This article has been accepted for publication in a future issue of this journal, but has not been
fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2014.2305593, IEEE Transactions on Vehicular Technology
Real-time Optimization for Power Management
Systems of a Battery/Supercapacitor Hybrid Energy
Storage System in Electric Vehicles
Mid-Eum Choi, Jun-Sik Lee, and Seung-Woo Seo, Member, IEEE
Abstract—Batteries mounted on electric vehicles (EVs) are
often damaged by high peak power and rapid charg-
ing/discharging cycles, which are originated from repetitive ac-
celeration/deceleration of vehicles especially in urban situations.
In order to reduce battery damage, the battery/super-capacitor
(SC) hybrid energy storage system (HESS) has been considered
as a solution because the SC can act as a buffer against large
magnitudes and rapid fluctuations in power. While the traditional
purpose of employing the HESS in EVs is to minimize the
magnitude/variation of the battery power or the power loss,
the previous approaches proposed for controlling the HESS
have some drawbacks; they did neither consider these objectives
simultaneously nor reflect real-time load dynamics for computing
the SC reference voltage. In this paper, we present a power
control framework consisting of two stages: one for computing
the SC reference voltage, and the other for optimizing the power
flowing through the HESS. In the presented framework, we
propose a methodology for calculating the SC reference voltage
considering the real-time load dynamics without given future
operation profiles. In addition, we formulate the HESS power
control problem as a convex optimization problem that minimizes
the magnitude/fluctuation of the battery power and the power
loss at the same time. The optimization problem is formulated so
that it can repeatedly be solved by general solvers in polynomial
time. Simulation results carried out on MATLAB show that the
magnitude/variation of the battery power and the power loss can
concurrently be reduced in real time by the proposed framework.
Index Terms—Battery, supercapacitor, convex optimization,
power management, hybrid energy storage, real-time control
I. I NTRODUCTION
I N recent years, Electric Vehicles (EVs) have started to be
spotlighted as zero-emission eco-friendly vehicles. Many
OEMs are showing various models of EVs with better per-
formance and improved driving distance. The performance
of EVs depends heavily on the performances of power-train
components, i.e., the motor, inverter, and battery. Among them,
the battery is considered as a key component because it has
a large impact on the acceleration and driving distance on a
single charge.
Copyright (c) 2013 IEEE. Personal use of this material is permitted.
However, permission to use this material for any other purposes must be
obtained from the IEEE by sending a request to pubs-permissions@ieee.org.
M. Choi, J. Lee, and S. Seo are with the Department of Electri-
cal and Computer Engineering, Seoul National University, (email: me-
choi@cnslab.snu.ac.kr, jslee@cnslab.snu.ac.kr, and sseo@snu.ac.kr).
This work was supported by the National Research Foundation of Ko-
rea(NRF) grant funded by the Ministry of Science, ICT and Future Plan-
ning(MSIP) (No.2009-0083495), and partially Institute of New Media and
Communications and Automation and Systems Research Institute, Seoul
National University.
0018-9545 (c) 2013 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.
Because the battery has a relatively low power density per
unit cell compared to other energy storage devices, it requires
a large number of cells to meet the power required from a
traction motor. One of the serious problems in operating a
large number of battery cells is that one or some cells can be
damaged by high peak power or rapid charging/discharging,
which are originated from repetitive acceleration/deceleration
of vehicles especially in urban situations. In order to supple-
ment these weaknesses, a battery/super-capacitor (SC) hybrid
energy storage system (HESS) has been proposed, in which
the SC can act as a buffer against large magnitudes and rapid
fluctuations in power because the SC has a high power density
comparing to the battery.
The battery/SC HESS can generally be categorized into
two types: passive and active. The passive HESS is easily
constructed by connecting the battery and the SC directly
in parallel [1]-[3]. However, the SC in the passive HESS
cannot provide or receive a large amount of power sufficiently
because the state of charge (SoC) of the SC is mostly fixed. In
contrast, the SC in the active HESS can provide a sufficient
amount of power to the motor because the SC can have a
different voltage from the battery thanks to DC/DC converters
[4]-[7]. The DC/DC converters in the active HESS make the
power flow controllable. Many researchers have consistently
proposed new DC/DC converter structures in order to improve
the performance of the DC/DC converter [8]-[11]. In [28], the
state-of-the-art energy storage topologies for HEVs and plug-
in HEVs are presented with specific examples.
When controlling the power flowing through the HESS, the
following objectives have mainly been considered. First, the
magnitude/variation of the battery power should be minimized
in order to extend the life of the battery; this is a fundamental
goal of the HESS [12][15][19]-[21]. Second, the power loss
should be minimized in order to increase the driving distance
[16]-[18]. Third, the SoC of the SC should be adjusted
according to future driving profiles [15][21][23]. Before the
EV decelerates in near future, for example, the SC should be
discharged beforehand in order to receive regenerative braking
energy.
Although these objectives should be accomplished at the
same time when the HESS power is controlled, the previous
approaches have not achieved these objectives simultaneously.
Moreover, while the real-time load dynamics should be con-
sidered when computing the reference voltage of the SC,
it has not been considered in all of the previous works.
Napoli et al [12] propose a control strategy for HESS in
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2014.2305593, IEEE Transactions on Vehicular Technology
which a battery, a fuelcell, and a SC are connected through a
multiple input DC/DC converter. In this approach, the authors
formulate linear equations which provide reference values for
the battery/fuelcell/SC power satisfying upper/lower bound
conditions. However, the objective functions to be minimized
are not described in this work. Lukic et al [14] propose several
HESS power control strategies, as follows: 1) the SoC of
the SC should be kept constant, 2) the load power should
be kept constant, and 3) the SC power should be maximized
considering the future load profiles. However, these strategies
are not suitable for application to a HESS in EVs for the
following reasons: for the first strategy, the SoC of the SC
should adaptively be adjusted in order to provide sufficient
power to the electric motor. For the second strategy, the
load power is not maintained constant in practice in general
driving situations. In the third strategy, it is assumed that the
future load profiles are previously known. Carter et al [15]
propose a rule-based strategy for HESS power control. In
this approach, described in Section IV, the HESS power is
controlled based on several state parameters, i.e., the motor
power, maximum battery power, reference voltage of the SC
and the SC voltage. Zandi et al [19] propose a power control
algorithm based on a Fuzzy Logic Control (FLC). The input
variables considered in this approach are the battery SoC, the
SC SoC, and the load power. The power split ratio between the
battery and the SC can be computed by using the membership
functions and the rule-bases, which are previously defined
based on the input variables. These rule-based control strategy
[15][19] are easily implemented in the HESS for real-time
operation. Schaltz et al [29] investigate the influence of the
battery/SC HESS sizing on the battery life time in a fuelcell
