Energy Conversion and Management 272 (2022) 116383

Contents lists available at ScienceDirect

Energy Conversion and Management

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

Hierarchical energy management strategy for fuel cell/ultracapacitor/

battery hybrid vehicle with life balance control
Jianjun Hu a, b, *, Zhouxin Wang a, Hao Du a, Lingbo Zou a
a
College of Mechanical and Vehicle Engineering, Chongqing University, Chongqing 400044, China
b
State Key Laboratory of Mechanical Transmission, Chongqing University, Chongqing 400044, China

A R T I C L E I N F O A B S T R A C T

Keywords: Fuel cell (FC) is an ideal power source for electric vehicles with high efficiency and little pollution. However,
Hierarchical control with its weak dynamic reaction, it needs to be used in combination with other energy storage devices, and the
Fuel cell vehicle coupling of multiple energy sources makes it difficult to fully utilize the system performance. Meanwhile, the cost
Composite energy storage
and durability of FC hinder its commercialization. This paper takes the composite energy storage system
Energy management
Life balance
composed of FC, battery (BAT) and ultracapacitor (UC) as the research object. In order to improve the system
Hardware-in-the-loop performance, a hierarchical energy management strategy is proposed based on the analysis of the relationship
between UC output power and vehicle demand power under dynamic programming (DP) strategy, taking into
account the vehicle energy consumption and energy source lifetime. A fuzzy control based on speed prediction
and a dynamic state-of-charge (SOC) threshold control optimization algorithm of UC based on working condition
recognition are designed to optimize the hierarchical control strategy. In the upper layer strategy, for realizing
the online real-time distribution of UC output power, the generalized regression neural network is used to learn
the relationship between UC output power and demand power under DP strategy; in the lower layer strategy, the
power of the FC and BAT is distributed using an equivalent energy minimization strategy. Based on above
strategy, this paper proposes a life balance control strategy for energy sources to maximize the lifetime of the
composite energy storage system. Simulation and experimental results show that the proposed strategy has a
comparable life-cycle average operating cost to the DP strategy with better life mileage.

1. Introduction source coupling. Reasonable parameter matching and effective energy

management strategies are therefore crucial [14].
Fuel cell (FC) is an ideal power source for electric vehicles with high Current studies in this field can be classified into rule-based strate­
efficiency and little pollution [1–3]. However, due to its weak dynamic gies and optimization-based strategies. Rule-based energy management
reaction, it is unable to track the changes in load whether the vehicle is strategy (EMS) is simple and intuitive, well developed, and requires no
beginning, braking, or accelerating [4,5]. To solve this problem, it must priori information, but it relies too much on expert experience and is
be integrated with other energy storage devices to create a composite extremely susceptible to changes in working conditions [15–17]. Yuan
energy storage fuel cell system [6,7]. A three-energy source system et al. [18] used a logical threshold filtering strategy to achieve power
composed of fuel cell, battery (BAT) and ultracapacitor (UC) is currently distribution in a FC/BAT/UC three-energy source system with FC as the
the most promising one. The advantages of each component can be primary energy source and a storage system consisting of BAT and UC as
utilized to their full potential in this system [8–11]. Ultracapacitor has the secondary energy source. Tao et al. [19] used fuzzy control and
the ability of fast charging and discharging [12,13], which makes up for frequency decoupling to distribute the demand power for FC/BAT/UC
the slow dynamic effect of fuel cells and the unsuitability of batteries for vehicle, which reduced the voltage fluctuation of FC and BAT and
high-current charging and discharging. Battery makes up for the low improved the service life. Ahmadi et al. [20] proposed a novel power
energy density of the ultracapacitor. However, design matching and sharing method based on fuzzy control, where the control parameters
control are made more challenging by the existence of multiple design are accurately adjusted by the genetic algorithm (GA), improving the
control degrees of freedom, multiple operating modes, and energy dynamic performance and the equivalent fuel economy of the fuel cell

* Corresponding author.
E-mail address: hujianjun@cqu.edu.cn (J. Hu).

https://doi.org/10.1016/j.enconman.2022.116383
Received 12 July 2022; Received in revised form 18 October 2022; Accepted 19 October 2022
Available online 29 October 2022
0196-8904/© 2022 Elsevier Ltd. All rights reserved.
J. Hu et al. Energy Conversion and Management 272 (2022) 116383

Fig. 1. The powertrain structure of FCHEV.

hybrid vehicle. Kaya and Hames [21] proposed a hydrogen fuel saving frequency power, enhancing the lifespan of fuel cell and the dynamic
control strategy and a life cycle saving control strategy. The strategies performance of electric vehicles.
were applied to different road models, including the stop–go road model In the above literature, when considering the FC life, instead of
and the uphill-downhill model. Both strategies improved the vehicle fuel directly taking into account the FC life degradation in the control
economy and life cycle of the system components compared with the strategy objective function, most of them extend the service life by
state machine control strategy. limiting its working state, such as limiting FC power fluctuation, setting
Optimization-based strategies can be further classified into global power upper and lower limits, or reducing the start/stop times of FC,
optimization strategies and local optimization strategies. The dynamic etc., which could not obtain the best optimization effect. Additionally, in
programming algorithm (DP) is a classical global optimization algo­ current studies, UC is mostly used as an auxiliary energy source to fill in
rithm that can obtain an ideal result [22–25]. Li [26] designed a the power gaps left by FC and BAT. However, since UC has a high power
multidimensional DP algorithm to solve an optimization problem that density, instantaneous high power charge and discharge capabilities,
takes into account both fuel economy and FC performance degradation and a long cycle life, it has the potential to significantly reduce the strain
rate. Although the control approach based on the global optimization on FC and BAT and enhance the dynamic performance of the vehicle if
algorithm can effectively increase the energy efficiency of the composite its utilization rate can be increased.
energy storage system, its drawback of requiring advanced knowledge of In this paper, for the three-energy source composite energy storage
future working conditions makes it challenging to implement in system, based on the analysis of the relationship between UC power and
practice. demand power under the DP strategy, a hierarchical energy manage­
Considering the real-time application of control strategies, local ment strategy with life balance control is proposed for composite energy
optimization strategies are developed, such as Model Predictive Control storage fuel cell vehicles, taking into account the vehicle energy con­
(MPC) [27–28] and Equivalent Consumption Minimum Strategy (ECMS) sumption, energy source life, and the balance of energy source life. This
[29–32]. Sellali et al. [27] designed a MPC framework for FC/BAT/UC strategy effectively increases the anticipated vehicle life mileage and
vehicle, which minimized battery degradation, prolonged the service lowers the comprehensive operating cost.
life and reduced economic cost. Although the MPC technique helps the The remainder of this paper is structured as follows. In Section 2, the
car get greater fuel economy, it still has heavy computational load and model of powertrain, including FC, BAT and UC, is described. In Section
lengthy computation time that make real-time control difficult. The 3, the hierarchical energy management strategy is proposed and opti­
ECMS does not need to anticipate future working conditions, and the mized. In Section 4, the hierarchical strategy is further improved by
computational effort is much smaller than that of the global optimiza­ adding a life balance control of energy sources; simulation and
tion strategy, but its optimization result is significantly affected by the hardware-in-the-loop testing are performed to verify its effectiveness.
equivalence factor. Li et al. [30] used an ECMS for power allocation of Finally, conclusions are drawn.
FC/BAT/UC trams to improve the system economy, but the equivalent
hydrogen consumption of UC was neglected in their ECMS objective 2. Power system modeling
function. Subsequently, Li et al. [31] proposed an online adaptive
equivalent consumption minimization strategy for composite energy The powertrain structure of the fuel cell hybrid electric vehicle
storage fuel cell vehicles, which adjusts the equivalence factor and the (FCHEV) in this study is shown in Fig. 1. An electric motor drives the bus
dynamic current change rate of FC according to the state of health of FC with a hybrid power source composed of a proton-exchange-membrane
and BAT to extend the service life of FC and BAT. Jiang et al. [33] fuel cell (PEMFC) system, a lithium-ion battery system and an ultra­
proposed a real-time energy management strategy based on a two- capacitor system. A unidirectional DC/DC converter has been used with
dimensional Pontryagin’s Minimal Principle, which can approximate the fuel cell system to enhance its voltage level towards the desired level
the optimal strategy of DP. Tao et al. [34] used the wavelet transform to and sustain this at the outputs. A bidirectional DC/DC converter has
separate the high-frequency power and used ECMS to distribute the low- been adopted at the side of the ultracapacitor system, where it works as a

