Applied Energy 185 (2017) 1654–1662

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

Model predictive control for power management in a plug-in hybrid

electric vehicle with a hybrid energy storage system q
Shuo Zhang a, Rui Xiong a,b,⇑, Fengchun Sun a,b
a
National Engineering Laboratory for Electric Vehicles, School of Mechanical Engineering, Beijing Institute of Technology, Beijing 100081, China
b
Collaborative Innovation Center of Electric Vehicles in Beijing, Beijing Institute of Technology, Beijing 100081, China

h i g h l i g h t s

An integrated power management for a PHEV with multi-energy sources was proposed.

A model predictive controller was used to regulate the power allocation of the HESS.

The robustness of the proposed approach was verified by three typical driving cycles.

The results show that the proposed control strategy can promote fuel economy greatly.

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

Article history: The fuel economy performance of plug-in hybrid electric vehicles (PHEVs) strongly depends on the power
Received 18 August 2015 management strategy. This study proposes an integrated power management for a PHEV with multiple
Received in revised form 19 November 2015 energy sources, including a semi-active hybrid energy storage system (HESS) and an assistance power
Accepted 8 December 2015
unit (APU). The HESS consists of battery packs and ultracapacitor packs. In the integrated control strategy,
Available online 28 December 2015
the output power between the battery packs and ultracapacitor packs is regulated by the model predic-
tive control strategy, while the output power between the APU and HESS is allocated by the rule-based
Keywords:
strategy. In the model predictive control process, a period of the future velocity will be predicted, and the
Plug-in hybrid electric vehicle
Hybrid energy storage system
dynamic programming algorithm will be applied to optimize the control strategy accordingly. The
Assistant power unit robustness of the proposed approach is verified by three typical driving cycles, including the
Model predictive control Manhattan cycle, CBDC cycle and UDDSHDV cycle. The results show that the proposed control strategy
Dynamic programming can promote fuel economy compared with the original control strategy, especially in the charge sustain
Power management mode under the MANHATTAN driving cycle (21.88% improvement).
Ó 2015 Elsevier Ltd. All rights reserved.

1. Introduction of the vehicle is low, the engine can be turned off and the vehicle
can be driven by the electric power system, which is helpful for
Fossil fuel depletion and serious air pollution have greatly improving the efficiency performance. Moreover, the application
encouraged the development of plug-in hybrid electric vehicles of a hybrid energy storage system (HESS), a combination of a bat-
(PHEVs) [1,2]. Compared with the pure electric vehicles, PHEVs tery and an ultracapacitor, can effectively reduce the charging/dis-
have a longer driving range because, when the battery’s state of charging rate of the battery and prolong the battery life. For a
charge (SoC) is low, the engine can keep the vehicle working for series PHEV system with HESS, the fuel economy performance
an additional driving range similar to that of a conventional vehi- highly depends on how the output power from the assistant power
cle. Conversely, when the vehicle speed or the power requirement unit (APU), battery packs and ultracapacitor packs is allocated.

q
This paper was presented at the 7th International Conference on Applied Energy 1.1. Literature review
(ICAE2015), March 28–31, 2015, Abu Dhabi, UAE (Original paper title: ‘‘Design and
evaluate of optimal control strategy for hybrid power system used in plug-in hybrid Much valuable work involving the power management of elec-
electric vehicle” and Paper No.: 88).
⇑ Corresponding author at: National Engineering Laboratory for Electric Vehicles, tric vehicles has been widely conducted by many researchers [3,4].
School of Mechanical Engineering, Beijing Institute of Technology, Beijing 100081, The most basic and widely applied power management strategy is
China. Tel./fax: +86 (10) 6891 4070. the rule-based control strategy [5], including fuzzy logic-based
E-mail address: rxiong@bit.edu.cn (R. Xiong). control strategy [6,7] that can be recognized as a combination of

http://dx.doi.org/10.1016/j.apenergy.2015.12.035
0306-2619/Ó 2015 Elsevier Ltd. All rights reserved.
 S. Zhang et al. / Applied Energy 185 (2017) 1654–1662 1655

