Proceedings of the ASME International Mechanical Engineering Congress & Exposition

IMEC2010
November 12-18, 2010, Vancouver, British Columbia, Canada

DRAFT-IMECE2010-38240

FORECASTING THE IMPACT OF PLUG-IN HYBRID ELECTRIC VEHICLES

PENETRATION ON ONTARIO’S ELECTRICITY GRID

Lena Ahmadi Woramon Unbangluang

University of Waterloo King Mongkut's University of Technology
Waterloo, ON, Canada Thonburi, Bangkok, Thailand
l2ahmadi@uwaterloo.ca patlompatlom@hotmail.com

Eric Croiset Ali Elkamel

University of Waterloo University of Waterloo
Waterloo, ON, Canada Waterloo, ON, Canada
ecroiset@uwaterloo.ca aelkamel@uwaterloo.ca

Peter L. Douglas Evgueniy Entchev Hong-Ming Ku

University of Waterloo CANMET Energy Technology Centre, King Mongkut's University of
Waterloo, ON, Canada Natural Resource Canada, Ottawa, Technology, Thonburi,
pdouglas@uwaterloo.ca ON, Canada Bangkok, Thailand
Evgueniy.entchev@nrcan-rncan.gc.ca hmku@inet.co.th

ABSTRACT from charging PHEVs is incorporated for electricity production

Vehicle emissions are a major concern in the development
of new automobiles. Plug-in hybrid electric vehicles (PHEVs)
have a large potential to reduce greenhouse gases emissions and
increase fuel economy and fuel flexibility. PHEVs are propelled
by the energy from both gasoline and electric power sources.
Penetration of PHEVs into the automobile market affects the
electrical grid and increasing the electricity demand has not
been fully investigated.
This paper studies effects of the wide spread adoption of
PHEVs on peak and base load demands in Ontario, Canada.
Long-term forecasting models of peak and base load demands
and the number of light-duty vehicles sold are developed. To
create proper forecasting models, both linear regression (LR)
and non-linear regression (NLR) techniques are employed,
considering different ranges in the demographic, climate and
economic variables.
The results from the LR and NLR models (LRM and
NLRM) are compared and the most accurate one is selected.
Furthermore, forecasting the effects of PHEVs penetration is
done through consideration of various scenarios of penetration
levels, such as mild, normal and aggressive ones. Finally, the
additional electricity demand on the Ontario electricity grid
planning purposes.
Keywords: Plug-in hybrid electric vehicle; electricity grid;
peak load demand; base load demand; load forecasting
INTRODUCTION
The transportation sector is one of the main energy
consumption sectors. Technologies for vehicles has been
developed over many decades to improve performance of
vehicles, reduce energy consumption, and reduce pollution
released into the environment. PHEVs combine the combustion
engine of conventional vehicles and the electric motor of
electric vehicles. PHEVs have greater fuel efficiency because
they consume less fuel than in conventional vehicles in which
gasoline is the only energy source. PHEVs battery can be
recharged by connecting into the electrical grid. This makes
PHEVs become ‘fuel flexible vehicles’ because they can use
both gasoline and electricity for propulsion.
One of the major challenges of PHEVs development is
battery selection. The main challenges with batteries are their
cost, performance and lifetime. At present, many researchers
attempt to develop PHEVs batteries that will overcome these
challenges [1, 2]. Another challenge of PHEVs is their
environmental impact. Although PHEVs can reduce the