2
J. Hu et al. Energy Conversion and Management 272 (2022) 116383

boost converter in discharging mode and as a buck converter in charging Table 1

mode. The battery system is passively connected to the bus. The pa­ Basic parameters of FCHEV.
rameters of the principal powertrain components are shown in Table 1. Item Parameter Value

Vehicle Gross weight, ma 2270 kg

2.1. Fuel cell modeling Curb weight, m 1870 kg
Frontal area, A 2.39 m2
Air resistance coefficient, CD 0.29
The FC model in this study is a steady-state model based on empirical
Air resistance coefficient, f 0.015
data. The mass consumption of hydrogen is related to the FC output Rolling radius, r 0.312 m
power [35], as shown in Eq. (1), where Pfc is the PEMFC system output Conversion coefficient of rotating 1.02
power; η fc is the PEMFC system efficiency; and LHV is the low heating mass,δ
Drivetrain efficiency, η 0.93
value of hydrogen.
Fuel cell Rated power, PFC 30 kW
∫t Battery Model name LP2270134
Pfc
mfc = dt (1) Rated pack capacity 22.528Ah (8.192
0 LHV⋅ηfc kWh)
Cullen efficiency 0.93
Fuel cell systems will face problems including cell aging and per­
Ultracapacitor Ultracapacitor energy 0.296 kWh
formance degradation as operating times rise. Previous studies have DC/DC Efficiency 0.9
demonstrated that fuel cell service life is correlated with the frequency converter
of starts and stops, load variation, idle time, and high power load Motor Peak power 80 kW
[36–38]. In this study, an empirical model of fuel cell life is developed by
referring to the method in literature [36], as shown in Eq. (2) and (3),
ΔP Table 2
Tf = (2) Fuel cell degradation parameters.
ΔPfcloss
Coefficient Value Definition
ΔPfcloss = kp (P1 n1 + P2 n2 + P3 t1 + P4 t2 ) (3) P1 0.00196 %/cycle A start-stop cycle of fuel cell stack
P2 0.0000593 %/ Power variation rate greater than 10 % Pfcmax
where Tf is the fuel cell service life, ΔP is the maximum allowed voltage cycle
degradation of the fuel cell (generally 10 %); ΔPfcloss is the fuel cell life P3 0.00126 %/h Output power less than 5 % Pfcmax
P4 0.00147 %/h Output power greater than 90 % Pfcmax
degradation rate; kp is the correction coefficients for different fuel cell kp 1.02 Correction factors for different fuel cell systems
systems; the coefficients P1, P2, P3 and P4 are determined experimen­
tally, with values referenced to the literature [36], as shown in Table 2;
n1, n2, t1, and t2 are the number of starts and stops per hour, the number
of large load changes, idling speed and high power running time, where, k is the number of time intervals after discretizing the working
respectively. condition by step dt; n is the discharge multiplier at the time point k; I1 is
the battery 1C discharge current; R is the gas constant; T is the Kelvin
temperature when the battery is in operation; B1 is the pre-exponential
2.2. Battery modeling
factor when the battery is discharged at 1C, calculated as Eq. (6).
The internal resistance model is used for the battery model. The B1 = 32350e− 0.3375
+ 4041e0.1271 (6)
battery voltage and internal resistance are obtained by checking tables
in the driving process. The state-of-charge (SOC) of battery, SOCbat, Generally, the battery life is considered to be terminated when the
cannot be obtained by direct measurement and is calculated based on percentage of capacity degradation reaches 20 %. The cumulative
the current integral, as in Eq. (4), ampere-hours at the end of battery life, Ah(20%), can be calculated ac­
⎧ ∫ cording to Eq. (7).
⎪
⎪ 1 t − 31329.7 ( )0.55
⎪
⎨ SOC 0 − ibat dt, ibat ⩾0 Qloss(20%) = B1 e RT Ah(20\% ) (7)
C 0
SOCbat = ∫ (4)
⎪
⎪
⎪ 1 t After calculation, when the lithium iron phosphate capacity degra­
⎩ SOC0 − ibat ηchg dt, ibat < 0
C 0 dation up to 20 %, the cumulative ampere-hour Ah(20%) = 18600A. The
battery capacity degradation percentage Lbatloss can be calculated as Eq.
where SOCbat is the current SOC of the battery; SOC0 is the initial SOC of (8).
the battery; C is the capacity of the battery; ibat is the current of the
battery; t is the time of charging and discharging. Ah
Lbatloss = (8)
The power battery capacity degradation model refers to the cycle life Ah (20%)
prediction model for lithium iron phosphate batteries proposed by Luo
et al. [39]. The battery capacity degradation model after N cycle
working conditions is as Eq. (5),

{ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ ( ) }0.55
∑ 0.55 dt
(5)
− 31329.7
Qloss = N⋅ (1.169e − 0.3375n + 0.146e 0.1271n )⋅e (− 0.1494+0.1494n) ⋅ nI1 ⋅B1 e RT , n⩽10
k
3600

3
J. Hu et al. Energy Conversion and Management 272 (2022) 116383

Fig. 2. The power distribution curve of the three energy sources and the relationship between UC output power and motor demand power.

2.3. Ultracapacitor modeling voltage of UC and Uuc_min is the minimum voltage of UC.
Uoc − Uuc min
The equivalent circuit model of capacitor and resistor in series is SOCuc = (10)
Uuc max − Uuc min
adopted for UC, and the UC current Iuc is calculated according to the
capacitor demand power Puc as shown in Eq. (9), where Uoc is the UC
3. Hierarchical energy management strategy and optimization
terminal voltage and Ruc is the UC internal resistance.
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
Taking into account the power coupling effect of the composite en­
Uoc − U 2oc − 4Ruc Puc
Iuc = (9) ergy storage system, a hierarchical energy management strategy is
2Ruc proposed in this section. A fuzzy control based on vehicle speed pre­
The SOCuc is calculated from Eq. (10), where Uuc_max is the maximum diction and a dynamic threshold control of ultracapacitor SOC based on

4
J. Hu et al. Energy Conversion and Management 272 (2022) 116383

Fig. 2. (continued).

working condition recognition are developed to optimize the strategy. source life economy includes the life degradation cost of FC and BAT.
∑
k=k max

3.1. Analysis of the output power of three energy sources under DP J= *

[a1 JH∗ 2 (k) + a2 Jele (k) + a3 Jfc* (k) + a4 Jbat
*
(k)] (11)
strategy k=1

a1 , a2 , a3 , a4 is the weighting factor of the corresponding cost item.

