Вы находитесь на странице: 1из 12

IET Generation, Transmission & Distribution

Research Article

ISSN 1751-8687
Multistage time-variant electric vehicle load Received on 27th October 2014
Revised on 26th July 2015
modelling for capturing accurate electric Accepted on 10th August 2015
doi: 10.1049/iet-gtd.2014.1019
vehicle behaviour and electric vehicle impact www.ietdl.org

on electricity distribution grids


Ahmed M.A. Haidar 1 ✉, Kashem M. Muttaqi 1, Mohammed H. Haque 2
1
School of Electrical, Computer and Telecommunications Engineering, University of Wollongong, NSW, Australia
2
School of Engineering, University of South Australia, SA, Australia
✉ E-mail: ahaidar@uow.edu.au

Abstract: Accurate electric vehicle (EV) load modelling is a daunting task because the proportion and characteristics of
these extra electrical loads in the grids are continuously changing. In many recent EV studies, EV load is assumed to
be a constant power load to analyse the effect of EV charging on electric network. However, such an assumption
would not be effective in real-time application and can give misleading results. This study proposes a multistage time-
variant EV load modelling technique based on the characteristics of EV load demand due to EV battery charging. To
capture the transitions between different stages of battery charging, the EV loads are modelled with a multistage
scheme and implemented ZIP model. A comparative study of energy losses and voltage profile is carried out on the
IEEE 69-bus test system for the use of a traditional constant load model, conventional ZIP load model and the
proposed multistage time-variant EV load model. Results show that the load models can significantly affect the energy
losses and terminal voltages, and provide misrepresentation if EV load is not accurately modelled. Therefore, for
proper analysis of EV loads during unidirectional charging, it is essential to have accurate models for predicting true
EV behaviour.

1 Introduction vehicle (PEV) penetration on distribution systems. The authors in


[4] applied large-scale distribution planning model to build the
There are potential issues that arise from large-scale deployment of optimal base-case network to supply the load demand and then
electric vehicles (EVs) in Australia and in other developed used the base-case network as a reference to assess the impact of
countries around the world which could bring major changes to PEV penetration for other cases using the constant load model. In
power system operation and power plant investment. Traditional [5], the potential impacts of PEV have been tested on the Western
power grids were designed over half a century ago and are not Australian electricity supply system using a constant load model.
suited to match the high penetration of this additional load with The conventional constant power model has been used for
atypical characteristics. The EVs could therefore have a disruptive modelling the impact of electric vehicle charging on distribution
impact on distribution systems if not integrated carefully, and the system components in [7–9]. Studies on charging control and
severity of this impact depends on their penetration into the coordination of (PEV) charging have been carried out in [11–13]
networks. Hence, there is a need to know whether the current grid in the context of uncertainty of the EV load behaviour.
meets the energy requirements in order to evaluate the grid Most of the above-mentioned investigations of EV impacts on
capability of handling a massive EV penetration [1]. The stable distribution systems have utilised a constant load model to
and reliable operation of a power grid is dependent on the system represent EV in the power flow analysis, for which load power
ability to match the electrical output of the generation to the was considered to be independent of variations in feeder voltage.
electrical loads of the system. The load behaviour is, therefore, The conventional constant power model might not be able to
very important and plays an important role in the system stability precisely represent the true behaviour of the PEVs. Such simple
and control. Furthermore, the realistic system behaviour can only models ignore the transition between different operational stages
be obtained using a proper load model that is required to minimise during EV charging; as each stage has its unique characteristics.
the operating cost while maintaining the system conditions with Only a few studies [14, 15] have considered the changes of
high security and reliability. However, load modelling is still charging rate with battery state of charge (SOC). Although in [15]
considered to be a complex problem due to its peculiarity while the ZIP model of PEV has been utilised, this model is simply
integrated into the grids [2]. focusing on individual charging control and there is a shortage of
The issue of EV charging affecting a power network is a a comprehensive investigation of the model. From the literature, it
commonly identified topical research issue. As it was shown in is clear that EV load modelling has not received much attention yet.
many studies [1, 3–10] that EVs are going to present a very In this paper, a multistage time-variant model has been developed
challenging load problem for electrical grids when their based on the ZIP modelling approach for the PEVs, which are
penetration increases, strong evidence was presented in [11–14] to connected to the distribution networks. Additionally, battery
confirm that sophisticated centralised control algorithms are the charging stages of the PEVs have been considered in the proposed
best management solution. However, this solution can be achieved modelling. The proposed model has been applied to determine the
by utilising an accurate modelling of the EV load during the accurate impacts of PEV on voltage profile and system losses. The
charging cycle to manage the problem. An approach based on the rest of the paper is structured as follows. In Section 2, some
use of a large-scale distribution planning model was proposed in important facts to be considered during the modelling of EV load
[4] for identifying the impacts of different levels of plug-in electric integrated to distribution systems are discussed. The proposed

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 16, pp. 2705–2716
& The Institution of Engineering and Technology 2015 2705

