Application of The Arithmetic Optimization Algorithm To Solv 2022 Results in
Application of The Arithmetic Optimization Algorithm To Solv 2022 Results in
Results in Engineering
journal homepage: www.sciencedirect.com/journal/results-in-engineering
A R T I C L E I N F O A B S T R A C T
Keywords: This article presents a methodology to solve to the Optimal Power Flow (OPF) problem in Direct Current (DC)
Optimal power flow networks using the Arithmetic Optimization Algorithm (AOA) and Successive Approximation (SA). This master-
Power flow slave methodology solves the OPF problem in two stages: the master stage estimates the solution to the OPF
Optimization algorithms
problem considering its constraints and variables, and the slave stage assesses the fitness of the solution proposed
DC networks
Electric power
by the master stage. To validate the methodology suggested in this article, three test systems cited multiple times
Optimization in the literature were used: the 10, 21 and the 69 nodes test systems. In addition, three scenarios varying the
allowable power limits for the Distributed Generators (DGs) are presented; thus, the methodology explores so
lutions under different conditions. To prove its efficiency and robustness, the solution was compared with four
other methods reported in the literature: Ant Lion Optimization (ALO), Black Hole Optimization (BHO), the
Continuous Genetic Algorithm (CGA), and Particle Swarm Optimization (PSO). The results show that the
methodology proposed here to reduce power losses presents the best solution in terms of standard deviation.
1. Introduction losses, CO2 emissions, network operating costs, and out-of-limit voltage
profiles and current [4–6]. An overview of DC networks and the OPF is
Microgrids are defined as micro- and macro-scale systems that can presented below.
operate with both Alternating Current (AC) and Direct current (DC) in
isolated systems or connected to a power grid. Their implementation in 1.1. DC networks and the power flow problem
DC networks is well-known because they can be used to integrate
distributed energy resources [1]. Non-conventional energy sources are Due to the progress made and the results obtained in power elec
one of the most important resources nowadays because they reduce CO2 tronics, it is possible to transmit and distribute energy as direct current
emissions significantly by decreasing the production of electric power [7]. DC networks have advantages over their alternating current coun
from fossil fuels. Due to the relevance of DC networks today, the sci terparts, e.g., phasors and reactive components; hence, they make it easy
entific community has identified the need to study ways to improve their to calculate their characteristic power variables. They also eliminate the
technical-economical aspects and increase their efficiency [2,3]. As a use of AC/DC converters, thus reducing power losses by conversion
result, it has been found that integrating Distributed Generators (DGs) [8–10].
into DC networks can improve their behavior and performance consid DC networks are usually composed of one slack generator (some
erably—which is known as Optimal Power Flow (OPF) and can be used times two), distribution lines, constant power loads, distributed gener
to study and optimize several network-related problems. However, some ators, and power storage devices [11,12]. To design them, after the DC
of their most frequent technical operating difficulties are electric power network and its components have been determined, it is necessary to
* Corresponding author.
E-mail addresses: [email protected] (J. Montano), [email protected] (O.D. Garzón), [email protected] (A.A. Rosales Muñoz),
[email protected] (L.F. Grisales-Noreña), [email protected], [email protected] (O.D. Montoya).
https://doi.org/10.1016/j.rineng.2022.100654
Received 3 June 2022; Received in revised form 2 September 2022; Accepted 19 September 2022
Available online 24 September 2022
2590-1230/© 2022 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
J. Montano et al. Results in Engineering 16 (2022) 100654
establish the amount of power that DGs should inject into the grid to quality solutions and low processing times, as well as the standard de
improve network operation and thus obtain economic, operating, viation of the optimization algorithms and the currents flowing through
technical, and environmental benefits [7]. The objective of OPF is to the lines, which were not considered in some studies.
improve the objective function set by the network operator. Here, the Hence, this paper proposes a new optimization technique based on a
objective function is the reduction of power losses (Ploss) associated master-slave methodology where the master stage uses the AOA and the
with power distribution considering that it has been widely used in the slave stage is performed by the SA. This methodology proposes a solu
specialized literature, according to a state-of-the-art review [13]. tion for the 21- and 69-bus test systems with distributed generation of
20%, 40%, and 60% of the power provided by the slack generator. The
1.2. The OPF problem in DC networks: state of the art solution quality, processing time, and robustness of the solution meth
odology are thoroughly analyzed.
Multiple investigations have focused on the OPF problem in DC
networks in recent years. They have proposed solutions such as the 1.3. Scope and contributions
implementation of algorithms based on commercial software and
sequential programming models using optimization techniques and This document addresses the OPF problem from the viewpoint of
numerical methods. The optimal power value that DGs in a network sequential programming methodologies, thus avoiding the use of com
should inject to comply with regulations and to minimize power losses mercial software, and proposes the application of the AOA combined
can be determined employing these solutions [14]. with SA. The purpose is to generate a highly effective hybrid method
Regarding the OPF solution methodology using commercial soft ology in terms of quality and time. The following are the main contri
ware, in Refs. [15–17], the objective function is to minimize power butions of this paper to the state of the art:
losses associated with power distribution. Due to the solution method
ology used in this paper, it is not possible to replicate the ones proposed ● A new application for the arithmetic optimization algorithm.
by those authors because they used commercial software. In Ref. [15], ● A new master-slave methodology that solves the OPF problem in DC
the convex relaxation method based on second-order conic program networks.
ming is used by means of MOSEK, which is commercial software. In that ● Better results (in terms of quality, repeatability, and computation
paper, the OPF problem is solved for a 16-bus DC network considering times) than those of other methodologies that have been imple
processing times and standard deviation. In Ref. [16], sequential convex mented to solve the OPF problem in DC networks.
