Bi-directional Co-evolution 𝑃𝑙𝑜𝑠𝑠 , 𝑄𝑙𝑜𝑠𝑠 Active and Reactive Power Loss
GA Genetic Algorithm a, b, c Thermal generator Cost parameters
NSGAII Nondominated Sorting Genetic Algorithms 𝑆𝑖 (𝑣𝑖 (𝑡)) Startup cost
SVC Shunt Var Compensator 𝑆𝑖 (𝑤𝑖 (𝑡)) Shut-down cost
AD Angle-based Density 𝑈𝑅𝑖 Ramp up rate
CV Constraint Violation 𝐷𝑅𝑖 Ramp down rate
VD Voltage Deviation 𝑈𝐻𝑖 Ramp up
CEL Cost of Energy loss 𝜏𝑗 transformer tap setting
PF Pareto Front 𝐶𝑖 (𝑡) Constrained generating capability of unit 𝑖 at hour 𝑡
PS Pareto Set 𝐹𝑖 (𝑃𝑖𝑡 ) Cost of ith at output power 𝑃𝑖𝑡 at current time t
HVI Hyper Volume Indicator 𝑡 = 1, … , 𝑇𝑛 Total number of hours
BCS Best Compromise Solution 𝑖 = 1, … , 𝑁𝑔 Total number of generators
FR Feasibility Ratio 𝑉𝐿 Load Bus voltage
CDP Constraint Domination Principle 𝑁𝐿 Number of Load buses
Binary decision vector of generator 𝑖 at hour 𝑡, 1 if unit
Indices/ Variables/ Parameters 𝑢𝑖 (𝑡)
is online and 0 elsewhere
𝑇𝑛 Span length of time 𝑡𝑢𝑝 and 𝑡𝑑𝑜𝑤𝑛 Minimum up and down time
𝑁𝐺 Total number of generators 𝑈𝑅𝑖 and 𝐷𝑅𝑖 Ramp up and down limits
𝑁𝜏 Total number of transformers 𝒙 Decision vector
𝑁𝐶 Total number of SVC ℎ(𝒙) Equality constraint Function
𝑛𝑙 Total Number of Transmission lines 𝑔(𝒙) Inequality constraint Function
𝑎𝑖 , 𝑏𝑖 , 𝑐𝑖 Cost parameters of thermal generators 𝛿𝑚𝑛 Voltage angle difference of branch between bus m and n
Total active and reactive load demand at time the transfer conductance of branch (line) q connecting
𝑃𝐷𝑡 , 𝑄𝐷𝑡 𝐺𝑞(𝑚𝑛)
𝑡 buses m and n
Binary startup and shut down states of generator i at time
𝑃𝑖𝑚𝑎𝑥 , 𝑃𝑖𝑚𝑖𝑛 Rated maximum and minimum of generator 𝑖 𝑣𝑖 (𝑡) and 𝑤𝑖 (𝑡)
t, 1 if unit is start up or shut down at time t, 0 elsewhere
𝑅(𝑡) System spinning reserve requirement at hour 𝑡 𝑆𝑙𝑞 Actual MVA branch flow limit

I. INTRODUCTION task. Due to this, numerous works in literature suggest

The unit commitment problem (UC) involves various methods for resolving this issue in the best
determining the optimal startup and shut down schedules possible way, making them a crucial area for operational
of thermal generators and the economic dispatch of the research advancements. To correctly operate the
online generators to meet the forecasted demand over a generators and reduce costs, UC must decide which
specific short-term time period (e.g., 24 hours). The UC generators to turn on and which to turn off based on
problem is one of the most important problems in power variation in load demand and subject to satisfy various
system operation. The UC problem is a day-ahead operational constraints [8-10] and the committed units
scheduling problem and comprises of two tasks: one is must also be economically dispatched [11]. The
determining the on/off status of the thermal units; the other computational time complexity of the UC problem
is the optimal power flow (OPF) which requires increases exponentially as the size of a network is
distributing the system load demand to the committed increased [12]. The biggest expense associated with the
thermal units [1]. The various past to recent survey papers unit commitment problem is fuel usage [13]. The authors
[2-6] describe that the UC is a complex np-hard, non- in [5] highlights the significance of accurately
convex, mixed-integer nonlinear problem where the cost representing these costs by noting that a utility would gain
of generators is minimized to obtain the best schedule of annually millions of euros from a 0.5% fuel savings in
units. Optimizing electricity generation provides electricity generation.
numerous benefits to market participants and end users To maximize profitability [5, 9], research is concentrated
[7]. As a result of the problem's magnitude and on refining the modeling detail. In previous decades, the
computational constraints, nevertheless, it is not a simple thermal foundation of power systems was often provided

2 VOLUME XX, 2017

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3351710

by coal plants. Fast-ramping gas turbines were employed for the solution of generation scheduling [26]. In [27], a
to meet demand peaks whereas combined cycles gas feasible and near-optimal solution to the UC problem is
turbines were reserved for times of high demand [14]. obtained by relaxing the ramping constraints and using
Since this organization was stable throughout time, step functions—in this function online generator can inject
effective asset management did not necessitate significant 100% of its capacity—to express the generating
modeling advancements. In power system generation capabilities. After that, GA is implemented to generate
scheduling, two duties are taken into account. One of these possible scheduling, and a heuristic method is applied to
is the unit commitment, which establishes the unit's start- solve UC. In [28], genetic algorithm were used to solve
up and shutdown schedules to reduce system fuel the UC problem. In recent years, researchers have shown
consumption. The other is economic dispatch, which interest and have looked for more effective ways to
allows system load demands to committed generating approach the UC problem. However, the non-convex UC
units to reduce the cost of power generation [15]. The problem has convergence issues when using classical
economic operation is noteworthy since a small decrease Lagrange relaxation (LR) approach [29]. These are
in fuel cost as a percentage results in significant system gradient-based and experience a significant bottleneck
operation cost savings [16]. The UC dilemma typically when they hit local minima [9]. In general, the UC
encompasses both of these decisions because they are problem is formulated as a nonconvex MINLP, and the
related. Finding the overall least expensive way to run the scale of this problem creates challenges to solving large
power system over the scheduling horizon is the goal. UC problems [30, 31]. The dramatic increase in the
Recently, many studies comprised of various single and efficiency of MINLP solvers has encouraged the thorough
multi-objective based on numerical, metaheuristic and exploitation of their capabilities [32].
hybrid combination of evolutionary and classical Linear approximation methods often simplify the unit
algorithms have been successfully applied to find the commitment problem by linearizing the cost functions,
optimal solution of UC problem. Classical Mixed integer which can lead to inaccurate results, especially in systems
Linear programming (MILP) based on Quadratic with nonlinear cost functions. This oversimplification may
programming given [17] and linear programming [18, 19], not account for the complexities and nuances of real-world
are the single objective optimization techniques which are power systems. Many unit commitment problems involve
used to minimize the cost subject to satisfy various nonlinear constraints, such as ramp rate limits, minimum
constraints. These techniques usually approximate the up and down times, and prohibited operating zones. Linear
nonlinear objective functions or constraints. approximation methods struggle to handle these nonlinear
The AI techniques have combined practical operational constraints effectively, leading to suboptimal solutions.
strategies with mathematical techniques to progress the Linear approximations may not capture the true operating
system models significantly. The mechanism of ANN characteristics of generating units accurately. This lack of
mimics the learning process of the human brain has been accuracy can lead to suboptimal schedules, increased
discussed in [5]. It has been proposed in several studies operating costs, and potential violations of system
looking at the unit commitment problem that the unit's constraints. Recently, various single objective
generating capability changes in steps from zero to the evolutionary algorithms (EAs) have been implemented to
rated capacity and vice versa [20] and startup and find the solution of UC problem these includes particle
shutdown costs proposed in [21], to avoid brittle failure of swarm optimization (PSO) [33], Coyote Optimization
before online of the unit. Further, most of the authors in Algorithm (COA) [34], Binary African Vultures
the literature consider unit step function to satisfy physical Optimization Algorithm (BAVOA) [10], monarch
constraints during online units such as ramp up and ramp butterfly optimization (MBO) [9], Gradient Based
down [22]. All activities are started as soon as the unit Optimizer (GBO) [35], Binaryfish migration optimization
reaches its rated capacity when employing a step function [36], binary particle swarm optimization (BPSO) [37],
to reflect changes in generating capability, which indicates improved PSO [38] and weighted improved crazy PSO
an unrealistic treatment of energy, especially when the unit (WICPSO) [39].
start-up is a lengthy process [23]. Similar to how it takes Single-objective EOAs typically aim to find a single
time for the turbine to cool down when a unit is shutting optimal solution based on a specific objective function,
down, so does it. Contrary to the scenario where the such as minimizing generation costs, emission rate, profit
changes in unit-producing capacity are treated as a step maximization etc. However, the UCP often involves
function, the remaining energy is to be utilized to meet the multiple conflicting objectives, including cost
load demand before the unit-generating capability minimization, reliability, and environmental impact.
declines to its lower limit [24]. Ramping up was once Single objective EAs explore the solution space through a
thought of as a dynamic dispatch in the economic dispatch. population of individuals, and they may not guarantee the
A dynamic procedure was carried out together with the exploration of the entire solution space. This limitation
economic dispatch to satisfy the ramping limits [25]. can result in the algorithm getting stuck in local optima,
Dynamic programming (DP) is considered to handle these missing global optimal solutions, and producing
ramping constraints in the economic dispatch sub-problem suboptimal schedules. Moreover, computational effort can
of the UC problem. DP-based algorithms are time- be challenging for large-scale UC problems with many
consuming, therefore, this study avoids DP to implement generating units and constraints. UC problems involves a

2 VOLUME XX, 2017

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3351710

variety of complex constraints, such as ramp rate limits, non-linear problem (MINLP) made even more complex by
minimum up and down times, and prohibited operating the presence of operational AC power flow constraints.
zones. Single-objective EAs may struggle to handle these Traditional optimization techniques struggle to find
constraints effectively, leading to solutions that violate optimal solutions for such intricate problems. In response,
technical or operational constraints. Furthermore, UC this paper employs a hybrid Genetic Algorithm (GA)
problem involves a variety of complex constraints, such as approach based on MINLP methodology to tackle the
ramp rate limits, minimum up and down times, and SCUC problem. For the multi-objective SCUC problem,
prohibited operating zones. Single-objective EAs may bidirectional coevolution (BiCo) based MOEA hybridized
struggle to handle these constraints effectively, leading to with the newly introduced variation operators to optimize
solutions that violate technical or operational limits. As
both single, bi and tri-objective functions. In the proposed
mentioned earlier, UC problem often involves multiple
algorithm, a straightforward binary encoding process is
conflicting objective functions. Single-objective EAs
adopted, where binary variables (u, v, and w in this paper)
cannot simultaneously optimize multiple objectives,
which limits their ability to explore trade-offs between are encoded as binary strings. If there are 𝑁𝑔 units and 𝑇𝑛
objectives effectively. scheduling periods in hours, each unit is either ON state
Multi-objective evolutionary algorithms (MOEAs), on (indicated by '1') or OFF (indicated by '0') at each hour.
the other hand, are specifically designed to address the Concatenating these binary strings for all 𝑁𝑔 units result
limitations of single objective EAs. They can in an 𝑁𝑔 -𝑇𝑛 bit string. Furthermore, multi-objective SCUC
simultaneously optimize multiple objectives, identify a set problem is solved through a two-step process. First, binary
of Pareto-optimal solutions that represent trade-offs, and coding is used to determine UC decision variables,
provide a more balanced and comprehensive approach to ensuring compliance with constraints such as minimum up
solving UC problems. As a result, in last two decades and down times, spinning reserves, and security
MOEAs attain increasingly favored to solve complex, constraints. In the second step, a wide range of
multi-objective optimization problems like UC. In the
nondominated solutions are obtained. Before evaluating
literature, various MOEAs based techniques were
the objective functions, all the constraints of SCUC
implemented to solve UC problem. These includes:
nondominated soring genetic algorithm (NSGAII) [40], problem must be addressed. Once all constraints are
multi-objective EA based on decomposition with the satisfied, the total operating cost, startup cost, cost of
binary variables are searched by GA search operators energy loss, and voltage deviation (VD) objective
(MOEA/D-GA) [41], multi-objective two-stage functions are computed. This approach ultimately yields
compromise programming (CP) [42], multi-objective an economically optimal unit commitment schedule that
based-on mixed-integer linear programming (MILP) [43] adheres to system operating constraints. This paper has the
and Multi-Objective Evolutionary Policy Search (MEPS) following four important contributions, which are
[44]. In the multi-objective UC (MOUC) problem mostly summarized as;
cost based, profit based, emission reduction based, and 1. A new formulation is implemented to find the solution
voltage stability index based objective functions are of single and multi-objective SCUC problem based
considered to find the optimal solution of UC problem. In on technical and economical objective functions. This
the context of recent literature, it is evident that the is the First attempt to propose a single and multi-
simultaneous consideration of combined technical and objective EAs to solve SCUC problem along with AC
economical objective functions in solving the multi- network constraints.
objective unit commitment (MOUC) problem has been 2. Efficiently solve the SCUC problem considering
largely overlooked. As a result, this paper introduces a network-based AC power flow constraints
novel formulation that integrates both economic and considering technical and economical single, bi and
technical objective functions to address the MOUC tri objective functions.
problem effectively. Moreover, IEEE 11units reliability 3. An ensemble algorithm based on integration of hybrid
test network and IEEE 6 units 30-bus test networks are real and binary coded operators’ strategy is developed
adopted to solve extensive case studies. Their comparison and extended with bidirectional coevolutionary
and analysis are presented to demonstrate the algorithm to solve Multi-objective SCUC problem.
effectiveness of the proposed hybrid strategy to solve unit 4. Week ahead and day ahead SCUC problems are
commitment of conventional thermal generators. In the solved on 11-unit IEEE RTS system and 6-unit IEEE
proposed formulation, UC decision variables and 30-bus test systems with and without security
generation scheduling variables are simultaneously constraints. Simulation results of proposed single and
obtained and applied to solve security constrained multi- multi-objective EAs have been compared and
objective UC problem. During optimization, step length analyzed with the recently implemented EAs and
binary decision variable is obtained by using crossover MOEAs.
and mutation operators of binary GA, whereas continuous The rest of this article is divided into the following
decision variables are searched using crossover and sections. The SCUC Problem formulation is provided in
mutation operators of real coded GA. Section II. In section III, specifics the proposed
The SCUC problem is a challenging mixed-integer methodology. Section IV presents proposed study cases

