Journal of Energy Storage
Journal of Energy Storage
Research papers
Keywords: This study addresses the problem of optimal operation of batteries in standalone and grid-connected
Energy management system Direct Current (DC) Microgrids (MGs) that include photovoltaic (PV) generators operating at maximum power
Batteries point. For that purpose, a mathematical model was formulated considering three objective functions: (1) the
DC microgrids
minimization of operating costs, (2) the reduction of energy losses associated with energy transport in DC
Parallel processing
MGs, and (3) the minimization of the total emissions of CO2 into the atmosphere produced by conventional
Metaheuristic methods
Variable power demand
generators. The model integrates a set of constraints that represent the operation of DC microgrids. Three
Variable renewable generation parallel versions of well-known solution methodologies were used here: Parallel Particle Swarm Optimization
Particle swarm optimization (PPSO), the Parallel Vortex Search Algorithm (PVSA), and the Parallel Ant Lion Optimizer (PALO). These
Vortex search algorithm methodologies were implemented to optimize the hourly power flow method based on successive approxi-
Ant lion optimizer mations and evaluate the objective functions and constraints. To validate the effectiveness of the solution
methods proposed in this paper, two test systems were used: a standalone network and a grid-connected
network (both in Colombia). Furthermore, to test the effectiveness of the proposed energy management system
(regarding solution, repeatability, and processing times) each solution methodology was executed 100 times
in each test system using MATLAB software. Based on the results of the simulations, the methodology with
the best performance for standalone grids was the PVSA. In turn, the methodology that achieved the best
results in the grid-connected network was the PALO. In both scenarios, the best average reductions in energy
fixed costs, variable costs, energy losses, and CO2 emissions were 0.0282%, 1.3039%, 9.1655%, and 0.1254%,
respectively. Although these were important benefits in technical and environmental aspects, the results of the
economic indicator suggest that the operation of electrical networks should consider variable costs in order to
improve the impact of energy management systems in the future.
∗ Corresponding author.
E-mail addresses: [email protected] (L.F. Grisales-Noreña), [email protected] (J.A. Ocampo-Toro), [email protected]
(O.D. Montoya-Giraldo), [email protected] (J. Montano), [email protected] (J.C. Hernandéz).
https://doi.org/10.1016/j.est.2023.107240
Received 28 December 2022; Received in revised form 18 March 2023; Accepted 23 March 2023
Available online 5 April 2023
2352-152X/© 2023 Elsevier Ltd. All rights reserved.
L.F. Grisales-Noreña et al. Journal of Energy Storage 65 (2023) 107240
Nomenclature 𝑆𝑂𝐶𝑖𝑚𝑖𝑛 and 𝑆𝑂𝐶𝑖𝑚𝑎𝑥 Maximum and minimum SOC allowed for
the battery at bus 𝑖.
𝛽 Normalized factor to regulate the penalty
𝑆𝑂𝐶0𝑖 Initial state of charge of the batteries
for the objective function.
installed at bus 𝑖.
𝛥𝑡 Duration of the time period 𝑡.
𝑆𝑂𝐶𝑓 𝑖 Final state of charge of the batteries in-
𝜇 𝑖𝑡𝑒𝑟+1 New center of the hyper-sphere in the
stalled at bus 𝑖.
PVSA.
𝑡𝑐 and 𝑡𝑑 Charging and discharging times of each
𝜇 𝑖𝑡𝑒𝑟 Center of the hyper-sphere in the PVSA.
battery.
𝐴𝑛𝑡𝑙𝑖𝑜𝑛𝑖𝑡𝑒𝑟+1 New best objective function as the ant lion
𝑡𝑐𝐵𝑖 and 𝑡𝑑𝐵𝑖 Charging and discharging times of a battery
in the new iteration.
installed at bus 𝑖.
𝐴𝑛𝑡𝑙𝑖𝑜𝑛𝑖𝑡𝑒𝑟 Best objective function as the ant lion in the
𝑉𝑖𝑚𝑖𝑛 and 𝑉𝑖𝑚𝑎𝑥 Minimum and maximum allowable voltages
current iteration.
at bus 𝑖.
𝐸 (𝑡) Reduction of CO2 emissions related to the
𝛾𝐶𝐺𝑖 CO2 factor emission related to the conven-
operation of conventional generators at
tional generators at bus 𝑖.
time 𝑡.
𝜙𝐵𝑖 Charge/discharge factor of the battery in-
𝐹𝐹 Fitness function.
stalled at bus 𝑖.
𝐻 Set of time periods.
𝐶𝐵𝑖 Nominal capacity of a battery at node 𝑖.
𝐻𝑃 𝐹 𝑆𝐴 Hourly power flow method based on SA.
𝐶𝐶𝐺𝑖 (𝑡) Energy purchasing cost of the conventional
𝐼𝑖𝑗𝑚𝑎𝑥 Maximum electric current allowed for the
generators located at bus 𝑖 at time 𝑡.
line between buses 𝑖 and 𝑗.
𝐶𝐸𝐸 (𝑡) Electric energy purchasing cost in the DC
𝐼𝑚𝑎𝑥 Maximum current limits in each line.
MG at time 𝑡.
𝐼𝑛𝑐𝑢𝑚𝑏𝑒𝑛𝑡𝑖𝑡𝑒𝑟+1 New incumbent, updated best solution in
𝐶𝑀𝐵𝑖 (𝑡) Maintenance cost of the batteries (energy
the population.
storage systems) installed in the DC MG at
𝐼𝑛𝑐𝑢𝑚𝑏𝑒𝑛𝑡𝑖𝑡𝑒𝑟 Current incumbent, best solution in the time 𝑡.
population.
𝐶𝑀𝐷𝐺𝑖 (𝑡) Maintenance cost of the PV DGs installed in
𝑖𝑡𝑒𝑟𝑁𝐼 Counter of non-improvement iterations. the DC MG at time 𝑡.
𝑖𝑡𝑒𝑟𝑁𝐼
𝑚𝑎𝑥 Maximum allowed number of 𝐶𝑀 (𝑡) Total maintenance cost of the DC MG at
non-improvement iterations. time 𝑡.
𝑖𝑡𝑒𝑟𝑚𝑎𝑥 Maximum number of iterations. 𝐶 (𝑡) Total operating cost of the DC MG at time 𝑡.
𝑖𝑡𝑒𝑟𝑁𝑚𝑎𝑥 𝐼 Maximum number of non-improvement 𝐺𝑖0 Constant resistance loads included in the
iterations. matrix, which do not produce power losses
𝑀𝑃 𝑇 Maximum required processing time. associated with energy transport.
𝑁 Number of individuals in the population. 𝐺𝑖𝑗 Component of the conductance matrix of
𝑁 Set of buses in the DC MG. the lines that interconnect buses 𝑖 and 𝑗.
𝑁𝐵 Number of batteries installed in the DC MG. 𝐼𝑖𝑗 (𝑡) Electric current that flows through the line
𝑁𝐶 Number of cores in the CPU of the that connects buses 𝑖 and 𝑗, at time 𝑡.
computer. 𝑃𝐶𝐺𝑖 (𝑡) Power produced by the conventional gener-
𝑁𝑃 Number of processes required. ator located at bus 𝑖 at time 𝑡.
𝑃𝑏 Power capacity of each battery. 𝑃𝑑𝑖 (𝑡) Power demanded by the load connected at
𝑃𝑗 Power demanded at bus 𝑗 related to the bus 𝑖 at time 𝑡.
load connected to this bus. 𝑃𝑙𝑜𝑠𝑠 (𝑡) Power losses in the DC MG at time 𝑡.
𝑃𝐵𝑖 (𝑡) Power stored by the batteries located at bus 𝑅𝑖𝑗 Electrical resistance of the connection line
𝑖 at time 𝑡. between buses 𝑖 and 𝑗.
𝑐ℎ𝑎𝑟_𝑚𝑎𝑥
𝑃𝐵𝑖 Maximum charging power allowed for the 𝑆𝑂𝐶𝐵𝑖 (𝑡) State of charge of the battery at bus 𝑖 and
battery located at bus 𝑖. time 𝑡.
𝑑𝑖𝑠𝑐ℎ_𝑚𝑎𝑥
𝑃𝐵𝑖 Maximum discharging power allowed for 𝑉𝑖 (𝑡) and 𝑉𝑗 (𝑡) Voltages at buses 𝑖 and 𝑗 at time 𝑡,
the battery located at bus 𝑖. respectively.
𝑚𝑖𝑛 and 𝑃 𝑚𝑎𝑥
𝑃𝐶𝐺𝑖 Minimum and maximum power genera-
𝐶𝐺𝑖 HPF Hourly power flow.
tion capacities allowed in the different
Acronyms
conventional generators.
𝑃𝐷𝐺𝑖 (𝑡) Power generated by the PV DG located at DC MG Direct current microgrid
bus 𝑖 at time 𝑡. PV Photovoltaic energy
𝑚𝑎𝑥 (𝑡) and 𝑃 𝑚𝑖𝑛 (𝑡) Maximum and minimum power limits for
𝑃𝐷𝐺𝑖 𝐷𝐺𝑖 BSS Battery storage system
the PV distributed generators at time 𝑡. PPSO Parallel particle swarm optimization
𝑃𝐷𝐺 Power rating of each PV DG. PVSA Parallel vortex search algorithm
𝑃𝐹 Penalty factor for the objective function. PALO Parallel ant lion optimizer
𝑟𝑖𝑡𝑒𝑟+1 New radius of the hyper-sphere in the DG Distributed generator
PVSA. RES Renewable energy source
𝑟𝑖𝑡𝑒𝑟 Radius of the hyper-sphere in the PVSA. ES Energy storage
𝑅𝑖𝑗 Electrical resistance of the line that con- DER Distributed energy resource
nects buses 𝑖 and 𝑗.
2
L.F. Grisales-Noreña et al. Journal of Energy Storage 65 (2023) 107240
3
L.F. Grisales-Noreña et al. Journal of Energy Storage 65 (2023) 107240
and operating limits while improving its technical, economic, and their solution, standard deviation, or processing times), they do provide
environmental conditions [23]. This is known as the Optimal Power important information about the integration of hybrid energy storage
Flow (OPF) problem. In this strategy, a power flow method should systems and the use of energy optimization techniques to enhance the
be selected to evaluate the objective function and the constraints system’s performance.
of each power configuration assigned by the proposed OPF to the In [37], the authors discuss a planning strategy for a Combined
batteries. Among the methods reported in the literature to solve the Cooling, Heat and Power (CCHP) system integrated with an electric
power flow problem in DC MGs, the most efficient option is successive vehicle charging station in a residential energy hub. Using optimization
approximations (SA) [24] because it achieves excellent convergence techniques including mixed-linear and non-linear programming, they
in short times. To solve the power flow every hour, the SA has been implemented a two-stage programming model to reduce the system’s
adapted as the Hourly Power Flow (HPF) method, which is widely operating costs and CO2 emissions. Compared to other programming
used in literature and was selected in this paper to solve the OPF methods, the results show that their strategy may produce significant
problem [6,25]. This methodology is explained in Section 3 of this savings in operating costs and reduce CO2 emissions. Nevertheless, they
paper. did not take into account variations in renewable energy generation
To solve the OPF problem, other authors have implemented ex- and climatic conditions, which might have an impact on the system’s
act methods based on specialized software. These methods stand out performance in the current global scenario.
due to their accuracy in terms of convergence [2]. For example, Gil Another study [38] solved the problem of optimal operation of
et al. [26] proposed a convex optimization method based on semidef- a DC microgrid using a predictive control technique. The objective
inite programming that operates batteries and renewable energy DGs functions in that case aimed at the reduction of three variables: energy
in a DC MG to reduce operating costs and energy losses. This method costs, power losses, and CO2 emissions. Nevertheless, said study did
presented excellent results compared to other methodologies. However, not compare its technique with other solution methodologies reported
the processing times it required were not analyzed. Furthermore, the in literature. Furthermore, it did not analyze the average solution or
maintenance costs of the DERs were not included in the proposed processing times.
mathematical model. The same problem occurs in [27], where the
Two papers [38,39] have analyzed the problem of optimal operation
authors proposed a convex mathematical model based on second-order
of DC microgrids that include different renewable energy sources and
cone programming for DC MGs to operate multiple batteries installed
batteries. They have employed different optimization techniques to
in a grid that includes renewable generation. In that case, the objec-
reduce energy costs and CO2 emissions. All these optimization methods
tive function was the reduction of operating costs and energy losses.
have obtained excellent results in the reduction of the objective func-
Like these methods, multiple energy managements systems based on
tions. However, these papers did not analyze the average solution or
specialized software have been reported in the last year. They use
processing times.
several kinds of programming: linear [23], mixed linear [28], and
Aghdam et al. [40] presented an optimization methodology for
quadratic [29], among others. However, the acquisition and imple-
the optimal programming of a virtual energy storage system in a re-
mentation of these solution methodologies increase the complexity and
configurable distribution network. The optimization technique they
costs of energy management systems [6].
