See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/230580285

Spatial and Temporal Model for Electric Vehicle Rapid Charging Demand

Conference Paper · October 2012

DOI: 10.1109/VPPC.2012.6422675

CITATIONS READS

15 113

3 authors, including:

Tachun Lin K. Thulasiraman

Bradley University University of Oklahoma
26 PUBLICATIONS   213 CITATIONS    223 PUBLICATIONS   2,453 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Dominator Sequences View project

Network Function Virtualization View project

All content following this page was uploaded by Tachun Lin on 08 July 2015.

The user has requested enhancement of the downloaded file.

 2012 IEEE Vehicle Power and Propulsion Conference, Oct. 9-12,2012, Seoul, Korea

Spatial and Temporal Model for Electric Vehicle

Rapid Charging Demand
Zhili Zhou Tachun Lin
IBM Research Collaboratory Department of Computing and Technology
IBM Research Cameron University
Singapore, Singapore 486048 Lawton, USA 73505
Email: [email protected] Email: [email protected]

Abstract-In this paper, we present a spatial and temporal environment and highway systems. The accurate estimation of
model of electric vehicle (EV) charging demand for rapid EV charging demands would help determine the power supply
charging stations located in the urban area. For this problem
of existing charging station, new charging station placement,
three lines of investigations have been pursued in the literature:
the first one is based on an assumption of a fixed charging
vehicle to grid (V2G) implementation, and the reliability of
station and charging time during the off-peak hours for EV existing power grid. This paper presents a spatio-temporal
charging demand; the second scenario has limited charging model of EVs' charging demand for rapid charging stations
stations at typical locations instead of a mathematical model; in urban environment.
and the third one is for charging demands in one way system,
Although the spatio-temporal EV charging demand is im­
such as charging stations nearby highway exists. Hence, from
the perspective of a distributed urban transportation system,
portant for EV adoption in urban or highway traffic system,
EV charging demand still has unidentified quantity which varies there are very few literature and research works on EV
with respect to charging station location and charging time. In charging demand at rapid charging stations. We summarize
this paper we presents mathematical models for EV charging these earlier works into three categories. The first one has
demand at rapid charging stations. The rapid charging station
the assumption that charging demands are fulfilled at a fixed
is formulated as a M M / /s queuing service system. A traffic
flow based model generates the arrival rate for EVs to arrive the
charging station with fixed charging time (see [3] etc), for
charging station. The computational results demonstrate that the instance, at a resident slow charging station in the evening or
proposed model is able to estimate the dynamic spatio-temporal at night. But the operation of special purpose or commercial
charging demands in the large-scale urban traffic environment. PEVs, such as taxis and police cars, requires them to be
Index Terms-electric vehicle, charging demand, energy con­
charged anytime at rapid charging stations instead of a single
sumption, energy management, queuing model
residential charging station with specified time. Second, the
survey on EV charging demands at various charging locations
I. INTRODUC TION
has been studied in [4] etc. They anticipated EV charging
EVs are increasingly being seen as a sustainable type of demands at residential, office, and shopping areas based on
transportation by countries worldwide which are more energy the survey from potential EV users. However, EVs are newly
efficient than traditional vehicles and can reduce gasoline con­ adopted by most cities hence there is no historical data to
sumption and carbon emissions (see [1] etc). Several automak­ help EV surveys and EV potential user estimation and no
ers, such as Nissan, Chevrolet, and Ford have manufactured mathematical models have been developed. Third, Bae and
commercial plug-in electric vehicles (PEVs) which appeared in Kwasinski [5] studied the spatio-temporal model of EVs on
market in 2010 and 2011 (see [2]). More and more traditional a one-way high way system using a fluid dynamic traffic
automakers are joining the design and manufacture of PEVs. model and the lvI/lvI/ s queuing system. They considered
Federal government of several countries, such as Singaporae traffic model for one direction on the highway and obtained the
and the U.S., have developed EV test-beds which can be used charging EV behavior by an assumed split-ratio for departure
to analyze benefits and feasibility of EVs' adoption for urban number of vehicles and charging vehicles at a highway exist.
transportation and power grid system. Kintner-Meyer et al. In this paper, to the best of our knowledge, our contributions
[3] concluded that the transportation electrification can fuel lay in the spatio-temporal charging model for EV charging
up to 84% of the U.S.'s light-duty vehicle fleet with existing demand with multiple charging stations in complicated urban
electricity infrastructure. However, they also demonstrated traffic environment, instead of charging demand in a single
that high plug-in hybrid electric vehicle (PHEV) and PEV charging station or in the one way traffic system.
penetration may lead to negative impact on extreme events The rest of the paper is organized as follows. In Section
or cause grid emergencies to existing power gird, which II, we provide the problem description and the corresponding
brings higher electricity cost, lower reliablity, and other side formulation. In Section III, we present the computational
effects. Hence, the spatio-temporal charging demands of EVs results of our model in Singapore traffic network with fast
are fundamental questions for EV adoption in urban traffic charging facilities. In Section IV, we provide a conclusion

978-1-4673-0954-7/12/$3\.00 ©2012 IEEE

345
and future research direction for this paper.