different control rules. Ref. [6] proposed a fuzzy logic-based control 1.2. Motivation and innovation
strategy for parallel hybrid vehicles to determine the output power
of the engine and motor/generator, and the results showed poten- This study proposes an integrated power management system
tial improvements. The merit of the rule-based control strategy is for PHEVs that includes three power sources: battery packs, ultra-
its easy implementation, but the performance of this type of con- capacitor packs and an APU. In the new control strategy, an MPC-
trol strategy usually depends on the engineering experience of based controller was integrated with the original control strategy.
the designer, and its robustness cannot be guaranteed. To further The output power of the engine and HESS is determined according
improve the performance of the power management for electric to the original control rules, and the control of the battery packs
vehicles, various intelligent algorithms have been introduced and ultracapacitor packs will be realized by the MPC controller.
[8,9]. In Ref. [8], particle swarm optimization algorithm was The integrated control strategy will be verified by three different
employed for the power management of a hybrid electric vehicle types of driving cycles: Manhattan drive cycle, CBDC drive cycle
and great improvement was obtained. In Ref. [9], the simulated and UDDSHDV drive cycle.
annealing algorithm was applied to determine the engine power
and maximum current coefficient for power-split plug-in hybrid 1.3. Organization of the paper
electric vehicles. Though the promotion of the efficiency perfor-
mance can be achieved, the optimal performance in real-time via The organization of this study is as follows: Section 2 illustrates the
an intelligent algorithm-based control strategy can hardly be modeling of the system and the original control strategy. Then, the
guaranteed. integrated control strategy and MPC formulation will be presented
Some researchers have attempted to optimize the power man- in Section 3. After that, the verification and discussion will be illus-
agement by various optimization algorithms such as convex opti- trated in Section 4. Finally, conclusions are provided in Section 5.
mization algorithm and PSO algorithm [10,11], dynamic
programming optimization (DP) [12–14] and Pontryagin’s mini-
mum principle (PMP) [15,16], among which the most powerful 2. System modeling and original control strategy
method is DP. The DP algorithm can locate the global optimal con-
trol strategy, but because DP-based power management requires 2.1. PHEV configuration and modeling
future road and vehicle speed information, this algorithm cannot
be applied directly. Moreover, when the model is complex, the cal- Many different types of topologies for the HESS have been devel-
culation time for the DP algorithm will be unacceptable. Another oped, including active HESS topologies, semi-active HESS topologies
popular optimization algorithm is PMP where an optimal control and passive HESS topologies [29–32]. In this study, a semi-active
strategy can be obtained by minimizing a Hamiltonian function HESS topology is applied in our target system, as displayed in
[15]. Though the PMP-based method cannot ensure the global opti- Fig. 1 [19]. The battery packs and ultracapacitor packs compose
mization, the control performance from PMP is sometimes similar the HESS, and the engine and generator compose the APU system.
to the performance from DP [16]. However, as the co-states in the The output power of the battery packs can be controlled by a DC/
Hamiltonian function are determined by a trial-and-error method, DC converter, while the output power of the ultracapacitor packs
the PMP-based method cannot be applied directly [17,18]. is controlled passively. To describe the dynamic performance of
To propose a robust and implementable control strategy, driv- the PHEV, a simplified but sufficiently complex backward simula-
ing pattern recognition (DPR)-based power management strategies tion model was constructed, including battery packs, ultracapacitor
have been developed [19]. For DPR-based methods, researchers packs, DC/DC converter, APU, vehicle and transmission system.
have attempted to design different control strategies according to
different driving conditions, and once the driving condition is iden- (1) Battery pack. To analyze the dynamic performance of a bat-
tified by the DPR module, the corresponding control strategy will tery pack, the Thevenin battery lumped parameters model
be applied [19,20]. Thus, the performance of DPR-based power has been selected, and its operation process can be given
management strongly depends on the identification precision. by the following equation [33]:
However, as the DPR module usually identifies the current driving (
condition according to historical driving information and assumes U_ D ¼  C D1RD U D þ C1D iL
ð1Þ
that the driving condition will not change, misidentification cannot U t ¼ U oc  U D  iL Ri
be avoided [21]. Some researchers have attempted to design the
power management via model predictive control (MPC), which will where UD and Ut denote the diffusion voltage and terminal
predict a period of vehicle velocity and then apply various opti- voltage for the battery pack, respectively, CD and RD denote
mization methods to design the control strategy based on the pre- the diffusion capacitance and the diffusion resistance for an
dicted velocity. The vehicle velocity could be predicted by an RC network, which is used to represent the dynamic voltage
artificial neural network algorithm [22], torque requirement expo- performances and the mass transport effects, iL denotes the
nentially decreasing model [23] or Markov Chain-based velocity battery output current, Uoc denotes the open circuit voltage,
predictor [24], and the control strategy optimization methods and Ri denotes the electrical resistance of various battery
could be the PMP algorithm [25], DP algorithm [26], nonlinear pro- components. The battery packs include 138 LiMn2O4
gramming algorithm [27] or quadratic programming algorithm lithium-ion battery cells, and they are connected in series.
[28]. The nominal capacity of the battery cells is 77 Ah, and the
Most of the above work focuses on the power management nominal voltage is 3.7 V. The upper and lower cutoff voltages
strategy for the vehicles only with two power sources, and few are 4.2 V and 3.0 V, respectively. The identification results for
of them for the PHEVs with three or more power sources, including these parameters including Uoc, CD, RD and Ri under different
battery packs, ultracapacitor packs and an APU. Though an adap- SoC levels are presented in Fig. 2 (the identification method
tive control strategy for a PHEV containing three energy sources comes from Ref. [34]).
has been studied in Ref. [19], the HESS is regarded as a single (2) Ultracapacitor pack. The ultracapacitor pack is modeled as a
source during its operation process. Because the output power of series combination of an idle capacitor and resistance Rc
the battery and ultracapacitor was determined by rule-based con- [32]. The operation process of the capacity can be expressed
trol once the required power is deterministic. by the following equation:
1656 S. Zhang et al. / Applied Energy 185 (2017) 1654–1662

APU

AC/DC
converter
Engine Generater

DC/AC
converter
HESS

Battery DC/DC
converter Transmission
pack Motor
system

Ultracapacitor

Fig. 1. The power system configuration of the PHEV.