Authorized licensed use limited to: Universidad Tecnologica de Pereira. Downloaded on February 24,2020 at 04:24:09 UTC from IEEE Xplore. Restrictions apply.
modelling technique and its implementation for distribution system It is important to mention that the voltage dependency of the
analysis are outlined in Section 3. Section 4 describes the test charger and the existence of exponential load models having wider
system and covers the approach of evaluating the proposed range of ZIP coefficients of battery charger load have been already
multistage time-variant EV ZIP load model. Section 5 discusses verified through laboratory testing by the authors in [23]. This
the results, while Section 6 provides some concluding remarks indicates the importance of using a multistage time-variant ZIP
based on this study. model as this model reflects the true characteristics of the charging
loads all the time and gives the accurate response for the load
flows, controller design, control settings and protection systems.
2 Problem statement As a result, the use of a ZIP model in its current form is not
sufficient to simulate a system containing a large number of EVs.
In the aforementioned studies, loads are modelled as constant powers There is a necessity to develop a technique to capture and
(active and reactive) and updated through a load flow for each time characterise the dynamics of the load demand due to EV battery
interval to observe a daily load profile. However, the load voltage charging. With this aim, this work is focusing on developing a
dependency has an important impact on the load power modelling approach for EV load in a multistage time-variant ZIP
consumption [16]. Therefore, voltage dependency of the actual model to study the true effect of EV integration. In this approach,
loads needs to be properly modelled. In fact, load in a load bus is the proposed EV load in a ZIP model utilises three different stages
a composite model (i.e. mix model) such as constant impedance, in progression with time to represent the power consumption by
constant current (CC) and constant power loads. In a time-variant EV; and each stage is governed by EV ZIP model which is
load representation, the coefficients of the mix model remain evaluated under various scenarios.
constant; the power consumption at the load bus varies with the
actual terminal voltage. In the traditional load composite model,
the model uses the same ZIP values for every load for given load 3 Proposed multistage ZIP model for EV load
classes (such as residential, commercial and industrial) throughout
the day in a daily load [14] profile. Since the load composition Since EV charging parks will be integrated in the existing
and the amount of load for each load class may change from one distribution systems in future, accurate load model that reliably
time instant to the next, the constant values of the ZIP model reflects underlying phenomena of the physical loads give better
parameters will not be representing the real behaviour of the load tuning of the control operation. This section will discuss in details
in a bus during the coincidental charging activities of EV that how to model multistage time-variant EV load and will also
occur at the distribution system. The energy consumption by an present how to implement this in terms of ZIP model for dynamic
EV is a function not only of the terminal voltage but also of other power flows that will capture the accurate EV characteristic for
variables defined by different ranges in battery SOC due to the grid-to-vehicle mode of operation. It is to be noted that EV with
changes in charging state and rate. Furthermore, voltage slow charging is addressed in this paper. EV as vehicle-to-grid and
dependency of the charging is a function of SOC and this can EV with fast charging are out of the scope of this paper.
cause different load characteristics for different SOC levels [17–
19]. Rate of battery charging mostly depends on the type of
battery. The real charging characteristic curves of a battery are 3.1 Polynomial model (PM)
depicted in Fig. 1 [17, 20]. As seen from this figure, current drawn
by the battery is constant at the beginning, and after reaching a The combination of constant current, constant power and constant
certain level of SOC, the voltage becomes constant but the current impedance loads is called the polynomial model. Accordingly, the
drawn gradually reduces and it reduces to a very small value polynomial model is an expansion of the exponential model [23–
when the battery reaches close to its maximum charge. Although 25], which has been widely used to represent the voltage
the charging profiles in this figure are obtained experimentally, the dependency of loads. It is also known as ZIP model and is
battery chemistries and the battery management system of the represented in the following form
various PEVs are different. Thus, the same profiles are not always
valid because different charging strategies can be applied
depending on the battery type [21, 22] and this will affect the PZIP = Po (ZpV 2 + IpV + Pp) (1)
demand profile on the grid side.
QZIP = Qo (ZqV 2 + IqV + Pq) (2)

where Po and Qo are the active and reactive load powers at the
nominal voltage (of 1.0 p.u.); V is the actual voltage; p and q are
the fractions; and Z, I, P are the constants (ZIP coefficients)
associated with impedance, current and power of active loads at
particular bus, respectively. The values of the ZIP coefficients
(ZIP parameters) depend on the load composition and hence the
system under study. The parameters of commonly used types of
loads are calculated through the measurement [23, 26]. To include
the effects of EV loads considering the percentage of each load
category in a bus and the composition of each load model in a
load category of the EV load behaviour in a ZIP model combined
with the base load at each bus, (3) and (4) of aggregate ZIP model
are the modification of (1) and (2)

PAg = Po (ZpAg V 2 + IpAg V + PpAg ) (3)

QAg = Qo (ZqAg V 2 + IqAg V + PqAg ) (4)

where ZpAg, IpAg and PpAg are the constants of the aggregate model
Fig. 1 Typical charging characteristic of a lithium battery [17, 20] and can be found by (5), and similarly, ZqAg, IqAg and PqAg can be

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 16, pp. 2705–2716
2706 & The Institution of Engineering and Technology 2015
Authorized licensed use limited to: Universidad Tecnologica de Pereira. Downloaded on February 24,2020 at 04:24:09 UTC from IEEE Xplore. Restrictions apply.
obtained Equation (7) indicates that the power on the DC side is load
dependent, as the battery is interfaced with the EV charger.
⎧ ⎫
⎨ ZpAg = Zp + ZpEV ⎬ Looking at the charger interface battery equivalent circuit in
Ip = Ip + IpEV (5) Fig. 2, it is evident that the change in SOC will change both the
⎩ PpAg = Pp + Pp ⎭ magnitude and angle of the equivalent impedance seen at the
Ag EV
charger terminals. Generally, the governing equation of terminal
voltage for the charger interfaced battery model, shown in the
EV active ZIP load can also be written from (3) and (5) as in (6), and circuit of Fig. 2, can be written as
similarly EV reactive ZIP load power can also be obtained
dIA
PEV = PEVo (Z pEV V 2 + I pEV V + P pEV ) (6) VA = RIA + L + VB (10)
dtL

3.2 Charger interfaced battery system where


A simple charger interface battery model behaviour is shown in R = RA + RD + RB


(11)
Fig. 2. The network behind the load is represented by an L = (LA + LD )
equivalent Thevenin voltage. The power balancing equation of the
battery model is given by As seen from the above equations, the system voltage dependency of
the charging is a function of VB and hence SOC. This system is not
PD = ID (VB − RB ID ) (7) linear and can cause different load characteristics for different SOC
levels. The dynamic equations characterising the conduction modes
where VD and ID are the voltage and current on the battery side, of EV charger in Fig. 2 are presented in detail in [28], and hence, the
respectively; RB is the battery internal resistance; and VB is the modelling is simplified here.
battery open-circuit voltage. The methodologies used to model Considering the efficiency of the charger ‘η’, the power on the DC
battery and formulate the open-circuit voltage of the battery are side of the battery charger at different levels of charging, can be
well documented [18, 19, 27], and hence not shown here. Battery expressed as
SOC is the gauge that is used to understand the amount of charge.
It is the remaining capacity in the battery divided by the rated
PD (S) = h VA (S)IA (S) cos w (12)
capacity of the battery. The open-circuit voltage and battery
internal resistance are functions of battery SOC. For the ampere
hour (Ah) capacity, the change can be expressed as Once the equivalent load impedance of charger interfaced battery, as
given in (11), and the Thevenin equivalent are known, the operating
t1 point at each point of the network ‘PV curve’ [30] for EV load
1 impedance ‘ZEVload’ is calculated using the following equations
DSOC = ID (t) dt (8)
Ah capacity t0
I A = V th /(Z th + Z EVload ) (13)