programming is used to solve the OPF problem for 10- and 21-bus test
systems considering processing times but not standard deviation. In
1.4. Structure of the article
Ref. [17], GAMS/CONOPT and the continuous genetic algorithm are
used to assess the OPF problem for a 10-bus test system, comparing the
This paper is organized as follows. Section 2 presents the mathe
solution obtained by the proposed methodology and the one obtained by
matical formulation of the OPF problem in DC networks, where the
GAMS; however, the processing times and the standard deviation are not
objective function is the reduction of power losses associated with power
considered. The cited studies used commercial software and small DC
distribution considering all the constraints of DC networks in a distrib
networks (10, 16, and 21 buses); thus, it is not possible to establish the
uted generation environment. Section 3 introduces the proposed opti
robustness of those methodologies. Besides, they did not consider the
mization technique to solve the OPF problem applying a master-slave
currents flowing through the power lines, the worst voltage in the sys
methodology that combines the AOA and SA. Section 4 details the
tem, and other parameters necessary to establish network performance.
optimization algorithms that are compared with the proposed solution
Aiming to promote the use of open source software and eliminate the
methodology, as well as the test systems used to perform the simula
use of its commercial counterpart, new sequential programming
tions. Section 5 reports the results obtained by the optimization algo
methods based on optimization algorithms and numerical methods have
rithms for the 10, 21 and 69 nodes test systems at different percentages
been suggested to solve the OPF problem; for instance, in Refs. [18–20].
of distributed power penetration. Finally, Section 6 draws the conclu
These techniques implement a master-slave solution methodology,
sions and suggests future research.
where the optimization algorithm is the master stage, and a numerical
method selected to solve the power flow problem is the slave stage. In
2. Mathematical formulation
Ref. [18], the multiverse optimization algorithm is used as the master
stage, and Successive Approximation (SA) as the slave stage. That paper
The state-of-the-art review above shows that the solution to the OPF
proposes a solution to the OPF problem for the 21- and 69-bus test
problem is related to the power flow and the constraints set by the
systems, considering the currents flowing through the lines but dis
network operator [21]. Once they are established, it is possible to
regarding processing times. In Ref. [19], the sine cosine algorithm is
determine the stability of the system, the current congestion in the lines,
proposed as the master stage; and SA, as the slave stage. In that article,
and the power losses, among others. Thus, in this case study, the power
the solution to the OPF problem is proposed only for the 21-bus system,
losses are the variable of interest, and the objective function is to
the currents flowing through the lines are not considered, and the pro
minimize them. The equations that compose the objective function and
cessing times and standard deviation obtained by the optimization al
the OPF constraints are presented below.
gorithms are disregarded. In Ref. [1], the authors propose a master-slave
methodology combining the Ant Lion Optimization (ALO) algorithm and
SA to solve the OPF problem in 21- and 69-bus systems, considering 2.1. Objective function
processing times but disregarding the standard deviation. Some of the
articles that suggest the use of open source software do not analyze the The objective function is defined as a specific problem to be solved or
computation time required by the methodologies; then, it is not possible minimized by applying an optimal flow. This paper addresses the
to determine the methodology’s performance. Also, among the cited reduction of power losses in a DC network.
papers, tuning is performed in only one, i.e., [18]; therefore, the tech
minPloss = vT GL v (1)
niques are not implemented under the same operating conditions.
After the state-of-the-art review above, we conclude that new tech In Equation (1), Ploss represents the function to be reduced, that is,
niques should apply sequential programming and encourage the use of the power losses in the system as a function of v, which is a vector
open source software to solve the OPF problem. Additionally, new containing all the voltage profiles calculated using the load flow; and GL
studies should detail the tuning of the techniques that produce high is the matrix that represents the conductivity effects of each line.
2
J. Montano et al. Results in Engineering 16 (2022) 100654
define the minimum and maximum power that the DGs can supply. minimum power level assigned to each generator in the OPF problem.
Equations (3) and (4) determine and limit the power to be injected into The first population is generated from random values (rand) in the [0 −
the microgrid, both the power coming from the main generator located 1] range to explore larger regions in the search space. They are multi
in the slack node and that supplied by the DGs, in such a way that there plied by the difference between the limits, thus allowing a greater dis
are no power losses due to over injection. Equation (5) includes vmin and tribution of individuals within the search space.
vmax, which define the maximum and minimum allowable voltage. This Xi(i,j) = ((ub − lb) ∗ rand ) + lb (8)
constraint ensures that the voltages resulting after power injection
comply with voltage regulations, avoiding sags and swells that desta The previous equation determines the values of individuals i in
bilize the system. One of the limits that facilitate a stable system is dimension j. To create the entire population of individuals Xi, we pro
Equation (6) because it contains the current flowing through the lines, pose a size nxd matrix, where n represents the number of individuals as
Iij, which must be lower than Imax viable solutions to the problem, and d represents the number of vari
ij . Finally, Equation (7) defines the
ables. In Equation (9), Xn is the n-th individual in matrix MXi. In the OPF
maximum penetration of DG, where α represents the allowable per
problem, the total number of columns d is the number of DGs in the DC
centage of penetration with respect to the power generated by the slack
network (other than the slack node), and the value in each column is the
node.
power they inject into the network.
