Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

292 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 14, NO.

1, JANUARY 2018

Adaptive Fuzzy Logic Control of


Fuel-Cell-Battery Hybrid Systems for
Electric Vehicles
Jian Chen , Senior Member, IEEE, Chenfeng Xu, Chengshuai Wu, and Weihua Xu

Abstract—In this paper, we propose an adaptive control creased complexity and brings about challenges to the synthesis
approach with fuzzy logic parameter tuning (AFLPT) for the of power management systems. In [7], several hybridization
energy management of electric vehicles that are using fuel topologies for FC vehicles have been categorized: FC-battery,
cell battery hybrid systems. The controller is adaptive to
different driving conditions including normal, regenerative, FC-supercapacitor, and FC-battery-supercapacitor. This paper
and overload conditions. Specifically, the power flow studies the FC-battery electric vehicle power systems, in which
between the fuel cell (FC) and the Li-ion battery is con- the FC functions as the main power source and the Li-ion battery
trolled in real time to maintain the battery state of charge functions as the secondary power source. Compared with other
(SOC) at a desirable level while satisfying the FC dynamic
rechargeable batteries, Li-ion battery has the advantages of
constraints. For guaranteeing performance in different
driving conditions, the FLPT is integrated with the adaptive high capacity, a large number of charge–discharge cycles and
controller. Moreover, theoretical properties of the designed reasonable cost [8]. This system is widely utilized in practical
controller are analyzed. Simulation and experiment results FC-battery electric vehicles [9], [10]. In a FC-battery hybrid
illustrate the effectiveness of the proposed strategy for power system, the FC is connected to the dc bus by a unidirec-
FC-battery hybrid systems in electric vehicles. tional dc–dc converter and the Li-ion battery is connected to
Index Terms—Adaptive control, electric vehicles, fuel cell the dc bus directly. Its energy consumption is less than that of a
battery hybrid system, fuzzy logic control, power manage-
ment.
FC-supercapacitor system [7] and the powertrain cost is much
lower than that of a FC-battery-supercapacitor system under
I. INTRODUCTION the same conditions [11]. The advantages discussed above are
the main reasons why FC-battery systems are widely utilized
NCREASINGLY serious energy crisis and greenhouse ef-
I fects motivate the utilization of renewable energy instead of
fossil fuels for cleaner, more reliable, and sustainable systems
in electric vehicles. Compared with FC-battery systems that
have two dc–dc converters, the power system with one dc–dc
converter studied in this paper supplies the load power more
[1], [2]. As one of the most promising clean energy sources, fuel
efficiently especially when the load fluctuates violently [12].
cells (FCs) have the advantages of relatively high efficiency and
Several control strategies have been proposed for the power
almost zero emissions, especially in the applications to hybrid
management of hybrid power systems. The goal is to coordinate
electric vehicles [3], [4].
the power flow between the FC and the auxiliary energy
Nevertheless, the slow dynamic response is a well-known
sources. In [13] and [14], intelligent control strategies based on
technical limitation of FCs. Therefore, FCs are usually utilized
artificial neural networks are utilized in FC hybrid vehicles. In
together with other energy sources, e.g., Li-ion batteries or
[15], a cascade control strategy for a FC-battery-supercapacitor
supercapacitors, to supply load power to electric vehicles in
system is developed to maintain both the voltage of the dc bus
a hybrid fashion. Those auxiliary energy sources can then
and state of charge (SOC) of the battery within expected ranges.
compensate the transient output power, improve the peak power
In [16], a combination of cascade control and lookup tables
capability, and absorb the regenerative energy during braking
is proposed to regulate the FC output power according to the
[5], [6]. This kind of hybridization, however, causes an in-
battery SOC in a FC-battery vehicle. Most of the work on power
sharing strategies is based on deterministic rules, as utilized
Manuscript received July 25, 2016; revised September 20, 2016; ac- in [17] and [18]. Generally, these rules depend on the analysis
cepted October 9, 2016. Date of publication October 19, 2016; date of of hybrid powertrain characteristics and the utilization of rich
current version January 3, 2018. This work was supported in part by the
National Natural Science Foundation of China under Grant 61433013 human experiences. In [19] and [20], model predictive control
and Grant 61304012, and in part by the Chinese Recruitment Pro- is implemented for the power distribution of FC vehicles.
gram of Global Youth Experts. Paper no. TII-16-0744.R1. (Corresponding However, most of the controllers involve high computational
author: J. Chen.)
All authors are with the State Key Laboratory of Industrial Control Tech- complexity which is not desirable for the real-time control of
nology College of Control Science and Engineering, Zhejiang University, the power systems in electric vehicles.
Hangzhou 310027, China (e-mail: jchen@zju.edu.cn). Fuzzy logic control has also been used to distribute the power
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. output between the FC and the auxiliary energy sources under
Digital Object Identifier 10.1109/TII.2016.2618886 different driving conditions, see, e.g., [21]–[23], owing to its
1551-3203 © 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications standards/publications/rights/index.html for more information.
CHEN et al.: ADAPTIVE FUZZY LOGIC CONTROL OF FUEL-CELL-BATTERY HYBRID SYSTEMS FOR ELECTRIC VEHICLES 293

and exchanged current are represented as Td , Rohm , and i0 ,


respectively. Note that similar to [25], this model is not utilized
in the subsequent controller design but only for simulations.

