Energy Reports 9 (2023) 5246–5268

Contents lists available at ScienceDirect

Energy Reports
journal homepage: www.elsevier.com/locate/egyr

Research paper

Optimal deployment of fast-charging stations for electric vehicles

considering the sizing of the electrical distribution network and traffic
condition
∗
Miguel Campaña a,b , , Esteban Inga b
a
Power Grid and Alternative Energies Research Group, Master of Electricity Program, Instituto Superior Universitario Sucre, Quito EC170148, Ecuador
b
Smart Grid Research Group, Master of Electricity Program, Universidad Politécnica Salesiana, Quito 170525, Ecuador

article info a b s t r a c t

Article history: The conventional vehicle fleet worldwide has contributed to the degradation of air quality due to CO2
Received 9 October 2022 emissions. Consequently, it has migrated from internal combustion to electric vehicles (EVs). However,
Received in revised form 1 April 2023 it is essential to ensure the deployment of electric vehicle charging station infrastructures (EVCSI) to
Accepted 19 April 2023
guarantee their interoperability for the development of electric mobility. Moreover, the sustainability of
Available online 1 May 2023
EVCSI depends not only on the capacity to meet demand but also on the adequate number of terminals
Keywords: in the different public charging stations (CS) to reduce waiting times for battery recharging. Then to
EV charging stations achieve an optimal sizing of charging stations, it is crucial to foresee the maximum number of vehicles
Electrical distribution networks that could use the different CS during a time interval. The sizing of CS must respond to real mobility
Georeferenced systems constraints and technical conditions, such as the capacity of vehicular flow, the capacity of the roads
Vehicle flow paths according to their geometry, the trajectories marked by the users, and the possible exit of operation of
Optimization some CS. Therefore, this paper addresses the problem considering four fundamental axes, which are:
Capacitated Multicommodity Flow Problem
stochastic analysis of heterogeneous vehicular flow, a solution to the transportation problem with the
Smart Grid
Smart City
capacitated multicommodity flow problem and Hungarian algorithm, analysis of the optimal number of
terminals considering loading times, and finally the proposed EVCSI will be validated using the CymDist
software for electrical engineering. Consequently, the computational complexity of the model is of a
combinatorial type and is defined as NP-hard given the multiple variables and constraints within the
transportation problem.
© 2023 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND
license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction forecast the demand, and guarantee the continuity of the electric
service, it is crucial to study and simulate the theoretical vehicular
Electric energy supply in charging infrastructures plays an flow in a georeferenced area. This way, it is possible to forecast
essential role in the sustainability of this new non-conventional the resources needed in each CS. It is essential to foresee the
mobility alternative. This new mobility concept contributes di- maximum demand of CS since it would help us to decide if the
rectly to reducing greenhouse gases, noise, and visual pollution electrical network can assume new loads or, if not, it is necessary
compared to conventional mobility. In Europe, it is estimated that to create new supply circuits for each CS. It would contribute
approximately 80% of the noise pollution caused on the streets is significantly to maintaining voltage stability in conventional dis-
generated by vehicles that use fossil fuels as a source of energy. tribution power systems and foresee the expansion of the power
In addition, human mobility plays a fundamental role in cities’ grid. Therefore, planning can be divided into four stages: (i) highly
development and economic growth. However, with conventional stochastic demand prediction, (ii) CS allocation, and sizing, (iii)
mobility, the environmental cost becomes critical and detrimen- minimum routing from demand to a power source to recharge
tal to future generations due to the radical climate change we the battery, and (iv) georeferenced simulation of the distribution
are currently undergoing. The most challenging problem is the power system that will support the charging infrastructure. Con-
supply chain management for EV charging; therefore, planning sequently, two optimizing variables are identified (i) operating
the deployment of CS (Liu et al., 2019b) is essential. To plan, costs and (ii) heterogeneous load distribution in georeferenced
scenarios (Rabbani et al., 2019; Campaña et al., 2021). Finally,
∗ Correspondence to: Smart Grid Research Group, Universidad Politécnica this paper observes a multi-objective approach to address the
Salesiana, Quito 170525, Ecuador. problems of heterogeneous vehicular traffic routing, CS sizing,
E-mail addresses: [email protected] (M. Campaña), and determining the optimal number of charging terminals for
[email protected] (E. Inga). EVs (Asna et al., 2023).

https://doi.org/10.1016/j.egyr.2023.04.355
2352-4847/© 2023 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/).
M. Campaña and E. Inga Energy Reports 9 (2023) 5246–5268

Nomenclature Qo Reactive power when the power supply

voltage corresponds to 1 p.u.
List of variables Vo Voltage when the supply voltage corre-
sponds to 1 p.u.
G Directed graph
P Active power model as a function of
ϕ Vehicular traffic voltage variation
V Vertex set Q Reactive power model as a function of
A Creation of spanning trees voltage variation
Ve # Electric vehicles p1 − p3 Active power parameters
ε Charging stations q1 − q3 Reactive power parameters
ζ Nodes involved in the road network V Supply voltage at p.u.
u, v Source–destination edges respectively
ι Track section length
ϖ Traffic capacity in ι
β Ordered pair of vertices in a track Depending on the number of terminals to recharge EV bat-
section (products)) teries, the service time in the CS could increase or decrease
Wβ Cost per traffic demand significantly. In addition, the number of charging terminals for
S1,2,3... Available routes on the road network EVs may result in the need for more or more petite sizing of
the electrical equipment required in the charging infrastructure.
fbeta Traffic flow
Consequently, for CS sizing, resources are allocated according to
i, j Indexes to relate the road network
the number of EVs expected to be served in a CS hourly. There-
κ Sites selected for CS location fore, the sum of the partial demands generated by each EV must
m Size of vector u, v. equal the resource assigned in the CS sizing problem. Candidate
δκ costcost, in terms of traffic as viewed locations for CS include parks, shopping malls, public–private
from a CS green spaces, and parking lots.
n # Vehicles, observing capacity criteria in Fig. 1 presents the architecture proposed in this article. It
each road section. shows a two-level spatial approach (I) demands based on ve-
ψ Road capacity hicular flow analysis and (II) electrical equipment for charging
M Subsets edges define enabled roads infrastructure. These demands are met by deploying a set of CS
L Subset of edges of a path in permitted public areas such as parking lots, parks, and high
φj Location variable concurrency areas. In addition, the equipment for the charging
infrastructure should consider forecasting vehicle demand based
θk Subset of assigned arcs
on the maximum traffic density generated on the road network.
τej Path enabled
Consequently, the heuristic proposed in this article will solve the
np Number of periods in years allocation of electric equipment in charging infrastructures, con-
q Vehicle density of each commodity sidering real transportation constraints under a spatial approach.
α Number of vehicles in ι Therefore, the objective is to solve multiple resource allocation
ts Time in hours problems in a georeferenced area of NP-hard computational com-
Θ Annual rate of increase in the vehicle plexity, relying on optimization tools, graph theory, and routing
fleet algorithms, thus reducing the computational complexity to a
s Vehicle spacing problem of NP type (Jiménez-Estévez et al., 2010).
Sp Travel speed of each EV
r(S ) Reduced-cost variable 1.1. Contributions and structure
T Indicates the transpose of the matrix
Level I analyzes vehicular traffic in the road network, observ-
χ∗k Optimal dual objective function
ing capacity and level of service criteria, allowing weights to the
ω Dual variable array road network’s different trajectories to enter the optimization
c Costs model. However, it is noted that this is not a trivial problem, and
b Inequality vector b it is a great challenge to solve, given that, in the analysis of vehic-
A Inequality matrix A ular traffic, geographical elements are involved (roads, orography,
x Independent variable overpasses), the dispositions of the vehicles are unknown. There-
π1,2 Dual variables fore, due to their high traffic variability, it is impossible to predict
M /M /s Multichannel queuing model the trajectory patterns with certainty (Jiang et al., 2018; Zhang
M # Available charging terminals and Ioannou, 2018; Song et al., 2019). Consequently, to develop
optimal routes in trajectories, it is crucial to pay special attention
λ Average arrival rate
to aspects that have to do with the variables of vehicular flow,
µ Average service rate in each channel
the probabilistic description of vehicular traffic, the distribution
η Number of EV units in system of vehicles in a road network, and the statistical distribution of
Ws Average time that an electric vehicle vehicular density. In this first level, the microscopic, Hungarian,
remains in the system and multicommodity flow problem models are used to identify
Po Probability the maximum vehicular flow the road network can experience
Po Active power when the supply voltage and direct the traffic to the different CS deployed in the area
corresponds to 1 p.u. of interest. In addition, it is possible to determine the number
of EVs that a CS can serve at the level, considering that one
of them goes out of service of the charging infrastructure. The
5247
M. Campaña and E. Inga Energy Reports 9 (2023) 5246–5268

Fig. 1. Proposed architecture for the deployment of electric charging stations.

exit from service of a CS will cause the vehicular flow to be • The organization and management of vehicular traffic are
reorganized, and the demand covered by the CS that went out solved with the Hungarian algorithm and the multiple-
of service will be directed to the available CS. Considering the product problem; a scalable model is proposed considering
eventual exit of a CS is undoubtedly a significant contribution hourly mobility patterns.
to the optimal sizing of CS and, in general, the charging infras- • The proposed academic model has industrial application
tructure. Level II analyzes the topology generated to achieve the potential for charging station infrastructure providers in
supply of electric power to each CS and thus meet the demand developing countries.
by massive introduction of EV in the vehicle flow management After this, this article is organized as follows: Section 2 briefly
processes (Montoya and Ramirez, 2012; Li et al., 2014; Nagarajan reviews related articles. Section 3 offers the traditional formula-
and Ayyanar, 2014; Neagu and Georgescu, 2014). In level II, the tion of the problem and the methodology to solve it. Section 4
M/M/s queuing theory model determines the number of charging contains the analysis of the results and the validation of the
terminals installed in a CS. Once the number of load terminals proposed model with the study of 2 cases. Finally, Section 5
is known, the transformer that will serve the CS is sized. Once presents the conclusions of this article.
the transformer sizing in kVA is known, the electrical network
is simulated with CymDist software to validate its operability, 2. Related works
considering georeferenced scenarios. Considering georeferenced
This article studies the characteristics of vehicular flow in a
scenarios for the simulation helps determine the conductors’
georeferenced area. It will help us dimension and allocate the
absolute length; with this, the technical losses in the conductors resources necessary for the correct operation of the EVCSI. A
and voltage drops are accurately obtained. The contributions of two-level trajectory approach is presented, considering the high
this article are as follows (Konara et al., 2023). stochasticity implied by heterogeneous vehicular traffic. This pa-
• A theoretical model for creating fast-charging station infras- per is closely related to three lines of research: vehicular flow
analysis in heterogeneous scenarios, allocation of the number
tructures is introduced based on accessible data obtained
of charging terminals for each CS, and optimization of EVCSI
from OpenStreetMap. In addition, the proposal is validated
required in electric mobility. This section summarizes some of the
using CymDist software for the construction and operational
most current and relevant works related to the proposed study.
simulation of the fast-charging infrastructure.
• The algorithm is designed to predict the number of EVs and 2.1. Assignment and location problem
minimize the trajectory of the vehicular flow by employing
a traffic rerouting system, considering the topology of the The location of energy supply centers for EVs has been de-
road network. veloped in Tang et al. (2018), Chung et al. (2018) considering
5248
M. Campaña and E. Inga Energy Reports 9 (2023) 5246–5268