1
emission from tailpipe to the environment, the emission shifts
to the power plant where electricity may be produced from
fossil fuel power plants. However, if electricity is produced
from nuclear, solar, hydro or wind power plants, emission from
the power plant becomes almost zero.
Ontario Power Generation (OPG), an Ontario-based
electricity generation company, attempts to minimize the
environment impact from their power plants in many ways.
OPG plans to phase out their coal plants by the end of 2014 [3].
Nowadays, approximately 78 percent of OPG’s electricity
production is generated from hydroelectric and nuclear
generating stations [4]. These stations produce no emissions
that contribute to global warming.
Several automobile manufacturers plan to produce PHEVs
commercially by the end of 2010 [5]. When PHEVs penetrate
the automobile market, gasoline consumption will decrease, but
electricity demand will subsequently increase. The next
challenge of PHEVs is to know whether the electricity grid is
capable of supplying the increased electricity demand from
charging PHEVs.
Electricity needs long-term load forecasts since it takes
many years to plan and increase capacity of existing power
plants and to build new power plants. For this reason, this study
focuses on long-term load forecasting to estimate whether the
supplied electricity can meet additional demands from charging
PHEVs.
Mohamed and Bodger [6] developed a model for long-term
forecast of the annual electricity consumption in New Zealand
by using the multiple LR method. Explanatory variables that
influence the forecast model were GDP, average price of
electricity, and population. Pao [7] developed LR and NLR
artificial neural network models to forecast the monthly
electricity consumption in Taiwan. National income,
population, GDP and consumer price product were proposed as
explanatory variables. Chui et al. [8] focused on the long-term
forecasting of electricity demand such as the annual energy,
peak load and base load demand using autoregressive, simple
LRMs and multiple LRMs. The list of potential explanatory
variables included GDP, GDP per capita, number of people
employed, population, dwelling counts, Heating Degree Days
(HDD), and Cooling Degree Days (CDD). Hadley [9] studied
the impact of PHEVs on the electricity grid of the VACAR sub-
region (North and South Carolina and much of Virginia) in
2018. He assumed that the market share for PHEVs in 2018 is
25%. Yu [10] investigated the impact of PHEVs charge profiles
on generation expansion by assuming high PHEVs penetration.
Hajimiragha et al. [11] performed a feasibility analysis of
optimally utilizing Ontario’s electricity grid for charging
PHEVs during off-peak period. They studied two different
transitions of PHEVs penetration.
In forecasting electricity consumption, most of the
literature employed weather, demographic and economic
variables as explanatory variables. Moreover, results from
NLRMs yield better forecast performance than those from
LRMs. Therefore, this study focuses on the impact of those
explanatory variables on the model and also compares the mean
error between LRMs and NLRMs in order to choose the best
model to represent peak and base load demands and the number
of light-duty vehicles sold. For PHEVs charging part, both peak
and off-peak charging are considered. Charging of PHEVs
during the off-peak period is the best case, in that it a lower
impact on the grid than charging during the peak period.
However, not all the people charge their PHEVs during the off-
peak period. There is a probability for those people to charge in
the peak-period. Thus, it is necessary to study the effect of
charging in both periods.
ISSUES AND ASSUMPTIONS
PHEVs are not commercially produced at the present time,
therefore, it is assumed that there is no PHEV in January, 2010.
Besides, only new light-duty vehicles are considered as
potentially new PHEVs since they have more potential to be
PHEVs.
Batteries capacity is another issue, especially for charging
patterns. Deeper battery charging and discharging cycles than
conventional hybrids are required for PHEVs. Since, battery
life is influenced by the number of full cycles, this may be
smaller than in traditional HEVs which do not deplete their
batteries as completely. In addition, design issues and trade-offs
against battery life, capacity, heat dissipation, weight, costs, and
safety are batteries limitations. In this work 80% safety factor
of 80% and 82% of charger efficiency are assumed.
To calculate the demand from charging, identification of
types of PHEVs that will penetrate the transportation sector is
essential. Based on the average commuting distance in Ontario,
PHEV-20 is assumed to be thw main PHEV that will penetrate
the light-duty vehicles sector. Another assumption is that no
PHEVs are retired during the period under study, 2010-2025.
FORECAST METHODS
Forecast methods are broadly classified into two basic
types: (i) qualitative methods and (ii) quantitative methods.
Qualitative methods normally use the opinions of experts to
predict future events. Such methods are often required when
historical data are either not available at all or scarce. In
contrast to qualitative methods, quantitative methods are based
on mathematical or statistical models. These methods involve
the analysis of historical data in an attempt to predict future
values. Quantitative methods can be classified into three kinds:
(i) time series models, (ii) regression models and (iii)
nonparametric models.
Regression models involve the identification of other
variables that are related to the variables to be forecasted. Once
these related variables have been identified, a statistical model
that describes the relationship between these variables (also
called independent variables, explanatory variables, regressors
or drivers) and variables to be forecasted (also called dependent
variables) is developed. Having determined this relationship,
the forecaster would use predicted future values of the
explanatory variables to predict future value of dependent
variables. There are two types of regression models: (i) LRM
and (ii) NLRM.
2
Linear regression model, LRM
LR means a regression that is linear with respect to the
parameters but not necessary linear with respect to explanatory
variables. The general form of a LRM can be written as
𝑌 = 𝑏� + ∑���� 𝑏� 𝑥� (1)
where b0 and bi represent unknown parameters, whereas xi is an
explanatory variable and Y is the dependent variable. b0 and bi
are referred to as the intercept and slope terms, respectively.
The intercept term, b0, tells what the expected value of Y would
be when all explanatory variables are zero. The slope term, bi,
tells how Y is expected to change for each one-unit increase in
xi.
Non-linear regression model, NLRM
NLRM refers to a model in which the parameters appear
nonlinearly. There are many forms of NLRM. Since in the
prediction of this study, an increase in explanatory variables
may lead dependent variable to gain slowly or to be constant,
an asymptotic regression model is employed to represent
