Demand Forecasting Associated with Electric Vehicle
Penetration on Distribution Systems
Andres F. Botero, and Mario A. Rios, Member, IEEE
Department of Electrical Engineering and Electronics
Universidad de los Andes
Bogota D.C., Colombia
mrios@uniandes.edu.co
Abstract—This paper proposes a methodology for forecasting
the electrical power demand associated to the penetration of
electric vehicles on a distribution system. The simulation model
is framed in a temporal and spatial base that considers a more
detailed analysis of some demographical aspects and technical
characteristics of the electric vehicles. This approach allows the
assessment of the impact of such penetration in the different
areas of a city, considering that each part of the network will
have several consumption behaviors. Additionally, it gives the
advantage to easily develop multiple case studies by adjusting
driver pattern probabilities, electric vehicle characteristics,
geographic and demographic parameters of the city and even
charging scenarios.
Index Terms-- Battery Charging Infrastructure, Demand
Forecasting, Electric Vehicle, Electric Vehicle Market
Penetration, Spatial and Temporal Analysis.
I. INTRODUCTION
Transport is without any doubt, one of the human activities
representing the greatest amount of emissions of greenhouse
gases worldwide, clearly due to its heavy reliance on fossil
fuels. Hence, the closest solution to this problem seems to be
the incorporation of vehicles powered by electric motors.
However, its widespread deployment brings a lot of
challenges for the electricity distribution system related not
only with operational terms but also with planning aspects.
Several researches have been performed on how to tackle
the resultant problems of incorporating these technologies into
the automobile market. Most of them have focused their
attention on technical short-term impacts of electric vehicles
(EVs) and its consequences on the equipment in the system.
However, an increasing penetration of EVs would press the
distribution network operators to develop and use reliable
forecasting models to estimate such additional energy
requirements. Thus, it is important to make additional efforts
in analyzing the impact on load profiles and develop models
that give clear signals to make decisions for system planning.
To achieve this goal, electrical demand projections will have
to take into account not only the overall growth rate and
energy consumption per customer, but also the aggregated
This work was supported in part by COLCIENCIAS, CODENSA,
EMGESA Grant 0197-2012.
demand due to the charging process of each one of the EVs in
the system. The big difficulty, in these forecasting models, is
found in the large number of uncertainties that arise when
evaluating the energy requirements and charging modes of the
EVs throughout the day.
Due to the lack of real-world data, and the worries
evaluating the impacts of EV load on the network, most of the
present literature establishes simple models of aggregate
arrival rates of EVs and their requests to charge, using Poisson
process or queuing theory, but without considering some
peripheral information that would affect the model. Some, as
[1], makes a very simple forecast of possible impacts on the
demand profile by just adding a certain number of EVs to each
residential agent but not evaluating possible driver behaviors
to compute energy requirements. Others like [2] and [3],
establish stochastic analysis in which only residential charging
is considered and the charging event is subject to the last
arrival of the day at home. Particularly, in [3] some parameters
are included to model the spatial and temporal diversity of the
EVs behavior in the system by defining some travel patterns
by agent based simulations.
This paper proposes a comprehensible methodology to
assess the impact on the demand profiles due to the charging
of plug-in electric vehicles (PEVs) in distribution networks.
The proposed methodology is outlined in a spatial and
temporal frame in which the use of real-world data is
imperative to obtain reliable results and make thoughtful
analysis of EVs load impacts on the network.
So, the general forecasting model is presented along with a
description of the variables involved and an explanation of the
simulation method used. Finally, a study case is presented
showing some important results of the model.
II. FORECASTING MODELS
A reliable forecasting model would have to take into
account a set of variables with uncertain behaviors in order to
estimate the additional daily load required for EVs charging.
To achieve this goal, this paper proposes a simulation model
framed in a temporal and spatial conception that considers a
more detailed analysis of some demographical aspects and
technical characteristics of the EVs. This approach allows the
assessment of the impact of such penetration in the different
areas of a city, considering that each part of the network will
have several consumption behaviors at the charging points.
Additionally, it gives the advantage to easily develop multiple
case studies by adjusting driver pattern probabilities, vehicle
characteristics, geographic and demographic parameters of the
city and even charging scenarios.