B. Battery Model
The dynamics of the Li-ion battery are described by an equiv-
alent internal resistance model [26], see Fig. 1, as
d Vc (t)
C(Vsoc ) Vc (t) = iB − , (3)
dt R2 (Vsoc )
VB (t) = Voc (t) − Vc (t) − iB R1 (Vsoc ) (4)
where R1 (Vsoc ) and R2 (Vsoc ) are the battery equivalent internal
Fig. 1. System structure. resistances, Vc (t) is the voltage of the capacitor C, VB is the
respective voltage of the system dc bus, and iB (t) is the output
effectiveness but simplicity in decision making for complex sys- current of the battery. The range 0–1 V of Vsoc corresponds to
tems. In [21] and [22], the fuzzy logic control is the only system range 0–100% of the Li-ion battery SOC. Lastly, Voc (t) is the
controller and generates the FC output power directly. In [23], battery open-circuit voltage with a nonlinear mapping to the
three driving conditions are categorized by the load power and SOC of the battery.
the control strategy can achieve a good performance, efficiency,
and durability. However, the changing rates of system key pa- C. Vehicle Load Model
rameters are usually not considered in these rule base design.
In this paper, an adaptive controller with the fuzzy logic As shown in Fig. 1, an ideal controlled current source that is
parameter tuning (FLPT) is proposed for the FC-battery hybrid connected to the system dc bus is utilized to express the power
power system in electric vehicles. The proposed adaptive con- consumed by the load, which has also been done in [19]. The
troller achieves real-time control and effective power allocation. vehicle load model is then expressed as
In particular, the power flow between the FC and the Li-ion d
battery is controlled to maintain the battery SOC at a desired L il (t) = −il (t)Rp + ip (t)Rp + VB (5)
dt
level. Additionally, to guarantee the control performance, the
where il (t) is the load current, Rp is the internal resistance of
FLPT is integrated with the adaptive controller considering
the ideal current source, and variable current ip (t) is unknown
different driving conditions and the characteristics of the hybrid
and varies with the load power. In order to describe the dynamic
power system.
behavior of the load more accurately, inductance L is used to
This paper is organized in the following sections. Section II
represent the imperfections of the load.
describes the model of the FC-battery electric vehicle power sys-
tem, which is the base for the controller design. In Section III,
the power management strategy is developed in detail, including D. DC–DC Converter Model
the adaptive controller, stability analysis, and the FLPT. Simu- The FC current and voltage vary with the demanded load
lations and experiments are discussed to validate the proposed power; thus, a unidirectional dc–dc converter that connects the
control strategy in Section IV. Finally, a brief conclusion is FC with the dc bus is necessary to ensure dynamic power out-
presented in Section V. puts. In this work, the unidirectional dc–dc converter is repre-
sented by its mean value equivalent model [27] as
II. MODEL DEVELOPMENT d
Ldc ifc (t) = −(1 − d(t))VB + Vfc (t) (6)
The topology of the hybrid system powered by an FC and a dt
Li-ion battery is presented in Fig. 1.
where Ldc is the inductance of the converter, d(t) is the duty
ratio of the converter, and ifc (t) is the current of the FC.
A. Fuel Cell Model
The FC block is implemented as a generic model to represent E. State-Space Model
the most popular types of FC stacks that are fed with hydrogen
In the dc–dc converter, a high-gain PID controller is adopted
and air [24]. The model can be expressed as follows:
to track the FC reference current i∗fc (t) which is controlled by the
 
ifc 1 adaptive controller. The steady state, i.e., ifc (t) = i∗fc (t), can be
E = Eoc − N A ln , (1) achieved after the transient time of the converter which is much
i0 sT3 d + 1
faster than the control cycle of the designed power management
Vfc = E − Rohm · ifc (2) strategy. Hence, the relationship between battery current iB (t)
where Vfc and ifc are the FC output voltage and current, re- and FC current i∗fc (t) can be acquired as follows:
spectively. Notation Eoc is the open-circuit voltage and N is Vfc ∗
the number of cells. The response time, internal resistance, iB = il − i . (7)
VB fc
294 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 14, NO. 1, JANUARY 2018

By substituting (4) and (7) into (3) and (5), the system state- to guarantee the dynamic performance of the system as
space model can be expressed as
î˙ p = β1 R̂p (il − i∗l ) + p1 , (14)
ẋ = A(x)x + G(x, u) (8)
˙
R̂1 = −β2 μ(VB − Voc∗ )(iB + il − i∗l ) + p2 , (15)
where the state vector x and control input u are defined as x(t)  ⎧
[ Vc il ]T ∈ R2 and u(t)  i∗fc (t), respectively, and A(x) and ⎨ −g1 , if îp ≥ ip,m ax and g1 > 0
p1 (t) = or îp ≤ ip,m in and g1 < 0 (16)
G(x, u) are defined as ⎩
0, otherwise
⎡ 1 1 ⎤ ⎧
− − ⎨ −g2 , if R̂1 ≥ R1,m ax and g2 > 0
⎢ CR2 C ⎥ p2 (t) = or R̂1 ≤ R1,m in and g2 < 0
A(x)  ⎣ ⎦ (9) ⎩
(17)
1 Rp + R1
− 0, otherwise
L L
⎡ ⎤
Vfc ∗ where g1 = β1 R̂p (il − i∗l ) and g2 = −β2 μ(VB − Voc∗ )(iB +
⎢ − i ⎥
CVB fc il − i∗l ) with β1 and β2 both being positive constant gains. From
G(x, u)  ⎢
⎣ R1 Vfc
⎥. (10)
∗ Voc Rp ⎦ (14) to (17), it is clear that îp (t) and R̂1 (t) are both bounded.
i + +
LVB fc L L The state errors of the system (8) are defined as

x̃(t)  x(t) − x∗ (t) = [ Ṽc ĩl ]T ∈ R2 (18)