NLRMs in this study. The equation of the asymptotic regression
model is given below.
𝑌 = 𝑏� + 𝑏� exp(𝑏� 𝑥) (2)
where b0, b1 and b2 are unknown parameters and x is
explanatory variable. When b0>0, b1<0, and b2<0, it gives the
Mistcherlich's model of the "law of diminishing returns". This
model initially increases quickly with increasing values of x,
but then it increases slowly and finally tapers off just below the
value b0.
For historical data of forecast variables which has seasonal
patterns, only LR or only NLR cannot create such pattern. One
technique to cope with this problem is combining LRMs and
NLRMs with seasonal time series. The general form of seasonal
time series model is
��� ���
𝑌 = 𝑏� + 𝑏� cos � � + 𝑏� sin � � (3)
� �
where b0, b1 and b2 are unknown parameters whereas t is time
index and L is time interval from one point to another point of
highest value of Y. Another technique is adding dummy
variables (x1, x2 and x3) into the model. The general form of a
model with dummy variables can be written as:
𝑌 = 𝑏� + 𝑏� x� + 𝑏� x� + 𝑏� x� (4)
where b0, b1, b2 and b3 are unknown parameters and x1, x2 and
x3 are dummy variables. This technique can create seasonal
patterns by assuming x1=1 and x2=x3=0 for winter, x2=1 and
x1=x3=0 for spring, x3=1 and x1=x2=0 for summer and lastly let
x1=x2=x3=0 for autumn.
METHODOLOGY
The methodology used in this study consists of five steps.
The detail of each step is described in the following.
STEP 1: Study and data collection
Different forecast techniques and model selection criteria
are studied to choose a suitable method. Both LR and NLR
techniques are employed to create proper forecast models.
Dependent variables are peak and base load demands (PEAK
and BASE) and light-duty vehicles sold (VEH). Peak load
demand is the maximum demand in each day normally
occurring between 9 a.m. and 9 p.m. For base load demand, it is
defined as the minimum amount of power that power plants
must make available to customers. It can be calculated by
averaging daily demands in weekday. Explanatory variables
that may impact the PEAK and BASE models are broadly
divided into three groups: (i) weather variables such as
temperature (T), relative humidity (RH) and wind speed (WS),
(ii) demographic variable such as population size (POP),
income (INC), number of employments (EMP), and (iii)
economic variable which is gross domestic product (GDP).
To forecast VEH, the number of new graduated students
(EDU) is also an important factor. People who get degrees from
both undergraduate and graduate level tend to buy new cars
more than others. Therefore, the number of graduated students
is one of explanatory variables to forecast the number of light-
duty vehicles sold.
Historical data of dependent and explanatory variables are
collected to fulfill the required data for developing the models.
Before achieving historical data of peak and base load
demand, outlier determination is an important step since they
can cause poor forecasting results. In this study, Statistics
Package for Social Science (SPSS) version 15.0 was used to
develop forecast models. SPSS also has a potential to identify
outliers among inputs by using boxplot. All outliers are omitted
from the data and replaced by the less extreme values.
An important possible issue with explanatory variables is
multicollinearity problems. Multicollinearity occurs when two
or more explanatory variables are highly correlated. As a result,
regression procedures may not be able to distinguish the
separate contributions of these variables to the dependent
variable, and the estimation of unknown parameters may be
unreliable. Ordinary multicollinearity is the situation in which
there is a close, but not perfect, linear relationship between
some of the explanatory variables in the sample data.
Multicollinearity is usually considered to be a data or sample
problem. The principle of parsimony (using the simpler model
when greater complexity does not provide significant benefits)
suggests that when two or more variables are highly correlated,
one of them should be omitted from the model. Matrix scatter
in SPSS is used for detecting multicollinearity. The scatter that
presents linear relationship between two explanatory variables
indicates multicollinearity problem.
3
STEP 2: Model development and selection
In this step, models for forecasted peak and base load
demands and light-duty vehicles sold are developed using
LRMs and NLRMs. Historical data of peak and base load
demands are monthly data. Considering simplicity of
forecasting, four months representing each season are used as
input in the model development. Hence, eight models are
developed to represent peak and base load demands in four
selected months. Four of them are used for forecasting peak
load demand and the rest are used for forecasting base load
demand. For historical light-duty vehicles sold, information is
provided by season. Therefore, one model is formulated to
represent light-duty vehicles sold in all seasons.
- Linear regression with SPSS
To formulate a LRM in SPSS, dependent and explanatory
variables need to be defined. Details of these variables are
discussed in the first step. A lists of variables used as input of
SPSS are summarized in Table 1.
Table 1 Inputs variables for linear regression
Dependent variables Explanatory variables
PEAK, ln(PEAK) T, RH, WS, POP, INC, EMP, GDP, ln(T), ln(RH),
ln(WS), ln(POP), ln(INC), ln(EMP), ln(GDP)
BASE, ln(BASE)
VEH, ln(VEH) POP, INC, EMP, EDU, GDP, ln(POP), ln(INC),
ln(EMP), ln(EDU), ln(GDP)
To determine which combination of explanatory variables
provides the best fit to the data, SPSS has an automated process
for variable selection called ‘stepwise regression’ in which the
regression equation is automatically estimated several times.
forecast future data. The equation of MAE can be written as
follows:
∑�
���|�� | ∑�
���|�� ��
�� |
MAE = = (5)
� �
where et is the error term, n is the total number of observation
and t is time index. yt and y� � are the observed and estimated
values, respectively.
STEP 3: Projection of forecast variables
The best models for forecasting peak and base load
demands and light-duty vehicles sold are used for projecting the
future value of those dependent variables from 2010 to 2025.
Future values of all explanatory variables shown in the selected
models are substituted into those models in order to predict
values of the dependent variables.
STEP 4: Study of PHEVs penetration and charging
pattern
Since PHEVs are not commercially produced at this
moment, this study assumes that there is no PHEV in January,
2010. Three transitions of PHEVs penetration in the light-duty
vehicles sold are shown in Fig. 1 assuming 10%, 30% and 50%
of PHEVs penetration in December, 2025, respectively. In this
study, only new light-duty vehicles are considered.
Before studying charging patterns, it is necessary to
identify which types of PHEVs will penetrate into the
transportation sector. To choose appropriate types of PHEVs
that match people’s lifestyle in Ontario, commuting distance
must be considered.
50
Transition 1
Percent of PHEVs penetration (%)

