Transportation Research Part A 94 (2016) 386396

Contents lists available at ScienceDirect

Transportation Research Part A

journal homepage: www.elsevier.com/locate/tra

Numerical analysis of electric bus fast charging strategies for

demand charge reduction
Nan Qin a,, Azwirman Gusrialdi b, R. Paul Brooker a, Ali T-Raissi a
a
Florida Solar Energy Center, University of Central Florida, 1679 Clearlake Rd, Cocoa, FL 32922, United States
b
Department of Electrical and Computer Engineering, University of Central Florida, Orlando, FL 32816, United States

a r t i c l e i n f o a b s t r a c t

Article history: Electric transit buses have been recognized as an important alternative to diesel buses with
Received 25 May 2016 many environmental benefits. Electric buses employing lithium titanate batteries can pro-
Received in revised form 18 August 2016 vide uninterrupted transit service thanks to their ability of fast charging. However, fast
Accepted 12 September 2016
charging may result in high demand charges which will increase the fuel costs thereby lim-
Available online 14 October 2016
iting the electric bus market penetration. In this paper, we simulated daily charging pat-
terns and demand charges of a fleet of electric buses in Tallahassee, Florida and
Keywords:
identified an optimal charging strategy to minimize demand charges. It was found that
Electric bus
Demand charge
by using a charging threshold of 6064%, a $160,848 total saving in electricity cost can
Lithium titanate be achieved for a five electric bus fleet, comparing to a charging threshold of 028%. In
Fast charging addition, the impact of fleet sizes on the fuel cost was investigated. Fleets of 4 and 12 buses
will achieve the lowest cost per mile driven when one fast charger is installed.
2016 Elsevier Ltd. All rights reserved.

1. Introduction

Due to increasing environmental pressure and unstable diesel fuel prices, the public transit agencies have been gradually
replacing legacy diesel buses with alternative fuel buses. There are over 70,000 transit buses operated by 800 transit agencies
in the United States (Dickens and Neff, 2014). The percentage of alternative fuel buses including compressed natural gas
(CNG), biodiesel, and hybrid electric buses rose from 8% in 2000 to 41% in 2012 (Bradley and Associates). The plug-in electric
bus (EB) is the newest player in the alternative bus arena and attracting increasing attention as it emits zero tail pipe green-
house gases and produces less noise. The price of electricity is not directly tied to the price of crude oil as it can be generated
from multiple domestic sources including solar, hydropower, coal, and natural gas, thus results in less price variation than
diesel (U.S. Energy Information Adminstration).
One of the major obstacles for EB market penetration is their high capital costs. Multiple studies conducted cost-benefit
analysis of EB by considering both capital costs and operational costs. For instance, Lajunen et al. analyzed the life cycle cost
of EB by taking into account of capital costs, electricity consumption, maintenance costs, battery system replacement costs,
and possible carbon emission costs (Lajunen, 2014; Lajunen and Lipman, 2016). Zhou et al. investigated the emission and
lifetime cost of several models of commercially available EBs in China and placed focus on traffic condition, passenger load,
AC operation and system charging efficiency (Zhou et al., 2016). Most research concluded that low fuel cost of EBs made
them competitive to diesel buses despite the high capital cost. However, a significant component of the electricity cost,

Corresponding author.
E-mail address: nqin@fsec.ucf.edu (N. Qin).

http://dx.doi.org/10.1016/j.tra.2016.09.014
0965-8564/ 2016 Elsevier Ltd. All rights reserved.
 N. Qin et al. / Transportation Research Part A 94 (2016) 386396 387

