Professional Documents
Culture Documents
fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2016.2550219, IEEE
Transactions on Smart Grid
1
1949-3053 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2016.2550219, IEEE
Transactions on Smart Grid
2
Vi,k,s Stochastic voltage magnitude at bus i, hour k, impacts of PEV charging on a real low voltage distribution
scenario s, p.u. network for various PEV penetrations is discussed in [10], it is
i,k Voltage angle at bus i, hour k, rad shown that PEV charging can have negative impacts in terms
i,k,s Stochastic voltage angle at bus i, hour k, of increased peak load, increase of power losses, overload of
scenario s, rad transformers and lines, voltage drop and increased voltage
asymmetry. Bidirectional dispatch coordination of PEVs in a
II. INTRODUCTION power grid is proposed in [11] using a robust optimization
The global demand for energy has been increasing rapidly, model; it is concluded that coordinating PEV
which imposes a large burden on the existing energy resources, charging/discharging with thermal generators and the ability of
and adversely impacts environmental pollution and global PEV aggregators to provide regulation reserve service are
warming. As all governments around the world move toward a beneficial to power grid operation. In [12], an optimization
green energy economy, the Plug-in Electric Vehicles (PEVs) model is developed to determine the optimal and maximum
have an increasingly important role to play, because of their penetration of PHEVs in the transport sector of Ontario
contribution to emissions reduction from the transportation considering the grid limitations.
sector. In Canada, almost 35% of the total energy demand is The travel patterns of light-duty vehicles in the U.S.
from the transport sector and it is the second largest source of obtained from the 2009 National Household Travel Survey
greenhouse gas (GHG) emissions [1]. The Government of (NHTS) is used in [13] to estimate the electric energy and
Ontario, Canada, has chalked out a path to move toward a power consumption of PEVs for two uncontrolled charging
green energy economy, aimed towards increased penetration of scenarios.
renewable energy sources and PEVs [2]. Penetration of PEVs The charging demand of PEVs is affected by different
into the market is expected to be large in the near future, and uncertain factors, such as the number of PEVs being charged
with their complex charging behavior, many technical simultaneously, their charging levels, battery capacity, and
problems related to their impact on the power grid need be charging duration. Monte Carlo Simulation (MCS) is used in
investigated. Since the charging behavior of PEVs is dependent [14] to generate virtual trip distances which consider driving
on a number of factors, their overall charging demand tends to habits, different vehicle models, etc., and hence formulates an
be uncertain. annual energy consumption model of light duty fleet of PEVs.
Several studies show that the power distribution grid can be In [15], a max-weight PEV dispatch algorithm, based on a
significantly impacted by high penetration levels of PEVs [3- queuing formulation integrated with renewable energy sources
5]. Electric vehicle (EV) charging will likely coincide with the is used to control the PEV charging in order to avoid costly
system peak demand and thus, in order to avoid overloading of distribution system infrastructure upgrades. A probabilistic
the distribution feeders, adequate load management schemes constrained load flow problem with wind generation and EV
need be in place [3]. In [4], a comprehensive study to assess demand is presented in [16] where the charging and
the implications arising from adoption of PEVs in Ontario, discharging processes are considered using M/M/ queuing
from the technical, consumer, policy, regulatory and market model. The spatial and temporal distribution of demand, based
points of view are presented and specific measures and on fluid dynamic traffic model and queuing theory is
approaches, and policy initiatives relevant to the Ontario developed in [17] to determine the EV charging demand for a
context is discussed. Different charging scenario studies rapid charging station. In [18], an EV demand model suitable
reported in [5] conclude that PEVs significantly increase for load flow studies is proposed wherein the EV demand is
demand side uncertainties and can potentially reduce the represented as a PQ buses with stochastic characteristics based
distribution transformer insulation life. In [6], the impact of EV on the concept of queuing theory. In [19], four different types
charging on a medium-voltage network and the Local of PHEVs are considered and factors that affect their charging
Distribution Companys (LDCs) benefits arising from the use behavior, e.g., differences in battery capacity and charging
of smart charging schemes is discussed. In [7] PEV charging level, is discussed. A single PHEV charging demand model is
control strategies are developed to mitigate distribution formulated and queuing theory is used to describe the behavior
transformer ageing that could result from load peaks caused by of multiple PHEVs.
PEV charging. It is noted that grid stability, need for In recent research, the Model Predictive Control (MPC)
infrastructure investment, and other challenges can be approach has been applied to various operational and control
mitigated by the LDC by adopting smart charging strategies. problems in the context of smart grids to consider the effect of
The impact of uncoordinated Plug-in Hybrid Electric uncertainties. In [20] a prediction based real-time charging
Vehicle (PHEV) charging on system peak load, losses, voltage method is proposed that considers the effect of future vehicles
and system load factor are discussed in [8] and is noted to have penetrating into the grid. In [21] an MPC based approach is
adversely affected the efficiency of the distribution grid. It is proposed for a minimum cost dispatch based operation while
shown in [9] that a 10% penetration of PEV may cause including the effect of PEV charging loads.