3. Proposed methodology
3
J. Montano et al. Results in Engineering 16 (2022) 100654
⎡ ⎤ ⎡ ⎤
Xi1,1 Xi1,2 ⋯ ⋯ Xi1,d Xi1 ⎧ (( ) )
⎢ Xi2,1
⎢ Xi2,2 ⋯ ⋯ Xi2,d ⎥ ⎢ ⎥
⎥ ⎢ Xi2 ⎥ ⎨ Best_Xi × MOP × ub(1,j) − lb(1,j) × μ + lb(1,j)
⎪ r2 < 0.5
MXi = ⎢
⎢ ⋮ ⋮ ⋮ ⋮ ⋮ ⎥ ⎢
⎥=⎢ ⋮ ⎥
⎥ (9) Xi(i,j) = Best_Xi (( ) )
⎣ ⋮ ⎪
⎩ × ub(1,j) − lb(1,j) × μ + lb(1,j) r2 > 0.5
⋮ ⋮ ⋮ ⋮ ⎦ ⎣ ⋮ ⎦ (MOP + ε)
Xin,1 Xin,2 ⋯ ⋯ Xin,d Xin
(14)
3.1.2. Calculation of the objective function This equation depends on a random value, r2, in the [0 − 1] range.
The slave stage evaluates the objective function of each individual This value allows determining the phase to be used, that is, if (r2 < 0.5),
(fitness function) and then assesses the impact of the viable solutions the M operator is used. Therefore, the exploration phase is based on
contained in MXi. Each value is stored in a nx1 matrix called MOXi, and Equation (14), using the individuals (Best_Xi) who provided the best
the losses in the power injected by each DG are calculated (see Equation solution so far employing the Math Optimizer Probability (MOP), the
(10)). In the OPF problem, each individual generated in the search space upper bounds (ub), lower bounds (lb), and a fixed value (μ = 0.5), which
and assigned to each Xi is evaluated and the values of the variables for is a parameter to control the search fitness. If the opposite occurs (i.e., r2
each possible solution are obtained. > 0.5), the D operator will be implemented using the best individuals,
⎡ ([ ]) ⎤ the upper and lower bounds, the MOP, and a parameter ϵ = 2.2204e− 16.
f ( [X1,1 , X1,2 , …, X1,d ] )
⎢ f X2,1 , X2,2 , …, X2,d ⎥ cont1/α
MOXi = ⎢ ⎥ (10) MOP = 1 − (15)
⎣ ⋮ ⎦ Maxiter1/α
([ ])
f Xn,1 , Xn,2 , …, Xn,d In Equation (15), MOP is a coefficient expressed as the value of the
The individuals with the best solution in the MOXi matrix (i.e., lowest function at the present iteration, the maximum number of iterations, and
power losses) are selected as the incumbent solution. Then, the objective a parameter α = 5 that determines the accuracy of the exploration phase
function is stored in Equation (9), and the configuration of its variables throughout the iterative process [22].
during the iterative cycle are stored in Equation (10). To establish the
best solution at the current iteration, this study proposes comparing the ● Exploitation phase: This search phase uses Addition (A) and Sub
solutions stored in the solutions vector given by the objective function. If traction (S) to take advantage of the search space because these
a better solution than the incumbent is obtained at a certain iteration, operators make it easy to approach a solution due to the low
the Best_Fob and Best_Xi values are updated. In Equation (12), Best_Xi is a dispersion of their search mechanism. Thus, the aim of the exploi
vector of 1xd, and a denotes the individuals that represent the best tation phase is to determine the most optimal solution (among the
objective function at each iteration. given solutions) at each iteration using Equation (16).
[ ( )] { (( ) )
Best_Fob = f Xi1,d (11) Xi(i,j) =
Best_Xi × MOP + ( ( ub(1,j) − lb(1,j) ) × μ + lb(1,j) ) r3 < 0.5
Best_Xi × MOP − ub(1,j) − lb(1,j) × μ + lb(1,j) r3 > 0.5
Best_Xi = [Xa ] (12) (16)
This equation, like the exploration phase, depends on a random
3.1.3. Exploration and exploitation phases value, r3, in the [0 − 1] range. If (r3 < 0.5), S is used to perform the
The AOA applies an iterative process using arithmetic operators ac exploitation based on Equation (16). The latter uses the individuals who
cording to the phase: exploration and exploitation (i.e., the movement provided the best solution obtained so far (Best_Xi) employing the MOP,
method of the optimization technique). The exploration phase uses the the addition of upper bounds (ub), lower bounds (lb), and parameter μ. If
Division (D) and Multiplication (M) operators, which obtain decision (r3 > 0.5), R is used following the same process, but doing a subtraction.
values under a high dispersion rate and compromise the exploitation Once the phase to be used is determined and it is applied to the
phase. Note that operators D and M cannot easily approach a feasible current iteration, the position of each individual in the iterations is
solution due to their high dispersion, unlike Addition (A) and Subtrac updated, making sure they respect the constraints of the problem; in this
tion (S), which are used in the exploitation phase and can easily case, not exceeding the power values set for the DGs. Subsequently,
approach a feasible solution. A and S can be used to start each search MOXi is updated by assessing the new individuals according to the
phase, which depends on a starting point obtained by calculating objective function and updating the incumbent problem at each itera
Equation (13), also known as the Math Optimizer Accelerated (MOA). In tion. This process is repeated until the stopping criteria established for
other words, exploration and exploitation are used to generate all the this problem are met.
new individuals in the iterative process in each particle and dimension
of the problem. 3.1.4. Stopping criteria
(
Max − Min
) The two stopping criteria used in the master stage are described
MOA = Min + Cont × (13) below:
Maxiter
This equation determines which phase will be executed in the iter ● The master stage will end after n consecutive iterations if the
ative process in the AOA (i.e., exploration or exploitation). Here, Cont incumbent of the problem is not updated. That is, the iterative pro
denotes the current iteration, ranging from an initial value of 1 to the cess ends when the objective function reaches a considerable number
maximum number of iterations, i.e., Maxiter. The value defined as the of iterations (a non-improvement counter) without finding a better
minimum of the accelerated function (Min = 0.2) and the maximum solution to the problem.
value (Max = 1) enable the MOA to determine a value to compare it with ● The computation ends when the optimization algorithm reaches the
a random value, i.e., r1. The process advances as follows: if (r1 < MOA), maximum allowable number of iterations, which is controlled by a
the exploration phase is used; otherwise, the exploitation phase is counter in the algorithm.
employed. The two phases are described below:
The iterative process in Algorithm 1 is presented below to help
● Exploration phase: It uses D and M to conduct a stochastic search. readers understand the proposed AOA-SA solution methodology.