The proposed model seeks to provide information to easily
evaluate the impact on electrical demand resulting from
expected penetration scenarios of EVs on the distribution
system. The simulation method, as shown in Fig. 1, would
take as inputs a set of information regarding the distribution
system composition, EVs technical characteristics, EVs’ user
behaviors and market considerations in order to construct a
model that provides results to assess the impact on electrical
demand in the distribution system.
Such forecasting model is clearly subject to a large number
of uncertain variables, making it very difficult to model with
analytical methods. Hence, the simulation alternative arise as
the most reliable option to face this challenge, blended with a
stochastic analysis that would help to capture the temporal and
spatial diversity of EVs integration in the system.
The first step in the proposed forecasting model is to seek,
organize and treat all the real data that can be found about the
region or distribution system that is going to be analyzed and
the behavior of the users within it.
Figure 1. General scheme of the proposed model.
Mobility of vehicles can be modeled in macro-simulation
or micro-simulation models. The first one implies a global
statistical analysis which cannot easily represent the dynamics
of the system. Meanwhile, micro-simulation strategies models
the individual behavior of drivers and their choices, making it
easier to provide a disaggregate output. The big advantage of
this approach is the ability to develop multiple scenario studies
by modifying driver behaviors, vehicle characteristics or even
the topology of the analyzed system.
The proposed methodology uses the micro-simulation
philosophy. The system is divided in many areas based on
demographical and geographical characteristics. These areas
are determined based on the economic level of the
neighborhood. The next step is to define which will be the
EVs types with possibilities in the market and clearly its
participation in it or its penetration level.
The next step is the computation of the energy
consumption of each EV in the system and it’s charging
behavior. Thus, the proposed forecasting model is based on a
Monte Carlo simulation method in order to drive the big
amount of uncertainties in the model. After the analysis of the
EVs’ user behaviors is established, it is straightforward to
compute the corresponding percentage of battery discharge for
each one of the EVs. Such results obtained for each EV are
then aggregated for each area and finally all the areas will
represent the forecasting of the entire system. Fig. 2 shows the
conceptual model.
Figure 2. Conceptual model for forecasting energy demand of PEVs.
Several parameters that affect the charging patterns of the
EVs are involved in the computation process. These
parameters must be characterized in order to deal with the
uncertainties in the model. Among these are: the distribution
system topology, the penetration level in the studied area,
available charging power, vehicle battery capacities, state of
charge, EVs user behaviors and daily EVs energy
requirements. Such parameters are explained below in a more
detailed way.
A. Distribution System Topology
The proposed model seeks to analyze the energy
requirements in different areas of the city since each will have
several consumption behaviors and characteristics, making it
mandatory divide the system in a certain number of areas and
discriminate its parameters. So, a spatial and temporal
characterization of distribution system is required.
The information needed for each one of the areas in the
system would be: area-ID, number of users, percentage of
residential, industrial users, and commercial users; social-
economic stratus, mean energy demanded in kWh, home-work
mean distance and deviation [km], home-commerce mean
distance and deviation [km], work-commerce mean distance
and deviation [km], mean velocity and deviation [km/s],
probability of charging at home, work or at commercial place.
The required information is related mostly to the
demographical and geographical characteristics of each area.
Such information would be crucial to define the penetration of
EVs in all areas. The charging probabilities in each place must
be also defined for each area as well as the distance
distribution parameters between places which have been
commonly modeled with a lognormal distribution function.
Generally, several papers assume only one charging
possible scenario in the last arrival at home [1]-[4], other like
[4] and [5] consider various scenarios but with a discrete
philosophy. In this paper, it is important to highlight that the
charging probabilities in home (HCPAn), work (WCPAn) and
commercial place (CCPAn) in the area An, could represent not
only the availability of chargers in each area but also would be
a very flexible way to simulate different charging scenarios,
for example if HCPAn=1, WCPAn =CCPAn=0, it is represented
the assumption of only home charging.
B. Market Framework
The penetration level is the first parameter that must be
tackled when analyzing the impact of PEVs on power demand
in a specific region, as it provides the number of EVs that will
interact within the electric distribution network.