Table 1
Category L7e-C EV characteristics.
Vehicle Autonomy Voltage Engine Maximum System Vehicle category
electric (km) terminal (V) (kW) speed (km/h) regenerative dimensions (m)
Tazzari Zero 200 230 15–25 90 ✓
Renault Twizy 100 230 17 80 ✓ height < 2.5
Audi Urban 73 230–400 15 100 – width < 1.5
Peugeot BB1 120 230 15 60 – length < 3.7

Table 2
Types of charging terminals for EVCSI (Campaña and Inga, 2019a).
Type Current (Amp) Time Recharge Power Owner INEC
Type (h) (%) kW Standard
Slow 16 AC 8 100 4–8 Public–Private
Semi-fast 32 AC 1.150 50–80 22 Public
Fast 63 AC 0.500 50–80 50 Public 61851
Ultra-fast 250–400 DC 0.170 50–80 350 Public
Change-Battery AC–DC 0.033 100 – Public

the routing problem and EV autonomy as decision variables. such as bicycles, cars, and public transport buses. However, it
Table 1 briefly outlines the characteristics of vehicles designed has not been possible to massively introduce EVs into the land
for urban use in the L7e-C category, where the average travel transportation system due to variables that do not make large-
length of 123 km is observed. The decision problem determines scale purchases attractive to potential users in urban and rural
the optimal number of CS, locations, allocation of users to each areas. These unattractive variables for consumers could be limited
CS, and economic dispatching policy. In Bi and Tang (2019), the autonomy, long charging times, battery life, high costs, and lack
problem is extended to a dynamic planning model by considering of EV charging infrastructure.
the itineraries due to EV usage. Recently, in Napoli et al. (2020), In Zhang et al. (2019), the problem is proposed to be addressed
the location problem has been studied by analyzing supply and by modeling mixed vehicle traffic with different charging ter-
demand, including the driver’s psychological component. In Luo minals and observing EV driving autonomy constraints; further-
et al. (2020), Balakrishna et al. (2014), in addition to deploying CS, more, the mathematical formulation and solution are addressed
the integration of distributed generation sources are suggested. by linear programming. In Vaziri et al. (2019), Lu et al. (2019), B
The authors have shown that integrating distributed gen- et al. (2018), Du et al. (2018), Moradipari and Alizadeh (2018), the
eration helps relieve unplanned loads on existing distribution CS allocation and deployment problem is addressed by partially
power grids, which have generally been operating in cities for considering several aspects that guide obtaining robust results;
several years. However, a rigorous analysis that considers road these aspects are: capacity constrained, trajectory probability,
network capacity and vehicle density analysis regarding the topol- travel time, flow conservation, charging time, spatial, tempo-
ogy of the road network defining traffic in cities could not be ral collaboration, terminal type, and traffic events. Planning, CS
demonstrated (see Table 2). deployment, and setting terminal type selection are essential
because EVs present technical constraints such as driving travel
range and limited payload. Terminals have been classified accord-
2.2. Resource allocation under optimal criteria
ing to the charging current the EV can support. The table below
briefly describes the different types of existing terminals and their
Both exact and approximate solution techniques have been
diverse public and private applications (Amry et al., 2022).
applied to solve the resource allocation problem. Accurate so-
As the charging current increases, the time to recharge an
lutions involve integer linear programming problems, whereas
EV battery decreases, and the types of charging terminals are
approximate solutions involve the development of heuristics and
different technology. Types I and II are applied in residential
metaheuristics (Campaña and Inga, 2019b). Table 3 provides the
areas, and type III is exclusively a public/private service. Type III
taxonomy of related studies with a summary of the problem
terminals can reach power ratings of 50–350 kW. Consequently,
statement, objective function, main constraints, solution meth-
this article is essential. It covers the EV charging infrastructure
ods, and other key features to address the vehicular traffic prob- planning problem in its integral form, passing through an analysis
lem, CS assignment, and sizing. of vehicular flow, modeling of high stochasticity, projection of
Recent developments have focused on solving planning prob- vehicle fleet increase with annual rates, and solution of the traffic
lems based on a convex multi-objective model. The model allocation (Singh et al., 2023; Antarasee et al., 2023). Additionally,
depends on the number of variables and constraints in the opti- routing problems and electrical equipment sizing by observing
mization problem. Among the issues that have not been charging models considering electrical simulation of charging
adequately considered are those characteristics that bring the infrastructure in geolocalized areas. Recall that, to size the CS; it
problem closer to real-world situations, such as the heterogeneity is necessary to model the vehicular traffic in trajectory according
of the vehicle fleet, vehicle flow routing, road network capacity, to the user’s need for battery recharging. Therefore, EVCSI must
and the multiple CS options that the user can resort to recharge be considered an essential factor during the planning process.
the EV batteries and be able to continue with his route. In addition, correct planning will allow determining whether or
not the different existing electric grids can assume new loads
2.3. Planning of charging infrastructure for EV due to the massive introduction of EVs. The main objective is
to guarantee the electric service’s quality, continuity, and safety.
The advancement of technology and the constant effort to In Campaña et al. (2017), Inga et al. (2017, 2019), a method
introduce new concepts of electric mobility on a large scale of cauterization and construction of minimum spanning trees
have encouraged the scientific community to develop products is proposed as an indispensable tool in electrical equipment’s
to reduce the CO2 emissions of conventional private and public optimization and planning processes (Chen and Chen, 2023; Kłos
transport. Today there are multiple electric mobility alternatives, and Sierpiński, 2023).
5249
 M. Campaña and E. Inga
Table 3
Taxonomy of related studies.
Model Problem Solution Trajectories
Author, year Theoretical Experimental Location Route Obj. function Constraints Exact Heuristic Meta-heuristic Density Urban Others
Allocation Traffic study
Gan et al. (2020) ✓ – ✓ ✓ Maximize Distance – ✓ – – – Elastic
Profit Service Standby time demand
Power grid Stochastic
capacity Newton–Raphson
NILP
Esteban et al. (2019) ✓ – ✓ ✓ Minimize Capacity – ✓ – – ✓ Scalable
Costs Distance multi-commodity capacities network
Flows
Demir et al. (2019) – ✓ – ✓ Minimize Flows – – ✓ – – Deterministic
Travel distance # Vehicles ILP
Costs Demand
fluctuation
NP-hard
Azadi Moghaddam Arani et al. (2019) ✓ – – ✓ Maximize Capacity ✓ – – ✓ – Multiple commodity
User Operation flows network
Routes Service Schedule Heterogeneous traffic
MILP
Branch and Bound Algorithm
Bevrani et al. (2019) ✓ – – ✓ Minimize Congestion ✓ – – – ✓ Multi-commodity network flow
Flow Expansion tree Flow reduction functions
5250

Routes Conservation NLP

flow Multiple nodes
Campaña and Inga (2019a) ✓ – ✓ ✓ Minimize Distance – ✓ – ✓ ✓ ILP
#Charging stations Minimal tree Graph theory
Trajectory Coverage Homogeneous nodes
Costs
Liu et al. (2019a) ✓ – – ✓ Minimize Capacity ✓ – – ✓ – Heterogeneous nodes
Power losses Congestion Optimal Solution
in distribution Charging congestion during
Maximize Flow fuzzy multi-objective model
Ghasemi et al. (2019) ✓ – ✓ ✓ Min. Transportation cost Demand ✓ ✓ ✓ ✓ – Probability approach
Minimize shortages Heterogeneous flow Planning periods
Max. humanitarian Capacity Multiple-objective PSO
Distance Stochastic demand
Campaña and Inga (2019c) ✓ – ✓ ✓ Min. Trajectories Coverage – ✓ – ✓ ✓ Segmentation
Min. Charge station Distance ILP
Min. Congestion road Capacity Need for Users

Energy Reports 9 (2023) 5246–5268

Neighborhood analysis
Current study ✓ – ✓ ✓ Min. Trajectories Capacity ✓ ✓ ✓ ✓ ✓ Revised-Simplex
Min. flow Distance MILP
Min. Charge stations Coverage Multi-commodity network flow
Max. Humanitarian Demand Heterogeneous nodes
Scalable
Multi-objective
NP-Hard
Non convex
M. Campaña and E. Inga Energy Reports 9 (2023) 5246–5268

3. Optimization problem description proposed model can solve the complex combinatorial problem
of allocating resources for electric vehicle charging in real time,
This section describes the problem through stochastic anal- considering traffic flow constraints. The methodology proposed
ysis of vehicular trajectory traffic and resource allocation for in this model also considers the strategic location of charging
EVCSI. In addition, this section presents the proposed strategy and centers in places such as shopping malls, parks, and parking
methodology, followed by the formal problem statement. lots, which ensures comprehensive and efficient coverage of the
charging infrastructure. In addition, using Dijkstra’s algorithm to
3.1. Stochastic analysis of vehicular traffic and resource allocation at find the shortest route for vehicles based on traffic density, along
EVCSI with simulation in MATLAB and CymDist, ensures the effective-
ness of the proposed model.
EV refueling stations in the public–private transport sector The model proposed in this study is relevant for scientific
are essential because they supply the primary energy to provide research and industrial application in electric vehicle charging
autonomy to the non-conventional vehicle fleet. It is necessary infrastructure. The methodology presented in this model has a
to plan for the massive introduction of EVs in daily traffic, which quantitative approach to allocating resources and minimizing
implies having heterogeneous vehicle networks where the main economic and social impacts. The model is highly flexible and
actors are conventional vehicles and public and private sector can be adapted to different scenarios and locations. The results
EVs. obtained from the simulation are based on geo-referenced data
Most of the research focuses on users and travel time reduc- from a free and public database, which increases the reproducibil-
tion by reorganizing vehicle traffic according to road capacity and ity and transparency of the study. In summary, the proposed
considering heterogeneous traffic networks. A fundamental detail model can potentially be a valuable resource for EV charging
is that the demand for vehicle charging is not centered on the infrastructure providers and urban planners in building efficient
nodes (electric or conventional vehicles) but on the flow of the and sustainable charging infrastructure.
vehicular traffic network. CS deployment planning can consider the optimal location of
Therefore, the vehicular traffic flow can simulate the demand different charging infrastructures and their influence on power
generated by EVs exposed in a heterogeneous road network. quality, operational safety, and the economics of system oper-
Consequently, the research aims to solve the problem of resource
ation. Consequently, it is defined as a typical multi-objective
allocation, traffic routing, and sizing of electric equipment for CS,
optimization problem. This paper will solve the planning problem
considering the heterogeneous density of vehicles and their high
by considering a multilevel: (i) traffic network and (ii) power
stochasticity.
distribution network. These categories may overlap, which means
We assume that installing a service center for EV battery
that two nodes of each type may be located in the same area.
charging cannot meet the generated demand due to the limited
Therefore, at level 1, the traffic network will characterize the flow
capacity of the CS on a stretch of road. This limitation will cause
of vehicles. Finally, at level 2, the capacity in each CS will be
critical traffic flow due to high coincidence factors for the battery
studied, considering the number of EVs to be served.
recharge requirement. Consequently, battery capacity is essential
for CS deployment and resource allocation. Therefore, we will
3.2. Proposed strategy and methodology
obtain the following considerations: (i) if there is a CS between
a start point and an endpoint of a path, the EV autonomy should
This paper proposes a theoretical model for resource allocation
be greater than the path length, and (ii) if there is only one CS on
and sizing of CS with traffic flow constraints in real hetero-
the path, the EV autonomy should be more than twice the path
length. geneous scenarios. The main objective is to optimize the re-
An important detail stated so far is that the algorithm knows sources of the loading infrastructure. A non-convex combinatorial
the multiple paths EV can take regarding Spatio-temporal con- problem is proposed under a mixed-integer linear programming
straints (distance, path, and flow variability). Furthermore, we structure. Accordingly, the column generation algorithm solves
consider that the maximum demand for battery recharging in a the multi-product flow problem frequently used in MILP.
given area will be provided by the maximum density of vehicles The column generation algorithm divides the problem into
circulating in a given time interval, which can be hourly. This two stages (i) restricted master problem (RMP) and (ii) subprob-
maximum demand or traffic density occurs when the distance lem. By solving the RMP with the minimum number of variables,
between cars and their travel speed is the minimum. we obtain the dual costs for each constrained master problem.
Based on microscopic models for the analysis of vehicle den- This solution is used in the objective function of the subproblem
sity, we study the maximum admissible density in a georefer- and solved. If the aim of the subproblem value is a negative
enced area considering the variables of spacing between vehicles reduced cost variable, it is added to RMP and cracked again. This
and the speed at which they travel on the road network. Using a process is iterative, and the stopping criterion occurs when the
random vector, under Poisson distribution, we generate the vari- objective value of the subproblem is greater than or equal to
ables of speed entry, vehicle spacing, and lengths of the vehicles zero. When this happens, the constrained master problem can be
participating in the road network. considered optimal.
Recall that we can identify whether it is a minibus, bus, or light In the first level, it is assumed that the driver will choose the
vehicle by their physical lengths, i.e., by introducing a random shortest path from the starting point (current location) to the
vector with sizes of the different types of cars, we can con- destination (CS location) by observing the distance and traffic
sider the diversity of vehicles that are part of the road network. flow constraints on each stretch of road. In addition, strategic
Consequently, the heterogeneity of cars participating in the road places such as shopping malls, parks, parking lots, customer ser-
network is evaluated to determine the maximum vehicle density vice centers, and conventional charging stations are considered
in a time interval and thus allocate resources and dimensions to for CS placement.
the EVCSI electrical equipment. At the second level, the CS capacity must meet the hourly
The proposed model for resource allocation and sizing electric demand for EV charging, determined by the type of terminal
vehicle charging centers is a significant breakthrough in charging or technology used for charging. Previous paragraphs mentioned
infrastructure optimization. Using a mixed integer linear pro- that the kind of charging depends on the batteries’ current capac-
gramming structure and the column generation algorithm, the ity. Also, the higher the present, the shorter the time it takes to
5251
M. Campaña and E. Inga Energy Reports 9 (2023) 5246–5268