These two operators, which are its main search strategies, are
modeled in Equation (14).
4
J. Montano et al. Results in Engineering 16 (2022) 100654
Algorithm 1
Hybrid AOA-SA optimization algorithm
The slave stage assesses the fitness of the solution suggested by the
master stage and solves the power flows to determine if the suggested
power injections for the DGs are adequate. The slave stage—in which
the objective function is evaluated—should be fast, accurate, and
effective because optimization algorithms should be efficient.
This paper suggests solving the OPF problem using an iterative
method based on SA [23], which was selected due to the convergence of Fig. 2. Hybrid AOA-SA optimization algorithm diagram.
its solution and its short processing time. This method is based on the
following equation: the flowchart are described below:
The proposed methodology required that, first: the data of the
Gdd ⋅vd = − D−d 1 (vd )Pd − Gdg ⋅vg (17)
electrical system must be load and the initial condition that represent
where Gdd is a positive symmetric matrix containing all the conductivity the OPF problem in DC networks, must be load too. The data correspond
effects of the distribution lines; vg, the voltage profile of the slack node; to electrical parameters, nodes connections, resistive in the branch,
and vd, the voltage at demand nodes. Through a mathematical devel percentage of power penetration (α), optimization algorithm parame
opment applied to Equation (17), it is possible to get a new equation and ters, among others. Second: the master stage (AOA) begins, in this stage
establish the nodal voltages at demand nodes. is generate the initial popualtion, next the stage slave evaluate the
objetive funcion for each particle in the initial population and the best
vd = − G−dd1 [D−d 1 (vd )Pd + Gdg ⋅ vg ] (18) incumbent is assigned for the firts population. Third: Based on the
incumbent and the initial parameters, the population is updated. Fourth:
To find these values in the vd system, an iterative process should be
checks if any of the stop criteria are met. Fifth: once a stop criterion is
implemented to find the nodal voltages with an almost null convergence
met, the optimization algorithm stops and the results obtained are
error. Thus, a t counter should be implemented in Equation (18). The
presented. It should be noted that these iterative processes are run over
following is the equation to find the voltage profiles:
and over again until the stop criteria are met in their entirety and when
vt+1
d = − G−dd1 D−d 1 (vtd )Pd + Gdg ⋅vg (19) this occurs the best result obtained for the OPF is selected and these
results are presented.
To explain the master-slave methodology in detail, which aims to
reduce the power losses in DC networks, in Fig. 2 is presented the
4. Test scenarios and considerations
flowchart of the methodology describe in this research document. The
flowchart presents all step necessaries to execute the optimization al
In this study, three test systems were used to validate the proposed
gorithm and obtain the solution to OPF problem. The steps presented in
5
J. Montano et al. Results in Engineering 16 (2022) 100654
The 21 nodes test system is presented in Ref. [24]. In Ref. [21], the
authors adapted this system to conduct tests in DC. This system,
composed of 21 buses and 20 lines, includes only constant power loads.
Its electrical configuration is presented in Fig. 9, and its parameters are
detailed in Table 2. Fig. 5. Electrical configuration of the 69-bus test system.
Table 2 shows the following data of the 21-bus test system (from left
to right): transmitting bus (from), receiving bus (to), resistance (R[pu])
Table 1
of the line connecting those buses, and power demand at that bus (P
Electrical parameters of the 10 nodes test system.
[pu]).
In the 21-bus test system, the slack node (Node 1) generates 5.8160 From To R [pu] P [pu] R [pu]
p.u. in its initial state. Besides, the voltage at the slack node is considered 1(slack) 2 0.0050 0 0
to be flat, that is, 1 p.u. The power losses in the system equal 0.27603 p. 2 3 0.0015 − 0.8 0
2 4 0.0020 − 1.3 0
u. A voltage and a base power of 1 kV and 100 kW, respectively, are
4 5 0.0018 0.5 0
assumed in this system. Data about the actual power and voltage in this 2 6 0.0023 0 2.0
system are available to validate the results in case real values are 6 7 0.0017 0 0
required. DGs were located in the system considering a state-of-the-art 7 8 0.0021 0.3 0
review and previous studies that have produced outstanding results. 7 9 0.0013 − 0.7 0
3 10 0.0015 0 1.25
Therefore, the DGs were located at nodes 9, 12, and 16, as shown in
[25].
Table 2
4.3. 69 nodes test system
Electrical parameters of the 21 nodes test system.
The 69 nodes test system is an adaptation of the classical AC 69-node From To R [pu] P [pu] From To R [pu] P [pu]
test system employed for power loss reduction via distributed generation 1(slack) 2 0.0053 − 0.70 11 12 0.0079 − 0.68
integration in AC networks, presented in Refs. [4,26,27]. The base 1 3 0.0054 0.00 11 13 0.0078 0.10
values used in this system to obtain p.u. values are 12.66 kV and 100 3 4 0.0054 − 0.36 10 14 0.0083 0.00
4 5 0.0063 − 0.04 14 15 0.0065 0.22
kVA. Additionally, modifications to the electrical configuration were 4 6 0.0051 0.36 15 16 0.0064 − 0.23
made in this study, as shown in Fig. 10. The parameters of this test 3 7 0.0037 0.00 16 17 0.0074 0.43
system are detailed in Table 3. Initially, in the 69-bus test system, the 7 8 0.0079 − 0.32 16 18 0.0081 − 0.34
power generation of the slack node is 40.4311 p.u., and the power losses 7 9 0.0072 0.80 14 19 0.0078 0.09
3 10 0.0053 0.00 19 20 0.0084 0.21
equal 1.5385 p.u. The slack node operates with a voltage of 1 p.u. In this
10 11 0.0038 − 0.45 19 21 0.0082 0.21
system, the generators are located at nodes 26, 61, and 66, as reported in
Ref. [13] (see Table 4).