2 VOLUME XX, 2017

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3351710

and simulation results. Section V brings the conclusion. of branch (line) 𝑞 connecting buses 𝑚 and 𝑛. 𝑛𝑙 shows the
total number of transmission lines. Second objective
II. MATHEMATICAL PROBLEM FORMULATION function is minimization of total cost of energy loss in the
A. OBJECTIVE FUNCTIONS entire time horizon, and it is computed as
𝑁𝑡

In this paper, solution of multi-objective SCUC 𝑡

𝑓2 = 𝐶(𝑃𝑙𝑜𝑠𝑠 ) = a × ∑ 𝑃𝑙𝑜𝑠𝑠 (4)
problem is obtained by considering both technical and 𝑡=1
𝑡
economic objective functions. In the Proposed techno- Where, 𝑃𝑙𝑜𝑠𝑠 power loss of the network at time t slot and
economic multi-objective SCUC problem formulation ‘a’ is the cost loss co-efficient equal to 0.01$/MWh.
involves a delicate balance between minimization of
operation cost, cost of energy loss, and maximizing 3. NETWORK VOLTAGE DEVIATION (𝒇𝟑 )
voltage stability index. Therefore, in this paper, Voltage deviation (VD) is a measure of voltage quality in
economical objective functions such as operation cost of the network. The index of deviation is also important from
thermal generators along with the startup cost and cost of security aspect. The indicator is formulated as cumulative
energy loss and one technical objective functions such as deviation of voltages of all load buses (PQ buses) in the
voltage deviation are considered to find the solution of network from nominal value of unity. Mathematically, VD
single and multi-objective SCUC problem. in the entire time horizon is computed as;
𝑇𝑛 𝑁𝐿

𝑓3 = 𝑉𝐷 = ∑ (∑ |𝑉𝐿𝑡𝑝 − 1|) (5)

1. TOTAL OPERATING COST (𝒇𝟏 )
𝑡=1 𝑝=1
The generation scheduling challenge seeks to reduce
the overall cost of system operation while adhering to the 𝑁𝐿 shows number of load buses, 𝑉𝐿𝑝 is the load bus
system operating limits. This objective function comprises voltage.
the fuel cost for producing electricity as well as the start- B. CONSTRAINTS
up and shut-down costs of thermal generators. The total The constraints for the unit commitment problem are:
operating cost of the UC problem can be given as: 1. OPF NETWORK CONSTRAINTS
𝑇𝑛 𝑁𝑔
During optimization all the optimal power flow (OPF)
𝑓1 = 𝐶(𝑃𝑔 ) = ∑ ∑[𝐹𝑖 (𝑃𝑖 (𝑡)) × 𝑢𝑖 (𝑡) + 𝑆𝑈𝑖 (𝑣𝑖 (𝑡))
𝑡=1 𝑖=1
(1) constraints must be satisfied. In this work, after
+ 𝑆𝐷𝑖 (𝑤𝑖 (𝑡))] optimal unit commitment, following nonlinear AC OPF
balanced equality constraints given in Eq. (6) to (7)
Whereas the Fuel cost of generator i at time t can be must be satisfied during NR load flow and these
computed as; constraints for the entire time horizon are given as;
𝑇𝑛 𝑁𝑔
𝐹𝑖 (𝑃𝑖 (𝑡)) = 𝑎𝑖 + 𝑏𝑖 ∗ 𝑃𝑖 + 𝑐𝑖 ∗ 𝑃𝑖2 (2) 𝑁𝑡

Additionally, a significant amount of energy must be used ∑ ∑ 𝑃𝑖 (𝑡)𝑢𝑖 (𝑡) = ∑ 𝑃𝐷𝑡 ∗ 𝑢𝑖 (𝑡) + 𝑃𝑙𝑜𝑠𝑠 (6)
𝑡=1 𝑖=1 𝑡=1
to bring the thermal unit online because the temperature 𝑁𝑡 𝑁𝑔 𝑁𝑡
and pressure of the unit must be moved slowly. As a result ∑ ∑ 𝑄𝑖 (𝑡)𝑢𝑖 (𝑡) = ∑ 𝑄𝐷𝑡 ∗ 𝑢𝑖 (𝑡) + 𝑄𝑙𝑜𝑠𝑠 (7)
of the commitment and de-commitment status, this energy 𝑡=1 𝑖=1 𝑡=1
enters the unit commitment problem as a start-up cost On the other hand, all the inequality constraints are
𝑆𝑈𝑖 (𝑣𝑖 (𝑡)) but does not result in any MW integration in handled using constraint domination principle (discussed
the system unit. Suppose the status of unit i at 𝑢𝑖 (𝑡 − 1) is in section), these inequality constraints are;
𝑉𝐺𝑚𝑖𝑛 ≤ 𝑉𝐺𝑖 ≤ 𝑉𝐺𝑚𝑎𝑥 ∀𝑖 ∈ 𝑁𝐺 (8)
OFF and it becomes ON in the next hour at 𝑢𝑖 (𝑡), then 𝑖 𝑖
𝑃𝐺𝑚𝑖𝑛 ≤ 𝑃𝐺 ≤ 𝑃𝐺
𝑚𝑎𝑥
∀𝑖 ∈ 𝑁𝐺 (9)
startup cost 𝑆𝑈𝑖 (𝑣𝑖 (𝑡)) is applied and it depends upon the 𝑄𝐺𝑚𝑖𝑛
𝑖 𝑖 𝑖
≤ 𝑄𝐺𝑖 ≤ 𝑄𝐺𝑚𝑎𝑥 ∀𝑖 ∈ 𝑁𝐺
𝑖 𝑖
(10)
number of hours that unit i has been recommitted. Shut 𝜏𝑗𝑚𝑖𝑛 ≤ 𝜏𝑗 ≤ 𝜏𝑗𝑚𝑎𝑥 ∀𝑗 ∈ 𝑁𝜏 (11)
down cost 𝑆𝐷𝑖 (𝑤𝑖 (𝑡)) is active if the status of unit 𝑢𝑖 (𝑡 − 𝑄𝑐𝑚𝑖𝑛 ≤ 𝑄𝐶𝑘 ≤ 𝑄𝐶𝑚𝑎𝑥 ∀𝑘 ∈ 𝑁𝐶 (12)
𝑘 𝑘
1) is ON at time 𝑡 − 1 and becomes off in the next period 𝑉𝐿𝑝 ≤ 𝑉𝐿𝑝 ≤ 𝑉𝐿𝑚𝑎𝑥
𝑚𝑖𝑛
𝑝
∀𝑝 ∈ 𝑁𝐿 (13)
𝑢𝑖 (𝑡). In the literature [1], most researchers neglect the 𝑆𝑙𝑞 ≤ 𝑆𝑙𝑞𝑚𝑎𝑥 ∀𝑞 ∈ 𝑛𝑙 (14)
shutdown cost because it is often modeled as a constant Where, 𝑉𝐺𝑖 is the voltage set point of ith generator, 𝑃𝐺𝑖 and
cost parameter in the thermal generators or its value is
𝑄𝐺𝑖 are the active and reactive power generation of ith
small compared to startup cost; therefore, in this paper
shut-down cost is also neglected during de-commitment. generator, 𝜏𝑗 shows the transformer tap setting of
connected in jth branch, 𝑄𝐶𝑘 is the MVAr injection at kth
2. COST OF ENERGY LOSS (𝒇𝟐 ) of shunt VAR compensator, 𝑉𝐿𝑝 is the pth load bus voltage
Due to inherent resistance of the transmission system, the and 𝑆𝑙𝑞 is MVA flow in the qth branch. Whereas min and
real power loss is unavoidable. The real power loss (in max superscripts are the minimum and maximum values
𝑀𝑊) in a transmission line is expressed as: associated variables.
𝑛𝑙

𝑃𝑙𝑜𝑠𝑠 = ∑ 𝐺𝑞(𝑚𝑛) [𝑉𝑚2 + 𝑉𝑛2 − 2𝑉𝑚 𝑉𝑛 cos (𝛿𝑚𝑛 )] (3) 2. SYSTEM SPINNING RESERVE
𝑞=1 REQUIREMENTS:
where 𝛿𝑚𝑛 = 𝛿𝑚 − 𝛿𝑛 is the voltage angle difference of This disparity ensures a technical requirement of power
buses 𝑚 and 𝑛. 𝐺𝑞(𝑚𝑛) signifies the transfer conductance systems, namely the availability of additional generation
capacity set aside for risky scenarios, such as the loss of a