Fig. 2. Methodological flowchart for the construction of MCFP-EVCSI.

charge 20% to 80% of the total capacity of the car battery. Differ- their destinations or CS, and (ii) the capacity of each intersection
ent CS locations and accommodations will cause other impacts of each road segment; this means that the traffic flow at each
on the electrical distribution network. Additional detail is that intersection does not have to exceed its capacity. Therefore, the
the Dijkstra algorithm is used to find the shortest path among multi-product flow problem combines several single-product is-
the multiple routes enabled to generate vehicle network routing sues, given the limited road capacity and the diversification of
based on vehicle density. heterogeneous vehicle density.
Fig. 2 shows the methodology for solving the resource allo- Moreover, the interaction between EVs denotes multiple com-
cation problem and sizing EV charging infrastructure. Four main plexities to solve the single-product flow problem independently.
study groups are proposed to guide the methodological devel- Consequently, the demand generated by each EV is related to
opment of this article: (i) scenario analysis, (ii) process, (iii) various supply options for this demand, and its origin–destination
simulation, and (iv) solution. In the analysis of scenarios, the node pair, which refers to the position of EV and CS, can identify
topology information of the road network of the study area is it. The CS located in the study area provides the battery recharge
extracted. In addition, different CS locations in a study area are supply.
known, and the nodes where the demand is generated to recharge In addition, we assume that the available route based on traffic
automobile batteries. Fig. 2 reveals that the georeferenced data flow and path capacity can acquire the supply of commodities to
are extracted from a free database published in OpenStreetMap. meet the demand generated by EVs. The optimal supply to the
Once the scenario is characterized, the experimental proce- CS is selected based on the optimal number of charging terminals
dure defined by the process and simulation detailed in Fig. 2 by observing Spatiotemporal collaboration, which involves mini-
is carried out. This process explains the actions, considerations, mum displacements with traffic flow constraints on the transport
and variables in the simulation. MATLAB and CymDist software network.
is used for the simulation. After the iterative process, the pro- Let us consider G(ϕ, V , A) a directed graph where ϕ corre-
posed heuristic returns the maximum vehicle density in the road sponds to the variable cost of vehicular traffic from an EV to
network during an hourly time interval. This data determines a CS; the distance calculation is performed with the equation
the optimal number of loading terminals required in the road haversine and is given in km. The set of vertices V represents the
network. Finally, this article is considered basic research because multiple nodes located in the study area as electric vehicles (Ve ),
of its experimental nature by applying high-potential simulation charging stations (ε ), and intersections (ζ ), therefore, Ve , ε, ζ inV
processes for EV charging infrastructure providers’ industrial ap- the displacement is given between two source–destination nodes
plications. It has a quantitative approach to determining resource Veu,v , εu,v , ζu,v inV and the set of edges u, v in A represent the
allocation and minimizing economic and social impacts. This origin and destination respectively.
model has an exploratory character, so it is essential to analyze A relates the multiple nodes existing in G resulting in the
different case studies. creation of spanning trees at each level. Each track section (ι) has
a given number of vehicles (α ). Each stretch of road is limited
3.3. Problem formulation in its traffic capacity (ϖ ) and is formed between two vertices at
a set of intersections (β ). The traffic demand or cumulative cost
This section presents some considerations and a formal de- (Wβ ) is reflected in each β . The available routes (S1,2,3... ), where
scription of the problem statement. The multiple product flow S ∈ V , are formed from a set of intermediate nodes β , which,
problem considers two necessary constraints: (i) travel demand, dimension the total vehicular flow fbeta by assigning a weight Wβ
which means that all products or EVs must be transported to on each path.
5252
M. Campaña and E. Inga Energy Reports 9 (2023) 5246–5268

A section of track can be unidirectional or bidirectional βi,j ̸ = Objective function:

βj,i . In other words, β ∪ S ∈ V . Let us consider that κ ⊆ V is
a subset of sites selected to locate CS where κ ∈ / β, Ve because t
∑
the sites selected to locate CS are assumed to be parks, shopping Min:Z(q) = qk θk (6)
malls, public parking lots. k=1
In addition, u = {i ∈ V : ∃β ∈ S | i = 1, 2, 3...m} denotes Subject to:
the subset of the origin nodes in a trajectory and v = {j ∈ V :
t
∃β ∈ S | j = 1, 2, 3...m} denotes the subset of destination nodes, ∑
where, m is the size of the vector {Ve ∪ ε ∪ ζ ∪ β}. θki = 1, i = 1, 2, . . . , u(k) (7)
k=1
In other words, the different trajectories (products) are con-
t
structed only from the β routes, which are associated with ζ ∑
vertices, where the maximum vehicle density generated in β is θkj = 1, j = 1, 2, . . . , v (k) (8)
∑m ∑ m
j=1 βi,j , ∀(i ̸ = j), ∈ Z .
+ k=1
reflected. This is possible, since, i=1 t
Finally, let us consider δκ the cost, in terms of vehicle density ∑
seen by a CS located in the road network ζj where j ∈ κ and n θki = fβ |ik , i = 1, 2, . . . , u(k) (9)
k=1
is the maximum number (capacity) of vehicles participating in a
t
longitudinal section of road (ψ ) of each CS. ∑
Consider M ⊆ A subsets of assigned edges defining the paths θkj = fβ |jk , j = 1, 2, . . . , v (k) (10)
enabled in the road network. Therefore, it is M = βi,j , βi ∈ u and k=1
t
j ∈ κ . We consider L ⊆ A to be a subset of edges of a path Li , Lj ∑
where i, j ∈ {κ ∪ β} with this we guarantee that there are no arcs θk = ϕ (11)
k=1
from other families. Next, we define the set of binary resource ∑
selection variables. θk ≤ |S | − r(S ), ∀S ⊆ V , S ̸ = 0 (12)
k∈S
• φj allocation variable ε, is 1 when it is found any node that
is part of the set j, j ∈ κ , 0 in any other way. θk , ∈ {0, 1} (13)
• θk subset of assigned arcs, if the arc k ∈ M it is used,
Eq. (6) corresponds to the objective function. Eqs. (7) and (8)
otherwise 0.
ensure that vehicular flow is directed from the service nodes
• τej path enabled, is 1 if the edge e ∈ L and j ∈ κ , otherwise to the demand nodes. Eq. (9), (10), and (11) correspond to the
0. vehicular concentration restrictions for the corresponding route,
S observing capacity criteria in each road section. Eq. (12) guar-
∑ ( veh )
fβ (m) = Wβ (m) km
(1) antees the search for the dual variables and their aggregation.
m=1 Finally, equation (13) declares the binary variable. Furthermore,
Ve np qk = [q1 , q2 , . . . , qm ]T where k = 1, 2, . . . , t and represents
∑ ∑ q(n) ( veh )
the vehicle density of each commodity. The cumulative capacity
Wβ (m,n) = (2)
of the EV is represented by fbeta = [Wbeta1 , Wbeta2 , Wbeta...m ]T ,
km
Sp(m)
m=1 n=1
np
and ϕ = [ϕ1 , ϕ2 , . . . , ϕ...m ]T respectively. The adjacency matrix
∑ α(m) ( veh ) between vertices and edges is represented by G = [Gij ]m×m . The
q(m) = hra
(3)
ts length of m × m is a function of the total number of vertices
m=1 j
in the scenario. Finally, fβ |ik = [fβ |1k , fβ |2k , . . . , fβ |m k ], and fβ |k =
np
∑ [fβ |1k , fβ |2k , . . . , fβ |m ] con k = 1, 2 , . . . , t is defined as follows.
α(m) = ϖ(m−1) ∗ Θ (v eh) (4) ⎧
k
i
m=1 ⎨ −Wβ |k
⎪ si i = u(k)
i = v (k)
A i i
∑ ι(n) + s(n) fβ |k = Wβ |k si (14)
⎪
⎩ 0
np
∑ s(n) (5) any other case
n=1
ϖ = (v eh)
⎧ j

m=1
np ⎨ −Wβ |k
⎪ si j = u(k)
j j
fβ |k = Wβ |k si j = v (k) (15)
Eqs. (1) and (2) are used to calculating the total and partial ⎪
0 any other case
⎩
concentration of vehicles in a specific length. Eq. (3) expresses a
vehicle flow on a stretch of the road network that corresponds to
3.3.1. Column generator
the frequency with which a given number of vehicles pass in a
In this article, we are going to use the column generation
specific time. Eq. (4) calculates the average number of cars in a
method. This method allows us to solve extensive linear pro-
lane as a function of the length of the road cross-section and the grams. To do so, we consider the following assumptions: (i) the
spacing between vehicles. Theta is an annual rate of increase in variables are non-basic, (ii) the variables will assume a value
the vehicle fleet. greater than or equal to zero in the optimal solution, (iii) an
It is assumed that the spacing between vehicles is uniform initial subset of variables must be considered to solve an initial
and that the maximum capacity of a road cross-section occurs problem and (iv) the problem can be divided into a primal and
at the minimum distance between vehicles and the minimum dual problem. The idea is to take advantage to generate variables
speed. Finally, the equation (5) is used to calculate the number of that have the potential to improve the objective function. The
cars as a cross-section length and spacing function. It is essential process of the algorithm is as follows.
to mention that a microscopic model has been used for the The primal problem is solved with a subset of variables con-
analysis of traffic flow, which considers ι, Sp, and s. Therefore, sidered; with this solution, we obtain the dual costs for each
the multi-product transportation problem can be formulated as constraint of the primal problem. We assume that χ∗k is the
follows: optimal value of the dual objective function that was solved by
5253
M. Campaña and E. Inga Energy Reports 9 (2023) 5246–5268

the revised simplex method; if the value of the dual objective is Subject to:
a negative reduced cost such that χ∗k < 0, ∀k = 1, 2, . . . , t this t
variable is added as a column to the primary problem and the
∑
θki = 1 (28)
problem is resolved iteratively.
When χ∗k ≥ 0, ∀k = k1 , k2 , . . . , kp , the algorithm stops, and we
k=1
t
can conclude that the primal problem is optimal. For each linear ∑
problem, there is a problem that is solved in parallel. The latter is θkj = 1 (29)
known as the dual problem. We consider the following issue to k=1
∑
solve the primal and dual problems. θk ≤ |S | − r(S ), ∀S ⊆ V , S ̸= 0 (30)
z = cx, (16) k∈S