0.2 (4) Engine-generator system. In the target PHEV, the APU is

mechanically decoupled from the transmission system,
Ri (Ω)

0.15 which indicates that the APU can always work according
to its optimal fuel rate line and is not limited to the current
0.1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 driving conditions. The optimal fuel rate line is developed by
0.4 combining the efficiency map of the engine and generator as
RD (Ω)

displayed in Fig. 3 [19]. The engine is a 1.9 L gasoline engine,

0.2
and the generator is an 80 kW permanent magnetic genera-
0 tor (the rated power for the generator is 130 kW). The rated
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
power of the APU is 75 kW. At each time step, the fuel con-
600
sumption of the engine can be obtained according to the
UOC (V)

500 output power of the APU.

(5) Vehicle and transmission system. As this study only considers
400 the PHEV’s longitudinal dynamics and does not consider its
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
400 other dynamic properties, the vehicle is regarded as a point
mass. Thus, the power requirement can be calculated by the
CD (F)

200 following equation:

0  
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 C ar A 2 dM dua ua
Pn ¼ Mgf cosðaÞ þ Mg sinðaÞ þ ua þ
SoC 21:15 3:6 dt 3:6gT
Fig. 2. The identified parameters of the battery pack. ð3Þ

where ua denotes the vehicle speed (km/h), M denotes the

U ct ¼ U co  Rc ic ð2Þ vehicle loaded mass (16,500 kg), f denotes the rolling resis-
tance coefficient (0.011), g denotes the gravitational accelera-
where Uct and Uco denote the terminal voltage of the ultraca- tion (9.81 m/s2), a denotes the grade of the road, Car denotes
pacitor pack and voltage across the capacitor, respectively. ic the air resistance coefficient (0.55), A denotes the windward
denotes the output current of the ultracapacitor. The ultraca- area of the vehicle (6.6 m2), d denotes the correction coeffi-
pacitor pack consists of 12 ultracapacitor modules connected cient of the rotation mass (1.03) and gT denotes the efficiency
in series. For each cell, the maximum voltage is 48 V, the of the transmission system (the efficiency of the transmission
resistance is 0.0063 X and the capacity is 165 F. Thus, the system was modeled as the fixed value 0.9).
maximum voltage of the ultracapacitor packs is 576 V, the
resistance is 0.0756 X and the capacity is 13.75 F.
(3) DC/DC converter. This study uses a response surface n(iDC,
PDC)-based efficiency map to describe the operation behavior 4.5
of the DC/DC converter, and it is shown in Table 1.
Fuel rate (g/s)

Table 1
Efficiency map of the DC/DC converter [32].
1.5

n(iDC, PDC) 10 kW 20 kW 30 kW 40 kW P50 kW

10 A 92% 95% 97% 95% 94% 0
0 20 40 60 80
50 A 91% 93% 96% 93% 92%
100 A 88% 91% 95% 92% 91% APU power (kW)
P150 A 82% 89% 92% 91% 90%
Fig. 3. The optimal fuel rate line of the APU.
 S. Zhang et al. / Applied Energy 185 (2017) 1654–1662 1657