- Non-linear regression with SPSS 45 Transition 2

NLR in SPSS does not have tool to choose the best 40 Transition 3
combination of explanatory variables unlike LR. Therefore, 35
selecting a set of explanatory variables should be done 30
manually. To eliminate the complexity, only multiple NLRMs 25
are considered which means only two explanatory variables are 20
used as input of the models. In addition, pairing of explanatory 15
variables which are highly correlated must be omitted to 10
prevent the multicollinearity problem. Logarithm terms of both 5
dependent and explanatory variables are not included. Possible 0
combinations of explanatory variables are illustrated in Table 2.
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025

Table 2 Possible explanatory variables combination for NLR Year

Dependent variables Combination of explanatory variables Fig. 1. Percent of PHEVs penetration in Ontario
PEAK T vs RH RH vs WS WS vs POP
T vs WS RH vs POP WS vs INC Table 3 compares commuting distances in Canada and
T vs POP RH vs INC WS vs EMP
T vs INC RH vs EMP WS vs GDP
Ontario. The average commuting distance in Ontario is 12.9 km
BASE
T vs EMP RH vs GDP (=8 miles). This implies that PHEV-20, which can travel twenty
T vs GDP miles without using its combustion engine, is appropriate for
people in Ontario. Therefore, this study assumes that only
After finding all possible LRMs and NLRMs, the next step PHEV-20 penetrates into the light-duty vehicles sold. Another
is model selection. Mean Absolute Error, MAE, is employed as assumption is that no PHEVs are retired during the period
a criterion for selecting the best model. The model that has the under study.
lowest MAE is chosen to represent the historical data and also

4
 Since most household outlets already contain 120 V/15 A Table 5 Details of charging scenarios
outlets, it is assumed that all PHEV-20 will be recharged Scenario
Percent of PHEVs
through this circuit every day. Charger requirements of PHEV- Peak period Off-peak period
20 with 120 V/15 A outlets are summarized in Table 4. 1 100 0
2 75 25
Table 3 Average commuting distance in Canada and Ontario 3 50 50
[12] 4 25 75
Commuters (people) 5 0 100
Commuting distance (km)
Canada Ontario
Less than 5 km 4,741,630 1,672,260 STEP 5: Comparisons of worse case scenario with
5 to 9.9 km 2,962,810 1,101,410 Ontario’s available resources
10 to 14.9 km 1,738,750 672,685 In this step, the worst case of penetration level and
15 to 19.9 km 1,095,465 475,410 charging scenario is chosen as the case study. The demand of
20 to 24.9 km 693,645 318,960 the worst case is compared with available resources in Ontario
25 to 29.9 km 461,250 213,460
to see whether they can supply increasing demand from PHEVs
30 km or more 1,376,340 640,470
charging.
Average commuting distance (km) 11.9 12.9

MODEL DEVELOPMENT
Peak load demand models
Table 4 Charger requirements for PHEV-20 under 120 V/15 A Using the above selection approach, LRMs for peak load
outlets [13, 14] demand forecast in January, May, August and October are
Rated pack size Charging Charger Charging chosen to be:
a b
Vehicle type (kWh) size rate time

JAN: ln(PEAK � ) = 9.7 + 5 ∗ 10�� GDP� − 9.1 ∗ 10�� T�

20 miles 8 miles (kW) (kW) (hour)
(8)
Compact Car 4.10 1.64 1.44 1.18 1.39
Mid-Sized Sedan 4.70 1.88 1.44 1.18 1.59
MAY: PEAK � = −36,901.3 + 4,138.6 ln(GDP� ) (9)
Mid-Sized SUV 6.30 2.52 1.44 1.18 2.13
Full-Sized SUV 7.40 2.96 1.44 1.18 2.51
AUG: PEAK � = 79,886 + 6,439.8 ln(GDP� ) + 5,205.1 ln(T� )
Average 5.63 2.25 1.44 1.18 1.91
Note: a An 80% required safety factor for continuous charging is used.
(10)
b
Charger efficiency is assumed to be 82%
OCT: PEAK � = −42,007 + 4,584.2 ln(GDP� ) (11)
New peak and base load demands represent PHEVs
charging in peak and off-peak period, respectively. They can be The peak load demand in January and August are a
calculated by adding the amount of PHEVs charging in each function of temperature and GDP, while the peak load demand
period with peak and base load demands obtained from in May and October are a function of GDP only. Temperature
regression models. Equations for calculation new peak and base has much effect on the winter and the summer. In the winter
load demands are shown below: (eq. (8)), the temperature is always below zero; therefore, the
lower the temperature is, the higher the peak load demand will
𝑃𝑒𝑎𝑘�,� = 𝑃𝑒𝑎𝑘�,� + 𝐶𝑅 × 𝑃𝐻𝐸𝑉𝑠� (6) be. This is because people need more electricity for space
cooling. On the other hand, during summer (eq. (10)),
���������
𝐵𝑎𝑠𝑒�,� = 𝐵𝑎𝑠𝑒�,� + (7) electricity consumption increases because more electricity is
�� ���
required for space heating. GDP is the only explanatory
where Peakn,i and Basen,i are new peak and base load demands variable which affects the peak load demand in all four months.
after adding the amount of PHEVs charging, respectively. GDP indicates the direction of economic growth. From eq. (8)
Peakr,i and Baser,i are peak and base load demands obtained to eq. (11), all coefficients of GDP are positive; hence, greater
from regression models, respectively. CR is the charger rate. number in GDP results in higher peak load demand.
PHEVsp and PHEVsop are number of PHEVs charging in peak The best NLRMs of peak load demand in four selected
and off-peak periods, respectively. BS is the battery size or the months are shown below:
rated pack size.
PHEVs can be recharged in both peak periods and off-peak JAN: PEAK � = 46,835.6 − 24,929.1(exp(8 ∗ 10�� T� )
periods. Details of each scenario are illustrated in Table 5. +exp(−6.2 ∗ 10�� GDP� )) (12)
Scenario 1 represents the worst case of charging scheme since
all PHEVs are assumed to be recharged during the peak period MAY: PEAK � = 17,911.3 − 73,141 exp(−10�� GDP� ) (13)
whereas Scenario 5 represents the best case which all PHEVs
are recharged during the off-peak period. AUG: PEAK � = 22,996.1 − 42,633.2(exp(−0.2 T� )
+exp(−7 ∗ 10�� GDP� )) (14)