HEV. In this approach, they show a power control strategy in
which the power of the battery/FC should be lower than such
threshold values. However, these approaches cannot provide
an optimal HESS power minimizing the magnitude/variation
of the battery power.
In order to compute the optimal HESS power for EVs,
several researchers have designed the problem of HESS power
control as an optimization problem. Moreno et al [16] for-
mulate an optimization problem which minimizes the power
loss in a DC/DC converter. In order to solve this problem, the
authors apply their idea to a neural network in which the HESS
power can be controlled in real time. Nevertheless, they do not
try to minimize the magnitude/variation of the battery power.
Wang et al [17][18] propose a methodology for sizing the
HESS and controlling the HESS power. In this approach, the
objective is also to minimize the power loss. When computing
the power loss, however, the loss caused by the DC/DC
converter is not considered despite the fact that it is relatively
large compared to the loss caused by the battery/SC. Moreover,
it is not attempted to minimize the magnitude/variation of the
battery power. Choi et al [20][21] formulate the HESS power
control problem as a convex optimization problem in an effort
to provide a global optimal solution. The objective of the
optimization problem is to minimize the magnitude/variation
of the battery current. However, this approach is based on
assumptions that are too ideal, i.e., the future operating profile
is previously given, and the power loss caused by the DC/DC
0018-9545 (c) 2013 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.
converters is zero. Bauman et al [22] propose an optimization
method of which objective is to maximize the efficiency and to
minimize the mass and the cost for determining the optimal
battery/SC size for the fuelcell/battery/SC HESS. However,
they do not attempt to compute the optimal HESS power
under dynamic load situations. Romaus et al [23] attempt
to optimize the power flowing through the battery/SC HESS
using Stochastic Dynamic Programming (SDP). In this opti-
mization problem, two objectives are considered; 1) the power
loss induced by the battery/SC HESS, and 2) the gap between
the SC voltage and the reference voltage of the SC, which is
determined experimentally based on typical driving situations.
The algorithm of the SDP consists of four steps: the initial
guess, an evaluation of the strategy, improving the strategy,
and exiting the iteration if costs converge. The strategy based
on the SDP algorithm has the following weaknesses: 1) the
SDP requires large computational overhead due to the two-
stage optimization process, i.e., the second and third step,
in a single iteration, and 2) the performance of the HESS
power control scheme can be worsen when the vehicle is
driven in unpredictable situations. In addition, they do not try
to minimize the magnitude/fluctuation of the battery power.
Hredzak et al [24] and Torreglosa et al [25] propose a strategy
based on the Model Predictive Control (MPC) algorithm. The
MPC algorithms show good performance if the motor power
changes with predictable patterns. Laldin et al [26] propose
the predictive algorithm of which objective is to minimize
the power loss. The authors present a Markov process with
nine states, which is defined from the speed/acceleration of the
driving cycle known previously. However, the accuracy of the
predicted value of the motor power decreases when the speed
of the vehicle sharply increases or decreases. In other words,
the SC cannot be used effectively when the motor provides or
requires an extremely high level of power.
In this paper, we investigate the real-time optimization of
the power management scheme for the battery/SC HESS in
EVs. Our contributions can be summarized as follows:
1) We design a new control framework for the real-time
control of the power flowing through the battery/SC HESS in
EVs. The framework consists of two parts: one for computing
the SC reference voltage based on the load dynamics, and the
other for optimizing the power flowing through the HESS.
2) We propose a methodology for computing the reference
voltage of the SC considering real-time load dynamics, i.e.,
the vehicle dynamics, characteristics of the motor, the driving
conditions, and the regenerative braking systems. With the
proposed method, the SC can handle the future power required
by the traction motor even when a future operation profile is
not given beforehand.
3) We formulate a convex optimization problem that min-
imizes the magnitude/variation of the battery power and the
power loss simultaneously. The formulated optimization prob-
lem is shown to become an equality constrained problem
which can repeatedly be computed by general solvers in
polynomial time. The reference voltage of the SC computed
by the proposed method is used as a given parameter in the
constraints of the formulated problem.
The remainder of this paper is organized as follows: Section
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2014.2305593, IEEE Transactions on Vehicular Technology
Pm = τ mωm
PE PB
motor battery
ηm = f m (τ m , ωm ) PSCa
PSC
Pwh = τ whωwh
SC
ηCONV = f CONV ( PSC )
Fig. 1. Schematic of the system architecture.
g max-tq Zm (k )
A B
Fig. 2. Motor efficiency map.
II describes the system architecture considered in this paper.
Section III proposes the power control framework for the
HESS in EVs and explains how the reference voltage of the
SC is computed considering previously mentioned factors. In
addition, the proposed optimization problem for controlling
the HESS power is presented. Simulation results and the
conclusion are given in Sections IV and V, respectively.
II. S YSTEM D ESCRIPTION
The system architecture considered in this paper is presented
in Fig. 1. The vehicle has a front-wheel driven powertrain
with an electric motor that yields a maximum power of
50 kW, and a gearbox of which the gear ratio assumed
to be fixed. The braking system of the EV consists of a
mechanical braking system and a regenerative braking system;
the mechanical/electrical braking torque is determined by the
braking strategy designed beforehand. The motor is connected
to the HESS through an inverter. In the HESS, the battery and
the SC are connected through a DC/DC converter which is
controlled by the proposed optimization strategy. Details will
be explained in the following subsections.
A. Powertrain Model
The resistance force at speed v(t) can be calculated by the
following equation [27]:
1 the ideal braking force of the EV. If the braking force required
Fv (v(t)) = ρa cD Af v(t)2 + FR (v(t)) + gmt sin ϑ(t), (1)
2 by the stroke of the brake pedal is 0.5mt g, for example,
where ρa , cD , and Af are the air density, drag coefficient,
and frontal area, respectively. Note that the force Fv (v(t))
0018-9545 (c) 2013 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.
is obtained by the summation of the aerodynamic drag, the
rolling resistance, and the grading resistance. The rolling
resistance FR (v(t)) can be derived by a fifth-order polynomial
function of the vehicle speed v(t)
FR (v(t)) = gmt cos ϑ(t){a0 + a1 v(t)
+ a2 v(t)2 + a3 v(t)3 + a4 v(t)4 + a5 v(t)5 }, (2)
where a0 , a1 , . . . , a5 are obtained from actual experimental re-
sults. In this work, we refer to the coefficients from Sciarretta’s
work [27].
The torque at the wheels required to accelerate the vehicle
is equal to the sum of the resistance torque and the dynamic
torque for acceleration. It is computed as
θv dv(t)
τwh (t) = rwh Fv (t) + , (3)
rwh dt
where rwh is the radius of the wheels and θv is the total
inertia of the vehicle. The rotational speed of the wheels can
be intuitively calculated as follows:
ωwh (t) = v(t)/rwh . (4)
The force at the wheels is delivered from/to the driving
motor through a gearbox as described in Fig. 1. According to
the gear ratio Gr , the torque and rotational speed at the motor
are determined as
τm (t) = τwh (t)/Gr , (5)
ωm (t) = ωwh (t)Gr . (6)
Using the values of τm (t) and ωm (t), the input power of
the motor-inverter can be calculated as

