Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

This article has been accepted for inclusion in a future issue of this journal.

Content is final as presented, with the exception of pagination.

IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS 1

Evaluating Electric Bus Operation for a Real-World


BRT Public Transportation Using
Simulation Optimization
Mariana Teixeira Sebastiani, Ricardo Lüders, and Keiko Verônica Ono Fonseca, Member, IEEE

Abstract—Growing concerns with environmental issues have tric Vehicles (PEV) and more specifically by Battery Electric
resulted in alternatives for urban mobility and public transporta- Vehicles (BEV) in a Bus Rapid Transit (BRT) system [6].
tion. Among the available options, battery-powered electric buses They combine the advantages of low cost and flexibility for
have been considered in terms of flexibility, sustainability, and
emission of pollutants. These buses have a plug-in recharge system defining routes and integrating different bus manufacturers in
that allows them to run without any extra power supply. However, a BRT with clean energy. However, the operation of this new
battery recharging during operation is necessary, and the number system is quite challenging as it depends on a well-planned
and location of charging stations are important issues for system schedule, and on charging stations to overcome the low bus
operation. This paper presents a discrete event simulation that operational autonomy [7]. For instance, other initiatives have
evaluates bus energy consumption using a mathematical model
that takes into account different load and friction forces. An been made with trolleybus, but they are not flexible enough
optimization strategy that utilizes a bi-objective genetic algorithm regarding energy cut-off or road blocking due to traffic or
is then associated with simulation (in a simulation optimization accidents. In addition, they still require permanent aerial cables
approach) to minimize both the number of charging stations to provide energy [8]. Another option is represented by battery
and the average extra time stopped in the station to recharge. swap stations [7], [9] in which vehicles replace low charge
Information, including passenger demand, bus speed, distances,
and route elevation profiles, has been obtained for Curitiba public batteries by pre-charged ones. As our focus is on the impact of
transportation. The parameters of buses’ mobility are based on charging time in the buses’ timetables during system operation,
real data acquired, filtered, and analyzed for six express routes the previous alternatives have not been considered here.
from raw data provided by the agency that controls public trans- Several works can be found in the literature concerning the
portation and urban development. Fast battery recharge with infrastructure of electric vehicles charging [2], [3], [10]. They
typical parameters is considered at charging stations. The results
show different arrangements for the number of stations, their present different approaches concerning estimates of energy
location, and delays in bus schedule, as well as the corresponding consumption and locations of charging stations. However, most
bus energy levels in different routes. of them focus on urban electric cars instead of public trans-
Index Terms—Transportation, electric vehicles, discrete-event portation. In [2], energy demand is estimated from population
systems, computer simulation, optimization, computational and economic activity indicators and a mixed-integer optimiza-
intelligence. tion model based on a classical covering problem is used to
locate charging stations. The same idea is found in [10], but
using simulation to estimate energy demand based on distances
I. I NTRODUCTION
traveled by vehicles that follow a probabilistic mobility model

N OWADAYS a new generation of electric urban vehicles


can be seen in big cities. They have advantage over fossil
fuel engines regarding emission and noise pollution [1]–[3]. At
between different urban areas. A linear integer programming
model is then used to determine location and capacity of charg-
ing stations. In [3], a genetic algorithm (GA)-based optimiza-
the same time, an increasing interest is emerging on current tion model minimizes missed trips that are subjected to budget
urban public transportation networks with clean energy [4], constraints. The energy consumption model is based on GPS
[5]. One promising alternative is represented by Plug-in Elec- historical data, considering that driving patterns for conven-
tional gasoline vehicles are maintained for electric ones. These
approaches can be useful for global energy estimation but fail
when more accurate energy measures should be provided [11].
There are also works taking into consideration only energy
Manuscript received June 8, 2015; revised September 29, 2015 and
December 18, 2015; accepted January 28, 2016. This work was supported by consumption estimation [4], [12], [13]. In general, they capture
CAPES. The Associate Editor for this paper was W.-H. Lin. complex details of electric bus technology. In [4], a dynamic
M. T. Sebastiani is with the Institute of Transportation Studies, University of model is built to evaluate energy consumption for sizing bat-
California at Irvine, Irvine, CA 92697-3600 USA (e-mail: marianat@uci.edu).
R. Lüders and K. V. O. Fonseca are with Federal University of Technology- teries to be carried by a bus. Several features are considered,
Paraná (UTFPR), 80230-901 Curitiba-PR, Brazil (e-mail: luders@utfpr.edu.br; such as bus design and interacting forces. A similar approach
keiko@utfpr.edu.br). is followed by [12] which also represents the interaction of
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. buses with local traffic by integrating the bus drive-cycles
Digital Object Identifier 10.1109/TITS.2016.2525800 within a standard urban traffic simulator. A very precise energy
1524-9050 © 2016 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 inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