used is called ‘‘linear mixed-entry programming’’, which has proven to
In [22,30–32], the authors present exhaustive reviews of microgrids
be successful in reducing congestion in distribution networks. Although
and highlight the significance of optimization and control techniques
they analyzed the average impact of the proposed solutions on the
for ensuring the stability of the latter. They provide a broad overview
objective function, they did not consider the standard deviation or
of the several approaches that have been adopted in the literature in
this field but do not go into the technical details of each technique or processing times of those solutions.
provide specific validation data. Instead, their articles discuss existing In recent years, solutions based on sequential programming methods
and future challenges for intelligent energy management in electrical (such as metaheuristic optimization methods) have been employed in
networks. They also stress the importance of energy managements sys- the literature to find a solution to the OPF problem in MGs where
tems based on sequential programming to avoid the implementation of DG PV panels are the main distributed energy resources and batteries
specialized software, thus reducing the acquisition and implementation are employed for ES. Metaheuristic optimization methods are widely
complexity of conventional solution methodologies. used as solution methodologies in this case due to their excellent per-
Other authors [33,34] proposed an energy management system formance and reduced complexity in solving non-linear problems with
for electric vehicles based on a consumption prediction model and continuous variables. For example, other authors have implemented the
an optimized power control strategy. The main objectives of their genetic algorithm (GA) [41], particle swarm optimization (PSO) [42],
system were to reduce electricity costs and improve the vehicle’s energy the ant colony algorithm, and the bee colony algorithm [43], among
efficiency. Although their results showed a significant decrease in elec- others.
tricity consumption, their study did not take into account the effects of Several studies have specifically examined the OPF problem in DC
ambient temperature, provide specific information on the optimization MGs that include batteries as their ES system. For instance, Wang
techniques they employed, or include a thorough evaluation of the et al. [21] used a PSO algorithm to solve the OPF in DC networks
electrical system. that consider DG and batteries. Their objective function was the re-
Different researchers [35,36] have used energy management tech- duction of energy costs while satisfying all the technical constraints
niques to operate hybrid AC/DC microgrids. To solve the OPF prob- that represent DC grids in a scenario of PV DGs and ES. Although they
lem, they have established the optimal operation of battery storage examined different test cases, they did not compare methodologies or
systems using different optimization methods: stochastic modeling, processing times, which means that an adequate performance analysis
mathematical programming, droop control, sliding mode control, rule- is not possible. Likewise, in [44], the authors presented a non-linear
base control, deadbeat control, model predictive control, fuzzy logic optimization model for the optimal operation of PV DGs, wind DGs,
control, optimization-based control, predictive control, and simulation. and batteries considering constant power loads. However, they did not
Their studies offer a critical assessment of several energy management consider the uncertainties of the primary sources, i.e., wind and solar
strategies and their success in improving energy efficiency, i.e., reduc- power generation. Consequently, the solutions they obtained do not
ing energy costs and CO2 emissions. Although they do not explicitly represent a real scenario, and the results are not representative of the
detail the performance of these optimization techniques (in terms of regions under study.
4
L.F. Grisales-Noreña et al. Journal of Energy Storage 65 (2023) 107240
Other researchers have investigated the optimal operation of bat- been as remarkable as in their AC counterparts [6]. Specifically, the
teries in MGs using solution methods such as multi-agent optimization topic of battery energy management in DC MGs considering RESs
based on market decisions, multi-objective PSO, bee colony optimiza- installed in the grid (mainly PV DGs) has not been sufficiently studied.
tion, and the GA [45,46]. In [6], the authors proposed a master–slave Therefore, currently, it is necessary to propose mathematical models
methodology. The master stage employed a continuous version of the and optimization strategies that consider variable power generation
genetic algorithm, the Monte Carlo method, and a parallel version of and demand, as well as all the constraints that describe the operation
the particle swarm optimization algorithm to solve the operation of of DC MGs under a DER scenario. These models should include limits
multiple batteries located in a DC microgrid that considers PV DGs for the line currents, which have been neglected in all the studies
operating under MPP conditions. The slave stage used the HPF method that address ES reviewed above. Furthermore, they must work with
based on SA for evaluating the objective functions and constraints grid-connected and standalone networks because most research so far
that represent the problem. In that case, the objective function was has focused only on one kind of electrical network. Finally, they must
composed of energy purchasing costs and energy losses. Their results implement the technical, economic, and environmental conditions of
demonstrate the effectiveness of Parallel Particle Swarm Optimization the grid in the objective function to have an impact on all these fronts,
(PPSO) in solving the problem they addressed in terms of average which is something smart grids do.
solution, standard deviation, and processing time. However, as the Energy management systems should provide a solution in a short
mathematical model they used did not consider the maintenance costs processing time in order to be able to analyze the largest possible num-
of the DERs, the solutions they obtained are not representative of real ber of scenarios during variations in power generation and demand in
life scenarios. electric grids [52]. In addition, new solution methodologies should be
In the literature, the number of studies conducted to improve the compared to those reported in literature to validate their effectiveness
environmental conditions of electrical networks (by means of the op- in terms of average solution and standard deviation, which guarantee
timal operation of the DERs that compose them) is limited compared that, every time the method is executed, the solution obtained will be
to those that have proposed energy management systems to improve as good as that reported by the authors.
their technical and economical conditions. In relation to improving the To solve the problems previously described, this paper formulates
environmental conditions of DC grids, a paper [47] proposed a strategy a mathematical model to reduce the operating cost, power losses,
for reducing gas emissions due to fuel consumption in a DC grid. The and CO2 emissions of DC MGs under a scenario of DERs (PV DGs
proposed control strategy reduced the speed of the generators installed + batteries), while considering all the technical and operating con-
in the network. In [48], the theta-crow search algorithm was used to straints that represent this problem. Furthermore, this paper proposes
solve the problem of optimal operation of ES in AC/DC grids. The an energy management system that implements a parallel version of
objective functions were the reduction of the operating cost and the a metaheuristic optimization technique – i.e., Parallel Particle Swarm
minimization of CO2 emissions. The excellent performance of the pro- Optimization (PPSO), the Parallel Vortex Search Algorithm (PVSA),
posed methodology was demonstrated in a test system. Nevertheless, no or the Parallel Ant-Lion Optimizer (PALO) – to solve the problem of
other methods were used for comparison, and the average solution and optimal operation of battery storage systems in a DC MG that includes
processing times were not analyzed. Currently, only a few published PV DGs operating at their maximum power point. These three solution
studies have aimed to improve the environmental conditions of DC methodologies were selected because they have achieved excellent
results in solving the OPF problem for electrical networks. In particular,
grids, and the existing ones have focused on reducing CO2 emissions.
the parallel processing added to the traditional methods seeks to reduce
Table 1 summarizes the state of the art of the field examined in this
their processing times., this paper formulates a mathematical model
study. It details (from left to right) several characteristics of the solution
to reduce the operating cost, power losses, and CO2 emissions of DC
methodologies that have been proposed in the literature: reference, op-
MGs under a scenario of DERs (PV DGs + batteries), while considering
timization technique(s), methodology/test system employed, objective
all the technical and operating constraints that represent this problem.
function, and type of performance validation (i.e., minimum results,
Furthermore, this paper proposes an energy management system that
average results, standard deviation, or processing times). This table
implements a parallel version of a metaheuristic optimization technique
indicates that the topic addressed in this paper is relevant and current.
– i.e., Parallel Particle Swarm Optimization (PPSO), the Parallel Vortex
Furthermore, it shows that economic, technical, and environmental
Search Algorithm (PVSA), or the Parallel Ant-Lion Optimizer (PALO) –
indicators have been widely implemented as objective functions. In
to solve the problem of optimal operation of battery storage systems
particular, optimization methods based on sequential programming
in a DC MG that includes PV DGs operating at their maximum power
have been the most commonly used in recent years. Finally, it is
point. These three solution methodologies were selected because they
possible to see that, currently, the performance of the new solution
have achieved excellent results in solving the OPF problem for electrical
methodologies should be evaluated by analyzing their minimum and
networks. In particular, the parallel processing added to the traditional
average results, standard deviation, and processing times in order to
methods seeks to reduce their processing times.
obtain excellent quality solutions in terms of repeatability and short
This study implemented two test scenarios: a grid-connected DC MG
processing times.
and a stand-alone DC MG. They used the data about power genera-
tion, power demand, energy cost, and CO2 emissions from two actual
1.3. Scope and main contributions
networks: a grid-connected MG in Medellín (Antioquia, Colombia) and
standalone MG in Capurganá (Chocó, Colombia), respectively. In order
Although the studies reviewed in the state of the art above have to evaluate the effectiveness and robustness of the solution methodolo-
demonstrated a good performance of the methods they have used, gies proposed here, each optimization method was executed 100 times
the number of scenarios they employed for comparison was limited, to analyze the best solution, average solution, standard deviation, and
and, in most cases, they did not analyze the average solutions and required processing times.
processing times required by the algorithms. Furthermore, an energy The main academic contributions and future industrial applications
management system that is highly efficient (in terms of solution quality, of this study are presented below:
repeatability, and processing times) should improve the most com- Academic contributions:
monly used technical, economic, and environmental indicators, i.e., it
should reduce energy costs, power losses, and CO2 emissions [10,28, • A mathematical model that represents the problem of optimal
49–51]. operation of batteries in grid-connected and standalone DC grids
In the literature reviewed above, it was found that the exploration to improve their technical, economic, and environmental condi-
and progress of energy management strategies in DC MGs has not tions by considering all technical and operating constraints that
5
L.F. Grisales-Noreña et al. Journal of Energy Storage 65 (2023) 107240
Table 1
Summary of recent literature about techniques, objective functions, and performance validation of energy management systems for DC microgrids.
Ref. Optimization techniques Methodology/ Test system Objective function Validation
[19] – Literature review – –
[20] – Proposed hierarchical control for AC – –
and DC microgrids
[6] PSO Test system based on the IEEE Reduction of the energy cost of Average and standard deviation
33-node system the microgrid of results
[21] Brute Force Algorithm Test system with different generation Minimization of the total annual Computation time
and load combinations operating cost
[22] Evolutionary Algorithms Literature review – –
[23] – Literature review – –
[24] Newton–Raphson Method Test system with two DC circuits in Power flow analysis in DC grids –
and Fixed-Point Method parallel
[25] PBIL, PSO Test system based on the IEEE Minimization of the total annual Average and standard deviation
33-node network operating cost of results
[2] – Literature review – –
[26] Semidefinite Programming Test system based on the IEEE-123 Minimization of the operating Computation time
Model node network cost
[27] Second-Order Cone Test system with different generation Minimization of the operating Computation time
Optimization Algorithm and load combinations cost
[28] Brute Force Algorithm Testing system with a distribution Minimization of the operating Average and standard deviation
network in Cauca, Colombia cost of results
[29] Mixed-Integer Linear Test system with different generation Minimization of the operating Computation time
Programming Algorithm and load combinations cost
[30] – Literature review – –
[31] – Literature review – –
[32] – Literature review – –
[33] PSO, ACO, GA Experimentation Minimization of a vehicle’s Average results
electricity consumption cost
[34] Supervised Learning Experimentation, heat demand Minimization of the annual Minimum results
Method forecasting operating cost and CO2 levels
[35] – Literature review – –
[36] GA Experimentation, stochastic Minimization of the annual Average results
scheduling operating cost and CO2 levels
6
L.F. Grisales-Noreña et al. Journal of Energy Storage 65 (2023) 107240
represent the problem, i.e., power generation limits; maximum • Reduction of the energy purchasing cost of conventional gener-
and minimum charging and discharging power in the batteries; ators and maintenance costs of PV DGs: Reduction of operating
state-of-charge limits for the batteries; initial and final state of costs (𝐸𝑐𝑜𝑠𝑡).
charge; and limits for the voltage and current in the lines of • Reduction of power losses associated with energy transport in DC
the electrical network. Thanks to its general formulation, the MGs: Reduction of power losses (𝐸𝑙𝑜𝑠𝑠).
same mathematical model can work with grid-connected and • Reduction of CO2 emissions related to conventional power gener-
standalone DC grids. ators: Reduction of CO2 emissions (𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠).
• New parallel versions of two metaheuristic algorithms, i.e., the
ALO and the VSA, to solve the problem of optimal power dispatch Considering these objective functions, the following subsection
of batteries in DC grids. presents the mathematical set of equations developed in this study for
• Two test systems that use data (about energy cost, CO2 emissions, the optimal operation of battery systems in DC MGs.
power demand, and PV generation) from two actual DC MGs
(one grid-connected and one standalone) located in two different 2.1. Reduction of operating costs
regions of Colombia. They could be used as baselines to evaluate
energy management systems in grid-connected and standalone Eq. (1) describes the expression used here to calculate the energy
microgrids in other places around the world.
purchasing costs for the DC MG in each period of time 𝑡 (𝐶𝐸𝐸 (𝑡)). In this
• A methodology based on average results and standard deviation
equation, the energy purchasing cost of the conventional generators
values for selecting the strategy with the best performance in
located at bus 𝑖 (𝐶𝐶𝐺𝑖 (𝑡)) can take two different values depending on
solving the problem of optimal operation of batteries in DC grids
the kind of electrical grid. In grid-connected MGs, this cost includes
while improving their technical, economic, and environmental
the price of purchasing power from the electrical grid to supply the
conditions.