2.2. Original control strategy In this study, an integrated power management is proposed as
displayed in Fig. 4. In the new strategy, the MPC is integrated with
For the power management of the PHEV, a type of Charge-Dep the above original control strategy to realize the power manage-
leting/Charge-Sustaining (CD/CS) strategy is applied in the original ment for the HESS (allocating the output power of the batteries
control strategy. In the CD condition, the PHEV will firstly work like and ultracapacitor when the output power of the HESS is deter-
a battery electric vehicle, and when the battery SoC (BSoC) depletes mined). For the operation of the APU, the original control rules will
to a predetermined level (where BSoC is 0.35), the PHEV will work still be applied. Specifically, the integrated control process includes
like a conventional HEV (the BSoC will be sustained at approxi- three main control procedures: power allocation control, lower
mately 0.3). The detailed original control strategy is displayed control and MPC control. For the power allocation control, the con-
below. troller will determine the output power between the APU and HESS
according to the original control rules. The MPC control will predict
(1) CD operation mode: when BSoC > 0.35, the vehicle is working the future power requirement for HESS, optimize the control strat-
with the CD mode in which the APU stops working (PAPU = 0) egy for HESS and output the control command IL to the lower con-
and the HESS provides the entire power requirement troller. The lower controller will control the APU controller and
(PHESS = Pn). For the HESS, the control rules of the battery HESS controller according to the received control command. To
are as follows: avoid frequent charging/discharging of the battery pack and make
If PHESS < 0 and USoC > USoCH, then Pb = PHESS and Puc = 0; the ultracapacitor work more efficiently, the lower controller will
If PHESS < 0 and USoC 6 USoCH, then Pb = 0 and Puc = PHESS; control the HESS according to two conditions: when the power
If 0 6 PHESS < PL, USoC > USoCL and BSoC > BSoCL, then Pb = - requirement of the HESS exceeds PL, the lower controller will con-
PHESS and Puc = 0; trol the HESS controller according to the control command from
If PL < PHESS, USoC > USoCL and BSoC > BSoCL, then Pb = PL and the MPC controller; and when the power requirement for the HESS
Puc = PHESS- PL; is lower than PL, the lower controller will control the DC/DC con-
If 0 6 PHESS, USoC 6 USoCL and BSoC > BSoCL, then Pb = PHESS verter to make the battery output the required power of the HESS,
and Puc = 0; which is similar to the original control. In this study, a series of
If 0 6 PHESS, BSoC 6 BSoCL and USoC > USoCL, then Puc = PHESS braking strategy was applied, which will make the HESS absorb
and Pb = 0; the braking energy as much as possible, and when the braking
If 0 6 PHESS, BSoC 6 BSoCL and USoC 6 USoCL, then Puc = 0 and power exceeds the absorption ability of HESS, the conventional
Pb = 0; friction braking mechanism will act and consume the remaining
power.
where USoC denotes the SoC of the ultracapacitor, USoCH and
USoCL denote the upper and lower control thresholds of the
3.2. MPC formulation
ultracapacitor and are initialized to be 0.99 and 0.51, respec-
tively, BSoCL denotes the battery control threshold for the bat-
The performance of the MPC-based control strategy highly
tery (set as 0.2), Pr denotes the rated power of the APU (75 kW)
depends on the prediction precision of the vehicle velocity. This
and PL denotes the power threshold (38.16 kW) for HESS.
study attempts to predict the future vehicle velocity by assuming
that the driver torque demand will decrease exponentially over
(2) CS operation mode: when BSoC 6 0.35, the vehicle is working
the prediction horizon and can be expressed by the following equa-
with the CS mode, in which the APU will work according to
tion [23]:
the following rules:
If Pn < 0, then PAPU = 0 and PHESS = Pn; ðis DtÞ
T Wheel ðk þ jÞ ¼ T Wheel ðkÞe d j ¼ 1; 2; . . . ; HF ð4Þ
If Pn > Pr, then PAPU = Pr and PHESS = Pn  PAPU;
If 0 6 Pn 6 Pr and BSoC 6 0.25, then PAPU = Pr until where Twheel denotes the driver torque demand acting on the
BSoC > 0.35 and PHESS = Pn  PAPU; wheels of the vehicle (positive for acceleration and negative for
If 0 6 Pn 6 Pr and BSoC > 0.25, then PAPU = 0 and PHESS = Pn; deceleration), j denotes the time index in the prediction horizon,
Dt denotes the sample time and is set to be one second, sd denotes
It is worth noting that, in the CS condition, the control rules for the the decay coefficient and is set to be seven, and HF denotes the
HESS are the same as with the CD condition when PHESS is given. length of the prediction horizon and is set to be ten in this study.
Then, the predicted speed can be expressed by the following
equation:
3. Integrated power management
T wheel ðkþjÞ
Rtire
 Mgf cosðaÞ  Mg sinðaÞ  21:15
C ar A 2
ua
3.1. Control logic of the integrated control strategy Vðk þ j þ 1Þ ¼ Vðk þ jÞ þ
M
To improve the fuel economy performance for the PHEV, the j ¼ 1; 2; . . . ; HF ð5Þ
MPC algorithm was applied to the new control strategy. The power
management strategy based on the MPC algorithm usually con-
tains three modules: the velocity prediction module, control strat- BSoC CSoC U D

egy optimization module, and control command selection module.

The velocity prediction module is used to predict a period of the Power allocation controller Lower controller
P PAPU iL PHEV
Pn ⎧ PHESS = Pn CD mode APU APU controller and
future vehicle speed according to the current or historical vehicle ⎨ model
⎩ PHESS = Pn − PAPU CS mode HESS controller
information. Then, the control strategy optimization module will
be applied to design the control strategy based on the predicted PHESS PAPU

velocity. After that, the control command selection module will IL

MPC controller
select the first control element of the optimized control strategy
as the control command, and the other control elements will be BSoC, CSoC, U D
abandoned. The three procedures will be repeated in each calcula-
tion step. Fig. 4. Operation process of the proposed power management.
1658 S. Zhang et al. / Applied Energy 185 (2017) 1654–1662

    ! 
where Rtire denotes the radius of the wheel. Based on the predicted U D ðk þ 1Þ exp Dt
0 U D ðkÞ
velocity, the control strategy for the HESS can be optimized. In this ¼ RD C D
BSoCðk þ 1Þ 0 1 BSoCðkÞ
study, the DP algorithm was applied to optimize the control strat- 0   1 ð6Þ
Dt
egy in the MPC controller. As the MPC control strategy is only used 1  exp RD C D RD
for the power management for the HESS, we only need to consider þ@ AiL ðkÞ
1
the state for the battery pack and ultracapacitor pack in the DP opti- 3600Q b

mization process. In the DP optimization process, the state variables

x(k) are the battery diffusion voltage UD(k), BSoC(k) and USoC(k). The USoCðkÞ  Q c  ic ðkÞ
USoCðk þ 1Þ ¼ ð7Þ
control variables are the battery output current iL(k). The state Qc
equations for the battery pack and ultracapacitor pack are pre-
where Qc and Qb denote the nominal capacity of the battery pack
sented below:
and ultracapacitor pack, respectively. The optimization target is to
locate the optimal control variable iL to minimize the cost function
presented below:
8
60
> X
N 1 X
N 1
Speed (km/h)

(A) MANHATTAN >

> J DP ¼ LðxðkÞ; uðkÞÞ ¼ Lb ðxðkÞ; uðkÞÞ þ LU ðxðkÞ; uðkÞÞ
40 >
<
k¼0 k¼0
ð8Þ
20 >
>
2
Lb ðxðkÞ; uðkÞÞ ¼ iL ðkÞU t ðkÞ þ iL ðkÞ Ri ðkÞ þ U 2D ðkÞ=RD
>
>
: 2
0
0 200 400 600 800 1000 1200 LU ðxðkÞ; uðkÞÞ ¼ ic ðkÞU ct ðkÞ þ ic ðkÞ Rc ðkÞ
Speed (km/h)