2 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS

estimation can be found in [13], which utilizes a specialized TABLE I


E XPRESS B US ROUTES IN C URITIBA
simulator to represent complex technical specifications. Based
on different bus-drive patterns, represented by selected routes,
a detailed cost-benefit analysis is made for replacing conven-
tional diesel buses by electric ones.
We propose a simulation optimization approach for allocat-
ing charging stations for full-electric buses (BEV) in public
transportation that integrates several features of energy demand
estimation into a single simulation model such as elevation
profile of routes, stops at traffic lights, driver’s behavior,
and bus dynamics. This simulation is suitable to evaluate
the feasibility of the system operation. Moreover, a genetic
bi-objective optimization algorithm is applied to obtain the
allocation of charging stations. Our contribution is focused
on developing more accurate estimations of energy consump-
tion than other optimization models. Particularly, the proposed
simulation model includes route elevation profiles and inter-
action with traffic lights, which are pointed out by [11] as
the most important factors in energy estimation. In addition,
our optimization approach is formulated for public transporta-
tion in a stochastic framework which is quite different from
the deterministic optimization approaches described above. A
bi-objective optimization algorithm is used in a problem often
formulated and solved for a single objective. This approach
provides a new strategy for decision-making where conflicting
objectives can be evaluated together and a final decision can
be taken based on priorities established a posteriori. Although
Fig. 1. Curitiba’s express bus routes and stops in a map.
our proposal has been applied in a real-world BRT system
found in Curitiba, Brazil, it is general enough to be applied This problem can be formulated by (1) as a bi-objective
to other cases by adjusting topologies and capabilities of the optimization
transportation system.
This paper is organized as follows. Section II presents the min f (x) = [f1 (x), f2 (x)]
x∈{0,1}n
problem description of the BRT system of Curitiba and high- subject to gi ≥ 0, i = 1, 2, . . . , nb (1)
lights the main issues that should be considered in order to n
replace diesel powered buses by electric ones. The methodol- with x as a n-dimension binary vector, f1 (x) = i=1 xi as the
ogy is presented in Section III where the proposed simulation number of active charging stations, f2 (x) as the average extra
optimization approach is discussed in details. Results are shown stop time obtained by simulation, gi (x) as the battery level for
in Section IV, and concluding remarks in Section V. bus i, and nb = 142 as the total number of buses. The binary
vector x represents bus stops allowed to have a charging station
(n = 118) and constraints gi guarantee that no bus runs out of
energy.
II. P ROBLEM D ESCRIPTION
Although the introduction of BEV buses has brought en-
Curitiba is a city in South Brazil with a public transportation vironmental benefits (low noise and less emissions), they are
network based on a Bus Rapid Transit (BRT) system. It means still expensive. Hence, the transition requires careful planning
that routes are chosen to run in exclusive lanes using high of the transport system operation to avoid the decrease of
capacity bi-articulated buses, which transport up to 250 passen- transported passengers or the increase of route completion time
gers per bus. Most of the buses in this network are currently (due to battery charging time compared to the actual diesel bus
powered by diesel. However, due to environmental concerns system) without going over budget. This scenario is evaluated
and sustainable principles, they should be gradually replaced by by simulation as discussed in Section III.
buses that use other sources of energy, like BEV buses. These This work considers six express bus routes listed in Table I
buses are powered by electric energy through internal batteries and shown in Fig. 1. Their corresponding labels are given by
that must be periodically recharged. In general, they have Table XII of the Appendix. Note that many routes share common
lower autonomy when compared to diesel buses. Therefore, roads and stops. Route 502/602 makes a circular path with each
the replacement of conventional buses by electric ones requires code representing one direction (clockwise and counterclock-
charging stations at intermediate route locations. The problem wise, respectively). These lines have been chosen because they
is to find solutions with good trade-off between the minimum represent the largest numbers of transported passengers. They
number of deployed charging stations and the extra stop time at give access to several other destinations through commuter sta-
bus stops (to charge batteries). tions using secondary routes operated by lower capacity buses.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

