IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 16, NO.
5, MAY 2020
Optimal Sizing and Energy Management for
Cost-Effective PEV Hybrid Energy
Storage Systems
Xiaoying Lu and Haoyu Wang , Senior Member, IEEE
Abstract—In battery/ultracapacitor (UC) hybrid energy
storage systems (HESS), sizing and energy management
strategies are crucial, which determine the system cost
and performance. However, research on these two prob-
lems in a coupled manner for plug-in electric vehicles is
still immature. This article aims at resolving this issue in
the perspective of minimizing the average operating cost.
Both manufacturing cost and system end-of-life timing are
incorporated. A quantitative battery degradation model is
employed to evaluate the battery dynamic capacity loss
and cycle life. Dynamic programming algorithm is then
deployed to achieve optimal power distribution between
battery and UC. Furthermore, the power management and
HESS optimal sizing strategies are unified into a single
cost-minimization problem. Combining those efforts, the
optimal size of the HESS with minimized average operating
cost is solved by simulated annealing method. Optimiza-
tion results illustrate that a minimum cost of 15.52 USD is
achieved with 72 UC cells and 7100 battery cells. A large
set of simulation data has proved the optimality of the opti-
mization results. Compared with the battery-only solution,
the proposed solution demonstrates 11.9% cost reduction
and 21.7% battery cycle life extension under the Urban
Dynamometer Driving Schedule. Moreover, the temperature
rise of the battery is reduced by 31.1%. Finally, based on
the optimal results, the energy management strategy is ex-
tended to fit real-time applications by utilizing Markov chain
and stochastic dynamic programming.
Index Terms—Battery degradation, cost-effective, energy
management, optimal sizing, ultracapacitor (UC).
Manuscript received May 30, 2019; revised August 22, 2019 and
November 8, 2019; accepted November 24, 2019. Date of publication
December 3, 2019; date of current version February 6, 2020. This work
was supported in part by the National Natural Science Foundation of
China under Grant 51607113, and in part by the Shanghai Sailing Pro-
gram under Grant 16YF1407600. Paper no. TII-19-2150. (Correspond-
ing author: Haoyu Wang.)
X. Lu is with the Power Electronics and Renewable Energies Lab-
oratory, School of Information Science and Technology, ShanghaiTech
University, Shanghai 201210, China, and also with the Cisco Systems,
Shanghai 200233, China (e-mail: xiaoying.lu1994@gmail.com).
H. Wang is with the Power Electronics and Renewable Ener-
gies Laboratory, School of Information Science and Technology,
ShanghaiTech University, Shanghai 201210, China (e-mail: wanghy.
shanghaitech@gmail.com).
Color versions of one or more of the figures in this article are available
online at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TII.2019.2957297
1551-3203 © 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: Anelis Plus consortium. Downloaded on January 11,2021 at 10:42:51 UTC from IEEE Xplore. Restrictions apply.
3407
I. INTRODUCTION
LUG-IN electric vehicles (PEVs) are considered as the
P most competitive candidate to revolutionize the internal
combustion engine-based transportation systems [1]. In PEV,
the energy storage system (ESS) is a key component, which
determines the vehicle performances, such as electric mileage,
manufacturing cost, and acceleration.
Many potential ESSs have been investigated and reported. In
PEV powering systems, flywheels, batteries, and ultracapacitors
(UC) are commonly considered [2]. A flywheel energy storage
(FES) has the advantage of high power density and theoretically
unlimited charge/discharge cycles. However, the extensive ap-
plication of FES has not been investigated in PEVs yet, owing
to the limitation of its high initial cost and immature technology
so far [3]. Rechargeable batteries possess good characteristics,
including high specific energy, flat discharge profile, low resis-
tance, and low memory effect. The battery types mainly include
nickel-based, lead-acid, metal-air based, and lithium-ion (Li-
ion) batteries. Among them, Li-ion battery dominates in PEV
applications due to its high specific energy, low self-discharge
rate, and low recyclability [4]. However, the cycle life of Li-ion
battery is affected by temperature and is reduced significantly in
high discharge rate [5]. Compared with the Li-ion battery, UC
exhibits the capabilities of longer cycle life and higher power
density. Thus, the hybridization of Li-ion battery and UC in
the hybrid energy storage system (HESS) is considered as an
effective solution to promise enhanced system performance [6].
Typically, there are two technical challenges associated with
the design of HESS: energy management strategy and sizing
optimization [7].
In battery/UC HESS-based PEV, the energy management
strategy directly determines the driving performance. Generally,
it is expected that UC absorbs the load dynamics, and battery
bears the average. The classical energy management strategies,
such as rule-based control [8] and filter-based control [9], have
been investigated in HESS due to their simplicity and low
computational burden. However, these control strategies are
mainly pre-set and cannot ensure real optimal solutions. Beyond
those pre-set strategies, a more advanced energy management
strategy is optimization-based strategy. The optimization objec-
tive is usually formulated and quantified into a cost function.
Substantial research efforts have been made to achieve different
objectives, including prolonged battery cycle life [10], reduced
3408
system cost [11], improved system efficiency [12], etc. Intelli-
gent algorithms such as dynamic programming (DP)[13], parti-
cle swarm optimization algorithm [14], and neural network [15]
have been utilized to solve these problems. Multiobjective op-
timization is also widely used in this field to reach the optimal
balance among different objectives [16], [17].
In the perspective of sizing optimization, the factors includ-
ing system weight, electric mileage, cost, and battery degrada-
tion are typically addressed [18], [19]. These sizing strategies
typically optimize system efficiency, weight, cost, and battery
degradation with the constraints of required stored energy and
power demand.