III. CONTROL DEVELOPMENT
A. Adaptive Control Design where x∗ (t)  [ Vc∗ i∗l (t) ]T is the reference state vector, and
Vc∗ = 0. Based on the adaptive update laws for i∗l (t), îp (t), and
In (8), the FC current is the only control input. The control
objective is to regulate the FC current to provide the main load R̂1 (t) in (8)–(15), the closed-loop error system is derived as
power and maintain the SOC of the battery at a desired level. follows:
The control input u(t) is designed based on the model in (8) as
x̃˙ = Ae (x)x̃ + Ge (x, u) (19)
follows:

VB where Ae (x) = A(x), and


u(t) = i∗fc = [h(t) − μ(VB − Voc∗ )] (11)
Vfc Ge =
where Voc∗ 
is the expected open-circuit voltage of the battery, − Cμ (VB − Voc∗ ) + T
C ḣ
+
μ ∈ R is a positive bounded control gain that can be designed V oc −V oc∗ +R p ĩ p
.
L̂ −L ∗
by the FLPT in the subsequent section. The second term in the L (VB
μ
− Voc∗ )(2R̂1 − R1 ) − T R 1 ḣ
L + L + L il
square bracket is utilized to protect the battery from overcharge
and overdischarge. In (11), h(t) is determined by the following B. Stability Analysis
differential equation:
In order to prove the stability of the closed-loop error system
(19), a candidate Lyapunov function V (t) is chosen as follows:
h + T ḣ = i∗l (12)
1 1 ĩ2p R̃2
where T is a positive time constant, and i∗l (t) is designed to V (t) = C Ṽc2 + Lĩ2l + + 1 (20)
track the transient load current, which is determined by 2 2 2β1 2β2
where C and L are positive and bounded according to the bat-
L̂i˙∗l = Voc∗ − (i∗l − îp )R̂p − 2μR̂1 (VB − Voc∗ ) (13) tery and load models. Therefore, V (t) is guaranteed to be non-
negative.
where L̂ and R̂1 (t) are the estimates of the inductance L in the Then, the derivative of V (t) can be obtained with (19) as
load and the battery inner resistance R1 , respectively. In (13), follows:
R̂p and îp (t) are defined as the estimates of the inner resistance  
and the current of the ideal current source shown in Fig. 1, 1 1 ∂C ĩp
V̇ = − (R1 + Rp )ĩl −
2
− V̇soc Ṽc2 + i̇p
respectively. To satisfy the system dynamic constraints, L̂ is R2 2 ∂Vsoc β1
configured as a positive constant much larger than Ldc . Lastly,
+ μ(VB − Voc∗ )(2R̂1 ĩl − R1 ĩl + Ṽc ) + (VB − Voc∗ )ĩl
R̂p is configured as a positive parameter.
Obviously, ip (t) is bounded due to the limits of the load, i.e., R̃1
ip,m in ≤ ip ≤ ip,m ax . Similarly, due to the limits of the battery + μR̃1 (VB − Voc∗ )(iB + ĩl ) − T R1 ḣĩl + Ṙ1
β2
internal resistance, R1 (t) is bounded with the upper bound
R1,m ax and the lower bound R1,m in . Inspired by [28], two adap- ĩp R̃1
+ (L̂ − L)ĩl i̇∗l + T ḣV̂c − p1 − p2 . (21)
tive update laws for îp (t) and R̂1 (t), respectively, are developed β1 β2
CHEN et al.: ADAPTIVE FUZZY LOGIC CONTROL OF FUEL-CELL-BATTERY HYBRID SYSTEMS FOR ELECTRIC VEHICLES 295