SEBASTIANI et al.: EVALUATING ELECTRIC BUS OPERATION FOR BRT USING SIMULATION OPTIMIZATION 3

Several assumptions are made: (i) charging stations can only The friction force F1 is given by (2). It depends on bus mass
be deployed at current bus stops; (ii) fast charging is considered m in kg and gravity force g = 9.8 m/s2
at bus stops; (iii) slow charging is made overnight for all buses;
(iv) uncertainty is characterized by random variables with given F1 = f · m · g (2)
probability distribution functions; (v) average bus speed is
obtained from real data and route elevation profiles are obtained with friction coefficient f = (0.0041 + 0.000041 · v · 2.24) ·
from third party applications for different route segments; Ch , bus speed v in m/s, and road friction coefficient Ch = 1.2.
(vi) regular and rush hours are considered during the day, The dragging force F2 is caused by air resistance against
with 60% and 100% of full bus occupancy (250 passengers), the bus’ movement and it depends on several shape design
respectively. properties. Essentially, it can be computed from bus frontal area
as in (3)
III. S IMULATION O PTIMIZATION F2 = 0.5 · ξ · Ca · A · v 2 (3)
The methodology adopted here for allocating charging sta-
tions is based on simulation optimization [14]. It has been cho- considering air density ξ = 1.2 kg/m3 , aerodynamic dragging
sen due to the nature of the problem: a stochastic optimization coefficient Ca = 1.17, and bus frontal area A = 10 m2 .
problem which performance involves a complex interaction be- Bus routes have different elevation profiles with possible
tween vehicle dynamics and random phenomena as the number climbing parts. In this case, an additional force F3 given by
of passengers and stops at traffic lights. In this case, simulation (4) is necessary to balance a slope of α degrees
includes as many details as necessary to capture the real-world
system behavior. F3 = m · g · sin(α). (4)
Discrete event simulation captures the relevant events that
The route elevation profiles have been obtained in a previous
affect energy consumption in a transportation network instead
analysis from altitude data of intermediate locations between
of providing particular results for standard drive-cycles. Tradi-
two stops. A weighted mean slope is then computed by consid-
tional methods of sizing battery packs for BEV are represented
ering intermediate distances.
by a combination of numerical modeling with physical testing
The resultant force F4 is necessary to drive the bus with a
of the target vehicle over legislative drive-cycles. These include
given acceleration a according to (5)
the Urban Dynamometer Driving Schedule (UDDS) within
North America, the JC08 Cycle within Japan, or Regulation F4 = m · a. (5)
101 within Europe. However, such drive-cycles fail to emulate
realistic driving conditions [15]. Precise energy consumption The total force F is then computed by (6) which generates a
models are important for vehicle design but they represent un- corresponding torque according to (7)
necessary computational burden for the analysis of operational
conditions in a transportation network. The discrete event sim- F = F1 + F2 + F3 + F4 (6)
ulation presented here models the basics of vehicle dynamics F ·r
combined with random phenomena, and considers express bus T = (7)
G
routes with shared resources such as lanes, terminals, stops and
charging stations. It is conceived for operations research, i.e., to with tire radius r = 0.5 m, transmission and gear ratio G =
investigate different operational plans considering randomness. 1 : 3 (first gear) and G = 1 : 1 (second gear) considering a
An evolutionary optimization algorithm variant of NSGA-II differential ratio of 8.83. It is assumed that every bus achieves
[16], [17] evaluates each simulated solution using a fitness mea- a final constant speed between two stops after an acceleration
sure and improves it until a satisfying solution is achieved. It is period of 10 s according to [19]. Therefore, the energy spent
recognized as one of the most prominent multiobjective evolu- between two stops is split into two parts: a pre-computed energy
tionary algorithms often used as reference, according to [18]. for bus acceleration plus the energy necessary to achieve the
The simulation requires a mathematical model for bus energy next stop with constant speed. This cycle is affected by the
consumption as well as for battery charging as discussed in driver’s efficiency under different driving conditions [21].
Section III-A. A discrete event simulation computes variables The corresponding total power is computed by (8)
of interest, including energy consumption for each bus in a
T · vm · 2 · π
round trip as well as charging time for different locations of P = (8)
charging stations (or solutions)—as discussed in Section III-B. 60000 · ηm · ηc
The optimization procedure is discussed in Section III-C. with angular engine speed vm and coefficients ηm and ηc to rep-
resent the engine’s and the bus driver’s efficiency, respectively.
A. Bus Energy Consumption and Battery Charging Models The angular engine speed vm is given by (9) and the
The mathematical model considered for computing the nec- engine’s efficiency is computed by ηm = −3 · 10−8 · vm 2
+
essary energy to run a bus is based on [19] and [20]. Basically, 0.0002 · vm + 0.638. In this work, the driver’s efficiency is
it is a dynamic model that takes into consideration the opposite considered to be constant at 92%, according to [19]
forces to the bus’ motion (friction and dragging forces) as well v · 60 · G
as ramp forces due to different route elevation profiles. vm = . (9)
2·π·r
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