For a three-energy source composite system, after determining the JH2 (k) is the FC hydrogen consumption cost; J ∗H2 (k) is its normalization.
output power of any two of the energy sources, the output power of the Jele (k) is the electricity consumption cost; J ∗ele (k) is its normalization.
other energy source can be calculated when the motor demand power is J fc (k) is the FC life degradation discounted cost; J ∗fc (k) is its normali­
known. Therefore, in this study, the output power of FC and UC, Pfc and
zation. Jbat (k) is the BAT life degradation discounted cost; J ∗bat (k) is its
Puc , are used as the control variables of DP strategy, and the SOC of BAT
normalization. JH2 (k), Jele (k), J fc (k), Jbat (k) are calculated as Eq. (12),
and UC are used as the state variables. With aims to achieve better
vehicle energy consumption economy and energy source lifetime
economy, the objective function shown in Eq. (11) is established. The
energy consumption economy includes the hydrogen consumption cost
of FC and the power consumption cost of BAT and UC, and the energy

5
J. Hu et al. Energy Conversion and Management 272 (2022) 116383

Fig. 2. (continued).

⎧
⎪
⎪ JH2 (k) = ΔmH2 × priceH2 battery price; Qbat is the battery pack capacity; Ubat is the battery pack
⎪
⎪
⎪
⎪ voltage.
⎪ J (k) = (Pbat (k) + Puc (k) ) × dt × priceele
⎪ ele
⎪
⎨ Based on the DP strategy, the power distribution of the three energy
Jfc (k) =
ΔPfcloss
× pricefc × Pfc (12) sources and the relationship between the UC output power and the
⎪
⎪
⎪
⎪ 10% motor demand power under nine typical cycle conditions are simulated
⎪
⎪
⎪
⎪
⎪
Ah on the Matlab/Simulink platform, as shown in Fig. 2. Among the nice
⎩ Jbat (k) = × pricebat × Qbat × Ubat
Ah (20%) typical conditions, MANHATTAN, NYCC and INDIA_URBAN are urban
conditions; UDDS, NEDC and IM240 are suburban conditions; HWFET,
where, ΔmH2 is FC hydrogen consumption; priceH2 is the hydrogen price; WVUINTER, and US06_HWY are highway conditions.
Pbat (k), Puc (k) is the BAT and UC output power at moment k; priceele is By analyzing Fig. 2, it can be found that there is a certain relationship
the residential electricity price; ΔPfcloss is the FC life degradation per­ between the UC output power and the motor demand power: 1) the UC
centage; pricefc is the FC price; Pfc is the FC system power; Ah is the BAT output power varies approximately linearly with the demand power. 2)
cumulative ampere-hours; Ah (20%) is the cumulative ampere-hours The scatter plots for all nine drive cycles have similar region 1, where the
when the battery capacity degradation up to 20 %; pricebat is the demand power at these moments is mainly provided by UC. 3) The

6
J. Hu et al. Energy Conversion and Management 272 (2022) 116383

Fig. 3. Hierarchical energy management strategy schematic.

energy management strategy is proposed in this paper, as shown in

Table 3
Fig. 3. Firstly, from information of drive cycle, the vehicle demand
MIVs for different variables.
power can be calculated. Secondly, by using General Regression Neural
Characteristic variable MIV Relative Cumulative Network (GRNN) in the upper layer strategy, UC output power is ob­
contribution contribution
tained. Lastly, the ECMS in the lower layer strategy can distribute the
Demand power, Pdem 0.6074 40.57 % 40.57 % output power of FC and BAT. Then, the power distribution of three
Vehicle speed, v − 0.4127 27.56 % 68.13 %
energy sources is clear at every moment.
Vehicle acceleration, a − 0.2623 17.52 % 85.65 %
Battery output power at the 0.1046 6.99 % 92.63 %
last moment, Pbat,pre 3.2.1. Upper layer energy management strategy
Ultracapacitor output power at 0.0953 6.36 % 99.00 % In the upper layer strategy, GRNN is used to learn the law of the
the last moment, Puc,pre relationship between the output power of UC and the demand power
Ultracapacitor charge state, − 0.0095 0.63 % 99.63 %
under the DP strategy in order to realize the online real-time distribution
SOCuc
Battery charge state, SOCbat 0.0034 0.23 % 99.86 % of the output power of UC. To improve prediction accuracy while
Fuel cell output power at the 0.0021 0.14 % 100 % reducing controller computing time, Mean Impact Value (MIV) analysis,
last moment, Pfc,pre which can reflect changes of the weight matrix in the neural network
and better evaluate the correlation of variables, is used for the screening
of neural network input variables [40]. The DP results under nine drive
scatter plots of drive cycle (c) to (i) all have similar region 2 and region
cycles in Section 3.1 were used as the training samples of the GRNN, and
3. Region 2 indicates pre-charging of the UC for the ensuing high-power
the eight feature parameters in Table 3 were initially selected as the
discharge demand, and region 3 indicates a hybrid drive of the UC with
input variables of the network. The MIVs of these variables are shown in
other power sources. 4) Region 4 in drive cycle (i) indicates that the
Table 3. Its sign represents the direction of the correlation, and the ab­
braking power of the working condition exceeds the power limit of the
solute value represents the relative importance of the impact [41].
UC. In order to brake safely and protect the UC, the UC recovers the
As shown in Table 3, the relative contributions of vehicle speed v,
braking energy with the maximum power at this time and completes
acceleration a, and demand power Pdem are all higher than 10 %, and the
braking together with the mechanical structure.
cumulative contribution reaches 85.31 %, which indicates that these
In summary, because the driving conditions information is known in
three variables have a greater influence on the output power of UC.
the DP strategy, the UC can adjust its SOC in advance according to the
Consequently, v, a, and Pdem are selected as the input variables of GRNN
power demand of future conditions. Therefore, the UC’s utilization rate
in this study.
is high, and the output ratio of FC and BAT can then be reduced. As a
result, the life degradation of both FC and BAT is slowed and their ser­
3.2.2. Lower layer energy management strategy
vice life is extended.
In the lower layer strategy, the power of the FC and the BAT is
distributed by the ECMS. The core idea is to convert the electrical con­
3.2. Hierarchical energy management strategy sumption of BAT and UC into equivalent hydrogen consumption and to
minimize the sum of equivalent hydrogen consumption and direct
From the above analysis, it is clear that under the DP strategy, in hydrogen consumption. The objective optimization function is shown in
order to reduce hydrogen consumption and FC life degradation, the FC Eq. (13),
operates at almost minimum power most of the time, and the majority of
the demand power is provided by the UC and the battery. To ensure that mH2 (t) = mfc (t) + kbat mbat (t) (13)
the UC has power available when future demand power is high and has
where mH2 (t) is the total hydrogen consumption at time t, mfc (t) is the
the ability to recover when future braking energy is high, the UC needs
direct hydrogen consumption of FC at time t, and mbat (t) is the
to have pre-charge and pre-discharge functions. Thus, a hierarchical