During charging process (t0 − t1), the converter acts as a battery V A = V th − I A Z th (14)
charger. In this situation, it consumes power from the grid.
Referring to Fig. 1, as the voltage is a function of the SOC at the Finally, the power for the equivalent load containing EV battery
DC side of a battery charger in the CC region, the power on the which is SOC dependent and having non-constant power factor is
grid side (AC) can be calculated as follows analytically obtained as

PA (S) = VA (S)IA (S) cos w (9) ∗


SA (S) = I A (S) · V A (S) = PA (S) + jQA (S) (15)

where VA and IA are the voltage and current on the AC grid side;
cosj is the power factor; and S is the state of charge (SOC). In the 3.3 Multistage EV load model
constant voltage (CV) region, the voltage at the DC side is
constantly maintained until 100% SOC level is reached and the The characteristics of energy consumption during charging activities
float charge compensates for the loss and is held at CV. of PEVs can be changed when charging is progressed with time. The
The typical equivalent circuit of single-phase battery charger as charging level used by EVs at car park is assumed to be level 1
shown in Fig. 2 can operate in different modes [27, 28] which which is a slow charging [27, 29]. It is similar to home charging
have been widely studied and modelled in the literature [27–29]. as used by EV owners at homes. Generally, PEV battery capacities

Fig. 2 Equivalent circuit of charger interface battery

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 16, pp. 2705–2716
& The Institution of Engineering and Technology 2015 2707

Authorized licensed use limited to: Universidad Tecnologica de Pereira. Downloaded on February 24,2020 at 04:24:09 UTC from IEEE Xplore. Restrictions apply.
Fig. 3 SOC and AC power demand profile of a typical EV battery [21, 22]

range from a few kWh to over 50 kWh. As the battery capacities are
of importance to determine reasonable charging profiles, in this Fig. 4 EV charging demand profile and DC voltage with respect to SOC
study, an 18 kWh battery capacity per PEV is selected because it
is expected that the lower end of battery sizes is more affordable
and more likely to initially dominate the Australian market. The least-squares fit, the ZIP parameters can be fitted to the measured
variations of AC power demand for typical lithium and lead-acid and simulated data with the constraints. The fitting process can be
EV batteries are shown in Fig. 3 [21, 22]. In fact, different factors formulated in terms of a simple convex quadratic optimisation
influence these characteristics, such as the type of EV batteries and problem [31]. Here, the fitting objective is to minimise the
charging level. The behaviour of these characteristics can be difference between the fitted approximation and the measured data
represented in terms of different charging stages and each stage for a finite number of voltage and SOC points within a range of
operation has a different set of characteristics. The variation of AC operating voltages. The problem can be written as
power as shown in Fig. 3 indicates that the AC power demand for
charging of the EV has decreased after 60% SOC (i.e. at 2.5 h)
nv   2
and in the second stage it is substantially decreasing until 90% Minimise ZpstEV Vi2 + IpstEV Vi + PpstEV − PEV
st
(Vi ) ,
i=1 (17)
SOC (i.e. from 2.5 to 4.75 h), and then it is dropped to almost
zero after 95% SOC (e.g. after 4.75 h) when the EV battery is st = 1, 2, 3
almost fully charged. Although the SOC peak is reached faster
with a high level of power, the topping charge will still be taking Subject to
a reasonably longer time. It is crucial to evaluate the EV load
demand with respect to the time instant of the relevant charging
ZpstEV + IpstEV + PpstEV = 1 (18)
stage. Due to three levels of load demand for EV charging as
discussed above, three-stage model (st = 1, 2, 3) is considered in
the development of the proposed modelling technique. where nv is the number of points selected within the voltage range
The proposed multistage EV ZIP is an alternative to the EV load for VA(S); Vi and P(Vi) are the ith pair of measured voltage and
model in (6) that enables a formulation of the EV behaviour on each active power; ZpstEV , IpstEV and PpstEV are the unknown variables to
charging stage. This model is a fitted stage ZIP model of a measured be calculated to find the optimum solution. Note that PEVo st
(16) is
[21, 22] and simulated charging profiles as given in (16) the active EV power at the nominal voltage of 1.0 p.u.. The data
for the synthesis in (17) can be obtained directly from load
measurements or previous synthesis by (11)–(15). The multistage
st
PEV = PEVo
st
(ZpstEV V 2 + IpstEV V + PpstEV ), st = 1, . . . , n (16) EV ZIP obtained by fitting a curve to simulated and experimental
measurement for EV charging is depicted in Fig. 4. The range of
st each stage can be fixed as in Table 1 and Fig. 4. Due to the lack
PEV is the power demand at each stage (st is stage) during the battery of specific experimental measurement data used to obtain the ZIP
charging activity. Constants ZpstEV , IpstEV , PpstEV are the parameters of values for different batteries with different capacities and
ZIP for load demand of EV on the first, second and third stages,
respectively, and calculated by curve fitting procedure. In a
real-time application, the values of the constant impedance and Table 1 Decision rules for EV multistage transition
current are recalculated as the load demand changes. The
presented model as given in (16) is an example to demonstrate the Charging time ZIP model ZIP parameter
concept of the proposed multistage EV load model. It is a general representation values
model and can be applied to batteries used in EVs under slow st1
ZpEV = −0.00374
charging. However, for different types of batteries (charging
st1
levels), the parameter values of the ZIP model may be slightly 0 < t < tstage1(e.g. <2.5 h) stage 1 IpEV = 0.3111
different and need to be adjusted accordingly. The proposed st1
PpEV = 0.6926
synthesis of (16) is the same type as (6), but with multistage. The st2
ZpEV = −0.00221
polynomial coefficients are to be determined. The multistage
st2
model approximates the load voltage dependency using impedance tstage1 < t < tstage2 (e.g. above stage 2 IpEV = 0.2322
2.5 h and below 4.75 h)
‘Z’ that represents the quadratic voltage dependency at each stage, st2
PpEV = 0.77
a current source ‘I’ that represents the linear voltage dependency at st3
ZpEV = −0.00165
each stage and power ‘P’ that represents constant power at each
st3
stage. Since the EV power demand is a function of SOC as seen tstage2 < t < tstage3 (e.g. above stage 3 IpEV = 0.1638
4.75 h)
from (9) and Fig. 2, this shows that the SOC will react to any st3
PpEV = 0.8379
sudden change in the input voltage. By conducting constrained