State-of-health (SOH) estimation for Li-ion batteries and
state-of-charge (SOC) estimation for UCs are also crucial prob-
lems in sizing and energy management optimization. The bat-
tery SOH estimation methods can be generally divided into
three categories: 1) physical model-based methods study the
physical and chemical effects occurred inside the battery to
evaluate its degradation [20]; 2) empirical model-based meth-
ods describe the law of battery degradation by fitting exper-
imental data [21]; 3) data-driven approaches take use of the
tested samples to extract feature variables, and further achieve
SOH prediction. Some signal processing methods, such as
Gaussian process regression [22], are utilized to identify the
feature variables more precisely. A variety of SOC estimation
methods have been investigated, including ampere-hour integral
method, black-box model-based estimation, and online model-
based approach, in which ampere-hour integral method is the
simplest and most widely used method; however, this method
highly depends on the initial SOC and requires accurate detec-
tion of operating current [23]. Alternatively, online model-based
approach is insensitive to the initial SOC and is able to be em-
bedded into real-time controllers: electrochemical models [24],
equivalent circuit models with the resistance-capacitance (RC)
network [17], and fractional-order models [25] have attracted
extensive interest.
Although both energy management and sizing have their
unique significances, it is indeed their combined effects that
determine the system performances. Several published works
consider these two problems in a coupled or decoupled way.
Zhang et al. [26] adopt a wavelet-transform-based power man-
agement algorithm to realize power allocation between battery
and UC, while the optimal sizing is derived considering battery
SOH, HESS weight, and cost. Shen et al. [27] conduct sizing
analysis based on tradeoffs between HESS weight and battery
capacity loss, while the energy management is implemented
using a rule-based control strategy. Similarly, energy manage-
ment and sizing optimization are considered as two decoupled
problems in [28]–[30]. Generally, those decoupled methods are
easy to implement. However, the enhancement of performance
is typically insignificant.
A more accurate approach to address the combined effects of
energy management and sizing is to integrate them in a coupled
manner [7], [31]–[33]. Hung and Wu [7] construct a multiple
for-loop structure with a preset cost function to achieve optimal
sizing and energy management. The optimization objective
is maximizing the stored energy. In [31], the optimal sizing
and energy management of a hybrid bus is solved by convex
Authorized licensed use limited to: Anelis Plus consortium. Downloaded on January 11,2021 at 10:42:51 UTC from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 16, NO. 5, MAY 2020
programming algorithm, in order to minimize a summation of
the diesel cost and the ESS cost. In [32], a dual-loop framework
based on DP algorithm and Pareto optimality is proposed to find
the quasi-optimal solution for energy management and HESS
sizing. Moreover, two methodologies are investigated in [33]
to study energy management and sizing problems in PEV
HESS, in which the second methodology employs an integrated
structure to yield a nonlinear optimization problem. Those
coupled optimization methods are more likely to maintain
a high optimization accuracy; they are, nevertheless, more
computation-intensive.
Although efforts have been spent to simultaneously optimize
HESS sizing and energy management, they mainly address the
tradeoffs among different optimization objectives [26]. Typ-
ically, several weights are assigned to different optimization
objectives [27], [34]. However, simply quantifying those trade-
offs by assigning them different weights based on personal
preference or experience has its intrinsic limitations.
Due to the aforementioned limitations of state-of-the-art re-
search, this manuscript provides a perspective to merge several
significant considerations together. The key idea is to quantita-
tively evaluate all factors in the perspective of a novel objective,
namely system average operating cost, which is defined as the
initial manufacturing cost divided by HESS cycle life. Those
factors can be literally briefed as follows: 1) the HESS size
determines the manufacturing cost; 2) large UC corresponds to
high vehicle weight and power demand, which leads to increased
manufacturing cost; 3) large UC helps to extend the battery
lifetime, which extends the battery end-of-life timing; and 4)
battery degradation is mapped to the operating cost. Those
considerations are integrated into a single cost-based objective
function. Therefore, the entire problem is rendered as how to
minimize the average operating cost during the entire life cycle
of the HESS.
The main novelties and contributions of this study are sum-
marized as follows. First, a unique cost-minimization principle
is proposed to derive the final optimization solution. Second, a
coupled dual-loop optimization structure is proposed to solve
the optimal sizing and energy management simultaneously: 1)
the inner loop addresses the energy management to minimize the
battery capacity loss using DP algorithm; 2) the outer loop solves
the optimized HESS sizing using simulated annealing (SA) al-
gorithm with reduced computation burden. Third, a quantitative
battery degradation model is built to evaluate both the dynamic
capacity loss and cycle life of the battery in a more precise way.
Fourth, based on the results of the dual-loop optimization, the
energy management strategy is extended to real-time by utilizing
Markov chain and stochastic dynamic programming (SDP).
This real-time implementation ensures both good optimization
accuracy and exempt of future driving information.
II. QUANTITATIVE BATTERY DEGRADATION MODEL
Battery cycle life (N ) is defined as the number of
charge/discharge cycles a battery is able to support before it
fails to meet the specific performance criteria. Generally, the
end-of-life battery capacity is defined as 80% of its original
capacity, namely, the battery capacity loss reaches 20%. Key
LU AND WANG: OPTIMAL SIZING AND ENERGY MANAGEMENT FOR COST-EFFECTIVE PEV HYBRID ENERGY STORAGE SYSTEMS 3409

TABLE I
VALUES OF B VERSUS CRATE

factors affecting battery cycle life are time, temperature, current

rate, and depth of discharge (DoD)[27], [35].
In this article, we build a quantitative battery degradation
model to measure both the dynamic capacity loss and cycle life
for batteries in automotive applications. The battery degradation
model is based on a semi-empirical model proposed by Wang
et al. [21]. This semi-empirical life model is established as
 
−Ea Fig. 1. Estimated Qloss versus Ah-throughput under different currents.
Qloss = B · exp (Ah )z (1)
RT
where Qloss is the battery capacity loss in percentage, B rep- where Cp represents the battery heat capacity, Q̇ is the rate of
resents the pre-exponential factor, Ea is the activation energy, heat generation, and qn the heat flux transmitted to the outside
R the gas constant (8.31 J/mol/K), T the absolute battery tem- which could be calculated by
perature, and Ah represents the Ah-throughput, the amount of
charge transported by the battery, while z is the power law factor qn = −hout (Tamb − Tsurf ) = hin (T − Tsurf ) (6)
with a value of 0.55. It is investigated that Ea decreases with
the increase of battery current. The relationship between them where Tsurf is the battery surface temperature, Tamb the ambient
is fitted as temperature far from the battery surface, and hout and hin the heat
transfer coefficients outside and inside the battery, respectively.