7
J. Hu et al. Energy Conversion and Management 272 (2022) 116383

Table 4 Table 5
Equivalent factors of typical cycles for DP optimization. Comparison of simulation results of each strategy.
Driving Drive cycle name Battery charge/ Average value Parameter DP Rule-based Hierarchical
condition discharge strategy strategy strategy
equivalence factor
Hydrogen consumption (g) 71.37 111.5 124.4
Urban INDIA_URBAN 0.9698/1.0211 0.9689/ Electricity consumption 2.995 2.27 2.425
MANHATTAN 0.9694/1.0206 1.0229 (kWh)
NYCC 0.9693/1.0206 Fuel cell life degradation (%) 2.4 × 10-3 4.04 × 10-3 2.909 × 10-3
Suburban UDDS 0.9696/1.0218 Battery capacity degradation 5.2 × 10-2 2.85 × 10-3 5.643 × 10-2
NEDC 0.9686/1.0227 (%)
IM240 0.9673/1.0237 Comprehensive operating 15.58 22.21 19.53
Highway HWFET 0.9676/1.0269 cost (yuan)
US06_HWY 0.9648/1.0317
WVUINTER 0.9696/1.0254
each condition is taken as the charge/discharge equivalent factor of the
battery in the ECMS.

3.2.3. Simulation analysis of multi-objective hierarchical energy

Pfc-max
C management strategy
The Worldwide Harmonized Light Vehicles Test Cycle (WLTC) con­
Pbat-max sists of low speed, medium speed, high speed, and super high speed
B
parts, reaching a maximum speed of 131 km/h. This test condition is
Puc-max close to the vehicle driving conditions on the real road, which can well
A
reflect the real energy consumption level of the vehicle. Therefore, in
this section, the WLTC condition is selected as the test driving condition,
and the proposed hierarchical strategy is compared with the DP strategy
D
and a rule-based strategy. The rule-based strategy is as follows: the en­
Puc-imax
ergy source provides power in the order of UC, BAT, and FC, and the
E strategy schematic is shown in Fig. 4. Under drive conditions, the UC
Pbat-imax provides demand power when it has enough power (point A); when the
F demand power is greater than the UC maximum power, the UC outputs
at the maximum power and drives together with the BAT (point B); if the
maximum output power of UC and BAT is exceeded by the demand
Fig. 4. Rule-based strategy schematic.
power, the shortage is then provided by FC (point C). Under braking
conditions, the braking energy is first recovered by the UC (point D);
equivalent hydrogen consumption of BAT at time t. kbat is the SOC
when the UC cannot recover more energy, the BAT is then used to
compensation factor of BAT, which can limit its SOC to a certain range.
recover the braking energy (point E); if neither BAT nor UC can recover
kbat can be calculated by Eq. (14),
energy, then mechanical braking is used to consume energy (point F).
2SOCbat − (SOCbat max + SOCbat min ) The simulation results are shown in Table 5. It can be seen that
kbat = 1 - μ⋅ (14) compared with the rule-based control strategy, the hydrogen con­
SOCbat max + SOCbat min
sumption, power consumption and battery capacity degradation under
where μ is used to balance the SOC of the BAT during the driving pro­ the hierarchical control strategy are all slightly increased, but the fuel
cess; SOCbat_max is the upper limit of SOCbat; SOCbat_min is the lower limit cell life degradation is significantly reduced. Because of the high cost of
of SOCbat. FC, the comprehensive operating cost under WLTC driving condition is
The equivalent hydrogen consumption of the BAT is stated as Eq. reduced by 12.07 %, but there is still a large gap with the DP strategy. To
(15), find out the reasons for the gap, the power of UC and its SOC variation
curves under the two energy management strategies are compared, as
σ⋅Pbat ⋅Cfcavg
mbat = (15) shown in Fig. 5.
Pfcavg
From Fig. 5, it can be seen that the utilization rate of UC is signifi­
where σ is the BAT charge/discharge equivalent factor; Cfcavg is the cantly higher under DP strategy, which is due to the fact that DP strategy
average FC hydrogen consumption; Pfcavg is the average FC output knows all the driving conditions information and can plan and adjust the
power. The equivalent factor is calculated based on Eq. (16), where ηdis SOC of UC according to the future conditions, so that it can better cope
is the battery discharge efficiency; ηchg is the battery charging efficiency; with the power demand of future conditions. As shown in Fig. 5(b),
under the hierarchical control strategy, the SOC of UC rapidly drops to
ηdis avg is the average battery discharge efficiency; ηdis avg is the average
the lower limit after releasing energy. Due to the lack of information on
battery charging efficiency.
working conditions, the UC can only rely on passive charging of braking
⎧
⎪ 1 energy without actively recharging itself, which leads to a low utiliza­
⎨ , P ⩾0
σ = ηchg avg ⋅ηdis bat (16) tion rate. Therefore, it is necessary to optimize the hierarchical energy
⎪
⎩ management strategy.
ηdis avg ⋅ηchg , Pbat < 0
The DP algorithm is used to find the optimal control sequence for the
3.3. Optimization of hierarchical energy management strategy
nine drive cycles in Section 3.1 and then the BAT charge/discharge
equivalent factor can be calculated using the control parameters of each
To improve the utilization of the UC, a fuzzy control optimization
component. The calculation results of the equivalent factor for each
algorithm based on speed prediction is proposed in this section to
drive cycle are shown in Table 4, where the optimal equivalent factors
modify the UC power from GRNN and then the current UC power can be
for each condition under the DP strategy do not differ much. Therefore,
adjusted according to the power demand in the future period. In addi­
the average value of the optimal charge/discharge equivalent factors for
tion, to give full play to the characteristics of UC, the optimal threshold

8
J. Hu et al. Energy Conversion and Management 272 (2022) 116383

Fig. 5. Ultracapacitor output power and SOC curve.

Fig. 6. Schematic diagram of the optimized hierarchical energy management strategy.

Fig. 7. The prediction results of LSTM network speed predictor under WLTC cycle.