60
(B) CBDC where N denotes the length of the optimization horizon, Lb denotes
40
the battery electricity consumption and LU denotes the ultracapac-
20 itor electricity consumption. To ensure safe and reasonable opera-
0 tion for the battery packs and ultracapacitor packs, some
0 200 400 600 800 1000 1200 1400
constraints are necessary, as displayed below:
Speed (km/h)

100
(C) UDDSHDV 8
> BSoC min 6 BSoCðkÞ 6 BSoC max
>
>
50 >
> USoC min 6 USoCðkÞ 6 USoC max
>
>
>
<i
0 L;min 6 iL ðkÞ 6 iL;max
0 200 400 600 800 1000 1200 ð9Þ
>
> ic;min 6 ic ðkÞ 6 ic;max
Time (s)
>
>
>
>
>
> U t;min 6 U t ðkÞ 6 U t;max
Fig. 5. The three different driving cycles. :

MPC strategy Original strategy

2 1 1
(a1) 2 (b1)
0.8
1.5
USoC

USoC

1.5 0.5
1600 1650 1700 1410 1450 1490 1530
1 1

0.5
0 5000 10000 0 1000 2000 3000 4000

600 200
(a2) 100
0 (b2) 100
400 400 0
-100
-100
Ib (A)

-200
Ib (A)

200 1600 1650 1700 200 1410 1450 1490 1530

0
0
-200
0 5000 10000 0 1000 2000 3000 4000

1500 200 400

(a3) 1500 (b3) 200
0 0
1000 -200
-200 1000
-400
Ic (A)
Ic (A)

-400
500 1600 1650 1700 500 1410 1450 1490 1530

0 0

-500 -500
0 5000 10000 0 1000 2000 3000 4000
Time (s) Time (s)

Fig. 6. Comparisons with the CD operation mode: The ultracapacitor SoC, the battery current and the ultracapacitor current under the Manhattan driving cycle are shown in
(a1), (a2) and (a3), respectively; and those are under the UDDSHDV driving cycle are shown in (b1), (b2) and (b3), respectively.
 S. Zhang et al. / Applied Energy 185 (2017) 1654–1662 1659

where BSoCmin and BSoCmax denote the constraints for battery SoC and U(k) denote the state variables and control variables as dis-
and are set as 0.2 and 1, respectively; USoCmin and USoCmax denote played below:
the constraints for ultracapacitor SoC and are set as 0.5 and 1, 
XðkÞ ¼ ½BSoCðkÞ; USoCðkÞ; U D ðkÞ; PAPU ðkÞ
respectively; iL,min and iL,max denote the constraints for battery out- ð11Þ
put current and are set as 154 A and 154 A, respectively; and ic,min UðkÞ ¼ ½AC ðkÞ; iL ðkÞ
and ic,max denote the constraints for ultracapacitor output current where Ac(k) denotes the APU on/off command, respectively. It could
and are set as 500 A and 500 A, respectively. be set to 1 or 0 to describe the on or off status of the APU.
To evaluate the performance of the proposed integrated control
strategy, we need to define another cost function, which will
4. Verification and discussion
consider the fuel economic performance containing the electricity
consumption from the HESS and fuel consumption from the APU.
The performance of the proposed integrated control strategy
The function is displayed below:
will be verified by three different types of driving cycles, the Man-
8 hattan driving cycle, UDDSHDV driving cycle and CBDC driving
>
> X
N1 X
N1
cycle, as displayed in Fig. 5. The Manhattan driving cycle and
>
> J¼ CðXðkÞ; UðkÞÞ ¼ C B ðXðkÞ; UðkÞÞ þ C APU ðXðkÞ; UðkÞÞ
>
> UDDSHDV driving cycle are extracted from software of the ADVI-
>
>
>
>
k¼0 k¼0
SOR [35,36], and the CBDC driving cycle is presented in Ref. [30].
< þC ðXðkÞ; UðkÞÞ U
The comparison between the new control strategy and original
>
>
2
C B ðXðkÞ; UðkÞÞ ¼ ðiL ðkÞU t ðkÞ þ iL ðkÞ Ri ðkÞ þ U 2D ðkÞ=RD ÞAe
>
> control strategy is carried out with two conditions: CD operation
>
> C ðXðkÞ; UðkÞÞ ¼ ði ðkÞU ðkÞ þ i ðkÞ2 R ðkÞÞA
>
> mode and CS operation mode. As the battery parameters vary
>
>
U c ct c c e
: _ under different SoC conditions, to ensure that the SoC variation
C APU ðXðkÞ; UðkÞÞ ¼ mðkÞA f
range includes all SoC conditions, twelve Manhattan driving cycles,
ð10Þ four UDDSHDV driving cycles and six CBDC drive cycles are applied
under both CD operation mode and CS operation mode. In this
where C(X(k), U(k)) denotes the usage costs (including the battery study, the highest speed of the target PHEV is 80 km/h, so when
cost CB (X(k),U(k)), ultracapacitor cost CB(X(k),U(k)), and APU cost the vehicle speed is higher than 80 km/h, the speed is recognized
CAPU(X(k),U(k))), Ae and Af denote the electric price and fuel price as 80 km/h. The maximum output/input power of the HESS will
(Ae is set to be 0.799 RMB per kWH, and Af is set to be 8.9908 be limited to 150 kW according to the vehicle company limit.
_
RMB per kilogram), mðkÞ denotes the fuel consumption rate and is The comparison between the new control strategy and original
a function of the APU output power as displayed in Fig. 3. X(k) control strategy under the CD operation mode is presented in
Fig. 6.
Fig. 6(a2) and (a3) show the output current of the battery pack
Table 2
and ultracapacitor pack under the Manhattan drive cycle. The com-
Fuel economy comparison with the CD operation mode.
parison of the ultracapacitor current indicates that the new control
12  MANHATTAN 6  CBDC 4  UDDSHDV strategy can use the ultracapacitor more frequently than the orig-
Original control 21.206 RMB 21.509 23.735 RMB inal control strategy, especially when the required power is regen-
strategy RMB erative power. On the contrary, the battery packs in the original
New control strategy 18.118 RMB 20.901 23.815 RMB
control strategy are more frequently used. As the new control
Improvement 14.56% 2.83% 0.34%
strategy tends to use ultracapacitor packs instead of battery packs,