demand charges, was overlooked by these studies. Demand charge is a fee determined based on the peak power during a
given electricity billing cycle, usually a month. It is calculated as the peak demand (measured in kW) multiplied by a demand
charge rate ($/kW). Peak power is determined as the highest power (averaged over a 15 or 30 min window) for the billing
cycle. The demand charge rate ranges from $6 to $25/kW in the US (U.S. Utility Rate Database). Demand charges represent as
much as 30% of a typical electricity bill for commercial and industrial buildings, but may constitute a much higher percent-
age in an EB charging station.
There are currently two general types of all electric transit buses in the U.S. market. One type uses 200300 kWh lithium
iron phosphate batteries and has a range of up to 155 miles. After the battery is depleted, the EB must be taken off the route
to recharge, usually with a 4050 kW charger and taking up to 5 h to complete. The second type of EB employs lithium tita-
nate batteries with a much smaller energy capacity (5572 kWh) and has a range of up to 30 miles. Due to the unique chem-
istry and relatively compact battery design, these EBs can tolerate repeated charging at high power, which in turn will
facilitate in-route charging. The typical recharge time for a lithium titanate EB is 10 min or less, which can be easily accom-
modated into the bus schedule. Although the total range of a lithium titanate EB is lower than the lithium iron phosphate, the
use of high speed, in-route charging can enable continuous operation of the lithium titanate EBs without the need to remove
them from service for recharging purposes.
High speed charging is accomplished through the use of a 500 kW DC fast charger that must be installed somewhere
along the bus route, enabling the bus to recharge during a typical layover (Proterra, Charging technologies). While this
approach facilitates continuous operation, the operational costs are significantly increased by electricity demand charges.
These demand charges are associated with the high power requirement of the in-route charging station, and can represent
a large portion of the total electricity costs. In the case of StarMetro, a City of Tallahassee transit agency which operates a
fleet of five lithium titanate EBs, it was reported that the demand charges consist of 75.2 8.6% of the total electricity bill
at their fast charger (StarMetro). Since the electricity cost of EBs plays a significant role in their economic viability, the
demand charge-induced fuel cost is a major road block in lithium titanate electric bus adoption.
Some strategies are proposed to mitigate the demand charges, including increasing electric bus efficiency, employing
energy transfer technology, using time-of-use (TOU) pricing, or temporarily suspending demand charges (Gallo et al.,
2014). You et al. propose a method based on EB battery switching strategy where depleted batteries are replaced with
charged ones at battery switching stations and the battery charging is scheduled so that the electricity cost and battery
degradation is minimized (You et al., 2015). Ding et al. consider an energy storage system which has the potential to reduce
the network integration cost for fast chargers and to reduce the charging cost of EBs via electricity price arbitrage (Ding et al.,
2015). They formulate and solve the problem of optimal charging/discharging of the energy storage system and the coordi-
nated charging of EBs with a goal of minimizing the total cost of investment and total charging cost. However, these strate-
gies may not be economically viable due to the added cost of the stand-by batteries or energy storage systems. In this paper,
we investigate a strategy to mitigate the demand charge by optimizing the charging schedules. Current research on electric
vehicle charging scheduling is mostly focused on consumer electric vehicles, with goals ranging from filling the overnight
electricity demand valley, minimizing power losses and improving voltage profile, to maximizing the total amount of energy
that can be delivered to the EVs over a period while satisfying the power network constraints (Gan et al., 2013; Deilami et al.,
2011; Iversen et al., 2014; Richardson et al., 2012; Karbasioun et al., 2014). In comparison to the consumer electric vehicles,
the charging scheduling of EBs is a unique problem due to its strict driving schedule and fixed routes. Paul et al. propose a
method to determine the optimal charging schedule of EBs operated in a city bus route in Japan (Paul and Yamada, 2014).
The proposed algorithm reduces the fuel cost by maximizing the travel distance of the EBs between charging events. How-
ever, this simulation is based on a Japanese electricity rate structure and does not take demand charges into consideration.
To the best of our knowledge, there is no study that systematically investigates methods to mitigate EB demand charges
through charging scheduling. In this paper we conduct a case study on the Tallahhassee EB Fleet containing five EBs and
one fast charger. We propose a realistic strategy to minimize demand charge while maintaining the current route and
scheduling requirements. Specifically, we perform numerical analyses to explore charging decision making strategies based
on the battery state of charge thresholds. The approach does not require any EB technology breakthrough or policy change,
and can be adopted at no additional costs to the transit agencies. This paper also describes the EB fleet size impact on the fuel
cost.
The rest of the paper is organized as follows: Section 2 describes the electric buses, the fast charging station, and oper-
ation route and schedule; Section 3 is the method section which describes an electric bus energy consumption model and
explains the electric bus charging strategy and demand charge modeling; Section 4 discusses the optimization results;
and Section 5 compares the economics of different fleet sizes.

2. Case description

2.1. Electric bus and fast charging station

The StarMetro bus fleet in Tallahassee includes five all-battery electric EcoRide BE-35 electric buses (EBs), manufactured
by Proterra. The EB employs a 72 kWh lithium titanate battery energy storage system and a 150 kW electric motor. The bat-
tery uses lithium titanate spinel oxide instead of graphite as anode materials. It has a lower energy density of 6070 Wh/kg
388 N. Qin et al. / Transportation Research Part A 94 (2016) 386396

in comparison to lithium phosphate battery of 80110 Wh/kg which is more commonly used by other EB manufactures
(Burke and Marshall, 2009). However, due to the higher operating potential of the anode, the solid electrolyte interphase
(SEI) layer is not formed, enabling the battery to accept much higher charge and discharge rates than other Li-ion battery
chemistries.
One 500 kW fast charging station was installed in a bus route to realize the uninterrupted service. When an EB
approaches, the station recognizes the bus and takes over control for the last 3040 feet, propelling the bus under the cate-
nary arm, and guiding it into its docking position. Charging begins when the driver engages the parking brake. The entire
charging process does not require drivers engagement.

