Sunteți pe pagina 1din 4

Proceedings of the Fourth International Conference on Machine Learning and Cybernetics, Guangzhou, 18-21 August 2005

OPTIMIZATION OF FRACTIONAL ORDER PID CONTROLLERS BASED ON


GENETIC ALGORITHMS
JUN-YI CAO, JIN LIANG, BING-GANG CAO

Department of Mechanical engineering, Xi’an Jiaotong University, Xi’an, 710049, China


E-MAIL: caojy@mailst.xjtu.edu.cn

Abstract: Fractional Order PI λ Dδ controllers are described by


The intelligent optimization method for designing fractional order differential equations. Expanding
Fractional Order PI λ Dδ controller (FOPID) based on derivatives and integrals to fractional orders can adjust
genetic algorithms is presented in this paper. Fractional control system’s frequency response directly and
calculus can provide new and good performance extension continuously. This great flexibility makes it possible to
for FOPID because of the arbitrary order of fractional design more robust control system. Letting control order be
calculus. However, the difficulties of designing FOPID
controllers increase. FOPID controllers’ parameters are
fractional, however, is not always straightforward. Several
composed of proportional constant, integral constant, methods have been reported for FOPID design. A method
derivative constant, derivative order and integral order, whose based on pole distribution of the characteristic equation in
design is more complex than that of conventional integer order complex plane was proposed [4]. B. M. Vinagre et.al
PID controller. The optimization design process based on presented a frequency domain approach based on the
genetic algorithms is analyzed in detail in this paper, in which excepted crossover frequency and phase margin [5]. A
the optimization performance target is the combination of state-space design method based on feedback poles
ITAE and control input, the numerical simulation of FOPID placement can be viewed in [6]. Doctor Ma Chengbin
controllers uses the method of Tustin operator and continuous provided a two-stage or hybrid approach: use conventional
fraction expansion. Simulation results show the proposed
method is highly effective.
(integer order) controller’s design method firstly and then
improve the performance of designed control system by
adding proper fractional order controller [7]. FOPID design
Keywords:
Fractional order controller; genetic algorithm; is a parameter optimization problem. A genetic algorithm is
optimization introduced to optimize FOPID parameters, and they consist
of proportional gain, integral gain, derivative gain,
1. Introduction derivative order and integral order.
This paper is organized as follows. Section 2 presents
Fractional calculus is a 300-year-old mathematical topic. the fractional Order PI λ Dδ controllers. Section 3 introduces
However, due to the unfamiliar idea of taking fractional the genetic algorithm. Section 4 presents the parameter
order, so few physical applications and limited optimization design process and simulation results. Some
computational power available at that time, fractional order conclusions are drawn in section 5.
calculus was not widely incorporated into control
engineering in past half century. In recent years, researchers 2. Fractional order PI λ Dδ controllers
reported that factional order differential equations could
model various materials more adequately than integer order Fractional order control systems are described by
ones. Especially, controllers making use of factional order fractional order differential equations. Fractional calculus
derivatives and integrals could achieve performance and allows the derivatives and integrals to be arbitrary order.
robustness results superior to those obtained with The FOPID controller is the expansion result of the
conventional (integer order) controllers [1][2][3]. Special conventional PID controller based on fractional calculus.
international symposiums and workshops organized by FOPID controllers design parameters have five and
ASME and IFAC were held to promote international improve the design flexibility.
exchange and cooperation in Fractional derivatives and
their applications research.

0-7803-9091-1/05/$20.00 ©2005 IEEE


5686
Proceedings of the Fourth International Conference on Machine Learning and Cybernetics, Guangzhou, 18-21 August 2005

