Mechanical Systems and Signal Processing 87 (2017) 17–29

Contents lists available at ScienceDirect

Mechanical Systems and Signal Processing

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

Fuel consumption optimization for smart hybrid electric

vehicle during a car-following process
Liang Li, Xiangyu Wang, Jian Song n
State Key Laboratory of Automotive Safety and Energy, Tsinghua University, Beijing 100084, China

a r t i c l e i n f o abstract

Article history: Hybrid electric vehicles (HEVs) provide large potential to save energy and reduce emis-
Received 19 December 2015 sion, and smart vehicles bring out great convenience and safety for drivers. By combining
Received in revised form these two technologies, vehicles may achieve excellent performances in terms of dynamic,
29 February 2016
economy, environmental friendliness, safety, and comfort. Hence, a smart hybrid electric
Accepted 1 March 2016
vehicle (s-HEV) is selected as a platform in this paper to study a car-following process
Available online 14 March 2016
with optimizing the fuel consumption. The whole process is a multi-objective optimal
Keywords: problem, whose optimal solution is not just adding an energy management strategy (EMS)
Smart hybrid electric vehicle to an adaptive cruise control (ACC), but a deep fusion of these two methods. The problem
Multi-objective optimal problem
has more restricted conditions, optimal objectives, and system states, which may result in
Fuel consumption optimization
larger computing burden. Therefore, a novel fuel consumption optimization algorithm
Car-following
Model predictive control based on model predictive control (MPC) is proposed and some search skills are adopted
in receding horizon optimization to reduce computing burden. Simulations are carried out
and the results indicate that the fuel consumption of proposed method is lower than that
of the ACC þ EMS method on the condition of ensuring car-following performances.
& 2016 Elsevier Ltd. All rights reserved.

1. Introduction

Environment pollution and petroleum problems have become more and more serious, which pushes vehicle technologies
into the aspects of energy conservation and environment protection [1]. Wherein, hybrid electric vehicles (HEVs) provide large
potential to save fuel consumption and reduce pollutant emission [2,3]. An HEV is a vehicle driven by more than one power
source, and usually refers to a combination of internal combustion engine (ICE) and electric motor (EM). It has various driving
mode under different conditions, such as ICE driving alone, EM driving alone, ICE and EM driving together, EM regenerative
braking and so on, which makes it easily to adjust IEC operation points in a high-efficiency area [4–6]. On another side, intel-
ligence and net-connection of vehicles are a trend, which can bring out great convenience and safety for drivers [7–10]. Advanced
driver assistant system (ADAS) is being applied in passenger vehicles gradually, which contains adaptive cruise control (ACC),
automatic parking, lane-change assistance, etc. [11]. By combining these two vehicular technologies, vehicles may achieve
excellent performances in terms of dynamic, economy, environmental friendliness, safety, and comfort. Therefore, in this paper, a
smart hybrid electric vehicle (s-HEV) is selected as a platform to study a car-following process with optimizing fuel consumption.
Many works have given a deep insight in the energy management strategy (EMS) for HEVs and the car-following control
which is a kind of ACC for intelligent/smart vehicles. Hu et al. studied energy efficiency of a series plug-in hybrid electric bus

n
Corresponding author.
E-mail address: daesj@mail.tsinghua.edu.cn (J. Song).

http://dx.doi.org/10.1016/j.ymssp.2016.03.002
0888-3270/& 2016 Elsevier Ltd. All rights reserved.
18 L. Li et al. / Mechanical Systems and Signal Processing 87 (2017) 17–29

(PHEB) with different EMSs and battery sizes based on the tank-to-wheel (TTW) analysis [12]. Li et al. proposed a correc-
tional DP-based EMS of PHEB, which improved the economic performance in a city-bus-route [13]. A MPC-based EMS is put
forward to solve the fuel consumption optimization which is formulated as a nonlinear constrained optimal control problem
in [14]. Besides, an adaptive energy management was proposed for a PHEV based on driving pattern recognition and
dynamic programming [15]. Among all the approaches, MPC-based method is one of most popular kind of EMSs [14,16].
Meanwhile, MPC is also the most common method to realize ACC. For example, Sarkouri et al. used a nonlinear MPC to
realize an automated switching to cruise control [17], and Li et al. proposed a fast online computation of MPC and applied it
to a fuel economy-oriented ACC [18]. Besides, supervised adaptive dynamic programming was also used for a full-range ACC
problem [19], and a method of kernel-based least squares policy iteration was proposed for a self-learning cruise control
[20]. In addition, the controls of vehicle dynamics, such as anti-brake system (ABS), traction control system (TCS), electronic
stability control (ESC), and tire force analysis, are also supposed to be considered in ACC [21–23].
However, most of studies are focused on EMS or ACC alone. As for the car-following of an s-HEV, it is a multi-objective optimal
problem, which has more restricted conditions, optimal objectives, and system states. Kural et al. integrated an ACC into EMS for
HEV to estimate the look-ahead battery energy and if necessary apply pre-discharge strategies to fully benefit the recuperation
energy during deceleration manoeuvres [24]. Luo et al. proposed a multi-objective decoupling hierarchical strategy [7] and a
coordinated control of tracking ability, fuel economy, and ride comfort [9]. The first method has a clear structure and each control
hierarchy uses a practical algorithm so that it can be applied in a real electric control unit. The second method uses a multistep
offline dynamic programming optimization and an online lookup table to realize the real-time control, which obtains better
performance than the first one. In fact, the solution of the problem is not supposed to simply add an EMS to an ACC in series (an
ACCþEMS method). An ACCþEMS method can solve the problem but is not optimal, because an ACC just considers dynamic car-
following performances and then an EMS can only decide the torque distribution of ICE and EM according to the desired vehicle
acceleration. It is worth noting that two different sequences of desired acceleration may achieve similar ACC performances but
have great differences in EMS results. Therefore, a deep fusion method with ACC and EMS is needed to solve the problem. The
contributions and novelties of this paper are as follows: 1) a car-following problem with optimizing fuel consumption for s-HEV
is formulated; 2) a novel fuel consumption optimization algorithm based on nonlinear MPC is proposed; 3) some search skills are
used in preceding horizon optimization to reduce the computing burden.
The rest paper is organized as follows. System dynamic models of an s-HEV are built in Section 2. Then, Section 3
introduces and formulates the problem. Next, an optimal fuel consumption control method is designed in Section 4. In
Section 5, simulations are carried out and results are analyzed. Finally, Section 6 concludes the paper.