4 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS

• Block 3: Proceed to the next stop until the last stop is


achieved.
• Block 4: Compute the energy spent between stops and
decrease battery level accordingly (discharge).
• Block 5: Compute boarding time at the current stop.
• Block 6: Is the current stop a charging station?
• Block 7: If so, charge battery during boarding time
(normal charge), increase battery level accordingly, and
evaluate energy necessary to the reach the next charging
station; else proceed to the next stop (block 3).
• Block 8: Is the current battery charge sufficient to achieve
the next charging station?
• Block 9: If so, proceed to the next stop (block 3); else
compute a non-null extra stop time for charging, increase
battery level accordingly (with extra charge) and proceed
to the next stop (block 3).
• Block 10: After simulating all buses for all routes, compute
Fig. 2. Typical battery charging curve. the average extra stop time over all extra time occurrences.
The simulation starts with entities (buses) being generated
TABLE II according to a pre-defined timetable (block 1). We assume that
P ROPERTIES OF O NE BATTERY PACK
a full battery charge is made overnight. Thus, the initial battery
level of a bus starting in a route is computed by decreasing the
energy spent between its garage and route from a full charged
(100%) battery (block 2). Distances between garages and routes
are presented in Table XVI of the Appendix.
A main loop is then established for each bus running between
two consecutive stops (block 3). Based on the average bus
speed and distance between stops, the corresponding spent
energy is then calculated according to Section III-A and the
The total energy spent during a time interval Δt = tf − ti is battery level is decreased accordingly (block 4). It takes into
then given by (10). This energy is decreased from the battery account the route elevation profile between stops, the number of
energy level after Δt stops at traffic lights, and the number of passengers according
to Section III-A. A dwell time is introduced for boarding of
tf
passengers following the real data shown in Table XV (block 5).
E= P dt. (10) If the current stop has a charging station (block 6), bat-
ti tery is charged during boarding time (block 7). Otherwise, a
new iteration is started with the next stop (block 3). Also in
A typical battery charging curve is given by Fig. 2 according
block 7, the minimum energy necessary to reach the next
to [22]. It can be divided into a linear part followed by an
active stop (with charging) is estimated to maintain at least
asymptotic function. The linear part occurs with constant elec-
20% of minimum battery level (block 7). This 20% level is
tric current and achieves 80% of the total energy level, which
set to take into account eventual stops at traffic lights and
takes 40% of the total charging time. The remaining charge is
intermediate stops without recharging. At the current stop,
accomplished using constant electric voltage with asymptotic
there is no information whether there will be stops at traffic
behavior. A fast battery charge can be done in 6 min according
lights ahead (it is a random variable). This safety margin of
to a bus manufacturer [23] as described in Fig. 2. The properties
20% was empirically determined through simulation. Values of
of one battery pack are given in Table II [24].
15% or less have yielded to run out of energy while values
of 25% (or more) have resulted in 10% (or more) residual
B. Simulation
battery level.
A computer simulation is built using MATLAB/SimEvents If the current battery charge is sufficient to reach the next
toolbox [25] to compute the battery’s energy level for each bus. charging station (block 8), a new iteration is started with the
MATLAB/SimEvents is a block diagram environment that im- next stop (block 3). Otherwise, an extra dwell time is introduced
plements common features of discrete event simulation [26]. The to charge the battery according to the minimum energy neces-
simulation model requires that charging stations have already sary to reach the next active stop (block 9) before starting a new
been assigned to bus stops. It is described by a sequence of ac- iteration (block 3). After all buses in all routes are simulated,
tions performed over entities (buses) according to the following. the average extra stop time is computed over the number of stop
• Block 1: For each route, generate a set of buses with a occurrences requiring extra charging time (block 10). This extra
corresponding itinerary. stop time is a fitness measure for optimization as discussed
• Block 2: For each bus in a route, assign its initial energy. in Section III-C.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