MG at bus 𝑖. This cost is reported by the local electricity distribution
• A simulation that demonstrates that, to improve the techni-
company, and it could be fixed or variable. In turn, in MGs that operate
cal, economic, and environmental conditions of standalone net-
in standalone mode, the energy purchasing cost is calculated based on
works, the operation of the batteries should be based directly
the cost of generating electricity using a diesel generator located at
on the power demand and PV generation curves. In turn, in
bus 𝑖. This electricity cost is fixed because the diesel price remains the
grid-connected networks, such operation should consider power
same during the entire day of operation. In Eq. (1), variable 𝑃𝐶𝐺𝑖 (𝑡)
demand, energy cost curves, and PV generation curves.
represents the power produced by the conventional generator located
Industrial applications: at bus 𝑖 in the period of time 𝑡, and 𝑁 denotes the set of buses in the
DC MG. Eq. (2) presents the expression used here for calculating the
• A mathematical model that represents all the technical and oper- total maintenance costs of PV DGs and batteries installed in the DC
ating constraints of DC grids. Its objective functions are based on grids (𝐶𝑀 (𝑡)). In this equation, 𝐶𝑀𝐷𝐺𝑖 (𝑡) and 𝐶𝑀𝐵𝑖 (𝑡) represent the
the most relevant economic, technical, and environmental factors maintenance cost of the PV DGs and that of the battery energy storage
for grid users, owners, and operators at present. systems installed in the DC MG, respectively. Since maintenance costs
• Identification of the most effective methodologies for solving are calculated based on the power management performed by the DERs,
the problem of optimal operation of batteries in grid-connected this equation should consider two variables: 𝑃𝐷𝐺𝑖 (𝑡) and 𝑃𝐵𝑖 (𝑡). They
and standalone DC microgrids considering the reduction of their describe the power generated by the PV DGs and that stored by the
operating costs, power losses, and CO2 emissions. batteries located at bus 𝑖 in period of time 𝑡, respectively. The total
• Energy management systems with processing times under two operating costs of the DC MG, i.e., 𝐶 (𝑡), are obtained from the sum of
minutes that enable the grids to make smart decisions based on the electricity purchasing costs and maintenance costs of the DC MG
solar radiance and power demand. (see Eq. (3)) in each period of time. The objective function for the
minimization of operating costs in the 24-hour mathematical model,
1.4. Paper organization i.e., 𝑂𝐹 1, is expressed in Eq. (4), which was adapted from [55].
∑( )
𝐶𝐸𝐸 (𝑡) = 𝐶𝐶𝐺𝑖 (𝑡) ∗ 𝑃𝐶𝐺𝑖 (𝑡) (1)
This paper is organized as follows. Section 2 presents the math- 𝑖∈𝑁
ematical formulation proposed in this study. Section 3 describes the ∑( )
solution methodologies proposed for solving the problem formulated 𝐶𝑀 (𝑡) = 𝐶𝑀𝐷𝐺𝑖 ∗ 𝑃𝐷𝐺𝑖 (𝑡) + 𝐶𝑀𝐵𝑖 ∗ 𝑃𝐵𝑖 (𝑡) (2)
𝑖∈𝑁
in Section 2. Section 4 details the electrical parameters and operating
conditions of the grid-connected and the standalone microgrids used 𝐶 (𝑡) = 𝐶𝐸𝐸 (𝑡) + 𝐶𝑀 (𝑡) (3)
as test scenarios. Section 5 reports and analyzes the simulation results
obtained by the proposed solution methodologies in the test scenarios. ( 24 )
∑
Finally, Section 6 draws the conclusions and proposes future work in 𝑂𝐹 1 = 𝑚𝑖𝑛 𝐶 (𝑡) (4)
this field. 𝑡=1
This section formulates a mathematical model to solve the problem Eq. (5) is used to calculate the power losses (𝑃𝑙𝑜𝑠𝑠 (𝑡)) in the DC MG
of optimal energy management of batteries in DC MGs. The model uses in each time period 𝑡. In (5), 𝑃𝑑𝑖 (𝑡) represents the power demanded by
three different objective functions to improve the economic, technical, the load connected at bus 𝑖 in time period 𝑡. 𝑉𝑖 (𝑡) and 𝑉𝑗 (𝑡) are the
and environmental conditions of DC MGs, while incorporating an entire voltages at buses 𝑖 and 𝑗, respectively, in time period 𝑡. 𝐺𝑖𝑗 denotes
set of constraints that represents DC grids in a scenario that includes the component of the conductance matrix of the lines that connect
DERs [53,54]. The proposed mathematical model takes into account buses 𝑖 and 𝑗. Finally, 𝐺𝑖0 represents the constant resistance loads
variable PV generation and demand by users over a 24-hour operation included in the conductance matrix, which do not produce power losses
divided into one-hour periods. The three objective functions and their associated with energy transport [43]. The objective function for the
respective indices are detailed below: minimization of energy losses in the proposed 24-hour mathematical
7
L.F. Grisales-Noreña et al. Journal of Energy Storage 65 (2023) 107240
model, i.e., 𝑂𝐹 2, is expressed in Eq. (6), which was adapted from [56]. The power charging and discharging limits for the battery located at
bus 𝑖 are calculated using Eqs. (13) and (14), respectively. In these two
∑( ) equations, 𝐶𝐵𝑖 is the nominal capacity of a battery at node 𝑖; and 𝑡𝑐𝐵𝑖
𝑃𝑙𝑜𝑠𝑠 (𝑡) = 𝑃𝐶𝐺𝑖 (𝑡) + 𝑃𝐷𝐺𝑖 (𝑡) ± 𝑃𝐵𝑖 (𝑡) − 𝑃𝑑𝑖 (𝑡)
𝑖∈𝑁 and 𝑡𝑑𝐵𝑖 are the charging and discharging times of a battery installed
( ) (5)
∑ ∑ at bus 𝑖.
= 𝐺𝑖𝑗 ∗ 𝑉𝑖 (𝑡) ∗ 𝑉𝑗 (𝑡) − 𝐺𝑖0 ∗ 𝑉𝑖2 (𝑡)
𝑖∈𝑁 𝑗∈𝑁
( 24 ) 𝑐ℎ𝑎𝑟_𝑚𝑖𝑛 𝐶𝐵𝑖
∑ 𝑃𝐵𝑖 =− ∀𝑖 ∈ 𝑁 (13)
𝑡𝑐𝐵𝑖
𝑂𝐹 2 = 𝑚𝑖𝑛 𝑃𝑙𝑜𝑠𝑠 (𝑡) (6)
𝑡=1 𝑑𝑖𝑠𝑐ℎ_𝑚𝑎𝑥 𝐶𝐵𝑖
𝑃𝐵𝑖 = ∀𝑖 ∈ 𝑁 (14)
𝑡𝑑𝐵𝑖
2.3. Reduction of CO2 emissions
Eq. (15) represents the state of charge (SOC) of a battery installed
at bus 𝑖 in time period 𝑡. In this equation, 𝑆𝑂𝐶𝐵𝑖 (𝑡) denotes the SOC
The third objective function is the reduction of CO2 emissions
of a battery at bus 𝑖 and time 𝑡, 𝜙𝐵𝑖 is the charge/discharge factor of
related to the operation of conventional generators in each time period
the battery installed at bus 𝑖, and 𝛥𝑡 is the duration of time period 𝑡.
𝑡, i.e., 𝐸 (𝑡) [57]. This equation did not consider CO2 emissions by the
Eq. (16) can be used to calculate 𝜙𝐵𝑖 .
PV DGs because these energy resources are considered an emission-free ( )
technology. Although their construction generates some environmental 𝑆𝑂𝐶𝐵𝑖 (𝑡) = 𝑆𝑂𝐶𝐵𝑖 (𝑡 − 1) − 𝜙𝐵𝑖 ∗ 𝑃𝐵𝑖 (𝑡) ∗ Δ𝑡
(15)
impacts, their electricity generation process does not [58]. 𝛾𝐶𝐺𝑖 denotes ∀𝑖 ∈ 𝑁, ∀𝑡 ∈ 𝐻
the CO2 emission factor of conventional generators. This parameter
1 1
takes different values depending on the kind of energy resources in- 𝜙𝐵𝑖 = 𝑐ℎ𝑎𝑟_𝑚𝑖𝑛
= 𝑑𝑖𝑠𝑐ℎ_𝑚𝑎𝑥
∀𝑖 ∈ 𝑁, ∀𝑡 ∈ 𝐻 (16)
stalled at bus 𝑖. This particular study used the CO2 emission values 𝑡𝑐𝐵𝑖 ∗ 𝑃𝐵𝑖 𝑡𝑑𝐵𝑖 ∗ 𝑃𝐵𝑖
associated with diesel power generation. Eq. (7) is used to calculate the To improve the technical, economic, and environmental impacts of
CO2 emissions of the system in each time period over the time horizon the energy management systems proposed for the problem studied here,
analyzed here. other studies in the literature have fixed the initial (𝑆𝑂𝐶0𝑖 ) and final
∑
𝑛
( ) (𝑆𝑂𝐶𝑓 𝑖 ) states of charge of the batteries installed in the DC grid (see
𝐸 (𝑡) = 𝛾𝐶𝐺𝑖 ∗ 𝑃𝐶𝐺𝑖 (𝑡) (7) Eqs. (17) and (18)). In this paper, all the batteries start and finish their
𝑖=1 operation day with a 50% state of charge, following the suggestion
The objective function for the minimization of CO2 emissions, in [6]. Furthermore, Eq. (19) establishes the SOC limits for the batteries
i.e., 𝑂𝐹 3, is expressed in Eq. (8), which was reformulated from [55]. installed at bus 𝑖, where 𝑆𝑂𝐶𝑖𝑚𝑖𝑛 and 𝑆𝑂𝐶𝑖𝑚𝑎𝑥 are the minimum and
( 24 ) maximum SOC, respectively, allowed for the kind of batteries used in
∑
𝑂𝐹 3 = 𝑚𝑖𝑛 𝐸 (𝑡) (8) these MGs. As this study considered lithium-ion batteries, the values of
𝑡=1 these two parameters are 0.1 and 0.9, respectively [27].
The problem of optimal energy management of batteries in a DC 𝑆𝑂𝐶𝐵𝑖 (24) = 𝑆𝑂𝐶𝑓 𝑖 ∀𝑖 ∈ 𝑁 (18)
MG is limited by technical and operating conditions (i.e., constraints),
which are represented in Eqs. (9) to (21). These equations were for- 𝑆𝑂𝐶𝑖𝑚𝑖𝑛 ≤ 𝑆𝑂𝐶𝐵𝑖 (𝑡) ≤ 𝑆𝑂𝐶𝑖𝑚𝑎𝑥 ∀𝑖 ∈ 𝑁, ∀𝑡 ∈ 𝐻 (19)
mulated considering the work reported in [25,26]. As a result, they
include limits for the line currents. These are very important constraints Finally, Eqs. (20) and (21) guarantee the voltage profiles and cur-
in electrical networks, in spite of the fact that they are often neglected rent limits of the DC grid. In them, 𝑉𝑖𝑚𝑖𝑛 and 𝑉𝑖𝑚𝑎𝑥 denote the minimum
in many studies in the literature. and maximum allowable voltages at bus 𝑖; 𝑅𝑖𝑗 , the electrical resistance
The first constraint is the power balance (see Eq. (9)). This equation of the line connecting buses 𝑖 and 𝑗; 𝐼𝑖𝑗 (𝑡), the electric current that flows
must be satisfied for all the sets of buses and hours contained in 𝑁 and through the line that connects buses 𝑖 and 𝑗 at time 𝑡; and 𝐼𝑖𝑗𝑚𝑎𝑥 , the
𝐻. maximum electric current allowed for the line between buses 𝑖 and 𝑗.
𝑃𝐶𝐺𝑖 (𝑡) + 𝑃𝐷𝐺𝑖 (𝑡) ± 𝑃𝐵𝑖 (𝑡) − 𝑃𝑑𝑖 (𝑡)
∑
= 𝑉𝑖 (𝑡) ∗ 𝐺𝑖𝑗 ∗ 𝑉𝑗 (𝑡) ∀𝑖, 𝑗 ∈ 𝑁, ∀𝑡 ∈ 𝐻 (9) 𝑉𝑖𝑚𝑖𝑛 ≤ 𝑉𝑖 (𝑡) ≤ 𝑉𝑖𝑚𝑎𝑥 ∀𝑖 ∈ 𝑁, ∀𝑡 ∈ 𝐻 (20)
𝑗∈𝑁
| |
𝑚𝑖𝑛 ) and maximum (𝑃 𝑚𝑎𝑥 ) allowable
Eq. (10) fixes the minimum (𝑃𝐶𝐺𝑖 |𝑉𝑖 (𝑡) − 𝑉𝑗 (𝑡)|
𝐶𝐺𝑖 𝐼𝑖𝑗 (𝑡) = | | ≤ 𝐼 𝑚𝑎𝑥
𝑖𝑗 ∀𝑖, 𝑗 ∈ 𝑁, ∀𝑡 ∈ 𝐻 (21)
power generation capacities for the conventional generators located at 𝑅𝑖𝑗
the buses that compose the DC MG .