II. PROBLEM DESCRIP TION AND FORMULATION

In this section, we provide the formal problem descrip­ (a) Cells without consideration of charging facility
tion and demonstrate mathematical model for spatio-temporal
charging demand of rapid charging stations. We let G(N, A)
denote the traffic network with node set N and arc set A.
We consider a model which integrates cell transmission model
and queuing model for transportation system and rapid charg­
ing facility service, respectively. The cell transmission based charging facility
transportation models are discussed by Daganzo in [6] and (b) Cells with consideration of charging facility
[7] and Ziliaskopoulos in [8]. The cell transmission model is
a hydrodynamic model utilizing differential equations, where
the difference equations are simplified with the assumption Fig. 2. Categorization of cells in cell transmission model

that a piecewise linear relationship between flow and density

in the cell level. It assumed that a cell of a segment on a
traffic link whose length is equal to the traveling distance of cells, respectively. Note here that every cell i E Cc would
free-flow vehicle during a fixed time interval T. be splitted into two new cells il and i2 which represent the
Based on the cell concept, Ziliaskopoulos in [8] categorized cell with regular traffic flow (bypassing the charging station)
the set of cells into five groups: (I) source cell, (2) sink cell, and the one with traffic entering entering a charging station,
(3) diverging cell, (4) merging cell, and (5) ordinary cell and respectively.
illustrated them in the following way. Next, we introduce variables utilized in the cell transmission
based EV traffic flow model.
X � : the number of vehicles contained in each cell i in time
interval t
ordinary cell diverging cell merging cell Nf : the maximum number of vehicles in cell i in time
interval t
Q� : the maximum number of vehicles that can flow in or
out from cell i during time interval t
source cell sink cell Y ; j : the number of vehicle moving from cell i to j during
time interval t
Fig. I. Categorization of cells in cell transmission model dJ : supply (inflow) at cell i in time interval t
d� : demand (outflow) at cell i in time interval t
We first present the model for the cell transmission based
A. Cell Transmission Model for EV Traffic Flow
traffic flow on source and sink cells, respectively.
Given the location of a rapid charging facility, we assume
that the traffic flow on a cell i at time t can then be splitted t t-l + dt-l - t-l, iECse,jE5-(i),tEI (1)
into two types, vehicles driving on the traffic link and vehicles
Xi = X
i i Yij
t t-l - dt-l + t-l, iECsk,jE5+(i),tEl (2)
staying on a charging station. After charging, the vehicles X i
= X
i i Yij
would be merged back to the traffic flow. Hence, we create
two cell for each rapid charging facilities. These cells are The initial value xO + i can be set to the initial traffic
connected with their upperstream and downstream cells. We condition of the network at the beginning of the time period of
assume that the split-ratio of traffic flow is known for regular interest. Constraints (1) and (2) are the flow mass conservation
driving and charging behavior and denote them as (Xl and (X2 constraints, which guarantee the flow coming in, going out of
with (Xl + (X2 = 1. In addition, we define the charging cell source and sink cells and the flow staying on cells would be
as a special set of the ordinary cells with notation Ce and all balanced.
successor and predecessor neighboring cells of charging cells Second, we present the flow conservation constraint for
are diverging and merging cells, respectively. ordinary, diverging, and merging cells.
We let I be the whole set of cells in the traffic network G
and I be the whole time horizon. We consider the discredited t t-l '"'" t-l + '"'" t-l,
time interval, t 1,2,. . . , III T. Given a cell i, we let
X
i
= X
i - � Yij � Yij
JEO-(i) jEO+(i)
= =

5+(i) be the set of successor cells of cell i and 5-(i) be the

set of predecessor cells of cell i. We let Cord, Cdivg, Cmrg, Csc, iECord U Cdivg U Cmrg' tEl (3)
Csk, Ccg, Ce be cell sets for ordinary cells, diverging cells,
merging cells, source cells, sink cells, and charging facility Third, we present bounded flow constraints for the flow