9
J. Hu et al. Energy Conversion and Management 272 (2022) 116383

Fig. 8. Condition recognition results of WLTC vehicle speed prediction.

of ultracapacitor SOC under different driving conditions is analyzed, and selected as the input of the fuzzy controller, and the UC power modifi­
a dynamic threshold control algorithm of ultracapacitor SOC based on cation coefficient k is obtained by the logic operation of the fuzzy
working condition recognition is proposed, which further strengthens controller. The modified UC output power Puc is shown in Eq. (17).
the adaptability of the hierarchical control strategy to different driving {
PGRNN ⋅(k + 1), k > 0
conditions. Puc = (17)
PGRNN ⋅(k − 1), k < 0
Fig. 6 shows the structure schematic of the optimized hierarchical
energy management strategy. Based on the future working condition
information provided by the vehicle speed prediction and the current 3.3.2. Dynamic threshold control of ultracapacitor SOC based on working
ultracapacitor SOC, the UC output power is corrected by fuzzy control condition recognition
and dynamic threshold control. Due to its low energy density, the UC’s SOC changes quickly, and
thus the threshold value of SOC has an impact on its utilization. In
3.3.1. Vehicle speed prediction and fuzzy control different driving conditions, the operation status and output ratio of UC
In this study, Long Short-Term Memory Networks (LSTM) is used for will also be different. In this section, by analyzing the characteristics of
speed prediction, and the training data consists of the nine standard different driving conditions, a SOC dynamic threshold control algorithm
working conditions used in Section 3.1. The prediction results of the of UC based on working condition recognition is proposed, and the UC’s
trained speed predictor under WLTC conditions are shown in Fig. 7, SOC threshold value is dynamically adjusted according to the results of
from which it can be seen that the predicted speed can follow the actual working condition recognition to further optimize its utilization rate.
speed and can provide proper information about the future speed for the The clustering algorithm used in [42] classifies the vehicle driving
hierarchical energy management strategy. conditions into urban congestion, suburban and highway conditions and
Since the pre-charge and pre-discharge of UC is mainly related to its constructs a condition recognizer based on GRNN to identify the vehicle
current SOC and future power demand Pnext, these two variables are driving conditions.
The dynamic threshold control strategy of ultracapacitor SOC is to

Fig. 9. Simulation results of hierarchical energy management strategy before and after optimization.

10
J. Hu et al. Energy Conversion and Management 272 (2022) 116383

Table 6 Table 7
Comparison of before and after optimization of hierarchical energy management Parameter setting of genetic algorithm.
strategy. Parameter Value
Parameter DP strategy Hierarchical Hierarchical
Population size 20
strategy strategy
Target parameters [μ, kbat_min, kbat_max]
before after
Initial population generation interval [0.4 0.65 1; 0.7 0.9 1.35]
optimization optimization
Crossover rate 0.7
Hydrogen consumption 71.37 124.4 120.6 Mutation rate 0.01
(g) Number of iterations 40
Electricity consumption 2.995 2.425 2.62
(kWh)
Fuel cell life degradation 2.4 × 10 -3
2.909 × 10 -3
2.375 × 10-3 4. Hierarchical energy management strategy with life balance
(%) control and its verification
Battery capacity 5.2 × 10-2 5.643 × 10-2 4.647 × 10-2
degradation (%)
Comprehensive operating 15.58 19.53 17.15 The service life of the composite energy storage system depends on
cost (yuan) the power source with the shortest life span, and the different expected
Energy cost (yuan) 4.80 6.55 6.53 lifetimes of the power sources will inevitably have a negative impact on
the lifetime of the composite energy storage system. Therefore, a hier­
archical energy management strategy with life balance control for FC
optimize the adaptability of hierarchical control strategy to different
and BAT is proposed in this section, aiming at a better life mileage and
working conditions and to better exploit the characteristics of UC. 1) In
economy of the studied fuel cell composite energy storage system
urban congestion conditions, since the vehicle speed is low and accel­
without replacing the energy source.
eration and deceleration are frequent, the UC needs to switch modes
frequently. Therefore, the upper limit of SOC should be larger to avoid
the shock of high current to the power battery due to insufficient
4.1. Life balance control of energy sources
available capacity of UC. 2) In suburban conditions, the time of vehicle
acceleration and deceleration is reduced compared to urban conditions,
In the composite energy storage system studied in this paper, the
but the vehicle speed increases and the instantaneous power required for
main power sources are FC and BAT, and UC is responsible for providing
acceleration and deceleration increases. Thus, the upper limit of the
instantaneous peak power or performing braking energy recovery. Since
ultracapacitor SOC should be moderate at this point so that it has
UC has excellent cycle life, only the life balance of FC and BAT is
enough charge to provide peak acceleration power and enough available
considered in this paper. To evaluate the comprehensive economy of the
capacity to recover the braking energy. 3) In the high-speed condition,
power system over its life cycle, this paper introduces the concept of the
the vehicle speed is high and the driving condition is stable. There are
life-cycle average operating cost, which is the ratio of the sum of the life-
few accelerations and decelerations. The UC works with a lower per­
cycle total energy cost and the power source configuration cost to the life
centage. Thus, its SOC upper limit value should be set low to avoid too
mileage of the composite energy storage system.
much charging that causes energy loss. The constructed condition
recognizer was used to recognize the speed prediction results of WLTC
4.1.1. Control parameter selection
conditions in Fig. 7, and the recognition results are shown in Fig. 8.
The life balance control strategy is proposed on the basis of the above
hierarchical energy management strategy, which still adopts an upper
3.3.3. Optimization effect of hierarchical energy management strategy
and lower hierarchical structure. The upper layer is responsible for the
The WLTC driving condition is selected for simulation. From Fig. 9
power distribution of the UC. Since the effect of the lifetime of UC is not
(b), it can be seen that the optimized composite energy storage system
considered in this study, the upper layer is kept consistent with the
can charge and discharge the UC in advance according to the future
optimized hierarchical strategy. In the lower level strategy, the ECMS is
demand, and the utilization rate of the UC has been significantly
used to allocate power between FC and BAT, and the allocation directly
improved compared with that before optimization. Table 6 shows that
affects the lifetime of both. Therefore, it is feasible to balance the service
the FC power fluctuation has been reduced as a result of improved UC
life of FC and BAT by selecting appropriate control parameters.
utilization. Compared with pre-optimization, the fuel cell life degrada­
Among the parameters of the lower control strategy, kbat is used as
tion has been reduced by 18.36 %, the battery capacity degradation has
the compensation coefficient of battery SOC, which can limit the SOC to
been reduced by 17.65 %, and the comprehensive operating cost of
a certain range. From Eq. (14), it can be seen that the parameter μ is used
WLTC conditions has been reduced by 12.19 %, narrowing the gap be­
to balance the battery SOC during the operation of the composite energy
tween the hierarchical energy management strategy and the DP strategy.
storage system and can indirectly affect the power distribution of the
This indicates that the optimization algorithm proposed in this section is
BAT. The output power of the BAT is limited by its charge state SOCbat.
able to reduce the lifetime degradation of FC and BAT by improving the
When SOCbat is higher, the BAT can output higher power and can reduce
utilization of UC, which results in a significant reduction of the
the fuel cell load; on the contrary, when the SOCbat is lower, it will limit
comprehensive operating cost.
its power output, and accordingly the FC load and life degradation will
From the results in Table 6, it can be found that the degradation rates
increase.
of FC and BAT are not the same, and the power source with a higher
In summary, the parameters μ and kbat can influence the life degra­
degradation rate will inevitably fail first and adversely affect the service
dation of FC and BAT by affecting the power distribution. Since there is a
life of the composite energy storage system. Therefore, in order to
correlation between μ and kbat, the maximum value and the minimum
extend the service life of the composite energy storage system and
value of kbat, kbat_max and kbat_max respectively, together with the
reduce the vehicle cost, it is necessary to optimize the power allocation
parameter μ are chosen as the control parameters of the life balance
of the composite system by taking the power source life balance into
control strategy.
account in the energy management strategy.
4.1.2. Genetic algorithm-based solution for control parameters
Since the energy management strategy proposed in this paper is a
complex nonlinear system, in which there are many parameters to be
optimized and certain correlations between them, GA is used to find the

11
J. Hu et al. Energy Conversion and Management 272 (2022) 116383

Fig. 10. Optimization results of genetic algorithm.