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 16, pp. 2705–2716
2708 & The Institution of Engineering and Technology 2015
Authorized licensed use limited to: Universidad Tecnologica de Pereira. Downloaded on February 24,2020 at 04:24:09 UTC from IEEE Xplore. Restrictions apply.
technologies, the presented ZIP values in Table 1 can be used for the loads are updated in each iteration of the power flow. Considering
EVs with different batteries as long as they are under slow charging. (19) that each bus load is an aggregation of different types of
Since the power on the grid is a function of the SOC as given in loads, it can be written as
(9), the corresponding charging load at each stage is, therefore, can
be derived for the EV load using multistage EV ZIP model given by PAg − fP (V , d) = 0 (23)
(16). EV load is modelled according to the actual EV power
consumption at any time instant, ‘t’ in a charging stage, ‘st’. The
range of each stage can be fixed as in Fig. 4. The transition of where PAg is the aggregate ZIP model as given in (3). In the case of
each stage can be determined by external signals in the case of the proposed multistage model, the coefficients of the aggregated
physical model [32]. In the multistage time-variant EV load in a model defined in (5) can be written as
ZIP model, the borderlines between stages are defined carefully.
As seen in Fig. 4, charging time for stage 1, tst1 is t0 to t1; ⎧ st ⎫
charging time for stage 2, tst2 is t1 to t2; and charging time for ⎨ ZpAg = Zp + ZpEV ⎪
⎪ ⎬
stage 3, tst3 is t3 to tf. In the case of battery under slow charging IpAg = Ip + IpstEV (24)

⎩ ⎪

(as shown in Fig. 4 and Table 1), we have considered the PpAg = Pp + PpstEV
borderlines for the three stages as t0 = 0, t1 = 2.5 h, t2 = 4.75 h and
tf = 8.5 h. These stages reflect the physical analysis presented in [1,
21, 22] considering the change in EV power demand as a function
Referring to Fig. 5, the active power flows through branches i and
of time. Given the percentage of EV load in a network bus, the
i + 1 can be written for constant power load as
proposed model can be used in the aggregate ZIP model expressed
in (5) for a given system voltage and SOC condition and also can
be extended for time-series analysis. F
Pi,i+1 = Pi,l
F
+ Pi+1 + Pi,i+1
Lo
(25)

3.4 System modelling where

The structure of distribution systems is generally radial [11, 33–35].   2


Lo
Pi,i+1 = Ri,i+1 Vi − Vi+1 Yi,i+1  (26)
The schematic of the distribution system with n number of buses is
depicted in Fig. 5. The bus voltages are found by solving the
Lo
mismatch equations (19) and (20). These equations are the general Pi,i+1 , Ri,i+1 and Yi,i+1 are the active power loss, resistance and
form of Newton–Raphson power flow equations. Details of the admittance of the line section between buses i and i + 1. Using
method are given in a number of standard textbooks (26) for time interval Δt, usually, the total active losses of
distribution system can be computed by (27) and similarly for
P − fP (V , d) = 0 (19) reactive power

Q − fQ (V , d) = 0 (20)
n−1
ToLo
PDt = Lo
PDt,(i,i+1) (27)
i=0
where

i+1 
  To simplify the problem, let the power flow at the end sending
    branches i, i + 1 be expressed as
fP (V , d) = Y ji Vj  Vi cos (di − dj − u ji ),
j=i−1 (21)
i = 1, 2, . . . , n
F
Pi,l = Pi,l+1
F
− Pi,i+1
Lo
− Pi+1 (28)

i+1 
 
    So far, many studies in the literature have assumed that the load Pi+1
fQ (V , d) = Y ji Vj  Vi sin (di − dj − u ji ), (base load combined with EV load), which is given in (28), is
j=i−1 (22) independent of the bus voltage. However, (28) can be extended to
i = 1, 2, . . . , n include the voltage dependence as an aggregated model by
substituting (3) into (29) as
P and Q are the total active and reactive power loads at each bus F,Ag F,Ag Lo,Ag Ag
which could be constant or a function of the bus voltage Pi,l = Pi,l+1 − Pi,i+1 − Pi+1 (29)
magnitude as given in (1)–(4); Vj and Vi are the voltages at buses j
and i with the corresponding phase angles δj and δi; Yji and θji are Lo,Ag
where Pi,i+1 is the total active loss of distribution system with the
the magnitude and angle of the branch ‘ji’ entry in the Y-bus aggregate ZIP model for each updated time interval Δt, which is
matrix [11, 34, 35]. The solution to the mismatch equations is obtained by including voltage dependence into (27) to get (30);
obtained by using the Newton–Raphson power flow method which F,Ag
Pi,l+1 is the power flow between buses i and i + 1 with the
is modified in order to take the voltage dependency of active and aggregate ZIP model
reactive power of loads into account. Thus, the active and reactive

ToLo,Ag

n−1
Lo,Ag
PDt = PDt,(i,i+1) (30)
i=0

As indicated earlier that when using the aggregate ZIP model


without multistage as in (6), the ZIP parameters are found by (5).
However, when using the aggregate ZIP model with multistage,
the ZIP parameters of multistage are found by (24). Thus, the
multistage ZIP model is employed with respect to the time for
different stages as proposed in (16) and implemented in the
iteration process of the dynamic power flows as demonstrated in
Fig. 5 Schematic of a radial distribution system with ‘n’ number of buses the algorithm as shown in Fig. 6. Please note that the obtained

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 16, pp. 2705–2716
& The Institution of Engineering and Technology 2015 2709

Authorized licensed use limited to: Universidad Tecnologica de Pereira. Downloaded on February 24,2020 at 04:24:09 UTC from IEEE Xplore. Restrictions apply.
Fig. 7 Probability distribution function of a daily driving distance

shown in Fig. 7 for initial state of charge


⎧ 
⎨ Erat − (D + Dem ) × Ekm
× 100, 0 , D ≤ Dmax
SOCin (%) = Erat

0, D . Dmax
(32)

where Erat is the rated energy of the battery (kWh); Dem is any
travelling distance for emergency case, using the information of
EV travelling a day, this value can be in the range of 5–10 km;
Dmax is the maximum travelling distance with a full state of
charge; and Ekm is the energy consumed from the battery per
kWh/km. Normally, the energy consumed from the battery for a
100 km trip is equal to 13 kWh (Erem = 0.13) which means that
∼90% of the EV battery capacity is only used.