2.2. Route and schedule description

The StarMetro currently operates all five EBs on a single route called the Canopy route. This route is illustrated in Fig. 1. It
is 6.5 miles long, with a fast charger (FC) located 4.1 miles from the west terminus and 2.4 miles from the east terminus. The
EBs leave the StarMetro headquarters 0.7 miles west of the west terminus in the morning, and return to it after the full day of
service in the evening. The bus service starts and ends at the FC from 6:15 AM to 7:35 PM with a 25-min departure interval.
The loop sequence is FC-west terminus-FC-east terminus-FC. The FC is located in a central plaza where passengers make con-
nections, bus drivers take short breaks or change shifts. All the buses operate continuously with no interruption during the
day.

3. Methods

3.1. Modeling of electric bus energy consumption

The scheduling model assumes the EB fuel efficiency (FE) to be a constant value during operation. Even though the empir-
ical FE of this route can be obtained from StarMetro EB operation data, it is limited to one bus type and one route. Therefore,
we employed an energy consumption model which gives flexibility to simulate different EB models and routes. The EB
energy consumption was modeled using a vehicle longitudinal dynamics method similar to the ones described in previous
literature (Filippo et al., 2014). This approach follows a certain velocity cycle at the wheels and back calculates the energy
consumption (Rios et al., 2014). The key for an accurate simulation is to choose a driving cycle that closely represent the real
world situation. According to the U.S. Environment Protection Agency (EPA) protocol, a transit bus route can be represented
by either the Manhattan cycle or the Orange County cycle, representing an urban or a suburban route, respectively (U.S.
Environmental Protection Agency, 2007). The following formula is used to determine the appropriate driving cycle:

Cycle Index 0:013  Percent Idle 6:025=Av erage Speed 1

If the Cycle Index is greater or equal to 1, the Manhattan cycle should be used. If it is less than 1, the Orange County cycle
should be used. Based on the information provided by StarMetro, the Percent Idle for electric buses on the Canopy route
is 44%, and the Av erage Speed is 8.8 mph. The Cycle Index was estimated to be 1.25. Thus the Manhattan cycle is used
throughout the simulation in this study. The simulation procedures are described in the Supplemental Materials. Constants
used in the model such as bus front area, weight, gear ratio, and regenerative braking efficiency are obtained from Proterra.
Other constants such as accessory load, rolling resistance coefficient and drag coefficient are estimated based on several ref-
erences (Chen, 2010; Zoroofi, 2008; Nam and Giannelli, 2005). These values are listed in Table S1 in the Supplemental
Materials.
We validated our model by comparing the simulated fuel efficiency with actual operational data provided by StarMetro.
The fuel efficiency predicted by this model is 2.69 kWh/mile while the operation data from August 2013 to March 2014
showed an average FE of 2.57 0.35 kWh/mile. It can be observed that the simulation result is within 5% of the real world
data. Hence, our model is appropriate for the numerical analysis. Note that even though more sophisticated energy con-
sumption models exist such as the forward method and the models provided by commercial simulation tools such as PSAT,
ADVISOR and AUTONOMIE (Gao et al., 2007; PSAT; Markel et al., 2002; Kim et al., 2012), this model may be easier to be
adopted by other transit agencies or researchers while still providing sufficient approximations.

0.7 miles 4.1 miles 2.4 miles

West Fast East

Headquarters Terminus Charger Terminus
Fig. 1. Illustration of the Canopy route.
 N. Qin et al. / Transportation Research Part A 94 (2016) 386396 389

3.2. Electric bus charging strategy and demand charge modeling

The charging strategy proposed in this study is a decision making process in which an EB only charges when its state of
charge (SoC) is below a certain charging threshold (CT) at the fast charger (FC). Within the decision making process, it is
assumed that all EBs used the same CT, as this policy is adopted in real-world operation. In order to compute the charging
threshold that minimizes monthly electricity cost, we formulated the following optimization problem. Table 1 shows a list of
notations used in the optimization. The value of the parameters are listed in Table 2.
First, we assume that the bus schedules are the same every day, and the monthly peak demand equals to PMAX
x . The
monthly fuel cost C x can then be computed according to the following equation:

C x Dtotal HEB Rbase PMAX

x Rdemand 2

where the P MAX

x Rdemand is the demand charge component. Since the Dtotal and HEB are independent of charging thresholds, the
only parameter that could influence the C x by adjusting charging threshold is the peak demand P MAX
x . To be more precise, the
P MAX
x can be represented as:
(P R sn1 )
i sn Pi;t;x dt
PMAX
x max ; n 2 N; i 2 I; 0 6 x 6 0:95 3
n Td

in which sn and sn1 are the time stamps at the beginning and end of a demand period, namely sn s1 n  1T d , and
P R sn1
Pi;t;x dt
s1 0. The term i sn
Td
calculates the power demand contributed by all the EBs that are charged within the specified
demand period. Hence, the problem can be formulated as the following optimization function:
n o
min PMAX
x 4
06x60:95