SEBASTIANI et al.: EVALUATING ELECTRIC BUS OPERATION FOR BRT USING SIMULATION OPTIMIZATION 5

TABLE III TABLE IV


S TRATEGY FOR E ACH P HASE OF THE O PTIMIZATION A LGORITHM PARAMETERS FOR THE B I -O BJECTIVE G ENETIC A LGORITHM

by a penalty method [28], [29]. In this case, the number of


C. Optimization active stations and extra time for battery charging have their
The optimization problem given by (1) is solved by a sim- values increased for solutions that run out of energy, making
ulation optimization approach. A metaheuristic optimization them less attractive when compared with feasible ones.
algorithm cycles between optimization and simulation until the
best solution is found for a specified number of optimization
IV. R ESULTS
cycles. This work utilizes the bi-objective optimization algo-
rithm of MATLAB Optimization toolbox, which is a controlled Several parameters are necessary to simulate variables of
elitist genetic algorithm variant of NSGA-II [27]. A controlled interest. These parameters are set for Curitiba’s case using data
elitist genetic algorithm favors solutions that can help diversity provided by URBS [30] (an agency that controls Curitiba’s
even if they have a lower fitness value. NSGA-II is a Pareto- public transportation). These data can be static as routes, stop
based optimization algorithm for which the general framework locations, traffic lights, among others, or dynamic as online bus
is given by Algorithm 1. In line 6, the population P is randomly position compiled from the BRT monitoring system.
initialized. A loop is then established in line 7 until a deter- This information is shown in Tables XIV–XVII of the
mined maximum number of generations is achieved within a Appendix. The number of buses per route in a day is given by
limited computational effort. In line 8, the fitness f (x) of all Table XIV. The average stop times for boarding or waiting at
individuals xin the current population P is evaluated by taking traffic lights are given by Table XV. Table XVI contains infor-
f1 (x) from ni=1 xi and f2 (x) from simulation. In line 9, pairs mation about the number of traffic lights at express routes and
of parent solutions are selected from the current population P to probability of stops at lights assuming a binomial distribution
compose the parent population P  through a tournament-based function. Rush and regular hours are shown in Table XVII.
selection. In line 10, an offspring population P  is generated The average bus speed between two consecutive stops has
from P  by crossover and mutation operations. The population been obtained by dividing the corresponding distance by the
of next generation is chosen in line 10 from P  and P  by time interval elapsed between these stops—measured by the
Pareto ranking and crowding distance. The best solutions are online BRT monitoring system. This time interval excludes stop
thus taken when the loop ends. times for boarding as well as at traffic lights. This procedure has
provided an average bus speed of 23 km/h with variations from
16 to 28 km/h.
Algorithm 1 Simulation optimization
An express bus weights 40.5 ton. When fully loaded
Input: P , FitnessFunction, Constraints, OptimizationOptions (250 passengers), its weight is increased by 15 ton, considering
Output: Non-dominated solutions (P ) an average weight of 60 kg per passenger. According to URBS,
1: // P : population of current generation  buses operate with full capacity in rush hours and with 60%
2: // FitnessFunction: ∀ x ∈ P , f1 (x) = ni=1 xi passenger occupancy otherwise. Moreover, about 1 ton should
3: // f2 (x) is evaluated by simulation be added to the bus weight due to one battery pack.
4: // Constraints: gi ≥ 0 Bus stops are distributed each 500 to 600 m with few
5: // OptimizationOptions: see Tables III and IV exceptions approaching to 1200 m. The average slopes between
6: P := Initialize(P ) stops are considered in the simulation through a route elevation
7: while termination condition is not satisfied do profile computed from altitude data using intermediate points.
8: Fitness := FitnessFunction(P ) They show slopes up to 8 degrees.
9: P  := Selection(P ) The simulation horizon is set to 8 h starting from 2:00 PM
10: P  := Genetic Operations(P  ) to include passenger demand of Peak 3 and Night 1. The later
11: P := Generation Update(P ∪ P  ) being the greatest challenge in terms of number of passengers
12: end while transported. The optimization algorithm runs with a maximum
of 200 generations before the best solutions are taken. For fit-
ness evaluation of each individual in each generation, five sim-
This approach is a “black box” type optimization in which ulation replications are executed. Each simulation replication is
options and parameters are set according to Tables III and IV, accomplished with a distinct sequence of randomly generated
respectively. They represent default choices for most optimiza- values to reduce the confidence interval of simulated variables
tion problems with good tradeoff between computational effort according to [26]. Although the number of replications used
and quality of solution according to [16]. Constraints are treated in the literature can vary according to different purposes and
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

6 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS

TABLE VI
N UMBER OF S TATIONS S HARED BY T WO OR
T HREE B US ROUTES FOR E ACH S OLUTION

TABLE VII
D ISTANCE B ETWEEN S TATIONS FOR E ACH S OLUTION ( M )

Fig. 3. Non-dominated solutions.

TABLE V
L OCATION OF C HARGING S TATIONS FOR E ACH S OLUTION

conditions, five to ten simulation replications are generally


applied, particularly when computational burden is a concern.
This process takes about 10 h to complete in a CPU Intel Core
i7/2.2 GHz with memory of 16 GB.
The Pareto’s border is shown in Fig. 3 representing the trade-
off between number of charging stations and extra stop time.
This extra time is beyond passengers’ boarding time but neces-
sary to charge batteries with enough energy to reach the next
charging station. It shows four solutions that cannot be im-
proved without penalizing one of the two objectives. These four
solutions are represented by 4, 5, 8, and 16 charging stations. It between stations is reduced only when 16 stations are con-
shows that extra time is reduced more than 50% by adding one sidered. Particularly, buses are charged in route 503 by only
station (from four to five stations). However, this ratio is re- one station until 16 stations are deployed. Another aspect that
duced as more stations are added until the upper bound number deserves attention is that routes with only one station can have
of 16 stations, where an increase on the number of stations has different minimum and maximum distances between consecu-
no effect on the extra time for charging. tive charging (depending whether the station is located in the
Table V presents the location of charging stations for the four half route distance or in the route start/end). For instance, route
solutions labeled 4-station, 5-station, 8-station, and 16-station. 503 has only one single station at route end yielding to a round
The location is identified by letters A to R which meaning is given trip distance of 20803.15 m between two consecutive charges.
in Table XIII and geographic position in Fig. 7 of the Appendix. After obtaining the solutions for the optimization prob-
Most stations presented in Table V are shared by different lem, a new simulation has been made for each solution with
routes as shown in Table VI. It shows an initial trend to increase 50 simulation replications to provide a proper confidence in-
the number of charging stations shared by more than one route. terval for each variable of interest. In the following tables, all
However, as the number of stations increases more routes have intervals are presented with 95% confidence level.
“private” stations (shared with no other routes). The extra stop time for charging can be seen in Table VIII.
Table VII shows the distance between stations for different It shows that the average extra charging time decreases signif-
solutions. It should be noticed that the average distance between icantly when increasing the number of stations, as expected.
stations decreases as the number of stations is increased (same However, the maximum time is not quite affected except for the
for the minimum distances). However, the maximum distance 16-station solution. It means that the worst case (maximum stop
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