346
staying on, coming in, and going out of a cell. • The number of charging pumps in a charging station
located in cell i is 8(i) with iEC(c).
I: Y7j ::; x�, i,jE C, tET (4) • The charging service time for all pumps is the same
YE8t and equal to fL, which is independently and exponentially
I: Y7j ::; Q�, i,jE C, tET (5) distributed.
All vehicles entering a charging station served by charg­
yE8t
•

ing pumps follow the first-in-first-out rules (FIFO). Mean­

while, vehicles leave charging stations directly when
i,jE C, tET (6) charging is finished.
Following the cell transmission model, vehicles entering the
i,jEC,tET (7) charging station are seperated from the traffic entering cell i
and have ratios Q2(i), iECe. Then, the following conclusion
holds.
i,jEC,tET (8)
Proposition i: Following the cell transmission traffic flow
model, the arrival rate for vehicles entering charging station
Here v is the free-flow speed and 11) is the backward prop­ A(i) Q2(i) LkE8+(i) yL with iEC(c).
=

agation speed. Constraint (4) presents that the flow between Based on above assumptions, the charging service in a
any two cells is bounded by the occupancy of the beginning charging station follows the !vI/!vI/8 queue, whose arrival rate
and end cells; constraints (5) and (6) define the remaining and the departure rates depend on the number of vehicles in
capacity of the beginning and ending cells; and constraints (7) the system, is known as a birth and death queueing model (see
and (8) represent the maximum flow which can get out of the [9]).
beginning cell and get into ending cells, respectively. Proposition 2: The departure rates of vehicles from a
The above constraints completely formulate the EV traf­ charging station in cell i with 8(i) charging pumps can be
fic flow model without considering charging facilities and formulated as follows.

{ IL8(Z),
their correponding ordinary cells' coming-in, going-out, and
staying-on flow. The objective function considered here is to fLxL , if x�, ::; 8(i),
minimize the total traveling time during the whole assignment Vt2t =

. if Z.ECc (11)
xL- > 8(i),
period T and formulated as follows:

minI: I: TX� (9) C. The Formulation for Charging Cells

x,y tEl iEC\C,
Now we discuss the model for charging cell i at time interval
Due to that T is a constant time interval, the above objective t, iEC(c) and tET
(9) could be rewritten as
minI:
Y
I: x� (10) iECc (12)
x, tEl iEC\C,
Without the consideration of charging stations, Ziliaskopou­ iECe (13)
los presented in [8] the necessary and sufficient condition for
the optimal condition of a system with a single destination iECc (14)
node.
Theorem 1 (Ziliaskopoulos [8]).' A necessary and sufficient
condition for system optimum dynamic traffic assignment on a With Condition (11) and Constraint (12), Constraint (12)
single destination network is that all used paths from any cell could be re-written as follows:
and departure time interval to the destination cell have cost
equal to the marginal cost of an additional unit of demand at t- t t
that cell and time interval, while all unused paths have cost D y t2J1 >
" - r'IIX t2 - !vIrl t (15)
jE8-(i)
I: y;;/:::-. fL8(i) - M(177 - 1)
higher than or equal to the marginal cost.
(16)
B. Charging Station Service Model jE8-(i)
In this section, we present the charging station service model X�2 - 8(i) ::; Mrl7, iECc (17)
as an !vI/!vI/8 queue with the following assumptions.
• The arrival process of vehicles to a rapid charging facility where !vI is a given big number. Constraint (17) determines
can be described by a Poisson process (and thus the the larger value between xh and 8(i). Then, Constraints (15)
interarrival time are exponentially distributed) with rate and (16) push the outflow from a charging station to be equal
A. to the departure rate of the charging station.

347
D. Cell Transmission Model for EV Traffic Flow on the OlD matrix, where the mean value of the OlD flow in
Now, we conclude the model for the EV traffic flow model the seed OlD matrix has a variance ±10% of the seed OlD
as follows: flow.
At the 20 mins time slot during the testing time horizon, the
minL L
x,y
x � total number of charging bumps in service and the charging
tET iEC\C, demand at charging stations located in the shopping area of
s.t.Constraints (1)-(8) and (15)-(17). (18) CBD and the airport are 6 and 258 KW, respectively. At the
40 mins time slot during the testing time horizon, the total
E. Spatia-temporal Charging Demand by Cell Transmission
number of charging bumps in service and the charging demand
Model and MIMls Queue at charging stations located in the shopping area of CBD and
In this section, we consider the spatio-temporal charging the airport are 8 and 344 KW, respectively.
demands based on the flow information generated from the
cell transmission traffic model and !vI/!vI/ s queue. IV. CONCLUSION AND FUTURE RESEARCH

