Você está na página 1de 9


3, MAY 2002

Fuzzy Logic Control for Parallel Hybrid Vehicles

Niels J. Schouten, Mutasim A. Salman, and Naim A. Kheir

Abstract—In this paper, a fuzzy logic controller is developed • Series-Parallel Combined System: Toyota Prius is an ex-
for hybrid vehicles with parallel configuration. Using the driver ample of this so-called dual system.
command, the state of charge of the energy storage, and the Because of the importance and potential benefits of the hy-
motor/generator speed, a set of rules have been developed, in
a fuzzy controller, to effectively determine the split between brid technology, many of the national laboratories, research in-
the two powerplants: electric motor and internal combustion stitutions, and universities are actively involved in this research
engine. The underlying theme of the fuzzy rules is to optimize the area. Hybrid technology is also one of the research topics for
operational efficiency of all components, considered as one system. a partnership between United States government and the auto-
Simulation results have been used to assess the performance of motive industry, i.e., Partnership for a New Generation of Ve-
the controller. A forward-looking hybrid vehicle model was used
for implementation and simulation of the controller. Potential hicles (PNGV). PNGV was established with the objective to
fuel economy improvement has been shown by using fuzzy logic, develop a new generation of vehicles. These future vehicles
relative to other controllers, which maximize only the efficiency should achieve up to three times todays average fuel economy
of the engine. (80 mi/gal 34 km/l), without compromising consumer expec-
Index Terms—Control strategies, emissions, fuel economy, fuzzy tations with respect to performance, comfort, safety, quality, and
logic, hybrid vehicles, optimization. cost of ownership.
In order to achieve these goals, it is very important to opti-
mize the architecture and components of the hybrid vehicle, but
as important is the energy management strategy that is used to

T HE search for improved fuel economy, reduced emission,

and affordable vehicles, without sacrificing vehicle per-
formance, safety, reliability, and other conventional vehicle at-
control the complete system. The energy management strategy
is implemented by a power controller. It controls the energy flow
between all components, and optimizes power generation and
tributes has made the hybrid technology (both thermal and elec- conversion in the individual components.
trical motorization) one of the challenges for the automotive The objective of this paper is to develop a power controller
industry. Hybrid systems, using a combination of an internal for a parallel hybrid vehicle (PHV) that will optimize the fuel
combustion engine (ICE) and electric motor (EM), have the po- economy. PHV power controllers presented in the past [2], [5]
tential of improving fuel economy by operating the ICE in the only optimize the ICE operation, thus not using the full potential
optimum efficiency range and by making use of regenerative of hybrid technology. The power controller presented here will
braking during deceleration. optimize the operation of all major PHV components: ICE, EM,
Within the last two decades, the automotive industry has in- and battery.
creasingly developed vehicles with hybrid power-trains. Many For the implementation of the controller, fuzzy logic has
concept vehicles, with several hybrid configurations, have been been used. Previous research at Oakland University [6], [7]
presented by several auto manufacturers. Few manufacturers and Ohio State University [8], [9] already indicated that fuzzy
have actually developed or have plans to develop hybrid vehi- logic control is very suitable for hybrid vehicle control. It is
cles for mass production [1]–[3]. a good method for realizing an optimal tradeoff between the
There are three different types of hybrid systems, (see [2], [4] efficiencies of all components of the PHV. Fuzzy logic control
for details): is tolerant to imprecise measurements and to component
• Series Hybrid: In this configuration, an ICE-generator variability. It also gives a systematic methodology for the
combination is used for providing electrical power to the development of a rule-based energy management strategy.
EM and the battery. Section II of the paper, explains the basics of parallel hybrid
• Parallel Hybrid: The ICE in this scheme is mechanically vehicles, and briefly describes the simulation model. The energy
connected to the wheels, and can therefore directly supply management strategy is presented in Section III, followed by the
mechanical power to the wheels. The EM is added to the description of the fuzzy logic power controller in Section IV. Fi-
drivetrain in parallel to the ICE, so that it can supplement nally, the simulation results and a brief analysis of these results
the ICE torque. are presented in Section V.