SEBASTIANI et al.: EVALUATING ELECTRIC BUS OPERATION FOR BRT USING SIMULATION OPTIMIZATION 7

TABLE VIII TABLE IX


E XTRA S TOP T IME FOR O NE BATTERY PACK ( S ) ROUND T RIP T IMES O BTAINED BY S IMULATION AND
BY C URRENT O PERATION F ROM URBS ( MIN )

TABLE X
E NERGY C ONSUMPTION BY S TATION FOR D IFFERENT S OLUTIONS (kWh)

TABLE XI
E XTRA S TOP T IME FOR T WO BATTERY PACKS ( S )

Fig. 4. Energy level of batteries in each route for 4-station solution.

Fig. 5. Energy level of batteries in each route for 16-station solution.

Fig. 6. Extra stop time for batteries with different full charge times.
time) would require high investment in number of stations to be
improved.
To illustrate energy consumption, Figs. 4 and 5 show battery The average battery level increases when 16 stations are
levels for a bus running in each of the six routes for 4-station deployed as shown in Fig. 5. In this case, the average battery
and 16-station solutions, respectively. These figures have been level is above 50%. The capacity to provide energy to buses is
obtained for the first bus entering a route in the first simulation not only important to keep buses running but also to maintain
replication. Note that the initial battery level is according to the high average battery levels during operation. This condition is
distance travelled between the garage and the route. fundamental to extend battery’s lifecycle and to reduce mainte-
Fig. 4 shows short intervals of charging at the beginning nance costs [20].
of operation. They represent battery charging during standard Another important issue is the impact of using electric buses
boarding time (no extra time is necessary). About 3 h after in the current round trip time. This is shown in Table IX where
the operation starts, charging times are extended to provide round trip times can be compared with field-collected data
batteries with the minimum necessary to keep buses running. compiled from URBS for different deployment solutions. Note
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

8 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS

Fig. 7. Geographic locations of charging stations. (a) 4-station solution. (b) 5-station solution. (c) 8-station solution. (d) 16-station solution.

that some similarity is expected as simulation with electric TABLE XII


L ABELS AND D IRECTIONS OF E XPRESS ROUTES
operation is set to be close to the BRT current operation with
diesel. By considering that 16-station solution gives no extra
stop time according to Table VIII, the last two columns of
Table IX can be compared to each other as only standard
boarding times are involved. From Table IX, round trip times
of the last two columns are similar, meaning that the current
transportation capacity is maintained.
It is also important to evaluate how much energy is demanded
at each charging station, providing electric power companies
more information to be used for planning the necessary infras- for 4-station solution (less than 1 s in average). However, extra
tructure to support bus operation (power requirements of charg- batteries imply extra weight and thus extra energy consumption.
ing stations, energy distribution, and energy demand). This is Even room for batteries should not be neglected as it reduces
shown in Table X considering an operation horizon of 8 h as passenger capacity.
mentioned before. It is also important to compare results for different full
A large range between minimum and maximum values should charging times (time for fully charging an empty battery).
be noticed from Table X. This could be an issue for power com- Fig. 6 shows the impact on extra stop time for full charge times
panies when dealing with load balancing. Another point that ranging from 6 min to 1 h. A significant difference can be seen
deserves attention is the total energy demanded for different so- from Fig. 6 between extra stop times for different number of
lutions. From Table X, the total average energy is 11454.80 kWh stations when longer full charge times are specified. Therefore,
for 4-station and 16747.20 kWh for 16-station. Thus a great a large number of stations are beneficial for the operation as it
number of stations contribute to maintain high levels of energy is less sensitive (or robust) against full charge time variations.
stored in bus batteries according to the graphics of Fig. 5.
By considering two battery packs, Table XI shows the cor-
V. C ONCLUSION
responding extra stop times. Note that a significant decrease is
observed when compared with Table VIII. The extra stop time A simulation optimization approach has been applied for
for charging has no significant impact on boarding time even evaluating the replacement of diesel-powered buses by electric
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

SEBASTIANI et al.: EVALUATING ELECTRIC BUS OPERATION FOR BRT USING SIMULATION OPTIMIZATION 9

TABLE XIII TABLE XV


B US S TOP ID S A SSOCIATED TO C HARGING S TATIONS M EAN S TOP T IME AT B US S TOPS AND T RAFFIC L IGHTS ( S )