6
J. Montano et al. Results in Engineering 16 (2022) 100654
Table 3
Electrical parameters of the 69 nodes test system.
From To R[pu] P[pu] From To R[pu] P[pu] From To R[pu] P[pu]
Table 4 Table 6
Parameters of the continuous methods employed in the master stage for the 10 Parameters of the continuous methods employed in the master stage for the 69
nodes test system. nodes test system.
Method AOA ALO BH CGA PSO Method AOA ALO BH CGA PSO
(BHO), continuous genetic algorithm (CGA), and particle swarm opti 5. Simulations and results
mization (PSO). Each of these five methods was used in the master stage
of the proposed master-slave methodology to solve the OPF problem. In This section presents the results obtained by the five optimization
said methodology, the master stage uses the optimization algorithm to methods implemented here to solve the OPF problem in DC networks
establish the power values to be injected by the DGs; in turn, the slave using the 10, 21 and 69 nodes test systems described in Section 4. The
stage employs the SA numerical method to establish the power loads. To simulations were conducted in MATLAB R2018b running on a work
confirm the effectiveness of the five methods, this study used the 10, 21 station with 16 GB of RAM, a 3.50-GHz Intel(R) Xeon (R) processor, a
and the 69 nodes test systems, which are frequently implemented in the 446-GB solid state drive, and Microsoft Windows 10 Pro for Worksta
literature for this purpose [2,4,18,24,28]. tions. Each solution methodology was run 100 times to evaluate its
To make a fair comparison of the five methods employed here to repeatability, accuracy, and standard deviation. The results in each
solve the OPF problem, the optimization algorithms were tuned. Such system are presented below.
tuning was performed using the PSO algorithm (presented in Ref. [4]) in
the two bus test systems so that every method could obtain the best
possible solution according to the objective function. The tuning pa 5.1. Results in the 10-bus test system
rameters were population size, [1–100]; maximum number of iterations
of each algorithm, [1–1000]; and non-improvement iterations, Table 7 presents the results obtained by the proposed methodology
[1–1000]. The population in the PSO algorithm was 10 individuals, and and the optimization algorithms after performing 100 runs for each of
the maximum number of iterations was 300. Finally, Tables 5 and 6 the test scenarios. To carry out this work, each of the DGs is allowed to
present the parameters of the five methods tuned for the 21- and 69-bus inject 20%, 40% and 60% of the power generation injected by the Slack
test systems, respectively. These parameters enable each optimization generator for the base case (system without distributed generation). The
method to find the best solution to the OPF problem according to the table mentioned above represents the information of the 10-node DC
objective function. system and is arranged as follows: in the first column is located the
optimization algorithm used to obtain the solution to the OPF problem
in DC networks, in the second column is the sum of the net power
Table 5 injected by the 3 DGs; in the third and fourth columns are the minimum
Parameters of the continuous methods employed in the master stage for the 21 Ploss and the average Ploss obtained by the optimization algorithms rep
nodes test system.
resented in kW, respectively. Similarly, the fourth and fifth columns
Method AOA ALO BH CGA PSO show the percentage of loss reduction achieved (%) with respect to the
Number of particles 64 79 67 52 49 Ploss of the base system. Column 5 shows the percentage of standard
Maximum number of iterations 783 769 317 592 679 deviation obtained by each optimization algorithm. Finally, the sixth
Maximum number of non-improvement 783 441 317 346 263 and seventh columns show the voltage farthest from 1 (worst voltage)
iterations
and the highest current flowing through the conductor lines for the
7
J. Montano et al. Results in Engineering 16 (2022) 100654
Table 7
Results of the simulation in the 10 nodes test system.
10 nodes test system
Method Total injected power [kW] Power Losses Reduction Worst V [p.u] Imax [A]
system under analysis. For this system a 500 kcmils conductor is selected
according to NTC2050.
Through Table 7 we obtain Figs. 6 and 7, through which the solutions
reached by each optimization algorithm after the 100 runs are
compared. In the first figure, the comparison in terms of minimum Ploss
between the AOA and the other optimization algorithms employed to
solve the OPF problem making use of the three penetration percentages
of 20%, 40% and 60% are presented. For the scenario that is allowed a
penetration percentage of 20%, the AOA achieves a minimum Ploss
reduction of 2.9615%, outperforming the CGA, BH, ALO and PSO by
0.0687%, 0.2487%, and 0.0008% and 0.0001% respectively. For the
case considering 40% distributed generation again the AOA is the
optimization algorithm that achieves the best minimum Ploss reduction,
achieving a reduction of 4.7839% and outperforming the other algo
rithms by an average percentage of 0.0677%. For the case considering
the 60% power injection by the GDs, again the AOA achieves the best
minimum Ploss reduction, achieving a reduction of 5.6218%, out
performing the PSO by 0.00004%, the CGA by 0.0005%, the ALO by
0.0653% and the BH by 0.0752%. The above analyzed results show that Fig. 7. Average results in the 10 nodes test system.
the AOA is the algorithm that allows obtaining the best results in terms
of minimum Ploss reduction for the 10-node system. In Fig. 7 it is possible to appreciate the average Ploss reduction per
centage obtained by each algorithm for the three allowed penetration
percentages, as well as the standard deviation and the average time
taken by each algorithm to obtain the best solution for the problem
addressed. For the case considering the average Ploss, the AOA achieves
an average reduction of 13.09 kW, tying with the PSO and being out
performed by the ALO by only 0.01 kW, by the CGA by only 0.02 kW and
the BH by only 0.03 kW. For the scenario that evaluates the standard
deviation, the AOA achieves an average percentage of 0.19%, being
surpassed only by the PSO by 0.1899%, but in turn surpasses the CGA by
11.41%, ALO by 13.56% and the BH by 21.48%. For the case where
processing times are considered, the AOA manages to solve the OPF
problem for the 10-node system in an average time of 6.39s, being
outperformed only by the BH and CGA by only 1.31s and 1.28s,
respectively; but outperforming the ALO by 1.28s and the PSO by 1.92s.