Ea = −370.3Crate + 31 700 (2) Q̇ is expressed as

where Crate is the battery current rate. It is defined as the battery Q̇ = Ib (Ub − Vb ) (7)
discharge rate relative to its nominal capacity. The values of B
versus Crate are listed in Table I [21]. where Ub is the open-circuit voltage of the battery cell, and Vb
is the cell terminal voltage.
A. Dynamic Capacity Loss Estimation
C. Battery Cycle Life Estimation
According to the damage accumulation theory [36], the bat-
tery degradation model in (1) can be also used to estimate the Based on (1) and (2), the curves of Qloss versus Ah with respect
battery dynamic capacity loss during a short period of time [37]. to different Crate are plotted in Fig. 1. As shown, 1) the increase
Based on this criterion, we derive the discrete formula of (1) as in Ah and the increase in Qloss are not linearly related; 2) the
variation of Crate has a significant impact on battery life. Those
 − z1
1 Ea z−1 factors bring some difficulties to the cycle life estimation in PEV
Qloss,k+1 − Qloss,k = ΔAh zB · exp
z · Qloss,k
z
(3) applications, in which batteries are experiencing nonuniform
RT
current profiles. Besides, due to the nonlinearity characteristic,
where Qloss,k+1 and Qloss,k are the accumulated capacity loss at the dynamic capacity loss of battery in one time instant is not
time instants tk+1 and tk , respectively. Hence, their difference only dependent on Ib , but also closely related to the initial state
is exactly the dynamic capacity loss during the time period from of the accumulated Qloss at the beginning of this step. To address
tk to tk+1 . ΔAh is the Ah-throughput from tk to tk+1 , which is this issue, we propose an iterative testbench to simulate the
expressed as accumulation of Qloss for a long time. Steps are as follows:
 tk+1 1) Assuming the initial battery SOC is 100%, define the
1
ΔAh = |Ib |dt (4) battery current profile during a certain drive cycle, and
3600 tk
set an initial Qloss .
where Ib is the battery current. 2) Start a new drive cycle. Calculate the accumulated Qloss
step by step according to (3), until this drive cycle ends.
B. Battery Thermal Model 3) Update SOC of the battery, and repeat step 2), un-
til the battery is fully discharged, which means this
The battery temperature T changes during the charging and
charge/discharge cycle ends. Then, the battery is charged
discharging process. According to the energy balance in the
to 100% SOC.
battery cell [38], it is derived that
4) Repeat steps 2) and 3), until the accumulated Qloss reaches
∂T 20%.
Cp = −qn + Q̇ (5) 5) Record the total charge/discharge cycles N .
∂t

Authorized licensed use limited to: Anelis Plus consortium. Downloaded on January 11,2021 at 10:42:51 UTC from IEEE Xplore. Restrictions apply.
3410 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 16, NO. 5, MAY 2020

TABLE III
BASIC SPECIFICATIONS OF THE UC CELL

TABLE IV
MAIN VEHICLE PARAMETERS

Fig. 2. Diagram of PEV powertrain with a full-active battery/UC HESS.

TABLE II
BASIC SPECIFICATIONS OF THE BATTERY CELL

B. UC Modeling
The BCAP0350 UC cell from Maxwell Technologies is con-
sidered in this work. The basic specifications of the UC cell
are listed in Table III. Since computation efficiency is crucial
in this study, RC model is adopted to emulate the behaviors of
the UC bank owing to its simplicity and acceptable accuracy, as
shown in Fig. 3(b) [17]. In this figure, the UC bank is equivalent
to an ideal capacitor (CC ) in series with an internal resistor
(RC ). Assuming that the UC bank consists of ncs UC cells in
series with ncp cells in parallel, the main parameters of the UC
Fig. 3. Equivalent circuit models for battery pack and UC bank. bank are
ncs ncp
UC = ncs Uc , RC = R c , CC = Cc (9)
ncp ncs
III. SYSTEM MODELING
where Uc and UC are the open-circuit voltages of the UC cell
The considered system configuration is a full-active bat- and the UC bank, respectively.
tery/UC topology. The PEV powertrain is shown in Fig. 2.
C. Powertrain Modeling
A. Battery Modeling
The main vehicle parameters are listed in Table IV. The power
In this article, we choose the LiFePO4 -26650-3300 battery demand is calculated in either traction mode or regenerative
cell supplied by AA Portable Power Corp as case study. The main braking mode
product specifications are shown in Table II. The PEV battery 
pack is composed of nbs battery cells in series and nbp cells in Pdemand,Tra = M gf v cos α + 0.5CD Aρv 3
parallel. Rint model is used to describe the battery dynamics, as
shown in Fig. 3(a): the battery pack is modeled as an equivalent dv  1
+ Mv + M gv sin α · (10)
circuit composed of an open-circuit voltage source (UB ) in series dt ηT ηm
with an internal resistance (RB )[12]. This model structure is 
linear in parameters and is relatively simple to implement in Pdemand,Reg = M gf v cos α + 0.5CD Aρv 3
practical applications [23]. Since the optimization is based on
dv 
the time-discretized problem, this simple static battery model is + Mv + M gv sin α · ηr (11)
applied. The basic parameters of the battery pack are expressed dt
as where g is the gravitational acceleration, v the PEV velocity, α
nbs the climbing angle, and ρ the air density [26], [37]. M is the total
UB = nbs Ub , RB = Rb , QB = nbs nbp Qb (8) vehicle loaded mass, including m0 and the HESS mass, which
nbp
is mathematically expressed as
where Ub is the open-circuit voltage of the battery cell, while
QB is the nominal capacity of the battery pack. M = m0 + 1.14(mb nbs nbp + mc ncs ncp ). (12)

Authorized licensed use limited to: Anelis Plus consortium. Downloaded on January 11,2021 at 10:42:51 UTC from IEEE Xplore. Restrictions apply.