This parameter (PenLevel) represents the percentage, of the
whole automobile fleet, that would be boosted by electric
motors. In this way, the number of EVs (NumEVs) within the
system can be computed as the total number of vehicles
(NumVs) in the region, multiplied by the penetration level; as:
NumEVs = NumVs × PenLevel (1)
The deployment of EVs will depend on several factors,
within which stand two key aspects: technology maturation
and consumer acceptance of new vehicle types. The first one
is in turn influenced by the battery performance and the access
to charging infrastructure, while the second one regards
technology costs, type of business model implemented,
government policies, among others.
Such diversity of factors and their interrelationship are the
cause of the adversity in making projections and justify the
impossibility to define one single scenario about the
penetration level of EV. In literature, several assumptions
have been made on these aspects, in [1] the penetration level is
only determined in a random way based on the number of
residential users in the system, while others [2]-[5] try to
define the penetration in a given horizon of years but keeping
a constant and almost proportional growth.
In this paper the EV yearly penetration level Penyear is
modeled as a logistic function that allows the consideration of
the mentioned factors, as:
Pen year = Penmax 1 + e −α ( x − β ) (2)
where Penmax is the maximum penetration that can be
reached a long term, x is the corresponding year and α, β are
the parameters modeling the projection curve in the long-term.
An easy way to easily determine these parameters is as follow:
α = 10 YearsTechM , β = Yearsini + YearsBoom (3)
Where YearsTechM represents the years needed to reach the
maturation of the technology, Yearsini is the first year of the
defined horizon and YearsBoom is the years that would take the
start of the penetration growth. It is clear that these parameters
can be defined with a deep analysis of all the factors
mentioned before, to obtain a reliable forecasting.
According to [6], the logistic function can be interpreted in
four quadrants. The first one, representing a period leading up
to the introduction of commercially viable EVs but with some
constrains primarily of technical matters, a second quadrant
symbolizing the early stages of commercialization with an
evolving technology and new manufacturing facilities
emerging; the third one with the ramp-up of EV production,
mature technology and a significant expansion in the capacity
to manufacture and distribute them, and finally the fourth
quadrant due the ultimate constraints in the market [6].
This representation gives us the opportunity to define the
penetration curve based on some considerations regarding the
scenario desired to simulate. However, there is a remaining
step needed to capture the spatial intention of the method and
is the distribution of the EVs into the different areas given its
particular characteristics, as explained in the last subsection.
Considering that the number of particular EVs will depend
on the stratus of the area and the number of residential users,
in this paper is proposed to make a weighted distribution of
the EVs throughout the areas, as follows:
n
EVn = Pn × EVT ∑P
i =1
i (4)
Pn = Pen + Run (5)
Pen = En × (Run × We ) ET (6)
where EVn is the number of electric vehicles assigned to
area n. Pn is the adjusted weighing for area n and depends on
the number of residential users of the area (Run) and the fit
value of the weighing of the area (Pen). The last one is
computed by the stratus of area n (En), the number of
residential users, a weighing value (We) to enhance the weight
of the stratus parameter and the sum of the all possible stratus
values (ET). This approach helps to obtain more realistic
forecasts of EV loads over space and time, providing the
opportunity to link the real data of the topology of the system
with the implemented model.
C. EV Technical Characteristics
Once the distribution topology has been defined and the
penetration on each area has been established, it is necessary
to define the characteristics of the EVs that are going to be
considered in the simulation.
Commonly, the only parameters considered when
including EV technical characteristics in impact models are
the battery types and charger levels and always in a static way
during the whole simulation. However, the proposed model
tries to consider, also, the technology evolution during the
whole horizon of analysis.
This is made by modifying two specific characteristics of
the EV: the battery capacity and the consumption rate. This
adjustment is not made in all year simulation cycles and
should depend proportionally to the technology maturation.
Hence, in order to make more close the linkages of the whole
model, it was defined a probability of occurring the
technology enhancement based on the penetration curve
already stated. So, considering that the derivate of the
penetration curve will provide information about the
proliferation of the technology, the probability of technology
evolution was defined as:
dPen year e −α ( x − β )
Pen year ' = = (7)
dx (1 + e α ( β ) )
− x−
(
EPTech = Pen year ' max Pen year '
where EPTech is the probability of technology enhancement
and Penyear’ is the derivate of the penetration curve defined
before.