2. System dynamic models

In this paper, a single-shaft parallel s-HEV equipped with an AMT is selected as the study platform, whose powertrain
structure is shown in Fig. 1. The powertrain system mainly contains a gasoline engine, an automated dry clutch, an electric
motor (EM), a power battery, a five-speed AMT, and a braking system [25]. The engine and EM can provide traction
independently for the vehicle, and the EM can also provide braking force coordinated with the braking system. The clutch
can connect or disconnect the engine with the powertrain to ensure that the engine can operate in high-efficiency area, but
the dynamics of clutch engagement and disengagement are not considered in this paper. The AMT can adjust operation
points for the engine and EM via changing transmission ratio [26].
The key parameters of the HEV are shown in Table 1, which are obtained from ADVISOR, and following text will
introduce and build simplified models of each part. Simplified models may neglect some dynamic characteristics of system
parts, but in this paper we emphasize the dynamics of the whole system far more than that of each part. Therefore, it is
proper and convenient using the simplified models to reflect system dynamics because the detailed dynamics of each part
have little effect on that of the whole system.

Clutch
Actuator I Actuator II
EM

Brake
Engine System

Inverter AMT
Battery pack

Fig. 1. Schematic graph of the single-shaft parallel HEV powertrain structure.

 L. Li et al. / Mechanical Systems and Signal Processing 87 (2017) 17–29 19

Table 1
Parameters of the single-shaft parallel HEV.

Components Description

Engine Type: four cylinder gasoline engine; Displacement: 1.9 L;

Maximum power: 63 kW @ 5500 rpm;
Peak torque: 145 N m @ 2000 rpm.
EM Type: permanent magnet synchronous motor (PMSM).
Maximum power: 49 kW; Maximum speed: 8000 rpm;
Peak torque: 274.4 N m as motor and  170 N m as generator.
Maximum current: 400 A.
Gear box Type: five-speed AMT; Average efficiency: 0.9;
Gear ratio: [3.79, 2.17, 1.41, 1, 0.86]; Final drive ratio: 4.55.
Battery Type: NiMH Spiral Wound;
Nominal cell voltage: 1.2 V; Nominal voltage: 288 V;
Total cells: 480 (12 cells  40 modules); Capacity: 13 A h.
Vehicle Mass: 1800/2300 kg; Frontal area: 2.66 m2;
Air resistance coefficient: 0.44;
Road resistance coefficient: 0.0125;
Tire radius: 0.343 m; Tire inertia: 4.829 kg m2 of each.

150
2
230 24 30
600 0
220
Fuel consumption rate (g/kWh)

240
120

240
260

260
500 220

230
Torque (Nm)

90 260

28
23

0
280
400 0 240
240

260

30
0
0
60
3200

30
260 35
80

0
300 280
300 400
30 350 350 450
400
200 0 400
450 450
5000 4000 50
3000 2000 100
1000 N m) 0
Speed (rp 0 150 ue( 0 1000 2000 3000 4000 5000
m) Torq
Speed (rpm)
3D surface MAP 2D contour MAP
Fig. 2. MAP chart for engine fuel consumption rate.

2.1. Engine model

As for an engine model, what we concerns in our study is the characteristics of fuel consumption but not the dynamics.
Therefore, the dynamic response of engine torque is simplified as a first-order process.

T Eng
T Eng ¼ ð1Þ
τEng s þ1
where T Eng , T Eng , and τEng are the output torque, the desired torque, and the time-delay constant of engine, respectively.
The fuel consumption rate of engine can be regarded as a function of engine torque and engine speed, if the engine
dynamics are neglected. Then, the fuel consumption rate can be expressed as follows.

be ¼ f Eng ðT Eng ; ωEng Þ ð2Þ

where ωEng is the output speed of engine, and be is the fuel consumption rate of engine whose unit is g=ðkWhÞ. The values of
be can be obtained by a lookup table based on fuel consumption MAP of engine as shown in Fig. 2. It is worth noting that the
engine cannot operate at a speed lower than the idle speed (about 701 rpm) and the fuel consumption rate is very high
when then engine outputs a small torque, so a minimum speed ωmin Eng and a series of minimum torques T Eng ðωEng Þ are defined
min

to ensure the engine operates at a high-efficiency area. Therefore, in Fig. 2(b), the shaded area refers to the area where the
engine is not allowed to operate. When the engine is going to come into the shaded area under a certain condition, there are
two ways to avoid it: one is that the clutch disengages and the engine operates at an idle state, and another is that the EM
operates as a generator to increase the engine output torque.
20 L. Li et al. / Mechanical Systems and Signal Processing 87 (2017) 17–29

280

95
250

93
95
80
100 65
200

85

91
90
EM Efficiency (%)

150 75

88
Torque (Nm)
80 100
80 95 93

93

95
50 93 91
70 85 91 91 88
85
80 88 88 85
75 85 80
0
60 65 85 88 85
88 80
85
91 88
-50 93 91
55 65
8000

93
-100
6000

91
-200 8

80 5
Spe 4000
-100 85 8

7
0

95
-150

93

91
ed ( 2000 100 55
rpm 200 ) -180
) 0 300 e (Nm
Torqu 0 2000 4000 6000 8000
Speed (rpm)
3D surface MAP 2D contour MAP
Fig. 3. MAP chart for EM efficiency.

0.45 340

Open-circuit Voltage
0.4 330
Charging Resistance
Discharging Resistance
0.35 320
Resistance (Ω )

Voltage (V)
0.3 310

0.25 300

0.2 290

0.15 280

0 20 40 50 60 75 80 100
SoC (%)
Fig. 4. Relationships between some parameters and SoC.