2 VOLUME XX, 2017

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3351710

committed thermal unit, to maintain supply security. It is C. DECISION VARIABLES

a linear inequality with continuous variables as well. The study presented in this paper considers load variation
𝑁
in 𝑇𝑛 time periods and computes upper and lower bounds
∑ 𝑃𝑖𝑚𝑎𝑥 𝑢𝑖 (𝑡) ≥ 𝑃𝐷𝑡 + 𝑅(𝑡) (15)
1=1
of decision variable in the entire time period. In UC
problem some constraints such as ramp rate (ramp up and
3. MINIMUM UP AND MINIMUM DOWN ramp down) and min up and min down constraints are
CONSTRAINTS dynamic and these constraints are not satisfied
To achieve technical constraints that lower the probability individually. Therefore, in this work, decision variables
of failure, this inequality is used to ensure that the unit is for each time period are stacked to form a large-scale
online for a minimum duration of time since it is started master problem. In this master problem objective
up or that it is offline for a minimum period since it is shut functions and constraints for each time period are treated
down. It uses integer variables and is linear. Once the unit as separate islands in a stacked network of entire time
is committed then it should not be de-committed horizon. Single stacked solution of entire time horizon
immediately and if the unit is de-committed it should not effectively optimizes the objective function and satisfy all
be recommitted after a minimum down. In the proposed the dynamic intertemporal constraints. The final stacked
method min down and min up constraints [45] can be UC problem allows greater flexibility in the management
computed as. of power systems, as it provides a single solution for the
𝑡
entire time horizon that can be used to make decisions in
∑ 𝑣 𝑦𝑖 ≤ 𝑢𝑡𝑖 (16)
real-time. The decision vector for the proposed problem,
𝑦=𝑡−𝑡𝑢𝑝 +1
𝑡 denoted as 𝒙, is defined in the study:
∑ 𝜔 𝑦𝑖 ≤ 1 − 𝑢𝑡𝑖 (17) 𝒙 = [𝑃𝑖𝑡 , 𝑉𝑖𝑡 , 𝜏𝑗 , 𝑆𝑉𝐶𝑘 , 𝑢𝑖 , 𝑣𝑖 , 𝑤𝑖 ] (20)
𝑦=𝑡−𝑡𝑑𝑜𝑤𝑛 +1
Whereas 𝑢, 𝑣 and 𝜔 for ith generator are the binary Where, subscript i, j and k show the indexes of generators,
decision variables for the entire time horizon, value of u is transformer and shunt VAR compensators, P and V are the
one shows the unit is online whereas zero shows output power of generators and voltage set point of ith
decommitment of ith unit. Value of 𝑣 is one shows that unit generator, 𝜏 is transformer tap ratio, 𝑢, 𝑣 𝑎𝑛𝑑 𝑤 are the
has changes it positions form 0 (at t-1 period) to 1 (at t unit commitment, startup and shut down binary variables,
time slot). On the other hand, 𝜔 is the indicator that shows 1 if unit is committed, startup or shut down and 0
that unit has change it status from 1 to zero. We can make elsewhere. In the proposed formulation decision variables
these sums given in Eqs. (16)-(17) "go in a circle" to create comprised of mixed integer variables.
a commitment plan that goes from the end of one time
period back to the beginning, and we make sure that this III. OPTIMIZATION METHOD
plan always works. Many real-world applications involve simultaneous
optimization of several objective functions, which are
4. RAMP-RATE LIMITS FOR UNIT GENERATION
CHANGES often conflicting with each other, and subject to a number
A thermal unit's power generation during the prior and of equality and inequality constraints. Unit commitment is
current time steps cannot differ by more than the ramping a mixed integer nonlinear problem (MINLP) and involve
rates thanks to this inequality. It employs continuous both integer and binary decision variables. Without loss of
variables and is a linear inequality, computed as; generality multi-objective optimization problem is stated
𝑃𝑖 (𝑡) − 𝑃𝑖 (𝑡 − 1) ≤ 𝑈𝑅𝑖 − 𝑚𝑎𝑥(𝑃𝑖𝑚𝑖𝑛 − 𝑈𝑅, 0) × 𝑣𝑖 (𝑡) (18) as;
𝑃𝑖 (𝑡 − 1) − 𝑃𝑖 (𝑡) ≤ 𝑚𝑎𝑥(𝑃𝑡𝑚𝑖𝑛 , 𝐷𝑅𝑖 )
− 𝑚𝑎𝑥(𝑃𝑡𝑚𝑖𝑛 − 𝐷𝑅, 0) × 𝑢𝑡𝑖 (19) 𝑚𝑖𝑛 𝐹(𝒙) = (𝑓1 (𝒙), 𝑓2 (𝒙), … , 𝑓𝑚 (𝒙))
𝑠. 𝑡. ℎ𝑖 (𝒙) = 0, 𝑖 = 1, … , 𝑝
Whereas 𝑃𝑖 (𝑡) shows the output power of ith generator at 𝑔𝑖 (𝒙) ≤ 0, 𝑖 = 𝑝 + 1, … , 𝑞 (21)
tth time slot. The generation level of ith generator cannot 𝒙 ∈ ℝ𝑛
increase or decrease in one hour by more than its ramp up 𝒙𝒊 ∈ {𝟎, 𝟏}
(𝑈𝑅𝑖 )/ ramp down (𝐷𝑅𝑖 ) limit, except when it is Where, 𝑓1 (𝒙), 𝑓2 (𝒙), … , 𝑓𝑚 (𝒙) are the m real valued
recommitted. When recommitted (turning on) it must be conflicting objective functions, m=1 is for the single
able to get from zero to its minimum generation level,
objective optimization and value of m is 2 or 3 for the
even if that gap is larger than its ramp up limit. Likewise,
when decommitted (turning off), it must be able to get multi-objective optimization. Whereas ℎ𝑖 (𝒙) and 𝑔𝑖 (𝒙)
from its minimum generation level down to zero, even if are 𝑝 and 𝑝 − 𝑞 nonlinear equality and inequality and 𝒙 is
that gap is larger than its ramp down limit. Constraints in the n-dimensional decision vector and 𝒙𝒊 is the subset of
Eq. (15-19) are dynamic/ intertemporal constraints; decision vector that holds sequence of binary constraints.
therefore, it is necessary to formulate a decision variable
that considers dynamic constraints appropriately.

2 VOLUME XX, 2017

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3351710

1) PARETO SET AND PARETO FRONT application of GAs, a simple binary solution was chosen
In the proposed formulation the ith degree of to encode a 𝑢, 𝑣 and 𝑤 variables. If 𝑁 represents the
constraint violation at a given decision vector 𝒙 can be number of units and T the scheduling period in hours, then
computed as; with the assumption that at every hour a certain unit can
be either ON or OFF. In such a string, a ‘1’ at a certain
max(0, 𝑔𝑖 (𝒙)), ∀ 𝑖 ≤ 𝑝 location indicates that the unit is ON at this particular hour
𝑐𝑖 (𝒙) = { (22) while a ‘0’ indicates that the unit is OFF. By concatenating
max(0, |ℎ𝑖 (𝒙) − 𝜖|) , 𝑒𝑙𝑠𝑒
the strings of the 𝑁 units an N‐H bit string is formed. In
Whereas, 𝜖 is the tollerance value used to relax the
the proposed formulation, UC problem is solved by two
equality constraints. Usually, over all constraint voilation separate measures. In the first step, UC decision variables
(CV) for all the constraints is computed as; are obtained using binary coding, in this step, all the
constraints of UC variables i.e., minimum up and down,
𝑞
spinning reserves and security constraints are checked.
𝐶𝑉(𝑥) = ∑ 𝑐𝑖 (𝒙) (23)
However, in the second step, decision variables of UC
𝑖=1
problems are passed to compute feasible solutions by
Decision vector x is feasible if CV(x) is zero, else it applying load flow techniques. Before, the evaluation of
is infeasible solution. Pareto set or Pareto optimal objective function it is desirable to repair the constraints
solutions (PS) are the set of all solutions that correspond of the OPF these are ramp up and ramp down. After
to feasible regions. Image of PS in the objective space is satisfying all the constraints given in Eq. (5)-(13), total
called Pareto Front (PF). operating cost along with start-up cost, Cost of energy loss
and VD objective functions are computed, and obtaining
2) PARETO DOMINANCE OR NONDOMINATED an economical unit commitment schedule, which satisfies
SOLUTIONS: the system operating constraints. The Unit Commitment
From the randomly selected two decision vectors, say (UC) problem represents a complex multi-objective
𝑥𝑢 and 𝑥𝑣 , 𝑥𝑢 is pareto dominance on 𝑥𝑣 if 𝑓(𝑥𝑢 ) ≤ 𝑓(𝑥𝑣 ) optimization challenge, involving dynamic and security
for all objective functions and for at least one objective constraints. Typically, in such scenarios, the Pareto-
function say 𝑗, 𝑓𝑗 (𝑥𝑢 ) < 𝑓𝑗 (𝑥𝑣 ), then 𝑥𝑢 is said to optimal solutions are located along the edges of these
dominate 𝑥𝑣 , and 𝑥𝑢 is considered non-dominated constraints. The primary goal of Multi-Objective
solutions. Evolutionary Algorithms (MOEAs) is to enhance the
3) CONSTRAINT DOMINATION PRINCIPLE (CDP): diversity and convergence of the Pareto Front (PF).
In this paper constraint domination Principle (CDP) However, accomplishing these objectives is far from
proposed in [46] is applied to handle infeasible solutions. straightforward, especially given the intricacies of
In this technique two solutions are randomly selected and network security constraints. While most MOEAs aim to
compared as: optimize the problem to emphasize feasible solutions, they
may give rise to the following two issues.
 If both solutions 𝑥⃗𝑢 and 𝑥⃗𝑣 are infeasible, select 𝑥⃗𝑢 1) Population becoming trapped within local feasible
if 𝐶𝑉(𝑥𝑢 ) < 𝐶𝑉(𝑥𝑣 ). areas or locally optimal feasible regions.
 𝑥𝑢 is feasible and 𝑥𝑣 is infeasible, select the 2) The driving force may be constrained as the
feasible one i.e., 𝑥𝑢 . population's evolution is limited to the feasible portion of
 If both 𝑥𝑢 and 𝑥𝑣 are feasible, then select 𝑥𝑢 if the search space.
for all the objective functions 𝑓𝑖 (𝑥𝑢 ) ≤ 𝑓𝑖 (𝑥𝑣 ). To address these challenges, the proposed algorithm
4) PROPOSED HYBRID EVOLUTIONARY explores the search space by coevolving two populations:
ALGORITHM the feasible main population (𝑃𝑡 ) and the representative
UC problem is a mixed integer non-linear problem infeasible archived population (𝐴𝑡 ) [47]. The proposed
(MINLP), and the optimal solution of MINLP is hard to algorithm effectively guides solutions towards the PF
find using classical optimization techniques. Moreover, from both the feasible main population and the infeasible
the complexity of proposed UC problem is highly archive population sides of the search space, a crucial
increased with the satisfaction of network security aspect in CMOP. Additionally, a novel angle-based
constraints. Classical optimization techniques are unable density (AD) selection scheme is introduced to update the
to find the optimal solution of such problem. In the last 𝑃𝑡 and 𝐴𝑡 . This scheme not only preserves search
few decades, evolutionary algorithms (EAs) were diversity, aiding the discovery of more feasible regions,
efficiently applied to solve the hard MINLP problems. but also keeps infeasible solutions close to the PF, thereby
Therefore, in this paper, hybrid real and binary coded accelerating the quest for Pareto-optimal solutions. To
Genetic Algorithm (GA) based methodology is applied to harmonize the interactions between the main and archive
solve the proposed UC problem. For Multi-objective UC populations and leverage their complementary
problem, bidirectional coevolution based MOEA information, the proposed algorithm incorporates a new
techniques proposed in [47] is hybrid with the proposed restricted mating selection mechanism. The flow diagram
variation operators has been applied to find the optimal of proposed algorithm is shown in Fig. 1.
solution of bi and tri objective functions. For the

2 VOLUME XX, 2017

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3351710

Input Data:
Load IEEE test system; 𝐺𝑚𝑎𝑥 = 500; Population size N=50; 𝑥 𝐿 and 𝑥 𝑈 bound
of DV; Number of current iterations 𝐺=0;

Initialization:
𝑃𝑡 ; % Randomly Generate Initial Main Population and evaluate
𝐴𝑡 = ∅ % initially empty archive population

𝐺>𝐺𝑚𝑎𝑥 Save Results

𝐶𝑡 = 𝑃𝑡 ∪ 𝐴𝑡 ; 𝑀𝑝 = ∅; // combine main and archive population

if ԡ𝐴𝑡 ԡ < 𝑁

Mating Selection 𝑀𝑝
𝑀𝑝 = binary_tournament_selection(𝑃𝑡 )
else
while 𝑀𝑝 < 𝑁
Randomly select 𝑥1 , 𝑥2 from 𝑃𝑡 and 𝑎1 , 𝑎2 from 𝐴𝑡
if 𝐶𝑉(𝑥1 ) < 𝐶𝑉(𝑎1 ) if 𝐴𝐷(𝑥2 ) > 𝐴𝐷(𝑎2 )
𝑀𝑝 = 𝑀𝑝 𝑈𝑥1 𝑀𝑝 = 𝑀𝑝 𝑈𝑥2
else else
𝑀𝑝 = 𝑀𝑝 𝑈𝑎1 𝑀𝑝 = 𝑀𝑝 𝑈𝑎2