τm (t)ωm (t)/ηm (t) if τm (t) ≥ 0
PE (t) = (7)
τm (t)ωm (t)ηm (t) otherwise
where 0 < ηm (t) < 1 is the motor-inverter efficiency, which
is computed as a function of τm (t) and ωm (t):
ηm (t) = fm (ωm (t), τm (t)) . (8)
The function fm can be experimentally obtained from a
bench test [33]; fm considered in this paper is presented in
Fig. 2.
B. Regenerative Braking System
One of the most important features of EVs is their ability
to convert significant amounts of braking energy into electric
energy that can be stored in the energy storage and then reused.
Because the braking torque required is generally much larger
than the torque that an electric motor can produce, mechanical
braking systems should coexist with the regenerative braking
system.
The braking force distribution strategy is shown in Fig. 3
[34]. Here, β-line represents a fixed ratio of the mechanical
braking force between the front and rear wheels, and I-line is
the operating points of the braking force should be placed
on the line of j = 0.5g; the point A and B in Fig. 3 are
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2014.2305593, IEEE Transactions on Vehicular Technology

0 40
0.40 TABLE I
o, Fbr /(mtg)

T HE BATTERY PARAMETERS
0.35
E -line VB0 battery constant voltage (V) 312.44
0 30
0.30 (mechanical brake)
ng force ratio

RB internal resistance (Ω) 7.2

KB polarisation voltage (V) 1.1307
0.25
AB exponential zone amplitude (V) 37.44
0.20 DB exponential zone time constant inverse ((Ws)−1 ) 9.4328e-08
Rear brakin

0.15 I -line
0.10 a A
(ideal braking) from that of the DC bus. Among all HESS topologies, the
specific topology in which only-the SC is connected through
Fbr-mech(A)

b B
0.05
the DC/DC converter, as described in Fig. 1, has mainly been
studied due to the fact that it has good controllability in that
B)

00
0.0 02
0.2 04
0.4 06
0.6 08
0.8 10
1.0
F br-mech(B

the HESS power can effectively be controlled by the DC/DC

Fbf -mech(A) F bf-reg (A)
converter connected to the SC. Moreover, it is possible for
Fbf -mech(B) F bf-reg (B) Front braking force ratio, Fbf /(mtg)
the battery in this topology to be damaged by too large peak
power over a very short time. However, the peak power flowing
Fig. 3. Braking forces varying with deceleration rate [34]. through the battery can be reduced rapidly by the DC/DC
converter in real-time. The low configuration cost is another
factor. Only one DC/DC converter is used to construct the
candidates for the braking force with 0.5mt g. When point A
HESS. We assume that the net power flowing in and out of the
is chosen as the braking point, the regenerative braking force
DC bus is equal to zero; that is, PB (t)+PSCa (t)−PE (t) = 0,
Fbf−reg (A) is determined by segment A-a because the EV
where PB (t), PSCa (t), and PE (t) are the SC power converted
considered here has a front-wheel-driven powertrain. The rest
by the DC/DC converter, the battery power and the power
of the braking force is supplied by the mechanical braking
delivered to the motor.
force of the front and rear wheels, denoted as Fbf−mech(A)
1) Battery/SC model: The P battery has max
a finite level of
and Fbr−mech (A), respectively. In other words, Fbf−reg (A) max
energy which is denoted as t PB (t)∆ = EB , where EB
+Fbf−mech(A) +Fbr−mech(A) is equal to the total braking
is the maximum energy which can be stored in the battery. The
force of 0.5mt g. In a similar way, we can compute the braking
internal voltage is modeled as in [31]:
force Fbf−reg (B), Fbf−mech(B), and Fbr−mech(B). As shown  max 
in Fig. 3, Fbf−reg (B) is larger than Fbf−reg (A), meaning that EB
VB,int (t) = VB0 − KB
the regenerative energy at point B is larger than that at point −EB (t)
A. Note that Fbf−reg (B) should be smaller or equal to the max
+ AB exp (−DB (EB − EB (t))) ,
max
maximum force that can be produced by the motor Fm . VB (t) = VB,int (t) − RB iB (t),
max
In other words, if Fbf−reg (B) is equal to Fm , we can
say that point B is an optimal operating point for obtaining where iB (t), VB (t), RB , and EB (t) denote the battery current,
the regenerative braking energy. Therefore, we compute the the battery voltage, the internal resistance of the battery, and
regenerative braking force Fbf−reg as follows: the energy stored in the battery respectively. The parameters
 VB0 , KB , AB , and DB , which are shown in TABLE I, can be
 0 if v(t) < v0 calculated using the discharge curve of the considered battery
Fbf−reg (t) = Fb (t) else if Fb (t) < Fmmax (9) (see [31]). The power loss induced by the battery can be
 max
Fm otherwise derived as follows:
2
where Fb (t) is the required braking force and v0 is the low

PB (t)
speed threshold; generally v0 is 15km/h [34]. The regenera- PB,Loss (t) = RB . (10)
VB (t)
tive energy that can be obtained at less than 15km/h is too low
because electromotive force is scarcely generated by the motor The energy stored in the SC, denoted as ESC (t), can be
at low rotational speeds. Therefore, Fbf−reg (t) is set to zero calculated using VSC (t) as follows:
when v(t) < v0 . The regenerative braking system described 1 2
ESC (t) = CSC VSC (t), (11)
in this section will be applied to compute a suitable SoC for 2
the SC. where CSC is the capacitance of the SC. The minimum voltage
min
of the SC (VSC ) is limited to half of the maximum voltage of
max
C. Battery/SC Hybrid Energy Storage Systems the SC (VSC ) because the efficiency of the DC/DC converter
connected to the SC becomes too low when the voltage of the
Fig. 1 presents the battery/SC HESS providing power max
SC (VSC (t)) is lower than VSC /2 as shown in Fig. 4.
(PE (t)) to the motor through an inverter. In the HESS, the
2) DC/DC converter model: The DC/DC converter controls
battery and the SC are connected through active DC/DC
the SC power denoted as PSC (t) and the output power of
converters. The HESS can be constructed with several types
the DC/DC converter denoted as PSCa (t). The relationship
of topologies, as explained in earlier work [14][20]. Generally,
between PSC (t) and PSCa (t) can be expressed as
the SC is connected through a DC/DC converter because a
DC/DC converter permits the voltage of the SC to differ PSCa (t) = PSC (t) − |Pconv,Loss (t)| , (12)

0018-9545 (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2014.2305593, IEEE Transactions on Vehicular Technology

A. Computation of SC Reference Voltage

As mentioned in the introduction, the reference voltage of
ref
the SC (VSC ) needs to be adjusted in order to maximize the
1.0 SC’s ability to support the peak power. If the future operation
of the motor can be known in advance, it is easy to control
0.9 ref
VSC
)

efficiently. However, it is difficult to obtain the future

Efficiency (%

0.8 operation of the motor in advance. In this subsection, we

ref
present a method for computing a suitable value of VSC
0.7
140
150 without knowledge of the future operating profile of the motor.
0.6
130
120 V)
Before the motor requires a large amount of power, the
( max
100 90 80 100
110
lta
ge SC should previously be charged up to VSC in order to
Output 70
current
60 50 90 vo provide sufficient power to the motor whenever needed. In
of the D 40 30 20 80 SC
C/DC c
onverte
r
10
(A)
many experiments, it was observed that the motor requires
a large amount of power, typically when the vehicle starts.
ref max
Therefore, VSC is set to VSC when the vehicle speed is zero.
Fig. 4. DC/DC converter efficiency map. This basic strategy allows us to control VSC ref
by considering
only the energy generated from regenerative braking.
ref
VSCref As described in Fig. 5, P1 computes VSC via τm (t−∆) and
W m (t  ') PB (t  ')
*
PSC (t ) ωm (t − ∆), which can be considered as the given parameters
Zm (t  ')   PSC (t )
VSC (t  ') 2 VSC (t  ') 2 at time t. From these two parameters, the future speed and
VB (t  ') PE (t ) torque at the wheels can be calculated as iteratively follows:
PWM fut fut
PI Duty vwh (k) = vwh (k − 1)
 