LU AND WANG: OPTIMAL SIZING AND ENERGY MANAGEMENT FOR COST-EFFECTIVE PEV HYBRID ENERGY STORAGE SYSTEMS
Fig. 4. Proposed dual-loop control flowchart.
It should be noted that the weight of the HESS contains a
1.14 times multiplier to account for the additional mass of the
two dc–dc converters [26].
IV. INTEGRATED DUAL-LOOP OPTIMIZATION
The optimal energy management problem plays a key role in
performance enhancement and is closely coupled with the sizing
problem. Thus, an integrated dual-loop optimization framework
is proposed to deal with the two problems simultaneously, as
shown in Fig. 4: 1) the inner loop derives the optimal energy
management strategy to minimize battery Qloss based on DP
approach; 2) the outer loop utilizes SA algorithm to obtain the
best sizing solution. The design philosophy is to minimize the
average operating cost, while the battery size is constrained to
the requested minimal electric mileage.
V. DP-BASED OPTIMAL ENERGY MANAGEMENT STRATEGY
The main target for the energy management is to minimize
the accumulated capacity loss for the battery pack during each
drive cycle, which can be expressed as
n
 Pcn = Ucn Ic,max the maximum UC cell power, and pdc =
min ΔQloss = (Qloss,k+1 − Qloss,k )
k=1
⎧ bank, respectively.
⎪
⎪ Pdemand = PB + PC
⎪
⎨ −IB,max ≤ IB ≤ IB,max
s.t. . (13)
⎪
⎪ −IC,max ≤ IC ≤ IC,max
⎪
⎩ Due to UC’s feature of low energy density, the electric mileage
10% ≤ SOCc ≤ 90%
where n is the total time instants the battery will experience in
each drive cycle. The SOC window of UC ranges from 10% to
90% to avoid overcharging and overdischarging.
DP algorithm is utilized to solve this problem, as shown in
Fig. 5. The UC bank voltage UC is the initial state of this
process, while it is discretized to m states from UC,min to UC,max .
The time step is set as 1 s. The circles in this figure form the
discretized solution space for performing the DP algorithm.
Starting from the initial state, UC could select any path to
Authorized licensed use limited to: Anelis Plus consortium. Downloaded on January 11,2021 at 10:42:51 UTC from IEEE Xplore. Restrictions apply.
3411
Fig. 5. DP flowchart.
the final state. Based on m-discretized UC voltage states and
n-discretized time steps, there are mn search paths in total. For
each time step, once UC is determined, all the other information
including UB , IB , SOCC , and Qloss can be derived. Then, DP
algorithm provides a backward search method to find the optimal
search path to minimize ΔQloss .
VI. SA-BASED OPTIMAL SIZING
A. Cost-Minimization Principle
The design philosophy of the HESS sizing is to minimize its
average operating cost. The average operating cost is defined as
the ratio between the total HESS cost and the cycle life (N ) of the
battery pack (the cycle life for UC is typically more than 500 000
cycles which is much longer than battery life, and therefore, it
is not supposed to expire before the battery), given as
nbs nbp (pb + Pbn pdc ) + ncs ncp (pc + Pcn pdc )
min cost =
N
(14)
where Pbn = Ubn Ib,max is the maximum battery cell power,
50 USD/kW the dc–dc converter cost. The two dc–dc converters
are designed to handle the power of the battery pack and the UC
B. Battery Sizing Constraints
of the vehicle is proportional to the stored energy in the bat-
tery pack. Therefore, the battery sizing is constrained to the
requested minimal electric mileage. It is known that the PEV
electric mileage is dependent on the driving conditions and
energy management strategy [39]. For a fair comparison, the
vehicle is designed to drive at Urban Dynamometer Driving
Schedule (UDDS) on a flat road (α = 0). Assume the requested
minimal driving mileage is 20 repetitions of such drive cycles
(approximately 240 km), while the DoD for battery is 90%.
Then, the relationship between the requested energy and the
3412
Fig. 6. SA flowchart.
total capacity of the battery pack is derived as
 tn
20 Pdemand (t) · dt ≤ 90% × 3600Vb Cb nbs nbp .
t0
According to (15), the number of battery cells nbs nbp is
related to the power demand of the vehicle.
C. SA Algorithm
SA algorithm is utilized to find the optimal sizing solution of
the HESS. As an integrated optimization problem, the problem
of solving the optimal energy management strategy by DP is also
nested in the SA algorithm. The name and inspiration for SA al-
gorithm originate from annealing in metallurgy. It simulates the
physical process of heating material, and then slowly lowers the
temperature to minimize the system energy [40]. In this article,
as a probabilistic technique, SA selects a possible HESS sizing
solution first, and updates this solution based on the Metropolis
criterion [41] to minimize the cost. The algorithm flowchart
for this work is shown in Fig. 6. Here, J is the time-varying
parameter to describe the temperature state, and the coefficient
r is used to update J. The solving steps are listed as follows:
1) Choose an initial guess for ncs ncp .
2) Calculate the minimum values for M and nbs nbp accord-
ing to (15).
Authorized licensed use limited to: Anelis Plus consortium. Downloaded on January 11,2021 at 10:42:51 UTC from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 16, NO. 5, MAY 2020
3) Run the loop for optimal energy management based on
DP according to (13) and record the optimal searching
path.
4) Estimate the battery cycle life N according to Section II.
5) Calculate the system average operating cost following
(14) based on the defined sizing.
6) Choose to accept this solution or update UC size accord-
ing to the principle of SA.
7) Repeat steps 2)–6) until the convergence conditions are
achieved.
Theoretically, as a probabilistic technique, SA may produce
different results each time it is used. However, if it is used to
solve a relatively low-dimensional optimization problem, and
the parameters of the algorithm are well designed, the optimal
solution can be obtained each time. The parameter selection is
essential to secure a global minimum within acceptable calcu-
lation times. Based on practical experience, reasonable initial
guess, and some test results, we have effectively tuned the
parameters, including the initial temperature state J, the cooling
coefficient r, and the convergence conditions. Finally, we would
also utilize a large set of simulation data to validate the accuracy
and consistency of the solved results.