The results discussed above show that the AOA can obtain good solu
tions in terms of average Ploss, obtaining an excellent standard deviation
for each penetration percentage, and managing to reach the solution of
this system in a short average time.
8
J. Montano et al. Results in Engineering 16 (2022) 100654
Table 8 shows the results of the five methods in the 21-bus test
system with 20%, 40%, and 60% of the power injected by DGs with
respect to the maximum power level injected by the slack node. This
table contains seven columns (from left to right): optimization method;
total power injected by the 3 DGs; minimum power losses (Ploss) ob
tained by each method [kW] and reduction in percentage compared with
the base case; average Ploss of each method [kW] and reduction in per
centage compared with the base case; standard deviation STD of each
method [%]; worst voltage of the system in p.u.; and maximum voltage
flowing through the distribution lines [A]. In the 21-bus test system, the
maximum allowable voltage was 520A.
Table 8 presents the results obtained by the optimization methods in
the 21 nodes test system. Figs. 8 and 9, which were obtained from said
table, compare the minimum and average Ploss obtained by the five
optimization algorithms. Fig. 8 compares the average minimum error in
Ploss in the 3 DG scenarios: 20%, 40%, and 60% of the power supplied by Fig. 8. Average reductions in power loss compared to AOA in the 21 nodes
the slack node compared to AOA. The AOA and PSO present the best test system.
responses in terms of Ploss reduction with 20% penetration, both
achieving a 52.2382% reduction compared with the base case, thus
outperforming ALO, the CGA, and BHO by 0.0039%, 0.0485%, and
0.4256%, respectively. With 40% penetration, the AOA and PSO ach
ieved the best reductions in Ploss, followed by ALO, which ranks in third
place with a reduction of 77.8220%, the CGA in fourth place with
77.8175%, and BHO in the fifth place with 77.6280%. With a penetra
tion of 60%, the AOA achieved a reduction percentage of 89.9061%,
which was outperformed by PSO and ALO by 0.0022% and 0.0010%,
respectively. However, the AOA outperformed the CGA and BHO by
0.0153% and 0.1069%, respectively.
Fig. 9 shows the differences in average Ploss between the proposed
optimization algorithm and the other four methods after running them
100 times. With 20% penetration, the AOA presented the best reduction
in average Ploss with 52.2338%, thus outperforming (in this order) PSO,
ALO, the CGA, and BHO by 0.1143%, 0.2983%, 0.3610%, and 3.2130%,
respectively. With 40% penetration, the AOA produced the best reduc
tion in Ploss with 77.7971%, thus outperforming ALO, PSO, the CGA, and
BHO by 0.0014%, 0.692%, 0.0700%, and 1.4755%, respectively. With a
penetration of 60%, the AOA achieved a reduction percentage of Fig. 9. Average results in the 21 nodes test system.
89.8930%, which was outperformed by ALO by 0.0069%. Nevertheless,
the AOA still outperformed PSO, the CGA, and BHO by 0.0439%,
0.0804% and 0.9134%, respectively.
Table 8
Results of the simulation in the 21 nodes test system.
21 nodes test system
Method Total injected power [kW] Power Losses Reduction Worst V [p.u] Imax [A]
9
J. Montano et al. Results in Engineering 16 (2022) 100654
Fig. 10. Average reductions in power loss compared to AOA in the 69 nodes Fig. 11. Average results in the 69 nodes test system.
test system.
11, which compare the minimum and average Ploss of the AOA with those
To assess the repeatability and precision of the optimization algo of the other four optimization algorithms.
rithms in solving the OPF problem in the 21 nodes test system, Fig. 9 Fig. 10 shows the difference in average minimum error in Ploss with
compares the standard deviation of the proposed optimization method 20%, 40%, and 60% of DG injection compared to AOA. In this figure,
with those of the other four optimization algorithms. With 20% power with 20% injection, the AOA presented the best solution to the OPF
injection, the AOA presents the best standard deviation, i.e., 0.0058%, problem with a reduction of 63.2854%, outperforming PSO by 0.0001%,
outperforming the CGA by 0.2729%, PSO by 0.7775%, ALO by ALO by 0.0481%, the CGA by 0.647%, and BHO by 0.8577%. With 40%
1.1341%, and BHO by 2.4123%. With 40%, the AOA achieved the best injection, PSO showed a reduction of 90.9048%, outperforming the AOA
standard deviation again with 0.055%, thus outperforming the CGA, by 0.0001%, followed by ALO, the CGA, and BHO with 90.8948%,
ALO, PSO, and BHO by 0.1802%, 0.7448%, 1.1880%, and 3.1824%, 90.8936%, and 90.5027%, respectively. With 60%, the AOA and PSO
respectively. With 60%, the AOA obtained a standard deviation of obtained the best value in Ploss reduction, i.e., 96.3888% each, followed
0.074%, outperformed only by ALO by 0.030%. In this scenario, the by the CGA with 96.3884%, ALO with 96.3876%, and BHO with
AOA outperformed the CGA by 0.375%, PSO by 1.932%, and BHO by 96.1755%.