We assume all the bus charging happens at the fast charger. Since only one fast charger is available, in the case when
more than one EB decide to charge at the same time, we apply the first in/first out (FIFO) queuing policy (i.e., the bus which
comes earlier at the FC will be charged first and the second charging event follows immediately afterwards). In order to solve
the optimization problem in Eq. (4), we propose the following algorithm summarized in Fig. 2. When an EB i passes the FC
the jth time, the model computes the remaining energy Ei;t and SoC X i;t and compares X i;t with the charging threshold x. If the
SoC is smaller than the CT, the EB is charged to 95% SoC. In this case, the model computes the charging start time stamp,
charging duration, end time stamp, and charging power (sSi;j , Li;j , sEi;j , and Pi;t;x ). If the SoC is larger than the CT, the model
checks if the EB can complete the next leg of the bus route and return to the FC with an SoC above 5%. If yes, the EB will
skip the charging. Otherwise, the EB charges to 95% SoC.

Table 1
Notations in the optimization problem.

Parameters
Td Demand period duration (in h) used by utility companies
HEB Fuel efficiency of an EB
T loop The travel time to finish one loop in a given route
T interv al The departure interval of EBs
T align The alignment time
Emax Battery capacity of an EB
Rbase Electricity cost per kWh
Rdemand Demand charge cost per kW
Dtotal Total distance traveled by an EB fleet in a month
Sets
I EB identifier I f1; 2; . . . ig n o
N Set of demand periods in a day, N 1; 2; . . . 24
Td

Variables
x Charging thresholds (decision variable), 0 6 x 6 0:95
Di;t Traveled distance of EB i at time t since the last charging event
Ei;t Remaining energy in the battery of EB i at time t Ei;t 0:95Emax  HEB Di;t
Ei;t
X i;t SoC of EB i at time t X i;t Emax
sSi;j Time stamp of the start of a charging event of EB i the jth time it passes a FC j 2 Z
Li;j Duration of a charging event of EB i the jth time it passes a FC
sEi;j Time stamp of the end of a charging event of EB i the jth time it passes a FC sEi;j sSi;j Li;j
P i;t;x Charging power of EB i at time t using a CT x
P MAX
x Peak demand using a charging threshold x
Cx Monthly electricity cost of the EB fleet using x
C permile Cost per mile C permile DCtotal
x
390 N. Qin et al. / Transportation Research Part A 94 (2016) 386396

Table 2
Parameter values used in the EB charging model.

Parameter Value Explanation

Td 0.5 h City of Tallahassee demand period duration
HEB 2.69 kWh/mile Based on the energy consumption model described in Section 3.1
T loop 114 min Based on the Canopy route described in Section 2.2
T interv al 25 min Based on the Canopy route schedule
Emax 72 kWh Based on Proterra EcoRide BE-35
Rbase $0.0636/kWh Based on Tallahassee utility rate as of November 2014
Rdemand $12.72/kW Based on Tallahassee utility rate as of November 2014
Dtotal 12,360 milesa Based on Canopy route and 5 EBs
a
Note: In order to serve from 6:15 AM to 7:35 PM, three EBs need to complete 6 loops while two will complete 5 loops. The daily service is assumed to be
the same for weekends and weekdays.

Event:
Compute:
EB i passes the FC
Ei,t and Xi,t
the jth time

yes
If:
Xi,t < x

Event: no
Charge battery to
95% SoC
If:
no Xi,t is enough for
the EB to return to
the fast charger
Compute: with > 5% SOC
Si,j , Li,j , Ei,j
and Pi,t,x yes

Event:
Pass the fast
charger without
charging

Fig. 2. Charging strategy for computing the optimal charging threshold

Fig. 3 presents the power curve during a charging event, as provided by Proterra. The amount of transferred energy during
the charge was determined by integrating the area under the power curve, also shown in the figure. Due to the character-
istics of lithium titanate batteries, the shape of the power curve is not effected by the starting SoC, so Fig. 3 can be used to
represent all charging events. In the model, the amount of energy transferred is calculated as 0:95Emax  Ei;t , and the charging
duration Li;j can be computed from the slop of the transferred energy curve. The end time stamp of a charging event sEi;j is
equal to sSi;j Li;j . When sSi;j 6 t 6 sEi;j , the P i;t;x is obtained from the power curve in Fig. 3, otherwise, P i;t;x 0. Since the charg-
ing events are discrete, if a charging event spans across two demand periods, the charging power is allocated to two periods.
The following paragraphs discusses additional assumptions and constrains used in the optimization strategy. An align-
ment time T align was incorporated in each loop to cover any charging events. T align is determined by T loop and T interv al by
the following equation:

T align jIjT interv al  T loop 5

where jIj is the cardinality of I. In the case of 5 EBs with 25 min departure interval on a 114 min loop, the T align is 11 min. In
transit bus scheduling, the typical alignment time to allow drivers to take a break is 1015 min. So the 11 min used in our
model is realistic. If the charging duration is shorter than T align , the bus will remain at the FC at the end of a loop until T align is
up. If the charging duration is longer than 11 min, the extra time is compensated by reducing the T align in the following loop.
 N. Qin et al. / Transportation Research Part A 94 (2016) 386396 391

450
Power Transferred Energy
400

350

Transferred Energy (kWh)

Charging Power (kW)
300

250
40
200
0.95Emax-Ei,t 30
150
20
100

50 10
Li,j
0 0
(Si,j) (Si,j+1) (Ei,j)
Time (Min)

Fig. 3. Charging power profile of a Proterra bus at a 500 kW fast charger.