Based on the Simulink model of the vehicle powertrain, GA is used to

Table 8 optimize the selected control parameters. The control parameter opti­
Comparison of simulation results for each energy management strategy.
mization range is determined according to reference [43] and the SOC
Parameter DP Optimized Hierarchical strategy limits of BAT, and here the SOC range of BAT is set as [0.3,0.9]. The GA
strategy hierarchical with life balance
parameters are set as shown in Table 7.
strategy control
The results of the GA optimization search are shown in Fig. 10, where
Hydrogen 71.37 120.6 122.7 μ is 0.571, kbat_min is 0.7015, kbat_max is 1.1994, and the optimal life-cycle
consumption (g)
Electricity 2.995 2.62 2.632
average operating cost is 0.683 yuan.
consumption (kWh)
Fuel cell life 2.4 × 10- 2.375 × 10-3 2.19 × 10-3 4.1.3. Simulation analysis
3
degradation (%) From the simulation results in Table 8, it can be seen that the fuel cell
Battery capacity 5.2 × 10- 4.647 × 10-3 4.38 × 10-3
3 and battery life degradation rate of the strategy proposed in this section
degradation (%)
Comprehensive 15.58 17.15 16.47 is approximately the same (because the end-of-life degradation rates of
operating FC and BAT are 10 % and 20 %, respectively). In addition, compared to
cost (yuan) the optimized hierarchical energy management strategy in Section 3.3,
Life mileage 89,500 100,130 106,180 the FC and BAT life degradation rates decrease by 7.79 % and 5.75 %,
expectancy (km)
Life-cycle average 0.683 0.707 0.683
respectively, but the energy consumption increases by 1.41 %. The
operating cost comprehensive operating cost under WLTC driving condition is reduced
(yuan) by 3.97 %, which is closer to the result of the DP strategy, while the life-
cycle average operating cost is reduced by 3.39 % compared with the
optimized hierarchical strategy, which is at the same level as the DP
optimal control parameters of the life balance control strategy.
strategy.
1) The construction of optimization objective function.
Since the optimization objective of the DP strategy is the optimal
In this study, life-cycle average operating cost is the optimization
comprehensive operating cost under the WLTC driving condition,
objective, and the objective function is shown in Eq. (18).
although its life-cycle average operating cost is at the same level as that
Fall mH2 ⋅priceH2 ⋅min(Lfc , Lbat ) + FS of the hierarchical strategy with life balance control, the life mileage
F(t) = = (18) expectancy under the DP strategy is not ideal. This is because its energy
Lall min(Lfc , Lbat )⋅L
source life degradation rate is high and the life balance of energy sources
Fall is the total cost over the vehicle life cycle; Lall is the expected is not considered. In contrast, though the hierarchical strategy with life
lifetime driving range; mH2 is the hydrogen consumption, including the balance control sacrifices some energy economy, its expected lifetime
direct hydrogen consumption of FC and the equivalent hydrogen con­ mileage increases by 6.04 % over the optimized hierarchical control
sumption from BAT and UC; Fs is the configuration cost of the composite strategy and by 18.64 % over the DP strategy.
energy storage system; L is the mileage of the simulation condition, and
the WLTC condition is still chosen as the simulation condition here; Lfc
and Lbat are the expected life cycles of the FC and BAT. Since the FC and 4.2. Hardware-in-the-loop testing and analysis
BAT life models used in this study are approximately linear, Lfc and Lbat
are calculated from the respective life degradation rates. To verify the effectiveness of the strategy proposed in this section, a
2) Genetic algorithm optimization search results. hardware-in-the-loop (HIL) real-time simulation platform is built based

12
J. Hu et al. Energy Conversion and Management 272 (2022) 116383

Fig. 11. Hardware-in-the-loop real-time simulation platform schematic.

Fig. 12. Hardware system of HIL.

on D2P and Speedgoat real-time simulator. The structure schematic is is used for closed-loop testing of the control strategy. The variation
shown in Fig. 11. curves or values of key parameters are derived for comparison with the
The hardware device is shown in Fig. 12. The Simulink Real-Time offline simulation results, as shown in Fig. 13,14 and Table 9.
Explorer is used to control the simulation model and make it run in As can be seen from Figs. 13 and 14, the SOC and power curves of the
the Speedgoat real-time simulator. Two upper computers monitor the BAT and UC in the hardware-in-the-loop real-time environment are
parameters in the real-time simulator and the controller respectively. basically consistent with the offline simulation results. The maximum
The Speedgoat can be connected to an external monitor for real-time difference between the hardware-in-the-loop and offline simulation re­
simulation data monitoring. sults of the power battery SOC is only 1.28 %, which indicates that the
The hardware-in-the-loop test uses the same operating conditions as expected control requirement is achieved in general.
the offline simulation, and the built hardware-in-the-loop test platform From the comparison results in Table 9, it can be seen that the

13
J. Hu et al. Energy Conversion and Management 272 (2022) 116383

Fig. 13. SOC and power comparison of BAT.

5. Conclusion
Table 9
Hardware-in-the-loop and offline simulation results. In this paper, a hierarchical energy management strategy is proposed
Parameter Offline Hardware-in-the- based on the analysis of the relationship between the UC output power
simulation loop and the motor demand power of nine typical driving conditions con­
Hydrogen consumption (g) 122.7 122.7 taining urban, suburban and high-speed driving conditions under the DP
Electricity consumption (kWh) 2.632 2.727 strategy, with the energy source life taken into account in the objective
Fuel cell life degradation (%) 2.19 × 10-3 2.23 × 10-3 function. In the upper layer strategy, the GRNN is used to learn the
Battery capacity degradation (%) 4.38 × 10-3 4.51 × 10-3 relationship between the UC power and the motor demand power under
Comprehensive operating cost (yuan) 16.47 16.67
Life mileage expectancy (km) 106,180 103,080
the DP strategy to obtain the optimal output power of UC offline and
Life-cycle average operating cost 0.683 0.702 apply it in real time. In the lower level strategy, the optimal equivalent
(yuan) factors for battery charging and discharging are obtained based on
typical driving conditions, which are used as the basis for power allo­
cation between the FC and BAT using the ECMS. In order to extend the
hardware-in-the-loop test shows a decrease in the optimization effect of
service life of the composite energy storage system and get better
the energy management strategy compared to the ideal environment
economy, fuzzy control based on vehicle speed prediction and dynamic
offline simulation test. The degradation rates of FC and BAT increased by
threshold control of ultracapacitor SOC based on working condition
1.83 % and 2.97 %, respectively. The life-cycle average operating cost
recognition are designed to optimize the hierarchical control strategy.
increased by 2.78 % and the life mileage expectancy decreased by
Based on the hierarchical energy management strategy, taking into
2.92 %. The overall error was less than 3 %, which is within the
account the life balance of energy sources, the hierarchical energy
acceptable range. The main reason for the error between hardware-in-
management strategy with life balance control is proposed. The upper
the-loop test and offline simulation results is that in the real-time
and lower limits of the battery SOC compensation coefficient kbat and the
environment, there is a certain delay and error in the signal interac­
coefficient μ for balancing the battery SOC are selected as the control
tion between the controller and the real-time simulator, resulting in lag
parameters, and GA is used to find the optimal control parameters. The
and distortion in the control effect, especially at the peak operating
simulation results show that the life-cycle average operating cost is
conditions.
reduced by 3.39 % under the proposed strategy compared to the hier­
In summary, the hierarchical energy management strategy with life
archical strategy without life balance control. The proposed strategy has
balance control proposed in this section can effectively increase the
a comparable life-cycle average operating cost to the DP strategy with
expected lifetime driving range of the vehicle with better economy by
better life mileage.
balancing the service life of the FC and the BAT.