Subject to: θk ∈ {0, 1} (31)

Ax ≤ b (17) Eq. (27) is the objective function of the secondary problem, so

χ∗k is the reduced cost variable, provided that χ∗k ̸= 0 can find
x≥0 (18) dual variables to be inserted in the main problem. Eqs. (28) and
The linear problem of Eqs. (16) and (17) obtains a dual coun- (29) are constraints that allow CS and points where demand is
terpart. generated to search for the best route within the road network.
In Eq. (30), the coefficient r is selected from the set S that
z = c ω, (19) will allow verifying the cost of the negative reduced variable.
Finally, the binary selection variable is declared with equation
Aω ≤ b (20)
(31). Fig. 3 shows the flowchart for resource allocation and sizing
ω≥0 of electrical equipment required in EVCSI.
(21)
Eq. (19)–(21) have the same terms, except for the time ω = 3.3.2. Optimal number of load terminals
CBV B− 1, which is a variable that is assigned to the equation (17) This section describes the methodology used to determine the
of the first formulation, where ω = CBV B− 1, which corresponds optimal number of charging terminals in each service station for
to an iteration of the revised simplex method and is known as recharging batteries in electric mobility. The charging terminals
a dual variable array. The variables CBV B− 1 and B− 1 correspond are in charge of providing the recharging service for batteries in
to the coefficients of the essential variables and the connectivity
electric mobility. Such service is stochastic and conforms to queu-
matrix relating vertices and edges, respectively. Consequently,
ing systems, specifically M/M/s systems. M/M/s systems consider
the problem illustrated in Eq. (6)–(12) can be separated into
multiple service stations and waiting times in the queue before
two primary and secondary issues, respectively. The equations
being served by a charging terminal. Consequently, the M/M/s
defining each primary and secondary problem are presented as
follows. system, applied in queuing theory, describes several customers
expected to arrive, followed by a queue generated before being
Primal–Primary Problem served by an available loading terminal. Once filled, the customer
Objective function: exits the system as illustrated in Fig. 4.
t
∑∑ The arrival of EVs to the CS is an independent event and satis-
Min: z(q) = qk Ψp θpk (22) fies stationarity conditions, Markov’s properties, and universality.
p∈β k=1 Consequently, the appearance of EVs at the CS is given according
to the Poisson process, and the time intervals between vehicles
Subject to:
assume a negative exponential distribution. For the output sys-
t
∑ ∑ tem, the charging time for each EV depends on the remaining
Ψp θpk = fβ |ik , i = 1, 2, . . . , u(k) (23) energy in the battery, so a negative exponential distribution is
p∈β k=1 used to describe the charging time. Another aspect to consider
t
∑ ∑ is that the charger can only serve one vehicle at a time and
Ψp θpk = fβ |jk , j = 1, 2, . . . , v(k) (24)
is serviced on a first-come, first-served basis, as illustrated in
p∈β k=1
t Fig. 4. On the other hand, the charging station can serve more
than one electric vehicle simultaneously because of its M/M/s
∑ ∑
Ψp θpk = ϕ (25)
p∈β k=1
characteristic for s > 1.
The system is affected by the initial state and the time elapsed
∑
Ψp = 1 (26) since the start of a transient condition. After sufficient time has
p∈β
elapsed, the system becomes independent of the initial state, and
The new set β is used in each iteration to identify the gener- the time elapsed becomes steady. Another consideration is that
ated routes, and Ψ is a partial solution to the primal problem. the number of customers entering the system must equal the
Eq. (22) is the objective function of the primal problem, and number of customers leaving. Consequently, the probability that
Eq. (23)–(25) are the vehicle concentration and road capacity con- satisfies this condition is:
straints, respectively, which allow restricting the solution space
to routes that only meet a little value. Eq. (26) guarantees a Po = [∑ ( )η ]
1
para M µ > λ
M −1 1 λ
( )M
λ
+ M1 ! µ Mµ (32)
η=0 η! µ M µ−λ
non-zero solution to the secondary problem.
η
Dual-secondary problem λ= station opening time (33)
Objective function:
Min: µ= 1
average load time (34)
t t t t
(27) Where M is the number of available charging terminals, λ is
∑ ∑ ∑ ∑
χ∗k = qk θk − π3 θki − π2 θkj − π1 θk − π0
k=1 k=1 k=1 k=1
the average arrival rate, µ is the average service rate in each
5254
M. Campaña and E. Inga Energy Reports 9 (2023) 5246–5268

Fig. 4. M/M/s queuing system scheme.

of costs of the feeder; additionally, thus the technical losses that

may affect depending on the longitudinal variable of the feeder.

3.3.3. EV charging model

The massive introduction of EVs into the electric distribution
system causes an increase in the electric current flowing through
the conductors of the distribution network. For correct planning
of charging infrastructures, it is necessary to guarantee the avail-
ability to recharge the batteries of electric vehicles when the EV
operator requires it. In this context, the present proposal seeks
to determine the maximum power that should be installed in
each CS so that the electric supply is guaranteed at the moment
required by the non-conventional mobility user. Therefore, it is
necessary to mathematically represent the characteristics of the
distributed loads. The load model represents the mathematical
relationship between the magnitude of voltage and active and
reactive power flowing in the feeder. In this paper, the CymDist
software is used, which allows simulation of the load under
two methodologies (i) exponential and (ii) polynomial. This pa-
per will use the polynomial model to determine the EV loading
model. Then, to select the polynomial model, it is necessary to
define coefficients and exponents. The exponents specify the load
type, such as constant power, constant current, and constant
impedance, while the coefficients define the relative proportion
of each load type. The active and reactive power expressions de-
scribed in the polynomial model (ZIP) are represented by Eqs. (36)
and (37), respectively.
( )
P = Po p1 V̄ 2 + p2 V̄ + p3 (36)
( )
Q = Qo q1 V̄ 2 + q2 V̄ + q3 (37)
V0 , P0 , Q0 represent the voltage, active and reactive power
when the supply voltage corresponds to 1 p.u in Eqs. (36) and
(37), p1 − p3 and q1 − q3 are parameters of the models that
Fig. 3. flowchart for solving the multicommodity flow problem for developing
electric vehicle charging station infrastructures. (MCFP-EVCSI).
when approaching 1 imply that the load behaves as a constant
impedance, a continuous current consistent or a regular power,
respectively. The independent parameter V of (38) is the p.u.
Supply voltage. If the actual supply voltage is equal to the voltage
channel, and η is the number of EV units in the system. Finally,
V0 , then V will also be equal to 1, as shown in (38).
Ws defines the average time an EV stays in the system.
V
( )M
λ V̄ = V0 (38)
µ µ
1 (35)
Ws =
(M −1)!(M µ−λ) 2 Po + µ The coefficients defining the polynomial static model are shown
Through Ws and Po curves as a function of the number of in Table 4 below.
terminals M, the optimum number of chargers is selected. Con- Finally, it is essential to model the load represented by the
sequently, it is possible to determine the CS topology and the electric vehicle on the grid, not only because of the increase
maximum installed power to be foreseen for CS operation. An- in current that this means but also because the electric charg-
other detail is that once the installed power and the topology of ers include in their model a harmonic current source based on
the charging stations are known, the feeder that will connect to a three-phase converter with a 12-pulse bridge, which repre-
the primary distribution transformers is laid through minimum sents the non-linear characteristic due to its power electronic
spanning trees. Therefore, It is possible to guarantee the reduction components.
5255
M. Campaña and E. Inga Energy Reports 9 (2023) 5246–5268

Fig. 5. Case A, Grover Beach with an area of 2.13 km2 , a city in San Luis Obispo County, US state of California.

Table 4 Table 5
EV charge static load model parameters for fast charger (480 V) (Tian et al., Simulation parameters.
2021). Deployment Density EV Variable
Coefficient Parameter Study cases A–B
p1 −0.1326 Study areas 0.99 y 2.13 km2
Active power p2 0.1816 Geographical area Urban
p3 0.9495 Geographic reference Latitude–Longitude
q1 – Allocation Annual rate Variable
Reactive power q2 – Scalability 4 years
q3 – Spacing Variable
Application Charging terminal AC–DC
Safety distance 3–6 m
Vehicular flow Variable
4. Analysis of results Vehicular concentration Variable
Vehicular speed [20 40 60] km/h
This section presents the results obtained by the MCFP-EVCSI Traffic Light 4.3 (m) ≤
heuristic model for the creation of EVCSI. The microscopic model Buses 15 (m)
has been used to simulate the input variables, which allows for Trucks 12 (m)
studying the relationship between speed, length, and separa-
tion between vehicles participating in the road network. Table 5
shows the simulation parameters, where the need to consider Fig. 6 presents the variation of vehicle concentration as a
the input variables is observed by analyzing heterogeneous and function of different incremental rates over 20 years. The max-
scalable vehicular traffic in georeferenced areas. In addition, as imum number of vehicles circulating in each case study is de-
this article is experimental, two case studies are developed to termined based on the maximum vehicle density. The total rate
validate the proposed heuristic. In this way, it will be possible corresponds to a theoretical value, which the designer of the
to dimension CS for EVs and allocate resources for energy supply vehicular network can adjust. Its value will depend on the studies
in charging infrastructures. and projections estimated in each zone of its different central
administrations.
4.1. Case study A The randomness of the scenario has been recreated using the
discrete Poisson probability, as suggested in the literature. The
Case study A is illustrated in Fig. 5, and it can find the topology analysis of the increase in the vehicle fleet has been performed
of the road network to be analyzed in this case study. We have over 20 years with four annual rates Θ as shown in Fig. 6. There is
important information such as (i) connectivity matrix, the direc- a technically linear trend in the first ten years of analysis, which
tion of the road, and weights assigned for each section included does not occur after the eleventh year. The slopes cease to be
in the optimization model. Consequently, the longitudinal rela- constant as the annual rate of the vehicle fleet increases. Another
tionships from an intersection i to a meeting j are known to the interesting detail revealed by Fig. 6 is that, as the average speed
model. Considering the actual data in each case study will allow at which vehicles circulate increases, vehicle density decreases,
us to approach reliable solutions and, in addition, will let us know i.e., they are inversely proportional. In purple (see Fig. 6) rep-
the maximum capacity of the road network in each cross-section. resented the empirical values that show the number of values
5256
M. Campaña and E. Inga Energy Reports 9 (2023) 5246–5268

Fig. 6. Case A: Microscopic traffic flow analysis using discrete Poisson probability.