Create Offspring 𝑄𝑡
Variation Operator
Decompose 𝑀𝑝 into binary 𝑀𝑝𝑏 and continuous variables 𝑀𝑝𝑐
Apply uniform crossover and bitflip mutation on 𝑀𝑝𝑏 to create binary Offspring 𝑄𝑏
After that Apply SBX and polynomial mutation to create continuous Offspring 𝑄𝑐
𝑄 = 𝑄𝑏 𝑈𝑄𝑐 , %Combine binary and continuous Offspring
𝑄𝑡 = 𝐸𝑣𝑎𝑙𝑢𝑎𝑡𝑒(𝑄); % Compute objective functions and 𝐶𝑉 of 𝑄

Update Main Population: 𝑃𝑡+1

𝐶𝑡 = 𝑃𝑡 ∪ 𝑄𝑡 ; // combine main and Offspring population
Update Archive Population: 𝐴𝑡+1 𝑆1 = {𝐶𝑖 ∈ 𝐶|𝐶𝑉(𝐶𝑖 ) = 0}; % feasible solutions of 𝐶𝑡
𝐶𝑡 = 𝑃𝑡 𝑈𝑄𝑡 𝑈𝐴𝑡 ; %Combine Population 𝑆2 = {𝐶𝑖 ∈ 𝐶|𝐶𝑉(𝐶𝑖 ) > 0}; % infeasible solutions of 𝐶𝑡
Add CV as m+1 objective function if ԡ𝑆1 ԡ > 𝑁
𝒱𝑡 = 𝐸𝑁𝑆(𝐶𝑡 ), % Implement ND sort ℱ = 𝐸𝑁𝑆(𝑆1 ); % Implement ND sort to find 𝑘 fronts
while ԡ𝒱𝑡 ԡ > 𝑁 𝑃𝑡+1 = ∅; 𝑖 = 1;
Select two solutions (𝑐𝑖 and 𝑐𝑗 ) w.r.t 𝐴𝐷 ቀ𝜃𝑐𝑖 ,𝑐𝑗 ቁ While ԡ𝑃𝑡+1 ԡ + ԡℱ𝑖 ԡ ≤ 𝑁
if 𝐶𝑉(𝑐𝑗 ) < 𝐶𝑉(𝑐𝑖 ) then 𝑃𝑡+1 = 𝑃𝑡+1 ∪ ℱ𝑖 ;
𝑖 = 𝑖 + 1;
𝒱𝑡 = 𝒱𝑡 \𝑐𝑖 ; % Delete 𝑐𝑖
else while ԡ𝑃𝑡+1 ԡ + ԡℱ𝑖 ԡ > 𝑁 do
𝒱𝑡 = 𝒱𝑡 \𝑐𝑗 ; % Delete 𝑐𝑗 Compute CD of ℱ𝑖 and delete a least one
𝑃𝑡+1 = 𝑃𝑡+1 ∪ ℱ𝑖 ;
𝐴 =𝒱
else
𝒫𝑡+1 = 𝑆1 ;
Sort 𝑆2 w.r.t CV and add first (𝑁 − ԡ𝑆1 ԡ) into 𝒫𝑡+1
FIGURE 1. Flow chart of proposed hybrid algorithm.

objective functions. After that check the terminating

As shown in flow chart, first load the data of proposed test
condition if satisfied terminate the algorithm and save the
network, parameters of UC problem and parameters of
result. If the terminating condition is not satisfied, in the
proposed algorithm and then evaluate the randomly
second step hybrid variation operators, such as mutation
generated initial population to evaluate and compute the
and crossover, is applied on the population that is in

VOLUME XX, 2017 1

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3351710

mating selection, binary tournament selection, to find the hand if the size of 𝑆1 > 𝑁 apply fast non-dominated
new population called Offspring (𝑄𝑡 ). sorting operator discussed in [46] on 𝑆1 to compute the
To enhance the convergence and diversity of the PF, it's rank of PF say ℱ1 , … , ℱ𝑘 (ℱ1 is the highest rank, ℱ2 is next
beneficial to encourage interaction and cooperation highest rank and so on) and assign the highest rank PF to
between main 𝑃𝑡 and archive 𝐴𝑡 population. The main 𝑃𝑡+1 than 2nd highest and process is continuous until size
population, which operates within the feasible search of 𝑃𝑡+1 is equal to N or greater than N. If the size of 𝑃𝑡+1
space, and the archive population, which explores is greater than N, than some of the solutions in the last
promising infeasible solutions. Parents for the mating pool front are eliminated using crowding distance (CD)
are selected using Binary tournament selection. If the size operator [46].
of the archive population ԡ𝐴𝑡 ԡ is smaller than the total Finally, update archive population 𝐴𝑡+1 which is
population size (𝑁), parents are chosen from the combined responsible to generate representative non-dominated
population of the 𝑃𝑡 and the 𝐴𝑡 . However, if the archive infeasible solutions by adding CV as one extra objective
size is equal to or larger than N, parents are alternately function and apply unconstrained fast non-dominated
selected from the main and archive populations based on sorting 𝑀 + 1𝑡ℎ objective function. Now, copy all the
their CV as defined in Eq. (9) and their angle-based infeasible non-dominated solutions into the archive 𝐴𝑡+1 .
density (AD). First to form mating pool (tournament If the number of solutions exactly matches the archive size
selection), in which two solutions are randomly picked or is less, proceed to the next stage. If not, utilize a
form 𝑃𝑡 (say 𝑥1 ) and 𝐴𝑡 (say 𝑎1 ) and select the one with truncation operator to remove surplus infeasible solutions
smaller CV. After that another two solutions are randomly based on their AD and CV. In the truncation process, pick
selected from 𝑃𝑡 (say 𝑥1 ) and 𝐴𝑡 (say 𝑎1 ) are randomly two solutions based on the smallest angle between them,
selected, and compared to select the one with the higher and eliminate the one with the greater constraint violation.
AD. In this proposed algorithm, the AD is calculated by
𝑖
normalize the objective functions using ideal 𝑍𝑚𝑖𝑛 and IV. SIMULATION RESULTS
𝑖
nadir 𝑍𝑚𝑎𝑥 points in the 𝐶𝑡 according to; To examine the reliability and efficiency of the proposed
𝑖
𝑓𝑖 − 𝑍𝑚𝑖𝑛 method, the IEEE RTS-test system (eleven units) [48] and
𝑓𝑖′ (𝑥𝑗 ) = 𝑖 𝑖
, 𝑖 = 1,2, … , 𝑚 (24)
𝑍𝑚𝑎𝑥 − 𝑍𝑚𝑖𝑛 IEEE 30-bus test network (six units) [49] are adopted to
The normalized objective functions are shown as find the solution of optimal UC problem. The number of
𝐹𝑖′ (𝑣𝑗 ) = (𝑓1′ (𝑣𝑗 ), 𝑓2′ (𝑣𝑗 ), … , 𝑓𝑚′ (𝑣𝑗 ) ). After that vector populations is selected as 40 and the maximum function
angle between two solutions say 𝑥𝑗 and 𝑥𝑘 is computed as evaluation is taken as 100000 when there are no network
𝑭′ (𝑥𝑗 ) ∙ 𝑭′ (𝑥𝑘 ) flow constraints and 200000 for the network flow
𝜃𝑥′ 𝑗 ,𝑥𝑘 = arccos | | 𝑤ℎ𝑒𝑟𝑒 𝑥𝑘 ∈ 𝒫𝑡 ∩ 𝑥𝑘
‖𝑭′ (𝑥𝑗 )‖ԡ𝑭′ (𝑥𝑘 )ԡ (25) constraints. The program is run on the Corei7 intel with
≠ 𝑥𝑗 MATLAB version R2023a. Furthermore, to validate the
Next, the solutions are ranked based on the angle between efficiency and superiority of the proposed method three
them, where a larger angle corresponds to a higher rank study scenarios are formulated to find the optimal solution
for the solution, making it a promising candidate for of UC problem with and without network security
mating selection. The idea is to mate one solution with a constraints by selecting single and multi-objective
favorable CV and another with a favorable AD value. This functions on a solution of IEEE 30-bus network. In the
strategy is expected to produce offspring that are not only Scenario 1, traditional UC problem without network
in converge to the Pareto Front (PF) but also exhibit good constraints of 11 units are duplicated 𝑇𝑛 times to find the
diversity. solution of optimal UC problem considering week ahead
Once the mating parents are selected, variation operator is planning. In this scenario, decision variables for each hour
applied to find the new solutions called Offspring. In the are stacked to form a master problem, where intertemporal
variation operators first decompose the population into constraints such as ramp up, ramp down, minimum up and
binary and continuous decision variables. After that down time are easily satisfied to find the near global
uniform crossover and bitwise mutation has been applied optimal solutions. In Scenario 2, various economical and
to vary binary decision variables to form binary part of technical single objective functions such as cost of thermal
Offspring and the well-known Simulated Binary generators, cost of energy loss and voltage deviation are
Crossover (SBX) and polynomial mutation techniques are considered to find the optimal solution of UC problem,
applied to generate the continuous offspring. After that there is also state-of-the-art single objective evolutionary
binary and continuous decision variables are combined algorithms are implemented and compared with the
and evaluate the Offspring population. In the next step, proposed hybrid evolutionary algorithm. In Scenario 3,
main population and archive populations are updated. various technical and economical multi-objective
Updated main population (𝐴𝑡+1 ) is obtained by combining objective functions, comprised of cost energy supplied
the previous main population (𝐴𝑡 ) and recently generated and startup cost, VD and CEL are simultaneously
offspring population (𝑄𝑡 ) to form combined population 𝐶𝑡 optimized and find the tradeoff between technical and
and extract the set of feasible 𝑆1 (where 𝐶𝑉 ≤ 0) and economical objective functions to compute the solution of
infeasible 𝑆2 (where 𝐶𝑉 > 0) solutions from 𝐶𝑡 . If the size security constrained UC problem.
of 𝑆1 < 𝑁 then sort the infeasible solutions 𝑆2 and select
the first 𝑁 − 𝑆1 solutions from sorted 𝑆2 . On the other

2 VOLUME XX, 2017

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3351710

SCENARIO 1: SIMULATION RESULTS OF SINGLE min up and down time constraints. During the sixth day,
OBJECTIVE WITHOUT NETWORK SECURITY when the load is maximum, it satisfies the reserve
CONSTRAINTS constraint. After examining the results of the sixth day, it
Table I shows the max, and min power capacity of each can be easily predicted that the 𝐺1 to 𝐺5 are highly
generator and their quadratic cost and constraint efficient generators and committed during base load
parameters. Simulation results of selected 24 hours of a conditions.
day is shown in Table II, whereas scheduled load demand The marginal cost of these generators is less compared to
of a week-ahead (168 hours) of a seven days and power 𝐺6 to 𝐺11 , hence committed during the entire proposed
produced by each committed generators with cumulative time horizon. Generators 𝐺6 to 𝐺11 are frequently
cost of active power generation are given in Fig. 2. decommitted and recommitted according to load profile
For better visualization, Table II shows the simulation and their cost curve. Further, Fig. 2 shows the scheduling
results of only day 6 (starting from 121 to 144 hours), of generators and cost curve for the entire 168 hours. Fig.
where the peak load has appeared analyzed. Table II 2(a) shows the variation load curve in week, Fig. 2(b)
shows that the proposed method can find a feasible clearly shows that the proposed algorithm gives optimal
solution, whereas all the solutions are feasible. It can be UC solutions subject to satisfying the UC constraints in all
noticed from Table II that all 11 generating units were seen the 168 time slots. The proposed method finds a near
to be committed from time 128hrs to 138hrs (at peak load globally optimal solution to the UC problem without
hours), and a majority of the units were operating at their violating any constraints in week ahead planning. To show
maximum power capacity in that time frame. Simulation the validation and performance of a proposed method, Fig.
results gives minimum cost subject to satisfying ramp rate, 2(b) on and off states of generators follow the load curve.
TABLE I. Generator data of RTS system for week ahead optimal UC.
Gen. # Pmin Pmax A B C SU RU RD MUT MDT
1 100 800 5 4 0.0010 415.0 5 5
2 100 800 5 6 0.0020 625.0 5 5
3 80 400 20 8 0.0025 676.0 1 3
4 80 400 20 10 0.0025 836.0 5 5
5 60 300 30 10 0.0020 637.2 4 1
6 60 300 30 12 0.0020 757.2 15 15 1 3
7 50 200 40 14 0.0015 743.7 2 5
8 50 200 40 16 0.0015 843.7 3 4
9 25 100 55 15 0.0012 430.7 5 5
10 25 100 55 17 0.0012 480.7 5 4
11 25 100 55 17 0.0012 480.7 1 1