4 Test system description and simulations

The IEEE 69-bus test system has been selected as a case study to
evaluate the effect of load modelling due to the integration of
EVs. This system is a 12.66 kV radial distribution system with 69
buses and 7 laterals, as shown in Fig. 8. It is a long radial system
Fig. 6 Algorithm of multistage EV ZIP model incorporated in the power with 47 load points with total load of 3.8 MW and 2.69 Mvar
flow iteration [34]. Three sectionalising switches have been added into the
standard system; these switches are operating as normally open.
The test system is employed with the base data assuming the same
results in Fig. 11 are directly related to (30), which is considered as a load variation across 24 h period similar to the real historical data
numerical example to clarify the model described by expressions of a daily load profile in the state of New South Wales (NSW) in
(27)–(30). Australia, as in Fig. 9. For the demonstration of the effectiveness
of the proposed modelling technique, time-series simulations were
carried out for every Δt. It is expected that the high EV
penetration level occurs during the working day as reported in
3.5 Initial state of charge [1, 5]. The EV penetration level refers to the ratio of EV load
demand that needs to be charged from the distribution grid system
To determine the initial state of charge [(SOCin)] of each EV before to the maximum level of EV load demand in the same grid
the next charging, it is important to know how deep EV was system. It is assumed that each bus of the network will have one
discharged during the day before next charging. On the basis of car park with 60 car spaces.
the driving habit statistic report in [1], it was found that the dally Time-variant ZIP models were included in the simulation for each
driving distance follows the lognormal distribution as illustrated in load class. First, the load supplied at the delivery point is categorised
Fig. 7 and is expressed by into load classes such as residential, commercial, industrial and EV
loads. Second, each category of load class is represented in terms of
  load components. Given the percentage of each load category in a
1
√ e −(In D−m) /2s ,
2 2
f ( D) = D.0 (31) bus and the composition of each load in a load category, the
Ds 2p equations given in Section 3 were used to compute the aggregate
ZIP model with relevant ZIP coefficients for the loads in each load
where D is the daily distance travelled in km which is a random bus. The percentage of each load class to the total load vary from
variable; μ is the mean and σ is the standard deviation of the bus to bus in the system and is dependent on weather and time
probability function [36]. After defining the random travelling during the day and season, according to the consumption patterns
distance, the initial state of charge can be calculated by (32) as and the underlying composition of each class. Therefore, the

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 16, pp. 2705–2716
2710 & The Institution of Engineering and Technology 2015
Authorized licensed use limited to: Universidad Tecnologica de Pereira. Downloaded on February 24,2020 at 04:24:09 UTC from IEEE Xplore. Restrictions apply.
Fig. 8 IEEE 69-bus distribution system with EV charging park

ToLo
relevant percentage of load composition for these models at each bus where PDt,Ag is the total added to the branch network losses due to
excluding the EV load model is hypothetically generated by referring EV penetration as given in (30) which can be the EV ZIP model
ToLo
to a typical daily load in the state of NSW in Australia. This with multistage or without multistage; PDt,ZIP is the network loss
simplification is made because the main focus in this work is to without EVs; NL is the number of lines in the system; NP is the
determine the accurate impact of EV load model on the number of periods in the load profile; and ΔtN is the number of
distribution grid system. charging periods. The lower the value of the index, the better the
In the distribution system, network losses and bus voltages largely benefits in terms of network loss reduction with respect to the load
depend on the load model and load profile [33, 37]. To measure the of the load model.
severity of impacts on the system losses and bus voltages for the ZIP
model due to EV penetration during a period of time Δt, the
following indices will be used. 4.2 Voltage profile index (VPI)

The voltage profile index (VPI) for the ZIP model is related to the
4.1 Energy loss index (ELI) voltage deviation between each bus ‘Vj’ and the root bus ‘V0’
considering the time-varying voltage magnitude and load demand
The ELI for the ZIP model is defined as at each time interval as given in (34), where NB is the number of
the buses in the system. The lower the value of the VPI, the better
NP the network performance is
Dt=1
ToLo
PDt,Ag − PDt,ZP
ToLo
ELI = 100 × × DtN (33) NP NB  
NL NP 
Dt j V0 − Vj 
VPI = 100 × (34)
NB NP

5 Results and discussion

To evaluate the effects of the ZIP load model compared with the
constant power model on the power flow results, initially load flow
solutions for the distribution system were obtained using the
nominal values of the system components. In solving the power
flow problem, the source voltage (root bus) was assumed to be
1.0 p.u. with a tolerance of 0.0001 p.u.. Fig. 10 depicts the voltage
magnitudes obtained using the constant power model and the ZIP
load model. The lowest voltage was observed on bus 65,
additionally, among the various load models, the constant power
provides the lowest voltage whereas ZIP model gives the higher
voltage due to the fact that the ZIP model represents lower loads
than the constant power load model. It also depends on the
composition of ZIP model (i.e. % of various components).
Compared with the constant power load model, the load demand
of the system for the ZIP model is decreased.
Fig. 9 Actual recorded composite load profiles during a typical day in To examine the impacts of the proposed time-variant ZIP load
summer models with and without multistage, the time-invariant constant

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 16, pp. 2705–2716
& The Institution of Engineering and Technology 2015 2711