Fig. 3 shows the behavior probability of technology
enhancement throughout the years. It can be seen that the
years with the greatest probability of enhancement are the
years of proliferation of EVs as shows the penetration curve.
Figure 3. Probability of technology enhancement based on a specific
penetration curve.
On the other side, the charging level has been modeled by
some authors as a random variable in all the possible places
for charging EVs [7], but considering the possible
standardization of such chargers this approximation could
represent just stochastic noise to the model. Thus, in this
paper, these parameters are limited to three charging levels
shown in Table I, which are just selected in the model given
the place where the charging occurs.
Finally, it must be defined the power curve of an
individual EV during its charging process. Most of the
literature assumes the power drawn out by the batteries in the
charging process as a constant value during certain time.
However, other like [8] and [9] assume the energy drawn out
by the battery as an exponential curve and [10] uses a discrete
approximation with five sections to model a fast charge
process. In this case, for planning purposes, it is a good
approximation to assume it as a constant value, given by:
DurationCh arg e = ERe quired Ch arg ing Power
TABLE I. CHARGING LEVELS FOR ELECTRIC VEHICLES
Level Power Level
Level 1 – 120 VAC 1.2 - 2.0 kW
Level 2 – 240 VAC 9 - 15 kW
Level 3 – 480 VAC 60 - 150 kW
D. User Behaviors
This paper proposes a methodology to use driver behaviors
capturing the spatial and temporal diversity of the system and
assuming the stochastic behavior of some variables but also
giving strategic importance to some of the parameters that can
be captured from real data. One aspect needed to fulfill this
2
objective is the probability, along the day, of users to be in
each one of the places where is being considering the
) (8)
possibility of finding EV charging infrastructure, which in this
case are home, work and commercial places [11] (Fig. 4).
The second aspect needed to match the information in Fig.
4, is a distribution of the permanence duration given the time
of arrival at each place. In this case, given the lack of
information, it was useful a cyclical simulation based on data
given in Fig. 5, to obtain some statics of this parameter.
Figure 4. Probability, along the day, of users to be in the places where can be
found EV chargers.
Then, considering these two variables, the proposed
methodology starts with a sequential probabilistic simulation
of one day period with a defined resolution, which in this case
was of minutes. In each one of these cycles, the place and
duration times are computed according to the given
information and all the other input data explained before.
After the user behavior has been established, it is
straightforward to compute the corresponding percentage of
battery discharge for each one of the EVs and, based on the
desired scenario, the charging events throughout the day.
III. FORECASTING TOOL
Table II explains step by step the process of using and
simulating with the forecasting tool and they way to set or
manipulate the input data needed for the model.
IV. CASE OF STUDY
As study case, it was formulated a test distribution system
with five substations (A to E) with a total of 14 feeders, as
shown in Fig. 5. Table III shows the customer characterization
for each feeder for the five substations. In addition, those
(9) feeders are also divided in areas, as Fig. 5 shows. On the other
hand, the penetration curve was defined as shown in Fig. 6.
The total number of vehicles considered is 60.000 vehicles.
Several simulation trials and scenario definitions were
carried out. Table IV shows the results obtained with the
model where it is combined the normal load growth of
customers and the electric vehicle consumption.