New strategy Original strategy

40
250 20
(a1) (a2) 0
0.5 0.26 200 -20
-40
PBat (kW)

150 -60
BSoC

0.4 0.24 100 -80

1000 1100 1200 1000 1100 1200
50
0.3 0
-50
0.2
0 5000 10000 0 5000 10000

100
250 (a3) 600 (a4) 100

200 50 0
PAPU (kW)

PUltra (kW)

400
150 -100
0
1000 1100 1200 200 1000 1100 1200
100

50 0

0
0 5000 10000 0 5000 10000
Time (s) Time (s)

Fig. 7. Comparison with the CS operation mode under Manhattan driving cycle.
1660 S. Zhang et al. / Applied Energy 185 (2017) 1654–1662

New strategy Original strategy

0.28 50
0.6 (a1) 300 (a2)
0
0.26

PBat (kW)
0.5 200 -50

BSoC
0.24
0.4 650 700 750 800 850 100 650 700 750 800 850

0.3 0

0.2
0 2000 4000 6000 0 2000 4000 6000

200 100
(a3) 400 (a4) 50
50 0
150 300
PAPU (kW)

PUltra (kW)
-50
0 200 -100
100 650 700 750 800 850 650 700 750 800 850
100
50 0
-100
0
0 2000 4000 6000 2000 4000 6000
Time (s) Time (s)

Fig. 8. Comparison with the CS operation mode under the CBDC driving cycle.

New strategy Original strategy

300 100
(a1) 0.29 (a2)
0.6 0.28 50
0.27 200 0
PBat (kW)

0.26
0.5 -50
0.25
BSoC

400 500 600 100 400 500 600

0.4

0.3 0

0.2
0 1000 2000 3000 4000 0 1000 2000 3000 4000

80 (a4) 100
200 (a3)
60 600
0
40
PAPU (kW)

PUltra (kW)

150 400 -100

20
0 400 500 600
100 400 500 600 200

50
0

0
0 1000 2000 3000 4000 0 1000 2000 3000 4000
Time (s) Time (s)

Fig. 9. Comparison with the CS operation mode under the UDDSHDV driving cycle.

the SoC of the ultracapacitor packs is usually sustained at a lower energy in the ultracapacitor is easily exhausted (as displayed in
level than that of the original control strategy, as displayed in Fig. 6 Fig. 6(b1)). Then, the battery packs have to provide all the power
(a1). The lower USoC level causes the ultracapacitor in the new requirement for the HESS, and the battery packs tend to work at
control strategy to have a greater potential to absorb the regener- high output current conditions more frequently than in the original
ative energy than that in the original control strategy. As the resis- control strategy (as displayed in Fig. 6(b2)). Although with the new
tance of the ultracapacitor packs is much lower than that of the control strategy the ultracapacitor packs can absorb the regenera-
battery packs and DC/DC converter, the frequent use of the ultraca- tive energy more effectively (as displayed in Fig. 6(b3)), the braking
pacitor is helpful for the promotion of the efficiency performance frequency under UDDSHDV driving condition is much lower than
for the PHEV. Table 2 shows that the efficiency performance of that under Manhattan condition. Moreover, the battery packs and
the system is promoted by 14.56% under the Manhattan driving DC/DC converter have a greater probability to work in a low effi-
condition. However, under UDDSHDV driving conditions, as the ciency condition, which causes the advantages and disadvantages
vehicle speed is usually high and the duration is usually long, the of the new control strategy to cancel each other out. That is why
 S. Zhang et al. / Applied Energy 185 (2017) 1654–1662 1661