Fig. 7. Vehicular traffic scalability with the incremental rate of 3% for case
studies A and B.

above and below the average value of the three analyzed speeds
in the 11 to 20 years. It means that approximately 95% of the
data are above and below the average. Its importance is visualized
in Fig. 6. Therefore, this article considers speed, geometric road
layout, spacing, and annual growth rates as fundamental variables Fig. 8. Vehicle density distribution located in geolocated scenario case A.
to achieve results that satisfy optimal solutions in real scenarios.
Fig. 7 illustrates the number of vehicles distributed over 24
hours for study cases A and B. The trend in Fig. 7 corresponds to rate. In addition, Fig. 7 shows that the highest vehicle density is
the weekly average of vehicular traffic behavior in its standard
distributed from 07:00 to 20:00 hours. Consequently, charging
working day conditions. The model proposed in this article is
stations should be designed to meet the demand during peak
evaluated over four years from year zero, and an incremental rate
traffic hours in a given area.
of 3% is considered for both case studies. This cumulative rate
has been taken as a reference from the literature. Therefore, this With Fig. 8, we illustrate, through the frequency graph, the
variable is adapted to any reality or case study that is specifically random distribution of vehicle density in an area in trajectories.
required. With the heat graph in Fig. 8, we can appreciate the areas of
The hourly variability of the number of vehicles circulating higher vehicle density represented by the yellow and red vertices.
in a given area responds to a normalized random vector consid- The adjacency matrix allows us to understand the data dis-
ering the original trend (year zero) and the annual incremental tribution within the case study. Fig. 9 shows two techniques for
5257
M. Campaña and E. Inga Energy Reports 9 (2023) 5246–5268

defines the vehicle density in Fig. 7. This inverse behavior is

linked to operating costs; this means that the design and sizing of
electric vehicle charging stations must function the peak demand
from 07:00 to 20:00 hours on ordinary days from Monday to
Friday. Therefore, the objective function has a minimum cost in
high-demand hours because it manages to justify the sizing of
the CS, and the cost of the objective function is maximum where
traffic demand is minimum; it is because, in those low-demand
hours, the recharging infrastructure is not used, so, somehow
there is an over-sizing in low demand hours.
Table 6 shows the resulting metrics for case study A. In each
column, respectively, the ID of the charging station, the geograph-
ical coordinates in latitude–longitude, the vehicle concentration,
and the number of vehicles in the hour of maximum demand
can be seen. The charging station with ID 4 should serve a max-
imum number of 49 cars during the period of maximum vehicle
demand. It is essential to remember that the analysis assumes
100 percent EV penetration in the road network. Furthermore,
Fig. 7 shows that the maximum demand in year 0 occurs at 19:00
Fig. 9. Case A: Distribution of connectivity matrix and its nodal relationship.
hours with 1151 vehicles distributed in the geographical area. The
load distribution of the cars is to be managed by each charging
station during the 19:00 hour of year 0. It is presented in Table 6.
reordering the sparse adjacency matrix: (i) the inverse Cuthill– The distribution of vehicle charging for the rest of the day is
McKee algorithm and (ii) column permutation. In solid blue color given by a percentage pattern that creates the trends in Fig. 7.
in Fig. 9 illustrates the inverse Cuthill–McKee algorithm, which However, in this case, we have considered the maximum demand
has a clear tendency to group its non-zero elements closer to the for vehicles in a time interval to estimate the maximum electrical
diagonal, preserving at the same time, the special relationship of power installed in the different CS distributed in the study area.
each edge (i, j) with its nodes i = j and j = i. For its part, the col- The installed electrical power of a CS should be directly pro-
umn permutation algorithm illustrated in Fig. 9allows reordering portional to the number of charging terminals required to meet
the columns of the sparse-type matrix in non-decreasing order the demand for EV battery charging. In addition, the number of
of non-zero count. This metric enables visualizing the different charging terminals will depend on the technology (50 MW–350
randomly assigned magnitudes of vehicle density that will be MW) and the number of vehicles requiring the service during
considered to evaluate the model proposed in this paper. There- periods of high demand. For example, if we consider the time
fore, Fig. 9 shows that the input data are stochastic, resulting from 19:00 in Fig. 7 and look at Table 6 illustrating the distribution
a microscopic analysis that allows us to understand vehicular of vehicle charging, we observe that the charging station with ID
traffic. Thus, we use the tiny model to solve MCFP-EVCSI as a 4 should serve 49 vehicles for 1 hour.
directed graph to model the traffic rate and the maximum vehicle
concentration at level 1. 4.2. Case study B
Fig. 10 illustrates the algorithm’s performance evaluated in
a scalable scenario from year 0 to year 4. Fig. 10(a) shows the Next, the model’s adaptability to a new case study is verified.
machine time to solve the primary problem. It is easily observed This case study is in Arroyo Grande County, San Luis Obispo
that the machine time used to solve the primary problem de- County, California, United States. In addition, they will analyze
pends on the time interval in which the analysis is developed; what happens with the different vehicle flows in each hourly
this response to the stochasticity of the vehicular flow in each interval with a growth rate of 3.0% and verify contingencies of
time interval during working days. Furthermore, with Fig. 10(a), up to N - 2. The contingency refers to the departure from the
the trend of machine time as a function of the annual growth rate operation of a CS. The scenarios are simulated in which CS has to
of the vehicle fleet is increasing with a minimum of 3 min in year go out of the process due to any event. Finally, the effect caused
0 and a maximum of 4.7 min in year 4. by a vehicular flow that each CS attend is analyzed.
The linear problem is constructed as the primary concern. Fig. 11 shows the geographic area in which case study B will
Therefore, the machine time is greater than the time spent on be analyzed. There are nine CS for VE identified with labels. An
the second problem, as illustrated by comparing Fig. 10(a and additional detail visualized in dark green is the bi-directionality
b). With Fig. 10(c), the observer’s machine time is presented, of each road section and how it relates to each intersection. In
which has the function of deciding on the growth of the column addition, Fig. 11 identifies the costs directly associated with the
generator, i.e., it is the one that verifies if there is reduced cost vehicle density on each roadway.
in its objective function to add it as a solution. Once a feasible Fig. 12 presents the graphical solution to the transportation
solution has been added, the algorithm repeats the iterations problem considering maximum vehicle flow during a time inter-
from the primary and secondary problems, and the observer val. The magnitude of the vehicle density generated at each link
makes decisions based on the negative cost variables. is visualized in black. The resulting vehicle density at each node is
A curious fact to note with Fig. 10(d) is that the cost of the easily identified in red, the sum of the flows entering each node
objective function in each scenario tested maintains the same minus the flows leaving. It is called the conservation of vehicle
trend, i.e., there is no deviation of data from year 0 to year flow. This condition is critical to maintaining the nodal equilib-
4. The movement in Fig. 10(d) responds to the immutability of rium criterion, which satisfies the equations of the mathematical
the scenario topology. That is, it increases vehicular traffic as problem statement. The negative red sign indicates the flows to
a function of time, but the topology of the case study remains be served by each recharging station; remember that the negative
the same. Finally, Fig. 10(d) shows that the objective function sign indicates that it is a supply node (recharging station). A
is inversely proportional to the hourly characteristic curve that fundamental condition for the convergence of the model is that
5258
M. Campaña and E. Inga Energy Reports 9 (2023) 5246–5268

Fig. 10. Case A: Performance of the algorithm considering scalability of 4 years with a growth rate of 3%.

the supply centers with negative signs cover the demand with a
positive signal at each intersection.
In the planning algorithms, a fundamental criterion for equip-
ment sizing is considering critical cases of high demand. This
consideration will allow us to approach the optimal sizing of
the resources required in the planning processes, guaranteeing
to cover the market. Consequently, Table 7 is presented the
result of the analysis case in Fig. 11 where the critical issue is
considered. The fundamental problem occurs at the maximum
density of vehicles driving in the geolocated area. These extreme
conditions occur at minimum travel speeds with minimum car
separation distances. Column one of Table 7 presents the index
of charging stations; column two illustrates the resulting con-
centration of vehicles to be served by each CS. The sign of the
second column is negative because the model considers it a node
contributing to the transportation system to form the equilibrium
equations, where the supply nodes must cover the sum of the
partial flows observed at each intersection. Consequently, we
denote those consumption nodes, and with the negative sign, we
identify the nodes that satisfy the equilibrium equations of the
multiple-product problem.
In addition, Table 7 in the third column shows the number of
vehicles each CS will have to serve under critical conditions of
high vehicle density without considering contingencies. Finally,
the last row of Table 7 identifies the maximum number of ve-
hicles in an instant of time, which in this case is 348 vehicles,
as the maximum admissible density in the area considered for
the analysis. Consequently, with the number of cars to be served
by each CS and the maximum traffic density in trajectory, we
Fig. 11. Case B: Weights assigned on each track section with microscopic
analysis in an area of 0.9887 km2 .
can size and foresee the necessary resources for constructing the
recharging infrastructure in the geolocalized areas.
5259
M. Campaña and E. Inga Energy Reports 9 (2023) 5246–5268

Table 6
Case A: Maximum partial demand in each CS at hour 19:00 with 32.97 km of road network in contingency N - 0
corresponding to year 0.
ID Longitude latitude Concentration Vehicles
Vehicular (#)
1 −120.625225787882 35.1271587155963 −0.77 25
2 −120.622727167999 35.1269678899083 −1.31 43
3 −120.617829873029 35.1267623853211 −0.64 21
4 −120.609994201077 35.1261458715596 −1.48 49
5 −120.615391220024 35.1256174311927 −1.29 42
6 −120.620588349380 35.1256467889908 −1.43 47
7 −120.624546163274 35.1248688073394 −1.09 36
8 −120.624326284724 35.1243403669725 −0.85 28
9 −120.619488956631 35.1240174311927 −1.43 47
10 −120.610473936095 35.1246045871560 −1.44 48
11 −120.611033626948 35.1230192660550 −0.67 22
12 −120.620568360421 35.1230779816514 −0.64 21
13 −120.622167477145 35.1233128440367 −0.85 29
14 −120.625865434572 35.1238266055046 −0.62 20
15 −120.625805467695 35.1222559633028 −0.61 20
16 −120.623466759484 35.1219183486239 −1.18 39
17 −120.620868194807 35.1221678899083 −0.92 30
18 −120.617310160094 35.1218889908257 −0.86 28
19 −120.615671065451 35.1218743119266 −0.99 33
20 −120.613552235790 35.1218449541284 −1.44 47
21 −120.610833737358 35.1199220183486 −1.21 40
22 −120.614731584375 35.1193642201835 −1.34 44
23 −120.616150800468 35.1204357798165 −0.96 32
24 −120.619588901427 35.1209495412844 −0.85 28
25 −120.625925401449 35.1213018348624 −0.97 32
26 −120.623666649075 35.1200247706422 −0.79 26
27 −120.619688846222 35.1195550458716 −0.81 27
28 −120.624606130151 35.1191293577982 −1.37 45
29 −120.617350138012 35.1186743119266 −1.19 39
30 −120.617250193216 35.1176027522936 −0.80 26
31 −120.618829320982 35.1176321100917 −1.09 36
32 −120.622327388818 35.1168981651376 −0.94 31
33 −120.613552235790 35.1176027522936 −0.61 20
34 −120.616050855673 35.1232541284404 −0.87 29
35 −120.608654940820 35.1195550458716 −0.62 20
Total −34.92 1151

Table 7
Case B: Maximum partial demand in each CS with 9,984 km of road network
in contingency N - 0.
ID Vehicle Concentration # Vehicles
1 −3.00 30
2 −5.33 53
3 −4.84 48
4 −2.93 29
5 −2.18 22
6 −3.72 37
7 −3.55 35
8 −4.72 47
9 −4.62 46
Total −34.89 348