Most of the works on PEV charging demand modeling that
unacceptable variations in voltage profiles if there is no
use queuing analysis [16-19], consider the arrival rates as a
regulation on PEV charging, while coordinated charging can
Poisson process, which is a stochastic process that assumes the
reduce system peak load, losses, and mitigate the impacts of PEVs have a constant arrival rate. Only a few have modeled
uncoordinated PEV charging in the distribution system. The
1949-3053 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2016.2550219, IEEE
Transactions on Smart Grid
3
the arrival rate as a non-homogeneous Poisson process [22]. A comprehensive set of studies considering load flow and
Moreover, the charging time is typically modeled by OPF analysis is carried out to examine the impact on LDC
exponential distributions with given upper and lower limits, operation, and to determine the optimal strategies for the
which are randomly assigned to each PEV, and the waiting LDC to improve its operational performance. Thereafter,
time is assumed to be infinite. Furthermore, it should be noted considering the PEV charging load model may have
that during the fast charging process, the charging power uncertainties, therefore to study the impact of such load
typically starts at a high rate, and drops off as the battery State model on LDC operation in the presence of uncertainties,
of Charge (SOC) approaches its full capacity, as per the battery stochastic load flow, stochastic OPF analysis, and MPC
charging behavior (BCB) of PEVs, discussed in [23]. This based analysis are carried out.
affects the charging time and needs to be considered in the The rest of this paper is organized as follows. The basics of
PEV load model, which has not been considered so far. the queuing model, the PEV BCB model, the PEV charging
Furthermore, in a fast charging station an infinite waiting time model and the distribution system operations model are
is not a reasonable assumption. In view of the above, the main presented in Section III. In Section IV, the system description
objectives of this paper are: pertaining to the case study carried out is presented. The results
Estimate the distribution of vehicles which are on the road and discussions are presented in Section V and the concluding
during the day using mobility statistics from the Waterloo remarks are presented in Section VI.
Region Transportation Tomorrow Survey (TTS) [24] and
construct the PEV arrival rate profile at a charging station III. MATHEMATICAL MODELING
based on customer convenience. Also, develop PEV arrival
rate profile based on customers response to PEV charging A. PEV Queuing Model
price. Queuing analysis is applied to estimate the total charging
Using the developed non-homogeneous Poisson process type power of PEVs. The PEV customers are considered to be
arrival rate profiles, the service time of PEVs modeled using served using M1/M2/N0 queue model at a PEV charging station,
their BCBs and waiting times, develop appropriate queuing where M1 denotes the arrival rate which varies from hour to
models to determine the 24-hour charging demand profile at hour of the day and is modeled as a non-homogeneous Poisson
a PEV charging station. process, the service time denoted by M2 includes the waiting
Integrate the developed PEV charging load model within a time and the charging time. The service time is modeled in this
distribution system operation framework to study the impact paper considering the PEV BCB. Poisson process is a
of PEV charging on distribution system operation
continuous-time stochastic process that counts the number of
considering optimal power flow (OPF), stochastic analysis,
arrivals in a given time interval where the time between each
and the Model Predictive Control (MPC) approaches.
pair of consecutive arrivals has an exponential distribution with
The novel features and contributions of this paper are as
follows: (mean of inter-arrival time) and each of these inter-arrival
The arrival of customers to charge their PEVs at the charging times are assumed to be independent of other inter-arrival
station has been modeled as continuous and random, non- times. It is useful for modeling arrival that occur independently
homogeneous Poisson processes, considering two different from each other [26]. Since the arrival of PEVs at the charging
arrival patterns- customers convenience, and customers station is a continuous-time stochastic process, Poisson process
response to PEV charging price. The first arrival rate profile has been considered in this work. In accordance to M1/M2/N0
makes use of the large database of mobility statistics queuing analysis [27], the probability of the number of PEVs
available from the Waterloo Region TTS [24]. The second charging simultaneously at an hour is modeled as a discrete
arrival rate profile considers customers response to PEV distribution, as follows:
charging price, where the price data is obtained from the
pk (n ) =
(N 0 )n pk (0 ) n = 1,2,3,...., N 0 (1)
winter Time-of-Use (TOU) prices applicable in Ontario,
n!