Then, the fuel consumption per second FC (L/s) can be calculated as follows.
T Eng U ωEng Ube
FC ¼ ð3Þ
3:6  109 U ρgas

where ρgas is the density of gasoline, whose value is 0:725 L=kg at the temperature of 20 °C.

2.2. EM model

Similar to the engine, the dynamics of the EM is not what we concerns and also can be simplified as a first-order process [27].
T EM
T EM ¼ ð4Þ
τEM s þ1
where T EM , T EM and τEM are the output torque, the desired torque, and the time-delay constant of EM, respectively.
The efficiency of EM can be obtained by a lookup table based on EM efficiency MAP as shown in Fig. 3 and expressed as
follows.
ηEM ¼ f EM ðT EM ; ωEM Þ ð5Þ

where ηEM is the efficiency of EM and ωEM is the output speed of EM.
Then, the power that the battery should provide to EM P bat can be calculated as follows.
P bat ¼ T EM U ωEM U ηαEM ð6Þ

where α is 1 when the EM operates as a generator and  1 when as a motor.

 L. Li et al. / Mechanical Systems and Signal Processing 87 (2017) 17–29 21

2.3. Battery model

The characteristics of battery parameters, such as open-circuit voltage, state of charge (SoC), state of health (SoH), are
complicated [28] and not the main concerns in our study. Therefore, the effects of temperature on battery are not con-
sidered, and then the relationships between open-circuit voltage, charging resistance and discharging resistance and SoC are
shown in Fig. 4. Considering the health of battery, the SoC is supposed to be restricted in a range from 0.5 to 0.75. It can be
found from Fig. 4 that the parameters do not change very much when the SoC varies from 0.5 to 0.75, so it is reasonable to
regard these parameters as constants.
For convenience, a simplified battery model with appropriate precision needs to be built, which simplify the battery as a
voltage source with an internal resistance. Then, the gradient of SoC with time can be written as follows [16].
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
dSoC V bat  V 2bat 4Rbat P bat
¼ ð7Þ
dt 2Rbat Q bat
where V bat , Rbat , and Q bat are the open-circuit voltage, the internal resistance, and the capacity of battery. It can be seen from
(7) that the maximum of battery is restricted by P bat rV 2bat =ð4Rbat Þ.

2.4. Vehicle model

The vehicle model in this paper refers to a longitudinal dynamic model of a vehicle, which is described as follows.
I veh ω
_ w ¼ T veh  T l ð8Þ
where I veh , T veh , and T l are the equivalent inertia, the driving torque on wheels and the resistant torque of vehicle, and ωw is
the rotational speed of wheels.
The driving torque on wheels is provided by engine, EM and braking system, and the resistant torque contains a friction
resistance, a gradient resistance, and an air resistance. Hence, they are expressed as follows [25].

I veh ¼ mveh r 2w þ4I w þ i2g i20 ðI Eng þ I EM Þ ð9Þ

T veh ¼ ηT ig i0 ðT Eng þ T EM Þ þ T Brk ð10Þ

 
1
T l ¼ mveh gð sin θ þ f r cos θÞ þ C D Av2veh U r w ð11Þ
2
where mveh is the mass of vehicle, r w is the radius of a wheel, I w is the inertia of a wheel, ig is the ratio of AMT, i0 is the ratio
of final drive, I Eng is the inertia of engine, I EM is the inertia of EM, ηT is the transmission efficiency, T Brk is the braking torque
provided by braking system which is a negative value, g is the gravitational acceleration, θ is the gradient of road, f r is the
fraction coefficient of road, C D is the air resistance coefficient, A is the frontal area of vehicle, and vveh is the longitudinal
speed of vehicle. Besides, there are some relationships between the speeds: ig i0 ωw ¼ ωEng ¼ ωEM and ωw r w ¼ vveh .

3. Problem formulation

As mentioned in Section 1, the fuel consumption optimization during a car-following process is a multi-objective
optimization problem based on system models. Hence, in this section, the problem is formulated as follows.

3.1. System states space equation

The system states are not only the vehicle states but also the battery states, so state vector is defined as
 T
x ¼ ½sveh ; vveh ; SoC T , and control vector is defined as u ¼ T Eng ; T EM ; T Brk ; ig , where sveh is the travel distance and the
acceleration of current vehicle. Then, considering that the road gradient and the states of preceding vehicle can be obtained
 T
for a smart vehicle by a radar, these external inputs are defined as d ¼ θ; sref sveh ; vref , where sref , vref , and aref are the
travel distance, the longitudinal speed and the acceleration of preceding vehicle. Therefore, the system states space equation
can be written as follows.
x_ ¼ f sys ðx; u; dÞ ð12Þ

According to the models built in Section 2, there are some constraints for control variables, which are:

Eng ðωEng Þ r T Eng r T Eng ðωEng ÞT EM ðωEM Þ r T EM rT EM ðωEM Þ ig A ½3:79; 2:17; 1:41; 1; 0:86
T min max min max
ð13Þ

where T min
Eng , T max
Eng , T min
and EM , T max
can be calculated by the MAPs of engine and EM. For convenience, we assume that the
EM
automatic shifting logic is dependent of EMS and only relevant to the vehicle longitudinal speed, which means ig ¼ f T ðvveh Þ.
 T
Then, the control vector can be simplified as u ¼ T Eng ; T EM ; T Brk .
22 L. Li et al. / Mechanical Systems and Signal Processing 87 (2017) 17–29

3.2. Dynamic optimizing indexes

As for a car-following process, the dynamic optimizing indexes usually include following distance Δs ¼ sref sveh , relevant
speed Δv ¼ vref  vveh , relevant acceleration Δa ¼ aref  aveh and vehicle jerk jveh ¼ daveh =dt, where aref and aveh are the
accelerations of preceding vehicle and current vehicle obtained by differential of longitudinal speed. The evaluations for
each dynamic index are analyzed as follows.
Considering the driving safety, minimum following distance is supposed to be larger than minimum braking distance.
The braking distance can be estimated as

v2veh
sBrk ¼ vveh τBrk þ ð14Þ
2aBrk

where sBrk is the braking distance, τBrk is the response time of braking system, and aBrk is the braking deceleration. Here, we
assume that τBrk is 0.5 s and maximum braking deceleration is 8 m/s2. Hence, minimum following distance Δsmin should be
set as follows.