The table below shows the optimal sizing of the recharging

stations considering the events or contingencies (N - 1 and N
- 2) generated in a geolocated area from the normal operating
conditions (N - 0). These contingencies respond to the elimination
of one or two recharge stations. In column 2 of Table 8, we show
the CS that go out of service, and from columns 3 to 11, we
present the flux density observed at each CS given a contingency.
The out-of-service charging stations are randomly selected.
When considering the contingency analysis, the traffic density
to meet the demand has to be reorganized and distributed with
the enabled charging stations. Looking at column three of Table 8
under normal operating conditions, the vehicle density to be
served by charging station ID 1 is 29.95 ≈ 30 vehicles in a
Fig. 12. Case B: Graphical solution to the multiple product problem.
time interval when the vehicle density is maximum. When the
model experiences the departure from operation of the charging
5260
M. Campaña and E. Inga Energy Reports 9 (2023) 5246–5268

Table 8
Case B: Optimal sizing of charging stations for EV based on multi-product flow problem.
Contingency Deactivated ID ID ID ID ID ID ID ID ID
(ID) 1 2 3 4 5 6 7 8 9
N - 0 – −3.00 −5.33 −4.84 −2.93 −2.18 −3.72 −3.55 -4.72 −4.62
N - 1 1 – −5.60 −4.38 −4.97 -4.08 −3.90 −4.20 −4.16 −5.08
2 −4.01 – -5.98 −4.99 −3.65 -5.96 −4.48 −5.03 −4.93
3 −4.27 -6.38 – −3.62 −4.03 −4.56 −4.35 −5.11 −5.12
4 -4.42 −4.32 −4.06 – −3.55 −5.34 −4.72 −5.24 −5.68
5 −4.41 −5.12 −4.81 −4.65 – −5.24 −3.40 −5.48 −4.75
6 −3.74 −4.37 −4.99 -6.05 −3.15 – −4.34 −5.15 -5.79
7 −4.25 −4.05 −5.31 −4.91 −3.53 −4.89 – -5.53 −4.91
8 −3.15 −4.98 −4.79 −4.73 −3.88 −5.12 −3.94 – −4.85
9 −4.02 −4.72 −5.41 −4.20 −3.35 −4.00 -4.86 −4.86 –
Average (µ) 4,03 4,94 4,97 4,77 3,65 4,88 4,29 5,07 5,14
Std. deviation (σ ) 0,40 0,72 0,57 0,65 0,31 0,65 0,43 0,40 0,36
µ+2∗σ 4,83 6,37 6,10 6,07 4,26 6,18 5,14 5,87 5,86
N - 2 1, 2 – – −4.67 −5.18 −4.89 −5.40 −5.54 −4.95 −5.42
2, 3 −4.05 – – −4.25 −4.59 −5.42 −4.43 −6.08 −5.02
8, 9 −4.13 −5.40 −5.49 −5.56 −5.12 −5.76 −5.46 – –
4, 5 −4.46 −4.89 −5.09 – – -6.08 -5.64 −5.71 −6.11
7, 9 −4.83 −4.01 -7.10 −5.73 −4.31 −5.70 – −5.93 –
5, 7 −4.38 −4.20 −4.71 −4.25 – −5.62 – -6.31 −5.16
3, 7 -5.37 −4.68 – −5.09 −4.97 −5.13 – −5.90 -6.14
3, 5 −4.15 −5.03 – −5.51 – −4.95 −4.36 −5.80 −4.78
2, 4 −5.20 – −6.08 – −5.22 −4.91 −5.06 −5.41 −6.00
5, 6 −4.19 −4.94 −5.38 -6.36 – – −4.89 −6.11 −5.79
7, 8 −4.30 −6.30 −4.50 −4.78 -5.27 −5.06 – – −5.46
1, 3 – −5.22 – −4.25 −3.51 −5.89 −4.41 −5.69 −5.09
3, 4 −4.77 -6.39 – – −3.84 −5.58 −5.07 −5.50 −6.08
Average (µ) 4,53 5,11 5,38 5,10 4,64 5,46 4,98 5,76 5,55
Std. deviation (σ ) 0,43 0,74 0,81 0,68 0,59 0,36 0,47 0,36 0,47
µ+2∗σ 5,39 6,58 7,00 6,46 5,82 6,19 5,93 6,48 6,50
Increment N - 1 (%) 61.0 19.5 26.0 107.2 95.4 66.1 44.8 24.4 26.8
Increment N - 2 (%) 79.7 23.5 44.6 120.5 167.0 66.4 67.0 37.3 40.7

station with ID 4, the thickness of vehicles to be served given the By multiplying the cumulative length in kilometers of the
N - 1 contingency is 48.22 ≈48. If the charging infrastructure road network by the vehicle density, we can obtain the num-
experiences the operation output of two CS with IDs 3 and 7 in ber of vehicles each CS should serve in each event studied. In
contingency N - 2, the traffic density to be served is 53.81 ≈ addition, Table 9 allows us to verify by how much the sizing
54 vehicles in a given time interval. Experience with the ID 1 of the CS should increase given a circumstance. Consequently,
charging station establishes that at N - 0, the vehicle density is the information presented so far is possible to plan the charging
three, with N - 1 and N - 2 at 4.83 and 5.39, respectively. This infrastructure for electric mobility in cities.
increase represents approximately 61% and 80% of the different Up to this point, it has been possible to foresee the maximum
traffic densities that will have to be served when the charging number of vehicles that could enter each CS. Now, we proceed
infrastructure disconnects the charging stations in the N - 1 and with the optimal sizing of charging terminals that a CS should
N - 2 contingencies, respectively. Likewise, we will perform the have to meet the demand generated by EV operators. To this end,
analysis for the recharging station with ID 5, which presents an the following considerations will be made:
increase in vehicle density of 95.4% in contingency N - 1 and an
increase of 167.6% in contingency N - 2 have to be served if one • The average charging time for 50 kW terminals is 30 min to
or two recharging stations are disconnected. reach 20–80%.
It is understood as the algorithm reorganizing the vehicle flow • The average charging time for 350 kW terminals will be 10
based on a vector of cost and capacity for each road section, min to reach 20–80%.
satisfying the mathematical variables of the linear equation. How- • The optimal number of loading terminals is determined at
ever, the equilibrium equations are satisfied for each iteration. 07:00 a.m. with the most vehicular traffic.
If we pay attention to detail, the charging station with ID 5 in
contingency N - 2 experiences increases in the density of vehicles 4.2.1. Optimum number of load terminals in case study B
that exceed 100% in the function of the regular operation in With Fig. 13, the optimal number of charging terminals to
contingency N - 0. In addition, the increase in vehicular flow will be installed in the CS with ID 1 (see Table 9) in event N - 0
depend on which charging station goes out of service. is presented. The metric in Fig. 13 illustrates that serving 30
Therefore, if the designer of the charging infrastructure knows EVs for 60 min requires 17 and 7 charging terminals with an
what percentage would increase or decrease the vehicle density output power of 50 kW and 350 kW, respectively. With this
to be served by each CS, they can make intelligent decisions to exciting data, it is possible to determine the installed capacity
foresee the electrical, civil, and logistic equipment necessary for required in the CS with ID 1. Consequently, if it is necessary to
the construction of charging infrastructures in electric mobility. establish a fast-charging infrastructure with 50 kW output power
Consequently, to determine the sizing of the CS, a rule of thumb at each charging terminal, the installed capacity in the CS should
is considered to find the magnitude of values. They are within be 850 kW. If the requirement in the infrastructure is ultra-fast
a band around the mean in a normal distribution with twice charging terminals, the energy required in the CS should be 2.45
the standard deviation width. Consequently, Table 8 with green MW to cover the estimated demand during high vehicular traffic.
indicates the maximum vehicular flow seen in each CS. The set of Finally, an additional detail illustrated in Fig. 13 is the average
CS will result in the EV charging station infrastructure. time an electric vehicle remains in the system. Consequently, it
5261
M. Campaña and E. Inga Energy Reports 9 (2023) 5246–5268

Table 9
Maximum partial demand at each charging station with 9,984 km of road network given a contingency.
Events ID 1 ID 2 ID 3 ID 4 ID 5 ID 6 ID 7 ID 8 ID 9
N - 0 30 53 48 29 22 37 35 47 46
N - 1 48 64 61 61 43 62 51 59 59
N - 2 54 66 70 65 58 62 59 64 65

Fig. 13. Graphical method for determining the optimal number of load terminals Fig. 14. Graphical method to determine the optimal number of load terminals
with event N - 0. considering N - 1 event.

is notorious for identifying that the higher the supply power at maintenance), the loading infrastructure is prepared to assume
the charging terminals for battery recharging, the shorter the EV the demand that will no longer be covered by the CS that goes
operator remains in the system. out of service. Consequently, with Fig. 14, the new number of
Fig. 14 shows the optimal number of load terminals consid- charging terminals in N - 1 increases to 26 and 10 charging
ering event N - 1. A reallocation of the number of terminals terminals with 50 kW and 350 kW output power at the service
required to meet the demand when a CS goes out of the system’s terminal, respectively. Consequently, the installed capacity in the
operation is evident. This reallocation is necessary because if a CS CS for the different charging technologies is 1.3 MW and 3.5 MW,
goes out of the process for any reason (preventive or corrective corresponding to an increase of 52.94% and 42.85%, respectively.
5262
M. Campaña and E. Inga Energy Reports 9 (2023) 5246–5268

Table 10
Charging infrastructure sizing applied to case study B with event
N - 1 with 50 kW charging terminals.
Loading # Loading Total power
station terminals (MW)
ID 1 26 1.30
ID 2 34 1.70
ID 3 32 1.60
ID 4 32 1.60
ID 5 23 1.15
ID 6 33 1.65
ID 7 27 1.35
ID 8 31 1.55
ID 9 31 1.55
Total 13.45

Table 11
Sizing of the loading infrastructure applied to case study B with event N - 1
with 350 kW loading terminals.
Loading # Loading Total power
Station Terminals (MW)
ID 1 10 3.50
ID 2 12 4.20
ID 3 12 4.20
ID 4 12 4.20
ID 5 9 3.15
ID 6 12 4.20
ID 7 10 3.50
ID 8 11 3.85
ID 9 11 3.85
Total 34.65

required in each CS considering the output of a CS, which will

give rise to the sizing of the EV charging infrastructure.