Table 3 work, the robustness of the proposed control strategy will be fur-
Fuel economy comparison with the CS operation mode. ther tested by real driving cycles collected in Beijing. After that
12  MANHATTAN 6  CBDC 4  UDDSHDV the strategy will be tested through bench test.
Original control 65.152 RMB 60.402 61.081 RMB
strategy RMB Acknowledgements
New control strategy 50.896 RMB 54.808 56.905 RMB
RMB
Improvement 21.88% 9.26% 6.84% This work was partially supported by the National Natural
Science Foundation of China (Grant No. 51507012), the National
Science & Technology Pillar Program (2013BAG05B00), the Excel-
the performance of the new strategy and original strategy is almost lent Young Scholars Research Fund of the Beijing Institute of Tech-
the same (the gap is only 0.34%, as displayed in Table 2). nology and Beijing Institute of Technology Research Fund Program
In the CS operation mode, the APU can work together with the for Young Scholars. Any opinions expressed in this paper are solely
HESS according to the battery SoC and vehicle power requirement. those of the authors and do not represent those of the sponsors.
The battery SoC fluctuates around 0.3. The performance compar-
ison between the new control strategy and original control strat- References
egy under different driving cycles is shown in Figs. 7–9.
[1] Roskilly AP, Palacin R, Yan J. Novel technologies and strategies for clean
With the Manhattan driving cycle, the new control strategy can
transport systems. Appl Energy 2015;157:563–6.
make the battery SoC vary more smoothly than that with the orig- [2] Bashash S, Moura SJ, Forman JC, Fathy HK. Plug-in hybrid electric vehicle
inal control strategy (as displayed in Fig. 7(a1)), which makes the charge pattern optimization for energy cost and battery longevity. J Power
Sources 2011;196(1):541–9.
APU postpone the turn-on time for charging the battery (as dis-
[3] Cordiner S, Galeotti M, Mulone V, Nobile M, Rocco V. Trip-based SOC
played in Fig. 7(a3)). From the output power comparison displayed management for a plugin hybrid electric vehicle. Appl Energy 2016;164:
in Fig. 7(a2), we can observe that the battery output power and 891–905.
working frequency are further reduced due to the operation of [4] Chau KT, Wong YS. Overview of power management in hybrid electric vehicles.
Energy Convers Manage 2002;43:1953–68.
the APU, while the ultracapacitor works frequently. This operation [5] Trovão JP, Pereirinha PG, Jorge HM, Antunes CH. A multi-level energy
property reduces the usage cost by 21.88%, as presented in Table 3. management system for multi-source electric vehicles–an integrated rule-
It is worth noting that, under the UDDSHDV driving cycle, the fuel based meta-heuristic approach. Appl Energy 2013;105:304–18.
[6] Schouten NJ, Salman MA, Kheir NA. Energy management strategies for parallel
economy performance was promoted by 6.84%; this is because, in hybrid vehicles using fuzzy logic. Control Eng Practice 2003;11(2):171–7.
the CS operation mode, the operation of the APU can reduce the [7] Gao D, Jin Z, Lu Q. Energy management strategy based on fuzzy logic for a fuel
power requirement for HESS, which will decrease the USoC con- cell hybrid bus. J Power Sources 2008;185(1):311–7.
[8] Chen SY, Hung YH, Wu CH, Huang ST. Optimal energy management of a hybrid
sumption rate for the new control strategy. Moreover, the fact that, electric powertrain system using improved particle swarm optimization. Appl
when the BSoC is lower than 0.25, the APU will continue working Energy 2015;160:132–45.
until the BSoC reaches 0.35, can further improve the operation per- [9] Chen Z, Mi CC, Xia B, You C. Energy management of power-split plug-in hybrid
electric vehicles based on simulated annealing and Pontryagin’s minimum
formance of the ultracapacitor for the new control strategy.
principle. J Power Sources 2014;272:160–8.
From Fig. 9(a1), we can observe that, as the BSoC of the original [10] Hu X, Moura SJ, Murgovski N, Egardt B, Cao D. Integrated optimization of
control strategy reaches 0.25 earlier than the new control strategy, battery sizing, charging, and power management in plug-in hybrid electric
vehicles. http://dx.doi.org/10.1109/TCST.2015.2476799.
the working period of the APU will be prolonged, as shown in Fig. 9
[11] Chen Z, Xiong R, Cao J. Particle swarm optimization-based optimal power
(a3). In terms of the performance under the CBDC driving cycle dis- management of plug-in hybrid electric vehicles considering uncertain driving
played in Fig. 8, the BSoC variation is similar to that under the Man- conditions. Energy 2015. http://dx.doi.org/10.1016/j.energy.2015.12.071. in
hattan driving cycle, and the working property of the battery pack press.
[12] Lin CC, Peng H, Grizzle JW, Kang JM. Power management strategy for a parallel
and ultracapacitor pack is between the Manhattan cycle and hybrid electric truck. IEEE Trans Control Syst Technol 2003;11(6):839–49.
UDDSHDV cycle. From Table 3, we can observe that, under the [13] Chen B, Wu Y, Tsai H. Design and analysis of power management strategy for
CBDC driving conditions, the fuel economy performance could be range extended electric vehicle using dynamic programming. Appl Energy
2014;113:1764–74.
improved by 9.26% compared with the original control strategy. [14] Pérez LV, Bossio GR, Moitre D, Garćıa GO. Optimization of power management
in an hybrid electric vehicle using dynamic programming. Math. Comput.
Simul. 2006;73(1):244–54.
5. Conclusion [15] Onori S, Tribioli L. Adaptive pontryagin’s minimum principle supervisory
controller design for the plug-in hybrid gm Chevrolet volt. Appl Energy
This study proposed an integrated optimal power management 2015;147:224–34.
[16] Serrao L, Onori S, Rizzoni G. A comparative analysis of energy management
approach for a PHEV launched with a hybrid energy storage sys- strategies for hybrid electric vehicles. J Dyn Syst, Measur Control 2011;133
tem. In the proposed approach, the output power between the (3):031012.
APU and HESS is determined according to the rule-based power [17] Kim N, Cha S, Peng H. Optimal control of hybrid electric vehicles based on
pontryagin’s minimum principle. IEEE Trans Control Syst Technol 2011;9
management strategy, and the output power of the battery pack (5):1279–87.
is determined according to the MPC controller with a receding time [18] Zhang S, Xiong R, Zhang C. Pontryagin’s Minimum Principle-based power
horizon. In the MPC controller, the torque demand of the next ten management of a dual-motor-driven electric bus. Appl Energy
2015;159:370–80.
seconds is predicted, and the DP algorithm is applied in the MPC [19] Zhang S, Xiong R. Adaptive energy management of a plug-in hybrid electric
controller to optimize the battery output current. vehicle based on driving pattern recognition and dynamic programming. Appl
The robustness of the proposed approach was verified by three Energy 2015;155:68–78.
[20] Jeon S, Jo S, Park Y, Lee J. Multi-mode driving control of a parallel hybrid
typical driving cycles, the Manhattan cycle, the CBDC cycle and the electric vehicle using driving pattern recognition. ASME J DYN SYST-T
UDDSHDV cycle. The performance between the integrated strategy 2002;124:141–9.
and the original strategy is compared under the CD and CS opera- [21] Langari R, Won J. Intelligent energy management agent for a parallel hybrid
vehicle-part I: system architecture and design of the driving situation
tion modes. In the CD operation mode, the performance of the pro-
identification. IEEE Trans Vehicle Technol 2005;54:925–34.
posed approach shows a significant improvement of 14.56% under [22] Sun C, Hu X, Moura SJ, Sun F. Velocity predictors for predictive energy
the MANHATTAN driving cycle. In the CS operation mode, the fuel management in hybrid electric vehicles. IEEE Trans Control Syst Technol
economy improvement under the UDDSHDV driving cycle is 6.84%, 2015;23(3):1197–204.
[23] Borhan H, Vahidi A, Phillips AM, Kuang ML, Kolmanovsky IV, Cairano SD. MPC-
and the improvements under the MANHATTAN driving cycle and based energy management of a power-split hybrid electric vehicle. IEEE Trans
the CBDC driving cycle are 21.88% and 9.26%, respectively. In future Control Syst Technol 2012;20(3):593–603.
1662 S. Zhang et al. / Applied Energy 185 (2017) 1654–1662