The models also assumes that each EB checks its SoC at the fast charger before beginning the morning service. If it is above
the selected CT, the EB starts the service without charging. If it is below the CT, the EB charges to 95% SoC and then starts the
service. At the end of the day, the EBs always charge up to 95% before returning to the headquarters. The rationale for requir-
ing the EBs to charge at the end of the day is twofold: (1) to ensure the EBs start off at the same SoC everyday so the sim-
ulation of one day is representative of all the service days and (2) to make sure the EBs have enough energy to return to the
headquarters and travel back to the FC the next day without the need to charge at headquarters. In real world operation,
charging stations with a lower power rating could be installed at headquarters for overnight charging.

4. Optimization result and discussion

We searched for the optimal CT by performing an exhaustive enumeration covering the full range CT of 095%, as shown
in Fig. 4. The optimal peak demand of 124 kW is achieved with CT of 6064%, while the highest peak demand of 213 kW is

Fig. 4. Peak demand and demand charge at different charging thresholds for 5 EBs.
392 N. Qin et al. / Transportation Research Part A 94 (2016) 386396

resulted from CT of 028%. Since the demand charge is proportional to the peak demand, the minimum and maximum
demand charges per month are calculated to be $1586 and $2703, respectively, also shown in Fig. 4. Assuming a typical
12-year transit bus service life, the optimal demand charge could save $160,848 compared to the highest demand charge
scenario.
The optimal demand charge composes 42.9% of the total electricity cost according to Eq. (2), significantly lower than the
real world 75.2 8.6% recorded by StarMetro. The StarMetro fleet had a much lower average total traveled miles of 4305 per
month than the 12,360 miles used in the model, due to lack of trained drivers, mechanical issues, scheduling issues, etc.
Despite the lower usage, the EB fleet had an average peak demand of 167 kW which is higher than the 124 kW of the optimal
charging strategy. Data on the charging behavior is not available. However we could estimate their charging behavior to be
similar to the 028% CT charging strategy as the drivers were trained to maximize the distance between charges. Even
though the real world data is not directly comparable with the model data due to the discrepancy between the total
mileages, it sheds some lights on how the demand charge can become a dominate portion of the fuel costs and thus need
to be minimized in order to improve the economic viability of the EBs.
The impact of CT on the daily demand profile is revealed in Fig. 5, in which CT of 20%, 40%, 60%, and 80% curves represent
CT of 028%, 2946%, 6064%, and 7795% scenarios, respectively. This representation will be used throughout the rest of the
P R tn1
Pi;t;x dt
i tn
paper. Each point on the curve represents a power reading for a demand period. Its value is calculated from Td
in Eq.
(3). The 20% and 40% profiles exhibit alternating deep valleys and high peaks. In contrast, the 60% and 80% curves show
smoother profiles which means that the total energy drawn from the grid is more evenly distributed in all the half hour win-
dows. As a result, lower peaks can be achieved.
Fig. 6 reveals the impact of CT on individual charging events. Each bar represents one charging event and each color rep-
resent one EB. The left and right borders of a bar represent the start and end time stamps, sSi;j and sEi;j , of a charging event, and
the width of the bar represents the charging length Li;j . The evolution of SoC for a single EB is shown in Fig. 7. In general, the
higher CT results in larger number of charging events and shorter Li;j . The charging events in the 40% and 20% CT scenarios
form clusters, which contributes to the big peaks and valleys in their demand profiles in Fig. 5. In contrast, the charging
events in the 80% and 60% CT scenarios are evenly distributed, which result in smoother demand profiles. The 80% CT results
in 61 charging events with Li;j ranging from 2.7 to 5 min. There are two overlapping events at 7:27 AM and 7:52 AM with an
overlap of 3 min each time, shown in a magnified region in Fig. 6. When overlap happens, the model allows the EB with a
later start time to wait until the event with an earlier start time is finished. This adjustment does not affect the subsequent
loop as both charging events can occur within the alignment time. The 60% CT resulted in 33 charging events, with Li;j rang-
ing from 5 to 6.6 min. There are no overlapping events and the minimum gap between two consecutive charging events is
14 min. For the 40% CT, there are a total of 23 charging events with Li;j of 6.610.5 min. No overlap happens and the mini-
mum gap between charging is 14.6 min. The maximum charging time 10.5 min is close to the alignment time 11 min, which
in real world may result in some stress of the scheduling as the EBs have little flexibility to realign the schedule. For the 20%
CT, the number of charging events is 23 and the Li;j ranging from 2.7 to 11.1 min. The 11.1 min Li;j slightly exceeds the

Fig. 5. Daily demand profiles using different charging thresholds.