 TABLE II. FORECASTING TOOL PROCESS DEFINITION TABLE III. CHARACTERIZATION OF THE DISTRIBUTION TEST CIRCUIT

1.0: START Number Number Number Average

# of
Feeder residential commercial Industrial Demand
2.0: Load input Data and Initialize Simulation Variables. customer customers customers (MWh/month)
Vehicles
Load Distribution System Characteristics, Characteristics of A11 9868 403 32 5553,4 5537
2.1: the different EV types, Initial Demand Curve of Areas. (From A21 8502 1122 198 1522,7 3400
Excel File) A22 4429 692 6 1674,7 3896
2.2: Define Penetration Curve. (Manual – Interface Tool) B11 7822 406 133 3309,5 3547
Define Simulation Method of Drivers Behavior and Drivers B12 6436 198 7 3426,8 4060
2.3: B21 11015 363 0 4000,3 4916
Patterns. (Manual – Interface Tool or From Excel File)
C11 6568 671 21 5276,7 3610
Define Additional Parameters: # of Monte Carlo Simulations,
C21 6386 427 22 4964,9 3758
2.4: # Vehicles, Parameters of Initial SOC Distribution and SOC C22 6982 1185 91 5234,6 3280
thresholds. (Manual – Interface Tool) D11 10105 841 115 3875,8 4323
Define Additional Parameters: # of Monte Carlo Simulations, D21 8057 509 77 3135,6 3871
2.5: # Vehicles, Parameters of Initial SOC Distribution and SOC D22 7746 1334 175 4565,9 4095
thresholds. (Manual – Interface Tool) E11 5198 390 182 2738,9 3754
2.6: Select Charging Strategy. (Manual – Interface Tool) E21 13876 247 15 3596,7 7953
3.0: Run Forecasting Model. (Manual – Interface Tool)
3.1: Compute Probability of Enhancement Curve
3.2: Monte Carlo Simulation Cycle. (Loop)
3.3: Years Cycle (Loop)
3.4: Modify Penetration for the Year
3.5: Modify Parameters Due to Evolution of Technology
3.6: Areas Cycle (Loop)
3.7: EV Types Cycle (Loop)
3.8: # of EVs Cycle (Loop)
Driver Behavior Definition According to the
3.9: Selected Method and Considering all the
Information and Parameters Stated Above. Figure 6. Market penetration curve of one of the scenarios simulated.
3.10: End of # of EVs Cycle
3.11: End of EV Types Cycle TABLE IV. RESULTS OF ENERGY DEMAND
3.12: End of Areas Cycle
3.13: End of Years Cycle | Demand [GWh/year]
3.14: End of Mote Carlo Simulation Cycle. Feeder Year Year 2030 Year 2046
Segment 2010 No EV With EV No EV With EV
4.0: Select Results to Plot. (Manual – Interface Tool)
A11 66,64 140,40 142,31 262,90 268,54
4.1: Plot Entire Energy System Forecasting A21 18,27 38,50 40,50 72,08 78,00
4.2: Plot Energy Forecasting By Areas A22 20,10 42,34 48,35 79,28 97,01
Plot Power Demand Curve for Selected Area. (Manual – B11 39,71 83,67 85,32 156,67 161,53
4.3:
Interface Tool) B12 41,12 86,63 87,69 162,22 165,33
5.0: END B21 48,00 101,13 102,95 189,37 194,73
C11 63,32 133,40 136,37 249,80 258,53
As shown in Table IV, the EV energy demand impact, as C21 59,58 125,52 127,41 235,04 240,61
C22 62,81 132,34 134,94 247,80 255,46
percentage growth, is different between several segments of D11 46,51 97,99 99,87 183,48 189,03
feeders; for example, the additional energy in EV in area B12 D21 37,63 79,27 81,04 148,44 153,64
is 1.9% while in area A22 is 22.4%. In this case, it is assumed D22 54,79 115,43 118,91 216,15 226,39
that the battery of the vehicle could be charged at home, or at E11 32,87 69,24 71,95 129,66 137,64
E21 43,16 90,93 94,43 170,27 180,58
work place or in commercial or public places.
Fig. 7 compares the demand profile of area A11-B for the
initial load considered for the area and the one resulting after
adding the EV load. In this case, it was assumed only
residential charging.
On the other hand, Fig. 8 shows the demand profile of area
A11-B, but with the possibility of work place and commercial
place charging, also. The result in this case shows a clear
reduction of the EV demand in the peak hours where probably
the user is at home.
Comparing Fig. 7 and Fig. 8, it can be seen how the
charging probability parameters are a very useful tool to
generate different charging scenarios and evaluate their impact
on the daily demand profile.
The error bars in Fig. 7 and 8 are the standard deviation
obtained after the mean value of all Monte Carlo cycles is
computed.
Figure 5. Test distribution system.

Figure 7. Demand profile of area A11-B in 2030 with a penetration level of
40% and stating HCP=1, WCP=CCP=0.
Figure 8. Demand profile of area A11-B in 2030 with a penetration level of
40% and stating HCP=0.2, WCP=CCP=0.4.