Canada [25].
where,
The queuing model proposed in this paper considers for the 1
N 0 1
first time, a detailed representation of the BCB of PEVs and (N 0 ) m + (N 0 ) N 0 1
the SOC, and the time taken to charge PEVs, including a pk (0) =
m! N 0! (1 )
(2)
finite waiting time. Such detailed models provide accurate m=0
information of the charging load on the distribution system,
taking into account different classes of PEVs. and is the occupation rate of the PEV charging station and is
These features render the proposed queuing model more calculated as follows:
realistic and accurate and also more generic and universal, 1
than the existing models. For example, homogenous Poisson k
= (3)
process models are a special case of non-homogeneous 1
Poisson process models, and similarly, the modeling of in N0
k
this work considers the SOC, the BCB, and the time of
The expected number of customers waiting at the charging
charging including a finite waiting time- which is the most
station, at hour k is given by Littles Law in queuing theory as
generic form of model for M2.
follows [27]:
1949-3053 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2016.2550219, IEEE
Transactions on Smart Grid
4
50
50 %
C. OPF Model for System Operation Including PEV Load
40 Once the PEV charging load is estimated using the queuing
30
20
model, the impact on system operation is examined by
10 10 Min 5 Min 7 Min 12 Min 25 Min formulating the following OPF model, with the objective of
0 minimization of feeder losses, as given below:
0 10 15 25 35 60
1
24 N N
( ( ))
Minutes 2 2
PLoss = G i , j V i , k + V j , k 2Vi , k V j , k cos j , k i , k (14)
Fig.1: Typical BCB of a Compact PEV during fast charging [23] 2 k =1 i =1 j =1
The SOC of the battery can be obtained as follows: The PEV charging load (Pchi,k) is included in the active
EC y power balance at bus i where the charging station is located.
SOC y = 1 y (6) The demand-supply balance for both active and reactive power
C Bat y
is given by the load flow equations as follows.
and the following conditions are imposed, from the above BCB N
of the PEV in Fig.1:
PGi ,k PDi ,k Pchi ,k = Vi ,k V j ,k Yi , j cos i , j + j ,k i ,k ( (15) )
0.2 if SOC y 0.2 j =1
N
SOC y = SOC y if 0.2 < SOC y 0.85
(7)
(
QGi , k QDi , k = Vi , kV j , k Yi , j sin i , j + j , k i , k )
(16)
0 . 85 if SOC y > 0 . 85 j =1
Once the SOC of a PEV is known, the required charging (or The voltage magnitudes at each bus at hour k are constrained
service) time of a PEV, given by T, is obtained from the BCB by their respective upper and lower limits.
(Fig.1) using the piece-wise linear relationship: Vi Min Vi , k Vi Max i Load buses (17)
SOCl y bl
T= l {1, 2, 3, 4} , y (8) The slack bus voltage magnitude and voltage angle, which is
al the substation bus, are fixed, as follows.
where al and bl are the slope and intercept of the linear equation Vsb, k = 1 p.u., sb, k = 0
obtained from the BCB and depend on the SOC of interval l. The substation capacity limit determines the maximum and
The charging current drawn by a PEV over the charging minimum active and reactive power transfer capacity over the
period T, at time k for PEV class y is given as follows: substation transformer.
EC y
I k , y = min( , I Max ) (9) PSiMin PSi , k PSiMax i = sb (18)
VTk
1949-3053 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2016.2550219, IEEE
Transactions on Smart Grid
5
N Yes
QGi,k ,s QDi,k = V (
i,k ,sV j ,k ,s Yi, j sin i, j + j ,k ,s i,k ,s ) (22) Calculate the expected power from (13)
j =1
1949-3053 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2016.2550219, IEEE
Transactions on Smart Grid
6
PEV charging station is arbitrarily located at bus-59. In this B. Modeling of Daily Driven Miles and PEV Arrival Rate, M1
paper, Level-3 charging is considered since high power level Fig. 5 shows the distribution of daily miles driven on all
charging is preferred at PEV charging stations, and thus IMax = vehicle driving days based on the TTS data. It is noted that the
63 amps and V = 400 volts. best fit for the TTS data (using EasyFit software [32]) is
Distribution systems are generally balanced by using various obtained with a lognormal distribution as evident from Fig.6.
load balancing schemes, and hence can be represented by So, the daily driven miles by the PEVs, DD, is modeled as a
single phase equivalents [31]. The unbalanced nature of lognormal distribution in this work, and is given by:
distribution system is more prevalent at the end-user level
DD y = e ( M + M f ) (26)
(residential customer level) but since this work considers a
PEV charging station load model, it is assumed to be connected where M and M are the mean and the variance of the
at 12.66 kV feeder level; and at this voltage level, the loads are lognormal distribution, respectively.
assumed to be balanced three-phase, and all line segments are 14
Percent of Vehicles %
three-phase, and perfectly transposed [31]. 12
With these assumptions, a single line-to-neutral equivalent 10
8
circuit for the feeder has been used, and a three phase
6
distribution system is represented by a single-phase equivalent.