Authorized licensed use limited to: Universidad Tecnologica de Pereira. Downloaded on February 24,2020 at 04:24:09 UTC from IEEE Xplore. Restrictions apply.
Fig. 11 Network loss of a ZIP load model with and without multistage

becomes large enough to observe significant impacts of the EV


Fig. 10 Voltage of some monitored buses for various load models load. The results for time-variant EV load in a ZIP model with
and without multistage are illustrated in Table 2 and graphically
shown in Figs. 11–16. As seen from Table 2, in all cases, the
power loads associated with the 69-bus test system were replaced energy loss index is lower in the multistage ZIP model than the
with the time-variant ZIP model and multistage ZIP model ZIP model because the power demand of EVs has decreased with
representing 48 EVs charging in each car park. The proposed time at three stages during the charging interval, whereas in the
indices were used to assess the performance of the developed ZIP ZIP model (without multistage), the power demand of EVs has
model implemented in the dynamic power flow program under decreased with time at only one stage. Thus, increasing the
different EV penetration levels. In this paper, the voltage deviation number of EVs loads leads to significant increase in power losses
has been used that can be defined as the difference between the when ZIP load model is used without considering multistage. It is
nominal and the actual voltages. Six cases were selected for each worth to mention that energy loss and voltage profile indices are
load model depending on the degree of EV penetration. The first the highest while charging during the peak time of commercial
case with no EV is taken as a reference case, while the next cases and industrial load demands because public charging during the
have an EV penetration of 5, 10, 15, 20 and 25% with respect to day is more demanding compared with the charging during
the maximum number of car spaces available in the car park. The overnight; these results are directly related to Fig. 9. From this
more centralised the charging time is, the more the total charging finding, it is revealed that the smaller the energy loss and voltage
power is required and thus a large impact on the power grid. profile indices, the better the voltage condition of the system.
Therefore, it is especially required to investigate the impacts to As seen from Fig. 11, due to charging with 5%, the comparison
distribution system when a large number of EVs are plugged in between the two models is showing that by using the multistate
for charging. Since the charging parks are located in public areas, ZIP model, the effects of charging profile can be captured clearly,
it is assumed that the charging in industrial and commercial areas which cannot be possible when only ZIP model without multistage
will occur at the time of EV owners arriving at the work places. is used. Fig. 12 shows the load demand profile for 15% EV
Thus the worst-case scenario is where the EVs are equally placed penetration; from this figure, the multistage ZIP model is very
in all charging parks and the most EVs are plugged in and start effective to reflect the behaviour of EV load demand at the
charging around 9 am when people usually arrived to their work. charging interval whereas this behaviour cannot be observed in the
The charging time interval is used as 9 am to 3 pm, and the ZIP model. As mentioned above, the worst-case scenario is
vehicles become fully charged at the end of the charging interval. considered where all EVs start charging at the same time (9 am)
The EV starts charging with an initial SOC as calculated in and continue charging until they become fully charged at 3 pm.
Section 3.5 [36, 38]. At every half an hour (similar to the interval
used in [14]), the power flow method embedded with the load
model has been run iteratively to determine the accurate voltage at
each bus until the convergence is achieved.
The impact of charging on the distribution system is demonstrated
by computing the above-mentioned indices related to voltage
deviations and the power loss for each charging period. The
consumed power by EV under 25% EV penetration in the system

Table 2 Energy loss index and VPI during charging interval (9 am–3
pm)

Penetration Energy loss index, % Voltage profile index, %


level, %
Zip load Zip load Zip load Zip load
model with model model with model
multistage multistage

5 0.157661 0.247291 2.667722 2.707974


10 0.347538 0.554597 2.743644 2.826945
15 0.568526 0.921306 2.822206 2.950945
20 0.822815 1.351476 2.903424 3.079835
25 1.112228 1.846129 2.987312 3.212530
Fig. 12 Load demand profile for 15% penetration

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 16, pp. 2705–2716
2712 & The Institution of Engineering and Technology 2015
Authorized licensed use limited to: Universidad Tecnologica de Pereira. Downloaded on February 24,2020 at 04:24:09 UTC from IEEE Xplore. Restrictions apply.
Fig. 15 Voltage deviation in a multistage ZIP model with 25% PEVs
Fig. 13 Load demand variation between ZIP load model with and without
compared with the ZIP model
multistage

Note that the end of charging process as seen in Fig. 12 coincides coincident peak that are affected by charging process of EV
with the end of peak demand, where the maximum hill in the battery at different stages.
demand profile of the system is reduced which is directly related Fig. 16 shows the voltage profile for penetration level of 10% in
to Figs. 9 and 11. two selected buses (which are bus 27 and bus 65). Clearly, there is
Network loss variation can be defined as the ratio of the difference a large voltage drop in the presence of EV for the ZIP load model
between total power losses of the load with EV represented with ZIP without multistage. On the contrary, the better voltage profile of
model and the total power losses of the base-case load without multistage ZIP model during EV charging is caused by different
additional EV represented with ZIP model to the total power charging stages which are employed in this model. Moreover, the
losses of the base-case load without additional EV represented portion of voltage that is independent of constant power gradually
with ZIP model. Load demand variation is also defined in a decreases in the second and third stages of this model.
similar manner. Figs. 13 and 14 depict the variation in total load For better representation of the multistage ZIP model, the
demand and system losses. The variation is higher when PEV correlation between the difference in voltage deviation and the
charging activities are high, especially during morning peak. It is difference in the power losses using ZIP load models with and
also shown in Fig. 14 that for a penetration of 5%, the network without multistage are quantified. From Fig. 17, it is seen that this
loss variation is close to 0.33 MW as seen from Fig. 11 which is relation is linear, the difference in energy losses between the ZIP
about 0.157661% as measured by ELI for the multistage ZIP model with and without multistage is increasing with an increase
model and 0.247291% for the ZIP model. Although the power in voltage deviation. ‘1%’ variation in voltage results in about
loss is high, the VPI for the EV penetration of 5% is acceptable. ‘1.7%’ variation in system loss.
Fig. 15 shows the voltage deviation for ZIP models with 25% of For further evaluation of ZIP load model, the impacts of EV
EV penetration during the charging interval 9 am–3 pm. From this charging loads on the power congestion level have been measured
figure, it is found that the smaller the deviation of bus voltage by implementing the ZIP model with and without multistage.
from the nominal voltage, the better the voltage condition of the Power congestion level is the ratio of power flow to the total flow
system. It is seen that ZIP load model has a very smooth profile, of power demand during a day. Table 3 has presented the power
while multistage model has a much more irregular profile and congestion level of some selected lines based on their location in
the test system. Normally, the loaded branches are expected to

Fig. 14 Network loss variation between ZIP load model with and without Fig. 16 Voltage profile in multistage ZIP model with 10% PEVs compared
multistage with the ZIP model

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 16, pp. 2705–2716
& The Institution of Engineering and Technology 2015 2713

Authorized licensed use limited to: Universidad Tecnologica de Pereira. Downloaded on February 24,2020 at 04:24:09 UTC from IEEE Xplore. Restrictions apply.
Table 4 Total network losses with different value of ZIP parameters for
25% EV penetration