𝑚𝑖𝑛 𝑚𝑎𝑥
2.5. Fitness function
𝑃𝐶𝐺𝑖 ≤ 𝑃𝐶𝐺𝑖 (𝑡) ≤ 𝑃𝐶𝐺𝑖 ∀𝑖 ∈ 𝑁, ∀𝑡 ∈ 𝐻 (10)
The maximum (𝑃𝐷𝐺𝑖 𝑚𝑎𝑥 (𝑡)) and minimum (𝑃 𝑚𝑖𝑛 (𝑡)) power limits to be To satisfy all the constraints that represent the problem of optimal
𝐷𝐺𝑖
the supplied by the PV distributed generators installed in the DC grid operation of DC MGs previously described, this study used the Fitness
in time period 𝑡 are established by Eq. (11). Function (𝐹 𝐹 ) in Eq. (22). This equation in composed of the objective
function under analysis (𝑂𝐹 𝑖), a normalization factor to regulate the
𝑚𝑖𝑛 𝑚𝑎𝑥
𝑃𝐷𝐺𝑖 (𝑡) ≤ 𝑃𝐷𝐺𝑖 (𝑡) ≤ 𝑃𝐷𝐺𝑖 (𝑡) ∀𝑖 ∈ 𝑁, ∀𝑡 ∈ 𝐻 (11) penalty (𝛽), and a Penalty Factor (𝑃 𝐹 ) for the objective function (if
one of the proposed solutions violates some constraints in the different
The battery storage and supply capacity of the battery located at bus
𝑐ℎ𝑎𝑟_𝑚𝑎𝑥 𝑑𝑖𝑠𝑐ℎ_𝑚𝑎𝑥 time periods that are being examined). In this study, 𝛽 was calculated in
𝑖 at hour 𝑡 are fixed by Eq. (12). In this equation, 𝑃𝐵𝑖 and 𝑃𝐵𝑖
a heuristic way, thus obtaining a value of 1000. This value guarantees
denote the maximum charging and discharging powers, respectively,
that all the solutions obtained by the proposed optimization method-
allowed for the kind of battery located at bus 𝑖.
ologies are feasible at the end of the iterative process by allowing
𝑐ℎ𝑎𝑟_𝑚𝑎𝑥 𝑑𝑖𝑠𝑐ℎ_𝑚𝑎𝑥
𝑃𝐵𝑖 ≤ 𝑃𝐵𝑖 (𝑡) ≤ 𝑃𝐵𝑖 ∀𝑖 ∈ 𝑁, ∀𝑡 ∈ 𝐻 (12) them to explore non-feasible regions. This kind of exploration of the
8
L.F. Grisales-Noreña et al. Journal of Energy Storage 65 (2023) 107240
solution space reduces the processing times required by the solution 3.1. Problem codification
methods [58].
The problem of optimal operation of batteries in electrical networks
𝐹 𝐹 = 𝐹 𝑂𝑖 + 𝛽 ∗ 𝑃 𝐹 ∀𝑖 ∈ 1, 2, 3. (22)
requires a continuous codification that can be used to establish the
{ } power to be supplied or stored by the batteries in each period over
⎛ ∑ 𝑐ℎ𝑎𝑟_𝑚𝑎𝑥 ⎞ the time horizon under analysis. For that purpose, this study used a
⎜ 𝑚𝑎𝑥 0, (𝑃𝐵𝑖 (𝑡) − 𝑃𝐵𝑖 ) ⎟
⎜ ⎟ continuous codification with a vector of size 1x(𝑁𝐵 ∗ |𝐻|), where 𝑁𝐵
⎜ | {𝑖∈ }| ⎟
| ∑ | represents the number of batteries installed in the DC MG; and |𝐻|, the
⎜ |
+ |𝑚𝑖𝑛 0, 𝑑𝑖𝑠𝑐ℎ_𝑚𝑎𝑥
(𝑃𝐵𝑖 (𝑡) − 𝑃𝐵𝑖 ) || ⎟
⎜ ⎟ number of periods over the time horizon.
| |
⎜ | { 𝑖∈ }| ⎟ This codification is presented in Fig. 1. In the latter, for each
⎜ ∑ ⎟
⎜ 𝑚𝑎𝑥 0, (𝑆𝑂𝐶𝐵𝑖 (𝑡) − 𝑆𝑂𝐶 𝑚𝑎𝑥 ) ⎟ battery located in the grid, the codification proposes a state of charge
𝑖
⎜ ⎟ associated with each time period. Note that the SOC of the batteries
⎜ | {𝑖∈ }| ⎟
| ∑ | starts and ends at 50%, following the suggestion in the previous section
∑⎜
24
+ ||𝑚𝑖𝑛 0, (𝑆𝑂𝐶𝐵𝑖 (𝑡) − 𝑆𝑂𝐶 𝑚𝑖𝑛 ) |
|
⎟
𝑃𝐹 = ⎜ | 𝑖
| ⎟ (23) (i.e., Section 2). For example, the SOC of Battery 1 starts at 50%.
⎜ | { 𝑖∈ } | ⎟ Then, the battery is discharged down to 40% in the second hour; and,
𝑡=1
⎜ ∑ ⎟
𝑚𝑎𝑥
⎜ +𝑚𝑎𝑥 0, (𝑉𝑖 (𝑡) − 𝑉𝑖 ) ⎟ in the 23rd hour, its SOC is 40%. Finally, it is charged to end the
⎜ ⎟
⎜ | { 𝑖∈ }| ⎟
operation day with a 50% SOC. Following the same methodology, this
⎜ | ∑ | ⎟ codification suggests the hourly behavior of the SOC of Batteries 2 and
⎜ + ||𝑚𝑖𝑛 0, (𝑉𝑖 (𝑡) − 𝑉𝑖𝑚𝑖𝑛 ) || ⎟
⎜ | | ⎟
k. This methodology can be used to solve the operation of batteries
| 𝑖∈ |
⎜ ⎧ ⎫ ⎟ installed in DC MGs as a function of their SOC. Note that, when the
⎜ ⎪ ∑ ∑ ⎪ ⎟ optimization methodologies propose a new solution in their iterative
⎜ +𝑚𝑎𝑥 ⎨0, (𝐼𝑖𝑗 (𝑡) − 𝐼𝑖𝑗𝑚𝑎𝑥 )⎬ ⎟
⎜ ⎪ 𝑖∈ 𝑗∈ ⎪ ⎟ process, it is necessary to guarantee that the SOC proposed for each
⎝ ⎩ ⎭ ⎠ battery satisfies all the constraints detailed in Section 2 of this paper.
The mathematical model previously described can be used for grid-
connected and standalone DC networks in any country or region. The 3.2. Master stage
only condition is that it should employ the economic, technical, and
environmental data from an actual electrical system and the region The master stage is responsible for solving the problem of optimal
where it is located. operation of batteries in DC grids. It proposes solutions to this problem
using the codification previously described and evaluates their impact
on the objective functions, which are later examined by the slave stage.
3. Proposed solution methodologies
Since the problem analyzed here is continuous non-lineal and non-
convex, three metaheuristic methods, i.e., the ALO, the VSA, and PSO,
This study proposes three different master–slave methodologies to were selected thanks to their excellent performance in solving optimal
solve the problem of optimal operation of batteries in DC MGs while power flow problems in DC grids, as reported in the literature [21,
improving their economic, technical, and environmental conditions. 58,59]. Because all these three optimization techniques are based on
The master stage of each of these methodologies employs a parallel populations, parallel processing can be used to simultaneously evaluate
version of a metaheuristic optimization technique, i.e., the Parallel Ant the objective functions of each one of the individuals that compose
Lion Optimizer (PALO), the Parallel Vortex Search Algorithm (PVSA), the population, thus reducing the processing time they require [6].
or Parallel Particle Swarm Optimization (PPSO). These optimization Therefore, this paper implemented the parallel versions of the three
methodologies were selected because they have achieved excellent methods mentioned above, i.e., the Parallel Ant Lion Optimizer (PALO),
results (in terms of solution, repeatability, and processing times) when the Parallel Vortex Search Algorithm (PVSA), and Parallel Particle
they have been used to solve optimal power flow problems in DC Swarm Optimization (PPSO), respectively.
grids [21,58,59]. Parallel processing was implemented to reduce their To control the exploration of the solution space by the three meth-
processing times because the operation of DERs makes it necessary to ods proposed here, two stopping criteria were applied: maximum num-
find the best solution as fast as possible [6]. In turn, the slave stages ber of iterations (𝑖𝑡𝑒𝑟𝑚𝑎𝑥 ) and maximum number of non-improvement
of all these three methodologies used the Hourly Power Flow method iterations (𝑖𝑡𝑒𝑟𝑁𝐼
𝑚𝑎𝑥 ). If any of these two stopping criteria is satisfied, the
based on SA (referred to here as HPFSA) due to its excellent results iterative process of the algorithm ends.
reported in literature in terms of convergence and processing times [6]. The following subsection details the main characteristic of each one
To explain the master–slave methodologies proposed in this paper, of these solution methods and their adaptation to parallel processing.
the following subsections describe their problem codification, their If readers of this paper need more information about one of the opti-
optimization methods, the parallel processing added to the latter, and mization techniques described below, they can consult the references
the HPFSA. cited in the respective subsection.
9
L.F. Grisales-Noreña et al. Journal of Energy Storage 65 (2023) 107240
3.2.1. Parallel Ant Lion Optimizer (PALO) After the fitness function of the initial population has been evaluated,
This optimization algorithm takes advantage of the hunting strate- the individual with the best objective function is selected as the ant lion
gies that ant lions use to obtain food. They create a cone-shaped trap at the current iteration (𝐴𝑛𝑡𝑙𝑖𝑜𝑛𝑖𝑡𝑒𝑟 ).
on the ground to hunt other ants, moving in randoms walks inside
𝑁𝑃 = 𝑁∕𝑁𝐶 (24)
the area where said trap is located to improve their possibilities of
obtaining food [58]. This optimization algorithm was formulated to
explore the solution space using the randoms walks of ants, which 𝑃 𝑇 = 𝑀𝑃 𝑇 ∗ 𝑁𝑃 (25)
represent individuals that compose the population. Then, the position The iterative process of the PALO starts after the initial values of the
of the ant lion in the solution space is updated to obtain a good-quality algorithm have been obtained. With the values of the initial population,
solution. Algorithm 1 below details the parallel version of the Ant Lion random and constant values, and 𝐴𝑛𝑡𝑙𝑖𝑜𝑛𝑖𝑡𝑒𝑟 , the population is updated.
Optimizer. Then, using parallel processing and the slave stage, the fitness function
Data: Load electrical data of the DC MG and PALO optimization of all the individuals that compose the new population is evaluated.
parameters Subsequently, the 𝐴𝑛𝑡𝑙𝑖𝑜𝑛𝑖𝑡𝑒𝑟+1 is updated. To check if the stopping
Generate the initial population in a random way; criteria have been met, first, the algorithm establishes if the current
Evaluate the fitness function of the population with the slave stage 𝐴𝑛𝑡𝑙𝑖𝑜𝑛𝑖𝑡𝑒𝑟+1 is better than the previous 𝐴𝑛𝑡𝑙𝑖𝑜𝑛𝑖𝑡𝑒𝑟 . If so, the counter of
using parallel processing; non-improvement iterations (𝑖𝑡𝑒𝑟𝑁𝐼 ) takes a value of zero; otherwise, 1
Select the best solution as the 𝐴𝑛𝑡𝑙𝑖𝑜𝑛𝑖𝑡𝑒𝑟 (incumbent); is added to the counter. If 𝑖𝑡𝑒𝑟𝑁𝐼
𝑚𝑎𝑥 has been met, the algorithm stops and
for 𝑖𝑡𝑒𝑟 = 0 ∶ 𝑖𝑡𝑒𝑟𝑚𝑎𝑥 do prints 𝐴𝑛𝑡𝑙𝑖𝑜𝑛𝑖𝑡𝑒𝑟 as the solution to the problem; otherwise, the iterative
Generate the new population using the last population and ant process of the algorithm continues. Then, the algorithm checks if 𝑖𝑡𝑒𝑟𝑚𝑎𝑥
lion information; has been met. If so, it stops and prints 𝐴𝑛𝑡𝑙𝑖𝑜𝑛𝑖𝑡𝑒𝑟+1 as the solution to
Evaluate the fitness function of the population with the slave the problem; otherwise, its iterative process continues.
stage using parallel processing;
Update 𝐴𝑛𝑡𝑙𝑖𝑜𝑛𝑖𝑡𝑒𝑟+1 (incumbent); 3.2.2. Parallel Vortex Search Algorithm (PVSA)
if 𝐴𝑛𝑡𝑙𝑖𝑜𝑛𝑖𝑡𝑒𝑟 ≤ 𝐴𝑛𝑡𝑙𝑖𝑜𝑛𝑖𝑡𝑒𝑟+1 then The vortex search algorithm imitates the behavior of vortices cre-
𝑖𝑡𝑒𝑟𝑁𝐼 =𝑖𝑡𝑒𝑟𝑁𝐼 + 1; ated in stirred fluids [62] by generating non-concentric hyper-spheres.
if 𝑖𝑡𝑒𝑟𝑁𝐼 = 𝑖𝑡𝑒𝑟𝑁𝐼
𝑚𝑎𝑥 then As more hyper-spheres are produced, this optimization algorithm re-
Solution achieved; duces their radius and relocates their center at each iteration. The
Result: Print 𝐴𝑛𝑡𝑙𝑖𝑜𝑛𝑖𝑡𝑒𝑟 as the solution to the problem; center of the concentric hyper-spheres is updated with information
break; about the best solution obtained at each iteration. In this paper, this
end algorithm uses the same stopping criteria as the PALO, and its iterative
else process is described in Algorithm 2.