Δsmin ¼ 2 þ 0:5 Uvveh þ0:0625 Uv2veh ð15Þ

As for maximum following distance, it depends on drivers' behaviors. Usually, it is also relevant to vehicle speed, so here
we set the maximum following distance Δsmax as follows.

Δsmax ¼ 10 þ vveh þ 0:0825 U v2veh ð16Þ

Then, according to the minimum and maximum following distance, optimal following distance area is defined as follows.

Δssup
opt ¼ α U Δsmax þ ð1  αÞ U Δsmin

Δsinf
opt ¼ α U Δsmin þ ð1  αÞ U Δsmax ð17Þ

where Δssup opt and Δsopt are the upper bound and the lower bound of optimal following distance area, and α is the adjustment
inf

coefficient. If α ¼ 0:5, then Δssup opt ¼ Δsopt , which means the optimal following distance area becomes an optimal line.
inf

As shown in Fig. 5, the following distance is supposed to be controlled in the optimal area. When Δs is out of the optimal
area, there should be a cost item, and when it is larger than maximum distance, the cost item should be very large. Con-
sidering the driving safety, Δs is not allowed to be smaller than minimum distance. Therefore, the cost function of Δs is
defined as follows.
8
>
>
>
þ 1; Δs o Δsmin
>  
>
> Δ s  Δsinf
> U π2 ; Δsmin r Δs o Δsinf
> f 1 U tan Δsmin  Δsinf
> opt

>
>
opt
>
opt
<
sup
Δ sinf þ Δ s
J Δs ¼ Δs  opt 2 opt ; Δsinf
opt r Δs r Δsopt
sup ð18Þ
>
>
>
>
>
>
>
> f 2 UðΔs  Δssup opt Þ ;
2
Δssup
opt o Δs r Δsmax
>
>
>
>
: f 3 UðΔs  Δsmax Þ þf 2 UðΔs  Δsopt Þ ; Δs 4 Δsmax
2 sup 2

where f 1 , f 2 , and f 3 are the cost coefficients, ensuring f 3 4 4 f 2 .

As for relevant longitudinal speed and acceleration, the evaluations also depend on drivers' behaviors, and here we use
their quadratic form to define the cost function.

J Δv ¼ g 1 U Δv2 J Δa ¼ g 2 U Δa2 ð19Þ

where g 1 and g 2 are the cost coefficients.

Following Distance (m)

160
Minimum Distance
120 Maximum Distance
Optimal Distance Area
80

40

0
0 20 40 60 80 100 120 140
Longitudinal Speed (km/h)
Fig. 5. Following distance with the longitudinal speed of vehicle.
 L. Li et al. / Mechanical Systems and Signal Processing 87 (2017) 17–29 23

At last, the vehicle jerk concerns the driving comfort, so it should be restricted by cost function.
8
< h1 Uj2 ; jveh 4 2 m=s3
veh
J jerk ¼ ð20Þ
: 0; j r 2 m=s3
veh

where h1 is the cost coefficient. During driving process, slight jerk is inevitable, so there is no cost item when jjveh j is less
than 2 m/s3. Besides, if the vehicle is not at a braking state, the maximum jerk is supposed to be less than 6 m/s3.
From the above, the cost function of dynamic optimizing indexes can be defined as follows.
Z tf
J Dyn ¼ ðJ Δs þJ Δv þJ Δa þ J jerk Þdt ð21Þ
0

where t f is the period of MPC predictive horizon that is described in Section 4.2. The cost functions in Eqs. (18)–(20) indicate the
Rt Rt Rt Rt
dynamic performances at a moment of car-following process, so the integrated indexes 0f J Δs dt, 0f J Δv dt, 0f J Δa dt, and 0f J jerk dt
refer to the totals of following distance, relevant speed, relevant acceleration, and jerk in the whole MPC predictive horizon,
respectively. Hence, J Dyn can reflect the car-following dynamic performances in MPC predictive horizon as a total cost function.

3.3. Economic optimizing indexes

In order to optimize fuel consumption, economic optimizing indexes should be defined for gasoline consumption of
engine and power consumption of EM. Generally, equivalent consumption minimum strategy (ECMS) is used in EMS of HEVs
to balance the consumption of engine and EM, which means power consumption is equivalent to fuel consumption by
corresponding rules. Considering that the electric power stored in battery is transformed by engine output power, the
equivalent fuel consumption of EM can be calculated as follows.
When engine outputs power and EM acts as a generator, we have

3:6  109 U ρgas UFC

P bat ¼ T EM U ωEM U ηEM ¼ T Eng2EM U ωEng U ηEM ¼ U ηEM ð22Þ
be

be U P bat
) FC EM ¼ ð23Þ
3:6  109 U ρgas U ηEM

where FC EM is the equivalent fuel consumption of EM, be is the average fuel consumption rate, and ηEM is the average
generation efficiency.
Considering the health of battery, the SoC of battery should be restricted in a range from 0.5 to 0.75, so an electric power
coefficient pe is defined as
8
< 0:001;
> SoC Z0:75
pe ðSoCÞ ¼ 6 U SoC þ 5; 0:5 o SoC o0:75 ð24Þ
>
: 1000; SoC r0:5

In order to avoid negative values of cost function, an additional item  pe FC min

EM is added. Therefore, the cost function of
economic optimizing indexes can be defined as follows.
Z tf
J Eco ¼ ðFC þpe FC EM  pe FC min
EM Þdt ð25Þ
0

where FC min
EM is the minimum equivalent fuel consumption of EM calculated by P min
bat .