TABLE II. UC values and ratings of scenario 1 (from 121 to 144 hours).
Time Generators Scheduled Power 𝟏𝟏 𝒇𝟏
(hours) 𝑮𝟏 𝑮𝟐 𝑮𝟑 𝑮𝟒 𝑮𝟓 𝑮𝟔 𝑮𝟕 𝑮𝟖 𝑮𝟗 𝑮𝟏𝟎 𝑮𝟏𝟏 ∑ 𝑷𝑮 𝒊 ($/h)
𝒊=𝟏
121 796.5 622.5 252.2 116.1 122.9 0 54.8 0 0 0 0 1964.9 14543.3
122 778.4 652.6 240.8 81.3 102.2 0 53.5 0 0 0 0 1908.7 13796.3
123 794.0 626.8 204.6 110.4 116.6 0 0 0 0 0 0 1852.6 12669.1
124 797.8 607.1 185.0 132.0 102.7 0 0 0 0 0 0 1824.5 12229.1
125 781.6 643.0 216.0 112.7 99.2 0 0 0 0 0 0 1852.5 12572.7
126 799.9 722.9 255.9 152.4 129.2 60 0 50 25 25 25 2245.5 20714.1
127 799.9 788.9 284.0 192.4 121.1 63.8 0 54.6 27.4 25.6 0 2357.8 18532.2
128 800 800 324.0 232.4 151.1 93.8 50 74.6 37.4 35.6 25 2646.6 22511.1
129 800.0 799.7 364.0 272.0 180.6 123.8 63.2 94.6 40.1 36.3 32.7 2806.9 24046.9
130 799.2 799.9 398.2 310.5 206.5 153.8 80.1 109.7 48.6 36.0 35.9 2975.4 26274.9
131 799.1 792.2 387.7 341.0 214.4 181.7 73.3 126.1 45.9 40.5 31.2 3033.4 27045.3
132 795.2 789.5 385.6 378.9 218.7 204.0 76.7 131.4 36.6 37.8 36.7 3089.3 27692.0
133 791.9 797.4 379.0 395.5 218.9 210.9 78.8 142.7 39.9 43.7 45.8 3147.9 28535.1
134 797.6 781.7 397.1 399.1 230.6 240.8 95.1 135.1 30.1 40.4 53.7 3201.6 29121.7
135 788.2 791.6 397.1 399.3 227.1 241.4 84.0 117.5 39.1 31.1 57.3 3173.6 28740.9
136 760.5 787.7 366.6 391.0 250.8 214.9 70.9 129.0 29.7 29.6 62.8 3090.5 27778.1
137 742.3 724.6 331.0 356.9 261.6 217.8 67.4 130.8 39.6 37.2 52.9 2973.2 26335.4
138 696.9 681.7 291.0 331.5 250.9 203.1 54.1 110.8 29.6 27.2 42.9 2720.3 23414.5
139 775.9 624.9 306.8 291.5 221.8 173.1 0 90.8 25.3 0 33.0 2542.9 21062.6
140 758.8 659.7 284.4 255.6 198.2 143.1 0 71.2 0 0 0 2370.9 18854.2
141 799.9 590.7 251.3 215.6 176.2 118.9 0 51.2 0 0 0 2203.8 16869.3
142 719.9 510.7 211.3 175.6 146.2 88.9 0 0 0 0 0 1833.2 13039.0
143 717.2 460.2 190.8 135.6 144.9 0 0 0 0 0 0 1648.6 11407.4
144 637.2 380.2 150.8 95.6 114.9 0 0 0 0 0 0 1378.6 9272.4

2 VOLUME XX, 2017

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3351710

FIGURE 2. Generators and demand of each hour in IEEE RTS system.

It is depicted in Fig. 2(b), that the maximum share of load and real coded GA-based methodology gives the near-global
supplied by generators is 1 to 6 (that means these generators optimal solution to the UC problem, and it is easy to
are economical to supply load demand without operational implement on any size of the real problem with less
constraints). Also shows that the optimal UC binary decision computational complexity.
variables and majority of units i.e. 𝐺1 to 𝐺6 , high-efficiency SCENARIO 2: SIMULATION RESULTS OF SINGLE
generators, were operating at their maximum power capacity OBJECTIVE CONSIDERING NETWORK SECURITY
in that time frame. It can be also noticed from Fig. 2(b) that CONSTRAINTS
all 11 generating units were seen to be committed from time
In this subsection, proposed UC formulation is applied on
128hrs to 138hrs, and a majority of the units were operating
IEEE 30-bus test system to find the solution of single
at their maximum power capacity in that time frame. Since
objective security constrained UC (SCUC) problem. In this
the min-up and min-down times of unit 4 are 5hrs each, it
section decision variables of SCUC problem is obtained by
stays de-committed during the 48th – 53rd hr and 145th – 149th
selecting three objective functions such as total operating
hr of 6hrs and 5hrs respectively. This is because the load
cost, cost of energy loss and voltage deviation. Table III
decreases and makes units 2 and unit 3 run on their minimum
shows the generator data of IEEE 30-bus test system,
power ratings. The power rating of unit 4 gradually increases
whereas network data is taken form [45]. Fig. 3 Shows the
at the instances of 55th, 79th, 102nd, and 126th hr where the
comparison between the state-of-the-art EAs and the
load increased, and it also forced maximum generators to go
proposed algorithm in terms of convergence plots of various
online. As compared to units 3, 5, 6, 7, 8, 10, and 11, the 9th
technical and economical objective functions applied to the
Unit is committed first in starting hours as it shares the
IEEE 30-bus 6-unit system.
economic power since it does not approach its maximum
The data reported in Fig. 3 are the minimum operation cost,
power during the whole course. Unit 11 remained de-
cost of energy loss and VD values in the convergence curve
committed most of the time due to its lowest min-up and
at each generation during the search. Convergence curve
min-down time i.e., 1hr. From Fig. 2(c), we can see that the
shows that the proposed algorithm and SHADE [50] finds
proposed algorithm gives minimum cost subject to satisfying
faster convergence compared to all other EAs. After 200
constraints. It also compares the overall cost of Lagrangian
generation in most of the cases proposed algorithm finds near
Relaxation (LR) and the proposed method, in which LR
global optimal solution compared to all the other EAs. In all
performs marginally better than the proposed algorithm in
the cases IMODE [51] is stuck in the local optimal solution.
some time slots 1. Whereas, in overall time slots proposed
From the comparative convergence curve, it is proved that
method beats the LR method. The proposed method gives a
search in both of the direction such as from feasible and
feasible solution compared to the LR method. In most of the
infeasible side gives faster rate of convergence to optimal
time slots, the LR method gives infeasible solutions.
solutions, but with slightly more computational complexity.
Moreover, multiple violations were observed by the LR
In the proposed algorithm high quality infeasible solutions
method due to the complexity of operational constraints. The
are obtained by converting single objective into bi-objective
above discussion exhibits that the proposed hybrid binary
(second objective function is the overall CV).

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3351710

FIGURE 3. Convergence plots of all the cases of competing algorithms.

After that, nondomination rule has been applied until all the 785.4 $/h respectively compared to all the other algorithms.
population members are jump into feasible region. However, minimum value of startup cost and VD is as shown
Computational cost of proposed algorithm is higher than in SHADE [50] that are 1473.4 $/h and 9.847 p.u, from the
other algorithms because in the proposed algorithm extra simulation results of case 1, it is shown that proposed
computation is required to find the representative infeasible algorithm find the minimum value of objective function at
solutions. These representative infeasible solutions help to the cost of increasing other non-optimizing functions. Also,
increase the convergence of proposed algorithm to find the in case 2, proposed algorithm competes all the other
near global optimal solution. Furthermore, Table IV shows algorithm to minimize the Cost of energy loss that is 719.5
the comparison of solution of SCUC problem considering $/h after that IMODE [51] find better value of objective
various technical and economical objective functions function. In case 3 proposed algorithm find better value
adopting recent evolutionary algorithms. Table IV describe compared to all the algorithms in all the proposed objective
that the proposed algorithm finds minimum cost of active functions. For the better visualization and analysis of
power generation (objective function in Case 1) and cost of decision vector of the proposed algorithm it is desirable to
energy loss of thermal generators that are19001.9 $/h and show the decision vector in Figure form.
TABLE III. Generator data of IEEE 30-bus system for week ahead optimal UC.
Gen. # Pmin Pmax A B C SU RU/RD
1 200 50 0 2 0.0200 250 80
2 80 20 0 1.75 0.0175 142 32
3 80 15 0 1 0.0625 129.062 32
4 50 10 0 3.25 0.0083 133.334 20
5 50 10 0 3 0.0250 132.5 20
6 40 12 0 3 0.0250 139.6 16

TABLE IV. Simulation results of SCUC problem of proposed and the recently implemented EAS
Case Algorithm Cost [$/h] CEL [$/h] VD [p.u] Startup Cost [$/h]
GA [52] 19515.9 848.4 15.798 1902.4
CSO [53] 19384.5 802.3 15.283 1885.8
Case 1 IMODE [51] 19095.3 794.7 16.844 1762.8
SHADE [50] 19538.7 912.8 9.847 1473.4
Proposed 19001.9 785.4 14.967 1762.8
GA [52] 19961.8 776.3 15.567 2044.4
CSO [53] 19768.6 742.9 15.177 1885.8
Case 2 IMODE [51] 19501.4 728.2 16.993 1904.8
SHADE [50] 19642.7 897.2 10.028 1473.4
Proposed 19399.6 719.5 14.785 1904.8
GA [52] 21188.1 1196.0 8.110 2171.5
CSO [53] 18279.5 799.6 3.900 1497.8
Case 3 IMODE [51] 18024.7 783.3 3.361 1495.0
SHADE [50] 21505.6 1223.8 3.298 1619.9
Proposed 17906.3 685.6 2.720 1625.9

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3351710

FIGURE 4. Heatmap of schedule power generation 𝑷𝒈 and binary SCUC variables.

Fig. 4 shows the sky-blue colored heat map of schedule decision vectors in various study cases. Fig. 5(a1) shows the
power of Pg, u, v and w decision variables. In this Fig. 4, dark box plot of transformer turns ratio of final optimal solution.
sky-blue color shows that output power is maximum Box chart of transformer tap ratio of all the cases is with in
whereas, light sky-blue shows minimum power produced by specified limit, whereas mean of case 1 and 2 is similar and
those generators and “0” value such as white color shows that dissimilar solution produced by case 3. Fig. 5(a2) shows the
generators is shut down. Fig. 4 also depicts that in entire time box plot of optimal MVAr injection of all the 9 SVCs of final
horizon proposed algorithm finds such solution that satisfy optimal solution. Box chart of SVC injections of all the cases
all the intertemporal constraints such ramp up and ramp is with in specified limit, whereas mean of case 1 and 2 is
down constraints and minimum up and down time similar and higher mean can show in case 3. Fig. 5 (a3) shows
constraints. Furthermore, Fig. 4 of heat map of schedule the box plot of optimal MVAr injection of all the committed
power generation clearly shows that during peak hours units of final optimal solution. Box chart of 𝑄𝑔 injection of
generators four and five are operated economically. Zero in all the cases is with in specified limit, whereas mean of case
the Fig. 4 shows that generators are shut down or 1 and 2 is similar and higher mean can show in case 3. Fig.
decommitted during a given time slot. Schedule output 5(b) shows the voltage profile of all the buses in the entire
power of generators in case 2 and 3 are similar except at time horizon of all the study cases. Circles outside the box
some locations of peak hour period. Furthermore, plot of Fig. 5(b) shows the outlier which are generator bus
comparison of other decision variables such as transformer voltage. Voltage at all the buses is with in specified limit.
tap settings, MVAr injection of static Var compensators Over the entire time horizon, the voltage profile of case 3 is
(SVC), reactive power generated and startup cost of all the ideal (near unity) compared to case 1 and 2. Considering
committed units of all the study cases in the entire time hour technical objective function such as VD emphasize the
are shown in Fig. 5. Usually, box chart shown in Fig. 5(a) decision variable near unity.
and (b) gives the statistical information of entire time horizon
in a single box, this gives minimum, maximum, median and
quartile values.
However, plot of each transformer of entire time period is
difficult to visualize hence box chart of most of the decision
variables are shown to justify the comparative values of