VII. RESULTS AND DISCUSSION
A. Sizing Optimization Results and Simulation Validation
By utilizing SA algorithm, a minimum system average oper-
ating cost of 15.52 USD is achieved with 4.33 kg UC cells. Since
mc for the example UC cell equals to 0.06 kg, the near-optimal
solution is 72 such UC cells. The number of the corresponding
battery cells is 7054. In order to meet the series-parallel specifi-
cation of the PEV battery pack, the near-optimal solution could
be selected as 7100 battery cells with nbs = 100 and nbp = 71.
We have run the algorithm for ten times, and it typically takes
(15) 37–40 computation times to reach the optimal solution. This
result also indicates that the proposed method is much more
efficient than using an exhaustive enumeration approach.
With the optimized design, the case study vehicle is equipped
with a 67-kWh usable battery pack (568 kg) and a small-sized
UC bank of 72 cells (4.32 kg). The total weight for the vehicle
is 1752 kg and the electric mileage is 240 km in the condition
of city test standard authorized by United States Environmental
Protection Agency. The results are reasonable for a middle-sized
PEV model. We can also conclude that with a small size auxiliary
UC bank and suitable energy management strategy, the system
average operating cost can be reduced effectively.
To validate the optimization results solved by SA algorithm, a
large set of simulation data is utilized to plot the curve between
UC mass and the average operating cost, as shown in Fig. 7. In
this figure, for each UC mass, the battery size is calculated based
on (15). Although it is an inequality constraint, the boundary
value of battery size is selected to calculate cost for this point.
As shown, the valley of the simulated curve is marked in this
figure, which is in accordance with the solution provided by
the adopted SA algorithm. Therefore, simulation results have
validated the optimality of the optimization results generated by
the SA algorithm.
LU AND WANG: OPTIMAL SIZING AND ENERGY MANAGEMENT FOR COST-EFFECTIVE PEV HYBRID ENERGY STORAGE SYSTEMS
Fig. 7. Relationship between UC mass and system average operating
cost.
Fig. 8. Velocity and power demand profiles of the vehicle under UDDS
(the power demand is calculated using the optimized HESS size).
Fig. 9. Power distribution between battery and UC in the optimized
HESS.
In simulation, the average operating cost for a battery-only
ESS is 17.62 USD, which demonstrates that the cost for the
optimized HESS is reduced by 11.9% compared with the battery-
only solution.
B. DP-Based Energy Management Strategy
The proposed DP-based energy management strategy is de-
rived based on the PEV with the optimized HESS running UDDS
drive cycle. The velocity profile during the drive cycle is shown
in Fig. 8. Along with the basic vehicle parameters in Table IV
as well as (10)–(12), the power demand profile for the PEV
equipped with the optimized HESS is also derived and plotted
in this figure.
The power distribution between battery and UC in the op-
timized HESS by utilizing the proposed energy management
strategy is shown in Fig. 9. It is demonstrated that the UC bank
Authorized licensed use limited to: Anelis Plus consortium. Downloaded on January 11,2021 at 10:42:51 UTC from IEEE Xplore. Restrictions apply.
3413
Fig. 10. Capacity loss path comparison for battery-only ESS and the
optimized HESS.
Fig. 11. Battery temperature comparison between battery-only ESS
and the optimized HESS.
is frequently charged/discharged while the battery pack realizes
reduced peak power and limited power change.
The battery capacity loss path during the first UDDS drive
cycle is simulated in Fig. 10. As shown, the accumulated Qloss
for the optimized HESS is 0.0308%, and this value is reduced
by 8.9% compared with that of the battery-only ESS (0.0338%).
This indicates that the proposed method is able to reduce battery
capacity loss, and further extend battery cycle life.
C. Temperature Comparison
The ambient temperature is set as 298 K. The battery heat
capacity is set as 73.2 J/K according to [38]. The battery
temperature variations during one drive cycle is illustrated in
Fig. 11. With the same heat dissipation system, the temperature
rise is approximately 0.31 K utilizing the proposed method,
versus 0.45 K of the battery-only ESS. The battery temperature
rise in the optimized HESS is about 31.1% less than that of
the battery-only ESS. Therefore, it is illustrated that the pro-
posed approach is more efficient in avoiding system overheating
and reducing energy loss, which is benefit for battery lifetime
extension [42].
D. Battery Cycle Life Estimation
The estimated charge/discharge cycle life for the battery pack
is described in Fig. 12. As shown, the battery cycle life for the
optimized HESS under 20 repetitions of the UDDS drive cycle
is 6241, while that for the battery-only ESS is 5129. This result
indicates that a 21.7% cycle life extension is achieved by the
proposed method.
3414
Fig. 12. Estimated battery cycle life for battery-only ESS and the
optimized HESS.
Note that a more significant improvement of battery cycle
life can be achieved by placing more UC cells. Similarly, other
performances including temperature rise and battery peak power
would also be improved in this way. However, the result that
comes with it is the increase in the cost of the UC and its linking
dc–dc converter, as well as the increase in the weight and power
demand of the vehicle under the same driving conditions. They
are all essential factors affecting the system average operating
cost. Note that we are trying to find the most economical choice
for the whole system, instead of simply boosting battery cycle
life by stacking UC cells.
VIII. REAL-TIME IMPLEMENTATION
Although DP provides an optimal energy management solu-
tion during the whole drive cycle, it could not work on its own in
real-time due to the requirements for future driving information.
In order to address this issue, an SDP-based real-time energy
management strategy is proposed. This strategy refers to the
optimal results obtained by the dual-loop optimization, and
develops a Markov chain to describe the stochastic process of the
driving cycle. By calculating the future probability distribution
of drive cycle and minimizing the cost function in a certain
horizon, an optimal energy management strategy is derived in
real-time.