𝑖𝑡𝑒𝑟𝑁𝐼 = 0; Similar to the PALO, the PVSA starts its iterative process by reading
end data about the electrical system and PVSA parameters. The latter
if 𝑖𝑡𝑒𝑟 = 𝑖𝑡𝑒𝑟𝑚𝑎𝑥 then are reported in Table 7. Then, with the maximum and minimum
Solution achieved; values assigned to the variables that compose the problem, the al-
Result: Print 𝐴𝑛𝑡𝑙𝑖𝑜𝑛𝑖𝑡𝑒𝑟+1 as the solution to the problem; gorithm generates the initial radius (𝑟𝑖𝑡𝑒𝑟 ) and center (𝜇 𝑖𝑡𝑒𝑟 ) of the
break; hyper-sphere. Subsequently, the initial population is obtained using the
end values mentioned above and a Gaussian distribution, thus spreading
end the individuals over the solution space and guaranteeing a symmetric
Algorithm 1: Iterative process proposed for the PALO. exploration. Afterward, parallel processing (i.e., the PALO) is used
to evaluate the fitness function of the population by means of the
The iterative process of the PALO starts by reading information slave stage. Finally, to obtain the initial conditions of the PVSA, the
about the electrical system, which includes line parameters, number best solution in the population is selected as the current incumbent
of buses, conventional generators, PV DGs, batteries, generation and (𝐼𝑛𝑐𝑢𝑚𝑏𝑒𝑛𝑡𝑖𝑡𝑒𝑟 ) based on the results obtained previously.
demand curves, energy costs, and emission factors, among other data. After the initial values of the PVSA have been obtained, the it-
In its first step, the algorithm reads the parameters of the PALO erative process starts by selecting the individual that represents the
because this and the other optimization algorithms used in this paper best solution as the new center of the hyper-sphere (𝜇 𝑖𝑡𝑒𝑟+1 ). After
were tuned using a discrete-continuous genetic algorithm [60,61] to that, the radius of the hyper-sphere is updated for the current iteration
achieve the best performance of each solution method. This strategy is (𝑟𝑖𝑡𝑒𝑟+1 ). The new population is obtained with these data and a Gaussian
described later in Section 5 of this paper. distribution. Then, the slave stage calculates the fitness function of the
Then, in the second step of the iterative process, the PALO generates population. Afterward, the best solution found thus far is selected as
the initial population of ant lions in a random way using the codifica- the incumbent of the problem at the current iteration (𝐼𝑛𝑐𝑢𝑚𝑏𝑒𝑛𝑡𝑖𝑡𝑒𝑟+1 ).
tion presented in Fig. 1. In the third step, the fitness function of the With the current and last value of the incumbent, the stopping criteria
individuals that compose the initial population is evaluated by the slave are evaluated. If one of the two criteria have been met, the PVSA stops
stage. For this purpose, the slave stage analyzes 𝑁, which is the number and prints the current center and incumbent as the solution to the
of individuals. To take advantage of parallel processing, the fitness problem; otherwise, the optimization algorithm continues its iterative
function of the whole population is evaluated simultaneously using the process.
different cores in the CPU of the computer (𝑁𝐶), thus reducing the
number of processes as described in Eq. (24). In said equation, 𝑁𝑃 3.2.3. ParallelParticle Swarm Optimization (PPSO)
represents the number of processes required to evaluate the 𝐹 𝐹 of 𝑁 PSO uses the hunting behavior that bird flocks and fish schools use
individuals that compose the population. The processing time required to search for food. It employs the cognitive and social knowledge of the
by the iterative process is described in Eq. (25), where 𝑀𝑃 𝑇 is the swarm to explore a solution space and obtain the best possible food.
maximum processing time required by the parallel iterative process, Over time, all the individuals that compose the population converge
and 𝑃 𝑇 is the total processing time required to evaluate the entire towards the best place to feed in the solution space [6]. This study used
population. The higher the 𝑁𝐶 and the better the performance of the the parallel version of PSO (i.e., PPSO), which is described in Algorithm
computer, the shorter the processing time of the optimization methods. 3.
10
L.F. Grisales-Noreña et al. Journal of Energy Storage 65 (2023) 107240
Data: Load electrical data and PVSA optimization parameters Data: Load electrical data and PPSO optimization parameters
Define (𝑟𝑖𝑡𝑒𝑟 ) and (𝜇𝑖𝑡𝑒𝑟 ) of the hyper-sphere; Generate initial population in a random way;
Generate the initial population using 𝜇 𝑖𝑡𝑒𝑟 , 𝑟𝑖𝑡𝑒𝑟 , and a Gaussian Evaluate the fitness function of the population with the slave stage
distribution; using parallel processing;
Evaluate the fitness function of the population with the slave stage Identify the best particle and swarm solution 𝐼𝑛𝑐𝑢𝑚𝑏𝑒𝑛𝑡𝑖𝑡𝑒𝑟 ;
using parallel processing; for 𝑖𝑡𝑒𝑟 = 0 ∶ 𝑖𝑡𝑒𝑟𝑚𝑎𝑥 do
Select the best solution in the population as the 𝐼𝑛𝑐𝑢𝑚𝑏𝑒𝑛𝑡𝑖𝑡𝑒𝑟 for Generate new individuals in the population using the
𝑖𝑡𝑒𝑟 = 0 ∶ 𝑖𝑡𝑒𝑟𝑚𝑎𝑥 do information contained in the best particle and solution in the
Select position of the best solution in the population as the swarm;
new center of the hyper-sphere 𝜇 𝑖𝑡𝑒𝑟+1 ; Evaluate the fitness function of the population with the slave
Update radius of the hyper-sphere 𝑟𝑖𝑡𝑒𝑟+1 ; stage using parallel processing;
Generate new population using 𝜇 𝑖𝑡𝑒𝑟+1 , 𝑟𝑖𝑡𝑒𝑟+1 , and a Gaussian Update best particle and solution in the swarm
distribution; (𝐼𝑛𝑐𝑢𝑚𝑏𝑒𝑛𝑡𝑖𝑡𝑒𝑟+1 );
Evaluate the fitness function of the population with the slave if 𝐼𝑛𝑐𝑢𝑚𝑏𝑒𝑛𝑖𝑡𝑒𝑟 ≤ 𝐼𝑛𝑐𝑢𝑚𝑏𝑒𝑛𝑖𝑡𝑒𝑟+1 then
stage using parallel processing; 𝑖𝑡𝑒𝑟𝑁𝐼 = 𝑖𝑡𝑒𝑟𝑁𝐼 + 1;
Select the best solution in the population as the 𝐼𝑛𝑐𝑢𝑚𝑏𝑒𝑛𝑡𝑖𝑡𝑒𝑟+1 ; if 𝑖𝑡𝑒𝑟𝑁𝐼 = 𝑖𝑡𝑒𝑟𝑁𝐼
𝑚𝑎𝑥 then
if 𝐼𝑛𝑐𝑢𝑚𝑏𝑒𝑛𝑡𝑖𝑡𝑒𝑟 ≤ 𝐼𝑛𝑐𝑢𝑚𝑏𝑒𝑛𝑡𝑖𝑡𝑒𝑟+1 then Solution achieved;
𝑖𝑡𝑒𝑟𝑁𝐼 = 𝑖𝑡𝑒𝑟𝑁𝐼 + 1; Result: Print the best solution in the swarm and
if 𝑖𝑡𝑒𝑟𝑁𝐼 =𝑖𝑡𝑒𝑟𝑁𝐼
𝑚𝑎𝑥 then 𝐼𝑛𝑐𝑢𝑚𝑏𝑒𝑛𝑖𝑡𝑒𝑟 as the solution to the problem;
Solution achieved; break;
Result: Print 𝜇 𝑖𝑡𝑒𝑟 and 𝐼𝑛𝑐𝑢𝑚𝑏𝑒𝑛𝑡𝑖𝑡𝑒𝑟 as the solution to end
the problem else
; 𝑖𝑡𝑒𝑟𝑁𝐼 = 0;
break; end
end if 𝑖𝑡𝑒𝑟 = 𝑖𝑡𝑒𝑟𝑚𝑎𝑥 then
else Solution achieved;
𝑖𝑡𝑒𝑟𝑁𝐼 = 0; Result: Print the best solution in the swarm and
end 𝐼𝑛𝑐𝑢𝑚𝑏𝑒𝑛𝑖𝑡𝑒𝑟+1 as the solution to the problem
if 𝑖𝑡𝑒𝑟 = 𝑖𝑡𝑒𝑟𝑚𝑎𝑥 then break;
Solution achieved; end
Result: Print 𝜇 𝑖𝑡𝑒𝑟+1 and 𝐼𝑛𝑐𝑢𝑚𝑏𝑒𝑛𝑡𝑖𝑡𝑒𝑟+1 as the solution to end
the problem; Algorithm 3: Iterative process proposed for PPSO.
break;
end
end
illustrated in Fig. 1 can be used to generate solutions for the problem
Algorithm 2: Iterative process proposed for the PVSA.
of optimal operation of batteries in DC MGs. However, this hourly
codification requires the implementation of an hourly power flow
method to evaluate the impact, on the fitness function, of the power
The iterative process of PPSO starts by reading data about the supplied and stored in each period over the time horizon analyzed here
electrical systems and PPSO parameters. The latter are detailed in (24 h in this particular case). Algorithm 4 describes the iterative process
Table 7. Then, the algorithm randomly generates the initial population
of the Hourly Power Flow method based on Successive Approximation
using the codification in Fig. 1 and the maximum and minimum values
(HPFSA) used in this paper [6,58].
fixed for the continuous variables. Subsequently, the fitness function of
each one of the particles that compose the swarm is evaluated using the
same parallel process described for the PALO. Finally, the best particle Data: Read data about electrical systems and HPFSA parameters ;
and solution in the swarm are identified, and the latter is selected as for 𝑡 = 1 ∶ 24 do
Load the power demanded by the loads located in the DC MG
the incumbent to the problem at the current iteration (𝐼𝑛𝑐𝑢𝑚𝑏𝑒𝑛𝑖𝑡𝑒𝑟 ).
at time 𝑡;
After the initial conditions of PPSO have been fixed, the iterative
Load the power generated by the PV DGs located in the DC
process starts by updating the position of the particles on the solution
MG at time 𝑡;
space. For this purpose, the algorithm uses the best position achieved
Load the power supplied or demanded by the batteries
by the particle and the swarm, as well as some random values and
installed in the DC MG at time 𝑡;
cognitive, social, and inertia constants that control the exploration and
Solve the DC power flow for the power data of time 𝑡 using the
convergence of said algorithm. Afterward, the fitness function of the
successive approximation method reported in [24];
particles is evaluated using parallel processing. The results are updated
Calculate the fitness function at time 𝑡 using Equation (22);
to obtain the best solution for each particle and the swarm, the latter
end
being the new incumbent solution to the problem (𝐼𝑛𝑐𝑢𝑚𝑏𝑒𝑛𝑖𝑡𝑒𝑟+1 ).
Add the 𝐹 𝐹 values obtained for each period of time;
Then, if one of its two stopping criteria has been met, the algorithm
Return the total 𝐹 𝐹 to the master stage;
stops and prints the current incumbent as the solution to the problem;
Algorithm 4: Pseudo-code proposed for the HPFSA.
otherwise, the optimization process continues.
3.3. Slave stage The HPFSA starts by reading the data about the electrical systems
and the power flow parameters. The electrical data depends on the test
In this study, the slave stage is responsible for evaluating the impact system under analysis (see next section). The power flow parameters
of each one of the solutions (proposed by the master stage) on the used in this paper were taken from [24]. In the latter, the authors
fitness function. As explained in the previous sections, the codification reported that the SA method obtained its best performance when the
11
L.F. Grisales-Noreña et al. Journal of Energy Storage 65 (2023) 107240
Table 2
Electrical parameters for the standalone DC MG (test system).