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3351710

FIGURE 5. Single objective decision variables (a1) transformer off nominal turns ratio (a2) Shunt MVAr
injection (a3) reactive power of generator (b) voltage profile of all the buses and (c) startup cost of entire
time horizon.
SCENARIO 3: SIMULATION RESULTS OF
Fig. 5(c) shows the startup cost of all the units and MULTIOB5WJECTIVE CONSIDERING NETWORK
cumulative startup cost obtained by proposed algorithm in SECURITY CONSTRAINTS
case 1, 2 and 3 is 1762.8, 1904.8, and 1625.9 respectively.
In this section, recently available MOEAs are implemented
Over all Simulation results of all the study cases shows that
to solve multi-objective SCUC problem and the results of all
economical objective functions such as operation cost
the implemented MOEAs are compared with the proposed
(comprised of cost of active power generation and startup
hybrid algorithm. Operation cost of active power generation,
cost) and cost of energy loss gives better objective functions
cost of energy loss and VD are the objective functions to find
whereas, minimum startup cost of generator and value of
the solution of SCUC problem along with the consideration
voltage profile variables are obtained ideal by minimization
of AC network constraints. Fig. 6(a1 to a3) shows the
of VD as objective function. In the literature mostly cost
comparison of Pareto Fronts (PFs) of all the study cases of
functions are considered to obtain the commitment and
final nondominated solutions of all the study cases of entire
schedule of thermal generators, whereas in this paper it is
time horizon. In the PF shown in Fig (a1 to a3), most recently
also proved the VD is also find the better results of decision
MOEAs are implemented to solve proposed multi-objective
vector compared to economical objective functions.
SCUC problem. These MOEAs includes NSGAII [46],
Therefore, in the next subsection multi-objective SCUC
ANSGAIII [54], AGEMOEAII [55], CCMO [56].
problem is solved by considering the trade-off between
Final PF as shown in Fig. 6(a1 to a3), clearly shows that
technical and economical objective functions of various two
proposed algorithm finds the better trade-off between bi and
and objective functions.
tri objective functions in terms of both convergence and
diversity. Fig. 6(a1 and a2) clearly shows that in case 1 and 2
proposed algorithm outperforms the other state of art
MOEAs. Whereas, in case 3 it difficult to judge the

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3351710

performance of proposed algorithm with the other state of membership function is computed, it is then normalized to
the MOEAs. In the literature most widely used metric obtain the normalized membership function μk .
𝑚
function is HVI that is used to find the performance of
𝑓 (𝑥) = ∑ 𝑓̃𝑖 (𝑥) (27)
MOEAs from the first to the last iteration. In this matric only 𝑖=1
the worst and best solutions are required from the results of
all the algorithm and select the reference point in the The value of 𝑁𝑑 represents the number of solutions in the
objective functions space such as (1,1,1)𝑚 , whereas m final PF (Pareto Front). The BCS can be determined by
shows the number of objective functions. Maximum value of finding the index with the highest 𝜇𝑘 value. Table 4 displays
HVI metric of a MOEA gives the information of better the BCS results of all algorithms of all the cases that utilized
convergence and diversity compared to other MOEAs of fuzzy decision-making rules. For better visibility, diversity
complicated PF. In the proposed study cases, HVI helps to and convergence of PF of proposed algorithm of case 2 and
compare performance of various MOEAs with the proposed 3 are as shown in Fig. 7. Also, Fig. 7 clearly shows that in
algorithm, especially in case 3 when the performance of the complicated study Cases 2 and 3, proposed algorithm
MOEAs is not judged to see the PF only. Fig. 6(b1 to b3) outperforms compared to other recently implemented
shows the HVI of all the implemented MOEAs. HVI in Fig. MOEAs. Fig. 7 also shows that proposed algorithm finds the
6 (b1 to b3) shows that after each 100 generation in case 1 and highly distributed nondominated PF in complex Case 2 and
2 and after each 150 generation in case 3 after this iteration 3. Table V gives Simulation results of all the study cases of
proposed algorithm find the feasible solutions. HVI of all the proposed algorithm and other recently implemented
study cases clearly shows that proposed algorithm has better MOEAs.
convergence and diversity compared to NSGAII [46], Table V describe that the proposed algorithm finds minimum
ANSGAIII [54], AGEMOEAII [55], CCMO [56]. The cost of active power generation (objective functions in Case
diamond shape in PF shows the best compromise solution 1) and cost of energy loss of thermal generators that are
(BCS) and, in this study its computed using fuzzy weight 19001.9 $/h and 719.5 $/h respectively compared to all the
functions. The fuzzy decision approach, as described in other algorithms. Maximum values of objective functions in
reference [57], involves first normalize the objective functions Case 1 are 19961.8 and 912.8 $/h, which are maximum
space called membership function (𝜇𝑚 𝑘)
as; compared to all the other MOEAs. Minimum and maximum
values shows that the proposed algorithm find the widely
1 𝑓𝑜𝑟 𝑓𝑚𝑘 ≤ 𝑓𝑚𝑚𝑖𝑛 distributed PF. In all the cases feasibility ratio (FR) of most
𝑘 𝑓𝑚𝑚𝑎𝑥 − 𝑓𝑚𝑘 of the algorithms is 100%, except CCMO [56] that some of
𝜇𝑚 = 𝑓𝑜𝑟 𝑓𝑚𝑚𝑖𝑛 < 𝑓𝑚𝑘 < 𝑓𝑚𝑚𝑎𝑥 (26)
the population members are stuck in infeasible region. HVI
𝑓𝑚𝑚𝑎𝑥 − 𝑓𝑚𝑚𝑖𝑛
{0 𝑓𝑜𝑟 𝑓𝑚𝑘 ≥ 𝑓𝑚𝑚𝑎𝑥 values of proposed algorithm are shown maximum compared
𝑘 to all the other algorithms.
The calculation of the membership function 𝜇𝑚 involves the
In case 2, proposed algorithm competes all the other
use of parameters m and k, which represent the number of
objective functions and population size respectively. Once the algorithm and values of objective functions 𝑓1 , 𝑓2 and 𝑓3 are
17088.4 $/h, 589.9 $/h and 2.777 p.u respectively.

FIGURE 6. Simulation results of multi-objective SCUC Problem (a) Final nondominated PF b) HVI of all the
study cases of entire time horizon of all the study cases.

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3351710

FIGURE 7. Final Nondominated solutions and BCS of Case 2 and 3 of proposed algorithm.

Table V. The simulation results of proposed and the recently implemented EAs available in the literature.
BCS_PF min (OF) max (OF) Startup
Method Cost CEL VD Cost CEL VD Cost CEL VD HVI FR Cost
NSGAII [46] 19639.7 802.3 15.910 19515.9 776.3 -- 19399.6 848.4 -- 0.144 100 1902.4
ANSGAIII [54] 19476.1 766.6 15.343 19384.5 742.9 -- 19768.6 802.3 -- 0.152 100 1885.8
AGEMOEAII [55] 19193.0 755.6 16.990 19095.3 728.2 -- 19501.4 794.7 -- 0.160 100 1762.8
CCMO [56] 19603.4 898.5 9.966 19538.7 897.2 -- 19642.7 785.4 -- 0.150 90 1473.4
Proposed 19103.5 745.1 15.002 19001.9 719.5 -- 19961.8 912.8 -- 0.164 100 1762.8
NSGAII [46] 21188.1 1196.0 8.110 21158.7 -- 8.1 21315.1 -- 9.3 0.110 93 2171.5
ANSGAIII [54] 18267.4 785.6 4.305 18265.2 -- 3.9 18279.5 -- 5.7 0.223 100 1497.8
AGEMOEAII [55] 18021.6 779.9 3.491 18021.5 -- 3.4 18024.7 -- 3.5 0.239 100 1495.0
CCMO [56] 21391.8 1202.2 3.371 21352.0 -- 3.3 21505.6 -- 6.4 0.132 100 1619.9
Proposed 17088.4 589.9 2.777 17086.1 -- 2.7 17906.3 -- 3.4 0.280 100 1353.8
NSGAII [46] 21179.8 1121.4 16.290 21155.3 1118.7 16.3 21212.3 1151.5 17.7 0.089 78 1631.4
ANSGAIII [54] 18856.8 828.6 3.265 18802.1 814.0 3.2 19019.7 844.7 3.5 0.082 100 1759.8
AGEMOEAII [55] 18975.2 772.8 2.860 18892.8 737.4 2.7 19305.9 807.0 3.9 0.092 100 1762.8
CCMO [56] 21149.9 1011.2 4.466 21083.8 851.8 4.4 21547.7 1011.2 10.8 0.039 100 1474.4
Proposed 17827.8 637.9 5.286 17679.3 584.7 4.9 18255.4 675.5 9.4 0.120 100 1212.6

However, in Case 3, proposed algorithm finds the minimum satisfy all the intertemporal constraints that includes ramp up
values of 𝑓1 and 𝑓2 that are17827.8 $/h and 637.9 $/h and down (ramp rate) constraints and min up and down time
respectively, whereas, minimum value of 𝑓3 is obtained by constraints. Fig. 8 of heat map of schedule power generation
AGEMOEAII [55] that is 2.860 p.u. Cumulative startup cost clearly shows that during peak hours generators four and five
in Case 2 and 3 is recorded by proposed algorithm that are are operated economically. Zero in the Fig. 8 shows that
1353.8 and 1212.6 $/h, where as in Case 1 minimum generators are shutdown or de-committed during a given
cumulative startup cost is figured by CCMO [56] that is time slot. Schedule output power of generators in case 2 and
1473.4 $/h. 3 are similar except at some locations of peak hour period.
For the better visualization and analysis of scheduling Furthermore, comparison of other decision variables such as
decision vector of proposed algorithm it is desirable to show transformer tap settings, MVAr injection of static Var
the decision vector in Figure form. Fig. 8 shows the sky-blue compensators (SVC), reactive power generated and startup
colored heat map of schedule power Pg, and binary decision cost of all the committed units of all the study cases in the
variables u, v and w of entire time horizon. In the Fig. 8, dark entire time hour are shown in Fig. 9. In Fig. 9(a) and (b), a
sky-blue color shows that output power is maximum box chart is typically employed to display statistical
whereas, light sky-blue shows minimum power produced by information for the entire time horizon in a single box. This
that generator and “0” value such as white color shows that box chart provides key statistical measures, including
generators is shut down. Fig. 8 also depicts that in the entire minimum, maximum, median, and quartile values of
time horizon proposed algorithm finds such solution that decision variables in the entire time horizon.

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3351710

FIGURE 8. Heatmap of schedule power generation 𝑷𝒈 and binary SCUC variables (u, v and w) of all the
cases.