According to (3), (4), (7), (11), and (12), the following equation where 0 < θ < 1, and
can be obtained:
−λm in (1 − θ)||x̃||2 + |S3 Ṙ1 | + |S4 i̇p | + |S5 | < 0,
−Ṽc − R1 ĩl Voc − Voc∗ − R̃1 T ḣ 
VB − Voc∗ = + . (22) |S3 Ṙ1 | + |S4 i̇p | + |S5 |
1 + μR1 1 + μR1 ∀||x̃|| > . (27)
λm in (1 − θ)
By substituting (22) into (21), the derivative of V (t) can be Therefore, the following expression always holds:
rewritten as follows:
V̇ < 0 ∀||x̃|| > γ (28)
V̇ = − x̃ Qx̃ + S1 Ṽc + S2 ĩl + S3 R̃1 + S4 ĩp + S5
T
where
ĩp μ(1 + μR̂1 ) R̃1 γ=
− p1 − (Voc − Voc∗ − T R1 ḣ)2 − p2
β1 (1 + μR1 )2 β2 ⎛ ⎛ ⎞⎞
(23) |S3 Ṙ1 | + |S4 i̇p | + |S5 | 2 · max(|S1 |, |S2 |)⎠⎠
sup⎝max⎝ , .
λm in (1 − θ) μλm in
where
⎡ ⎤ Based on (28) and the selected Lyapunov function V (t) in (20),
1 1 ∂C 
⎢ R2 − V̇ soc 0 ⎥ μ(1 + μR̂1 ) 1 R1 x̃(t) will eventually converge to the domain {x̃| ||x|| < γ}.
Q=⎣ 2 ∂V soc ⎦+ ,
(1 + μR1 )2 R1 R12 Therefore, the closed-loop error system (19) with adaptive con-
0 R1 + Rp troller (11)–(15) is stable and the control objectives are satisfied.
2μ(1 + μR̂1 ) R̃1
Remark 1: According to the stability analysis, in the steady
S1 = (Voc − Voc∗ − T R1 ḣ) + T ḣ, S3 = , state, the state-space variable Vc (t) will eventually converge to
(1 + μR1 )2 β2
 zero, which means that the open-circuit voltage Voc (t) remains
2μ(1 + μR̂1 )R1 unchanged due to the dynamical circuit relations. Therefore, the
S2 = + 1 (Voc − Voc∗ − T R1 ḣ)
(1 + μR1 )2 battery current iB (t) will eventually converge to zero and the
battery SOC will converge to a constant. In this way, the battery
+ (L̂ − L)i̇∗l , can be protected from overcharge and overdischarge.
ĩp
S4 = , S5 = μ(VB − Voc∗ )(Voc − Voc∗ − R̂1 T ḣ). C. Fuzzy Logic for Different Driving Conditions
β1
In the adaptive controller, the values of μ and Voc∗ have a
Considering the relatively slow dynamics of the battery SOC, significant influence on the control performance. If they are
the following assumption is made. set as constants under different driving conditions, the control
Assumption 1: Under all the operating conditions of the bat- performance could be degraded. Thus, an FLPT is developed
tery, the following relationship between R2 (Vsoc ) and C(Vsoc ) to improve the adaptive control performance under different
is always satisfied: driving conditions.
  The input variables of the FLPT are chosen as
1  1 ∂C 
>  V̇soc  . (24) 1) the estimation of load current, i.e., i∗l ;
R2 2 ∂Vsoc 2) the absolute value of the derivative of i∗l , i.e., |i̇∗l |;
3) the deviation between the battery voltage and the ex-
Since μ is positive and bounded, the matrix Q is symmetric
pected battery voltage Vref , i.e., ΔVB , which is obtained
positive definite if Assumption 1 is satisfied. Based on (14)–(17),
from
the following inequalities always hold:
ΔVB  VB − Vref . (29)
ĩp (t)p1 (t) ≥ 0, R̃1 (t)p2 (t) ≥ 0. (25)
Instead of using fuzzy logic controllers to generate output
powers of the FC and the storage devices as has been done in
Let λm in denote the minimum eigenvalue of the matrix Q,
the literature [21] and [22], the outputs of the FLPT in this paper
then by the Rayleigh Ritz inequality [29], the inequalities can
are the parameters μ and ΔVfl which are obtained as
be obtained as follows:
ΔVfl  Vref − Voc∗ . (30)
V̇ ≤ − x̃ Qx̃ + S1 Ṽc + S2 ĩl + S3 R̃1 + S4 ĩp + S5
T
Remark 2: Vref and the parameter Voc∗
defined in (11) are
≤ − λm in (1 − θ)||x̃||2 + |S3 Ṙ1 | + |S4 i̇p | + |S5 | different concepts, and Voc∗ is equal to Vref if and only if the
adaptive controller is used alone. With the introduction of the
+ (2 · max(|S1 |, |S2 |) − λm in θ||x̃||)||x̃|| FLPT, Voc∗ varies with load shifting in different control cycles.
≤ − λm in (1 − θ)||x̃||2 + |S3 Ṙ1 | + |S4 i̇p | + |S5 |, However, Vref is the expected battery open-circuit voltage and
usually set as a constant in the whole vehicle runtime.
1 To guarantee the effectiveness and reliability of controller, the
∀||x̃|| > 2 · max(|S1 |, |S2 |) (26)
μλm in rules of the FLPT are specified based on the dynamic response
296 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 14, NO. 1, JANUARY 2018

TABLE I
RULE BASE FOR THE NORMAL DRIVING CONDITION

TABLE II
RULE BASE FOR THE REGENERATIVE BRAKING CONDITION

Fig. 2. Membership functions of input and output variables. (a) Input


variable i∗l , (b) input variable ΔV B , (c) input variable |i̇∗l |, (d) output
variable μ, and (e) output variable ΔV fl .