rwh fut fut



+∆ τwh (k − 1) − rwh Fv vwh (k − 1) (14)
Motor Eff. DC/DC Conv. θv
Eff. Map
Map
 
fut θv fut


τwh (k) = ad + rwh Fv vwh (k − 1) , (15)
rwh
Fig. 5. Flowchart of the proposed power control strategy
where
fut
where Pconv,Loss (t) is the power loss induced by the DC/DC vwh (0) =rwh ωm (t − ∆) /Gr
fut
converter, the determination of which depends on the SC τwh (0) =Gr τm (t − ∆) .
current and the SC voltage. The value of Pconv,Loss (t) is
computed as [32]: Eq. (15) is equivalent to (3), and Eq. (14) can be derived
 2 from (15). Here, k ∈ [1, K] where K = T /∆ is the time of the
PSCa (t) fut
near future from t to t + T − 1. If vwh fut
(k) ≤ 0, then vwh (k) =
Pconv,Loss (t) = DRon , (13)
VB (t)(1 − D) 0. The force Fv (·) in (14) and (15) can be calculated using
where D is shown at the bottom of this page. (1). The parameter ad in (15) is the minimum acceleration
The Ron is the drain-source resistance of the switch which (maximum deceleration), which can be provided by the ego-
is included in the DC/DC converter. The efficiency map of the vehicle; that is,
DC/DC converter is shown in Fig. 4. The detailed architecture dv(τ )
ad = min . (16)
of the battery/SC HESS will be described in the next section. ∀τ dτ
fut
The calculated value of τwh (k) is provided by the me-
III. P OWER C ONTROL S YSTEM FOR HESS chanical braking and the regenerative braking systems. As
fut
In this section, we propose the power control framework presented in the previous section, τwh (k) should be converted
for the battery/SC HESS in EVs. The proposed power control into regenerative braking energy to the greatest extent possible.
fut fut
framework consists of two parts, as described in Fig. 5; P1 , Therefore, τwh (k) is provided by the motor only if τwh (k)
ref
which computes the SC reference voltage (VSC ) based on the is not larger than the maximum torque that can be provided by
max fut max
vehicle dynamics; and P2 , which optimizes the power flowing the motor (τm (k)). If τwh (k) is larger than τm (k), the
through the HESS. The details of P1 and P2 are explained in rest of the braking torque is supported by mechanical braking.
the following subsections. Therefore, the future rotational speed and torque of the motor

q 2
VB + VSC 2 − Ron PSCa /VB2 ± {VB + VSC (2 − Ron PSCa /VB2 )} − 4 (VB + VSC ) VSC
D= .
2 (VB + VSC )

0018-9545 (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2014.2305593, IEEE Transactions on Vehicular Technology

can be calculated as follows:

fut fut
ωm (k) = Gr vwh (k)/rwh (17)
fut

 0 vwh (k) < v0
fut
τm (k) = fut
τ fut (k)/Gr τwh max
(k)/Gr < τm (k) (18)
 whmax
−τm (k) otherwise,
where rwh is the radius of the wheels and Gr is the gear ratio.
max
Here, τm (k) can be determined by the maximum torque
function, denoted as gmax−tq , which is a function of ωm (k),
as described in Fig. 2. For example, a motor operating at
rotational speed A can provide braking torque up to 100Nm.
max
Similarly, τm (k) = 50Nm if the motor operates at point B.
fut fut
Using ωm (k) and τm (k), the power delivered from the
motor-inverter can be calculated as
pfut fut fut
E (k) = ωm (k)τm (k)ηm (k), (19)
where ηm (k) is the motor-inverter efficiency as computed by Fig. 6. The power loss of DC/DC converter when VB = 300(V ) and
(8). Note that pfut
E (k) ≤ 0. VSC = 150(V ). The dashed line is obtained by (13) and the solid line
Although the regenerative braking energy delivered from is its approximated piecewise linear function.
.
the motor is charged to not only the SC but also the battery,
we set the energy charged to the SC equal to the regenerative
braking energy in order to minimize the peak power delivered • Deadzone-linear penalty function:
to the battery. The energy charged to the SC in the near future ψ(u) = max (0, |u| − σ) , where σ > 0 is a given
with duration T can be computed by parameter that determines the width of the penalty-free
K
zone. Using the l1 −norm function and the deadzone-
fut
X linear penalty function, the piecewise linear penalty func-
pfut


ESC = E (k) + |Pconv,Loss (k)| ∆, (20)
tion can be designed as ψ(u) = max (|u|, 2|u| − σ) .
k=1
• Log barrier penalty function:
where |Pconv,Loss (k)| is the power loss of the DC/DC con-
fut −σ 2 log(1 − (u/σ)2 ) |u| < σ

verter as computed by (13). From the computed value of ESC ,
ψ(u) =
the reference voltage of the SC can be calculated using (11) ∞ |u| ≥ σ,
as follows:
where σ > 0 is a given parameter that determines the
s
ref E fut
VSC = 2 SC + (VSC max )2 . (21) width of the distribution range of u.
CSC
Using penalty functions which are convex, the power control
fut
Note that ESC ≤ 0 because only the regenerative braking problem can be formulated as a penalty function approxima-
ref
energy is considered when computating VSC . Nevertheless, tion problem [35]. The objective function can be formulated
the SC can deliver the required future motor power sufficiently as the sum of the penalty functions, i.e.,
ref min
because the stored energy in the SC, i.e., ESC − ESC , can be
fut
used for pE > 0. If the vehicle speed is zero, for example, α1 ψ1 (PB (t)) + α2 ψ2 (PB (t) − PB (t − ∆))
the SC needs to be fully charged in order to provide enough
+ α3 ψ3 (PB (t), VB (t)) + α4 ψ4 (PSCa (t), VSC (t), VB (t))
power to the motor in near future. ref


+ α5 ψ5 ESC (t), ESC , (22)
B. Computation of the optimal SC power where ESC (t) is the energy stored in the SC. It can be
ref
In this subsection, we formulate a convex optimization calculated from VSC (t) using (11). Similarly, ESC , which is
ref
problem to optimize the power flow in the HESS. The the reference energy of the SC, can be derived from VSC . The
objective of the optimization problem is to minimize the values of α1 , . . ., α5 are trade-off parameters. The terms ψ1
magnitude/variation of the battery power and the energy loss. and ψ2 are a function of the magnitude and the variation of the
The formulated problem is solved in P2 , as described in Fig. battery power. The terms ψ3 and ψ4 correspondingly refer to
5. By solving the proposed optimization problem, the optimal the power loss caused by the battery and the DC/DC converter
power flowing through the HESS can be obtained. respectively. The final term ψ5 is added to adjust the energy
1) Objective function: In order to design the objective ref
stored in the SC based on ESC , which is computed by P1 .
function of the optimization problem, penalty functions ψ(·) 2) Design of the penalty functions: It is important to design
can be applied. A common characteristic of penalty functions penalty functions that minimize the objectives efficiently. A
is that the function value increases as the absolute input value large PB (t), especially PB (t) > PBthre where PBthre is a
increases. The representative penalty functions are described threshold of the battery power which can shorten the battery’s
as follows [20][35]: life. Therefore, the function ψ1 should have more of a handicap
p
• lp −norm function: ψ(u) = |u| , where p is 1, 2, or ∞. when PB (t) > PBthre , as expressed by

0018-9545 (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2014.2305593, IEEE Transactions on Vehicular Technology