 N. Qin et al. / Transportation Research Part A 94 (2016) 386396 393

Fig. 6. Charging events under different charging thresholds

100
CT=80%

0
100
CT=60%
SoC (%)

0
100
CT=40%

0
100
CT=20%

0
6:00 AM 9:00 AM 12:00 PM 3:00 PM 6:00 PM 9:00 PM
Time

Fig. 7. Evolution of SoC for a single EB under different charging thresholds

alignment time of 11 min. As mentioned previously, the model allows the EBs to use the alignment time in the subsequent
loop to make up the time difference. One overlapping event happens at 6:45 PM under this CT.
The detailed analysis of Fig. 6 indicates that the 60% CT not only results in the lowest peak demand, but also results
in suitable charging length, comfortable minimum gap, and no overlapping charging events. Thus 60% CT can be easily
implemented in the real world operation. The CT based charging strategy can also be translated to a scheduling table
based on loop numbers and directions, as shown in Table 3. For example, when using the 80% CT, an EB needs to charge
every time it passes the FC. When using the 60% CT, the EB should be charged at the FC when traveling westbound. The
scheduling table can be more easily adopted by drivers, and will produce the same demand charge as the CT based
strategy.
It is worth mentioning that even the optimal charging strategy results in localized peaks in the demand profile. In order to
further shave the peaks, an energy storage system such as a battery can be employed. For instance, the peak demand of the
80% CT scenario is defined by a single peak of 134 kW at 7:30 AM (Fig. 5). Assuming a stationary lithium battery is available
to supply power in conjunction with the grid, a 15 kWh battery with a 80% discharge cycle is able to shave the 134 peak to
394 N. Qin et al. / Transportation Research Part A 94 (2016) 386396

Table 3
Daily charging pattern of an EB with different charging strategies.

CT L1W L1E L2W L2E L3W L3E L4W L4E L5W L5E L6W L6E End
80 1 1 1 1 1 1 1 1 1 1 1 1 1
70 1 1 0 1 0 1 0 1 0 1 0 1 1
60 1 0 1 0 1 0 1 0 1 0 1 0 1
50 0 1 0 1 0 1 0 1 0 1 0 1 1
40 0 1 0 0 1 0 0 1 0 0 1 0 1
20 0 0 1 0 0 1 0 0 1 0 0 1 1

Note: L1W represents when EB traveling westbound in the loop 1, and L1E represents when EB traveling eastbound. Number 1 indicates charging events
and 0 indicates no charging events.

110 kW which is the next highest peak. This simple action can save the demand charge by $44,000 (Rdemand  24  144) for a
12 years operation period. Based on the current lithium battery price of $268/kWh (U.S. Department of Energy), the net sav-
ing is $39,980. For situations where peak demand frequently appears throughout the day (e.g. the 60% CT scenario), a more
sophisticated mathematical model is needed to simulate the size and the charging/discharging of the battery between the
electric bus charging events. The topic of using backup energy source to reduce demand charge will be investigated in
the future studies.

5. Economic analysis of the impact of EB fleet sizes

The fast charger is an essential component in the lithium titanate based electric bus system, and its capital and installa-
tion costs represent a significant investment. For example, the capital and installation cost of StarMetros 500 kW FC was
$1.165 million. As a result, the early adoption of EBs will likely start with only one FC and a fleet of several EBs. It is necessary
to investigate the maximum number of EBs that can be accommodated by one fast charger, and how fleet sizes impact the
demand charges and fuel costs. The Canopy route was used to simulate a series of fleet sizes. The departure intervals T interv al
and alignment time T align are adjusted to fit all EBs into the schedule, according to Eq. (5). A smaller T interv al represents cases of
a busier route with a larger ridership while a bigger T interv al simulates a smaller ridership. The T interv al and T align for fleet size of
414 EBs are shown in Table S2 in the Supplemental Materials. Ideally, the maximum fleet size that one FC can accommodate
is the one that results in no overlapping charging events. The simulation results, shown in Table 4, presents the numbers of
overlaps and minimal gaps between two consecutive charging events. It can be seen that CT of 20% and 80% generally tend to
result in more overlaps than CT of 40% and 60%. A larger fleet size also results in smaller minimum gap and greater number of
overlapping events. Theoretically, the maximum fleet size that results in no overlaps is 12 when CT of 40% is used, and 13
when CT of 60% is used.
We also modeled the optimal CT that results in the lowest demand charges for each fleet size. The CT of 60% results in
the lowest peak demand for every fleet sizes, as shown in Fig. 8. Therefore, the economic analysis of fleet sizes is con-
ducted by using demand charges calculated from CT of 60%. The cost per mile C permile is used to compare the cost efficiency
of different fleet sizes and can be computed as C permile DC x . Fig. 9 shows the C permile as a function of the fleet sizes. The
total