TABLE III
constraints of the FC and the Li-ion battery. The membership RULE BASE FOR THE OVERLOAD DRIVING CONDITION
functions (MFs) of the input and output variables are depicted
in Fig. 2. The MFs for the input variable i∗l are designed as 1)
small (S); 2) middle (M); and 3) big (B); small, middle, and big
correspond to the regenerative braking driving condition, normal
driving condition, and overload driving condition, respectively,
in the driving cycle.
The MFs for input |i̇∗l | are low (L), middle (M), and high
(H), which represent the load-shifting speed. The MFs for input
ΔVB are negative big (NB), negative middle (NM), negative
small (NS), zero (Z), positive small (PS), positive middle (PM),
and positive big (PB), which represent the deviation degree the battery works in the charging mode and the output
between the current SOC and the expected SOC of the battery. power of the FC would be decreased depending on the
Similarly, the MFs of the output variable ΔVfl are NB, NM, NS, battery SOC level. At this moment, in order to absorb the
Z, PS, PM, and PB. The MFs for μ are big low (BL), middle regenerative braking energy as much as possible, μ and
low (ML), small low (SL), middle (M), small big (SB), middle |ΔVfl | reduce properly, as shown in Table II.
big (MB), and big big (BB). 3) Overload operating condition: Whenever i∗l is big (B), it
Remark 3: In the adaptive controller, if the range of μ is implies that the load power is greater than the FC-rated
from 0 to 20, the bigger the μ, the faster the output response of power and the battery works in the discharge mode. At
the FC, which means that the load power demand can be better this moment, the FC output power should be regulated
satisfied. With regard to ΔVfl , the battery SOC will converge to more smoothly. Thus, ΔVfl can almost remain unchanged
its expected level with smaller errors when |ΔVfl | increases but and μ should be decreased properly to achieve a smoother
still bounded. FC response, as shown in Table III.
The rules of the synthesis process of the FLPT during different Remark 4: The FC-rated output voltage and power will de-
driving conditions are subject to the principles listed as follows: crease after long periods of driving. Due to the degradation
1) Normal driving condition: Whenever i∗l is middle (M), mechanism, the FC transient performance could be degraded
which implies that the load power is positive and smaller and the system dynamic constraints could also be changed [30].
than the FC-rated power, the FC supplies the main load Under this condition, the rule base in Tables I–III should be
demand and the battery is regulated to its own reference adjusted correspondingly as follows. Under the overload driv-
voltage. At this moment, μ and ΔVfl are sensitive to the ing condition, to protect the FC from the fuel starvation, the
load-shifting speed and the deviation between the current values of μ and ΔVfl in Table III should be decreased accord-
and reference battery voltages, as shown in Table I. ingly due to the degraded FC-rated power; Under the normal
2) Regenerative braking driving condition: Whenever i∗l is driving condition and the regenerative driving condition, due to
small (S), which implies that the load power is negative, the transient performance degradation, the values of μ and ΔVfl
CHEN et al.: ADAPTIVE FUZZY LOGIC CONTROL OF FUEL-CELL-BATTERY HYBRID SYSTEMS FOR ELECTRIC VEHICLES 297

Fig. 5. Comparison of FC output in different strategies.


Fig. 3. Load power.

Fig. 6. Comparison of open-circuit voltage of battery.


Fig. 4. FC and battery output power.

in Tables I and II should be reduced properly, especially when


|i̇∗l | has values of M and H.
Remark 5: The FLPT inputs consider |i̇∗l | (the changing rate
of the load current estimation). Under the normal driving condi-
tion, the FLPT outputs are sensitive to the load current changing Fig. 7. FLPT output. (a) Parameter μ and (b) parameter ΔV fl .
rate and should be increased properly when the MF of |i̇∗l | is
H. Under the overload driving condition, the FC output should
be regulated more smoothly. Therefore, when |i̇∗l | has values presents the load power demand in the simulations. The load
of M or H, the FLPT outputs should be decreased accordingly inductor L and the resistor Rp are 0.01 mH and 10 Ω. The other
compared with the outputs under the normal driving condition. parameters in the control strategy are determined by multiple
Under the regenerative driving condition, all the regenerative repeated simulation experiments and are selected as follows:
energy is absorbed by the battery. Therefore, the FLPT outputs
should not be sensitive to the load current changing rate but to β1 = 0.015, β2 = 0.5, T = 10,
the battery SOC even if |i̇∗l | has values of M or H. Note that |i̇∗l | is
L̂ = 0.98 mH, R̂p = 10.3 Ω, Vref = 48 V.
already calculated in (13) and only utilized in the FLPT lookup
tables; therefore, it will almost not increase the computational The effectiveness of the proposed adaptive controller is illus-
complexity for the real-time control. trated in Fig. 4 with the parameters μ = 7 and Voc∗ = Vref = 48 V.
It is clear that the FC provides the main load power demand with
IV. SIMULATION AND EXPERIMENTAL RESULTS a slow dynamic and the Li-ion battery provides the transient
load power demand. Comparisons are made for the adaptive
A. Simulation Results controller with and without the FLPT in terms of the FC output
To evaluate the effectiveness of the proposed control strategy power and the battery open-circuit voltage, which are depicted
and the rationality of considering the load current estimation in Figs. 5 and 6, respectively. From Fig. 5, it is clear that, with the
changing rate, two groups of comparative simulations are car- FLPT, the FC can track the load power more effectively under the
ried out: the comparison of the adaptive controller with and same constraints of transient current, especially when the load
without the FLPT, and the comparison of using the FLPT with fluctuates violently. From Fig. 6, it is clear that, with the FLPT,
and without the input |i̇∗l |. The FC-rated current and voltage in the Li-ion battery open-circuit voltage fluctuates more closely
the simulation are 50 A and 39 V. The Li-ion battery capacity to the reference voltage Vref on the whole, which means the cor-
and rated voltage are 50 Ah and 48 V. The FC current changing responding battery SOC is closer to the expected level. Besides,
rate is limited to less than 8 A/s, which corresponds to the con- Figs. 5 and 6 show the comparison results with and without |i̇∗l |
d ∗
troller constraint −8 < dt il (t) < 8. The expected SOC level in the FLPT inputs. From the zoom-ins in Fig. 5, we can observe
of the Li-ion battery is 55–65%, which corresponds to the ex- that, with |i̇∗l | in the FLPT inputs, the FC output approximates
pected open-circuit battery voltage Vref range 48 ± 0.4 V. Fig. 3 the load demand better under the same system constraints, which
298 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 14, NO. 1, JANUARY 2018

Fig. 8. Test vehicle. Fig. 12. Power of the battery and FC.