Manuscript received February 23, 2000; revised September 14, 2001. Man-
uscript received in final form January 28, 2002. Recommended by Associate II. BASICS OF PHVs
Editor M. Jankovic. This work was supported by Partnership for New Genera-
tion of Vehicles, under contract to Oakland University. Fig. 1 presents a block diagram of a PHV with an EM and an
N. J. Schouten and N. A. Kheir are with the Electrical and Systems Engi- ICE. For this particular configuration, the ICE and EM power
neering Department, Oakland University, Rochester, MI 48309 USA.
M. A. Salman is with General Motors Corporation, Warren, MI 48090 USA. are combined downstream of the transmission. Alternatively, the
Publisher Item Identifier S 1063-6536(02)03420-6. power could also be combined upstream of the transmission.
1063-6536/02$17.00 © 2002 IEEE


Fig. 1. Block diagram of the parallel hybrid vehicle.

For both the upstream and downstream configuration, there

are five different ways to operate the system, depending on the
flow of energy: 1) provide power to the wheels with only the
ICE; 2) only the EM; or 3) both the ICE and the EM simultane-
ously; 4) charge the battery, using part of the ICE power to drive important to have the battery SOC charge-neutral for each of the
the EM as a generator (the other part of ICE power is used to cycles.
drive the wheels); 5) slow down the vehicle by letting the wheels The specific PHV configuration, used throughout the paper,
drive the EM as a generator that provides power to the battery consists of the following components:
(regenerative braking). • compression ignition direct injection (CIDI) engine:
A power controller is needed to manage the flow of energy 55 kW;
between all components, while taking into account the energy • permanent magnet motor: 20 kW continuous, 40 kW peak;
available in the battery. The power controller adds the capability • advanced battery: 40 kW, 2 kWh;
for the components to work together in harmony, while at the • manual transmission: five speed;
same time optimizes the operating points of the individual com- • total test vehicle mass: 1100 kg.
ponents. This is clearly an added complexity not found in con- The size of the components was chosen to achieve the PNGV
ventional vehicles. vehicle performance requirements given in Table I.
For controller analysis and design, the PNGV systems
analysis toolkit (PSAT) models [10] are used to simulate the III. ENERGY MANAGEMENT STRATEGY
PHV in the Matlab/Simulink environment. These models were
This section describes the energy management strategy,
developed under direction and contributions of the three major
which is the philosophy behind the power controller. The
United States car companies (DaimlerChrysler, Ford, and
energy in the system should be managed in such a way that:
General Motors) and the United States government. The model
architecture is “forward looking,” which is much more suitable 1) The driver inputs (from brake and accelerating pedals) are
for controller analysis and design than a backward looking satisfied consistently (driving the PHV should not “feel”
architecture (see [11] for a discussion on forward and backward different from driving a conventional vehicle).
looking models, and see [4], [12] for a description of backward 2) The battery is sufficiently charged at all times.
looking models). 3) The overall system efficiency of the four basic compo-
Apart from the model of the PHV, PSAT also includes a nents (ICE, EM, battery, and transmission) is optimized.
model of a driver, and a set of driving cycles given as speed While the PHV is operated, the power controller should de-
versus time profiles. The driver’s objective is to track the termine how much power is needed to drive the wheels based on
speed versus time profile. The driver achieves his objective the driver inputs, and how much is needed to charge the battery.
by modulating the brake and accelerator pedals. The power Then it should split the power between ICE and EM. If the bat-
controller uses these driver inputs to compute the commands tery needs to be charged, negative power needs to be assigned to
for the several local controllers, such as the ICE, EM, battery, the EM, and the ICE should provide the power for both driving
and transmission controllers. the wheels and charging the battery.
The fuel economy for composite urban/highway travel is The power-split strategy can be used to optimize the effi-
computed by letting the driver track the profiles described ciency of all four basic components of the PHV, because it deter-
in the SAE J1711 standard. The composite fuel economy mines the operating points of the components. Once the power
(Composite_FE) is given as is split in an optimal way, the power generation/conversion of
the individual components also has to be optimized. To identify
Composie FE city FE highway FE the optimal operating points, the efficiency maps of the compo-
nents were used.
The city-driving test standard starts from ambient initial con-
ditions (also known as “cold start”). It is composed of two urban A. Efficiency Maps
cycles separated by a 10-min soak. In our analysis, the engine Fig. 2 presents a contour plot of the efficiency of a generic
is assumed to be warm and only the second urban cycle is used. CIDI engine in the speed-torque plane. Superimposed on the
For the highway cycle only one cycle with warmed-up engine contour plot is the optimal efficiency curve. For a given ICE
is used. In order to have a meaningful fuel economy value, it is power level, the optimal efficiency curve defines the optimal