4
2
0
Fig.6: Distribution of daily driven distance per vehicle using EasyFit [32]
1949-3053 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2016.2550219, IEEE
Transactions on Smart Grid
7
It is important to point out that, the arrival rates modeled in V. RESULTS AND DISCUSSIONS
this paper are based on some assumptions pertaining to
vehicles on the road, charging price and how PEVs arrive for A. PEV Charging Load Using Queuing Analysis
charging at the charging station. Such assumptions are In this section the effect of PEV charging on the distribution
necessary in order to understand the impact of PEV charging system performance is examined. Queuing analysis is used to
on the distribution grid but, need be validated with realistic model the 24-hour PEV charging demand at the charging
data from ground level surveys. station. The objective is to determine the optimal distribution
Moreover, as per Scenario-2, the arrival rate would be high system operation considering PEV charging demand while
at night since the PEV charging price is low at these hours. minimizing the system losses using the OPF model. The
However, considering that the probability of charging during probability distribution of N0 is obtained from (1) and shown in
night is low, because of customer inconvenience, the arrival Fig.10 which is used as an input to the M1/M2/N0 queuing
rate is modified appropriately, as shown in Fig. 9, where the model [27]. Different NCap values are examined within the
removed arrival data of early hours are indicated. Also to be proposed queuing model and it is noted that when NCap is high,
noted that since home charging has been ignored in this work, the probability of simultaneously charging of NCap number of
there will be no effect on the early morning arrival rates. The PEVs, i.e., P(NCap) is very low, as seen from Fig.10. Because
two arrival rate profiles, as discussed above, are modeled as of this low probability, there is insignificant impact on the total
non-homogeneous Poisson processes with mean k which is the expected charging demand for high values of NCap. By various
time dependent number of expected car arrivals at a charging trial runs it is noted that beyond NCap =17, the effect on
station throughout the day. expected charging demand does not change significantly and
hence, NCap =17 was chosen for the paper.
It is noted from Fig.11 that the expected PEV charging
demand is low for low arrival rates (M1 = 40, i.e., a PEV
arriving every 40 minutes) and the discrete distribution pattern
of the demand as a function of N0 is skewed normal with a low
mean value of N0. As the arrival rate increases, i.e., M1 = 25 (a
PEV arriving every 25 minutes), and then for M1 = 10, the
expected PEV charging demand increases and the distribution
Fig.9: Number of PEVs arriving per hour at the charging station (Arrival rate)
1949-3053 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2016.2550219, IEEE
Transactions on Smart Grid
8
Fig.15: Total PEV expected charging demand for all queuing model
1949-3053 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2016.2550219, IEEE
Transactions on Smart Grid
9
Fig.20: PEVs charging, overflow, waiting: not considering BCB, > NCap Fig.23: Impact of PEV charging on losses during optimal operation of system
1949-3053 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2016.2550219, IEEE
Transactions on Smart Grid
10
Fig.25: Expected voltage profile at Bus 65 for optimal operation Fig.29: Expected active power transfer, optimal operation
1949-3053 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2016.2550219, IEEE
Transactions on Smart Grid
11
Fig.32: Comparison of optimal operation versus MPC for Scenario-1 Fig.36: Comparison of optimal operation versus MPC for Scenario-1
1949-3053 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2016.2550219, IEEE
Transactions on Smart Grid
12
1949-3053 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2016.2550219, IEEE
Transactions on Smart Grid
13
[27] Adan, I., and Resing, J., Queueing theory, available at:
http://www.win.tue.nl/_iadan/queueing. pdf, Department of
Mathematics and Computing Science, Eindhoven University of
Technology, The Netherlands, 2002.
[28] Y. Cao, Y. Tan, C. Li, and C. Rehtanz, "Chance-constrained
optimization-based unbalanced optimal power flow for radial
distribution networks," IEEE Trans. Power Del, vol. 28, no. 3, July
2013.
[29] F. Allgwer, R. Findeisen, and Z. Nagy, Nonlinear model predictive
control: from theory to application, J. Chin. Inst. Chem. Engrs., Vol.
35, No. 3, 299-315, 2004.
[30] M. E. Baran and F. F. Wu, Optimal sizing of capacitor placed on ra-
dial distribution systems, IEEE Trans. Power Del. , vol. 4, no. 1, pp.
735743, Jan. 1989.
[31] W. H. Kersting, Distribution System Modeling and Analysis, CRC
Press, New Mexico, 2002.
[32] EasyFit Software [Online].Available:
http://www.mathwave.com/en/home.html.
ACKNOWLEDGMENTS
The authors would like to thank P. Fisher, and G. Keyworth, Regional
Municipality of Waterloo and the Data Management Group, Department of
Civil Engineering, University of Toronto for providing the data to perform this
study.
1949-3053 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.