4.686%. Fig. 11 shows the differences in average Ploss between the AOA and
the other four optimization algorithms. In the 20% scenario, the AOA
exhibited the best average power loss reduction, i.e., 63.2840%, out
5.3. 69 nodes test system performing PSO, the CGA, ALO, and BHO by 0.1392%, 0.3867%,
0.5924%, and 3.7985%, respectively. With 40% power injection, the
Table 9 presents the results obtained by each optimization algorithm AOA presented the best Ploss average reduction with 90.9021%, followed
analyzed in the 69 nodes test system, where the DGs were enabled to by the CGA with 90.8006%, PSO with 90.6968%, ALO with 90.8948%,
inject 20%, 40%, and 60% of the power supplied by the slack node. It is and BHO with 87.8768%. With 60% power injection, the AOA and PSO
organized as Table 8, but the maximum allowable current was 335A in showed a reduction of 96.3888%, outperforming the CGA by 0.0156%,
this case. The results presented in Table 9 are illustrated in Figs. 10 and
Table 9
Results of the simulation in the 69 nodes test system.
69 nodes test system
Method Total injected power [kW] Power losses Reduction Worst V [p.u] Imax [A]
10
J. Montano et al. Results in Engineering 16 (2022) 100654
ALO by 0.1142%, and BHO by 1.7997%. y la Innovación, Fondo Francisco José de Caldas}; Instituto Tecnológico
As with the 21 nodes test system, Fig. 11 shows the differences in Metropolitano; Universidad Nacional de Colombia; and Universidad del
standard deviation between the AOA and the other four optimization Valle under the Project entitled “Estrategias de dimensionamiento,
methods in the 69-bus system. This figure can be used to evaluate the planeación y gestión inteligente de energía a partir de la integración y la
accuracy and repeatability of each method in finding a good quality optimización de las fuentes no convencionales, los sistemas de almace
solution. As shown in Fig. 11, the AOA presented the best standard de namiento y cargas eléctricas, que permitan la generación de soluciones
viation with 20% penetration, i.e., 0.0015%, thus outperforming PSO by energéticas confiables para los territorios urbanos y rurales de
0.4057%, the CGA by 0.4166%, ALO by 1.1578%, and BHO by 4.4925%. Colombia”, which is part of the research program entitled “Estrategias
With 40% DG, the AOA exhibited a standard deviation of 0.0164%, para el desarrollo de sistemas energéticos sostenibles, confiables, efi
outperforming the other optimization algorithms by an average of cientes y accesibles para el futuro de Colombia”.
5.3788%. With 60% penetration, the AOA obtained a standard deviation
of 3.94 × 10− 6, which was outperformed by PSO by only 3.35 × 10− 6. References
However, the AOA outperformed the CGA by 0.2977%, ALO by
6.3320%, and BHO by 21.5430%. This analysis proves that the AOA [1] O. Garzon-Rivera, J. Ocampo, L. Grisales-Noreña, O. Montoya, J. Rojas-Montano,
Optimal power flow in direct current networks using the antlion optimizer, Statist.
method is the best option for obtaining a high quality solution every Optimizat. Informat. Comput. 8 (4) (2020) 846–857.
time the algorithm is run. [2] A. Garces, Uniqueness of the power flow solutions in low voltage direct current
grids, Elec. Power Syst. Res. 151 (2017) 149–153.
[3] O.D. Montoya, W. Gil-González, V.M. Garrido, Voltage stability margin in dc grids
6. Conclusions with cpls: a recursive Newton–raphson approximation, IEEE Transact. Circuits Syst.
II: Express Briefs 67 (2) (2019) 300–304.
This article presented a new optimization method for the Optimal [4] L.F. Grisales-Noreña, D. Gonzalez Montoya, C.A. Ramos-Paja, Optimal sizing and
location of distributed generators based on pbil and pso techniques, Energies 11 (4)
Power Flow (OPF) problem in Direct Current (DC) networks using a (2018) 1018.
master-slave methodology. In the latter, the master stage is performed [5] O.D. Montoya, D.A. Giral-Ramírez, L.F. Grisales-Noreña, Optimal economic-
by the Arithmetic Optimization Algorithm (AOA), which evaluates environmental dispatch in mt-hvdc systems via sine-cosine algorithm, Results Eng.
13 (2022), 100348.
possible solutions to the OPF problem considering its constraints. Sub
[6] L.F. Grisales-Noreña, O.D. Montoya, W.J. Gil-González, A.-J. Perea-Moreno, M.-
sequently, the slave stage evaluates the fitness of the solutions to the A. Perea-Moreno, A comparative study on power flow methods for direct-current
problem (proposed by its master counterpart) and uses Successive Ap networks considering processing time and numerical convergence errors,
proximations (SA) to solve the OPF problem. Three scenarios, two sys Electronics 9 (12) (2020) 2062.
[7] L.F. Grisales-Noreña, O.D. Montoya, C.A. Ramos-Paja, An energy management
tems, and four other methods were implemented to validate the system for optimal operation of bss in dc distributed generation environments
proposed methodology. The three scenarios were defined to evaluate the based on a parallel pso algorithm, J. Energy Storage 29 (2020), 101488.
performance of the algorithms when the power injected by DGs was [8] P. Rault, X. Guillaud, F. Colas, S. Nguefeu, Investigation on Interactions between
Ac and Dc Grids, in: 2013 IEEE Grenoble Conference, IEEE, 2013, pp. 1–6.
limited to 20%, 40%, and 60% with respect to the power injected by the [9] O.D. Montoya, W. Gil-González, L. Grisales-Noreña, An exact minlp model for
slack node. The 21- and 69-bus test systems, frequently reported in the optimal location and sizing of dgs in distribution networks: a general algebraic
literature, were also employed for this validation. modeling system approach, Ain Shams Eng. J. 11 (2) (2020) 409–418.