Fig. 4. Efficiency map of the battery. The arrows indicate increasing efficiency.
Fig. 2. Efficiency map and optimal curve of the ICE. The arrows indicate
increasing efficiency.

Fig. 5. Block diagram of the energy flow: (1) mechanical path and
(2) electrical path.

The EM speed is directly related to vehicle speed, because

it cannot be controlled by gear shifting. Therefore, the EM ef-
ficiency can only be optimized, by optimizing the power at a
given EM speed. The efficiency is optimal for EM speeds be-
Fig. 3. Efficiency map, continuous and peak power curves of the EM. The tween 320 and 430 rad/s. The optimal power for this region is
arrows indicate increasing efficiency.
approximately 10 kW.
Fig. 4 presents a generic efficiency plot of an advanced bat-
operating point in the speed-torque plane. Once the optimal tery in the state of charge (SOC)-power plane. It can easily be
speed and torque are known, the torque is controlled by varying seen that the battery operates most efficiently for high SOC and
the throttle angle and the speed by shifting gears of the auto- low power levels. For power-split control, this indicates that the
mated manual transmission (note that for a given vehicle speed, SOC should be as high as possible by frequently charging at low
a higher gear number will result in a lower engine speed). power level. In this study, it is assumed that the highest admis-
For the generic CIDI map used in this paper, the ICE effi- sible SOC is 0.9 for charging, and 0.97 for regenerative braking.
ciency on the optimal curve is highest for engine speeds between
230 and 320 rad/s, corresponding to an ICE power between 30 B. Power Split Strategy
and 50 kW, with the absolute optimum at 47 kW. Therefore, the
power-split strategy should preferably result in an ICE power Now that the efficiency characteristics of the components are
level in this range. The efficiency on the curve is lowest for known, it is possible to formulate the power-split strategy. The
very low and very high engine speeds; therefore the ICE power difference between using the ICE or the EM to drive the wheels
should preferably not be below 6 kW or over 50 kW. is explained in Fig. 5. When the ICE is used (path 1), the en-
Fig. 3 presents a contour plot of the efficiency of the per- ergy flows directly from the ICE through the transmission to
manent magnet motor also in the speed-torque plane. The effi- the wheels. When the EM is used (path 2), energy first flows
ciency plot for motoring and generating mode is the same, only from the ICE through the transmission to the EM, operated as
the sign of the torque is reversed. Superimposed on the plot are a generator, for charging the battery; later the energy will flow
the continuous and peak power curves. from the battery to the EM, operated as motor, to the wheels.

Fig. 6. Simplified block diagram of the power controller.