3.4. Comprehensive optimizing index

Finally, considering both dynamic indexes and economic indexes, a cost function of comprehensive optimizing index is
defined as

J Opt ¼ J Dym þ λ U J Eco ð26Þ

where λ is a coefficient to adjust the proportion of dynamic indexes and economic indexes in cost function. The value of λ
should be determined with considering the values of J Dym and J Eco . The magnitudes of J Dym and J Eco are 102 and 10  1 in most
conditions, respectively, so the magnitude of λ should be chosen as 103 to balance the proportion of dynamic indexes and
economic indexes, which sets J Dym and J Eco at a same magnitude.
Therefore, the problem in this paper is to optimize the cost function (26) to satisfy the system (12) and the restriction (13).
24 L. Li et al. / Mechanical Systems and Signal Processing 87 (2017) 17–29

Fig. 6. Schematic of optimal fuel consumption control method based on MPC.

4. Fuel consumption optimization method

In this section, a fuel consumption optimization method based on a nonlinear MPC is proposed to solve the problem, as shown
in Fig. 6. The optimization algorithm needs enumerate all eligible control vectors to search the optimal solution because the system
is nonlinear, which may result in large computing burden. Hence, some search skills are used in the optimization algorithm.

4.1. Discretization of predictive model

According to Eqs. (7)–(12), the discretization of system predictive model can be written as follows.
1
sveh ðk þ1Þ ¼ sveh ðkÞ þ vveh ðkÞ U T s þ aveh ðkÞ U T 2s vveh ðk þ 1Þ ¼ vveh ðkÞ þ aveh ðkÞ UT s
2
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
V bat  V 2bat  4Rbat P bat ðkÞ
SoCðk þ 1Þ ¼ SoCðkÞ  UTs ð27Þ
2Rbat Q bat

where T s is the sample time and set as 0.1 s in this paper, and the expressions of aveh ðkÞ and P bat ðkÞ are
1  
aveh ðkÞ ¼ η U ig ðkÞ Ui0 ðT Eng ðkÞ þ T EM ðkÞÞ þT Brk ðkÞ T l ðvveh ðkÞ; θðkÞÞ Urw
I veh T
P bat ðkÞ ¼ ηEM ðkÞ U T EM ðkÞ U ωEM ðkÞ ð28Þ

The discretization of predictive model can also be written in a simplified form as (12):
xðk þ 1Þ ¼ f dsys ðxðkÞ; uðkÞ; dðkÞÞ ð29Þ

4.2. Optimizing Algorithm for MPC

It can be seen that the predictive model (29) is nonlinear due to the nonlinearities of T l and SoC. Therefore, the predictive
outputs xðk þ iÞ; i ¼ 1; 2; ⋯; P will be a really complex function of xðkÞ, uðkÞ, and dðkÞ, so that we cannot calculate an analytical
solution for optimal control law easily in receding horizon optimization. That is to say, all eligible control vectors are supposed
to be enumerated to search the optimal solution. However, it may result in large computing burden, so some simplifications
for MPC and some search skills for the preceding horizon optimization are used to improve the computing efficiency.
Firstly, predictive horizon P and control horizon m of MPC should be determined. Considering that the sample time is set
as 0.1 s, it is proper to set P less than 10 to ensure the whole predictive horizon is less than 1 s. Then, m is set as 1, which
means the control vector is assumed to be a constant during predictive horizon. Besides, we also assume that the accel-
eration of preceding vehicle is a constant during predictive horizon.
Secondly, the acceleration of current vehicle aveh and the torque of EM T EM are selected as enumerated objectives. aveh is
restricted by the maximum jerk of current vehicle jmax 3
veh which is set as 6 m/s , and the eligible maximum and minimum
max
accelerations (amax and amin ) for ðk þ 1ÞT s are aveh ðkÞ 7 jveh UT s , respectively. Meanwhile, T EM is also restricted by T min
EM and
T max
EM . In fact, we can calculate T veh by using aveh and T Eng by using T veh and T EM . As for T Brk , it is set as zero unless the EM
cannot provide enough braking torque, and if so T EM is set as T min EM and the rest braking torque is complemented by the
braking system. Hence, the enumeration range can be decreased.
 L. Li et al. / Mechanical Systems and Signal Processing 87 (2017) 17–29 25

Thirdly, two loops are used to enumerate aveh and T EM . The loop of aveh is set as the outer one and that of T EM is set as the
inner one. In the outer loop, the dynamic states of current vehicle can be calculated and then the cost of dynamic indexes is
obtained. It is obvious that if the cost of only dynamic indexes is more than current optimal cost, the result must be worse
and there is no need to enumerate T EM and calculate the cost of economic indexes. This is an important search pruning,
which can reduce the computing burden effectively. Meanwhile, if the speed of engine is lower than minimum value, the
vehicle should be driven by EM alone and there is no need to enumerate T EM , too. In addition, during enumeration, if one of
T Eng and T EM is out of restricted range, the loop should continue.
Finally, based on the skills of search pruning in last paragraph, it can be inferred that the earlier the optimal solution
appears, the better the effect of search pruning is. Hence, a stochastic enumeration strategy is used to get an eligible aveh .
To sum up, the steps of algorithm for receding horizon optimization of MPC are listed as follows:
Algorithm for receding horizon optimization of MPC
n o
Initialize J min ; amin ; amax ; skveh ; vkveh ; Δsk ; vkref ; ikg and etc: ;
if Δsk o Δsmin then
emergency braking;
end if
while aveh A ½amin ; amax  do
calculate T veh ;
if T veh 4 T max max
Eng þ T EM then continue;

fskveh ðiÞ; vkveh ðiÞ; jkveh ðiÞg’f ðskveh ; vkveh Þ; i ¼ 1; 2; ⋯; P;

n o
calculate J kΔs ðiÞ; J kΔv ðiÞ; J kΔa ðiÞ; J kjerk ðiÞ ; i ¼ 1; 2; ⋯; P;
PP n o
J Dys ¼ J kΔs ðiÞ þ J kΔv ðiÞþ J kΔa ðiÞþ J kjerk ðiÞ ;
i¼1
if J Dys 4 J min then continue;
if ωEng o ωmin
Eng then

fT Eng ¼ 0; T EM ¼ T veh ; T Brk ¼ 0g or fT Eng ¼ 0; T EM ¼ T min min

EM ; T Brk ¼ T veh  T EM g;
calculate Q kEM ðiÞ;
P
P
J Eco ¼ pe Q kEM ðiÞ;
i¼1
if J Dys þ λ U J Eco o J min then
n o n o
þ1 þ1 þ1 þ1 kþ1
update J min ; T kEng ; T kEM ; T kBrk and skveh ; vveh ;
end if
else
for T EM ¼ T min max
EM : 1: T EM do
fT Eng ¼ T veh  T EM ; T Brk ¼ 0g or fT Eng ¼ 0; T Brk ¼ T veh  T EM g;
if T Eng not in ½T min max
Eng ; T Eng  then continue;

calculate fFC k ðiÞ; FC kEM ðiÞg;