(28), the term costB × Nreduced/Nnormal means the battery

ψ1 = max (0, |u1 | − 1) , (23) cost lost by PBmax , and the term T × (Nnormal − Nreduced)
is total hours of battery use. The factor α2 can similarly
where u1 = PB (t)/PBthre . Similarly, the function ψ2 can also be designed with the cost of the battery damage caused by
be represented as thre
|PB (t) − PB (t − ∆)| − PB,var . We assume that the tradeoff
ψ2 = max (0, |u2 | − 1) , (24) factor α1 and α2 can experimentally be determined. In addi-
tion, α3 and α4 can simply be designed with the cost per unit
thre
where u2 = (PB (t) − PB (t − ∆))/PB,var .
The threshold power because the functions ψ3 and ψ4 are functions of the
thre thre
values, i.e., PB and PB,var , can be determined arbitrarily or power.
experimentally. The function ψ5 , which needs to be as close The function 0 ≤ ψ5 ≤ 1 can be considered as representing
to zero as possible to minimize the gap between ESC (t) and the potential damage of the battery; for example, PB (t)
ref
ESC , can be expressed as increases when ψ5 increases. Therefore, the last tradeoff factor
ψ5 = |u5 | , (25) α5 can be defined as
α5 = α1 . (29)
ref max min



where u5 = ESC (t) − ESC ESC −ESC . Note that
ref
|u5 | ≤ 1 because ESC (t) and ESC are in ESC max min
, ESC . 4) Optimization problem: The problem concerning the con-
The ψ3 and ψ4 are a function of the power loss induced by trol of the power flow in the HESS can be formulated as a
the battery and the DC/DC converter respectively. Therefore, convex optimization problem which is solved using P2 , as
these penalty functions can be designed as follows: follows:
 2 min. Eq.(22)
PB (t)
ψ3 = RB , (26) s.t. PB (t) + PSCa (t) − PE (t) = 0,
VB (t)
 2 PSCa (t) = PSC (t) − |Pconv,Loss (t)| , (30)
PSCa (t) D
ψ4 = Ron , (27) ESC (t) − ESC (t − ∆) = −PSC (t)∆,
VB (t) (1 − D)2 min max
ESC ≤ ESC (t) ≤ ESC .
where ψ4 is equal to (13). Because ψ3 and ψ4 are not convex,
we need to make these penalty functions convex. In (26), In this formulation, the control variables are PB (t), PSC (t),
VB (t) can be replaced with VB (t − ∆), which can be known PSCa (t), and ESC (t). The voltage of the SC can easily be
at time t, in order to make ψ3 convex. This is acceptable computed from ESC (t) using (11). However, this formulation
because the difference between VB (t) and VB (t − ∆) is is not a convex optimization problem because the second
sufficiently small. In (27), similarly, VB (t) and VSC (t) can be constraint function regarding PSC (t) is not an affine function;
replaced with VB (t − ∆) and VSC (t − ∆). Because VB (t) and the equality constraint functions of a convex problem must
VSC (t) were replaced by the known parameters, the power loss be affine. In order to make the second constraint affine, we
equation (27) becomes a function of PSCa (t). This function replace Pconv,Loss (t) with Pconv,Loss (t − ∆), which can be
can be designed as a piecewise linear function which is a determined at time t. Because the time gap ∆ is sufficiently
convex function as shown in Fig. 6. Note that the shape small, it is acceptable to assume that the difference between the
of the piecewise linear function is determined depending on DC/DC converter efficiencies at time t−∆ and t are negligible.
VB (t − ∆) and VSC (t − ∆). Moreover, the accumulated error caused by the negligible
3) Adjustment of tradeoff factors: The tradeoff factors, i.e., difference between Pconv,Loss (t − ∆) and Pconv,Loss (t) does
α1 , . . ., α5 ≥ 0, affect the weight of the functions included not diverge because the given parameters included in the
in the objective function. Generally, the weight of the penalty formulated problem are updated at each instant. Therefore,
function becomes larger as the corresponding tradeoff factor the problem regarding the control of the power flow in the
increases. If ψ(u) = |u|, for example, αi ψ(u) > αj ψ(u) for HESS can be expressed as a convex optimization problem, as
all u, where αi > αj . If ψ(u) is designed as the deadzone follows:
linear function, αi ψ(u) > αj ψ(u) for |u| > σ and αi ψ(u) =
min. Eq.(22)
αj ψ(u) = 0 for |u| ≤ σ, where αi > αj . Note that the
function has no penalty when |u| ≤ σ. s.t. The constraints of Eq.(30) where
Here, α1 , which is the cost of the battery damage caused second equality is substituted with
by |PB (t)| − PBthre , can be derived approximately from PSCa (t) = PSC (t) − |Pconv,Loss (t − ∆)| .
Nreduced In order to obtain the optimal power of the SC denoted
costB × =
Nnormal ∗
as PSCa (t), the formulated problem needs to be solved by a
max
P solver.
α1 × Bthre × T × (Nnormal − Nreduced), (28)
PB 5) Computation by solver: A general form of the convex
where costB is the price of the battery, Nnormal is the initial problem can be described as follows:
cycle life of the battery, Nreduced is the battery cycle life minimize f0 (x)
reduced by PBmax which is a maximum available power of subject to fi (x) ≤ 0, i = 1, · · · , m
the battery, and T is the time of battery use with PBmax . In Ax = b,

0018-9545 (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2014.2305593, IEEE Transactions on Vehicular Technology

Algorithm 1 The barrier method. TABLE II

1: (0)
given feasible x, t ← t > 0, µ > 1, tolerance ǫ > 0. T HE V EHICLE PARAMETERS

2: repeat ρa air density (kg/m3 ) 1.205

3: Compute x∗ by minimizing tf0 + φ, subject to Ax = b, cD drag coefficient 0.32
starting at x Af frontal area (m2 ) 2.31
mt vehicle mass (kg) 1680
4: x ← x∗ θv vehicle total inertia (kg·m2 ) 145
5: t ← µt rwh wheel radius (m) 0.29
6: until m/t < ǫ Gr gear ratio 6.45
a0 rolling resistance coefficient 8.80·10−3
a1 rolling resistance coefficient -6.42·10−5
a2 rolling resistance coefficient 9.27·10−6
a3 rolling resistance coefficient -3.30·10−7
a4 rolling resistance coefficient 6.68·10−9
a5 rolling resistance coefficient -4.46·10−11
CSC SC capacitance (F) 3400/56
max
VSC maximum voltage of SC (V) 2.7·56
PBthre the threshold of the power causing damage to 15
the battery (kW)
thre
PB,var the threshold of the power variation causing 1.5
damage to the battery (kW/s)
max
Pconv maximum power of DC/DC converter (kW) 50

IV. S IMULATION R ESULT

Fig. 7. The operating profile: the speed profile (FTP75) and the power
required by the motor. The required power is computed by using the dynamic
model described in Section II-A. . and the optimization problem. In the simulation, we considered
where f0 , . . . , fm : Rn → R are convex and twice continu-
ously differentiable, and A ∈ Rp×n with rank A = p < n.
In order to solve the problem with a guaranteed specified
accuracy level of ǫ, the general form needs to be changed
to an equality constrained problem
minimize (m/ǫ)f0 (x) + φ(x)
subject to Ax = b,
where ǫ is the tolerance to guarantee the sub-optimality, m
is the number
Pm of inequality constraints in the problem, and
φ(x) = − i=1 log(fi (x)).
The barrier method, a typical solver, can solve the equality
constrained problem in polynomial time [35]. A simple version
of the barrier method is described by Algorithm 1, where µ is
the size of the Newton step. At each iteration, x∗ is computed
by Newton’s method. The iteration terminates when t ≥ m/ǫ.
The proposed formulation shown in the previous subsection
can be presented with the equality constraint problem. The
total number of iterations of the barrier method is
log(m/t(0) ǫ)
  