In path 1 the mechanical power produced by the ICE will TABLE II

immediately be used to drive the wheels. In path 2 the same
mechanical power will be converted to electrical, chemical,
electrical, and back to mechanical. Inherent to these power
conversions are losses. These additional losses considerably
lower the efficiency of using the EM for driving the wheels.
Therefore, using the EM (path 2) will only be efficient in a few
(extreme) situations, when directly using the ICE (path 1) is
very inefficient.
The minimum losses due to power conversions are deter-
mined by the efficiency of the EM, first used as a generator
and then as a motor, and of the battery, first used to store and
then to release energy. The maximum round-trip efficiency
of the battery used in our simulations is 0.95 and of the EM
(motor/generator) 0.88, so that the maximum efficiency of the
power conversions in path 2 is 84% (minimum additional losses
16%). Therefore, it is only justified to use the EM (path 2), in
the cases that directly using the ICE (path 1) is at least 16% less
efficient than using the ICE in path 2 (for charging the battery). [13]–[16]. Fig. 6 presents a block diagram of the power con-
For a good energy management strategy the average ICE ef- troller, the fuzzy logic controller (FLC) being the main part.
ficiency in path 2 (ICE used for charging the battery), will be The basic idea of an FLC is to formulate human knowledge and
close to the optimal ICE efficiency. Using the ICE efficiency reasoning, which can be represented as a collection of if–then
map (Fig. 2), it was determined that the efficiency on the op- rules, in a way tractable for computers. Table II presents a list
timal curve only drops below 16% of the absolute optimum for of if–then rules that represent the energy management strategy
power levels below 6 kW and over 50 kW. Therefore, when the described in the previous section (a description of the rule
power command is below 6 kW, only the EM is to be used to base will follow in the next section). The variables in Table II
drive the wheels. Between 6 and 50 kW, only the ICE is to be preceding “then”: SOC, , and , correspond to the
used to drive the wheels, and the ICE can also produce addi- inputs of the FLC (Fig. 6): battery state of charge, driver power
tional power to charge the battery. When the power command is command, and EM speed, respectively. The variables following
over 50 kW, the EM is to be used to complement the ICE, be- “then”: (generator power), and scaling factor correspond
cause in this case the ICE is again very inefficient and close to to the outputs.
its maximum (55 kW). The first part of a rule (preceding “then”), called the
The EM power for driving the wheels is computed using this antecedent, specifies the condition (i.e., the combination of
strategy. When the driver power command is between 6 and inputs) for which a rule holds. The second part (following
50 kW, the EM power for charging the battery still needs to “then”), called the consequent, is the corresponding control
be optimized. The next section discusses how the driver power action, i.e., the controller output. The antecedents in Table II
command is computed and how fuzzy logic is used to compute contain linguistic terms (low, high, optimal, etc.) that reflect
the optimal generator power. human knowledge of the PHV. The antecedents are defined as a
combination of individual conditions, using the logical AND and
IV. FUZZY LOGIC POWER CONTROLLER NOT operators (in general, it is also possible to use the logical
OR). The linguistic terms, the connectives, and the if–then
This section first discusses the basics of the fuzzy logic con- relations need to be defined in a way tractable for computers.
troller that was used to implement the power controller, and then The linguistic terms are represented by fuzzy sets. Consider
the power controller itself will be presented in more detail. the variable “driver power command” in Fig. 7. This variable is
represented by the linguistic terms “normal” and “high.” These
A. Fuzzy Logic Control Basics linguistic terms are represented by two fuzzy sets that are de-
The general energy management strategy described in the fined by the two membership functions in Fig. 7. The member-
previous section has been implemented using fuzzy logic ship functions define the degree of membership of the vari-

Inference and Aggregation can be combined into a single


with the number of rules.