[24] Liu J, Peng H. Modeling and control of a power-split hybrid vehicle. IEEE Trans [30] Bauman J, Kazerani M. An analytical optimization method for improved fuel
Control Syst Technol 2008;16(6):1242–51. cell–battery–ultracapacitor powertrain. IEEE T Veh Technol 2009;58:3186–97.
[25] Ngo V, Hofman T, Steinbuch M, Serrarens A. Predictive gear shift control for a [31] Cao J, Emadi A. A new battery/ultracapacitor hybrid energy storage system for
parallel hybrid electric vehicle. In: Proc. IEEE Veh Power Propuls Conf 2011. p. electric, hybrid, and plug-in hybrid electric vehicles. IEEE Trans Power Electron
1–6. 2012;27:122–32.
[26] Moura SJ, Fathy HK, Callaway DS, Stein JL. A stochastic optimal control [32] Zhang S, Xiong R, Zhou X. Comparison of the topologies for a hybrid energy-
approach for power management in plug-in hybrid electric vehicles. IEEE storage system of electric vehicles via a novel optimization method. Sci China
Trans Control Syst Technol 2011;19(3):545–55. Technol Sci 2015;58(7):1173–85.
[27] Uthaichana K, Bengea S, DeCarlo R, Pekarek S, Zefran M. Hybrid model [33] Sun F, Xiong R. A novel dual-scale cell state-of-charge estimation approach for
predictive control tracking of a sawtooth driving profile for an HEV. In: Proc series-connected battery pack used in electric vehicles. J Power Sources
Amer Control Conf 2008. p. 967–74. 2015;274:582–94.
[28] Ripaccioli G, Bemporad A, Assadian F, Dextreit C, Cairano SD, Kolmanovsky IV. [34] Sun F, Xiong R, He H. A systematic state-of-charge estimation framework for
Hybrid modeling, identification, and predictive control: An application to multi-cell battery pack in electric vehicles using bias correction technique.
hybrid electric vehicle energy management. In: Hybrid Systems: Computation Appl Energy 2016;162:1399–409.
and Control. Berlin, Germany: Springer-Verlag, 2009. p. 321–35. [35] National Renewable Energy Laboratory: Advanced Vehicle Simulator ADVISOR,
[29] Lukic SM, Wirasingha SG, Rodriguez F, Cao J, Emadi A. Power management of ver 3.2, 2001.
an ultracapacitor/battery hybrid energy storage system in an HEV. IEEE [36] Lin CC, Jeon S, Peng H, Lee JM. Driving pattern recognition for control of hybrid
Vehicle Power Propul Conf 2006:1–6. electric trucks. Vehicle Syst Dyn 2004;42(1–2):41–58.