2.1. Fractional calculus numbers. As shown in Figure 1, the fractional order


PI λ Dδ controller generalizes the conventional integer
There are several definitions of fractional derivatives order PID controller and expands it from point to plane.
[8]. Perhaps the best known is the Grunwald-Letnikov This expansion could provide much more flexibility in PID
definition. The m order fractional derivative of continuous control design.
function f (t ) is given by:
[ t −m ]
m − m h j m d m f (t )
a Dt f ( t ) = lim h ∑ ( −1)   f ( t − jh ) = (1)
h →0 j =0 j dt m
t −m  m
Where   is a truncation;   is binomial
 h  j
m  m  m ( m −1)...( m − j +1)
coefficients,   =1 ,   = , it can be
0  j j!
Figure 1. PID controllers with fractional orders
m Γ ( m +1)
replaced by Gamma function,   = .
j
  j ! Γ ( m − j +1) As shown in the Riemann-Liouville definition,
The general calculus operator (including fractional order fractional order systems have an infinite dimension. To
and integral order is defined as realize fractional order controllers perfectly, all the past
inputs should be memorized. Several proper approximation
 α α ℜ(α ) > 0
 d / dt methods by finite differential or difference equation were
Dα = 1
a t  ℜ(α ) = 0 (2) proposed in recent researches, such as Sampling Time
t −α ℜ(α ) < 0 scaling, Short memory principle, Tustin Taylor Expansion,
 a ( dτ )
∫ Lagrange function interpolation method.
The Laplace transform of the derivative of f (t ) is Because fractional order systems have an infinite
given by: dimension, the digital realization of FOPID keeps
somewhat difficulty. The above FOPID controller (6) can
{ }
L 0 Dtα f (t ) = sα F ( s ) −[ 0 Dtα −1 f (t )]t =0 (3) be approximated using different discrete methods, which is
given by:
Where F(s) is the Laplace transform of f (t ) . Gc ( z ) = k p + ki wi ( z ) + kd wd ( z ) (7)
The Laplace transform of the integral of f (t ) is given
Where wi ( z ) is the discrete approximation equation
by:
{ }
−λ
L 0 Dt−α f (t ) = s −α F ( s ) (4) of fractional order integral s , wd ( z ) is the discrete
δ
approximation equation of s . The higher the order of
2.2. Fractional order controllers approximation equation, the closer the discrete model is
approximate the real fractional order systems.
The differential equation of fractional order PI λ Dδ 3. Genetic algorithms
controller is described by:
−λ δ Genetic Algorithms (GA) is global optimizing ones,
u (t ) = K e (t ) + K D e (t ) + K D e (t )
p i t d t
(5)
based on natural selection and genetics mechanisms. They
The continuous transfer function of FOPID is obtained use a parallel procedure and structured strategy, but random,
through Laplace transform, which is given by: aiming to reinforce searching of high aptitude points. GA
−λ δ can be able to overcome complex non-linear optimization
G (s) = K + K s
c P i
+K s
d
(6)
tasks like non-convex problems, non-continuous objective
It is obvious that the FOPID controller not only need functions [9][10].
design three parameters K p K i and K d , but also design Generally speaking, GA consists of three
two orders λ,δ of Integral and Derivative controllers. fundamental operators: reproduction, crossover and
mutation. Given a optimization problem, GA encode the
The orders λ,δ are not necessarily integer, but any real parameters designed into finite bit strings, and then run

5687
Proceedings of the Fourth International Conference on Machine Learning and Cybernetics, Guangzhou, 18-21 August 2005

iteratively using the three operators in random way but 4.1. Representation of parameters
based on the fitness function evolution to perform the basic
tasks of copying strings, exchanging portions of strings as From the equation (6), five parameters K p K i K d
well as changing some bits of strings, and finally find and
decode the solutions to the problem from the last pool of and λ , δ are required to be designed, according to control
mature strings. objectives. Based on the scope of parameters to be
optimized, binary strings with different length are used to
3.1. Genetic operators describe them. This paper adopts ten bits string to represent
one parameter, so the GA operating objective consists of
Reproduction. Reproduction is a process by which the fifty bits string, as shown in Figure 2.
strings with larger fitness values can produce accordingly
with higher probabilities large numbers of their copies in
the new generation.
Crossover: crossover is a process by which the
1001000111 ………… 10010 …………1001000000
systematic information exchange between two strings is ……… ………
implemented using probabilistic decisions. Kp Ki Kd λ δ
Mutation: mutation is a process by which the chance 50 bits string
for the GA to reach the optimal point is reinforced through
just an occasional alteration of a value at a randomly Figure 2. Binary valued encoding of parameters
selected bit position. Mutation always allows small
probability. Suitable mutation probability can avoid earlier 4.2. Selection of Genetic Operators
maturation.
The Rank-based fitness assignment is selected in
3.2. Evaluation function fitness calculations of reproduction operator, because its
simplicity and robustness. Crossover uses the binary valued
In the every generation, each of the strings is decoded uniform crossover with probability 0.6. Mutation adopts
to be its corresponding actual parameter. This function binary valued mutation with mutation probability 0.01.
deals with the evaluation of each chromosome generated by
algorithms. For the present problem, the arm is to obtain 4.3. Fitness function
comprehensive control performance criterion, including
rising time, steady-state tracking error and overshoot. To evaluate the controller performance, there are
always several criterions of control quality:
3.3. Stop criteria ∞ ∞
J1 = ∫ e 2 (t )dt , J 2 = ∫ | e(t ) | dt (8)
0 0
The stop criteria used was the one that defines the J1 can track error quickly, but easily give rise to
maximum number of generations to be produced. When oscillation. J2 can obtain good response, but its selection
genetic algorithms, the new populations generating process performance is not good. For getting good dynamic
is finished, and the best solution to complete the generation performance and avoiding large control input, the following
number is the one among the individuals better adapted to control quality criterion is used in this paper.
the evaluation function. ∞
J = ∫ ( w1 | e(t ) | + w2u 2 (t ))dt (9)
0
4. Optimization of FOPID controllers The fitness function is: F=1/J.