C permile decreases as the fleet size increases with the exception of the case of 4-EB. A larger fleet generally means more total
miles driven which diminishes the percentage of the demand charges. In the case of 4-EBs, the demand profile is flat with
no peaks, as shown in the insert of Fig. 9, thus results in a low demand charge percentage despite of the low Dtotal . Our
simulation reveals that when real world factors such as bus schedule, route, and how charging events aligns with utility
company billing window were taken into consideration, the impact of fleet sizes on the cost efficiency should be analyzed
on a case by case basis.

Table 4
Number of overlapping events and minimum gaps for fleet size of 414 EBs using different charging strategies.

Fleet size Number of overlapping events/minimum gap (min)

80% 60% 40% 20%
4 0/8.0 0/18.0 0/21.5 1/NA
5 2/NA 0/14.0 0/14.6 1/NA
6 26/NA 0/9.0 0/10.5 4/NA
8 43/NA 0/2.0 0/5.5 5/NA
10 4/NA 0/2.0 0/2.5 3/NA
11 60/NA 1/NA 0/1.0 8/NA
12 55/NA 1/NA 0/0.0 18/NA
13 73/NA 0/0.0 12/NA 43/NA
14 82/NA 1/NA 13/NA 41/NA
 N. Qin et al. / Transportation Research Part A 94 (2016) 386396 395

Fig. 8. Peak demand as a function of fleet sizes and charging strategies.

Fig. 9. Cost per mile of different fleet sizes using CT of 60%; insert: their demand profiles.

6. Conclusion

Electric buses using lithium titanate batteries are a very attractive option in the alternative fuel transit bus arena with
advantages such as pure electricity operation, zero tail pipe emissions, and the capability for uninterrupted in-route charg-
ing. The high demand charge induced fuel cost sets an obstacle in its wide deployment. This paper investigated the factors
that impact demand charges such as charging strategies and fleet sizes, by using an electric bus energy consumption model
coupled with a charging strategy model. It was demonstrated that simply adopting an appropriate charging strategy could
lead to significant savings in demand charges, without occurring additional cost of backup batteries or the need for policy
changes. The model reveals that more frequent charging may lead to a reduction in demand charges, contradicting to the
popular practice where drivers maximize distance between charges. This finding might inspire transit agencies to modify
the EB driver training programs in order to achieve most savings. In cities where a higher demand charge rate or a smaller
demand period duration (e.g. 15 min) are used, the savings could be even more significant. It was also found that the cost per
mile can be reduced when the fleet size and charging strategies were both optimized.
396 N. Qin et al. / Transportation Research Part A 94 (2016) 386396

The EB energy consumption model and the charging strategy model described in this paper can be used to study the
scheduling and demand charges of electric transit bus systems with different driving cycles and routes. In addition, it can
be used to explore methods such as battery energy storage systems to reduce demand charges, as well as conduct compar-
ative studies of EB with different battery technologies. In the future work, we also aim to search the optimal charging strat-
egy where varied charging thresholds of each bus and each node are allowed.

Acknowledgment

This work was supported by an award to the Florida Solar Energy Center, University of Central Florida, as part of Grant No.
DTRT13-G-UTC51 from the U.S. Department of Transportations University Transportation Centers Program. The authors
would like to thank Ralph Wilder from City of Tallahassee, StarMetro and Tom Gillman from City of Tallahassee, Utility
Customer Operations for providing electric bus operation data and electricity usage data. The authors would also like to
thank Sam Shartzer from Proterra for providing the electric bus charging power profiles.

Appendix A. Supplementary material

Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.tra.
2016.09.014.