After evaluating the daily demand profiles impact, the
attention went to the micro-area analysis of the system to see
how the characteristic defined could affect the energy demand
impact in each area. For this, a 3D graphic showing the annual
energy demanded in each area for each one of the years in the
period of analysis was obtained (Fig. 9). Finally, after the
aggregation of all the areas in the system, the entire energy
demand forecasting is obtained. Fig. 10 shows as the load
growth in long-term including PEV loads.
Figure 9. Annual energy EV demand in each area.
Figure 10. Energy forecasting for the whole system.
As mentioned before, the simulation tool was implemented
with the intention of running all the desired scenarios and
modifying all the input parameters to evaluate its impact on
the model. So the graphs that have been shown are only one
demonstration piece of the big amount of alternatives that the
model gives us to evaluate.
V. CONCLUSIONS
This paper has proposed a methodology for forecasting the
electrical power demand associated to electric vehicles on a
distribution system; taking into account a temporal and spatial
conception that considers a detailed analysis of demographical
aspects of EVs utilization and its technical characteristics.
This approach allows the assessment of the impact several
penetration levels of EVs in the different areas of a
distribution system, considering that each part of the network
will have distinctive consumption behaviors.
The defined model is an useful tool for distribution
systems not only for the evaluation of possible impacts in the
system, but also as a business tool for the analysis of user
behaviors, in order to implement commercial strategies and
define requirements to satisfy their clients necessities.
Strategic consideration of the available real data in the
development of models is a crucial factor to obtain reliable
results, as actual driver behaviors would be almost maintained
due to the imperative necessity of keeping up with user’s
mobility requirements.
REFERENCES
[1] D. J. Kang, S. Park, “A Case Study on the Grid Impact of PHEV to
Distribution Power System”, in Proc. 44th Hawaii International
Conference on System Sciences, 2011.
[2] K. J. Yunus, M. Reza, H. Zelaya-De La Parra, K. Srivastava, "Impacts
of Stochastic Residential Plug-In Electric Vehicle Charging on
Distribution Grid," in Proc. Innovative Smart Grid Technologies
(ISGT), 2012 IEEE PES , 16-20 Jan. 2012.
[3] S. Acha, K.H. van Dam, N. Shah, "Modelling spatial and temporal
agent travel patterns for optimal charging of electric vehicles in low
carbon networks," in Proc. 2012 IEEE Power and Energy Society
General Meeting, 22-26 July 2012.
[4] L. Gan, Z. Xiao-Ping, "Modeling of Plug-in Hybrid Electric Vehicle
Charging Demand in Probabilistic Power Flow Calculations," IEEE
Transactions on Smart Grid, vol.3, no.1, pp. 492-499, March 2012.
[5] C. Farmer, P. Hines, J. Dowds, S. Blumsack, "Modeling the Impact of
Increasing PHEV Loads on the Distribution Infrastructure," in Proc.
43th Hawaii International Conference on System Sciences, 5-8 Jan.
2010.
[6] P. J. Balducci, “Plug-in Hybrid Electric Vehicle Market Penetration
Scenarios”, U.S. Department of Energy, PNNL-17441, Sept. 2008.
[7] Sungwoo Bae; Kwasinski, A., "Spatial and Temporal Model of Electric
Vehicle Charging Demand," IEEE Transactions on Smart Grid, vol. 3,
no. 1, pp. 394-403, March 2012.
[8] P. Bauer, Yi Zhou, J. Doppler, N. Stembridge, "Charging of electric
vehicles and impact on the grid," in Proc. 13th International
Symposium Mechatronika, pp.121-127, 2-4 June 2010.
[9] I. Roncancio, M.A. Rios, "Spectral and steady state impact assessment
of PHEV on distribution systems," in Proc. 2013 Workshop on Power
Electronics and Power Quality Applications (PEPQA), 6-7 July 2013.
[10] EV Battery Monitoring. [Online] Available:
http://batteryuniversity.com/learn/article/Charging_lithium_ion_batteri
es
[11] C&M, Duarte Guterman y Cia Ltda, “Socioeconomic characterization
of Bogotá and its Region” (in spanish), Bogotá City Hall, Traffic and
Transport Secretary, 2005.