Node 𝑖 Node 𝑗 𝑅𝑖𝑗 (Ω) 𝑃𝑗 (kW) 𝐼𝑚𝑎𝑥(A)
1 2 0.0140 0 195
2 3 0.7463 0 145
3 4 0.4052 297.5 85
4 5 1.1524 0 70
5 6 0.5261 255 70
6 7 0.7127 0 55
7 8 1.6628 212.5 55
8 9 5.3434 0 20
9 10 2.1522 266.05 20
2 11 0.4052 85 70
Fig. 2. Electrical configuration proposed for the standalone DC MG (test system). 11 12 1.1524 340 55
12 13 0.5261 297.5 40
13 14 1.2358 19125 25
14 15 2.8835 106.25 20
maximum number of iterations was 1000, and the convergence error 15 16 5.3434 255 20
3 17 1.2942 255 55
was 1x10−10 . 17 18 0.7027 127.5 40
After the input data for the HPFSA have been defined, the iterative 18 19 3.3234 297.5 40
process starts by analyzing every hour over the time horizon (24 h). The 19 20 1.5172 340 20
20 21 0.7127 85 20
HPFSA reads, in each hour, the power generated and demanded by the 4 22 8.2528 106.25 20
loads, the PV DGs, and the batteries installed in the DC MG. Afterward, 5 23 9.1961 55.25 20
this information is employed to solve the power flow problem using the 6 24 0.7463 69.7 20
8 25 2.0112 255 20
successive approximation method reported in [24]. Then, the voltage
8 26 3.3234 63.75 20
profiles obtained by the SA method and the data on the electrical 26 27 0.5261 170 20
system are implemented to calculate the fitness function using Eq. (22).
After all the time periods have been analyzed, the 𝐹 𝐹 values obtained
for each period are added, and the total 𝐹 𝐹 is returned to the master
stage.
In Table 2, the five columns show (from left to right) the sending
Each test scenario (i.e., the standalone and the grid-connected DC
node; the receiving node; the resistance of the line that connects
MGs) includes 3 PV DGs and 3 lithium-ion batteries as DERs. Table 4
the sending and receiving nodes (𝑅𝑖𝑗 ); the power demanded at the
shows the nodes where the PV DGs are located in each test system
receiving node (𝑃𝑗 ) due to the load connected to this node; and, finally,
and their respective power ratings (𝑃𝐷𝐺 ). Since this paper focuses on
the maximum current limits for each line (𝐼𝑚𝑎𝑥). It should be clarified the operation of the batteries, the PV DGs are assumed to operate at
that these maximum line currents were calculated according to the their maximum power point (MPP), while considering power varia-
Colombian technical standard (NTC 2050) [63] and using the results tions related to solar radiation, ambient temperature, and type of PV
obtained after running a power flow with the basic structure of the technology (see Section 4.4).
standalone test system (i.e., without PV DGs or batteries). Finally, this This paper considers three different types of lithium-ion batteries:
standalone DC MG considered a maximum voltage limit of +/- 10% of A, B, and C. Table 5 details (from left to right) the battery number, its
the nominal voltage, as established in NTC 1340 standard [64], which location node in the DC MG, type, power capacity (𝑃𝑏 ), charging time
describes the voltage regulation in Colombia. (𝑡𝑐 ) in hours, and discharging time (𝑡𝑑 ) in hours.
12
L.F. Grisales-Noreña et al. Journal of Energy Storage 65 (2023) 107240
Table 3
Electrical parameters for the grid-connected DC MG (test system).
Node 𝑖 Node 𝑗 𝑅𝑖𝑗 (Ω) 𝑃𝑗 (kW) 𝐼𝑚𝑎𝑥(A)
1 2 0.0922 100 320
2 3 0.4930 90 280
3 4 0.3660 120 195
4 5 0.3811 60 195
5 6 0.8190 60 195
6 7 0.1872 200 95
7 8 17114 200 85
8 9 10300 60 70
9 10 10400 60 55
10 11 0.1966 45 55
11 12 0.3744 60 55
12 13 14.680 60 40
13 14 0.5416 120 40
14 15 0.5910 60 25
15 16 0.7463 60 20
16 17 12890 60 20
17 18 0.7320 90 20
2 19 0.1640 90 30
19 20 15042 90 25
20 21 0.4095 90 20
21 22 0.7089 90 20
3 23 0.4512 90 85
23 24 0.8980 420 70
24 25 0.8900 420 40
6 26 0.2030 60 85
26 27 0.2842 60 85 Fig. 4. (a) Average PV power generation in standalone and grid-connected MGs.(b)
27 28 10590 60 70 Average power demand in standalone and grid-connected MGs.
28 29 0.8042 120 70
29 30 0.5075 200 55
30 31 0.9744 150 40 Table 6
31 32 0.3105 210 25 Economic and environmental parameters of the electrical networks analyzed here.
32 33 0.3410 60 20 DC MG 𝐶𝐶𝐺𝑖 𝐶𝑀𝐷𝐺𝑖 𝐶𝑀𝐵𝑖 𝛾𝐶𝐺𝑖
(USD/kWh) (USD/kWh) (USD/kWh) (kg/kWh)
Grid-connected 0.1302 0.0019 0.2913 0.1644
Table 4
Standalone 0.2913 0.0019 0.2913 0.2671
Locations and power ratings of the PV DGs in each test system.
DG Standalone network Grid-connected network
Node 𝑃𝐷𝐺 (kW) Node 𝑃𝐷𝐺 (kW)
1 5 1012.5 12 1125 are given in p.u. (per unit) so that they can be extrapolated based on
2 9 1188 25 1320 the installed power of each generator.
3 19 899.1 30 999
To establish the power demand curves and the average consumption
in Capurganá and Medellín in 2019, this study obtained data from
Table 5 reports published by their respective network operators, i.e., IPSE
Location and characteristics of the batteries in the two test systems. (Instituto de Planificación 𝑦 Promoción de Soluciones Energéticas para
Standalone network (rural) Zonas no Interconectadas) [53] and EPM (Empresas Públicas de Medel-
Battery Node Type 𝑃𝐵 (kW) 𝑡𝑐 (h) 𝑡𝑑 (h) lín) [6]. The power demand curves in Capurganá and Medellín are
1 3 C 2000 5 5 shown in Fig. 4b.
2 8 A 1000 4 4
3 19 B 1500 4 4
4.5. Parameters used in the objective functions proposed in this study
Grid-connected network (urban)
Battery Node Type 𝑃𝐵 (kW) 𝑡𝑐 (h) 𝑡𝑑 (h)
Table 6 details four parameters for the grid-connected and stan-
1 6 C 2000 5 5
2 14 A 1000 4 4 dalone MGs: specific energy purchasing costs of conventional genera-
3 31 B 1500 4 4 tors installed in the grid; battery cost; DG maintenance costs; and CO2
emissions coefficients. In the standalone DC MG, the energy purchasing
cost is calculated based on power generation by diesel generators, and
the CO2 emissions coefficient depends on this type of generation. In
4.4. PV generation and demand power curves for standalone and grid-
the grid-connected DC MG, the energy purchasing cost is associated
connected DC MGs
with the grid energy cost fixed by EPM (i.e., the grid operator in this
The PV generation and demand power curves for the standalone region), and the CO2 emissions coefficient depends on the type of power
and grid-connected DC MGs were obtained from an average day in generation technologies used in the grid [55]. The costs of batteries
2019 [55]. This year was selected so that the data would not be and DG maintenance are assumed to be the same for both types of
affected by the COVID-19 pandemic. The PV power generation curves, DC MG [25,67]. These costs are calculated as a function of the power
radiation, and ambient temperature were obtained from NASA POWER generated and demanded by the DERs. The CO2 emissions coefficient
reports for Capurganá and Medellín [65], with 1-hour steps between of the PV DGs is considered to be equal to zero because the operation
measurements. Considering these data and poly-crystalline silicon PV of these DERs does not generate CO2 emissions. That is, although a CO2
panels, the PV generation in each region was determined [57,66]. emissions footprint remains in the construction of solar panels, this is
Fig. 4a shows the PV power generation curves for the standalone and considered a prior action and not a consequence of using solar panels
grid-connected networks. The values in the PV power generation curves as DGs in a DC MG [68].
13
L.F. Grisales-Noreña et al. Journal of Energy Storage 65 (2023) 107240
Table 7
Optimization parameters tuned for the solution methods employed in this study: PALO,
PVSA, and PPSO.
Method Optimization parameters Value
Number of particles 32
PALO Maximum number of iterations 428
Maximum number of non-improvement iterations 349
Number of particles 129
Maximum number of iterations 1049
PVSA
Maximum number of non-improvement iterations 544
Radius reduction interval (x) 0.027
Number of particles 191
Maximum number of iterations 1928
Maximum number of non-improvement iterations 860
Fig. 5. Variable energy costs for the grid-connected DC MG [7]. PPSO Cognitive component (C1) 0.7063
Social component (C2) 1.8098
Minimum inertia 0.2
Maximum inertia 0.7231
All the values in Table 6 are considered constant over all the time
intervals. However, the grid-connected DC MG considers two scenarios
for the objective function of the reduction of operating costs. Selecting Table 8
Results of the simulations using three optimization methodologies in the standalone DC
one scenario or the other depends on the power purchasing cost. In
MG.
the first scenario, said cost is considered fixed because it is constant
Minimum objective function
for most users in grid-connected networks, as reported in Table 6. In
Method 𝐸𝑐𝑜𝑠𝑡 (USD) 𝐸𝑙𝑜𝑠𝑠 (kWh) 𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠 (kg CO2 )
the second scenario, the energy purchasing cost is variable, changing
Base case 15366.0235 383.2191 14072.4694
every hour as shown in the variation curve in Fig. 5. This variable
PALO 15364.1413 341.3382 14061.2944
cost scenario exists in Colombia for unregulated users, while the second PVSA 15363.9433 341.1321 14061.2717
scenario includes this variation in p.u. applied to the fixed energy cost PPSO 15364.3663 342.0684 14061.4095
used in the first scenario. The purpose of implementing these two cost Mean objective function
scenarios is to compare and analyze the impact of fixed and variable
Method 𝐸𝑐𝑜𝑠𝑡 (USD) 𝐸𝑙𝑜𝑠𝑠 (kWh) 𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠 (kg CO2 )
costs on battery energy management in DC MGs.
PALO 15365.9063 341.7322 14061.3960
PVSA 15364.8569 341.6536 14061.3810
5. Simulation results PPSO 15365.7065 343.8800 14062.0259
Standard deviation [%]
This section presents and analyzes the results obtained by the Method 𝐸𝑐𝑜𝑠𝑡 𝐸𝑙𝑜𝑠𝑠 𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠
master–slave methodologies proposed above. These methodologies
PALO 1.0555 0.2097 0.0545
solve the problem of optimal operation of battery storage systems in PVSA 0.5919 0.2682 0.0858
standalone and grid-connected DC MGs by controlling the charging and PPSO 0.7630 1.3853 0.3940
discharging actions of the batteries in the network, under a scenario of Average processing time [s]
distributed generation by solar panels that operate at their maximum
Method 𝐸𝑐𝑜𝑠𝑡 𝐸𝑙𝑜𝑠𝑠 𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠
power point. All the simulations in this study were performed in
PALO 18.2570 18.2434 18.3840
MATLAB® 2022𝑎 on a computer with the following features: Intel(R) PVSA 41.2985 41.0999 41.3089
Xeon(R) E5-1660 v3 3.0-GHz CPU, 16-GB DDR4 RAM, 2.5’’ 480-GB PPSO 82.3534 85.5945 82.3862
solid state hard drive, and Windows 10 Pro operating system. In order
to determine which methodology produces the best results in terms
of solution, repeatability, and processing times, each one of them was
executed 1000 times to evaluate the average values they obtained for energy transport given in kW; and levels of CO2 emissions (𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠)
per kg produced by conventional power generation (a diesel generator
the objective functions, processing times, and standard deviation.
in this case). To analyze Table 8, one should start with the base
To guarantee a fair comparison between the optimization algo-
case, a scenario without batteries that considers power injection by PV
rithms used here and obtain the best results for the problem under
generators operating at their MPP. Then, the base case can be compared
study, each one of the optimization techniques was tuned using a
to the results obtained by the PALO, the PVSA, and PPSO in terms
discrete-continuous genetic algorithm [60,61]. This GA used a popu-
of three objective functions (i.e., 𝐸𝑐𝑜𝑠𝑡, 𝐸𝑙𝑜𝑠𝑠, and 𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠). This
lation of 40 individuals and 400 iterations. Furthermore, the following
table presents the results of the three optimization algorithms for the
ranges were implemented to tune its parameters: number of particles
three objective functions, detailing their minimum objective function
or individuals, [1–100]; number of iterations, [1–1000]; and number of
(i.e., best solution), mean objective function (i.e., average solution),
non-improvement iterations, [1–1000]. PPSO used two ranges: cognitive standard deviation percentage, and average time required in seconds.
and social components, [0–2]; and inertia parameter, [0.01–1]. Finally, Note that all the solutions obtained by the optimization algorithms
the PVSA incorporated a parameter known as radius reduction interval, satisfy the current line limits and nodal voltage profiles imposed by the
i.e., 𝑥, with a [0–1] range. The parameters obtained by this GA for each electrical network under analysis, which is achieved by implementing
algorithm are presented in Table 7. This table details (from left to right) the fitness function given in Section 2.
the optimization algorithm, the parameters that were tuned, and the Fig. 6 presents the reductions obtained by three optimization meth-
result of the tuning process. ods (i.e., the PALO, the PVSA and PPSO) with respect to the base
case (a scenario without batteries supplying or storing energy). These
5.1. Standalone network: simulation and results minimum and average reductions were achieved in three objective
functions: 𝐸𝑐𝑜𝑠𝑡, 𝐸𝑙𝑜𝑠𝑠, and 𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠. These results also show that
Table 8 presents the results obtained in the case of the standalone (with respect to the base case) the PALO obtained minimum reductions
DC MG. It details (from left to right) the optimization method (Method); of 0.0122%, 10.8831%, and 0.0792% in 𝐸𝑐𝑜𝑠𝑡, 𝐸𝑙𝑜𝑠𝑠, and 𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠,
operating cost (𝐸𝑐𝑜𝑠𝑡) in USD$; power losses (𝐸𝑙𝑜𝑠𝑠) associated with respectively. In addition, the PALO achieved average reductions of
14
L.F. Grisales-Noreña et al. Journal of Energy Storage 65 (2023) 107240
Fig. 6. Minimum and average reductions in economic, technical, and environmental objective functions obtained by three optimization methods in a rural standalone DC network.