Fig. 9. Driving route.

Fig. 13. Estimation of R 1 .

Fig. 10. Relationship between V oc and SOC(V soc ).

Fig. 14. Comparison of FC output with load power.

Fig. 15. FLPT output in the experiment. (a) Parameter μ and (b) pa-
Fig. 11. Experimental load power. rameter ΔV fl .

tery in the vehicle are, respectively, 100 Ah and 50 V. The current


demonstrates the effectiveness of considering the load current
changing rate of the FC is limited to less than 10 A/s. Owing to
estimation changing rate. Furthermore, Fig. 6 shows that, with
the immeasurability of the battery SOC, the internal mapping be-
|i̇∗l | in the FLPT, the battery open-circuit voltage deviates less
tween the SOC and the open-circuit voltage (Voc ) of the battery
from the reference voltage (Vref ) than that without |i̇∗l | in the
used in the experiments is measured based on the experimental
FLPT inputs. The FLPT outputs are shown in Fig. 7. Depending
procedure carried out in [8], which is depicted in Fig. 10.
on the simulation results mentioned above, the proposed control
We first validate the proposed adaptive controller with the
strategies for power management are proved to be effective.
parameters μ = 5 and Voc∗ = Vref = 49.5 V. The load demand
power, the FC, and battery output power (Pfc and Pbat ) in this
B. Experimental Results test cycle are depicted in Figs. 11 and 12, respectively. Note
In order to further verify the effectiveness of the proposed that the FC provides the main load power demand with a slow
controller, the control strategy is deployed in our experimental dynamic and the Li-ion battery, as an auxiliary power source,
FC-battery electric vehicle, as shown in Fig. 8. And the reference provides the transient load power demand. The estimation of the
ranges of some parameters in the controller are obtained from the battery inner resistance R1 is depicted in Fig. 13, which switches
simulations. The FC-rated current and voltage in the vehicle are frequently along with the battery mode change between charging
50 A and 39 V. The capacity and rated voltage of the Li-ion bat- and discharging.
CHEN et al.: ADAPTIVE FUZZY LOGIC CONTROL OF FUEL-CELL-BATTERY HYBRID SYSTEMS FOR ELECTRIC VEHICLES 299

TABLE IV
BATTERY SOC MAINTENANCE EXPERIMENTS

A B C

Tests 1 2 3 1 2 3 1 2 3

V start (V) 49.325 49.898 50.253 49.346 49.825 50.188 49.280 49.867 50.204
V end (V) 49.301 49.872 50.225 49.334 49.814 50.174 49.276 49.860 50.198
S start 23.7% 49.1% 73.1% 26.7% 43.9% 65.7% 19.4% 46.3% 68.5%
S end 20.6% 46.8% 69.8% 25.8% 42.8% 64.9% 19.1% 45.9% 68.1%