P
P
J Eco ¼ fFC k ðiÞ þ pe FC kEM ðiÞg;
i¼1
if J Dys þ λ U J Eco o J min then
n o n o
þ1 þ1 þ1 þ1 kþ1
update J min ; T kEng ; T kEM ; T kBrk and skveh ; vveh ;
end if
end for
end else
end while

5. Simulations and results

In this section, an s-HEV model is built in MATLAB/SIMULINK based on the data in ADVISOR. The results of proposed
method are analyzed and also compared with the results of ACC þEMS method.

Table 2
Simulation parameters.

Parameters Symbols Values Parameters Symbols Values

Predictive horizon P 8 Distance cost efficient I f1 0.2

Control horizon m 1 Distance cost efficient II f2 2
Average fuel consumption be ðg=ðkW hÞÞ 240 Distance cost efficient III f3 100
Average generation efficiency ηEM (%) 88 Speed cost efficient g1 0.01
Cost function efficient λ 1000 Acceleration cost efficient g2 0.01
Optimal distance efficient α 0.4 Jerk cost efficient h1 0.1
26 L. Li et al. / Mechanical Systems and Signal Processing 87 (2017) 17–29

Longitudinal Speed (km/h)

130
120
Preceding vehicle
90 Current vehicle

60

30

0
0 200 400 600 800 1000 1200 1400 1600
Time (s)

Fig. 7. Results for the longitudinal speed of preceding vehicle and current vehicle.
Travel Distance (km)

35
30
Preceding vehicle
20 Current vehicle

10

0
0 200 400 600 800 1000 1200 1400 1600
Time (s)

Fig. 8. Results for the travel distance of preceding vehicle and current vehicle.

5.1. Simulation parameters

The parameters of HEV are shown in Section 2 Table 1. The mass of vehicle is set as 2000 kg with half-load. Then, cycle
ABR02 is selected as the driving cycle of preceding vehicle, which lasts 1639 s and where the average speed is about 70 km/h
and the maximum speed is about 129.2 km/h. In addition, the road gradient is assumed as zero during the cycle. The
parameters of proposed method are listed in Table 2.

5.2. Results analysis

The longitudinal speed of preceding vehicle and current vehicle is shown in Fig. 7. The preceding vehicle follows the
ABR02 driving cycle. It can be seen that the speed of current vehicle tracks that of preceding vehicle well, although there are
many conditions of rapid acceleration and deceleration.
As shown in Fig. 8, during the whole process, the vehicles travel for approximate 31.9 km. In Fig. 9, because the following
distance is relevant to the longitudinal speed, it changes with time and its curve shape is similar to that of speed. Besides,
the actual following distance is between the maximum and minimum following distance, which means the car-following
result is satisfying. It is worth noting that, combined with Fig. 7 and from the zoom area, we can see that the following
distance varies from neighborhood of maximum to neighborhood of minimum during deceleration process, and varies
oppositely during acceleration process. That is because in this way the acceleration or deceleration of current vehicle will be
more assuasive than preceding vehicle, which may reduce not only the jerk of vehicle but also fuel consumption.
Fig. 10 shows the jerk of preceding vehicle and current vehicle. Obviously, the jerk of current vehicle is smaller than that
of preceding vehicle. The jerk of current vehicle is restricted in a range of 7 5 m=s3 during most time. In fact, if we see the
gear position variations in Fig. 11, it can be seen that the large jerks occur at the shifting points, because the transmission
ratio changes after a shifting process, which results in the driving torque changing acutely. Therefore, although the detailed
process of shifting is not considered here, the jerks caused by shifting cannot be avoided.
As mentioned in Section 3.1, the shifting strategy is defined as a simple one-parameter shifting method, which means the
gear position is only depended by longitudinal speed. For convenience, the neutral gear position is not considered, so the
gear position during the whole process is shown in Fig. 11 without the neutral gear position.
Fig. 12 shows the operation points of engine and EM during the while process. As mentioned before, one of the advantage
of a HEV is that the operation points of engine can be adjusted in the high-efficiency area. Hence, as shown in Fig. 12, a
majority of engine operation points concentrate in the area where be is less than 240 g=ðkW hÞ, while the EM operation
points spread dispersedly. Besides, the points where engine does not output torque are not plotted in the figure, so it seems
that the points in left is less than that in right.
Finally, Figs. 13 and 14 show the variations of SoC and fuel consumption with time. The SoC is set as 65.00% initially, and
the value becomes 66.24% at last. Besides, during the whole process the SoC is in a range from 60% to 70%, which is beneficial
to battery health. The fuel consumption is 2.975 L gasoline at last. Assuming that the average fuel consumption rate is
240 g=ðkW hÞ, we can convert the incremental energy of battery (SoC varies from 65.00% to 66.24%) to the saving gasoline
(0.0165 L). Therefore, the equivalent fuel consumption is 2.959 L gasoline. Then, the fuel consumption per hundred kilo-
meters is approximate 9.276 L/(100 km). It is worth noting that in this simulation the engine is at idle state, and if the idle-
stop technology is used the fuel consumption will be reduced more.
 L. Li et al. / Mechanical Systems and Signal Processing 87 (2017) 17–29 27

150

Following Distance (m)

120 Maximum following distance
Actual following distance
90 Minimum following distance
75
60
50
30
zoom 25
0
0 200 400 600 800 1000 1200 01400 1600
1235 1255
Time (s)

Fig. 9. Results for the following distance between preceding vehicle and current vehicle.

