You are on page 1of 6

Optimal Speed Control of Hybrid Electric

Vehicle Using GWO based Fuzzy-PID Controller

Priya Bisht Mrs. Jyoti Yadav


ICE Division ICE Division
NSIT, Sector 3, Dwarka New Delhi, NSIT, Sector-3, Dwarka New
India priyabisht.bisht9@gmail.com Delhi, India bmjyoti@gmail.com

Abstract​— The Hybrid electric vehicle (HEV) is getting efficiency, less costs in power. By controlling the servo
more attention now a days due to limited conventional motor, the speed of hybrid electric vehicle can be controlled.
energy resources and environment issues. The limited The different control techniques such as Linear quadratic
range of battery power of electrical vehicle (EV) regulator (LQR) controller, state feedback controller (SFC),
motivates the researchers to explore HEV which Observer based controller and fuzzy Logic Controller (FLC)
combines both electrical machines and Internal have been explored by various research groups for speed
combustion (IC) engine to produce power. The objective control of HEV. The traditional control methods utilized in
of this research work is speed control Hybrid Electric HEV are found unsatisfactory due to indefinite input output
vehicle (HEV) by controlling the throttle position of the relationship and unspecified external distractions. To deal
motor. Moreover, the designed controller needs to give with such problems in the near past, different kind of
smooth throttle movement with minimized the steady controllers such as fuzzy logic controllers (FLC) has been
state error. To fulfill the desired objective two control recommended.
schemes has been employed. To optimize the The performance of FLC is often limited by un availability
parameters of the controller, Grey Wolf Optimization of systematic procedure to make rule base and membership
(GWO) and Particle Swarm Optimization (PSO) selection of input and output. The response of controller
technique are used. The comparative analysis of also depends on values of controller parameters. Many
response using different control schemes has been authors suggested different optimization technique to obtain
carried out. The observed results demonstrate the the value of controller parameters [1][2]
superiority of proposed scheme in terms of lessen fuel The objective designing [10] the research is to give the
utilization of the HEV queue and improving traffic smooth throttle movement, to reduce the steady state error
smoothness. and to regularize a Selected Vehicle speed (SV). With the
help of MATLAB software HEV model is simulated. In the
Keywords— Electric Vehicle, Fuzzy logic controller, PID present work Conventional controller PID and intelligent
controller, Particle swarm optimization, Greywolf optimization. Fuzzy-PID controllers are applied to control speed of HEV
[3][4][5]. The comparative analysis of results using various
I. I​NTRODUCTION control schemes has been done.
As vehicle numbers increase, vehicle internal combustion
emissions also increase, which in turn increases the
The following sections provide overview about the Hybrid
greenhouse effect. So, the concern for the environment has
Electric Vehicle (HEV) dynamics. The observability and
been increased exponentially and the use of electric vehicles
controllability analysis has been done in the section 3. The
has become so needy. So, hybrid electric vehicles (EV) are
section 4 gives information about various control schemes
getting more attention because it has no emission which will
and optimization techniques. The results and discussion are
help to reduce global warming [1].
explained in the section 5. The section 6 concludes the
As the electric power is generated at remotely located
research work.
plants, it’s easier to regulate the emission than those
emerging from IC engine vehicles. Even the provided
electric power used by Electric vehicle batteries (EV) can be II. HYBRID ELECTRIC VEHICLE SYSTEM
created by more environmentally sustainable
A. Modelling of the Electric vehicle
non-conventional renewable energy sources. In addition,
the EV has not had pollutants and it therefore does not Electronic throttle engine’s schematic diagram is shown in
cause any contamination to the environment. The limited fig.1. in which different controller have been used to control
range of battery power range of EV’s motivates the the DC servo motor [2].
automobiles industry experts to shift to other alternatives
like HEV. The HEV uses both electric machine and IC The various dynamics of this electric vehicle are represented
propellers to distribute the propulsion power [2]. as:
The various research groups are exploring low costly,
Compact sized EVs, high range/high performance

978-1-7281-9785-2/20/$31.00 ©2020 IEEE


115

Authorized licensed use limited to: Tsinghua University. Downloaded on December 18,2020 at 17:47:10 UTC from IEEE Xplore. Restrictions apply.
B. Modelling of Electric Vehicle
m v = F e (θ) − αν 2 − F g (1) Transfer function obtained from the state space variable A,
t
B, C and D to check the feedback of the system and to apply
Fe
τc t =− F e (θ) + F e1 (θ) (2) various controllers on the same is modelled in equation (8).
V (s) 5
8.29*10
F e1 (θ) = F 1 + γ √θ (3) θ(s)
= s2 +5s
(8)

Open Loop Stability Analysis.