4.2.2. Optimal sizing of electrical equipment required in EV charging

infrastructure
The number of charging terminals required in each CS is
illustrated in Tables 10 and 11 considering N - 1 events. Table 10
shows the optimum number of charging terminals for electric
vehicles weighing a charging time of 30 min from 20–80% charge.
This type of terminal provides a service power of 50 kW. On
the other hand, with Table 11, the charging time is 10 min with
a terminal charging power of 350 kW. By purchasing the two
tables, it is evidenced that the average installed power is 1.49 and
3.85 MW for fast and ultra-fast charging terminals, respectively.
The average number of charging terminals with fast charging
Fig. 15. Graphical method to determine the optimal number of load terminals
technology is 30 units, while for ultra-fast charging terminals, it
considering N - 2 events. is 11. On the other hand, the average time for the electric vehicle
operator to remain in the system is 44 min with fast charging
terminals and 18 min with ultra-fast charging terminals.
Therefore, the proposed model has demonstrated the ability to
Fig. 15 illustrates the number of charging terminals required
adapt to different case studies that allow us to analyze contingen-
to be installed in the CS with ID 1, considering the exit from
cies or events in the charging infrastructure, which are very useful
the operation of two CS belonging to the charging infrastructure.
when estimating the power installed in each CS. It will enable us
Thus, it is evidenced that 29 and 11 charging terminals should
to consider the electrical equipment, civil works, and logistics to
be installed at the service station with ID 1 when two CS go
construct adequately planned charging infrastructures.
out of operation. Consequently, the sizing for the CS with ID 1 The sizing and selection of the transformer for each CS are
is 1.45 MW and 3.85 MW for each type of technology, 50 kW and shown in Table 12. Based on the electrical power required to
350 kW, respectively. supply the demand and considering a safety factor of 25%, the
The terminal charging technology directly relates to the supply transformer capacity to be installed is selected. Additional detail
power for battery recharging. Therefore, with Figs. 13–15, it is is that the final capacity of the transformer is chosen based on
demonstrated that it is possible to determine the optimal number commercial transformers.
of charging terminals depending on the terminal charging tech- The CymDist software is used to determine the power flow
nology to be considered in the CS by applying M/M/s queuing of the loading infrastructure. Fig. 16 illustrates the route to be
theory. In this case, 50 kW and 350 kW were used, which offer 30 followed by the primary feeder to the CS. The feeder route is
and 10-minute charging times, respectively. Next, the graphical constructed using a minimum-spanning tree. Consequently, it is
method will be applied to determine the number of terminals ensured that the length of the conductor is the minimum required
5263
M. Campaña and E. Inga Energy Reports 9 (2023) 5246–5268

Table 12
Sizing and selection of three-phase transformer for each charging station.
Loading Total power Security factor Total power Transformer
station (MW) 25% (MW) calculated (MVA) 3 phases (MVA)
ID 1 1.30 1.63 1.65 2
ID 2 1.70 2.13 2.15 2.5
ID 3 1.60 2.00 2.02 2
ID 4 1.60 2.00 2.02 2
ID 5 1.15 1.44 1.46 2
ID 6 1.65 2.06 2.10 2.5
ID 7 1.35 1.69 1.71 2
ID 8 1.55 1.94 1.96 2
ID 9 1.55 1.94 1.96 2
Total 13.45 16.83 17.0 19

Fig. 16. Case B: Optimal routing of the primary feeder to the fast-charging stations.

to provide electrical power to each CS for EV. An additional detail caused by magnetization currents and by the characteristics of
shown with Fig. 16 is the load flow with power factor 0.99, the conductor.
transformer primary voltage 22 kV, and secondary voltage 480 The average currents expected in the primary and secondary
V. Furthermore, a safety factor of 25% in addition to the installed of the transformer installed in the different CS are 39 A and 1784
power (see Table 10) is considered for transformer sizing. Amp, respectively, as shown in Table 15. The voltage drop from
With Table 13, you can see the different lengths of each the source to the various nodes of the electrical circuit of the EV
conductor section. In addition, the maximum current flowing in charging infrastructure is presented below.
the various areas of the primary feeder can be observed. The Verifying that the voltage drop does not violate the admissible
maximum current reflected in the source is 361.2 amperes. An- limits is essential. Fig. 17 shows the voltage drop from the source
other detail of interest shown in Table 13 is the losses in kW to each node energizing the transformer installed at each load
that occur due to the characteristics of the conductors and trans- station. Another detail observed is the conductors’ length, i.e., at
formers. Consequently, with Table 13, it is possible to foresee the approximately 1200 meters from the source, the CS with ID 1 is
conductors required to construct the EV charging infrastructure. located without violating the voltage drop limits.
The transformer’s rated capacity, power throughput, total Fig. 18 shows the total current THD as a function of frequency.
losses, and load percentage are illustrated in Table 14. The av- Since this is a three-phase balanced system, harmonics 3, 5, 7, 9,
erage chargeability of the transformer is approximately 70 per- and 11 average 9.404%. The average individual distortion factor
cent, representing a 30 percent reserve, given that a commercial (IHD) is 3.502%, so it is inferred that the limits indicated using the
transformer is considered. Another detail of interest shown in IEEE 519-1992 standard, where it is specified that for currents of
Table 14 is the transformer’s active and reactive power losses 50 < 100, the TDD 12% are not violated.
5264
M. Campaña and E. Inga Energy Reports 9 (2023) 5246–5268

Table 13
Primary feeder power flow for fast charging terminals, 50 kW.
Node Node Longitude Step power Average PF Maximum current Losses Charge
source destination (m) (kW) (%) (A) (kW) %
167 168 25.7 1539.7 99.90 40.5 0.11 47.59
176 177 30.9 1581.1 99.89 41.6 0.14 48.98
197 169 91.2 4011.5 99.93 105.5 1.01 49.64
162 167 115.7 4615.8 99.93 121.2 1.70 57.04
199 200 182.1 1634.5 99.93 43.0 0.86 50.59
169 172 190.5 1141.3 99.95 30.0 0.44 35.31
167 197 196.1 7189.6 99.93 188.8 8.82 74.04
197 199 217.9 3169.2 99.92 83.3 1.52 39.22
199 198 252.6 1533.2 99.90 40.4 1.05 47.47
162 164 312.4 1341.3 99.96 35.2 0.50 22.40
162 175 378.4 1585.9 99.90 41.7 1.68 49.03
169 176 433.2 2869.2 99.93 75.5 2.47 35.51
162 174 458.0 1686.8 99.93 44.3 2.29 52.14
176 163 468.9 1285.7 99.94 33.8 1.37 39.81

Table 14
Power flow in transformers for fast charging terminals, 50 kW.
Charge Nom Cap Step power Primary Current Total losses Total losses Charge
Station (kVA) (kW) (A) (kW) (kvar) (%)
ID 1 2000 1284,3 33,9 6,0 49,6 63,9
ID 2 2500 1684,6 44,3 7,8 68,0 67,1
ID 3 2000 1584,3 41,7 8,5 75,2 78,8
ID 4 2000 1580,9 41,6 8,5 75,1 78,6
ID 5 2000 1140,9 30,0 4,9 39,0 56,8
ID 6 2500 1633,7 43,0 7,4 64,0 65,1
ID 7 2000 1340,8 35,2 6,4 53,8 66,7
ID 8 2000 1539,6 40,5 8,1 70,8 76,6
ID 9 2000 1532,2 40,4 8,1 70,5 76,2

Table 15
Primary and secondary side transformer current.
Charging Primary Secondary
station (Amp) (Amp)
ID 1 33.8 1550.0
ID 2 44.3 2030.2
ID 3 41.7 1909.3
ID 4 41.6 1907.1
ID 5 30.0 1374.4
ID 6 43.0 1969.7
ID 7 35.2 1614.0
ID 8 40.5 1852.8
ID 9 40.3 1848.1

Fig. 18. Case B: Analysis of current vs. total frequency distortion in each CS.

The voltage distortion at the low voltage nodes as a function

of time is presented in Fig. 19. If we compare it with Fig. 20, it
can be identified that there is higher THD on the low voltage side.
Another detail shown in Fig. 19 is the minimum voltage variation
in each node serving the different charging stations. It is because
the system analyzed is balanced since only the electrical network
of the charging infrastructure is analyzed.
As viewed from the source, the THD is illustrated in Fig. 20.
There is more significant distortion on the low voltage side of
each of the transformers. This statement is contrasted with Ta-
ble 16. There is a total distortion at the nodes connecting the
primary and secondary sides of the transformer; also, the source
Fig. 17. Case B: Voltage drop from the source to the charging stations.
is presented. The red color shows the nodes that violate the limits
of the IEEE 519-1992 standard (see Table 16). Additional detail is
that if Table 16 is evaluated with the IEEE 519-2014 standard,
5265
M. Campaña and E. Inga Energy Reports 9 (2023) 5246–5268

Fig. 19. Case B: Distortion analysis voltage vs. time in each CS node.
Fig. 20. Case B: Voltage vs. time distortion analysis at the source.

Table 16
Total harmonic voltage distortion at transformer and source nodes. 6. Conclusions
Charge Node HV LV
station (#) (%) (%)
This article has made planning EVCSI in geolocated scenarios
ID 1 183 2.47 5.02 possible based on free information from OpenStreetMap. This
ID 2 191 2.18 4.88
ID 3 195 2.15 5.34 information contains the knowledge of the topology of the road
ID 4 185 2.40 5.55 network on which the case studies are analyzed.
ID 5 187 2.26 4.51 With the proposed methodology, the maximum number of
ID 6 203 2.31 4.92
vehicles traveling on a given road network is predicted by ob-
ID 7 189 2.11 4.80
ID 8 193 2.01 5.11 serving its topology to allocate resources and size the EV charging
ID 9 201 2.33 5.39 infrastructure.
Source 167 2 The heuristic has proven scalable and validated through case
studies and iterative processes that address contingency con-
cepts. Therefore, the main contribution of this work is its high
the established limits would not be violated. In the IEEE 519- potential to become a computational model and infrastructure
1992 standard, a maximum THD of 5% is considered for voltages manager for planning and resource allocation in electric mobility.
<= 1 kV. In the IEEE 519 - 2014 standard, 8% is considered for The heuristic provides finite and time-scalable solutions. The
voltages <= 1 kV. hourly analysis of vehicle flow has allowed identifying the max-
Consequently, the methodology proposed in this article has imum and minimum demands experienced by each CS, which
allowed the construction of the electrical network to energize contributes to coordinating and managing EV battery charging
the different CS that will serve multiple EV users. The proposed processes.
topology and sizing of the transformers have been simulated with The CS sizing criteria are based on empirical statistical tech-
CymDist software, validating the functionality of the proposed niques, considering the mean with two standard deviations. Con-
model. The importance of the simulation lies in verifying the sequently, the M/M/s queuing system determines the optimal
operational performance of the charging infrastructure and thus number of charging terminals.
being able to foresee corrective actions in favor of the charging Once the optimal number of charging terminals is found, the
infrastructure to be built.
electrical equipment can be sized to meet the demand for battery
recharging in electric mobility. In addition, the power flow study
5. Discussion
with load models is included to make a reading of the operation
and functioning of the electrical network.
Implementing effective and efficient charging infrastructure
is a crucial challenge to encourage the widespread use of elec- The study of harmonics, when modeling the load, becomes
tric vehicles and achieve a cleaner and more sustainable energy fundamental since power electronics are used in the load ter-
transition. Although the model proposed in this article provides a minals. This study will allow foreseeing the necessary equip-
valuable contribution to the optimization of charging infrastruc- ment with optimality criteria. Finally, this combinatorial prob-
ture resources, it is essential to consider the model’s limitations lem needs more global solutions since its complexity increases
and seek a more comprehensive and holistic approach to address exponentially as the topology grows.
this challenge.
In addition to the factors mentioned above, it is also necessary
CRediT authorship contribution statement
to consider regional differences in demand for electric vehicles
and the availability of renewable energy resources, as this can sig-
nificantly affect the location and size of charging stations. A more Miguel Campaña: Conceptualization, Methodology, Software,
comprehensive approach that considers these considerations can Formal analysis, Investigation, Writing – original draft, Writing –
help ensure that charging infrastructure is implemented fairly review & editing, Visualization. Esteban Inga: Conceptualization,
and equitably across regions, encouraging greater acceptance and Methodology, Validation, Writing – review & editing, Visualiza-
adoption of electric vehicles nationwide. tion, Supervision.
5266
M. Campaña and E. Inga Energy Reports 9 (2023) 5246–5268