5
OCT: PEAK � = 19,899.2 − 18,566 exp(−4.8 ∗ 10�� GDP� )
(15)
The same trends are found as with LRMs. Temperature
affects base load demands in January and August, while GDP
affects base load demand in all four months. All coefficients of
NLRMs follow the law of diminishing returns which b0>0,
b1<0, and b2<0. It means that the results of the models first
increase quickly when increasing temperature and GDP and
then increase more slowly. For eq. (12), the coefficient of
temperature is positive; however, when multiplying with
temperature in the winter which is always negative, this term
will be negative which follows the law of diminishing returns.
All temperature are given in °C.
Base load demand models
Using the same selection approach as in the case of peak
load demand, the best LRMs and NLRMs of base load demand
in January, May, August, and October are shown below:
LRMs:
JAN: BASE� = 13,545.5 − 176.2 T� + 9.4 ∗ 10�� GDP� (16)
MAY: BASE� = −36,998.5 + 4,048.2 ln(GDP� ) (17)
AUG: BASE� = 73,229 + 5,650.9 ln(GDP� ) + 5,644.7 ln(T� )
(18)
OCT: BASE� = −36,509 + 4,047.3 ln(GDP� ) (19)
NLRMs:
JAN: BASE� = 57,741.6 − 22,428.9(exp(8.3 ∗ 10�� T� )
+exp(−5.5 ∗ 10�� GDP� )) (20)
MAY: BASE� = 17,260.4 − 20,676.8 exp(−5.8 ∗ 10�� GDP� )
(21)
AUG: BASE� = 21,072.6 − 42,799.7(exp(−0.1 T� )
+exp(−7.4 ∗ 10�� GDP� )) (22)
OCT: BASE� = 17,632.9 − 16,266.4 exp(−5.4 ∗ 10�� GDP� )
(23)
Trends for base load demand forecast are similar to those
of peak load demand forecast. Both LRMs and NLRMs of base
load demand forecast in January and August depends on the
temperature and GDP and those of base load demand forecast
in May and October depend only on GDP.
Light-duty vehicles sold
Best LRMs and NLRMs are shown below:
LRMs:
ln (VEH� ) = 11.1 − 0.1 x� + 0.3 x� + 0.1 x�
+ 2.1 ∗ 10�� GDP� (24)
NLRMs:
VEH� = 75,841.4 − 6,851.4 x� + 24,372.9 x� + 7,784.4 x�
−39,580.2 exp (−6.1 ∗ 10�� GDP� ) (25)
where x1, x2 and x3 are dummy variables. Both LRMs and
NLRMs of light-duty vehicles sold forecast consist of these
dummy variables and GPD. From eq. (24) and eq. (25), the
number of light-duty vehicles sold increases when increasing
GDP because people have more potential to buy new vehicles
when the economic growth is positive.
MODEL SELECTION
Comparisons are made among LRMs and NLRMs. MAE is
employed as the criterion to see which model yields the most
accurate results. MAEs of all regression models of peak and
base load demands and light-duty vehicles sold are compared in
Table 6.
For peak load demand, NLRMs of all four months yield
lower MAEs than LRMs. Therefore, NLRMs represented in eq.
(12) to eq. (15) are selected to represent peak load demand in
January, May, August, and October, respectively.
When comparing between LRMs and NLRMs of base load
demand, NLRMs in May, August, and October gives smaller
MAEs. The opposite result is found in January. The LRM of
January yields lower MAE than NLRM. However, the
difference between MAE of LRMs and NLRMs is very small
(approximately 1.8%). Therefore, the LRM represented by eq.
(16) is employed to represent base load demand in January and
NLRMs represented by eq. (21) to eq. (23) are used to illustrate
base load demand in May, August, and October, respectively.
For light-duty vehicles sold, the LRMs gives better results
than NLRMs, but there is a slight difference between MAE of
both regression models (approximately 0.4%). In this case, the
LRM represented by eq. (24) is chosen to represent the number
of light-duty vehicles sold due to lower mean absolute error of
the model.
In summary, most of NLRMs yield lower MAE than
LRMs. This implies that the relationship between forecast
variables (peak and base load demands) and explanatory
variables (temperature and GDP) are not always linear.
Although LRMs give better results than NLRMs in some cases
(base load demand in January and light-duty vehicles sold), the
difference between MAEs of both regression models is
insignificant.
PROJECTION OF FORECAST VARIABLES
From the previous section, the best models of peak and
base load demands and light-duty vehicles sold depend upon
the temperature and GDP. The projections of peak and base
load demands, without PHEVs, and light-duty vehicles sold
until 2025 are shown in Fig. 2 and Fig. 3, respectively.
6
Table 6 Model comparisons of temperature substituting in the regression models is assumed
Forecast variables
MAE
Selected models
to be constant. Therefore, the difference of peak load demand
LRM NLRM from 2010 to 2025 depends only upon GDP.
1. Peak load demand When considering the projection of light-duty vehicles sold
- January 307.6 300.3 NLRM (eq. (12)) (Fig. 3), it was found that most of light-duty vehicles are sold in
- May 323.7 273.0 NLRM (eq. (13)) the spring. This brings about greater number of light-duty
- August 443.8 420.7 NLRM (eq. (14)) vehicles sold in the spring. The rank of light-duty vehicles sold
- October 415.2 391.4 NLRM (eq. (15)) from high to low is: spring > summer > autumn > winter.
2. Base load demand
- January 327.6 333.4 LRM (eq. (16)) EFFECTS OF PHEVS PENETRATION
- May 336.4 308.9 NLRM (eq. (214)) In the study of PHEVs penetration, three transitions are
- August 386.2 360.5 NLRM (eq. (225)) assumed to represent PHEVs penetration from 2010 to 2025
- October 390.3 372.9 NLRM (eq. (236)) which is shown in Fig. 1. Other assumptions used in PHEVs
3. Light-duty vehicles sold 6,537.6 6,564.5 LRM (eq. (24)) charging demand calculations are listed below:

30,000
Peak load
- Only PHEV-20 penetrates into Ontario’s transportation sector.
28,000 Base load - No PHEVs are retired from 2010 to 2025.
Peak load (IESO) - All PHEVs are recharged through the circuit during the peak
26,000
period every day (worse case scenario).
Demand (MW)

24,000

22,000 Figure 4 represents accumulative numbers of PHEVs in the

20,000
Ontario’s transportation sector in various transitions of PHEVs
penetration levels. The total number PHEVs at the end of 2025
18,000
for Transition scenario 1, 2 and 3 are 178,076, 534,214 and
16,000 890,362, respectively.
14,000
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025

1,000,000
Transition 1
900,000 Transition 2
Year 800,000 Transition 3
PHEVs (vehicles)

Fig. 2. Projection of peak and base load demands 700,000

600,000
110,000 500,000
Light-duty vehicles sold (vehicles)

105,000 400,000
100,000 300,000
95,000 200,000
90,000 100,000
85,000 0
80,000
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
75,000
70,000 Year
65,000 Fig. 4. Accumulative numbers of PHEVs in the Ontario’s
60,000 transportation sector
55,000
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025

New peak and base load demands after adding PHEVs into
transportation sector can be calculated by eq. (6) and eq. (7),
Year
respectively. As illustrated in Fig. 5, the peak load demand of
Fig. 3. Projection of light-duty vehicles sold
PHEVs for Transition 3 is the highest since this transition
assumes the greatest amount of PHEVs penetration which is
As shown in Fig. 2, the highest peak and base load
50% of new vehicles in December, 2025. Additional peak load
demands of each year normally occur in January which is
demands in December, 2025 from PHEVs charging for
approximately 26,000 MW and 21,000 MW, respectively. More
Transitions 1, 2 and 3 are 210.3 MW, 630.8 MW and 1,051.3
electricity is required for space heating in the winter, resulting
MW, respectively.
in greater amount of peak and base load demands in January.
IESO also published peak load demand forecast in Ontario
EFFECTS OF CHARGING PATTERN
from 2010 until 2015. Comparing peak load demand from
In the study of charging pattern, three assumptions used in
regression models with IESO forecast, there is a slight
PHEVs charging demand calculations are shown below:
difference (approximately 3%). It might be because projection