The fig.2 shows the open loop response of HEV in absence
F e => Engine Force, throttle position function, of any controller. It clearly indicates that the open loop
F g => Gravitaional Force, A road grade function’, system is unstable.
θ => throttle position, v->speed of the vehicle. From the system equation , we have proved below that the
ASSUMPTION system is controllable and observable.
1. Gravitational force (F​g​) is 30% of vehicle’s weight. Characteristic equations of system A in equation (4) are
2. Time constant for hybrid vehicle’s engine normally is given by
in range [0.1, 1] sec., In the present work research we have [λI − A] = 0 (9)
taken 0.2 s. On solving above relation with equation 4 with MATLAB,
the eigenvalues for the Open-Loop system are
λ1 = 0 and λ2 =− 5.

Fig. 1​. Complete Simulink Model of the Electric


Vehicle System

Table 1.​ Numerical values for the parameters.


Fig. 2.​ Open Loop Response of the Electric Vehicle System
Constant Notation Value​e
Mass​s​of hybrid m 1000kg III. O​BSERVABILITY​ A​ND​ C​ONTROLLABILITY​ A​NALYSIS
vehicle The state eqn. for the LTI system is represented by
Coefcient for Alpha(α) 4N/(m/s)​2
drag ẋ (t) = Ax (t) + B u(t) . (10)
Coefcient​i​for Gamma( γ ) 12500​i​N
Here, A represents n x n system matrix, B represents n x r
Engine​i​Force
input or control matrix and u represents the r x 1
Engine​i​Idle F​i 6400​i​N input-vector matrix. To find the controllability, one of the
Force method for checking rank of the matrix M= [B AB A​2​B….
TimeConstant of τe 0.2​i​second A​n-1​B]. If the matrix rank M is equal to order of M then we
engine can say that the system is controllable in equation (4).
M = (1 * 109 ) * [0 0.0008 0.83 − 4.145 ]
We can see clearly that the rank of matrix M is equal to the
State space Equations for system equations are shown below order of M, so it can be deduced that the system is
controllable.
A = [0 0.001 0 − 5 ] (4) In the same manner, to find the observability of the system,
one of the method for checking the matrix rank N=[C CA
CA​2​….C A​n-1​]. If the matrix rank of N is n then we can say
B = 0 8.29 * 10
8 (5) that the system is observable in equation (4).

N = [1 0 0 0.001 ]
C = [1 0] (6)
We can clearly see that rank of matrix N is equal to that of
A and also equal to the order of N, hence we can say tha the
D = [0] (7) system is observable. Now, as we have proved that the
system is controllable hence we can use controller to control
the system so that it can become stable system. In the next