The validation of the adaptive control approach with FLPT The battery SOC values are calculated by Fig. 10. Making a
(AFLPT) is divided into two parts: the effectiveness of the FC comparison between the tests of groups A and B in Table IV,
output power regulation and the battery SOC maintenance. In it is clear that the adaptive controller can maintain the battery
Fig. 14, comparisons are made for the adaptive controller with SOC in the desired level more effectively than the average output
and without the FLPT in terms of the FC output power. It is strategy in group A. The SOC maintenance deviation in group
clear that, with the FLPT, the FC output can catch up with the B is about 1.2%. And making a comparison between the tests
load power more effectively under the same current constraint, of groups B and C, it is clear that the AFLPT performs better
especially when the load power demand fluctuates violently on the battery SOC maintenance than the adaptive controller in
(from 330 to 450 s), which means that the AFLPT has a better group B. The SOC maintenance deviation in group C is about
adaptability on satisfying the load power demand in comparison 0.5%. Depending on the experiment results mentioned above,
with the adaptive controller. In the highlighted areas of Fig. 14, the proposed control strategy for power management is proved
with the FLPT, the FC output decreases faster than that without to be effective.
the FLPT under the same system dynamic constraints, which
demonstrates the effectiveness of the proposed control strategy V. CONCLUSION
under the typical regenerative braking driving condition. This is
further clearly shown in the zoom-in in Fig. 14. The FLPT out- In this paper, an adaptive controller with the FLPT is de-
puts in the experimental driving cycle are presented in Fig. 15. veloped to manage the power flow in the FC-battery electric
It is clear that the more violently the load fluctuates, the more system. The proposed control strategy adjusts the power flow
significantly the μ and ΔVfl vary accordingly. The appropriate between the FC and the battery in real time without the need to
variation of these two parameters improve the FC output a lot, predict the system behavior. In order to achieve control objec-
which is the reason why the AFLPT performs better on the FC tives in the adaptive controller, two parameter update laws are
output power regulation. developed for the immeasurable inner resistance of the battery
The battery SOC maintenance capability of the AFLPT is and the current of the ideal controlled current source. The sta-
illustrated by another group of experiments. From Fig. 10, the bility analysis based on Lyapunov theory shows the rationality
battery SOC can be estimated by the battery open-circuit voltage of the adaptive controller. The FLPT is designed to improve the
(Voc ). To validate the controller more effectively, the following performance of the adaptive controller by adjusting the parame-
experimental process is designed: first, before every driving test, ters μ and ΔVfl appropriately under different driving conditions.
the value of the battery open-circuit voltage is denoted as Vstart Simulation and experimental results validate the effectiveness
and the corresponding battery SOC is denoted as Sstart . Then, the of the proposed adaptive controller and FLPT on power man-
FC system is launched and the driving test is processed along agement.
the routes depicted in Fig. 9. After arriving at the terminal point,
the FC system is shutdown immediately and the Li-ion battery is REFERENCES
kept at standby at least half an hour till the battery open-circuit [1] J. Twidell and T. Weir, Renewable Energy Resources. Evanston, IL, USA:
voltage remains unchanged. Finally, the value of the battery Routledge, 2015.
open-circuit voltage is denoted as Vend and the corresponding [2] W. Su, H. Eichi, W. Zeng, and M.-Y. Chow, “A survey on the electrification
of transportation in a smart grid environment,” IEEE Trans. Ind. Informat.,
battery SOC is denoted as Send . Note that each experiment cycle vol. 8, no. 1, pp. 1–10, Feb. 2012.
has the same vehicle driving route, however, all the load demand [3] U. Eberle, B. Müller, and R. von Helmolt, “Fuel cell electric vehicles
curves differ from each other more or less due to the different and hydrogen infrastructure: Status 2012, ” Energy Environ. Sci., vol. 5,
no. 10, pp. 8780–8798, 2012.
driving and traffic conditions. To obtain experimental results [4] D. Feroldi, M. Serra, and J. Riera, “Design and analysis of fuel-cell hybrid
more accurately, the validation driving experiments are divided systems oriented to automotive applications,” IEEE Trans. Veh. Technol.,
into three groups: A, B, and C, where each group has three vol. 58, no. 9, pp. 4720–4729, Nov. 2009.
[5] M. Jang, M. Ciobotaru, and V. G. Agelidis, “Design and implementa-
driving tests. In the test group A, the FC output power is set tion of digital control in a fuel cell system,” IEEE Trans. Ind. Informat.,
as a constant which is the average FC output power calculated vol. 9, no. 2, pp. 1158–1166, May 2013.
from test groups B and C. The proposed adaptive controller [6] I. L. Sarıoğlu, O. P. Klein, H. Schröder, and F. Küçükay, “Energy man-
agement for fuel-cell hybrid vehicles based on specific fuel consumption
and the proposed AFLPT are utilized in the test groups B and due to load shifting,” IEEE Trans. Intell. Transp. Syst., vol. 13, no. 4,
C, respectively. All the test results are presented in Table IV. pp. 1772–1781, Dec. 2012.
300 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 14, NO. 1, JANUARY 2018