7
- Only PHEV-20 penetrates into Ontario’s transportation sector
using Transition 3 (most aggressive penetration level).
- No PHEVs are retired from 2010 to 2025.
- All PHEVs are recharged through the circuit every day.
Results of peak and base load demands for different
charging scenarios after adding PHEVs into the transportation
sector are shown in Fig. 6. It was found that the peak load
demand from charging pattern in Scenario 1, which represents
charging only during the peak period, is the highest among all
scenarios because this scenario has the greatest number of
PHEVs being recharged in the peak period. For Scenario 5, its
peak load demand is similar to the peak load demand when
there is no PHEVs penetration because the number of PHEVs
being recharged in the peak period in Scenario 5 is assumed to
be zero. Additional peak load demands in December 2025 from
PHEVs charging in Scenario 1 to Scenario 5 are 1,051.3 MW,
788.5 MW, 525.7 MW, 262.8 MW and 0 MW, respectively.
COMPARISONS OF WORSE CASE STUDY WITH
ONTARIO’S AVAILABLE RESOURCES
Transition 3 with 50% of PHEVs penetration in December,
2025 and Scenario 1 in which all PHEVs are assumed to be
recharged in peak period are selected as the case study to
compare with Ontario’s available resources. As illustrated in
Fig. 7, in the beginning of 2010 in which there is no PHEVs
penetration into the transportation sector, the supply is less
than the average peak load demand by about 2,855 MW. At the
end of 2025 in which the total number of PHEVs is 890,362
vehicles, peak load demand is greater than the supply by about
6,042 MW. Therefore, it can be concluded that available
resources in Ontario cannot afford the increasing demand from
charging PHEVs between 2010 and 2025. In addition, since
Ontario exports electricity to nearby province and USA, the
increasing amount from PHEVs charging can reduce the
quantity of electricity exported from Ontario.

28,000 27,500
No penetration
27,000 26,500
Transition 1
Peak load demand (MW)

26,000 25,500
Transition 2
24,500
Demand (MW)
25,000 Transition 3
24,000 23,500
23,000 22,500
22,000 21,500
21,000 20,500
19,500
20,000
18,500
19,000
17,500
18,000
17,000 1 2 3 4 5 6 7 8 9 10 11 12
Month
1 2 3 4 5 6 7 8 9 10 11 12 (a) No penetration (Peak) Scenario 1 (Peak)
Month Scenario 2 (Peak) Scenario 3 (Peak)
Fig. 5. Comparisons of peak load demand in different Scenario 4 (Peak) Scenario 5 (Peak)
transition levels in 2025 22,500
22,000
For the base load demand, Scenario 5 in which all PHEVs 21,500
21,000
are recharged during the off-peak period has the highest base
Demand (MW)

20,500
load while base load demand for Scenario 1 in which no 20,000
PHEVs are recharged during the off-peak period is similar to 19,500
the base load demand with no PHEVs penetration. From Fig. 6, 19,000
18,500
the base load demand in all scenarios is not much different. 18,000
Additional base load demands in December, 2025 from PHEVs 17,500
charging in Scenario 1 to Scenario 5 are 0 MW, 20.9 MW, 41.7 17,000
16,500
MW, 62.6 MW and 83.5 MW, respectively.
When comparing between additional peak and base load 1 2 3 4 5 6 7 8 9 10 11 12
Month
demands in all scenarios, it was found that the peak load (b) No penatration (Base) Scenario 1 (Base)
demand has more effect from PHEVs charging than the base Scenario 2 (Base) Scenario 3 (Base)
load demand. The reason is in the equations used for Scenario 4 (Base) Scenario 5 (Base)
calculating peak and base load demands, which are completely
different. Additional peak load demand employs a charger rate Fig. 6. Peak load demands (a) and base load demands (b) in
multiplied by the number of PHEVs, whereas additional base different charging scenarios in 2025
load demand is calculated from averaging the total rated pack
sized of PHEVs per day.

8
 28,000 load demands in December 2025 from PHEVs charging in
Supply
Peak load (case study) Transitions 1, 2 and 3 are 210.3 MW, 630.8 MW, and 1,051.3
26,000 MW, respectively.
Five different scenarios of the charging pattern are
24,000 developed since not all people charge their PHEVs during the
Demand (MW)