Proposition 3: The steady state of !vI/!vI/ s queue for a In this paper we presented a spatio-temporal model for
charging station is commercial EV charging demand at rapid charging stations.
We proposed a traffic model based on dynamic traffic as­
(19) signment in the urban traffic network and formulated the
charging service at each placed rapid charging station as
If the above condition does not hold, then, the queue size of the an !vI/!vI/ s queue. Then, we obtained the spatio-temporal
charging station would be increased without limitation. Hence, charging demand at the charging station by combining the
there would be no limiting probabilities for this M/M/ s traffic flow model with charging station service model.
model. REFERENCES
Proposition 4: The spatio-temporal charing demand at
[1] M. Duvall. E. Knipping. M. Alexander, L. Tonachel, and C. Clark,
charging station i at time t with i E Cc and t E T is
{ X�2'
"Environmental assessment of plug-in hybrid electric vehicles-volume
I: Nationwide greenhouse gas emissions," Electric Power Research
if L s( i)
<
(3B2t with B2t
x
=
(20) Institute,Palo Alto,CA,Tech. Rep. 1015325,2007.
s(i), otherwise, [2] L. Dickerman and J. Harrison,"A new car,a new grid," IEEE Power &
Energy Magazine, vol. 8,pp. 55-61,2010.
where (3 is the average charging power per pump. [3] M. Kintner-Meyer,K. P. Schneider,and R. G. Pratt,"Impacts assessment
of plug-in hybrid vehicles on electric utilities and regional us power
III. COMPUTATIONAL RESULTS grids: Patt 1: Technical analysis," Paciffic Northwest National Labora­
tory,Richland,WA,Tech. Rep. PNNL-SA-61669,2007.
In this section, we report our computational report for the [4] J. Axsen and K. S. Kurani, "Anticipating plug-in hybrid vehicle en­
spatio-temporal charging demand at rapid charging facility. We ergy impacts in california: Constructing consumer-informed recharge
profiles," Transportation Research: Part D, Transport and Environment,
explore our computational report in Singapore traffic networks. vol. 15,pp. 212-219,2010.
First, the total testing time horizon considered is 1 hour [5] S. Bae and A. Kwasinski,"Spatial and temporal model of electric vehicle
with small time intervals T, each has 30 seconds. Then, there charging demand," IEEE Transactions on Smart Grid, vol. 3,pp. 394-
403,2012.
are 120 time slots in our computation. We consider the major [6] C. F. DAGANZO, " T he cell transmission model: A simple dynamic
roads, including expressway and major arterial roads, in Singa­ representation of highway traffic consistent with the hydrodynamic
pore. We assume that the speed limits on the expressway and theory," Transportation Research Part B: Methodological, vol. 28, pp.
269-287,1994.
arterial road to be 90 kmlhour and 45 km/hour, respectively. [7] --," T he cell transmission model. ii. network traffic," Transportation
We set the upper bound for flow volume on the express way to Research Part B: Methodological, vol. 29,pp. 79-93,1995.
be 150 vehicles/min (see [10]) and 50 vehicles/min for major [8] A. K. Ziliaskopoulos,"A lineat programming model for the single des­
tination system optimum dynamic traffic assignment problem," Trans­
arterial road. With a small time interval T 30 seconds, there =

portation Science, vol. 34,pp. 37-49,2000.

are total 180 cells under testing. We select 5 charging stations [9] S. M. Ross,Introduction to P robability Models, 10th ed. Academic
in Singapore traffic networks, which are located in the airport, Press,December 2009.
[10] P. Olszewski and L. Xie,"Modeling the effects of road pricing on traffic
the shopping area of the Central Business District (CBD), the in singapore," Transportation Research Part A, vol. 39, pp. 755-772,
business area of CBD, the developing area at the east and the 2005.
north sides of Singapore. In each charging station, we assume
that there are 4 charging pumps. We assume that the fixed ratio
of the traffic flow entering charging station over the regular
traffic flow is 3/7.
Based on the EVs' traffic information during a morning
peak hour, we generate the origin/destination (OlD) matrix.
The EV penetration is around 10% of regular traffic in our
model. In our experiment, we fix the original and destination
node pairs to be the same as the seed (peak hour) OlD matrix.
The flow between OlD node pairs is randomly generated based

348

View publication stats