116

Authorized licensed use limited to: Tsinghua University. Downloaded on December 18,2020 at 17:47:10 UTC from IEEE Xplore. Restrictions apply.
section we will analyse the performance of the different
types of controllers and will compare the results. As we have designed the Fuzzy PID controller, optimization
can be used to find the better values of parameters of the
IV. CONTROLLER controller which can be very helpful in speed control of
A. PID Controller HEV. In this research, we have optimized the parameters
using the technique called Particle swarm optimization
In PID controller, the control actions form part of derivative,
(PSO) and GreyWolf Optimization(GWO) technique and
integral and proportional actions. The proportional part
compared the results of both.
helps to reduce the error, integral part improves the steady
state by integrating that error and the derivative section C. PARTICLE SWARM OPTIMISATION (PSO)
improves the transient performance of the system by It is computation method developed by James Kennedy &
deriving that error thereby improvising the transient Russell Eberhart. Engineering contains complex
performance [5]. e(t) Error signal distinguishes between mathematical problems and the concept of PSO is capable
rated speed versus output speed. of solving these existing problems [6]. It shows basic
evolutionary computation characteristics incorporating
e(t) = V ref (t) − V (t) (11) initialization with a population of random solutions and
looking optima by updating generations.
u(t) in time domain is the output of the controller where PSO, a population-based optimization algorithm, which is
pulling the attention of the embedded computing community
o K p is the proportional gain, K d is derivative gain
[7]. It is a proficient tool for some continuous multimodal
and K I is integral gain and multi-dimensional problem classes.
t
de (t) Particle swarm optimization Algorithm:
u (t) = K p * e (t) + K d * dt
+ K I ∫ e (t) * dτ (12)
0 Let us consider every possible solution as a bird which can
be called as particle.
B. FUZZY Controller Fitness values are defined for every particle.
We have checked the fuzzy logic controller (FLC) Particles with best performance are updated.
performance for our model of a vehicle. The speed is being Similarly, create a group of particles with best performance
maintained constant despite of changes in road grade, and update it.
resistance due to wind or any other kind of variable. Figure Then the velocities of particles are updated by observing the
2 displays the vehicle block diagram with [4]. best performance of the group and single particle.
Four components to design the FLC are used namely Rule In general, the particle swarm optimization algorithm is
base, fuzzication, inference​ ​engine and defuzzication. composed of the two formula [7] i.e.
● formula for updating position and
● formula for updating velocity

Velocity updating formula:

v tin+1 = v tin + c1 * r1 * (ptbest − v t+1 − xtin) + c2 * r2 * (g tbest − xtin)


(13)
Fig.3.​ Membership Functions of input/output
Position updating formula:
Table 2.​ Rate of change of error ( de )
dt , error (e) and FLC Output
Rule Base
xtin+1 = xtin + v tin+1 (14)
​e
de
PB P Z N NB The formula for updating the velocity of particles had
dt
changed to equation given below, when Eberhart and Shi
added inertial weight ​ω into the main velocity formula, and
PB PB PB PB P Z
the position updating formula will remain unchanged:
P PB PB P Z N v (t+1)in = ω * v tin + c1 * r1 * (ptbest − v t+1 − xtin) + c2 * r2 * (g tbest − xtin)
(15)
N P Z N NB NB
V. GREY WOLF OPTIMIZATION
Z PB P Z N NB
Over a period of time, meta-heuristic technique of
optimizing has become very popular. One of them is
NB Z N NB NB NB
Genetic Algorithm (GA) [8]. Now the question raised was
why meta-heuristics have turn out to be so usual. The

117

Authorized licensed use limited to: Tsinghua University. Downloaded on December 18,2020 at 17:47:10 UTC from IEEE Xplore. Restrictions apply.
→ → → → → → → → → → → →
answer found was divided mainly into four reasons: local W 1 = ||W α (t) − P 1.Dα|| W 2 = ||W β (t) − P 2.Dβ || W 3 = ||W δ (t) − P 3.Dδ||
optima avoidance, simplicity, derivation-free mechanism, (21)
and flexibility [8]. Keeping these four reasons in mind, we → → →
→ W +W 2 +W 3
have used Grey Wolf Optimization (GWO) algorithm. This W (t + 1) = 1 3
optimization technique is based on the Grey Wolfs. (22)
GWO algorithm imitates the control order or hierarchy as
well as the pattern to hunt by grey wolves in nature. D. Attacking the prey (exploitation):
For simulating the leadership hierarchy, four types of Grey Whenever the prey becomes still, the grey wolves complete
wolves are employed which are alpha(α), beta(β), omega( ω their hunting by attacking that prey. To design the approach
) and delta( δ ) [9]. precisely, we reduce the p→ value. The varying array of P


A. Social hierarchy: is reduced by p→ [5] i.e., P is a arbitrary value in the range
To design mathematical model, we consider: [-p, p] where p is declined from 2 to 0 in the path of
iterations. The next position of a search agent is any position
● Alpha (α)- fittest ranking b/w its present position and the location of the pray when