References

Burke, A., Marshall, M., 2009. Performance characteristics of lithium-ion batteries of various chemistries for plug-in hybrid vehicles. In: EVS24 International
Battery, Hybrid and Fuel Cell Electric Vehicle Symposium. Stavanger, Norway, May 1316.
Bradley M.J., Associates LLC, Comparison of Modern CNG, Diesel and Diesel Hybrid-Electric Transit Buses: Efficiency and Environmental Performance,
Technical Report, Concord, MA.
Chen, D., 2010. The Effects of Drive Cycle Accessory Load and Degree of Hybridization on Fuel Economy and Emissions for Hybrid Electric Buses Masters
thesis. University of Alabama, Tuscaloosa.
Deilami, S., Masoum, A.S., Moses, P.S., Masoum, M.A.S., 2011. Real-time coordination of plug-in electric vehicle charging in smart grids to minimize power
losses and improve voltage profile. IEEE Trans. Smart Grid 2 (3), 456467.
Dickens, M., Neff, J., 2014. 2014 Public Transportation Fact Book Appendix B: Transit Agency and Urbanized Area Operating Statistics. American Public
Transportation Association, Washington, DC.
Ding, H., Hu, Z., Song, Y., 2015. Value of the energy storage system in an electric bus fast charging station. Appl. Energy 157, 630639.
Filippo, G.D., Marano, V., Sioshansi, R., 2014. Simulation of an electric transportation system at the Ohio state university. Appl. Energy 113, 16861691.
Gallo, J., Bloch-Rubin, T., Tomic, J., 2014. Peak Demand Charges and Electric Transit Buses, Technical Report, Calstart, October.
Gan, L., Topcu, U., Low, S.H., 2013. Optimal decentralized protocol for electric vehicle charging. IEEE Trans. Power Syst. 28 (2), 940951.
Gao, D.W., Mi, C., Emadi, A., 2007. Modeling and simulation of electric and hybrid vehicles. Proc. IEEE 95 (4), 729745.
Iversen, E.B., Morales, J.M., Madsen, H., 2014. Optimal charging of an electric vehicle using a markov decision process. Appl. Energy 123, 112.
Karbasioun, M., Lambadaris, I., Shaikhet, G., Kranakis, E., 2014. Optimal charging strategies for electrical vehicles under real time pricing. In: 2014 IEEE
International Conference on Smart Grid Communications, Venice, November, pp. 746751.
Kim, N., Kwon, J., Rousseau, A., 2012. Comparison of powertrain configuration options for plug-in HEVS from a fuel economy perspective. In: SAE 2012
World Congress & Exhibition.
Lajunen, A., 2014. Energy consumption and cost-benefit analysis of hybrid and electric city buses. Transport. Res. Part C: Emerg. Technol. 38, 115.
Lajunen, A., Lipman, T., 2016. Lifecycle cost assessment and carbon dioxide emissions of diesel, natural gas, hybrid electric, fuel cell hybrid and electric
transit buses. Energy 106, 329342.
Markel, T., Brooker, A., Hendricks, T., Johnson, V., Kelly, K., Kramer, B., OKeefe, M., Sprik, S., Wipke, K., 2002. Advisor: a systems analysis tool for advanced
vehicle modeling. J. Power Sour. 110 (2), 255266.
Nam, E.K., Giannelli, R., 2005. Fuel Consumption Modeling of Conventional and Advanced Technology Vehicles in the Physical Emission Rate Estimator
(PERE) Draft, Technical Report. U.S. Environmental Protection Agency.
Paul, T., Yamada, H., 2014. Operation and charging scheduling of electric buses in a city bus route network. In: IEEE 17th International Conference on
Intelligent Transportation Systems (ITSC), pp. 27802786.
Proterra, Charging Technologies <http://www.proterra.com/wp-content/uploads/2015/05/Tearsheets_ChargingTechnologies.pdf> (accessed: 2016-08-09).
PSAT (Powertrain System Analysis Toolkit) <http://www.transportation.anl.gov/modeling_simulation/PSAT/> (accessed: 2015-09-10).
Richardson, P., Flynn, D., Keane, A., 2012. Optimal charging of electric vehicles in low-voltage distribution systems. IEEE Trans. Power Syst. 27 (1), 268279.
Rios, J., Sauras-Perez, P., Gil, A., Lorico, A., Taiber, J., Pisu, P., 2014. Battery Electric Bus Simulator - A Tool for Energy Consumption Analysis. SAE Technical
Paper 2014-01-2435.
StarMetro, The demand charge was calculated from monthly electricity comsumption at the StarMetro fast charging location from January 2014 to July
2014. The cost is based on the electricity rate as of Novermber 2014.
U.S. Department of Energy <http://energy.gov/sites/prod/files/2016/06/f32/vto000_cooper_2016_o_web.pdf> (accessed: 2016-08-17).
U.S. Energy Information Adminstration <http://www.eia.gov/petroleum/gasdiesel/> (accessed: 2015-09-10).
U.S. Environmental Protection Agency, 2007. Smartway Fuel Efficiency Test Protocol for Medium and Heavy Duty Vehicles, Technical Report.
U.S. Utility Rate Database <http://en.openei.org/wiki/Utility_Rate_Database> (accessed: 2015-09-10).
You, P., Yang, Z., Zhang, Y., Low, S.H., Sun, Y., 2015. Optimal charging schedule for a battery switching station serving electric buses. IEEE Trans. Power Syst.
PP (99), 111.
Zhou, B., Wu, Y., Zhou, B., Wang, R., Ke, W., Zhang, S., Hao, J., 2016. Real-world performance of battery electric buses and their life-cycle benefits with respect
to energy consumption and carbon dioxide emissions. Energy 96, 603613.
Zoroofi, S., 2008. Modeling and Simulation of Vehicular Power Systems Masters thesis. Chalmers University of Technology, Goteborg, Sweden.