TABLE XVI
ROUTE P ROPERTIES
TABLE XIV
N UMBER OF B USES BY ROUTE D URING THE D AY

ones in a real-world BRT public transportation of Curitiba,


Brazil. This transportation system is composed by six express
routes with 118 stops served by different number of buses
according the period of the day. Although simulation utilizes
a simplified model for bus dynamics found in the literature, it
allows computation of energy spent in basic trajectories. The TABLE XVII
deployment of charging stations has been optimized by using RUSH ( IN B OLD ) AND R EGULAR H OURS OF T RANSPORTATION
a metaheuristic bi-objective algorithm that considers a tradeoff
between number of charging stations and extra time for battery
charging. The objective is to operate the bus fleet with minimum
overhead in stop time while avoiding buses to run out of
energy. Results from optimization have provided alternatives
for deployment of stations with average values of extra stop
time ranging from zero (for 16-station) to 15 s (for 4-station). the scenario significantly as hybrid vehicles do not have severe
However, cases far from average should also be taken into autonomy constraints.
account as they have a significant impact on passengers’
waiting time (e.g., some cases require additional 2 min to A PPENDIX
complete charging after passenger boarding). A large num- Geographic locations of charging stations for different solu-
ber of stations contribute for system operation with high tions are shown in Fig. 7. Tables XII–XVII contain data from
levels of energy stored in batteries (more than 50%) which Curitiba city.
can benefit battery’s lifecycle. In this case, the average extra
stop times for charging are less sensitive to full charge time ACKNOWLEDGMENT
variations as well. Although results have been presented for
a particular scenario of Curitiba, the methodology of simu- The authors acknowledge URBS, Volvo do Brasil, and the
lation optimization can be applied to other cases, requiring Municipality of Curitiba for technical support.
adjustments to provide an initial evaluation of system oper-
ation feasibility. The proposed model is also clearly suitable R EFERENCES
to smart grids planning. By adding restrictions related to [1] V. Tikka, J. Lassila, H. Makkonen, and J. Partanen, “Case study of the load
the availability of energy supply points for charging stations, demand of electric vehicle charging and optimal charging schemes in an
urban area,” in Proc. 3rd IEEE PES Innov. Smart Grid Technol. Europe,
the model can provide useful information to energy utilities and 2012, pp. 1–7.
public transportation operation to set expansion plans. Future [2] I. Frade, A. Ribeiro, G. Gonçalves, and A. P. Antunes, “Optimal location
work includes developing of models for energy consumption of charging stations for electric vehicles in a neighborhood in Lisbon,”
Transp. Res. Rec., vol. 2252, pp. 91–98, 2011.
based on more detailed bus manufacturer’s information. Plug-in [3] J. Dong, C. Liu, and Z. Lin, “Charging infrastructure planning for promot-
hybrid electric vehicles should also be considered. They change ing battery electric vehicles: An activity-based approach using multiday
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