25
15
Jerk (m3/s)

5
0
-5

-15 Preceding vehicle

Current vehicle
-25
0 200 400 600 800 1000 1200 1400 1600
Time (s)

Fig. 10. Results for the jerk of preceding vehicle and current vehicle.

5
Gear Position

4
3
2
1
0 200 400 600 800 1000 1200 1400 1600
Time (s)

Fig. 11. Results for the gear position of current vehicle.

150 280
250 93

95
80

2 65
230 24 3 0 200
85

0
91

22 0
24 0

24 0

120
26 0 150 75
88
26 0

Torque (Nm)
Torque (Nm)

100
23 0

22 0
80
93

90
26 0

95
28

50
23

85 91 9
0

91
28 0
0

24 0 85
80 88 88
75 85
24 0 0
26 0

30

65 858
0

85 88 91
0

60
320 0

30

260 35 -50
80

0 65
28 0
93

300 40 0 -100
91

35 0 8
80

45 0
85 8
75

30 350
95

40 0 -150
400 55
20 450 450 -180
800 2000 3000 4000 5000 0 1000 2000 3000 4000 5000
Speed (rpm) Speed (rpm)
Fig. 12. Operation points of engine (Right) and EM (Left).

75
70
SoC (%)

65
60
55
50
0 200 400 600 800 1000 1200 1400 1600
Time (s)

Fig. 13. Results for the SoC of current vehicle battery.

5.3. Comparison with ACCþ EMS

The ACC þEMS method is also carried out in simulation, which refers to that a MPC-based ACC is applied to obtain the
desired acceleration of current vehicle for next step, and then a MPC-based EMS is used to distribute the requested torque of
28 L. Li et al. / Mechanical Systems and Signal Processing 87 (2017) 17–29

Fuel Consumption (L)

3.0
2.5
2.0
1.5
1.0
0.5
0
0 200 400 600 800 1000 1200 1400 1600
Time (s)

Fig. 14. Results for the fuel consumption of current vehicle.

Table 3
Comparison of simulation results.

Method Fuel consumption Increment of SoC Fuel consumption per hundred kilometers

Fuel consumption optimization 2.975 L 1.24% 9.276 L/(100 km)

ACC þ EMS 3.109 L 0.16% 9.740 L/(100 km)

80

70
Following Distance (m)

60

50

40

30
Maximum
20 Minimum
Optimal Area
10 Actual I
Actual II
0
0 10 20 30 40 50 60
Time (s)

Fig. 15. Comparison of the following distance of two methods in first 60 s.

engine and EM according to the desired acceleration. This strategy ensures the dynamic following performance to be
optimal firstly and optimizes fuel consumption secondly. The cost functions of dynamic indexes and economic indexes are
same as that of fuel consumption optimization method (proposed method), so the results of simulation are also similar to
that of proposed method. Here, we only analyze some differences.
The SoC and fuel consumption at last are 65.16% and 3.109 L gasoline. The equivalent fuel consumption is 3.107 L gasoline,
so the fuel consumption per hundred kilometers is approximate 9.740 L/(100 km). Compared with the results before, the
fuel consumption of ACCþEMS method is 5% more than that of proposed method. The comparisons of these two methods
are shown in Table 3.
It is worth noting the differences of the following distance of two methods. For convenience, the first 60 s of the whole
process is taken out to analyze the following distance, as shown in Fig. 15, where the solid line (Actual I) refers to the results
of proposed method and the dot line (Actual II) refers to the results of ACC þEMS. It can be seen that the dot line is within
the optimal following distance area while the solid line is out of the optimal area sometimes, which means the ACC þEMS
has a better following performance because the ACC is prior to EMS. That is to say, the fuel consumption optimization
method sacrifices dynamic performance to obtain the better economic performance. In fact, a driver is not sensitive to the
reduction of dynamic performance in this extent, but the improvement of economic performance is concerned by a driver.

6. Conclusions

This paper proposes a method of fuel consumption optimization by using a nonlinear MPC for s-HEV during a car-
following process, which fuses the ACC and EMS but is not a simple addition of two methods. To reduce the computing
burden, some search skills are used in optimizing algorithm of MPC. Simulations are carried out in MATLAB/SIMULINK, and
results show that for a 2000 kg SUV, on the condition of ensuring the car-following performances, the fuel consumption per
hundred kilometers is 9.276 L/(100 km), which is 5% less than that of ACCþEMS method. The proposed method may have a
disadvantage of large computing burden, but as the onboard computers develop so fast the computing burden is not an
intricate problem. The studies in the future will focus on two aspects: one is studying more high-efficiency and better-effect
methods to achieve more comfort and economic car-following process, and another is taking the shifting process, mode
 L. Li et al. / Mechanical Systems and Signal Processing 87 (2017) 17–29 29

transition and engine idle-stop technology into account for the car-following process of s-HEVs. Besides, the bench tests and
vehicle experiments are also the emphases we are working on.

Acknowledgment

The authors are very grateful to the China government by the support of this work through the National Science Fund for
Excellent Young Scholars of the People's Republic of China (Grant no. 51422505), and the National Key Technology R&D
Program of the Ministry of Science and Technology (Grant no. 2013BAG14B01).