m(µ − 1 − log µ)
Nbf = +c
log µ µ
≈ O(m log m),
where c = log2 log2 (1/ǫnt ) [35]. Here, ǫnt denotes the
tolerance in Newton’s method. In the results, the optimal
power PB∗ (t) and PSCa ∗
(t) can directly be derived from
∗
x (t) provided by the barrier method. Based on the value of
∗ max
PSCa (t), the PI controller described in Fig. 5 adjusts the PWM
duty cycle of the DC/DC converter to keep the SC power close
∗
to PSCa (t).
A simulation program was implemented in MATLAB in
order to verify the aforementioned HESS system architecture
the HESS topology presented in Fig. 1. The considered driving
cycle, which is a speed profile termed FTP75, is shown in Fig.
7. Using this speed profile, the power required by the electric
motor can be derived using the power train model described
in Section II-A. Because we assume that the future operating
profile cannot be known in advance, the optimal power of a
battery/SC is computed at each instant. The vehicle parameters
applied in the simulation are shown in Table II [27]. The
simulation results were obtained by solving the proposed opti-
mization problem, which minimizes the magnitude/variation of
the battery power and the power loss. The solver implemented
in the simulation is SeDuMi [35] which is a MATLAB-based
solver for solving convex optimization problems.
For a performance comparison, the rule-based strategy [15],
the SC voltage control based on the vehicle speed [16], and a
fuzzy logic strategy [19] were considered in the simulation. In
the rule-based algorithm, the HESS power is controlled based
req
on several states, i.e., the motor power (PM ), the threshold of
the power causing damage to the battery (PBthre ), the reference
ref
voltage of the SC (VSC ), and the voltage of the SC (VSC ), as
follows:
• PMreq
> PBthre : the SC supplies PM req
− PBthre . If the SC
req
does not have sufficient energy to supply PM − PBthre ,
the SC supplies as much power as possible and the battery
supplies the rest.
req req
• PM ≤ PBthre : the battery provides PM . If VSC ≤ VSC ref
,
the battery supplies power to both the motor and the SC.
req
• PM < 0 (regenerative braking): the SC receives as much
power as possible and the battery receives the rest.
In the SC voltage control strategy based on p the vehicle
speed, the SC voltage is determined by VSC 1 − v/vmax
where v is the speed of the vehicle and vmax is the maximum
speed of the vehicle.

0018-9545 (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2014.2305593, IEEE Transactions on Vehicular Technology

60
40
SC power (kW) 20
0
-20
-40
-60
0 200 400 600 800 1000 1200 1400 1600 1800
Times (s)
40
Battery power (kW)

30
20
10
0
-10
-20
-30
-40
0 200 400 600 800 1000 1200 1400 1600 1800
Times (s)
160
140
SC voltage (V)

120
100
80
60
0 200 400 600 800 1000 1200 1400 1600 1800
Times (s)
355
350
Battery voltage (V)

345
340
335
330
325
320
315
0 200 400 600 800 1000 1200 1400 1600 1800
Times (s)

Proposed algorithm
Rule-based algorithm
SC voltage control based on speed
Fuzzy logic algorithm

Fig. 8. The simulation results: the SC power, the battery power, the SC voltage, and the battery voltage.

30
Battery power (kW)

20
10
0
-10
-20
200 250 300 350 400 450 500 550 600 650 700
Times (s)

Proposed algorithm
Rule-based algorithm
SC voltage control based on speed
Fuzzy logic algorithm

Fig. 9. The battery power from t = 200 to 700 computed by the proposed strategy and the other strategies, i.e., the rule-based strategy, the SC voltage
control based on the vehicle speed, and the fuzzy logic strategy.