Cases Stages Constant Constant Constant Total network


impedance current current loss (MW)
during PEV
charging

1 1 −0.00374 0.3111 0.6926 6.634041


2 0.002211 0.2322 0.7700
3 −0.00165 0.1638 0.8379
2 1 −0.1824 0.9949 −0.1773 6.50373
2 0.3240 0.8482 −0.1722
3 −0.00565 0.6638 0.3419
3 1 −0.00174 0.4111 0.5906 6.589255
2 −0.02211 0.3521 0.6700
3 0.2824 0.8949 −0.1773

Fig. 17 Correlation of voltage and loss variations between ZIP models with
and without multistage

Table 3 Total power congestion level of selected lines with 15% PEVs
Line number Multistage ZIP model ZIP model

L12 (12–13) 26.554631 35.953817


L27 (3–28) 38.886466 49.578219
L35 (3–36) 33.383367 43.645546
L52 (9–53) 5.2293392 7.8683221

reach high-power congestion level. As seen from Table 3, the most


stress is observed in the power congestion level of the ZIP load
model. The same can be observed from Fig. 18, it is seen that the Fig. 19 Network losses for 25% EV penetration with different value of ZIP
flow of power in the ZIP model during a day is higher than the parameters as in Table 4
flow of power in the multistage ZIP model.
Since the EV charging load is incredibly growing in the whole
system, it is necessary to demonstrate in the simulation process the the ZIP values of Table 1 are used in the first case, while the ZIP
effect of different parameters of the multistage ZIP model on the values used in cases 2 and 3 were hypothetically generated, which
network losses. For this investigation, the standard system was are within the limits to meet the constrained as in (18). It is
tested with different ZIP values as shown in Table 4. There are observed from these results that the system behaviour can be
three cases that have been considered with different ZIP values for slightly changed when the ZIP values at different stages are
each stage. The obtained results are graphically shown in Fig. 19 different due to the difference in the battery capacities and
and demonstrated in the last column of Table 4. Please note that technologies and their charging systems.
To assess the sensitivity of each ZIP load model, the topological
structure of the distribution system was reconfigured by adding tie
lines (the tie switches are connected between 46–65, 27–67 and
35–57) as seen in Fig. 8. It can be evidently observed from
Fig. 20 that in the reconfigured system the multistage ZIP model is
more accurate in representing the load demands and network
losses compared with the ZIP model. The obtained results here are
directly related to Figs. 11 and 12 that show the effectiveness of
the multistage ZIP model. Despite the new configuration of the
system, the realistic behaviour of the system during charging
activities can be seen clearly with the multistage ZIP model. This
is an indication that the proposed model can be effectively applied
to all kinds of topologies of the distribution systems.

6 Conclusion

In this paper, a multistage, time-variant load modelling technique has


been developed for EV charging load model and imbedded in ZIP
model. The EV battery charging load in a distribution grid has
been analysed by considering constant load model, and ZIP load
model without multistage and ZIP model with multistage
Fig. 18 Power congestion level with 15% PEVs ‘L12 (12–13)’ voltage-dependent load models. The impacts of EV charging parks

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 16, pp. 2705–2716
2714 & The Institution of Engineering and Technology 2015
Authorized licensed use limited to: Universidad Tecnologica de Pereira. Downloaded on February 24,2020 at 04:24:09 UTC from IEEE Xplore. Restrictions apply.
Fig. 20 Load demand and network loss profiles for 25% EV penetration when tie switches are closed