The fuzzy logic controller (FLC) described in this paper is of
the Sugeno–Takagi type [17]. A conceptually different type of
FLC is the Mandami FLC [18], for which the consequent part
Fig. 7. Membership functions for driver power command.
of the rules (the control actions) are also represented by fuzzy
sets. For Mandami’s approach the inference and aggregation act
on the fuzzy sets of the consequents, instead of single (“crisp”)
able in the two fuzzy sets. For driver power command larger than
values, and the result of both steps is again a fuzzy set. A fifth
50 kW the degree of membership in the fuzzy set “high.”
step (defuzzification) is needed to replace this fuzzy set by a
equals 1 and the degree of membership in the fuzzy set “normal”
single (“crisp”) value (such as the center of area of the fuzzy
equals 0. For normal driver power command less than
set) that acts as one of the controller outputs.
30 kW, it is the other way around. One can imagine that there
is a gradual transition from normal to high driver power com-
mand. This transition is represented by the overlapping interval B. Power Controller
in Fig. 7 (between 30 and 50 kW). The values in this interval Fig. 6 presents a simplified block diagram of the power con-
belong to both fuzzy sets with various degrees of membership, troller. The first block converts the driver inputs from the brake
e.g., 45 kW belongs to “normal” with membership 0.25 and to and accelerator pedals to a driver power command. The sig-
“high” with membership 0.75. nals from the pedals are normalized to a value between zero
In order to work with the fuzzy rules in Table II the fuzzy and one (zero: pedal is not pressed, one: pedal fully pressed).
sets need to be combined using the logical connectives AND and The braking pedal signal is then subtracted from the acceler-
NOT (and in general also OR). For this purpose, the logical con- ating pedal signal, so that the driver input takes a value between
nectives from conventional Boolean logic have been extended 1 and 1.
to their fuzzy equivalents. There are several ways to implement The negative part of the driver input is send to a separate
these fuzzy equivalents, but in general the minimum operator is brake controller that will compute the regenerative braking and
used for AND [e.g., ], the simple comple- the friction braking power required to decelerate the vehicle.
ment (e.g., ) for NOT, and the maximum operator The controller will always maximize the regenerative braking
for OR [e.g., ]. power, but it can never exceed 65% of the total braking power
Using the fuzzy sets and the fuzzy set operators, it is possible required, because regenerative braking can only be used for the
to design a fuzzy reasoning system (inference system) that can front wheels.
act as a fuzzy controller. The reasoning mechanism used in this The positive part of the driver input is multiplied by the max-
paper can be separated in four main steps. imum available power at the current vehicle speed. This way all
1) Fuzzification: The membership degrees of the three input power is available to the driver at all times.
values of the FLC are computed using the membership The maximum available power is computed by adding the
functions (Figs. 7–9). maximum available ICE and EM power. The maximum avail-
2) Degree of Fulfillment: The degree of fulfillment for the able EM and ICE power depends on EM/ICE speed and EM/ICE
antecedent of each rule (Table II) is computed using temperature, and is computed using a two–dimensional look-up
the fuzzy logic operators. This degree of fulfillment table with speed and temperature as inputs. However, for a given
determines to which degree the th rule is valid. For vehicle speed, the ICE speed has one out of five possible values
example: Table II, rule 3: (SOC), (one for each gear number of the transmission). To obtain the
1 . maximum ICE power, first the maximum ICE power levels for
3) Inference: This operation represents the if–then implica- those five speeds are computed, and then the maximum of these
tion. The degree of fulfillment of the antecedent of each values is selected.
rule is used to modify the consequent of that rule accord- Once the driver power command is computed, the fuzzy logic
ingly. This is done by multiplying the degree of fulfill- controller (Fig. 6) computes the optimal generator power for the
ment of the antecedent with the consequent of EM in case it is used for charging the battery and a scaling factor
rule . For example Table II, rule 2: kW. for the EM in case it is used as a motor. This scaling factor is
4) Aggregation: For each controller output , the results (close to) zero when the SOC of the battery is too low. In that
of the inference step are combined into a single value. case the EM should not be used to drive the wheels, in order
This is done by taking the average of the inference re- to prevent battery damage. When the SOC is high enough, the
sults weighted by the degrees of fulfillment of the rules. scaling factor equals one.