The Markov process is based solely on its present state, and
hence independent from the future and past states. This means
the SDP-based optimal energy management does not need any
historical or future data and could be implemented in real-time.
Of course, the probability distribution information for the case
study road is needed. In order to acquire this information, the
typical drive cycles are needed, which should have already been
reported by the government or other authorized agencies and
are simple to obtain. Then, we could use the above information
to predict the future data based on the present data during driv-
ing process, and implement the real-time optimization strategy
simultaneously.
Specifically, the Markov transition probability between each
driving state can be extracted from the existing driving
data [43]. The system is described by the state vector xk =
[Pdemand,k , vk , Uc,k ]T . Similar with DP, the control signal of SDP
is the discretized UC voltage UC .
ΔQloss is dependent on the driving states, and the expectation
of ΔQloss can be calculated by using the statistical distribution
Authorized licensed use limited to: Anelis Plus consortium. Downloaded on January 11,2021 at 10:42:51 UTC from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 16, NO. 5, MAY 2020
Fig. 13. Comparison of battery power, UC voltage, and capacity loss
trajectories of DP and SDP.
of driving states. At each time step, the present driving state
is used to calculate the distribution of the driving states in the
next l quasi-steps according to the pre-statistic Markov transition
probability between states. Then the expectation of ΔQloss of the
next l quasi-steps is searched and backward minimized. Only
the optimized control action at the first quasi-step is applied in
the real-time control.
The simulation results for the SDP-based energy management
strategy are shown in Fig. 13. The battery power and UC voltage
trajectories in one UDDS drive cycle are compared with those of
DP algorithm. Both battery power trajectories are relieved from
the peak power to some degree, especially for the regenerative
braking power. However, when applying DP, regenerative energy
is solely stored in UC, while with the SDP strategy, a small
part of the regenerative energy is stored in the battery. For SDP,
although the utilization of UC is less than DP, the optimization
effect on battery capacity loss is very close to that of DP. As
shown in Fig. 13, the accumulated capacity loss in one UDDS
cycle for SDP is 0.0315%, which is close to that value of DP
(0.0308%), and is 6.9% lower than battery-only ESS.
IX. CONCLUSION
Based on cost-minimization principle, this article suggested a
novel dual-loop framework which enabled optimized sizing and
energy management simultaneously for a PEV equipped with
a battery/UC HESS. A quantitative battery degradation model
considering thermal effect was developed to evaluate both the
dynamic and long-time capacity loss for automotive applications
LU AND WANG: OPTIMAL SIZING AND ENERGY MANAGEMENT FOR COST-EFFECTIVE PEV HYBRID ENERGY STORAGE SYSTEMS
with varying load profile. This degradation model also facilitates
the following optimization method.
Within the inner loop, the energy management strategy to min-
imize battery capacity loss during each drive cycle was solved
using DP algorithm, whereas this energy management loop was
nested in the outer loop regarding sizing optimization. In the
outer loop, vehicle electric mileage requirements, battery cycle
life, costs and masses of battery, UC as well as dc–dc converters
were comprehensively considered and mapped into the average
operating cost. Finally, SA algorithm was utilized to find the
most cost-effective solution for battery and UC sizing. Results
indicate that by employing 7100 battery cells with 72 UC cells,
the average operating cost for vehicle use under 20 repetitions
of the UDDS drive cycle could be minimized to 15.52 USD.
The optimality of the optimization results was validated by a
large set of simulation data. Compared with the battery-only
ESS, the average operating cost for the optimized battery/UC
HESS was reduced by 11.9%. The accumulated capacity loss
and temperature rise for the battery pack during one UDDS drive
cycle were decreased by 8.9% and 31.1%, respectively, and the
battery cycle life was prolonged by 21.7%.
Finally, based on the optimal results, the DP energy man-
agement strategy was extended to real-time applications with
Markov chain and SDP algorithm. Exempt from historical or
future data, SDP guarantees real-time applications while ensur-
ing the comparable optimization effects to DP.
This article provided instructive guidance on how to design
the most cost-effective battery/UC HESS for PEVs, and the pro-
posed energy management strategy solved by DP and SDP can be
also used in other energy management systems with predefined
sizing. Future work will address the considerations of further
reducing the computing time and enhancing the optimization
accuracy of SDP.
REFERENCES
[1] Amin, R. T. Bambang, A. S. Rohman, C. J. Dronkers, R. Ortega, and
A. Sasongko, “Energy management of fuel cell/battery/supercapacitor
hybrid power sources using model predictive control,” IEEE Trans. Ind.
Inform., vol. 10, no. 4, pp. 1992–2002, Nov. 2014.
[2] M. Hannan, M. Hoque, A. Mohamed, and A. Ayob, “Review of energy
storage systems for electric vehicle applications: Issues and challenges,”
Renewable Sustainable Energy Rev., vol. 69, pp. 771–789, 2017.
[3] G. Ren, G. Ma, and N. Cong, “Review of electrical energy storage system
for vehicular applications,” Renewable Sustainable Energy Rev., vol. 41,
pp. 225–236, 2015.
[4] E. Chemali, M. Preindl, P. Malysz, and A. Emadi, “Electrochemical and
electrostatic energy storage and management systems for electric drive
vehicles: State-of-the-art review and future trends,” IEEE J. Emerging Sel.
Topics Power Electron., vol. 4, no. 3, pp. 1117–1134, Sep. 2016.
[5] Q. Wang, P. Ping, X. Zhao, G. Chu, J. Sun, and C. Chen, “Thermal runaway
caused fire and explosion of lithium ion battery,” J. Power Sources,
vol. 208, pp. 210–224, 2012.
[6] C. G. Hochgraf, J. K. Basco, T. P. Bohn, and I. Bloom, “Effect of
ultracapacitor-modified PHEV protocol on performance degradation in
lithium-ion cells,” J. Power Sources, vol. 246, pp. 965–969, 2014.
[7] Y.-H. Hung and C.-H. Wu, “An integrated optimization approach for a
hybrid energy system in electric vehicles,” Appl. Energy, vol. 98, pp. 479–
490, 2012.