Fig. 14. SOC and power comparison of UC.

14
J. Hu et al. Energy Conversion and Management 272 (2022) 116383

A hardware-in-the-loop simulation platform based on D2P and promising versatile tool for organic catalysis and electrical capacitance. Colloids
Surf, A 2020;587:124335. https://doi.org/10.1016/j.colsurfa.2019.124335.
Speedgoat is built to experimentally validate the proposed energy
[14] Fathy A, Rezk H, Nassef AM. Robust hydrogen-consumption-minimization strategy
management strategy. The hardware-in-the-loop test results are basi­ based salp swarm algorithm for energy management of fuel cell/supercapacitor/
cally consistent with the offline simulation, and their errors are within batteries in highly fluctuated load condition. Renewable Energy 2019;139:147–60.
acceptable limits, indicating that the proposed energy management https://doi.org/10.1016/j.renene.2019.02.076.
[15] Lü X, Wang P, Meng L, Chen C. Energy optimization of logistics transport vehicle
strategy can achieve the expected control and optimization effects in the driven by fuel cell hybrid power system. Energy Convers Manage 2019;199:
real-time environment. 111887. https://doi.org/10.1016/j.enconman.2019.111887.
[16] Dao HV, To XD, Truong HVA, Do TC, Ho CM, Dang TD, et al. Optimization-Based
Fuzzy Energy Management Strategy for PEM Fuel Cell/Battery/supercapacitor
CRediT authorship contribution statement Hybrid Construction Excavator. International journal precision engineering
manufacturing-green Technology 2021;8:1267–85. https://doi.org/10.1007/
s40684-020-00262-y.
Jianjun Hu: Conceptualization, Methodology, Resources, Supervi­ [17] He H, Wang X, Chen J, Wang Y. Regenerative Fuel Cell-Battery-Supercapacitor
sion, Funding acquisition. Zhouxin Wang: Writing – review & editing, Hybrid Power System Modeling and Improved Rule-Based Energy Management for
Software, Visualization, Data curation. Hao Du: Validation, Formal Vehicle Application. J Energy Eng 2020;146(6):04020060. https://doi.org/
10.1061/(ASCE)EY.1943-7897.0000708.
analysis, Investigation. Lingbo Zou: Writing – original draft.
[18] Yuan X, Yan G, Li H, Liu X, Su C, Wang Y. Research on energy management
strategy of fuel cell–battery–supercapacitor passenger vehicle [C]. 2021 8th
Declaration of Competing Interest International Conference on Power and Energy Systems Engineering. Energy
Reports, 2022. https://doi.org/10.1016/j.egyr.2021.11.244.
[19] Tao F, Zhu L, Fu Z, Si P, Sun L. Frequency decoupling-based energy management
The authors declare that they have no known competing financial strategy for fuel cell/battery/ultracapacitor hybrid vehicle using fuzzy control
interests or personal relationships that could have appeared to influence method. IEEE Access 2020;8:166491–502. https://doi.org/10.1109/
ACCESS.2020.3023470.
the work reported in this paper. [20] Ahmadi S, Bathaee SMT, Hosseinpour AH. Improving fuel economy and
performance of a fuel-cell hybrid electric vehicle (fuel-cell, battery, and ultra-
Data availability capacitor) using optimized energy management strategy. Energy Convers Manage
2018;160:74–84. https://doi.org/10.1016/j.enconman.2018.01.020.
[21] Kaya K, Hames Y. Two new control strategies: For hydrogen fuel saving and extend
Data will be made available on request. the life cycle in the hydrogen fuel cell vehicles. Int J Hydrogen Energy 2019;44
(34):18967–80. https://doi.org/10.1016/j.ijhydene.2018.12.111.
[22] Liu Y, Liang J, Song J, Ye J. Research on Energy Management Strategy of Fuel Cell
Acknowledgements Vehicle Based on Multi-Dimensional Dynamic Programming. Energies 2022;15
(14):5190. https://doi.org/10.3390/en15145190.
This work was funded by the National Key Research and Develop­ [23] Tao S, Chen W, Gan R, Li L, Zhang G, Han Y, et al. Energy management strategy
based on dynamic programming with durability extension for fuel cell hybrid
ment Program of China under Grant No. 2018YFB0105402. tramway. Railway Engineering Science 2021;29(3):299–313. https://doi.org/
10.1007/S40534-021-00247-W.
References [24] Hou S, Gao J, Zhang Y, Chen M, Shi J, Chen H. A comparison study of battery size
optimization and an energy management strategy for FCHEVs based on dynamic
programming and convex programming. Int J Hydrogen Energy 2020;45(41):
[1] Zhang J, Cheng M, Luo X, Li H, Luo L, Cheng X, et al. Current status of the research
21858–72. https://doi.org/10.1016/j.ijhydene.2020.05.248.
on key technologies of vehicle fuel cell stack. J Automotive Safety Energy 2022;13
[25] Ali AM, Ganbar A, Söffker D. Optimal Control of Multi-Source Electric Vehicles in
(1):1–28. https://doi.org/10.3969/j.issn.1674-8484.2022.01.001.
Real Time Using Advisory Dynamic Programming. IEEE Trans Veh Technol 2019;
[2] Wang Z, Mao J, He Z, Liang F. Fuzzy Control Based on IQPSO in Proton-Exchange
68(11):10394–405. https://doi.org/10.1109/TVT.2019.2941523.
Membrane Fuel-Cell Temperature System. J Energy Eng 2020;146(5):04020044.
[26] Li Z. Research on energy management strategy of fuel cell vehicle based on multi-
https://doi.org/10.1061/(ASCE)EY.1943-7897.0000691.
dimensional DP algorithm [D]. Guangdong University of Technology; 2021.
[3] Rezk H, Nassef AM, Abdelkareem MA, Alami AH, Fathy A. Comparison among
https://doi.org/10.27029/d.cnki.ggdgu.2021.001785.
various energy management strategies for reducing hydrogen consumption in a
[27] Sellali M, Ravey A, Betka A, Kouzou A, Benbouzid M, Djerdir A, et al. Multi-
hybrid fuel cell/supercapacitor/battery system. Int J Hydrogen Energy 2021;46(8):
objective optimization-based health-conscious predictive energy management
6110–26. https://doi.org/10.1016/j.ijhydene.2019.11.195.
strategy for fuel cell hybrid electric vehicles. Energies 2022;15(4):1318. https://
[4] Zhang M, Chen J, Lan X, Xiao C. Simulation on staggered parallel boost converter
doi.org/10.3390/en15041318.
with double integral sliding mode control. J Phys Conf Ser 2021;1983(1):012076.
[28] Chen H, Xiong R, Lin C, Shen W. Model predictive control based real-time energy
https://doi.org/10.1088/1742-6596/1983/1/012076.
management for hybrid energy storage system. CSEE J Power Energy Syst 2021;7
[5] Truong HVA, Dao HV, Do TC, Ho CM, To XD, Dang TD, et al. Mapping Fuzzy
(4):862–72. https://doi.org/10.17775/CSEEJPES.2020.02180.
Energy Management Strategy for PEM Fuel Cell–Battery–Supercapacitor Hybrid
[29] Mounica V, Obulesu YP, Chen M. Hybrid power management strategy with fuel
Excavator. Energies 2020;13(13):3387. https://doi.org/10.3390/en13133387.
cell, battery, and supercapacitor for fuel economy in hybrid electric vehicle
[6] Lü X, Qu Y, Wang Y, Qin C, Liu G. A comprehensive review on hybrid power system
application. Energies 2022;15(12):4185. https://doi.org/10.3390/en15124185.
for PEMFC-HEV: Issues and strategies. Energy Convers Manage 2018;171:1273–91.
[30] Li H, Ravey A, N’Diaye A, Djerdir A. A novel equivalent consumption minimization
https://doi.org/10.1016/j.enconman.2018.06.065.
strategy for hybrid electric vehicle powered by fuel cell, battery and
[7] Badji A, Abdeslam DO, Becherif M, Eltoumi F, Benamrouche N. Analyze and
supercapacitor. J Power Sources 2018;395:262–70. https://doi.org/10.1016/j.
evaluate of energy management system for fuel cell electric vehicle based on
jpowsour.2018.05.078.
frequency splitting. Math Comput Simul 2020;167:65–77. https://doi.org/
[31] Li H, Ravey A, N’diaye A, Djerdir A. Online adaptive equivalent consumption
10.1016/j. matcom.2019.02.014.
minimization strategy for fuel cell hybrid electric vehicle considering power
[8] Zhou D, Al-Durra A, Gao F, Ravey A, Matraji I, Simões MG. Online energy
sources degradation. Energy Convers Manage 2019;192:133–49. https://doi.org/
management strategy of fuel cell hybrid electric vehicles based on data fusion
10.1016/j. enconman.2019.03.090.
approach. J Power Sources 2017;366:278–91. https://doi.org/10.1016/j.
[32] Hu X, Zou C, Tang X, Liu T, Hu L. Cost-Optimal Energy Management of Hybrid
jpowsour.2017.08.107.
Electric Vehicles Using Fuel Cell/Battery Health-Aware Predictive Control. IEEE
[9] Swarnkar A, Maherchandani JK. Performance Analysis of Hybrid Fuel Cell/
Trans Power Electron 2019;35(1):382–92. https://doi.org/10.1109/
Battery/Supercapacitor Electric Vehicle for Different Battery State of Charge Levels
TPEL.2019.2915675.
[C]. In: 2018 International Conference on Recent Innovations in Electrical,
[33] Jiang H, Xu L, Li J, Hu Z, Ouyang M. Energy Management and Component Sizing
Electronics & Communication Engineering (ICRIEECE). IEEE; 2020. https://doi.
for a Fuel Cell/Battery/Supercapacitor Hybrid Powertrain based on Two-
org/10.1109/ICRIEECE44171.2018.9008909.
Dimensional Optimization Algorithms. Energy 2019;177:386–96. https://doi.org/
[10] Thounthong P, Rael S. The benefits of hybridization. IEEE Ind Electron Mag 2009;3
10.1016/j.energy.2019.04.110.
(3):25–37. https://doi.org/10.1109/mie.2009.933885.
[34] Tao F, Zhu L, Ji B, Si P, Fu Z, Li X. Energy Management Strategy Using Equivalent
[11] Li C, Liu G. Transient behavior comparison of fuel cell only and fuel cell hybrid
Consumption Minimization Strategy for Hybrid Electric Vehicles. Security and
powertrain [C]. UKACC International Conference on Control 2010. IET, 2013.
Communication Networks 2020;2020:1–8. https://doi.org/10.1155/2020/
https://doi.org/10.1049/ic.2010.0354.
6642304.
[12] Eivazzadeh-Keihan R, Taheri-Ledari R, Mehrabad MS, Dalvand S, Sohrabi H,
[35] Fu Z, Zhu L, Tao F, Si P, Sun L. Optimization based energy management strategy for
Maleki A, et al. Effective Combination of rGO and CuO Nanomaterials through Poly
fuel cell/battery/ ultracapacitor hybrid vehicle considering fuel economy and fuel
(p-phenylenediamine) Texture: Utilizing It as an Excellent Supercapacitor. Energy
cell lifespan. Int J Hydrogen Energy 2020;45(15):8875–86. https://doi.org/
And Fuels 2021;35(13):10869–77. https://doi.org/10.1021/acs.energyfuels.
10.1016/j.ijhydene.2020.01.017.
1c01132.
[13] Eivazzadeh-Keihan R, Taheri-Ledari R, Khosropour N, Dalvand S, Maleki A,
Mousavi-Khoshdel SM, et al. Fe3O4/GO@ melamine-ZnO nanocomposite: A