Fig. 8. Membership functions for state of charge.

Fig. 10. Operating points, efficiency map, and optimal curve of the internal
combustion engine for the fuzzy logic controller.

zero. Rule 8 prevents battery charging when the driver power

demand is high and the SOC is not too low. Charging in this sit-
uation will shift the ICE power level outside the optimum range
(30–50 kW). Finally, when the SOC is not too low (rule 9), the
scaling factor is one.
The third block in Fig. 6 computes the final values for the ICE
power and EM power , using , , and
the scaling factor (SF). normally is the driver power com-
mand added to the desired generator power
Fig. 9. Membership functions for electric motor (EM) speed.
, and is minus the desired generator power
. There are two exceptions.
The three inputs of the fuzzy logic controller are: driver power 1) When is smaller than the threshold
command , SOC and EM speed . The member- value (SF 6 kW), then kW, and
ship functions (MFs) for driver power command are presented .
in Fig. 7. Fuzzy set “normal” represents the power range for 2) When is larger than the max-
“normal” driving conditions, “high” represents the range only imum ICE power at current speed ,
used during high acceleration and high speed. The power range then and
for the transition between normal and high (30–50 kW) is the .
optimal range for the ICE. Finally, when is positive (EM used as motor), has
Fig. 8 presents the MFs for SOC. Fuzzy sets “too low” and to be multiplied by the scaling factor SF .
“too high” represent the ranges where the SOC should not be.
Fuzzy set “normal” represents the range where the SOC should C. Gear Shifting Control
be and “low” acts as a buffer between “normal” and “too low.”
The desired ICE power level is used by the gear shifting con-
Fig. 9 presents the MFs for EM speed. Fuzzy set “optimal”
troller to compute the optimum gear number of the automated
represents the optimal speed range. The MF drops relatively
manual transmission. First, the optimal speed-torque curve is
fast, since the efficiency also drops fast for speeds outside the
used to compute the optimal ICE speed and torque for the de-
optimal range (Fig. 3).
sired ICE power level. The optimal ICE speed is then divided
The rule base is presented in Table II. If the SOC is too high
by the vehicle speed to obtain the desired gear ratio. Finally, the
(rule 1) the desired generator power will be zero, to pre-
gear number closest to the desired gear ratio is chosen.
vent overcharging the battery. If the SOC is normal (rules 2 and
3), the battery will only be charged when both the EM speed
is optimal and the driver power is normal. If the SOC drops to V. SIMULATION RESULTS
low, the battery will be charged at a higher power level. This The power controller has been implemented and simulated
will result in a relatively fast return of the SOC to normal. If with PSAT using the driving cycles described in the SAE J1711
the SOC drops to too low (rules 6 and 7), the SOC has to be in- standard.
creased as fast as possible to prevent battery damage. To achieve Figs. 10–12 present the operating points of the ICE, EM, and
this, the desired generator power is the maximum available gen- battery plotted on the efficiency maps. The operating points in
erator power and the scaling factor is decreased from one to Fig. 10 are close to the optimal curve, which indicates that the

Fig. 11. Operating points, efficiency map, continuous and peak power curves
of the electric motor (EM) for the fuzzy logic controller.
Fig. 13. Operating points, efficiency map, and optimal curve of the ICE for
the default PSAT controller.