[8] H. Yin, W. Zhou, M. Li, C. Ma, and C. Zhao, “An adaptive fuzzy
logic-based energy management strategy on battery/ultracapacitor hybrid
electric vehicles,” IEEE Trans. Transp. Electrification, vol. 2, no. 3,
pp. 300–311, Sep. 2016.
Authorized licensed use limited to: Anelis Plus consortium. Downloaded on January 11,2021 at 10:42:51 UTC from IEEE Xplore. Restrictions apply.
3415
[9] Q. Xu, J. Xiao, X. Hu, P. Wang, and M. Y. Lee, “A decentralized power
management strategy for hybrid energy storage system with autonomous
bus voltage restoration and state-of-charge recovery,” IEEE Trans. Ind.
Electron., vol. 64, no. 9, pp. 7098–7108, Sep. 2017.
[10] Z. Chen, C. C. Mi, R. Xiong, J. Xu, and C. You, “Energy management of
a power-split plug-in hybrid electric vehicle based on genetic algorithm
and quadratic programming,” J. Power Sources, vol. 248, pp. 416–426,
2014.
[11] J. Bauman and M. Kazerani, “A comparative study of fuel-cell–battery,
fuel-cell–ultracapacitor, and fuel-cell–battery–ultracapacitor vehicles,”
IEEE Trans. Veh. Technol., vol. 57, no. 2, pp. 760–769, Mar. 2008.
[12] J. Shen and A. Khaligh, “A supervisory energy management control
strategy in a battery/ultracapacitor hybrid energy storage system,” IEEE
Trans. Transp. Electrification, vol. 1, no. 3, pp. 223–231, Oct. 2015.
[13] A. Santucci, A. Sorniotti, and C. Lekakou, “Power split strategies for hy-
brid energy storage systems for vehicular applications,” J. Power Sources,
vol. 258, pp. 395–407, 2014.
[14] Q. Zhang, W. Deng, and G. Li, “Stochastic control of predictive power
management for battery/supercapacitor hybrid energy storage systems of
electric vehicles,” IEEE Trans. Ind. Inform., vol. 14, no. 7, pp. 3023–3030,
Jul. 2018.
[15] M. Ibrahim, S. Jemei, G. Wimmer, and D. Hissel, “Nonlinear autoregres-
sive neural network in an energy management strategy for battery/ultra-
capacitor hybrid electrical vehicles,” Electric Power Syst. Res., vol. 136,
pp. 262–269, 2016.
[16] H. Yin, C. Zhao, and C. Ma, “Decentralized real-time energy management
for a reconfigurable multiple-source energy system,” IEEE Trans. Ind.
Inform., vol. 14, no. 9, pp. 4128–4137, Sep. 2018.
[17] X. Lu, Y. Chen, M. Fu, and H. Wang, “Multi-objective optimization-based
real-time control strategy for battery/ultracapacitor hybrid energy man-
agement systems,” IEEE Access, vol. 7, pp. 11640–11650, 2019.
[18] J. Cao, W. Du, H. Wang, and M. McCulloch, “Optimal sizing and control
strategies for hybrid storage system as limited by grid frequency devia-
tions,” IEEE Trans. Power Syst., vol. 33, no. 5, pp. 5486–5495, Sep. 2018.
[19] R. Bindu and S. Thale, “Sizing of hybrid energy storage system and
propulsion unit for electric vehicle,” in Proc. IEEE Transp. Electrification
Conf., 2017, pp. 1–6.
[20] M. Galeotti, L. Cinà, C. Giammanco, S. Cordiner, and A. Di Carlo,
“Performance analysis and SOH (state of health) evaluation of lithium
polymer batteries through electrochemical impedance spectroscopy,” En-
ergy, vol. 89, pp. 678–686, 2015.
[21] J. Wang et al., “Cycle-life model for graphite-LiFePO4 cells,” J. Power
Sources, vol. 196, no. 8, pp. 3942–3948, 2011.
[22] Z. Wang, J. Ma, and L. Zhang, “State-of-health estimation for lithium-ion
batteries based on the multi-island genetic algorithm and the Gaussian
process regression,” IEEE Access, vol. 5, pp. 21286–21295, 2017.
[23] S. Nejad, D. Gladwin, and D. Stone, “A systematic review of lumped-
parameter equivalent circuit models for real-time estimation of lithium-ion
battery states,” J. Power Sources, vol. 316, pp. 183–196, 2016.
[24] S. Allu, B. V. Asokan, W. A. Shelton, B. Philip, and S. Pannala, “A general-
ized multi-dimensional mathematical model for charging and discharging
processes in a supercapacitor,” J. Power Sources, vol. 256, pp. 369–382,
2014.
[25] L. Zhang, X. Hu, Z. Wang, F. Sun, and D. G. Dorrell, “Fractional-order
modeling and state-of-charge estimation for ultracapacitors,” J. Power
Sources, vol. 314, pp. 28–34, 2016.
[26] L. Zhang, X. Hu, Z. Wang, F. Sun, J. Deng, and D. G. Dorrell, “Multiobjec-
tive optimal sizing of hybrid energy storage system for electric vehicles,”
IEEE Trans. Veh. Technol., vol. 67, no. 2, pp. 1027–1035, Feb. 2018.
[27] J. Shen, S. Dusmez, and A. Khaligh, “Optimization of sizing and battery
cycle life in battery/ultracapacitor hybrid energy storage systems for
electric vehicle applications,” IEEE Trans. Ind. Inform., vol. 10, no. 4,
pp. 2112–2121, Nov. 2014.
[28] M. Masih-Tehrani, M.-R. Ha’iri-Yazdi, V. Esfahanian, and A. Safaei,
“Optimum sizing and optimum energy management of a hybrid energy
storage system for lithium battery life improvement,” J. Power Sources,
vol. 244, pp. 2–10, 2013.