on the distribution system were quantified using the proposed 7 Jeffrey, S., Paul, D., Jiarui, D., et al.: ‘Modeling the impact of electric vehicle
multistage time-variant load modelling technique and compared charging on heat transfer around underground cables’, Electr. Power Syst. Res.,
2013, 97, pp. 76–83
with the results obtained from the use of ZIP model without 8 Yunfei, M., Jianzhong, W., Nick, J., et al.: ‘A spatial–temporal model for grid
considering multistage. The results have revealed that the proposed impact analysis of plug-in electric vehicles’, Appl. Energy, 2014, 114, pp. 456–465
modelling technique can provide the accurate results compared 9 Claus, K.: ‘On the synergy between large electric vehicle fleet and high wind
with the existing models. It has been observed that the voltage penetration – an analysis of the Danish case’, Renew. Energy, 2011, 36,
profile, system losses and system load demand depend pp. 546–553
10 Robert, C., Lingfeng, W., Mansoor, A.: ‘The impact of plug-in hybrid electric
significantly on the load and hence the use of accurate load model vehicles on distribution networks: a review and outlook’, Renew. Sustain.
is necessary to determine the true impacts of EV load on the Energy Rev., 2011, 15, pp. 544–553
networks. It is established that EV integration into the power grid 11 Deilami, S., Masoum, A., Moses, P., et al.: ‘Real-time coordination of plug-in
should not be simply planned based on the results obtained from electric vehicle charging in smart grids to minimize power losses and improve
the constant power load model which has been traditionally used, voltage profile’, IEEE Trans. Smart Grid, 2011, 2, (3), pp. 456–467
12 Rotering, N., Ilic, M.: ‘Optimal charge control of plug-in hybrid electric vehicles in
as it does not provide the accurate EV load behaviour. It is also deregulated electricity markets’, IEEE Trans. Power Syst., 2011, 26, (3),
established that the traditional ZIP model in its current form is not pp. 1021–1029
suitable for planning of EV penetration in the electricity systems. 13 Kristien, C., Edwin, H., Johan, D.: ‘The impact of vehicle-to-grid on the
It is necessary to use a time-variant, multistage load modelling distribution grid’, Electr. Power Syst. Res., 2011, 81, pp. 185–192
technique to properly represent the EV load behaviour in a 14 Qian, K., Zhou, C., Allan, M., et al.: ‘Modeling of load demand due to EV battery
charging in distribution systems’, IEEE Trans. Power Syst., 2011, 26, pp. 802–810
time-series simulation. The proposed load modelling approach will 15 Sortomme, E., Negash, A., Venkata, S., et al.: ‘Multistate voltage dependent load
enable power utilities to quantify, in a realistic manner, the true model of a charging electric vehicle’. Proc. Int. Conf. on Transportation
effects of EV charging load on the networks as well as on the Electrification Conf., Dearborn, MI, June 2012, pp. 1–5
overall system demand while planning for high EV penetration in 16 Price, W., Wirgau, K., Murdoch, A., et al.: ‘Load modeling for power flow and
the future. transient stability computer studies’, IEEE Trans. Power Syst., 1988, 3, (1),
pp. 180–187
17 Buchmann: ‘Battery University: charging lithium-ion’. Available at http://www.
batteryuniversity.com/learn/article/charging_lithium_ion_batteries, accessed on
March 2013,
7 References 18 Tremblay, O., Dessaint, L.: ‘Experimental validation of a battery dynamic model
for EV applications’, World Electr. Veh. J., 2009, 3, pp. 1–10
1 Speidel, S., Jabeen, F., Olaru, D., et al.: ‘Analysis of Western Australian electric 19 Johnson, V.: ‘Battery performance models in ADVISOR’, J. Power Sources, 2002,
vehicle and charging station trials’. Proc. Int. Conf. on Australian Transport 110, pp. 321–329
Research Forum, Perth, Australia, September 2012, pp. 1–12 20 Mi, C., Masrur, M., Gao, D.: ‘Hybrid electric vehicles: principles and applications
2 Mota, L., Mota, A.: ‘Load modeling at electric power distribution substation using with practical perspectives’ (John Wiley and Sons Press, USA, 2011, 1st edn.)
dynamic load parameters estimation’, Electr. Power Energy Syst., 2004, 26, (10),
21 Mendoza, A., Argueta, J.: ‘Performance characterization-GM EV1 Panasonic lead
pp. 805–811
acid battery’ (Southern California Edison, 2000), pp. 1–57
3 Clement-Nyns, K., Haesen, E., Driesen, J.: ‘The impact of charging plug-in hybrid
electric vehicles on a residential distribution grid’, IEEE Trans. Power Syst., 2010, 22 Madrid, C., Argueta, J., Smith, J.: ‘Performance characterization – 1999 Nissan
25, (1), pp. 371–380 Altra-EV with lithium-ion battery’ (Southern California Edison, 1999), pp. 13–42
4 Fernandez, L., Roman, T., Cossent, R., et al.: ‘Assessment of the impact of plug-in 23 Les, M., Behnam, D.: ‘Laboratory measurements and models of modem loads and
electric vehicles on distribution networks’, IEEE Trans. Power Syst., 2011, 26, (1), their effect on voltage stability studies’, IEEE Trans. Power Syst., 1998, 13, (2),
pp. 206–213 pp. 584–592
5 Mullan, J., Harries, D., Braunl, T., et al.: ‘Modelling the impacts of electric vehicle 24 José, R., Hamed, A., Lincol, B.: ‘Linear power-flow formulation based on a
recharging on the Western Australian electricity supply system’, Energy Policy, voltage-dependent load model’, IEEE Trans. Power Deliv., 2013, 28, (3),
2011, 39, pp. 4349–4359 pp. 1682–1690
6 Hugo, M., Tiago, S., Zita, V., et al.: ‘Evaluation of the electric vehicle impact in the 25 IEEE Task Force on Load Representation for Dynamic Performance: ‘Standard
power demand curve in a smart grid environment’, Energy Convers. Manage., load models for power flow and dynamic performance simulation’, IEEE Trans.
2014, 82, pp. 268–282 Power Syst., 1995, 10, (3), pp. 1302–1313

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 16, pp. 2705–2716
& The Institution of Engineering and Technology 2015 2715

Authorized licensed use limited to: Universidad Tecnologica de Pereira. Downloaded on February 24,2020 at 04:24:09 UTC from IEEE Xplore. Restrictions apply.
26 Lindén, K., Segerqvist, I.: ‘Modelling of load devices and studying load/system 32 Ashtari, A., Bibeau, E., Shahidinejad, S., et al.: ‘PEV charging profile prediction
characteristics’. Technical Report No. 131L, Chalmers University of Technology, and analysis based on vehicle usage data’, IEEE Trans. Smart Grid, 2012, 3, (1),
Göteborg, Sweden, 1992 pp. 341–350
27 Marra, F., Traholt, C., Larsen, E., et al.: ‘Average behavior of battery-electric 33 Kersting, W.: ‘Distribution system modeling and analysis’ (CRC Press, Taylor and
vehicles for distributed energy studies’. Proc. Int. Conf. on Innovative Smart Francis Group, USA, 2007, 2nd edn.)
Grid Technologies, Gothenburg, October 2006, pp. 1–7 34 Baran, M., Wu, F.: ‘Optimal capacitor placement on radial distribution systems’,
28 Sainz, L., Mesas, J., Balcells, J.: ‘On electric vehicle battery charger modeling’. IEEE Trans. Power Deliv., 1989, 4, (1), pp. 725–734
Proc. Int. IEEE Power Quality Conf., 2011, pp. 1–5. Available at http://www. 35 Chin, H.: ‘Optimal shunt capacitor allocation by fuzzy dynamic programming’,
icrepq.com/icrepq’11/350-sainz Electr. Power Syst. Res., 1995, 35, pp. 133–139
29 Haidar, A.M.A., Kashem, M., Danny, S.: ‘Technical challenges for electric power 36 Chen, J., Ricardo, T., Diogo, S., et al.: ‘Method to assess the power quality impact of
industries due to grid-integrated electric vehicles in low voltage distributions: a plug-in electric vehicle’, IEEE Trans. Power Deliv., 2014, 29, (2), pp. 958–965
review’, Energy Convers. Manage., 2014, 86, pp. 689–700 37 Dharmakeerthi, C., Mithulananthan, N., Saha, T.: ‘Impact of electric vehicle fast
charging on power system voltage stability’, Electr. Power Energy Syst., 2014,
30 Leon, M., Vargas, R.: ‘Local voltage stability assessment for variable load
57, pp. 241–249
characteristics’. Master thesis, The University of British Columbia, 2009
38 Olivier, T., Louis-A., D., Abdel-Illah, D.: ‘A generic battery model for the dynamic
31 Stephen, B., Lieven, V.: ‘Convex optimization’ (Cambridge University Press, simulation of hybrid electric vehicles’. Proc. Int. Conf. on Vehicle Power and
New York, USA, 2004) Propulsion, Arlington, TX, September 2007, pp. 284–289

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 16, pp. 2705–2716
2716 & The Institution of Engineering and Technology 2015
Authorized licensed use limited to: Universidad Tecnologica de Pereira. Downloaded on February 24,2020 at 04:24:09 UTC from IEEE Xplore. Restrictions apply.

Вам также может понравиться