[7] J. Bauman and M. Kazerani, “A comparative study of fuel-cell- [29] W. J. Rugh, Linear System Theory, vol. 2. Upper Saddle River, NJ, USA:
battery, fuel-cell-ultracapacitor, and fuel-cell-battery-ultracapacitor vehi- Prentice-Hall, 1996.
cles,” IEEE Trans. Veh. Technol., vol. 57, no. 2, pp. 760–769, Mar. 2008. [30] B. Wahdame “Analysis of a fuel cell durability test based on design
[8] M. Chen and G. A. Rincon-Mora, “Accurate electrical battery model of experiment approach,” IEEE Trans. Energy Convers., vol. 23, no. 4,
capable of predicting runtime and I-V performance,” IEEE Trans. Energy pp. 1093–1104, Dec. 2008.
Convers., vol. 21, no. 2, pp. 504–511, Jun. 2006.
[9] Z. Jiang and R. A. Dougal, “A compact digitally controlled fuel cell/battery
hybrid power source,” IEEE Trans. Ind. Electron., vol. 53, no. 4, pp. 1094–
1104, Jun. 2006.
[10] X. Hu, N. Murgovski, L. M. Johannesson, and B. Egardt, “Optimal di-
mensioning and power management of a fuel cell/battery hybrid bus via
Jian Chen (M’06–SM’10) received the B.E. de-
convex programming,” IEEE/ASME Trans. Mechatron., vol. 20, no. 1, pp.
gree in detection technology and instrumen-
457–468, Feb. 2015.
tation and the M.E. degree in control sci-
[11] B. Vural, S. Dusmez, M. Uzunoglu, E. Ugur, and B. Akin, “Fuel consump-
ence and engineering from Zhejiang University,
tion comparison of different battery/ultracapacitor hybridization topolo-
Hangzhou, China, in 1998 and 2001, respec-
gies for fuel-cell vehicles on a test bench,” IEEE Trans. Emerg. Sel. Topics
tively, and the Ph.D. degree in electrical engi-
Power Electron., vol. 2, no. 3, pp. 552–561, Sep. 2014.
neering from Clemson University, Clemson, SC,
[12] C. Carrejo, R. Giral, E. Arango, C. Ramos-Paja, C. Olalla, and C. Alonso,
USA, in 2005.
“Fuel cell hybrid topologies selection and control,” in Proc. IEEE Int.
From 2006 to 2008, he was a Research Fel-
Conf. Power Eng., Energy Electr. Drives, 2011, pp. 1–6.
low with the University of Michigan, Ann Arbor,
[13] Y. Ates, O. Erdinc, M. Uzunoglu, and B. Vural, “Energy management of a
MI, USA, where he was involved in fuel cell mod-
FC/UC hybrid vehicular power system using a combined neural network-
eling and control. In 2008, he joined IdaTech LLC, Bend, OR, USA, where
wavelet transform based strategy,” Int. J. Hydrog. Energy, vol. 35, no. 2,
he was involved in the National Fuel Bus Program. In 2013, he joined
pp. 774–783, Jan. 2010.
the Department of Control Science and Engineering, Zhejiang Univer-
[14] M. Karthik and K. Gomathi, “Design, modeling and simulation study of a
sity, where he is currently a Professor in the College of Control Science
cascaded optimal neural network based fuel cell powered electric vehicle,”
and Engineering. His research interests include modeling and control
in Proc. IEEE 2nd Int. Conf. Electr. Energy Syst., 2014, pp. 71–76.
of fuel cell vehicles, battery management, visual servo techniques, and
[15] P. Thounthong, S. Rael, and B. Davat, “Energy management of fuel
nonlinear control.
cell/battery/supercapacitor hybrid power source for vehicle applications,”
J. Power Sources, vol. 193, no. 1, pp. 376–385, 2009.
[16] M. Pahlevaninezhad, P. Das, J. Drobnik, G. Moschopoulos, P. K. Jain, and
A. Bakhshai, “A nonlinear optimal control approach based on the control-
Lyapunovfunction for an ac/dc converter used in electric vehicles,” IEEE
Trans. Ind. Informat., vol. 8, no. 3, pp. 596–614, Aug. 2012.
[17] D. Feroldi, M. Serra, and J. Riera, “Energy management strategies based Chenfeng Xu was born in 1992. He received
on efficiency map for fuel cell hybrid vehicles,” J. Power Sources, the Bachelor’s degree in automation engineer-
vol. 190, no. 2, pp. 387–401, 2009. ing from Central South University, Changsha,
[18] F. R. Salmasi, “Control strategies for hybrid electric vehicles: Evolution, China, in 2014. He is currently working toward
classification, comparison, and future trends,” IEEE Trans. Veh. Technol., the M.E. degree in control science and engineer-
vol. 56, no. 5, pp. 2393–2404, Sep. 2007. ing at Zhejiang University, Hangzhou, China.
[19] J. P. Torreglosa, P. Garcia, L. M. Fernandez, and F. Jurado, “Predictive His research interests include hybrid fuel
control for the energy management of a fuel-cell-battery-supercapacitor cell/battery system energy management and
tramway,” IEEE Trans. Ind. Informat., vol. 10, no. 1, pp. 276–285, fuzzy logic control.
Feb. 2014.
[20] A. Taghavipour, M. Vajedi, N. L Azad, and J. McPhee, “Predictive power
management strategy for a PHEV based on different levels of trip in-
formation,” Engine Powertrain Control, Simul. Model., vol. 3, no. 1,
pp. 326–333, 2012.
[21] D. Gao, Z. Jin, and Q. Lu, “Energy management strategy based on fuzzy
logic for a fuel cell hybrid bus,” J. Power Sources, vol. 185, no. 1, Chengshuai Wu was born in 1992. He received
pp. 311–317, 2008. the Bachelor’s degree in automation engineer-
[22] H. Hemi, J. Ghouili, and A. Cheriti, “A real time fuzzy logic power ing, in 2014 from Zhejiang University, Hangzhou,
management strategy for a fuel cell vehicle,” Energy Convers. Manage., China, where he is currently working toward the
vol. 80, pp. 63–70, 2014. Ph.D. degree in control science and engineer-
[23] A. A. Ferreira, J. A. Pomilio, G. Spiazzi, and L. de Araujo Silva, “En- ing.
ergy management fuzzy logic supervisory for electric vehicle power sup- His current research interests include hybrid
plies system,” IEEE Trans. Power Electron., vol. 23, no. 1, pp. 107–115, fuel cell/battery systems, nonlinear control, and
Jan. 2008. port-Hamiltonian systems.
[24] O. Tremblay et al., “A generic fuel cell model for the simulation of
fuel cell vehicles,” in Proc. IEEE Veh. Power Propulsion Conf., 2009,
pp. 1722–1729.
[25] M. Hilairet, M. Ghanes, O. Béthoux, V. Tanasa, J. Barbot, and D.
Normand-Cyrot, “A passivity-based controller for coordination of con-
verters in a fuel cell system,” Control Eng. Pract., vol. 21, no. 8,
pp. 1097–1109, 2013. Weihua Xu received the B.E. degree in automa-
[26] L. Gao, S. Liu, and R. A. Dougal, “Dynamic lithium-ion battery model tion engineering and the Ph.D. degree in control
for system simulation,” IEEE Trans. Compon. Packag. Technol., vol. 25, science and engineering from Zhejiang Univer-
no. 3, pp. 495–505, Sep. 2002. sity, Hangzhou, China, in 1998 and 2003, re-
[27] E. Van Dijk, J. Spruijt, D. M. O’sullivan, and J. B. Klaassens, “PWM- spectively.
switch modeling of dc–dc converters,” IEEE Trans. Power Electron., She is currently an Associate Professor in
vol. 10, no. 6, pp. 659–665, Nov. 1995. the College of Control Science and Engineer-
[28] T. Das and S. Snyder, “Adaptive control of a solid oxide fuel cell ultra- ing, Zhejiang University. Her research interests
capacitor hybrid system,” IEEE Trans. Control Syst. Technol., vol. 21, include adaptive control of uncertain linear sys-
no. 2, pp. 372–383, Mar. 2013. tems, modeling and control of distributed energy
systems, and nonlinear control.

You might also like