[10] O.D. Montoya, A convex opf approximation for selecting the best candidate nodes
The results allow us to conclude that the AOA-SA method proposed for optimal location of power sources on dc resistive networks, Engineering Science
in this article is the best option to solve the OPF problem because it and Technology, Int. J. 23 (3) (2020) 527–533.
achieves the best standard deviation (as shown in the figures and tables [11] N. Chaudhuri, B. Chaudhuri, R. Majumder, A. Yazdani, Multi-terminal Direct-
Current Grids: Modeling, Analysis, and Control, John Wiley & Sons, 2014.
above) in the solution to the OPF problem. Thus, this method is the most [12] E. Prieto-Araujo, A. Egea-Alvarez, S. Fekriasl, O. Gomis-Bellmunt, Dc voltage droop
likely to produce an optimal solution in terms of power supplied by each control design for multiterminal hvdc systems considering ac and dc grid
generator and significant reductions in power losses. dynamics, IEEE Trans. Power Deliv. 31 (2) (2015) 575–585.
[13] L.F. Grisales-Noreña, D. Gonzalez Montoya, C.A. Ramos-Paja, Optimal sizing and
The figures also show that the reductions in power losses obtained by
location of distributed generators based on pbil and pso techniques, Energies 11 (4)
the methods do not present significant differences. However, if we (2018) 1018.
analyze them in terms of kilowatts or megawatts, it is evident that the [14] L. F. Grisales-Noreña, O. D. Garzón Rivera, J. A. Ocampo Toro, C. A. Ramos-Paja,
most efficient method, that is, the proposed method, can reduce power M. A. Rodriguez Cabal, Metaheuristic Optimization Methods for Optimal Power
Flow Analysis in Dc Distribution Networks.
losses, thus generating significant savings. The tables indicate that this [15] J. Li, F. Liu, Z. Wang, S.H. Low, S. Mei, Optimal power flow in stand-alone dc
method (i.e., AOA-SA) also kept the voltage and current within their microgrids, IEEE Trans. Power Syst. 33 (5) (2018) 5496–5506.
limits, which is a good indicator of its performance and also proves that [16] O.D. Montoya, W. Gil-González, A. Garces, Sequential quadratic programming
models for solving the opf problem in dc grids, Elec. Power Syst. Res. 169 (2019)
the constraints of the OPF problem were respected. 18–23.
[17] O. Montoya, W. Gil-González, L. Grisales-Noreña, Optimal Power Dispatch of Dgs
Credit author statement in Dc Power Grids: A Hybrid Gauss-Seidel-Genetic-Algorithm Methodology for
Solving the Opf Problem.
[18] A.A. Rosales-Muñoz, L.F. Grisales-Noreña, J. Montano, O.D. Montoya, A.-J. Perea-
conceptualization, methodology, formal analysis, investigation, and Moreno, Application of the multiverse optimization method to solve the optimal
writing-original draft preparation, Jhon Montano, Oscar Daniel Garzon, power flow problem in direct current electrical networks, Sustainability 13 (16)
(2021) 8703.
and Luis Fernando Grisales-Noreña. Writing-original draft preparation [19] J. Giraldo, O. Montoya, L. Grisales-Noreña, W. Gil-González, M. Holguín, Optimal
Jhon Montano, Oscar Daniel Garzón, Andres Alfonso Rosales Muñoz, power flow solution in direct current grids using sine-cosine algorithm, in: Journal
Luis Fernando Grisales-Noreña, and Oscar Danilo Montoya. of Physics: Conference Series vol. 1403, IOP Publishing, 2019, 012009.
[20] O.S. Velasquez, O.D. Montoya Giraldo, V.M. Garrido Arevalo, L.F. Grisales Norena,
Optimal power flow in direct-current power grids via black hole optimization, Adv.
Declaration of competing interest Electric. Electron. Eng. 17 (1) (2019) 24–32.
[21] O.D. Montoya, L.F. Grisales-Noreña, D. González-Montoya, C. Ramos-Paja,
A. Garces, Linear power flow formulation for low-voltage dc power grids, Elec.
The authors declare that they have no known competing financial
Power Syst. Res. 163 (2018) 375–381.
interests or personal relationships that could have appeared to influence [22] L. Abualigah, A. Diabat, S. Mirjalili, M. Abd Elaziz, A.H. Gandomi, The arithmetic
the work reported in this paper. optimization algorithm, Comput. Methods Appl. Mech. Eng. 376 (2021), 113609.
[23] O.D. Montoya, V.M. Garrido, W. Gil-González, L.F. Grisales-Noreña, Power flow
analysis in dc grids: two alternative numerical methods, IEEE Transact. Circuits
Acknowledgments Syst. II: Express Briefs 66 (11) (2019) 1865–1869.
[24] A. Garcés, On the convergence of Newton’s method in power flow studies for dc
This research and the APC were funded by Minciencias through the microgrids, IEEE Trans. Power Syst. 33 (5) (2018) 5770–5777.
11
J. Montano et al. Results in Engineering 16 (2022) 100654
[25] O.S. Velasquez, O.D. Montoya Giraldo, V.M. Garrido Arevalo, L.F. Grisales Norena, [27] S. Sultana, P.K. Roy, Krill herd algorithm for optimal location of distributed
Optimal power flow in direct-current power grids via black hole optimization, Adv. generator in radial distribution system, Appl. Soft Comput. 40 (2016) 391–404.
Electric. Electron. Eng. 17 (1) (2019) 24–32. [28] W. Gil-González, O.D. Montoya, E. Holguín, A. Garces, L.F. Grisales-Noreña,
[26] S. Kaur, G. Kumbhar, J. Sharma, A minlp technique for optimal placement of Economic dispatch of energy storage systems in dc microgrids employing a
multiple dg units in distribution systems, Int. J. Electr. Power Energy Syst. 63 semidefinite programming model, J. Energy Storage 21 (2019) 1–8.
(2014) 609–617.
12