10 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS

travel data,” Transp. Res. C, Emerging Technol., vol. 38, pp. 44–55, [24] D. Perrotta, B. Ribeiro, R. Rossetti, and J. Afonso, “On the potential of
Jan. 2014. regenerative braking of electric buses as a function of their itinerary,”
[4] P. Sinhubera, W. Rohlfsa, and D. Sauera, “Study on power and energy Procedia—Social Behav. Sci., vol. 54, pp. 1156–1167, Oct. 2012.
demand for sizing the energy storage systems for electrified local pub- [25] SimEvents User’s Guide, MathWorks Inc., Natick, MA, USA, 2005.
lic transport buses,” in Proc. IEEE Veh. Power Propulsion Conf., 2012, [Online]. Available: http://tinyurl.com/q4sxya9
pp. 315–320. [26] J. Banks, J. S. Carson, II, B. L. Nelson, and D. M. Nicol, Discrete-Event
[5] G. Tzeng, C. Lin, and S. Opricovic, “Multi-criteria analysis of alternative- System Simulation, 5th ed. Englewood Cliffs, NJ, USA: Prentice-Hall,
fuel buses for public transportation,” Energy Policy, vol. 33, no. 11, 2009.
pp. 1373–1383, Jul. 2005. [27] K. Deb, A. Pratap, S. Agarwal, and T. Meyarivan, “A fast and elitist
[6] U. Choi, H. Jeong, and S. Jeong, “Commercial operation of ultra low floor multi-objective genetic algorithm: NSGA-II,” IEEE Trans. Evol. Comput.,
electric bus for Seoul City route,” in Proc. IEEE Veh. Power Propulsion vol. 6, no. 2, pp. 182–197, Apr. 2002.
Conf., 2012, pp. 1128–1133. [28] T. P. Runarsson and X. Yao, “Stochastic ranking for constrained evolution-
[7] C. Zhu and X. Chen, “Optimizing battery electric bus transit vehicle ary optimization,” IEEE Trans. Evol. Comput., vol. 4, no. 3, pp. 284–294,
scheduling with battery exchanging: Model and case study,” Procedia— Sep. 2000.
Social Behav. Sci., vol. 96, pp. 2725–2736, Nov. 2013. [29] C. A. C. Coello, “Theoretical and numerical constraint-handling tech-
[8] N. Leemput et al., “Electrification of trucks and buses in an urban environ- niques used with evolutionary algorithms: A survey of the state of the art,”
ment through continuous charging,” in Proc. IEEE Int. Elect. Veh. Conf., Comput. Methods Appl. Mech. Eng., vol. 191, no. 11/12, pp. 1245–1287,
2012, pp. 1–8. Jan. 2002.
[9] J. Yang and H. Sun, “Battery swap station location-routing problem with [30] Urbanization of Curitiba S/A, URBS, Curitiba, Brazil, 2015. [Online].
capacitated electric vehicles,” Comput. Oper. Res., vol. 55, pp. 217–232, Available: http://www.urbs.curitiba.pr.gov.br
Mar. 2015.
[10] X. Xi, R. Sioshansi, and V. Marano, “Simulation-optimization model for
location of a public electric vehicle charging infrastructure,” Transp. Res.
D, Transp. Environ., vol. 22, pp. 60–69, Jul. 2013.
[11] F. Baouchea, R. Billota, N.-E. El Faouzi, and R. Trigui, “Electric vehicle Mariana Teixeira Sebastiani received the bach-
charging stations allocation models,” in Proc. Transp. Res. Arena Conf., elor’s degree in mechatronics engineering from
2014, pp. 1–10. Pontifical Catholic University of Paraná, Curitiba,
[12] D. Perrotta et al., “Route planning for electric buses: A case study Brazil, and the master’s degree in electrical and
in Oporto,” Procedia—Social Behav. Sci., vol. 111, pp. 1004–1014, computer engineering from Federal University of
Feb. 2014. Technology-Paraná, Curitiba. Since 2015, she has
[13] A. Lajunen, “Energy consumption and cost-benefit analysis of hybrid and been working toward the Ph.D. degree with the
electric city buses,” Transp. Res. C, Emerging Technol., vol. 38, pp. 1–15, Institute of Transportation Studies, University of
Jan. 2014. California, Irvine, CA, USA.
[14] M. C. Fu, “Overview of the handbook,” in Handbook of Simulation Opti-
mization, 3rd ed., M. C. Fu, Ed. New York, NY, USA: Springer-Verlag,
2015, pp. 1–7.
[15] R. Shankar and J. Marco, “Method for estimating the energy consumption
of electric vehicles and plug-in hybrid electric vehicles under real-world
driving conditions,” IET Intell. Transp. Syst., vol. 7, no. 1, pp. 138–150, Ricardo Lüders received the M.S. and Ph.D. de-
Mar. 2013. grees in electrical engineering from State University
[16] K. Deb, Multi-Objective Optimization Using Evolutionary Algorithms. of Campinas (Unicamp), Campinas, Brazil. His sab-
New York, NY, USA: Wiley, 2001. batical leave was with University of Michigan, USA,
[17] C. Fonseca and P. Fleming, “An overview of evolutionary algorithms in 2009. In 1997, he joined the Federal University
in multiobjective optimization,” Evol. Comput., vol. 3, no. 1, pp. 1–16, of Technology-Paraná (UTFPR), Curitiba, Brazil,
Spring 1995. where he is currently an Associate Professor. His re-
[18] C. A. C. Coello, G. B. Lamont, and D. A. V. Veldhuizen, Evolutionary search interest includes formalisms, simulation, and
Algorithms for Solving Multi-Objective Problems, 2nd ed. New York, optimization of discrete event systems with applica-
NY, USA: Springer-Verlag. tions in transportation, manufacturing, and industrial
[19] D. Perrotta, A. Teixeira, H. Silva, B. Ribeiro, and J. Afonso, “Electrical automation.
bus performance modeling for urban environments,” SAE Int. J. Alterna-
tive Power, vol. 1, no. 1, pp. 34–45, Apr. 2012.
[20] J. Larminie and J. Lowry, Electric Vehicle Technology Explained.
Chichester, U.K.: Wiley, 2003. Keiko Verônica Ono Fonseca received the
[21] J. C. Ferreira, J. de Almeida, and A. R. da Silva, “The impact of M.S. degree from State University of Campinas
driving styles on fuel consumption: A data-warehouse-and-data-mining- (Unicamp), Campinas, Brazil, and the Ph.D. degree
based discovery process,” IEEE Trans. Intell. Transp. Syst., vol. 16, no. 5, from Federal University of Santa Catarina (UFSC),
pp. 2653–2662, Oct. 2015. Florianópolis, Brazil, both in electrical engineering.
[22] H. Hõimoja, A. Ruffer, G. Dziechciaruk, and A. Vezzini, “An ultrafast Her sabbatical leave was with Dresden University
EV charging station demonstrator,” in Proc. Int. Symp. Power Electron., of Technology (TUD), Germany, in 2013. In 1980,
Elect. Drives, Autom. Motion, 2012, pp. 1390–1395. she joined Federal University of Technology-Paraná
[23] Sustainability Report 2014, Volvo Group, Göteborg, Sweden, 2014. (UTFPR), Curitiba, Brazil, where she is currently a
[Online]. Available: http://www3.volvo.com/investors/finrep/sr14/pdf/SR Full Professor. Her research interest includes smart
_2014.pdf grids and smart cities.

You might also like