0.0035%, 10.6459%, and 0.0772% in the same objective functions, Fig. 7(d) details the processing times required by the methodolo-
respectively. In turn, the PVSA exhibited the most significant reductions gies proposed here. In it, the PVSA is in second place, with average
in terms of 𝐸𝑐𝑜𝑠𝑡, 𝐸𝑙𝑜𝑠𝑠, and 𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠. reductions in processing times of 49.8522% (𝐸𝑐𝑜𝑠𝑡), 51.9830% (𝐸𝑙𝑜𝑠𝑠),
Similarly, to establish the impact of the proposed solution method- and 49.8594% (𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠) compared to those required by PPSO. In this
ologies on the objective functions of the standalone network (described case, the first place was taken by the PALO, which reduced the times
in Section 2), the average reductions should be calculated based on required by the PVSA by 126.2065% (𝐸𝑐𝑜𝑠𝑡), 125.2859% (𝐸𝑙𝑜𝑠𝑠), and
the energy purchasing cost with respect to the base case. In this 124.7008% (𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠). Although PPSO is a fast solution methodology,
case, 1.17 USD$ was obtained as the operating cost on an average it presented the worst results in terms of objective functions. This
day. Therefore, the average annual cost would be 425.8 USD$. On an demonstrates that its short processing times are due to a more limited
average day of operation in the network, 𝐸𝑙𝑜𝑠𝑠 and 𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠 were exploration of the solution space. As a result, PPSO is not considered a
reduced by 41.57 kW and 11.09 kg of CO2 , respectively, generating an good solution methodology for the optimal operation of battery systems
average annual reduction of 15.17 MW and 4.05 CO2 tons, respectively. in standalone DC MGs.
These results highlight the importance of adequate energy management Fig. 8 presents the branch currents and voltage profiles obtained
systems for operating battery systems installed in standalone electrical by the optimization methods after they analyzed all the objective
networks. In particular, energy management can improve the technical, functions. Fig. 8(a) indicates that all the currents stayed under the
environmental, and economic conditions of a standalone electrical maximum limit in the 26 branches that compose the standalone DC
network that uses diesel generation. In this type of network, PVSA grid. Furthermore, Fig. 8(b) shows that all the voltage profiles obtained
obtained the best results for the three objective functions in terms of in the different time periods analyzed here remained within the maxi-
minimum and average reductions, as well as standard deviation (under mum and minimum voltage limits. These results demonstrate that the
1%).
mathematical model and optimization methodologies proposed here
Fig. 6 details the reductions obtained by the PVSA with respect to
satisfy all the operating constraints that represent DC grids.
the PALO and PPSO, which are better illustrated in Fig. 7. Fig. 7(a)
Based on this analysis, the PVSA is the best methodology in this case
shows the minimum reductions obtained by the PVSA compared to
because it offers the best solution quality, repeatability, and processing
those achieved by the two other solution methodologies, i.e., 0.0020%
time to solve the problem described above in a standalone DC MG. Fi-
in 𝐸𝑐𝑜𝑠𝑡, 0.1670% in 𝐸𝑙𝑜𝑠𝑠, and 0.0006% in 𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠. Fig. 7(b)
nally, Fig. 9 details the behavior of the batteries’ SOC in the standalone
presents the improvement obtained by the PVSA but in terms of average
DC grid when different objective functions are used. In all the scenarios
reductions. In this figure, compared to the PALO and PPSO, the PVSA
in the standalone grid, the batteries supplied energy during the peak
obtained average reductions of 0.0062%, 0.3352%, and 0.0023% in
demand hours, i.e., from 1:00 to 3:00 and from 19:00 to 24:00 h (see
𝐸𝑐𝑜𝑠𝑡, 𝐸𝑙𝑜𝑠𝑠, and 𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠, respectively. These results make the PVSA
Fig. 4(b)). In turn, they stored energy when the power demand was low
the best methodology in this case because it produced the greatest im-
and solar radiation was high (from 5:00 to 15:00 h). The SOC of all the
provements in the economic, technical, and environmental conditions
of the standalone DC MG. batteries started and ended the operation day within the limits fixed
After each algorithm was executed 1000 times, the standard de- by the proposed methodology. This behavior of the SOC indicates that
viation showed that the repeatability of the solutions depended on battery operation is directly related to the power demand curve, which
the optimization technique (see Fig. 7(c)). This figure indicates that, should be taken into account to improve the technical, economic, and
with respect to PPSO, the PVSA reduced the standard deviation of environmental conditions of standalone networks.
𝐸𝑐𝑜𝑠𝑡, 𝐸𝑙𝑜𝑠𝑠, and 𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠 by 43.922%, 80.6393%, and 78.211%,
respectively. In relation to the results obtained by the PALO, the PVSA 5.2. Grid-connected network: simulation and results
achieving a reduction of 22.4197% of standard deviation presented
for 𝐸𝑐𝑜𝑠𝑡. In contrast, the PALO achieved standard deviations of 𝐸𝑙𝑜𝑠𝑠 Table 9 reports the results obtained for the grid-connected DC MG.
and 𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠 that were 27.9314% and 57.6163% lower than those It is organized as Table 8, but this one includes two types of energy
produced by the PVSA, respectively (numbers in red in the figure). purchasing costs for the CGs, i.e., fixed (𝐹 𝑖𝑥𝑒𝑑𝐸𝑐𝑜𝑠𝑡) and variable
Nevertheless, regarding technical and environmental conditions, the (𝑉 𝑎𝑟𝐸𝑐𝑜𝑠𝑡).
PVSA presented standard deviation values of 0.2682% and 0.0858%, Table 9 and Fig. 10 detail the reductions in 𝐹 𝑖𝑥𝑒𝑑𝐸𝑐𝑜𝑠𝑡, 𝑉 𝑎𝑟𝐸𝑐𝑜𝑠𝑡,
which are excellent results in terms of solution repeatability. 𝐸𝑙𝑜𝑠𝑠, and 𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠 obtained by the three optimization methods
15
L.F. Grisales-Noreña et al. Journal of Energy Storage 65 (2023) 107240
Fig. 7. Reductions obtained by the PVSA (compared to the PALO and PPSO) in the standalone DC MG.
Fig. 8. Current and voltage values obtained by three solution methodologies in the standalone DC MG.
compared to the base case. In particular, they show that all the method- a year if the cost is fixed to 32671.15 USD$ a year if it is variable.
ologies achieved reductions in the objective functions proposed for Such annual reduction in energy purchasing costs can reach 96.24%
the grid-connected DC MG. Unlike in the standalone DC MG, in this if the cost is considered variable. Consequently, energy management
case, variable energy purchasing costs can be analyzed. As a result, systems that include batteries should implement variable energy costs.
the batteries can be charged when electricity is more economical and In addition, average daily reductions of 101.63 kW and 16.69 Kg CO2
discharged when the cost is higher, thus increasing the possibility were achieved for 𝐸𝑙𝑜𝑠𝑠 and 𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠, respectively. After an average
of obtaining profits from the energy management of power storage year of operation, this can result in approximately 37.095 MW and
devices. 6.093 CO2 tons less, respectively. These results demonstrate that using
Fig. 10 presents the reductions obtained by the three methodologies optimization methodologies for an adequate operation of batteries can
in the grid-connected DC MG compared to the base case. Using the data improve the technical, economic, and environmental conditions of
in this figure, it is possible to calculate average minimum reductions grid-connected DC MGs.
of 0.0572% in 𝐹 𝑖𝑥𝑒𝑑𝐸𝑐𝑜𝑡𝑠, 1.3085% in 𝑉 𝑎𝑟𝐸𝑐𝑜𝑠𝑡, 7.6690% in 𝐸𝑙𝑜𝑠𝑠, Fig. 11 illustrates the reductions obtained by the PALO with respect
and 0.1754% in 𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠. After 1000 executions of each optimization to the other solution methods. In particular, Fig. 11(a) presents the
methodology, the objective functions presented the following aver- minimum reductions in objective functions obtained by the PALO com-
age reductions compared to the base case: 0.9535% in 𝐹 𝑖𝑥𝑒𝑑𝐸𝑐𝑜𝑡𝑠, pared to PVSA and PPSO: −0.0072% (𝐹 𝑖𝑥𝑒𝑑𝐸𝑐𝑜𝑡𝑠), 0.0193%
3.4266% in 𝑉 𝑎𝑟𝐸𝑐𝑜𝑠𝑡, 7.2112% in 𝐸𝑙𝑜𝑠𝑠, and 0.9656% in 𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠. (𝑉 𝑎𝑟𝐸𝑐𝑜𝑠𝑡), 0.2192% (𝐸𝑙𝑜𝑠𝑠), and 0.0085% (𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠). Although the
These results demonstrate that all the solution methodologies used PALO was outperformed by the PVSA in 𝐹 𝑖𝑥𝑒𝑑𝑐𝑜𝑠𝑡, the difference is
in this study improved the economic, technical, and environmental minimal. However, regarding 𝑉 𝑎𝑟𝑐𝑜𝑠𝑡, 𝐸𝑙𝑜𝑠𝑠, and 𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠, the PALO
conditions of the grid-connected DC MG. In particular, the PALO offers obtained better results than the two other solution methods employed
the best performance in terms of solution quality. here. Fig. 11(b) shows a similar behavior in average reductions ob-
The results illustrated in Fig. 10 indicate that the optimization tained by the optimization methodologies. In particular, compared to
algorithms can further reduce 𝐸𝑐𝑜𝑠𝑡 when it is variable. For example, the PVSA and PPSO, the PALO achieved the following average reduc-
on an average day of operation, when the energy purchasing cost tions: 1.2091% (𝐹 𝑖𝑥𝑒𝑑𝐸𝑐𝑜𝑡𝑠), 4.2358% (𝑉 𝑎𝑟𝐸𝑐𝑜𝑠𝑡), 6.6357% (𝐸𝑙𝑜𝑠𝑠),
is fixed, it can be reduced by 3.36 USD$. In contrast, when it is and 1.3288% (𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠).
variable, it can be reduced by 89.51 USD$. This extra reduction in These results indicate that the PALO is the best methodology (in
daily energy purchasing costs means going from saving 1226.40 USD$ terms of minimum and average results) to improve the technical,
16
L.F. Grisales-Noreña et al. Journal of Energy Storage 65 (2023) 107240
Fig. 9. State of charge of the batteries in the grid-connected DC grid according to the best solution found by the PALO.
Fig. 10. Minimum and average reductions in economic, technical, and environmental objective functions obtained by three optimization methods in the grid-connected DC MG.
economic, and environmental conditions of the grid-connected DC MG, Fig. 12 indicates that the solutions obtained by the proposed
which considers fixed and variable energy purchasing costs for the CGs. methodologies in the grid-connected MG respect the voltage and cur-
To calculate the standard deviations of their solutions, the solution rent limits. Fig. 12(a) demonstrates that all the currents in the branches
algorithms were executed 1000 times each (as in the case of the stan- that compose this electrical network were lower than the maximum
dalone DC MG). In Fig. 11(c), the PALO reduced the standard deviation limit established for each branch (32 branches in total in this particular
values of 𝐹 𝑖𝑥𝑒𝑑𝐸𝑐𝑜𝑠𝑡, 𝑉 𝑎𝑟𝐸𝑐𝑜𝑠𝑡, 𝐸𝑙𝑜𝑠𝑠, and 𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠 by 66.5859%, case). Furthermore, during the entire day of operation, the voltage
60.2921%, 84.4552%, and 86.9253%, respectively, on average. There- profiles respected the maximum and minimum voltage limits that were
fore, the PALO can be used in grid-connected DC networks because set for all the 33 buses (see Fig. 12(b)).
it obtained the best results in terms of solution quality, with lower Fig. 11(d) reports the average processing times required by the pro-
standard deviations. Moreover, the PALO offers the best performance posed methodologies in the grid-connected DC MG. In that figure, the
(regarding minimum reduction, average reduction, and standard devia- PALO ranks third, behind the PVSA. The PVSA outperformed the PALO
tion) to improve the technical, economic, and environmental conditions in the average processing times of 𝐹 𝑖𝑥𝑒𝑑𝐸𝑐𝑜𝑠𝑡 by 30.1684%, 𝑉 𝑎𝑟𝐸𝑐𝑜𝑠𝑡
of standalone and grid-connected DC MGs. by 31.7240%, 𝐸𝑙𝑜𝑠𝑠 by 29.5019%, and 𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠 by 26.6234%. In
17
L.F. Grisales-Noreña et al. Journal of Energy Storage 65 (2023) 107240
Fig. 11. Reductions obtained by the PALO (compared to the PVSA and PPSO) in the grid-connected DC MG.