Optimization of FOPID controllers firstly needs design 4.4. Simulation researches


the optimization goal, and then encode the parameters to be
searched. Genetic operator is running until the stop The following is control objective:
condition is satisfied. The decoding values of the last 400
chromosome are optimized parameters of the FOPID G ( s )= (10)
2
s +50 s
controller.
The sampling period is 0.001s. The input is step signal.
For reducing the time of optimization, The initial range of
parameters are selected, these are Kp:[0,20], Ki:[0,1],

5688
Proceedings of the Fourth International Conference on Machine Learning and Cybernetics, Guangzhou, 18-21 August 2005

Kd:[0,1], λ :[0,1] and δ :[0,1] . The fractional order 4. Conclusions


controller can be digitally realized using the method of
Tustin operator and continuous fraction expansion, where It has been demonstrated that the parameters
λ δ
sampling time is 0.001s; the approximation model order is optimization of fractional order PI D controller based
6. The population of initial generation is 50. w1=0.999; on genetic algorithms. According to optimization target, the
w2=0.001; the number of generation is 200. After 200 proposed method can search the best global solution for
generations of genetic operation, the best J is 32.0752, FOPID controllers’ parameters. Based on GA, the design
K p = 2.7566 , Ki =0.0029 , K d =1.000 and λ =0.7908,δ =0.4848 . and application of FOPID will be appeared in various
fields.
The following figures are the optimization results. The
Figure 3 is the variety of performance function J of
optimization process. The Figure 4 is the dynamic response References
after optimization.
[1] Petras I., Dorcak L. and Kostial I., “Control quality
enhancement by fractional order controllers”, Acta
44 Montanistica Slovaca, Kosice, Vol 3, No. 2, pp.
143-148, 1998.
42 [2] Chengbin Ma and Hori Y., “Backlash vibration
suppression in torsional system based on the fractional
40 order Q-filter of disturbance observer”, The 8th IEEE
International Workshop on Advanced Motion Control,
Best J

38 Kamasaki, Japan, pp.577 –582, 2004.


[3] Fonseca Ferreira N.M. and Tenreiro Machado J.A.,
36 “Fractional-Order Hybrid Control of Robotic Manipulators”,
Proceedings of the 11th International Conference on
Advanced Robotics, Coimbra, pp.393-398, June, 2003.
34 [4] Petras, I., “The fractional order controllers: methods for their
synthesis and application”, J. Electrical Engineering, Vol.50,
32 No.9-10, pp.284-288, 1999.
0 50 100 150 200 [5] B. M. Vinagre, Podlubny I., Dorcak L. and Feliu V., “On
number of generation Fractional PID Controllers: A Frequency Domain
Figure 3. Variety of performance function Approach”, Proceedings of IFAC Workshop on Digital
Control - PID’00, Terrassa, Spain, 2000
[6] Dorcak, L., Petras, I., Kostial, I. and Terpak, J., “State-space
1.4 controller design for the fractional-order regulated system”,
Proceedings of the ICCC'2001, Krynica, pp. 15 - 20. May,
1.2 2001.
[7] Chengbin Ma and Hori Y., “The application of fractional
1 order PID controller for Robust two-inertia speed control”,
Proceedings of the 4th International power electronics and
0.8 motion control conference, Xi’an, Aug. 2004.
[8]
yout

Podlubny I., “Geometric and Physical Interpretation of


0.6 fractional integration and fractional differentiation”,
Fractional calculus and Applied analysis, Vol 5, No. 4,
0.4 pp.367-386, 2002.
[9] Mitsukura, Y., Yamamoto, T., and Kaneda, M., “A design of
0.2 self-tuning PID controllers using a genetic algorithm”,
Proceedings of the American Control Conference, San
0 Diego, pp.1361-1365, June 1999.
0 0.1 0.2 0.3 0.4 [10] Paiyi Huang and Yungyaw Chen, “Design of PID
time(s) controller for precision positioning table using genetic
Figure 4. Dynamic response after optimization algorithms”, Proceedings of the 36th IEEE Conference
on Decision and Control, San Diego, pp. 2513 – 2514,
Dec.1997.

5689

S-ar putea să vă placă și