However, visualizing the line plots of each decision variable profile of case 3 is ideal, approaching unity, unlike cases 1
over the entire time period can be challenging. Therefore, and 2.
box charts for most decision variables are presented to This highlights that considering technical objective functions
facilitate a comparison of decision vectors in various study such as Voltage Deviation (VD) places an emphasis on
cases. In Fig. 9(a1), you can observe the box plot representing decision variables approaching unity. In Fig. 9(c), the startup
the transformer turns ratio of the final optimal solution. The costs of all units over the entire time periods are presented.
box charts for transformer tap ratios in all cases fall within The cumulative startup costs obtained by the proposed
specified limits. The means of Cases 1 and 2 are similar, algorithm in Cases 1, 2, and 3 are 1762.8 $/h, 1353.8 $/h, and
while case 3 produces a dissimilar solution with a different 1212.6 p.u, respectively. The simulation results across all
mean. Moving to Fig. 9(a2), we have found the box plot study cases reveal that economic objective functions, such as
illustrating the optimal MVAr injection for all 9 SVCs in the operational cost (comprising cost of active power generation
final optimal solution. The box charts for SVC injections in and startup cost) and cost of energy loss, yield superior
all cases stay within the specified limits. In Fig. 9(a3), the box objective function values. Notably, the minimum startup cost
plot shows the optimal MVAr injection of all committed of the generator and the values of voltage profile variables
units in the final optimal solution. Similar to the previous are ideal in case 3, where all three objective functions are
cases, the box chart for 𝑄𝑔 injection in all cases complies minimized concurrently. It's worth mentioning that this
with the specified limits. The means of cases 1 and 2 are paper demonstrates that Voltage Deviation (VD) can yield
similar, while case 3 features a higher mean. Fig. 9(b) depicts better results of SCUC problem in decision vectors
the voltage profiles of all buses throughout the entire time compared to economic objective functions.
horizon for all study cases. Circles outside the box plots
indicate outliers, which correspond to the voltage of
generator buses. Voltage levels at all buses remain within the
specified limits. Over the entire time horizon, the voltage

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3351710

FIGURE 9. Multi-objective objective decision variables of SCUC (a1) transformer off nominal turns ratio (a2)
Shunt MVAr injection (a3) reactive power of generator (b) voltage profile of all the buses and (c) startup
cost of entire time horizon.
the cost of energy supplied, startup and shutdown costs of
V. CONCLUSION
generators, the cost of energy loss, and voltage deviation.
This paper, embarked on a significant journey to address the This versatile formulation has proven its effectiveness in
intricate challenge of Unit Commitment (UC), specifically resolving both single and multi-objective SCUC problems,
focusing on Security Constrained UC (SCUC). The SCUC placing emphasis on a harmonious interaction between
problem is central to the optimization of power generation technical and economical objective functions. Through
unit scheduling within the bounds of numerous operational rigorous testing, our proposed algorithm has exhibited
constraints. In recent decades, there has been a increasing exceptional performance across various scenarios,
interest in the application of evolutionary algorithms (EAs) employing the 11-units IEEE RTS system and the 6-unit
to tackle large-scale multi-objective MINLP challenges. This IEEE 30-bus test system, both with and without security
pioneering approach integrates single and multi-objective constraints. We have successfully demonstrated the
EAs to navigate the complexities of SCUC, with a unique algorithm's ability to discover solutions that approach global
emphasis on incorporating intricate AC network constraints optimality, surpassing other contemporary EAs. The
through hybrid binary and real coded operators. The integration of our search operator with a multi-objective
development of an ensemble algorithm, uniting hybrid real coevolutionary algorithm, operating seamlessly with both
and binary coded operators and extended by a bidirectional feasible and infeasible solutions, has shown outstanding
coevolutionary algorithm, fortifies the methodology to performance in addressing multi-objective SCUC problems.
handle multi-objective SCUC problems effectively. Our Simulation results have been rigorously compared to various
study introduces a novel formulation, leveraging three recently implemented Multi-Objective Evolutionary
conflicting objective functions to address the SCUC Algorithms (MOEAs), clearly demonstrating the superiority
problem. These objectives encompass the minimization of of our proposed algorithm in terms of convergence and

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3351710

diversity. Our research also introduces an innovative and Generators," International Review of Electrical Engineering (IREE),
vol. 17, no. 2, pp. 105-117, 2022.
highly effective approach for tackling the SCUC problem [14] T. Jiang et al., "Exploiting flexibility of combined-cycle gas turbines
within the confines of AC network constraints. The in power system unit commitment with natural gas transmission
outcomes presented in this paper illustrate impressive constraints and reserve scheduling," International Journal of
achievements in terms of optimality and performance, Electrical Power & Energy Systems, vol. 125, p. 106460,
2021/02/01/ 2021, doi: https://doi.org/10.1016/j.ijepes.2020.106460.
particularly in the context of multi-objective optimization.
[15] A. J. Wood and B. F. Wollenberg, Power generation, operation, and
This work sets the stage for future advancements in the field control. John Wiley & Sons, 2012.
of power system optimization, with the potential to [16] A. Bhadoria, S. Marwaha, and V. K. Kamboj, "An optimum forceful
revolutionize the way we address UC challenges. generation scheduling and unit commitment of thermal power system
using sine cosine algorithm," Neural Computing and Applications,
vol. 32, no. 7, pp. 2785-2814, 2020/04/01 2020, doi:
ACKNOWLEDGMENT 10.1007/s00521-019-04598-8.
The authors extend their appreciation to the Deanship of [17] A. Grey and A. Sekar, "Unified solution of security-constrained unit
Scientific Research at Northern Border University, Arar, commitment problem using a linear programming methodology,"
Saudi Arabia, for funding this research work through the IET Generation, Transmission &amp; Distribution, vol. 2, no. 6, pp.
856-867. [Online]. Available: https://digital-
project number ‘‘NBU-FFR-2023’’. library.theiet.org/content/journals/10.1049/iet-gtd_20070367
[18] M. Nozarian, H. Seifi, M. Sheikh-El-Eslami, and H. Delkhosh,
REFERENCES "Hydro Thermal Unit Commitment involving Demand Response
[1] N. Yang et al., "A Comprehensive Review of Security-constrained resources: a MILP formulation," Electrical Engineering, pp. 1-18,
Unit Commitment," Journal of Modern Power Systems and Clean 10/05 2022, doi: 10.1007/s00202-022-01651-z.
Energy, vol. 10, no. 3, pp. 562-576, 2022, doi: [19] Z. Shao, Q. Zhai, Z. Han, and X. Guan, "A linear AC unit
10.35833/MPCE.2021.000255. commitment formulation: An application of data-driven linear power
[2] Q. P. Zheng, J. Wang, and A. L. Liu, "Stochastic Optimization for flow model," International Journal of Electrical Power & Energy
Unit Commitment—A Review," IEEE Transactions on Power Systems, vol. 145, p. 108673, 2023/02/01/ 2023, doi:
Systems, vol. 30, no. 4, pp. 1913-1924, 2015, doi: https://doi.org/10.1016/j.ijepes.2022.108673.
10.1109/TPWRS.2014.2355204. [20] M. E. Nazari and Z. M. Fard, "A novel heuristic optimisation
[3] N. P. Padhy, "Unit commitment-a bibliographical survey," IEEE algorithm for solving profit-based unit commitment for thermal
Transactions on Power Systems, vol. 19, no. 2, pp. 1196-1205, 2004, power generation with emission limitations," International Journal
doi: 10.1109/TPWRS.2003.821611. of Sustainable Energy, vol. 41, no. 6, pp. 675-698, 2022/07/03 2022,
[4] N. Muralikrishnan, L. Jebaraj, and C. C. A. Rajan, "A doi: 10.1080/14786451.2021.1971979.
Comprehensive Review on Evolutionary Optimization Techniques [21] Y. Yin, C. He, T. Liu, and L. Wu, "Risk-Averse Stochastic Midterm
Applied for Unit Commitment Problem," IEEE Access, vol. 8, pp. Scheduling of Thermal-Hydro-Wind System: A Network-
132980-133014, 2020, doi: 10.1109/ACCESS.2020.3010275. Constrained Clustered Unit Commitment Approach," IEEE
[5] H. Abdi, "Profit-based unit commitment problem: A review of Transactions on Sustainable Energy, vol. 13, no. 3, pp. 1293-1304,
models, methods, challenges, and future directions," Renewable and 2022, doi: 10.1109/TSTE.2022.3150918.
Sustainable Energy Reviews, vol. 138, p. 110504, 2021/03/01/ 2021, [22] J. H. Zheng, J. J. Chen, Q. H. Wu, and Z. X. Jing, "Reliability
doi: https://doi.org/10.1016/j.rser.2020.110504. constrained unit commitment with combined hydro and thermal
[6] L. Montero, A. Bello, and J. Reneses, "A Review on the Unit generation embedded using self-learning group search optimizer,"
Commitment Problem: Approaches, Techniques, and Resolution Energy, vol. 81, pp. 245-254, 2015/03/01/ 2015, doi:
Methods," Energies, vol. 15, no. 4, 2022, doi: 10.3390/en15041296. https://doi.org/10.1016/j.energy.2014.12.036.
[7] V. Krishnan et al., "Co-optimization of electricity transmission and [23] W. Hou and H. Wei, "Data-driven robust day-ahead unit
generation resources for planning and policy analysis: review of commitment model for hydro/thermal/wind/photovoltaic/nuclear
concepts and modeling approaches," Energy Systems, vol. 7, no. 2, power systems," International Journal of Electrical Power & Energy
pp. 297-332, 2016/05/01 2016, doi: 10.1007/s12667-015-0158-4. Systems, vol. 125, p. 106427, 2021/02/01/ 2021, doi:
[8] M. Premkumar et al., "An efficient and reliable scheduling algorithm https://doi.org/10.1016/j.ijepes.2020.106427.
for unit commitment scheme in microgrid systems using enhanced [24] H. O. Howlader, O. B. Adewuyi, Y.-Y. Hong, P. Mandal, A.
mixed integer particle swarm optimizer considering uncertainties," Mohamed Hemeida, and T. Senjyu, "Energy Storage System
Energy Rep, vol. 9, pp. 1029-1053, 2023/12/01/ 2023, doi: Analysis Review for Optimal Unit Commitment," Energies, vol. 13,
https://doi.org/10.1016/j.egyr.2022.12.024. no. 1, doi: 10.3390/en13010158.
[9] V. Kumar, R. Naresh, and V. Sharma, "Profit based unit commitment [25] C. Wang and S. M. Shahidehpour, "Effects of ramp-rate limits on
problem solution using metaheuristic optimisation approach," unit commitment and economic dispatch," IEEE Transactions on
International Journal of Systems Science: Operations & Logistics, Power Systems, vol. 8, no. 3, pp. 1341-1350, 1993, doi:
vol. 10, no. 1, p. 2037026, 2023/12/31 2023, doi: 10.1109/59.260859.
10.1080/23302674.2022.2037026. [26] Y.-Y. Hong and G. F. D. G. Apolinario, "Uncertainty in Unit
[10] A. Abuelrub and B. Awwad, "An improved binary African vultures Commitment in Power Systems: A Review of Models, Methods, and
optimization approach to solve the UC problem for power systems," Applications," Energies, vol. 14, no. 20, doi: 10.3390/en14206658.
Results in Engineering, vol. 19, p. 101354, 2023/09/01/ 2023, doi: [27] B. Postolov and A. Iliev, "New metaheuristic methodology for
https://doi.org/10.1016/j.rineng.2023.101354. solving security constrained hydrothermal unit commitment based on
[11] J. K. Skolfield and A. R. Escobedo, "Operations research in optimal adaptive genetic algorithm," International Journal of Electrical
power flow: A guide to recent and emerging methodologies and Power & Energy Systems, vol. 134, p. 107163, 2022/01/01/ 2022,
applications," Eur J Oper Res, vol. 300, no. 2, pp. 387-404, doi: https://doi.org/10.1016/j.ijepes.2021.107163.
2022/07/16/ 2022, doi: https://doi.org/10.1016/j.ejor.2021.10.003. [28] A. Agarwal and K. Pal, "Optimization of Unit Commitment Problem
[12] A. Hussain and A. Ismail, "Operation cost reduction in unit Using Genetic Algorithm," International Journal of System
commitment problem using improved quantum binary PSO Dynamics Applications (IJSDA), vol. 10, no. 3, pp. 21-37, 2021, doi:
algorithm," International Journal of Electrical and Computer 10.4018/IJSDA.2021070102.
Engineering (IJECE), vol. 10, p. 1149, 04/01 2020, doi: [29] F. Zhuang and F. D. Galiana, "Towards a more rigorous and practical
10.11591/ijece.v10i2.pp1149-1155. unit commitment by Lagrangian relaxation," IEEE Transactions on
[13] H. M. Aung Myo Win, "IGDT-Based Realistic Scheduling of Power Systems, vol. 3, no. 2, pp. 763-773, 1988.
Thermal Power Generators Under Integration of Wind Turbine