Fig. 12. Operating points and efficiency map of the battery for the fuzzy logic

ICE has been operated close to optimal efficiency. The operating Fig. 14. Operating points, efficiency map, continuous and peak power curves
points of the EM (Fig. 11) are mainly in the optimal speed range of the EM for the default controller.
of 320–430 rad/s. The operating points of the battery (Fig. 12)
are at a relatively high SOC (between 0.77 and the maximum (FLC)]. Therefore, for the default controller the operating points
0.9), and the power level is relatively low, both resulting in high of the ICE (Fig. 13) are closer to the optimal curve than for the
efficiency. FLC (see Fig. 10).
These results have been compared with the results for the de- However, the efficiency of the other components is better for
fault controller in PSAT. This controller only optimizes the ICE the FLC than for the default controller, and therefore the overall
efficiency, and does not optimize EM, battery or transmission efficiency is better for the FLC. Figs. 14 and 15 present the op-
efficiency. It optimizes the ICE efficiency by adjusting both the erating points of the EM, and battery for the default controller.
ICE speed (using the transmission gear ratio) and torque to make The operating points of the EM are much more spread out over
the operating point as close as possible to the optimal curve. the map, and are not particularly in the optimal speed range of
For the default controller the efficiency of all components is 320–430 rad/s (compare to Fig. 11), which explains the lower
lower, except for the ICE efficiency. The default controller is EM efficiency compared to the FLC. The operating points of the
more effective in the optimization of the ICE, because there is battery are in the desired SOC range, but the average charging
no tradeoff between the ICE efficiency and the efficiency of the and discharging power is higher (compare to Fig. 12), thus de-
other components of the PHV [unlike the fuzzy logic controller creasing the battery efficiency.

in both tables). For the FLC the losses in the ICE, EM, battery,
and drivetrain are smaller for both the urban and highway cycles.
The vehicle losses, energy for accessories and friction braking
are approximately the same, because both controllers are simu-
lated with the same vehicle and the same driving cycles.
Although the ICE efficiency is higher for the default con-
troller, the ICE losses for the FLC are smaller. This is because
the ICE produces less energy for the FLC. Less energy is needed
to complete the cycle, since the efficiency of the other compo-
nents is higher than for the default controller.
The overall improvement of the FLC for an urban cycle equals
6.8% and for a highway cycle 9.6%.

In this paper, a fuzzy logic-based power controller for PHVs
has been presented. This power controller optimizes the energy
flow between the main components of the PHV and optimizes
Fig. 15. Operating points and efficiency map of the battery for the default the energy generation and conversion in the individual compo-
controller. nents (ICE, EM, transmission, and battery). The efficiency maps
of the components have been used to design the controller.
TABLE III The power controller first converts the accelerator and brake
NORMALIZED LOSSES FOR THE DEFAULT CONTROLLER pedal inputs of the driver to a driver power command. The driver
power command, state of charge of the battery, and electric
motor speed are then used by a fuzzy logic controller to com-
pute the optimal generator power and a scaling factor for the
electric motor. The driver power command, optimal generator
power, and scaling factor are used to compute the optimal ICE
and EM power. Furthermore, the efficiency of the ICE for a
given power level is optimized using an optimal speed-torque
curve, and using gear shifting to control the speed of the ICE.
The power controller ensures that the driver inputs (from
brake and accelerator pedals) are satisfied consistently, the
battery is sufficiently charged at all times, and the fuel economy
of the PHV is optimized.
TABLE IV Simulation results, using the driving cycles described in the
SAE J1711 standard, show potential improvement by using
fuzzy logic, over other strategies that optimize only the ICE
In future research, the robustness of the fuzzy logic controller
will be investigated in more detail. The control structure will be
used to design a controller that minimizes the emissions and
maximizes the fuel economy. Adaptive/learning elements will
be added to the controller, to enable on-line controller optimiza-
tion. Combined component size and controller optimization will
also be studied.