[29] A. Jaafar, C. R. Akli, B. Sareni, X. Roboam, and A. Jeunesse, “Sizing
and energy management of a hybrid locomotive based on flywheel and
accumulators,” IEEE Trans. Veh. Technol., vol. 58, no. 8, pp. 3947–3958,
Oct. 2009.
[30] R. Sadoun, N. Rizoug, P. Bartholomeüs, B. Barbedette, and P. Le Moigne,
“Influence of the drive cycles on the sizing of hybrid storage system
battery-supercapacitor supplying an electric vehicle,” in Proc. IEEE 37th
Annu. Conf. IEEE Ind. Electron. Soc., 2011, pp. 4106–4112.
3416
[31] X. Hu, N. Murgovski, L. M. Johannesson, and B. Egardt, “Comparison of
three electrochemical energy buffers applied to a hybrid bus powertrain
with simultaneous optimal sizing and energy management,” IEEE Trans.
Intell. Transp. Syst., vol. 15, no. 3, pp. 1193–1205, Jun. 2014.
[32] L. Xu, C. D. Mueller, J. Li, M. Ouyang, and Z. Hu, “Multi-objective
component sizing based on optimal energy management strategy of fuel
cell electric vehicles,” Appl. Energy, vol. 157, pp. 664–674, 2015.
[33] R. E. Araujo, R. de Castro, C. Pinto, P. Melo, and D. Freitas, “Combined
sizing and energy management in EVs with batteries and supercapacitors,”
IEEE Trans. Veh. Technol., vol. 63, no. 7, pp. 3062–3076, Sep. 2014.
[34] G. Li, Z. Yang, B. Li, and H. Bi, “Power allocation smoothing strategy for
hybrid energy storage system based on Markov decision process,” Appl.
Energy, vol. 241, pp. 152–163, 2019.
[35] B. Lunz, Z. Yan, J. B. Gerschler, and D. U. Sauer, “Influence of plug-in
hybrid electric vehicle charging strategies on charging and battery degra-
dation costs,” Energy Policy, vol. 46, pp. 511–519, 2012.
[36] M. Safari, M. Morcrette, A. Teyssot, and C. Delacourt, “Life-prediction
methods for lithium-ion batteries derived from a fatigue approach. I.
Introduction: Capacity-loss prediction based on damage accumulation,”
J. Electrochemical Soc., vol. 157, no. 6, pp. A713–A720, 2010.
[37] Z. Song, H. Hofmann, J. Li, X. Han, X. Zhang, and M. Ouyang, “A
comparison study of different semi-active hybrid energy storage system
topologies for electric vehicles,” J. Power Sources, vol. 274, pp. 400–411,
2015.
[38] C. Forgez, D. V. Do, G. Friedrich, M. Morcrette, and C. Delacourt,
“Thermal modeling of a cylindrical LiFePO4 /graphite lithium-ion battery,”
J. Power Sources, vol. 195, no. 9, pp. 2961–2968, 2010.
[39] H. A. Bonges III and A. C. Lusk, “Addressing electric vehicle (EV) sales
and range anxiety through parking layout, policy and regulation,” Transp.
Res. Part A: Policy Pract., vol. 83, pp. 63–73, 2016.
[40] F. Y. Vincent, A. P. Redi, Y. A. Hidayat, and O. J. Wibowo, “A simulated
annealing heuristic for the hybrid vehicle routing problem,” Appl. Soft
Comput., vol. 53, pp. 119–132, 2017.
[41] A. R. Al-Roomi and M. E. El-Hawary, “Metropolis biogeography-based
optimization,” Inf. Sci., vol. 360, pp. 73–95, 2016.
[42] Z. Ma, J. Jiang, W. Shi, W. Zhang, and C. C. Mi, “Investigation of
path dependence in commercial lithium-ion cells for pure electric bus
applications: Aging mechanism identification,” J. Power Sources, vol. 274,
pp. 29–40, 2015.
[43] C. Vagg, S. Akehurst, C. J. Brace, and L. Ash, “Stochastic dynamic
programming in the real-world control of hybrid electric vehicles,” IEEE
Trans. Control Syst. Technol., vol. 24, no. 3, pp. 853–866, May 2016.
Authorized licensed use limited to: Anelis Plus consortium. Downloaded on January 11,2021 at 10:42:51 UTC from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 16, NO. 5, MAY 2020
Xiaoying Lu received the B.S. degree in mi-
croelectronics from the Chongqing University
of Posts and Telecommunications, Chongqing,
China, in 2016, and the M.S. degree in micro-
electronics and solid-state electronics from the
Shanghai Institute of Microsystem and Informa-
tion Technology, Chinese Academy of Sciences,
Shanghai, China, in 2019.
She was an integrated circuit (IC) Validation
Engineer Intern with the Freescale Semiconduc-
tor Limited, Shanghai, China, in Spring 2015.
From 2016 to 2019, she worked as a Graduate Research Assistant with
the Power Electronics and Renewable Energies Laboratory, School of
Information Science and Technology, ShanghaiTech University, Shang-
hai, China. She is currently a Hardware Engineer of Board-Mounted
Power (BMP) Team with the Cisco Systems (China) Research and
Development Company, Ltd., Shanghai, China, where she is involved
with Cisco core-switching products.
Haoyu Wang (S’12–M’14–SM’18) received the
bachelor degree (with distinguished honor) in
electrical engineering from the Zhejiang Univer-
sity, Hangzhou, China. He received the Ph.D.
degree in electrical engineering from the Univer-
sity of Maryland, College Park, MD, USA.
Since 2014, he has been a Tenure Track As-
sistant Professor in Electrical Engineering with
the School of Information Science and Technol-
ogy, ShanghaiTech University, Shanghai, China.
His research interests include power electron-
ics, plug-in electric and hybrid electric vehicles, the applications of wide
bandgap semiconductors, renewable energy harvesting, and power
management integrated circuits.
Dr. Wang is an Associate Editor of IEEE TRANSACTIONS ON TRANS-
PORTATION ELECTRIFICATION, and an Associate Editor of CPSS Transac-
tions on Power Electronics and Applications.