● Beta(β)-second best ranking random values of P are in [-1,1].
● Delta( δ )- third best ranking
● Omega ( ω )- Lowest ranking grey wolf. E. Objective Function of GWO
For optimization, the objective function has been taken as
B. Encircling Prey the performance index for simulation period of maximum
During a hunt, these wolves circles around the prey. time of 25 seconds by using the sum of the integral of
Mathematically, it can be expressed as [8]: square error (ISE) of the system.
VI. RESULTS AND DISCUSSION
→ → → →
D = ||V · W P (t) − W (t)|| (16)
The HEV model and controller are simulated with the help
→ → → of Matlab software. The Fig.2 shows the open loop of the
(t + 1) = ||W P (t) − P .D|| (17) system. It is observed that the system is completely unstable
and. In the presnt work PID and FuzzyPID controller are
used in the system to speed control of HEV. Also to
→ →
Here ‘t’ is the present iteration, P and V are coeff. vectors, optimise the controller’s parameters to improve the
→ → transient and steady state response we have used the GWO
W P is the positional vector of pray, and W is positional technique and compared the results with the PSO technique.
vector of the GWO. T ss
→ →
P as well as V are calculated as given below: Objective function = ∫ e2 (t) .dt (23​)
0

→ →
P = 2p.q →
1−p (18)

V = 2.q→2 (19)

where components of p are decreasing from [2, 0] along the
path of iterations. The vectors q→1, q→2 are random vectors in
range [0,1].
C. Hunting ​Fig.4 ​Convergence curve for the GWO using the cost
Hold the skill to identify the place of the prey and make a function used in equation used above
circle around them. Generally, the hunt is directed by alpha.
The hunting can also be joined by beta and delta now and Table 3.​ Controller parameters
then [8]. However, we have no idea about the position of the
optimum value in a concrete existence search space. CONTROLLER K​P K​I K​D K​PI K​E
In order to calculate the ​hunting actions mathematically of Parameters
grey wolves, we presume that the ‘α’, ‘β’ and ′δ′ contains PID 0.6 0.73 0.09 - -
desirable information regarding the position of the preys.
Hence, we keep the better solutions acquired till now and PID WITH PSO 0.41 1.05 0.02 - -
update the position of more greywolves by using ‘α’ and ‘β’ PID WITH 0.29 0.85 0.06 - -
position [4].
GWO
→ → → →
Dα = ||V 1 · W α − W || ,
→ → → →
Dβ = ||V 2 · W β − W || , FUZZY PID 0.62 1.21 0.06 0.18 0.5
→ → → →
Dδ = ||V 3 · W δ − W || (20) WITH GWO
Fig.5 and table.4 shows the comparative analysis of all
controllers. fig.6, shows the output of PID controller with

118

Authorized licensed use limited to: Tsinghua University. Downloaded on December 18,2020 at 17:47:10 UTC from IEEE Xplore. Restrictions apply.
PSO, we have compared it with the basic PID controller and
the results are not that much satisfied. After this we have
used GWO technique with PID controller, the result of the
controller is shown in fig. 7. The results of this has
improved to an extent.

Fig.8 shows the output of FuzzyPID with GWO controller


and compared with the PID controller. On comparing the
results, we can clearly observe that the results are much
improved and the rise time, settling time, peak overshoot,
undershoot has been improved to a great level.