References

[1] D. Crolla, D. Cao, The impact of hybrid and electric powertrains on vehicle dynamics, control systems and energy regeneration, Veh. Syst. Dyn. 50
(2012) 95–109.
[2] H. Zhang, J. Wang, Y. Wang, Optimal dosing and sizing optimization for a ground vehicle diesel engine two-cell selective catalytic reduction system,
Trans. Veh. Technol. http://dx.doi.org/10.1109/TVT.2015.2476760.
[3] H. Zhang, J. Wang, Y. Wang, Nonlinear observer design of diesel engine selective catalytic reduction systems with NOx sensor measurements, IEEE/
ASME Trans. Mechatron. 20 (4) (2015) 1585–1594.
[4] X. Hu, J. Jiang, B. Egardt, D. Cao, Advanced power-source integration in hybrid electric vehicles: multicriteria optimization approach, IEEE Trans. Ind.
Electron. 62 (12) (2015) 7847–7858.
[5] X. Zhu, H. Zhang, D. Cao, Z. Fang, Robust control of integrated motor-transmission powertrain system over controller area network for automotive
applications, Mech. Syst. Signal Process. 58–59 (2015) 15–28.
[6] C. Yang, X. Jiao, L. Li, Y. Zhang, L. Zhang, J. Song, Robust coordinated control for hybrid electric bus with single-shaft parallel hybrid powertrain, IET
Control Theory Appl. 9 (2) (2014) 270–282.
[7] Y. Luo, T. Chen, K. Li, Multi-objective decoupling algorithm for active distance control of intelligent hybrid electric vehicle, Mech. Syst. Signal Process.
64–65 (2015) 29–45.
[8] C. Hu, H. Jing, R. Wang, F. Yuan, M. Chadli, Robust H1 output-feedback control for path following of autonomous ground vehicles, Mech. Syst. Signal
Process. 70–71 (2016) 414–427.
[9] Y. Luo, T. Chen, S. Zhang, K. Li, Intelligent hybrid electric vehicle ACC with coordinated control of tracking ability, fuel Economy, and ride comfort, IEEE
Trans. Intell. Transp. Syst. 16 (4) (2015) 2303–2308.
[10] G. Tagne, R. Talj, A. Charara, Design and validation of a robust immersion and invariance controller for the lateral dynamics of intelligent vehicles,
Control Eng. Pract. 40 (2015) 81–92.
[11] M. Lio, F. Biral, E. Bertolazzi, M. Galvani, P. Bosetti, D. Windridge, A. Saroldi, F. Tango, Artificial co-drivers as a universal enabling technology for future
intelligent vehicles and transportation systems, IEEE Trans. Intell. Transp. Syst. 16 (1) (2015) 244–263.
[12] X. Hu, N. Murgovski, L.M. Johannesson, B. Egardt, Energy efficiency analysis of a series plug-in hybrid electric bus with different energy management
strategies and battery sizes, Appl. Energy 111 (2013) 1001–1009.
[13] L. Li, C. Yang, Y. Zhang, L. Zhang, J. Song, Correctional DP-based energy management strategy of plug-in hybrid electric bus for city-bus-route, IEEE
Trans. Veh. Technol. 64 (7) (2015) 2792–2803.
[14] H. Borhan, A. Vahidi, A. Phillips, M. Kuang, I. Kolmanovsky, S. Carrano, MPC-based energy management of a power-split hybrid electric vehicle, IEEE
Trans. Control Syst. Technol. 20 (3) (2012) 593–603.
[15] S. Zhang, R. Xiong, Adaptive energy management of a plug-in hybrid electric vehicle based on driving pattern recognition and dynamic programming,
Appl. Energy 155 (2015) 68–78.
[16] L. Li, S. You, C. Yang, B. Yan, J. Song, Z. Chen, Driving-behavior-aware stochastic model predictive control for plug-in hybrid electric buses, Appl. Energy
162 (2016) 868–897.
[17] P. Sarkouri, A. Ordys, Nonlinear model predictive control approach in design of adaptive cruise control with automated switching to cruise control,
Control Eng. Pract. 26 (2014) 160–177.
[18] S. Li, Z. Jia, K. Li, B. Cheng, Fast online computation of a model predictive controller and its application to fuel economy-oriented adaptive cruise
control, IEEE Trans. Control Syst. Technol. 16 (3) (2015) 1199–1209.
[19] D. Zhao, Z. Hu, Z. Xia, C. Alippi, Y. Zhu, D. Wang, Full-range adaptive cruise control based on supervised adaptive dynamic programming, Neuro-
computing 125 (2014) 57–67.
[20] J. Wang, X. Xu, D. Liu, Z. Sun, Q. Chen, Self-learning cruise control using kernel-based least squares policy iteration, IEEE Trans. Control Syst. Technol.
22 (3) (2014) 1078–1087.
[21] L. Li, G. Jia, J. Chen, H. Zhu, D. Cao, J. Song, A novel vehicle dynamics stability control algorithm based on the hierarchical strategy with constrain of
nonlinear tyre forces, Veh. Syst. Dyn. 53 (8) (2015) 1093–1116.
[22] Y. Wei, C. Oertel, Y. Liu, X. Li, A theoretical model of speed-dependent steering torque for rolling tyres, Veh. Syst. Dyn. http://dx.doi.org/10.1080/
00423114.2015.1111391.
[23] L. Li, X. Ran, K. Wu, J. Song, Z. Han, A novel fuzzy logic correctional algorithm for traction control systems on uneven low-friction road conditions, Veh.
Syst. Dyn. 53 (6) (2015) 711–733.
[24] E. Kural, B. Aksun, Integrated adaptive cruise control for parallel hybrid vehicle energy management, IFAC Pap. OnLine 48–15 (2015) 313–319.
[25] Y. Sun, L. Li, B. Yan, C. Yang, G. Tang, A hybrid algorithm combining EKF and RLS in synchronous estimation of road grade and vehicle' mass for a hybrid
electric bus, Mech. Syst. Signal Process. 68–69 (2016) 416–430.
[26] X. Zhu, H. Zhang, Z. Feng, Speed synchronization control for integrated automotive motor-transmission powertrain system with random delays, Mech.
Syst. Signal Process. 64–65 (2015) 46–57.
[27] B. Alt, F. Antritter, F. Svaricek, M. Schultalbers, Multivariable speed synchronisation for a parallel hybrid electric vehicle drivetrain, Veh. Syst. Dyn. 51
(3) (2013) 321–337.
[28] Y. Zou, X. Hu, H. Ma, S. Li, Combined state of charge and state of health estimation over lithium-ion battery cell cycle lifespan for electric vehicles,
J. Power Sources 273 (2015) 793–803.