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3351710

[30] H. Z. Jangkung Raharjo, "Unit Commitment Effects in Economic Sustainable Energy, vol. 10, no. 3, pp. 1280-1289, 2019, doi:
Scheduling of Generators," International Review of Electrical 10.1109/TSTE.2018.2865454.
Engineering (IREE), vol. 16, no. 3, pp. 247-256, 2021. [46] K. Deb, A. Pratap, S. Agarwal, and T. Meyarivan, "A fast and elitist
[31] L. Yang, J. Jian, Y. Wang, and Z. Dong, "Projected mixed integer multiobjective genetic algorithm: NSGA-II," IEEE Transactions on
programming formulations for unit commitment problem," Evolutionary Computation, vol. 6, no. 2, pp. 182-197, 2002, doi:
International Journal of Electrical Power & Energy Systems, vol. 68, 10.1109/4235.996017.
pp. 195-202, 2015/06/01/ 2015, doi: [47] Z. Z. Liu, B. C. Wang, and K. Tang, "Handling Constrained
https://doi.org/10.1016/j.ijepes.2014.12.054. Multiobjective Optimization Problems via Bidirectional
[32] A. Viana and J. P. Pedroso, "A new MILP-based approach for unit Coevolution," IEEE Transactions on Cybernetics, vol. 52, no. 10, pp.
commitment in power production planning," International Journal 10163-10176, 2022, doi: 10.1109/TCYB.2021.3056176.
of Electrical Power & Energy Systems, vol. 44, no. 1, pp. 997-1005, [48] C. Grigg et al., "The IEEE Reliability Test System-1996. A report
2013/01/01/ 2013, doi: https://doi.org/10.1016/j.ijepes.2012.08.046. prepared by the Reliability Test System Task Force of the
[33] B. Yang et al., "Unit Commitment Comprehensive Optimal Model Application of Probability Methods Subcommittee," IEEE
Considering the Cost of Wind Power Curtailment and Deep Peak Transactions on Power Systems, vol. 14, no. 3, pp. 1010-1020, 1999,
Regulation of Thermal Unit," IEEE Access, vol. 8, pp. 71318-71325, doi: 10.1109/59.780914.
2020, doi: 10.1109/ACCESS.2020.2983183. [49] P. Biswas, P. Arora, R. Mallipeddi, P. Suganthan, and B. Panigrahi,
[34] E. S. Ali, S. M. A. Elazim, and A. S. Balobaid, "Implementation of "Optimal placement and sizing of FACTS devices for optimal power
coyote optimization algorithm for solving unit commitment problem flow in a wind power integrated electrical network," Neural
in power systems," Energy, vol. 263, p. 125697, 2023/01/15/ 2023, Computing and Applications, pp. 1-22, 11/05 2020, doi:
doi: https://doi.org/10.1016/j.energy.2022.125697. 10.1007/s00521-020-05453-x.
[35] M. Said, E. H. Houssein, S. Deb, A. A. Alhussan, and R. M. [50] R. Tanabe and A. Fukunaga, "Success-history based parameter
Ghoniem, "A Novel Gradient Based Optimizer for Solving Unit adaptation for Differential Evolution," in 2013 IEEE Congress on
Commitment Problem," IEEE Access, vol. 10, pp. 18081-18092, Evolutionary Computation, 20-23 June 2013 2013, pp. 71-78, doi:
2022, doi: 10.1109/ACCESS.2022.3150857. 10.1109/CEC.2013.6557555.
[36] J.-S. Pan, P. Hu, and S.-C. Chu, "Binary fish migration optimization [51] K. M. Sallam, S. M. Elsayed, R. K. Chakrabortty, and M. J. Ryan,
for solving unit commitment," Energy, vol. 226, p. 120329, "Improved Multi-operator Differential Evolution Algorithm for
2021/07/01/ 2021, doi: Solving Unconstrained Problems," in 2020 IEEE Congress on
https://doi.org/10.1016/j.energy.2021.120329. Evolutionary Computation (CEC), 19-24 July 2020 2020, pp. 1-8,
[37] H. Anand, N. Narang, and J. S. Dhillon, "Multi-objective combined doi: 10.1109/CEC48606.2020.9185577.
heat and power unit commitment using particle swarm optimization," [52] K. Deb, K. Sindhya, and T. Okabe, "Self-adaptive simulated binary
Energy, vol. 172, pp. 794-807, 2019/04/01/ 2019, doi: crossover for real-parameter optimization," in Proceedings of the 9th
https://doi.org/10.1016/j.energy.2019.01.155. annual conference on genetic and evolutionary computation, 2007,
[38] A. Rezaee Jordehi, "An improved particle swarm optimisation for pp. 1187-1194.
unit commitment in microgrids with battery energy storage systems [53] R. Cheng and Y. Jin, "A Competitive Swarm Optimizer for Large
considering battery degradation and uncertainties," Int J Energ Res, Scale Optimization," IEEE Transactions on Cybernetics, vol. 45, no.
vol. 45, no. 1, pp. 727-744, 2021, doi: 2, pp. 191-204, 2015, doi: 10.1109/TCYB.2014.2322602.
https://doi.org/10.1002/er.5867. [54] H. Jain and K. Deb, "An Evolutionary Many-Objective Optimization
[39] A. Shukla and S. N. Singh, "Multi-objective unit commitment with Algorithm Using Reference-Point Based Nondominated Sorting
renewable energy using hybrid approach," IET Renewable Power Approach, Part II: Handling Constraints and Extending to an
Generation, vol. 10, no. 3, pp. 327-338, 2016, doi: Adaptive Approach," Ieee Transactions on Evolutionary
https://doi.org/10.1049/iet-rpg.2015.0034. Computation, vol. 18, no. 4, pp. 602-622, Aug 2014, doi:
[40] Y. F. Li, N. Pedroni, and E. Zio, "A Memetic Evolutionary Multi- 10.1109/tevc.2013.2281534.
Objective Optimization Method for Environmental Power Unit [55] A. Panichella, "An improved Pareto front modeling algorithm for
Commitment," IEEE Transactions on Power Systems, vol. 28, no. 3, large-scale many-objective optimization," presented at the
pp. 2660-2669, 2013, doi: 10.1109/TPWRS.2013.2241795. Proceedings of the Genetic and Evolutionary Computation
[41] A. Trivedi, D. Srinivasan, K. Pal, C. Saha, and T. Reindl, "Enhanced Conference, Boston, Massachusetts, 2022. [Online]. Available:
Multiobjective Evolutionary Algorithm Based on Decomposition for https://doi.org/10.1145/3512290.3528732.
Solving the Unit Commitment Problem," Ieee T Ind Inform, vol. 11, [56] Y. Tian, T. Zhang, J. Xiao, X. Zhang, and Y. Jin, "A coevolutionary
no. 6, pp. 1346-1357, 2015, doi: 10.1109/TII.2015.2485520. framework for constrained multiobjective optimization problems,"
[42] R. Mena, M. Godoy, C. Catalán, P. Viveros, and E. Zio, "Multi- IEEE Transactions on Evolutionary Computation, vol. 25, no. 1, pp.
objective two-stage stochastic unit commitment model for wind- 102-116, 2020.
integrated power systems: A compromise programming approach," [57] E. Davoodi, E. Babaei, and B. Mohammadi-ivatloo, "An efficient
International Journal of Electrical Power & Energy Systems, vol. covexified SDP model for multi-objective optimal power flow,"
152, p. 109214, 2023/10/01/ 2023, doi: International Journal of Electrical Power & Energy Systems, vol.
https://doi.org/10.1016/j.ijepes.2023.109214. 102, pp. 254-264, 2018/11/01/ 2018, doi:
[43] A. G. Trojani, J. M. Baigi, and M. S. Moghaddam, "Stochastic https://doi.org/10.1016/j.ijepes.2018.04.034.
Security-constrained Unit Commitment Considering Electric
Vehicles, Energy Storage Systems, and Flexible Loads with
Renewable Energy Resources," Journal of Modern Power Systems AAMIR ALI received the B.E., M.E, Ph.D degrees in electrical engineering
and Clean Energy, vol. 11, no. 5, pp. 1405-1414, 2023, doi: from QUEST, Nawabshah, Pakistan. He is currently working as an Assistant
10.35833/MPCE.2022.000781. Professor in the department of Electrical Engineering, Quaid-e-Awam
[44] G. M. C. Leite, S. Jiménez-Fernández, S. Salcedo-Sanz, C. G. University of Engineering Science and Technology, QUEST, Nawabshah,
Marcelino, and C. E. Pedreira, "Solving an energy resource Sindh, Pakistan. His main research interests are power system optimization,
management problem with a novel multi-objective evolutionary grid connected and islanded operation of distributed generation, smart grid
reinforcement learning method," Knowledge-Based Systems, vol. and multi-objective evolutionary algorithms.
280, p. 111027, 2023/11/25/ 2023, doi:
https://doi.org/10.1016/j.knosys.2023.111027. ARSALAN ALI is working towards his master’s degree and main research
[45] A. J. Lamadrid, D. Muñoz-Alvarez, C. E. Murillo-Sánchez, R. D. interests are power system optimization, grid connected and islanded
Zimmerman, H. Shin, and R. J. Thomas, "Using the Matpower operation of distributed generation and smart grid.
Optimal Scheduling Tool to Test Power System Operation
Methodologies Under Uncertainty," IEEE Transactions on MUHAMMAD USMAN KEERIO was born in Nawabshah, Pakistan, in
December 1965. He received the bachelor’s degree in electrical (power)

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3351710

engineering from the Mehran University of Engineering and Technology,

Jamshoro, in 1991, the master’s degree in control engineering from NUST,
Karachi, in 2002, and the Ph.D. degree in controls and robotics from the
Beijing Institute of Technology, China, in 2008. His research interests
include electrical and control engineering, curriculum development,
institutional accreditation, neural networks, and optimization.

NOOR HUSSAIN MUGHERI received the B.E., M.E, Ph.D degrees in

electrical engineering from QUEST, Nawabshah, Pakistan. He is currently
working as an Assistant Professor in the department of Electrical
Engineering, Quaid-e-Awam University of Engineering Science and
Technology, QUEST, Nawabshah, Sindh, Pakistan. His main research
interests are power system optimization.

GHULAM ABBAS received the B.Eng. degree in electrical engineering

from QUEST, Nawabshah, Pakistan, in 2017, and the M.Eng. degree in
electrical engineering from Southeast University, Nanjing, China, in 2019,
where he is currently pursuing the Ph.D. degree in electrical engineering.
His research interests include distributed generations' integration, planning,
and optimization.

EZZEDDINETOUTI received the B.S. and master’s degrees in electrical

engineering in 1997 and 2005 from the Higher National College of
Engineers of Tunis (Tunisia) and the National College of Engineers of Tunis
(Tunisia) respectively, all in electrical engineering. He received the Ph.D.
Degree in induction generators, power quality and electric drives from the
National College of Engineers of Monastir (Tunisia) and Artois University
(France) in 2013. In 2014, he joined the Laboratory of Industrial Systems
Engineering and Renewable Energies (LISIER), University of Tunis,
Tunisia, as an Assistant Professor and Researcher. His research interests are
control of electrical systems, power electronics and renewable energy.

MOHAMMED HATATAH received the B.Sc. degree in electrical

engineering from King Abdulaziz University, Saudi Arabia, in 2007, the
MBA degree from King Abdulaziz University, Saudi Arabia, in 2010, the
M.Sc. degree in power system from Pennsylvania State University, USA, in
2015, and the Ph.D. degree in power electronics from University of
Pittsburgh, USA, in 2021. In 2005, he did a six-month internship at Saudi
Aramco, Saudi Arabia, where he worked on distribution equipment design.
From 2006 to 2011, he was an Electrical Planning Engineer at Saudi
Electricity Company, Saudi Arabia, where he worked on the regional
planning (110kV). Dr. Hatatah is a member of IEEE PELS and PES. His
research interests include power system, renewable energy, and control and
optimization of power converter.

AMR H. YOUSEF is working a HOD at the department of Electrical

Engineering Department, University of Business and Technology, Jeddah,
Saudia Arabia. He holds Ph.D. in Electrical and Computer Engineering from
Old Dominion University, Norfolk, Virginia USA.

MOUNIR BOUZGUENDA received his B.S. degree in Electrical

Engineering the Pennsylvania State University, USA, in 1985. He also
received his M.S. and Ph.D. degrees in Electrical Engineering from Virginia
Polytechnic Institute and State University, USA in 1988 and 1992,
respectively. Dr. Mounir taught in Virginia, Maryland and Washington, DC
and Tunisia. He also worked as a consultant with Standard Technologies
Institute, Maryland and the Temple Group, Washington DC and Computer
Engineering Services, Sfax-Tunisia. Dr. Mounir joined Sohar University,
Oman in 2000 as a Senior Lecturer and Sultan Qaboos University-Oman as
an Associate Professor in 2009. Currently, he is teaching in the Electrical
and Computer Engineering. His research interests include renewable energy
systems, power systems and power electronics. He has authored and co-
authored many technical papers in these areas.

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4