0018-9545 (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2014.2305593, IEEE Transactions on Vehicular Technology
Accumulated power loss (kWs)
8e+3
6e+3
4e+3
2e+3
0
0 200 400 600 800 1000 1200 1400 1600 1800
Times (s)
Proposed algorithm
Rule-based algorithm
SC voltage control based on speed
Fuzzy logic algorithm
Fig. 10. The accumulated power loss computed by the proposed strategy and
the other strategies, i.e., the rule-based strategy, the SC voltage control based
on the vehicle speed, and the fuzzy logic strategy.
In the fuzzy logic strategy, the input variables considered are
the battery voltage, the SC voltage, and the load power. The
membership functions for the battery voltage and the SC volt-
age are listed as the following five categories: very negative,
negative, zero, positive, and very positive. The membership
functions of the load power are categorized into three types:
recovery, normal, and overload. Because the HESS considered
in an earlier work [19] and that in this paper are different with
regard to the maximum voltage of the energy storage devices,
we modified the scale of the input of the membership functions
and the values used for the rule bases.
Simulations were carried out with several trade-off factors,
Sα = (α1 , α2 , α3 , α4 , α5 ). Fig. 8 shows the changing pattern
of the SC power, the battery power, the SC voltage, and the
battery voltage as computed by the proposed strategy and
the other strategies, i.e., the rule-based strategy, the strategy
with SC voltage control based on the vehicle speed, and
the fuzzy logic strategy. In Fig. 9, the battery power from
t = 200 to 700 is shown in detail. The black line in
Fig. 8 was obtained by the proposed strategy with Sα =
(0.1481, 0.01481, 1/3000, 1/3000, 0.1481). The average com-
putation time for a single iteration of the proposed algorithm
is 42.8 ms and the maximum time is 78.1 ms. This value can
change depending on the solver used.
As shown in Fig. 8, the battery controlled by the rule-
based strategy tends to be discharged when the motor power
increases strongly, indicating that the rule-based strategy can
effectively keep the battery power within PBthre . However, it
is not discharged in general urban driving situations. This
indicates that the rule-based control method is not appropriate
for urban situations, in which the motor frequently requires
considerable and fluctuating power due to the repeated stop-
and-go driving patterns used in such situations. The battery
power controlled by the rule-based strategy is relatively large
compared to the other values when PB (t) < 0, as the
SC cannot absorb the large amount of the power provided
by the motor. The speed-based SC voltage control strategy
and the fuzzy logic strategy can reduce the battery power
more effectively than the rule-based strategy in urban driving
0018-9545 (c) 2013 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.
10
1.2e+4
Accumulated power loss (kWs)
1.0e+4
8.0e+3
6.0e+3
4.0e+3
2.0e+3
0.0
0 200 400 600 800 1000 1200 1400 1600 1800
Times (s)
Fig. 11. The accumulated power loss computed by the proposed
strategy with Sα = (0.1481, 0.01481, 1/3000, 1/3000, 0.1481) (solid
line), Sα = (0, 0, 1/3000, 1/3000, 0.1481) (dotted line), and Sα =
(0.1481, 0.01481, 0, 0, 0.1481) (broken line).
situations as presented from the cases when t = 400 to 700 in
Fig. 9. However, these strategies cannot prevent the battery
power from exceeding the threshold which causes damage
to the battery when the motor requires an extremely large
amount of power, as shown from the cases when t = 200
to 300. It is observed that the proposed strategy can restrict
the battery power below PBthre with variation below PB,var thre
.
In other words, the proposed method can reduce the mag-
nitude/fluctuation of the battery power effectively and make
the SC receive more regenerative energy from the motor.
Thus, the life of the battery can be extended by the proposed
optimization method.
Fig. 10 shows the comparison of the accumulated power
losses as computed by the four strategies. As shown in Fig. 10,
the lowest power loss is obtained by the fuzzy logic strategy,
not our proposed strategy. However, the difference between the
amount of the power loss provided by the fuzzy logic strategy
and that provided by our strategy is small. The reasons why the
proposed strategy does not show the best performance in terms
of the power loss are as follows: 1) our strategy minimizes
not only the power loss but also the magnitude/fluctuation of
the battery power, and 2) the SC voltage is influenced by the
SC reference voltage, which depends on the load dynamics.
Note that the SC reference voltage is not considered in the
fuzzy logic strategy. Fig. 11 shows the effect of the tradeoff
factors when using the proposed strategy. The dotted line
was obtained with Sα = (0, 0, 1/3000, 1/3000, 0.1481) which
means that the magnitude/fluctuation of the battery power
was not considered in the simulation. In contrast, the broken
line was obtained with Sα = (0.1481, 0.01481, 0, 0, 0.1481)
which means that the power loss was not considered in
the simulation. As shown in Fig. 11, the power loss com-
puted with Sα = (0, 0, 1/3000, 1/3000, 0.1481) is lower
than that computed with Sα = (0.1481, 0.01481, 0, 0, 0.1481)
because the objective function formulated with Sα =
(0, 0, 1/3000, 1/3000, 0.1481) imposes more of a penalty to
the power loss function. In other words, it is clear that the
power loss term of the objective function in the proposed
strategy, i.e., ψ3 and ψ4 , can effectively reduce the power loss
induced by the battery/SC HESS.
Table III shows the statistics of the battery power computed
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2014.2305593, IEEE Transactions on Vehicular Technology
TABLE III
S IMULATION RESULTS
Maximum magnitude of the battery power (W)
Average magnitude of the battery power (W)
Standard deviation
Maximum variation of the battery power (W/s)
Average variation of the battery power (W/s)
Standard deviation
by four strategies, which can be obtained from Fig. 8. As
shown in Table III, the maximum magnitude/variation of the
battery power calculated by our strategy, i.e., 15000 (W)
and 1500.2 (W/s) respectively, are much smaller than that
computed by the other strategies. This simulation result means
that the proposed method can be used for increasing the life
of the battery effectively because the damage of the battery
is exponentially increased as the magnitude/variation of the
battery power increase. Although the average battery power
computed by our strategy is slightly larger than that controlled
by the rule-based strategy, the difference between these two
values is not large.
V. C ONCLUSION
In this paper, real-time optimization for the power manage-
ment systems for the battery/SC HESS in EVs was investi-
gated. We described a power control framework to control the
power flowing through the HESS and proposed a methodology
for computing the reference voltage of the SC in order to adjust
the SoC of the SC without a future operating profile. Finally,
we formulated a convex optimization problem to achieve the
following goals: minimizing the magnitude/fluctuation of the
battery power, minimizing the energy loss, and minimizing
the gap between the SC voltage and its reference value. We
reformulated the problem of the HESS power control into
an equality constrained problem, which can be solved by
a general solver such as the barrier method within a few
iterations in a polynomial time. As a result, the life of the
battery can be extended and the size of the battery can be
reduced through our scheme because it effectively minimizes
the magnitude/fluctuation of the battery power. In addition, the
proposed scheme can be applied for sizing the SC because the
amount of charging/discharging energy can be analyzed by a
function which computes the SC reference voltage and by a
function which computes the optimal SC power.
R EFERENCES
[1] J. P. Zheng, T. R. Jow, and M. S. Ding, ”Hybrid Power Sources for Pulsed
Current Applications,” IEEE Trans. Aerosp. Electron. Syst., vol. 1, no. 1,
pp. 288-292, Jan. 2001.
[2] H. W. Brandhorst Jr. and Z. Chen, ”Achieving a High Pulse Power System
through Engineering the Battery-Capacitor Combination,” in Proc. 16th
Annu. Battery Conf. Applications Advances, pp. 153-156, Jan. 2001.
[3] R. A. Dougal, S. Liu, and R. E. White, ”Power and Life Extension of
Battery-Ultracapactor Hybrids,” IEEE Trans. Compon. Packag. Technol.,
vol. 25, no. 1, pp. 120-131, Mar. 2002.
[4] D. Liu, and H. Li, ”A Three-port Three-phase DC-DC Converter for
Hybrid Low Voltage Fuel Cell and Ultracapacitor,” in Proc. Conf.
Industrial Electronics, pp. 1369-1374, Jun. 2003.
0018-9545 (c) 2013 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.
11
Proposed Rule-based. Speed based. Fuzzy logic.
15000 32641 27432 29677
4065.14 3929.95 5264.55 4729.66
5243.81 5358.79 5435.77 5050.26
1500.2 16323.8 18320.9 16485.26
616.22 1706.88 2049.86 1974.80
690.29 2320.84 2884.14 2435.74
[5] S. Pay and Y. Daly, ”Effectiveness of Battery-Supercapacitor Combination
in Electric Vehicles,” in IEEE Proc. Power Tech. conf., vol. 3, pp. 1-6,
Jun. 2003.
[6] A. C. Baisden and A. Emadi, ”Advisor-based Model of a Battery and an
Ultra-capacitor Energy Source for Hybrid Electric Vehicles,” IEEE Trans.
Vehic. Tech., vol. 53, no. 1, pp.199-205, Jan. 2004.
[7] L. Gao, R. A. Dougal, and S. Liu, ”Power Enhancement of an Actively
Controlled Battery/Ultracapacitor Hybrid,” IEEE Trans. Power. Electron.,
vol. 20, no. 1, Jan. 2005.
[8] J. Cao, and A. Emadi, ”A New Battery/UltraCapacitor Hybrid Energy
Storage System for Electric, Hybrid, and Plug-In Hybrid Electric Vehi-
cles”, in IEEE Trans. Pow. Elect., vol. 27, no. 1, pp.122-132, Jan. 2012.
[9] O. C. Onar, and A. Khaligh, ”A Novel Integrated Magnetic Structure
Based DC/DC Converter for Hybrid Battery/Ultracapactor Energy Storage
Systems”, in IEEE Trans. Smart Grid., vol. 3, no. 1, pp.296-307, Mar.
2012.
[10] M. B. Camara, H. Gualous, F. Gustin, and A. Berthon, ”Design and New
Control of DC/DC Converters to Share Energy Between Supercapacitors
and Batteries in Hybrid Vehicles”, in IEEE Trans. Veh. Tech., vol. 57, no.
5, pp.2721-2735, Sep. 2008.
[11] M. B. Camara, H. Gualous, F. Gustin, A. Berthon, and B. Dakyo,
”DC/DC Converter Design for Supercapacitor and Battery Power Man-
agement in Hybrid Vehicle Applications - Polynomial Control Strategy”,
in IEEE Trans. Indust. Elect., vol. 57, no. 2, pp.587-597, Feb. 2010.
[12] A. D. Napoli, F. Crescimbini, F. G. Capponi, and L. Solero, ”Constrol
Strategy for Multiple Input DC-DC Power Converters Devoted to Hybrid
Vehicle Propulsion Systems,” in Proc. 2002 IEEE International Sympo-
sium on Industrial Electronics, pp. 1036-1041, May. 2002.
[13] Y. Zhang, Z. Jiang. and X. Yu, ”Control Strategies for Bat-
tery/Supercapacitor Hybrid Energy Storage Systems,” in IEEE Energy
2030 Conf., pp.1-6, Nov. 2008.
[14] S. M. Lukic, S. G. Wirasingha, F. Rodriguez, J. Cao, and A. Emadi,
”Power Management of an Ultracapacitor/Battery Hybrid Energy Storage
System in an HEV,” in IEEE Vehic. Power and Prop. Conf., pp.1-6, Sep.
2006.
[15] R. Carter, A. Cruden, and P. J. Hall, ”Optimizing for Efficiency or
Battery Life in a Battery/Supercapacitor Electric Vehicle”, in IEEE Trans.
Veh. Tech., vol. 61, no. 4, pp.1526-1533, May. 2012.
[16] J. Moreno, M. E. Ortuzar, and J. W. Dixon, ”Energy-Management
System for a Hybrid Electric Vehicle, Using Ultracapacitors and Neural
Networks”, in IEEE Trans. Indust. Electr., vol. 53, no. 2, pp.614-623,
Apr. 2006.
[17] L. Wang, E. G. Collins. Jr, and H. Li, ”Optimal Design and Real-Time
Control for Energy Management in Electric Vehicles”, in IEEE Trans.
Veh. Tech., vol. 60, no. 4, pp.1419-1429, May. 2011.
[18] L. Wang, and H. Li, ”Maximum Fuel Economy-Oriented Power Manage-
ment Design for a Fuel Cell Vehicle Using Battery and Ultracapacitor”,
in IEEE Trans. Indust. Appli., vol. 46, no. 3, pp.1011-1020, May/Jun.
2010.
[19] M. Zandi, A. Payman, J. P. Martin, S. Pierfederici, B. Davat , and F. M.
Tabar, ”Energy Management of a FuelCell/Supercapacitor/Battery Power
Source for Electric Vehicular Applications”, in IEEE Trans. Veh. Tech.,
vol. 60, no. 2, pp.3882-3895, Feb. 2011.
[20] M. Choi, S. Kim, and S. Seo, ”Energy Management Optimization in a
Battery/Supercapacitor Hybrid Energy Storage System,” in IEEE Trans.
Smartgrid, vol. 3, no. 1, pp.463-472, Mar. 2012.
[21] M. Choi, and S. Seo, ”Robust Energy Management of a Bat-
tery/Supercapacitor Hybrid Energy Storage System in an Electric Ve-
hicle,” in IEEE Elec. Veh. Conf., pp.1-5, 2012.
[22] J. Bauman, and M. Kazerani, ”An Analytical Optimization Method for
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2014.2305593, IEEE Transactions on Vehicular Technology