15
J. Hu et al. Energy Conversion and Management 272 (2022) 116383

[36] Pei P, Chang Q, Tang T. A quick evaluating method for automotive fuel cell [40] Dombi GW, Nandi P, Saxe JM, Ledgerwood AM, Lucas CE. Predication of Rib
lifetime. Int J Hydrogen Energy 2008;33(14):3829–36. https://doi.org/10.1016/j. fracture injury outcome by an artificial neural network. The Journal of Trauma
ijhydene.2008.04.048. 1995;39(5):915–21. https://doi.org/10.1097/00005373-199511000-00016.
[37] Fletcher T, Thring R, Watkinson M. An energy management strategy to [41] Wu Q, Wang Yu, Wang Y, Wang J, Lan L, Deng Y, et al. Ablation state assessment of
concurrently optimise fuel consumption & PEM fuel cell lifetime in a hybrid SF 6 circuit breaker contacts based on BP neural network and mean impact value
vehicle. Int J Hydrogen Energy 2016;41(46):21503–15. https://doi.org/10.1016/j. [C]. Energy Rep 2022;8:874–83. https://doi.org/10.1016/j.egyr.2022.02.237.
ijhydene. 2016.08.157. [42] Shi Y, Gao F, Zhang G, Qiang G, Gao Y. A hybrid energy optimization control for
[38] Chen H, Pei P, Song M. Lifetime prediction and the economic lifetime of Proton urban rail trains. 155–9, 164 Railway Standard Design 2019;063(011). https://doi.
Exchange Membrane fuel cells. Appl Energy 2015;142:154–63. https://doi.org/ org/10.13238/j.issn.1004-2954.201901080005.
10.1016/j. apenergy.2014.12.062. [43] Zu G. Study on energy management strategy of dual motor driving pure electric
[39] Luo Y, Wang F, Yu H, Liu X. A study on the driving-cycle-based life model for vehicle based on driving cycle identification. Chongqing University, 2017.
LiFePO4 battery. Automotive Engineering 2015;37(8):881–5. https://doi.org/ Available from https://d.wanfangdata.com.cn/thesis/D01397114. 2022.
10.3969/j.issn. 1000-680X, 2015.08.004.

16