Table 4. Comparative analysis of response parameters using Fig. 7​ ​System Output using Fuzzy-PID Controller tuned
various controllers with PSO
Response
paramet PID PID FUZZY PID FUZZY
ers W PID W PI
IT WI I D
H TH T W
PS PS H IT
O O G H
W G
O W
O
RISE TIME 3.5 4.3 3.85 4.2 3.4
(s )
SETTLING 0.8 0.78 0.57 0.60 0.45
TIME(s)
OVERSHOO [0.73 to [0.75 to [0.75 to 0.75 to [.77 to
T 1.8 1. 1.93 1. 1.
RANGE 6] 61 ] 8] 9]
(%) ] Fig. 8​ System Output using Fuzzy-PID Controller
UNDERSHOO [0.55 to [1.63 to [1.71 to [1.71 to 1.93 to tuned with GWO
T 0.4 0. 0.18 0. 0.
RANGE 7] 36 ] 26 13
(%) ]] ] ] Table 5.​ Comparison with benchmark methods
Response FUZZY PID Literature [1]
STEADY 0.15% 0.30% 0.10% 0.15% .05% Parameters WITH
STATE
ERROR GWO
Rise Time (s) .4 N/A
Settling Time (s) 0.45 0.50
Overshoot [0.77 to 1.9] 0.6%
Range i.e. .59%
Undershoot Range [1.93 to 0.13] N/A
Steady State Error 0.05% 0.1%

VII. CONCLUSION
Optimal control strategy for speed control of HEV is
evaluated in the present research. The systems open loop
output is unstable. PID and Fuzzy-PID controllers are
designed for speed control of HEV. Response of controller
is dependendent on the value of controller parameters.
Therefore GWO and PSO techniques are used in order to
Fig. 5. Graphical analysis of response parameters using various optimize the parameters of controllers. The comparative
controllers analysis of the controllers performance has been done. It is
observed from the results transient as well steady state
response is improved using GWO tuned Fuzzy controller
which shows the superiority of the GWO tuned Fuzzy-PID
as compared to other designed controllers. .
R​EFERENCES
[1] A.K. Yadav, Prerna Gaur, S.K. Jha., J.R.P. Gupta, and A.P. Mittal,
“Optimal Speed Control Of Hybrid Electric Vehicles,” Journal Of
Power Electronics, Vol. 11, No.4, 2011.
[2] I. Husain “ Electric and Hybrid Vehicles,” Design Fundamentals, 2nd
edition, CRC Press, Taylor and Francis Group, New York, Chap. 1,
Fig. 6.​ ​System output using PID Controller tuned with 2011.
PSO [3] R. Mayr, and O. Bauer, “Safety issues in intelligent cruise control,”
inProc. IEEEIntelligent Transportation Systems, pp. 970-975,1999.

119

Authorized licensed use limited to: Tsinghua University. Downloaded on December 18,2020 at 17:47:10 UTC from IEEE Xplore. Restrictions apply.
[4] K.T.Oner ,E. Cetinsoy , E.Sirimoglu, C. Hancer, T.Ayken , and M. [8] Z.M. Gao, and Juan Zhao, “An Improved Grey Wolf Optimization
Unel, “LQR and SMC stabilization of a new unmanned aerial Algorithm with Variable Weights. Computational Intelligence and
vehicle,” World Academy of Science, Engineering and Technology Neuroscience,” Article ID 2981282, 2019.
58, 2009 [9] S. Cheng, Q. Qin, J.Chen, and Y. Shi, “ Brain storm optimization
[5] K.H. Ang , G. Chong , Yun Li, “PID Control System Analysis, algorithm,” a review. Artif Intell Rev 46(4):445–458, 2016.
Design, and Technology,” IEEE Transactions on Control Systems [10] S.K. Katsikas., D.T. Tsahalis , D.A. Manolas, and Xanthakis A
Technology, Vol. 13, Issue 4, 2005. Genetic Algorithm for Operation Optimization of an Industrial
[6] Song Mei-Ping, and Gu Guo-Chang, “Research on Particle Swarm Cogeneration System. Computers Chem. Engineering, Vol. 20,
Optimization,” A Review. inProc IEEE Conference of Machine Suppl., pp. S1107-S1112,1996.
Learning and Cybernetics, 2005.
[7] E.G. Gonzalo, and J.L.Fernandez-Martinez, “A Brief Historical
Review of Particle Swarm Optimization(PSO),”. Journal of
Bioinformatics and Intelligent Control, Vol. 1, 13-1, 2012.

120

Authorized licensed use limited to: Tsinghua University. Downloaded on December 18,2020 at 17:47:10 UTC from IEEE Xplore. Restrictions apply.

You might also like