12

Improved Fuel Cell-Battery-Ultracapacitor Powertrain,” in IEEE Trans. Jun-Sik Lee received the M.S. degree in 2007 and
Veh. Tech., vol. 58, no. 7, pp.3186-3197, Sep. 2009. is currently working toward the Ph.D. degree in the
[23] C. Romaus, D. Wimmelbucker, K. S. Stille, and J. Bocker, ”Self- Department of Electrical and Computer Engineering,
Optimzation Energy Management Considering Stochastic Influences for Seoul National University, Korea.
a Hybrid Energy Storage of an Electric Road Vehicle”, in IEEE Intern. He is a Researcher with the Intelligent Vehicle
Elect. Mach. and Driv. Conf., pp.67-74, 2013. IT (IVIT) Research Center funded by the Korean
[24] B. Hredzak, V. G. Agelidis, and M. S. Jang, ”A Model Predictive Control Government and Automotive Industries. He is a
for a Hybrid Battery-Ultracapacitor Power Source”, in IEEE Trans. Power senior engineer with Korea Testing Certification. His
Electr., pp.1469-1479, 2014. current research areas include vehicular communi-
[25] J. P. Torreglosa, P. Garcia, L. M. Fernandez, and F. Jurado, ”Pre- cation, computer and network security, and system
dictive Control for the Energy Management of a Fuel-Cell-Battery- optimization.
Supercapacitor Tramway”, in IEEE Trans. Indust. Inform., vol. 10, no. 1,
pp.276-285, Feb. 2014.
[26] O. Laldin, M. Moshirvaziri, and O. Trescases, ”Predictive Algorithm
for Optimizaing Power Flow in Hybrid Ultracapacitor/Battery Storage
Systems for Light Electric Vehicles”, in IEEE Trans. Power Electr., vol.
28, no. 8, pp.3882-3895, Aug. 2013.
[27] A. Sciarretta, M. Back, and L. Guzzella, ”Optimal Conrol of Parallel
Hybrid Electric Vehicles,” in IEEE Trans. Cont. Sys. Tech., vol. 12, no.
3, pp.352-363, May. 2004.
[28] A. Khaligh, and Z. Li, ”Battery, Ultracapacitor, Fuel Cell, and Hybrid
Energy Storage Systems for Electric, Hybrid Electric, Fuel Cell, and Plug-
In Hybrid Electric Vehicles: State of the Art”, in IEEE Trans. Veh. Tech.,
vol. 59, no. 6, pp.2806-2814, Jul. 2010.
[29] E. Schaltz, A. Khaligh, and P. O. Rasmussen, ”Influence of Bat-
tery/Ultracapacitor Energy-Storage Sizing on Battery Lifetime in a Fuel
Cell Hybrid Electric Vehicle”, in IEEE Trans. Veh. Tech., vol. 58, no. 8,
pp.3882-3891, Oct. 2009.
[30] H. Yeo, S. Hwang, and H. Kim, ”Regenerative Braking Algorithm for a
HEV with CVT Ratio Control During Deceleration,” in Proc. of the Inst.
of Mech. Eng., Part D: Journal of Automobile Engineering, vol. 220, no.
11, pp.1589-1600, Nov. 2006.
[31] O. Tremblay, L. Dessaint, and A. Dekkiche, ”A Generic Battery Model
for the Dynamic Simulation of Hybrid Electric Vehicles,” in Proc. IEEE
Vehicle Power Propuls. Conf., pp 284289, Sep. 2007.
[32] R. W. Erickson, ”Fundamentals of Power Electronics, 2nd edition,” Seung-Woo Seo (M97) received the B.S. and M.S.
2004. degrees from Seoul National University, Seoul, Ko-
[33] T. A. Burress, S. L. Campbell, C. L. Coomer, C. W. Ayers, A. A. rea, and the Ph.D. degree from Pennsylvania State
Wereszczak, J. P. Cunningham, L. D. Marlino, L. E. Seiber, and H. T. Lin, University, University Park, PA, USA, all in electri-
”Evaluation of the 2010 Toyota Prius Hybrid Synergy Drive System”,Praise the Lord cal engineering.
in Energy Efficiency and Renewable Energy FreedomCAR and Vehicle He is a Professor of Electrical Engineering with
Technologies, Mar. 2011. Seoul National University and Director of the Intel-
[34] M. Ehsani, Y. Gao, A. Emadi,”Modern Electric, Hybrid Electric, and ligent Vehicle IT (IVIT) Research Center funded by
Fuel Cell Vehicles (Fundamentals, Theory, and Design)”, CRC press, the Korean Government and Automotive Industries.
2010 He was with the Faculty of the Department of
[35] S. Boyd, and L. Vandenberghe,”Convex Optimization,” Cambridge Uni- Computer Science and Engineering, Pennsylvania
versity Press, 2004. State University, and served as a Member of the Research Staff with the
Department of Electrical Engineering, Princeton University, Princeton, NJ.
In 1996, he joined the Faculty of the School of Electrical Engineering,
Institute of New Media and Communications and Automation and Systems
Research Institute, Seoul National University. He has served as Chair or
a Committee Member at various international conferences and workshops,
including INFOCOM, GLOBECOM, PIMRC, VTC, MobiSec, Vitae, etc. He
also served for five years as a Director of the Information Security Center with
Seoul National University. His research areas include vehicular electronics for
intelligent vehicles, communication networks, computer and network security,
and system optimization.

Mid-Eum Choi received the B.S. degree in elec-

trical engineering from Korea University, Seoul, in
2008. He is currently working toward the Ph.D. de-
gree with the Department of Electrical and Computer
Engineering, Seoul National University, Korea.
He is a Researcher with the Intelligent Vehicle IT
(IVIT) Research Center funded by the Korean Gov-
ernment and Automotive Industries. His current re-
search areas include intelligent vehicular technology,
vehicular communication, and system optimization.

0018-9545 (c) 2013 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.