off-peak period. By assuming Transition 3 of PHEVs

22,000 penetration, Scenario 1 has the highest peak load demand,
whereas Scenario 5 has the highest base load demand.
20,000 Additional peak load demands in December 2025 from PHEVs
charging in Scenario 1 to Scenario 5 are 1,051.3 MW, 788.5
18,000
MW, 525.7 MW, 262.8 MW, and 0 MW, respectively. Also,
additional base load demand in December, 2025 from PHEVs
16,000
charging in Scenario 1 to Scenario 5 is 0 MW, 20.9 MW, 41.7
MW, 62.6 MW, and 83.5 MW, respectively.
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
Finally, the demands for the worse case, assumed
Month Transition 3 of PHEVs penetration and Scenario 1 of charging
Fig. 7. Comparisons of peak load demand with available pattern, is compared with available resources in Ontario. It was
resources in Ontario found that at the end of 2025 in which the total number of
PHEVs is 890,362 vehicles, supply is less than the peak load
CONCLUSIONS demand by about 6,04278 MW. Therefore, it can be concluded
PHEVs have the potential to penetrate Ontario’s that the energy supply sources will not meet current and future
transportation sector in the near future. They can impact the electricity demand of Ontario generating sector for the worse
electricity grid if a greater number of PHEVs is recharged to case scenario.
the grid at the same period. Since planning and increasing
capacity of existing power plants or even constructing new ACKNOWLEDGMENTS
power plants takes many years to complete, long-term peak and The authors would like to thank CANMET Energy Technology
base load forecast models are developed. Also, long-term Centre, Natural Resource Canada.
forecast of the number of light-duty vehicles sold is necessary
to see how many PHEVs can penetrate the transportation REFERENCES
sector. LR and NLR techniques are employed as forecast [1] Guerfi A., Dontigny M., Charest P., Petitclerc M., Lagacé
techniques to study the relationship between explanatory M., Vijh A. and K. Zaghib, 2010, “Improved electrolytes
variables and forecast variables. Weather, demographics, and for Li-ion batteries: Mixtures of ionic liquid and organic
economic variables are explanatory variables that can impact electrolyte with enhanced safety and electrochemical
forecast models. performance”, Journal of Power Sources, Vol. 195, pp.
It was found that both the temperature and the GDP have 845–852.
greater impact on peak and base load demands in the winter and
the summer. For peak and base loads in the spring and the [2] Amjad S., Neelakrishnan S. and Rudramoorthy R., 2010,
autumn, they depend only on the GDP. Likewise, GDP is the “Review of design considerations and technological
only factor that has an impact on the number of light-duty challenges for successful development and deployment of
vehicles sold. Comparing between LR and NLR by MAE, it plug-in hybrid electric vehicles”, Renewable and
was found that NLR most accurately describes peak and base Sustainable Energy Reviews, Vol. 14, pp. 1104–1110.
load demands in all seasons except for the base load demand in
the winter. Moreover, for light-duty vehicles sold forecast, the [3] Ontario Power Generation, 2009, Biomass Energy
LRM has lower errors than the NLRM. However, the [Online], Available: http://www.opg.com/power/fossil/
differences in the error of both regression models of base load biomass.asp [2010, December 15].
demand in the winter and light-duty vehicles sold are relatively
small. [4] Ontario Power Generation, 2008, 2008 Sustainable
Most accurate regression models which have the lowest Development Report, Ontario, Canada, p. 3.
MAEs are chosen to project peak and base load demands and
the number of light-duty vehicles sold from 2010 to 2025. [5] The California Cars Initiative, How Carmakers are
Three different transitions of PHEVs penetration levels for Responding to the Plug-In Hybrid Opportunity
light-duty vehicles are developed to study the effects of these [Online], Available: http://www.calcars.org/carmakers.
levels on the electricity grid. By assuming that all PHEVs are html [2009, January 9].
recharged during the peak period, Transition 3 whose
penetration of PHEVs into the light-duty vehicles in December
2025 is 50% has the highest peak load demand. Additional peak

9
[6] Mohamed Z. and Bodger P., 2005, “Forecasting electricity
consumption in New Zealand using economic and
demographic variables”, Energy, Vol. 30, pp. 1833–1843.

[7] Pao H. T., 2006, “Comparing linear and nonlinear

forecasts for Taiwan’s electricity consumption”, Energy,
Vol. 31, pp. 2129–2141.

[8] Chui F., Elkamel A., Surit R., Croiset E. and Douglas
P.L., 2009, “Long-term electricity demand forecasting for
power system planning using economic, demographic and
climatic variables”, European J. Industrial Engineering,
Vol. 3, No. 3, pp. 277-304.

[9] Hadley S. W., 2006, Impact of Plug-in Hybrid Vehicles

on the Electric Grid, OAK Ridge National Laboratory,
U.S., pp. 1-18.

[10] Yu X., 2008, “Impact Assessment of PHEV Charge

Profiles on Generation Expansion Using National Energy
Modeling System”, Power and Energy Society General
Meeting - Conversion and Delivery of Electrical
Energy in the 21st Century, U.S., pp. 1-5.

[11] Hajimiragha A., Canizares C. A., Fowler M. W. and

Elkamel A., 2009, “Optimal Transition to Plug-in Hybrid
Electric Vehicles in Ontario-Canada Considering the
Electricity Grid Limitations”, IEEE Transactions on
Industrial Electronics, Special Issue on Plug-in Hybrid
Vehicles, pp. 1-12.

[12] Statistics Canada, 2006, Commuting Distance [Online],

Available: ht tp : // www1 2 .st at ca n. gc .ca /ce n s u s -
recensement/2006/dp-pd/tbt/Rp-eng.cfm?TABID=1&LANG=
E&A=R&APATH=3&DETAIL=0&DIM=0&FL=A&FRE
E=0&GC=35&GID=837983&GK=1&GRP=1&O=D&PI
D=90655&PRID=0&PTYPE=88971,97154&S=0&SHO
WALL=0&SUB=0&Temporal=2006&THEME=76&VID
=0&VNAMEE=&VNAMEF=&D1=0&D2=0&D3=0&D4
=0&D5=0&D6=0#FN2 [2010, January 15].

[13] Electric Power Research Institute (EPRI), 2001,

Comparing the Benefits and Impacts of Hybrid
Electric Vehicle Options, California, U.S., pp. 6-14, 4/9-
4/11, B/2-B/14.

[14] Electric Power Research Institute (EPRI), 2002,

Comparing the Benefits and Impacts of Hybrid
Electric Vehicle Options for Compact Sedan and Sport
Utility Vehicles, California, U.S., pp. 2/25-2/26, A/3-
A/17.

10