Declaration of competing interest Du, L., Song, G., Wang, Y., Huang, J., Yu, Z., Ruan, M., 2018. Model for expressway
network : A network flow approach. IEEE Intell. Transp. Syst. (January),
The authors declare that they have no known competing finan- 107–120. http://dx.doi.org/10.1109/MITS.2017.2776130.
Esteban, I., Roberto, H., Sandra, C., 2019. Capacitated multicommodity flow
cial interests or personal relationships that could have appeared
problem for heterogeneous smart electricity metering communications us-
to influence the work reported in this paper. ing column generation. Energies 13 (1), 1–21. http://dx.doi.org/10.3390/
en13010097.
Data availability Gan, X., Zhang, H., Hang, G., Qin, Z., Jin, H., 2020. Fast charging station
deployment considering elastic demand. IEEE Trans. Transp. Electrif. PP (XX),
Data will be made available on request 1. http://dx.doi.org/10.1109/tte.2020.2964141.
Ghasemi, P., Khalili-Damghani, K., Hafezolkotob, A., Raissi, S., 2019. Uncer-
tain multi-objective multi-commodity multi-period multi-vehicle location-
Acknowledgments allocation model for earthquake evacuation planning. Appl. Math. Comput.
350, 105–132. http://dx.doi.org/10.1016/j.amc.2018.12.061.
This work was financially funded by the Smart Grid Research Inga, E., Céspedes, S., Hincapié, R., Cárdenas, A., 2017. Scalable route map
Group (GIREI) of Universidad Politécnica Salesiana, Ecuador for advanced metering infrastructure based on optimal routing of wireless
(Project: Electric vehicle charging in buildings and its impact on heterogeneous networks. IEEE Wirel. Commun. 24 (April), 1–8. http://dx.doi.
org/10.1109/MWC.2017.1600255, URL https://ieeexplore.ieee.org/document/
the sizing and planning of electrical distribution networks) and
7909154/.
the Power Grids and Smart Cities (RECI), Salesian Institutions of Inga, E., Hincapié, R., Céspedes, S., 2019. Capacitated multicommodity flow
Higher Education (IUS). problem for heterogeneous smart electricity metering communications us-
ing column generation. Energies 13 (1), 1–21. http://dx.doi.org/10.3390/
References en13010097, URL https://www.mdpi.com/1996-1073/13/1/97.
Jiang, J., Yue, Y., He, S., 2018. Multimodal-transport collaborative evacuation
Amry, Y., Elbouchikhi, E., Le Gall, F., Ghogho, M., El Hani, S., 2022. Electric vehicle strategies for urban serious emergency incidents based on multi-sources
traction drives and charging station power electronics: Current status and spatiotemporal data. In: Leibniz International Proceedings in Informatics,
challenges. Energies 15 (16), 1–30. http://dx.doi.org/10.3390/en15166037. LIPIcs. Vol. 114. No. 35. pp. 1–8. http://dx.doi.org/10.4230/LIPIcs.GIScience.
Antarasee, P., Premrudeepreechacharn, S., Siritaratiwat, A., Khunkitti, S., 2023. 2018.35.
Optimal design of electric vehicle fast-charging station’s structure using Jiménez-Estévez, G.A., Vargas, L.S., Marianov, V., 2010. Determination of feeder
metaheuristic algorithms. Sustainability (Switzerland) 15 (1), 22. http://dx. areas for the design of large distribution networks. IEEE Trans. Power Deliv.
doi.org/10.3390/su15010771. 25 (3), 1912–1922. http://dx.doi.org/10.1109/TPWRD.2010.2042468.
Asna, M., Shareef, H., Prasanthi, A., 2023. Planning of fast charging stations with Kłos, M.J., Sierpiński, G., 2023. Strategy for the siting of electric vehicle charging
consideration of EV user, distribution network and station operation. Energy stations for parcel delivery service providers. Energies 16 (6), 2553. http:
Rep. 9, 455–462. http://dx.doi.org/10.1016/j.egyr.2023.01.063. //dx.doi.org/10.3390/en16062553.
Azadi Moghaddam Arani, A., Jolai, F., Nasiri, M.M., 2019. A multi-commodity net- Konara, K.M.S.Y., Kolhe, M.L., Ulltveit-Moe, N., Balapuwaduge, I.A., 2023. Optimal
work flow model for railway capacity optimization in case of line blockage. utilization of charging resources of fast charging station with oppor-
Int. J. Rail Transp. 7 (4), 297–320. http://dx.doi.org/10.1080/23248378.2019. tunistic electric vehicle users. Batteries 9 (2), http://dx.doi.org/10.3390/
1571450. batteries9020140.
B, B.G., Ozlen, M., Neumann, F., 2018. Parallel Problem Solving from Nature - Li, J., Ma, X.Y., Liu, C.C., Schneider, K.P., 2014. Distribution system restoration
PPSN XV - 15th International Conference, Coimbra, Portugal, September 8- with microgrids using spanning tree search. IEEE Trans. Power Syst. 29 (6),
12, 2018, Proceedings, Part I. Vol. 1. Springer International Publishing, pp. 3021–3029. http://dx.doi.org/10.1109/TPWRS.2014.2312424.
Liu, G., Kang, L., Luan, Z., Qiu, J., Zheng, F., 2019a. Charging station and power
69–81. http://dx.doi.org/10.1007/978-3-319-99253-2.
network planning for integrated electric vehicles (EVs). Energies 12 (13),
Balakrishna, P., Rajagopal, K., Swarup, K.S., 2014. AMI / GIS Based Distribution
1–22. http://dx.doi.org/10.3390/en12132595.
System Load Flow for Extended Situational Awareness. IEEE, pp. 1–6.
Liu, M., Richa, A.W., Rost, M., Schmid, S., 2019b. A constant approximation for
Bevrani, B., Burdett, R.L., Bhaskar, A., Yarlagadda, P.K., 2019. A Multi Commodity
maximum throughput multicommodity routing and its application to delay-
Flow Model Incorporating Flow Reduction Functions. Springer US, http:
tolerant network scheduling. In: Proceedings - IEEE INFOCOM. 2019-April.
//dx.doi.org/10.1007/s10696-019-09349-4.
IEEE, pp. 46–54. http://dx.doi.org/10.1109/INFOCOM.2019.8737402.
Bi, X., Tang, W.K., 2019. Logistical planning for electric vehicles under time-
Lu, G., Zhou, X., Mahmoudi, M., Shi, T., Peng, Q., 2019. Optimizing resource
dependent stochastic traffic. IEEE Trans. Intell. Transp. Syst. 20 (10),
recharging location-routing plans: A resource-space-time network modeling
3771–3781. http://dx.doi.org/10.1109/TITS.2018.2883791.
framework for railway locomotive refueling applications. Comput. Ind. Eng.
Campaña, M., Inga, E., 2019a. Despliegue óptimo georreferenciado de estaciones
127, 1241–1258. http://dx.doi.org/10.1016/j.cie.2018.03.015.
de carga vehicular pública considerando capacidad de flujo y distancias
Luo, L., Wu, Z., Gu, W., Huang, H., Gao, S., Han, J., 2020. Coordinated allocation
máximas habilitantes. I+D Tecnológico 15 (2), 68–78. http://dx.doi.org/10.
of distributed generation resources and electric vehicle charging stations
33412/idt.v15.2.2248.
in distribution systems with vehicle-to-grid interaction. Energy 192, http:
Campaña, M., Inga, E., 2019b. Heuristic method for optimal deployment of
//dx.doi.org/10.1016/j.energy.2019.116631.
electric vehicle charge stations using linear programming. Commun. Comput.
Montoya, D.P., Ramirez, J.M., 2012. A minimal spanning tree algorithm for
Inf. Sci. 1096 CCIS, 247–258. http://dx.doi.org/10.1007/978-3-030-36211-
distribution networks configuration. In: IEEE Power and Energy Society
9_20.
General Meeting. pp. 1–7. http://dx.doi.org/10.1109/PESGM.2012.6344718.
Campaña, M., Inga, E., 2019c. Optimal allocation of public charging stations
Moradipari, A., Alizadeh, M., 2018. Electric vehicle charging station network
based on traffic density in smart cities. In: 2019 IEEE Colombian Conference
equilibrium models and pricing schemes. arXiv. arXiv:1811.08582.
on Applications in Computational Intelligence, ColCACI 2019 - Proceedings.
Nagarajan, A., Ayyanar, R., 2014. Application of Minimum Spanning Tree
http://dx.doi.org/10.1109/ColCACI.2019.8781986.
Algorithm for Network Reduction of Distribution Systems. IEEE, pp. 1–5.
Campaña, M., Inga, E., Cárdenas, J., 2021. Optimal sizing of electric vehicle
Napoli, G., Polimeni, A., Micari, S., Andaloro, L., Antonucci, V., 2020. Optimal
charging stations considering urban traffic flow for smart cities. Energies
allocation of electric vehicle charging stations in a highway network: Part
14 (16), 1–16. http://dx.doi.org/10.3390/en14164933.
1. Methodology and test application. J. Energy Storage 27 (October 2019),
Campaña, M., Inga, E., Hincapié, R., 2017. Optimal placement of universal data
101102. http://dx.doi.org/10.1016/j.est.2019.101102.
aggregation points for smart electric metering based on hybrid wireless. In:
Neagu, B.C., Georgescu, G., 2014. Wind farm cable route optimization us-
CEUR Workshop Proceedings. Vol. 1950. pp. 6–9.
ing a simple approach. In: Electrical and Power Engineering (EPE), 2014
Chen, J., Chen, H., 2023. Stations considering user selection preferences. Energies
International Conference and Exposition on, Iasi. Vol. 1. No. Epe. pp.
16 (4), http://dx.doi.org/10.3390/en16041794.
1004–1009.
Chung, B.D., Park, S., Kwon, C., 2018. Equitable distribution of recharging stations Rabbani, M., Zenouzzadeh, S.M., Farrokhi-asl, H., 2019. Solving a bi-objective
for electric vehicles. Socio-Econ. Plan. Sci. 63, 1–11. http://dx.doi.org/10.1016/ multi-commodity two-echelon capacitated location routing problem. JISE 12
j.seps.2017.06.002. (3), 249–268.
Demir, E., Huckle, K., Syntetos, A., Lahy, A., Wilson, M., 2019. Vehicle Routing Singh, P.P., Wen, F., Palu, I., Sachan, S., Deb, S., 2023. Electric vehicles charging
Problem: Past and Future. Springer International Publishing, pp. 97–117. infrastructure demand and deployment: Challenges and solutions. Energies
http://dx.doi.org/10.1007/978-3-030-14493-7. 16 (1), http://dx.doi.org/10.3390/en16010007.

5267
M. Campaña and E. Inga Energy Reports 9 (2023) 5246–5268

Song, D., Tharmarasa, R., Zhou, G., Florea, M.C., Duclos-Hindie, N., Kirubarajan, T., Vaziri, S., Etebari, F., Vahdani, B., 2019. Development and optimization of a
2019. Multi-vehicle tracking using microscopic traffic models. IEEE Trans. horizontal carrier collaboration vehicle routing model with multi-commodity
Intell. Transp. Syst. 20 (1), 149–161. http://dx.doi.org/10.1109/TITS.2018. request allocation. J. Clean. Prod. 224, 492–505. http://dx.doi.org/10.1016/j.
2804894. jclepro.2019.02.043.
Tang, X., Bi, S., Zhang, Y.-J.A., 2018. Distributed routing and charging scheduling Zhang, Y., Ioannou, P.A., 2018. Integrated control of highway traffic flow.
optimization for internet of electric vehicles. IEEE Internet Things J. PP (c), 1. J. Control Decis. 5 (1), 19–41. http://dx.doi.org/10.1080/23307706.2017.
http://dx.doi.org/10.1109/JIOT.2018.2876004, URL https://ieeexplore.ieee.org/ 1402715.
document/8491301/.
Zhang, X., Rey, D., Waller, S.T., Chen, N., 2019. Range-constrained traffic assign-
Tian, H., Tzelepis, D., Papadopoulos, P.N., 2021. Electric vehicle charger static
ment with multi-modal recharge for electric vehicles. Netw. Spat. Econ. 19
and dynamic modelling for power system studies. Energies 14 (7), http:
(2), 633–668. http://dx.doi.org/10.1007/s11067-019-09454-9.
//dx.doi.org/10.3390/en14071801.

5268