Fig. 12. Current and voltage values obtained by the solution methodologies in the grid-connected DC MG.
turn, PPSO takes the first place in this regard because its average re- batteries took advantage of the high solar radiation between 12:00 and
ductions in processing times were 97.7972% (𝐹 𝑖𝑥𝑒𝑑𝐸𝑐𝑜𝑠𝑡), 115.7790% 16:00 h and were charged up to the maximum SOC. Then, as the energy
(𝑉 𝑎𝑟𝐸𝑐𝑜𝑠𝑡), 63.0153% (𝐸𝑙𝑜𝑠𝑠), and 88.0285% (𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠) with respect cost and power demand increased, the batteries expended all the stored
to those of the PALO. Importantly, the PALO required an average energy between 16:00 and 21:00 h. Afterward, during the hours of low
processing time of 108.75 s to improve the economic, technical, and electricity cost, the batteries were charged up to the maximum SOC
environmental conditions of the grid. This is a short time considering before starting a new day of operation.
an entire day of operation of an electrical system. As in the standalone The operation scheme of the batteries produces reductions in energy
case, the PPSO was the fastest method, but it produced the worst losses and CO2 emissions, but they do not depend on the energy
results in terms of the objective functions for the grid-connected DC cost curve. Therefore, the operation of the batteries was defined as
MG. Therefore, it is not considered a suitable solution for the problem a function of peak demand hours, maximum energy losses, and CO2
studied here. emissions. This meant lower PV power generation due to the absence
Finally, Fig. 13 reports the state of charge of all the batteries of solar radiation. The batteries started with little activity between 1:00
(in percentage) when each one of the four objective functions was and 3:00 h. Then, they supplied power to the system from 2:00 to 8:00.
implemented in the grid-connected DC MG. This figure shows that, Subsequently, all the batteries started charging to take advantage of the
when the energy purchasing cost was fixed in the grid-connected DC solar radiation until 15:00. Finally, the batteries started discharging
MG, the batteries did not use the stored energy during the first hours of when there was peak demand and low solar radiation, i.e., between
the day, charging until the maximum SOC was reached when there was 17:00 and 22:00 h. Thus, the batteries were charged at the end of
maximum solar radiation. Therefore, the energy stored in the batteries the day until they reached the final SOC established for these devices.
was used during peak demand hours while respecting the final SOC. This dynamic demonstrates the relationship between battery operation
Other authors can use this information to compare new methodologies and input data (energy costs, power demand, and PV generation), as
in future studies. well as the importance of batteries for energy management—which
When the cost was variable, the charging process was defined as can improve the economic, technical, and environmental conditions of
a function of the energy cost and power demand curves. Hence, the grid-connected DC networks.
batteries were charged when the electricity price was low (see Fig. 5), The results analyzed in this document (including processing times)
using a little power between 6:00 and 11:00 h. After that, all the indicate that the PALO achieved the best performance in solving the
18
L.F. Grisales-Noreña et al. Journal of Energy Storage 65 (2023) 107240
Fig. 13. State of charge of the batteries in the grid-connected DC MG according to the best solution found by the PALO.
19
L.F. Grisales-Noreña et al. Journal of Energy Storage 65 (2023) 107240
required for solving the problem. And (vi) it works with any kind of These results demonstrate the importance of variable energy costs
DC grid or location. when energy storage systems are considered in energy management
Three different parallel optimization methodologies (i.e., PPSO, the systems. This is because energy storage devices should be charged when
PALO, and the PVSA) were proposed for a master–slave strategy that the electricity cost is low and supply power when said cost is high. In
solves the mathematical formulation to find the optimal operation of general, the technical and environmental improvements were similar in
grid-connected and standalone DC MGs (in particular, the charging the grid-connected and standalone DC MGs. The PALO obtained, on an
and discharging scheme of the batteries in the grid). As a result, the average year, a reduction of 37.095 MW in energy losses and 6.093 tons
economic, technical, and environmental conditions of the DC MGs were in CO2 emissions. This demonstrates that using these methodologies
improved. This strategy also included an hourly power flow method promotes an adequate operation of the batteries in a grid-connected
based on successive approximation to evaluate the fitness function. The DC MG, thus improving its technical, economic, and environmental
novelty of this strategy is that it applied parallel versions of the ALO conditions.
and the VSA to the operation of DC grids. The simulation results demonstrated that, to improve the technical,
These solution methodologies were validated using the operation economic, and environmental conditions of standalone networks, the
profiles of an average day of electricity generation and demand in operation of their batteries should be directly based on power demand
a standalone DC MG in Capurganá (Chocó, Colombia) and a grid- and PV generation curves. In turn, in grid-connected networks, their
connected DC MG in Medellín (Antioquia, Colombia). The three operation should be defined based on power demand, energy cost
methodologies mentioned above were validated in these two test sys- curves, and PV generation curves.
tems, executing each algorithm 1000 times to evaluate its effectiveness The simulation results also indicate that grid-connected DC MGs
in terms of minimum solution, average solution, standard deviation, should consider a variable energy purchasing price because it enables
and average processing time. The simulations in the grid-connected and them to reduce their operating cost by 96.24% compared to a scenario
standalone DC MGs, as well as a statistical analysis, established which of fixed energy price. Therefore, in Colombia, it is necessary to promote
solution methodology in particular achieved the best performance for laws and regulations that implement variable costs for all kinds of users:
the three objective functions used in this paper. As a result, grid owners regulated and unregulated. However, current regulations only offer
and operators can design versatile and adaptive energy management this possibility to unregulated users, which are usually big companies
systems that produce the best results (in economic, technical, or envi- and industries. This regulation should also be applied to standalone
ronmental terms) based on the goals that are set for the operation of networks so that their energy management systems can produce more
the microgrid, which could change over its useful life. economic benefits. Furthermore, economic incentives for this type of
Another important contribution of this study is that the mathemat- DC MGs can promote the use of renewable energy resources.
ical model and solution methodologies proposed here can be adapted
to grid-connected and standalone DC networks in any country by using
different economic, technical, and environmental data. This is possible 6.3. Main limitations of this study and future work
because the mathematical model was defined and described in a general
way. Consequently, it can work with grid-connected and standalone DC The main limitation of this work is that the PV distributed gener-
networks, as demonstrated in the Results section of this manuscript. The ators operated at their maximum power point, imitating the regular
input data for these optimization methodologies should be energy and operation of these devices in Colombia. Nevertheless, operating these
maintenance costs; the technical and environmental parameters of the devices between 0 W and the maxim power allowed in each hour could
distributed energy resources considered in the network (DGs and ES); result in more dynamic energy management systems with a stronger
and the type of electrical network under analysis. impact on the objective functions used in each case. Furthermore, this
study did not consider the implementation of a multi-objective func-
tion, which would allow grid operators and owners to obtain benefits
6.2. Main results
in two different objective functions at same time, e.g., operating cost
and environmental conditions. Finally, this study did not consider the
On the one hand, this paper showed that the PVSA achieved the
re-location of DGs or ES, which would increase the economic, techni-
best performance in the standalone DC MG in terms of reductions in
cal, and environmental benefits of the proposed energy management
operating costs (with fixed energy purchasing costs), energy losses, and
systems.
CO2 emissions. This solution methodology was also the best in terms of
Future studies in this field can investigate new optimization method-
solution quality and repeatability because it obtained the best minimum
ologies that offer better solutions (in terms of effectiveness and pro-
and average solutions after the algorithms were executed 1000 times
cessing times) using multi-objective functions based on technical, eco-
each. Its standard deviation values were under 1.12%, and it required
nomic, and environmental aspects of standalone and grid-connected
short processing times (less than 18.29 s on average). In an average
MGs. Finally, their mathematical formulations could integrate the opti-
year of operation, the PVSA would reduce energy costs by 425.8
mal operation of PV generators and other renewable energy resources
USD$; energy losses, by 15.17 MW; and CO2 emissions, by 4.05 CO2
to improve the operating conditions of the DC MG.
tons. These results demonstrate that the energy management system
proposed here obtains excellent results in technical and economical
terms. However, due to its fixed energy cost, a standalone network is Declaration of competing interest
not an attractive scenario for the operation of batteries because these
devices take power from the grid for their operation. The authors declare that they have no known competing finan-
On the other hand, the PALO obtained the best results in the cial interests or personal relationships that could have appeared to
grid-connected DC MG in terms of minimum and average reduction influence the work reported in this paper.
of the objective functions; average standard deviation (0.78%); and
processing times (under 108.75 s on average). In this scenario, the
PALO reduced the operating costs by 1226.40 USD$ when the energy Data availability
purchasing price was considered fixed. In turn, it reduced them by
32671.15 USD$ when said price was variable (i.e., a 96.24% decrease). No data was used for the research described in the article.
20
L.F. Grisales-Noreña et al. Journal of Energy Storage 65 (2023) 107240
21
L.F. Grisales-Noreña et al. Journal of Energy Storage 65 (2023) 107240
[54] L. He, S. Zhang, Y. Chen, L. Ren, J. Li, Techno-economic potential of a [61] J.J. Montano, L.F.G. Noreña, A.F. Tobon, D.G. Montoya, Estimation of the
renewable energy-based microgrid system for a sustainable large-scale residential parameters of the mathematical model of an equivalent diode of a photovoltaic
community in Beijing, China, Renew. Sustain. Energy Rev. 93 (2018) 631–641. panel using a continuous genetic algorithm, IEEE Lat. Am. Trans. 20 (4) (2022)
[55] L.F. Grisales-Noreña, J.A. Ocampo-Toro, A.A. Rosales-Muñoz, B. Cortes-Caicedo, 616–623.
O.D. Montoya, An energy management system for PV sources in standalone [62] C. Altintasi, O. Aydin, M.C. Taplamacioglu, O. Salor, Power system harmonic and
and connected DC networks considering economic, technical, and environmental interharmonic estimation using Vortex Search Algorithm, Electr. Power Syst. Res.
indices, Sustainability 14 (24) (2022) 16429. 182 (2020) 106187.
[56] J. Cao, W. Du, H.F. Wang, S. Bu, Minimization of transmission loss in meshed [63] C.E. Colombiano, NTC 2050, Minist. Minas Energía (1998).
AC/DC grids with VSC-MTDC networks, IEEE Trans. Power Syst. 28 (3) (2013) [64] Instituto Colombiano de Normas Técnicas y Certificación (ICONTEC), Tensiones
3047–3055. y frecuencia nominales en sistemas de energÍa elÉctrica en redes de servicio
[57] O. Montoya, W. Gil-González, L. Grisales-Norena, Linear–based Newton–Raphson pÚblico NTC1340, Bogotá DC (2004).
approximation for power flow solution in DC power grids, in: 2018 IEEE 9th [65] O.S. Velasquez, O.D. Montoya Giraldo, V.M. Garrido Arevalo, L.F.
Power, Instrumentation and Measurement Meeting, EPIM, IEEE, 2018, pp. 1–6. Grisales Norena, Optimal power flow in direct-current power grids via
[58] B. Cortés-Caicedo, L.F. Grisales-Noreña, O.D. Montoya, M.A. Rodriguez-Cabal, black hole optimization, Adv. Electr. Electron. Eng. 17 (1) (2019) 24–32.
J.A. Rosero, Energy management system for the optimal operation of PV [66] O.D. Montoya, L. Grisales-Noreña, D. González-Montoya, C. Ramos-Paja, A.
generators in distribution systems using the antlion optimizer: A Colombian Garces, Linear power flow formulation for low-voltage DC power grids, Electr.
urban and rural case study, Sustainability 14 (23) (2022) 16083. Power Syst. Res. 163 (2018) 375–381.
[59] J. Ocampo-Toro, O. Garzon-Rivera, L. Grisales-Noreña, O. Montoya-Giraldo, W. [67] A. Garces, Uniqueness of the power flow solutions in low voltage direct current
Gil-González, Optimal power dispatch in direct current networks to reduce grids, Electr. Power Syst. Res. 151 (2017) 149–153.
energy production costs and 𝐶𝑂2 emissions using the antlion optimization [68] Z. Hasan, M.E. El-Hawary, Optimal power flow by black hole optimization
algorithm, Arab. J. Sci. Eng. 46 (10) (2021) 9995–10006. algorithm, in: 2014 IEEE Electrical Power and Energy Conference, IEEE, 2014,
[60] M.S. Ismail, M. Moghavvemi, T. Mahlia, Characterization of PV panel and global pp. 134–141.
optimization of its model parameters using genetic algorithm, Energy Convers.
Manage. 73 (2013) 10–25.
22