To be able to compare the efficiency of the default controller [1] N. Hattori, S. Aoyama, S. Kitada, I. Matsuo, and K. Hamai, “Config-
uration and operation of a newly developed parallel hybrid propulsion
and the FLC, the battery energy at the beginning and the end of system,” in Proc. Global Powertrain Congr., Detroit, MI, Oct. 6–8, 1998.
each cycle has to be the same. This is done by using an algorithm [2] B. K. Powell, K. E. Bailey, and S. R. Cikanek, “Dynamic modeling and
that iteratively adapts the initial SOC until the SOC at the end of control of hybrid electric vehicle powertrain systems,” IEEE Contr. Syst.
Mag., pp. 17–33, Oct. 1998.
the cycle equals the initial SOC. This is the basis for the charge [3] J. M. Miller, A. R. Gale, and A. Sankaran, “Electric drive subsystem
sustaining part of the SAE J1711 test procedure. for a low-storage requirement hybrid electric vehicle,” IEEE Trans. Veh.
Table s III and IV present the normalized losses for the urban Syst., vol. 48, pp. 1788–1796, Nov. 1999.
[4] Z. Rahman, K. L. Butler, and M. Ehsani, “Designing parallel hybrid
and highway cycle, respectively. The losses are normalized with electric vehicles using V-ELPS 2.01,” in Proc. Amer. Contr. Conf., San
respect to the total loss for the default controller (which is 100% Diego, CA, June 1999, pp. 2693–2697.

[5] M. Ehsani, Y. Gao, and K. L. Butler, “Application of Electrically [11] K. B. Wipke, M. R. Cuddy, and S. D. Burch, “ADVISOR 2.1:
Peaking Hybrid (ELPH) propulsion system to a full-size passenger car User-friendly advanced powertrain simulation using a combined
with simulated design verification,” IEEE Trans. Veh. Syst., vol. 48, pp. backward/forward approach,” IEEE Trans. Veh. Technol., vol. 48, pp.
1779–1787, Nov. 1999. 1751–1761, Nov. 1999.
[6] N. Jalil and N. Kheir, “Energy management studies for a new genera- [12] L. Guzzella and A. Amstutz, “CAE tools for quasistatic modeling and
tion of vehicles (Milestone #5: Fuzzy logic for the series hybrid),” Tech. optimization of hybrid powertrains,” IEEE Trans. Veh. Technol, vol. 48,
Rep., Dept. Elect. Syst. Eng., School Eng. Comput. Sci., Oakland Univ., pp. 1762–1770, Nov. 1999.
Rochester, MI, Mar. 1998. [13] U. Kaymak, R. Babuška, and H. R. van Nauta Lemke, “Fuzzy con-
[7] , “Energy management studies for a new generation of vehicles trol—Theory and design,” J. A, vol. 36, no. 3, 1995.
(Milestone #6: Fuzzy logic for the parallel hybrid),” Tech. Rep., Dept. [14] K. Hirota, Ed., Industrial Applications of Fuzzy Technology. Tokyo,
Elect. Syst. Eng., School Eng. Comput. Sci., Oakland Univ., Rochester, Japan: Springer-Verlag, 1993.
MI, Mar. 1998. [15] D. Driankov, H. Hellendoorn, and M. Reinfrank, An Introduction to
[8] B. M. Baumann, “Intelligent control strategies for hybrid vehicles using Fuzzy Control. Berlin, Germany: Springer-Verlag, 1993.
neural networks and fuzzy logic,” Master’s thesis, Dept. Elect. Eng., [16] C. C. Lee, “Fuzzy logic in control systems,” IEEE Trans. Syst., Man,
Ohio State Univ., Columbus, 1997. Cybern., vol. 20, pp. 404–435, 1990.
[9] H. Kono, “Fuzzy control for hybrid electric vehicles,” Master’s thesis, [17] T. Takagi and M. Sugeno, “Fuzzy identification of systems and its appli-
Department of Electrical Engineering, The Ohio State University, 1998. cation to modeling and control,” IEEE Trans. Syst., Man, Cybern., vol.
[10] S. T. McBroom, “Toolkit for tomorrow’s car,” in Technology Today: 15, pp. 116–132, 1985.
Southwest Research Institute Publications, Spring 1997. [Online]. Avail- [18] E. H. Mandami, “Applications of fuzzy algorithms for control of simple
able: www.swri.org. dynamic plant,” Proc. IEE, pp. 1585–1588, 1974.