Вы находитесь на странице: 1из 6

Design of Robust PID Controller With Disturbance Rejection For Motor

Using Immune Algorithm


Dong Hwa Kim, Jae Hoon Cho
Dept. of Instrumentation and Control Eng., Hanbat National University,
16-1 San Duckmyong-Dong Yuseong-Gu, Daejon City, Korea, 305-719.
E-mail: kimdh@hanbat.ac.kr, Hompage: ial.hanbat.ac.kr
Tel: +82-42-821-1170, Fax: +82-821-1164

Abstract To improve the performance of PID controller tuning for


In this paper, design approach of PID controller with rejection processes with changing dynamic properties, several tuning
function against external disturbance in motor control system is strategies for PID controller have been proposed, for
proposed using immune algorithm. Up to the present time, PID example, automatic tuning PID, adaptive PID, and intelligent
Controller has been used to operate for AC motor drive because
controller. These controllers have recalibration features to
cope with little a priori knowledge and significant changes in
of its implementational advantages in practice and simple
the process dynamics [4].
structure. However, it is not easy to achieve an optimal PID gain
However, the PID controller parameters are still computed
with no experience, since the gain of the PID controller has to be
using the classic tuning formulae and, as noted above, these
manually tuned by trial and error in the industrial world. To
do not provide good control performance in all situations, for
design disturbance rejection function, disturbance rejection example, for unstable systems with time delay. De Paor used
conditions based on H f are illustrated and to decide the the Modified Smith Predictors to cope with unstable and
performance of response for the designed PID controller, an integrating process with long time delay.
ITSE (Integral of time weighted squared error) is used. Hence, In order to provide consistent, reliable, safe and optimal
parameters of PID controller are selected by immune algorithm to solution to industrial control problems as described above,
obtain the required response. many approaches for PID control schemes and tuning
techniques have been presented. These schemes generally
Key Words: PID control; Disturbance control; Immune consist of four basic parts: model estimation, desired system
algorithm, PID tuning; Motor control. specifications, optimal tuning mechanism and an online PID
controller. The model estimation provides a
parametric/non-parametric model for the process. The
1. Introduction desired system specifications include a set of control
requirements of the process [6].
A Proportional Integral Derivative (PID) controller has In this paper, an intelligent tuning method by an immune
been widely used in the most industrial processes despite algorithm is suggested for robust control with disturbance
continual advances in control theory. This is not only due to rejection function on control system of motor control loop.
the simple structure which is theoretically easy to understand
but also to the fact that the tuning technique provides 2. PID Controller Tuning With Disturbance
adequate performance in the vast majority of applications. Rejection Function By Immune Algorithms
However, it cannot effectively control such a complicated or
fast running system such as motor control system, since the A. Condition for Disturbance Rejection
response of a plant depends on only the three parameters (P,
I, and D) and its gain has to be manually tuned by trial and In Fig. 1, the disturbance rejection constraint can be given
error in the industrial world. Most of the PID tuning rules by
developed in the past years use the conventional method
such as frequency-response methods [1]. This method needs Y w( s )
a highly technical experience to apply since they provide max G . (1)
d ( t )D d 1  K ( s, c)G ( s ) f
simple tuning formulae to determine the PID controller
parameters. In case of the Ziegler-Nichols rule tuning
technique, it often leads to a rather oscillatory response to Here, G 1 is constant defining by the desired rejection
set-point changes as the following system [2]:
level and x denotes the H f -norm, which is defined as
(a) Plant which the system has non-linearities such as f
directionally dependent actuator and plant dynamics
(b) System that various uncertainties, such as modeling error G ( s) max G ( jZ ) . (2)
and external disturbances, are involved in the system. f Z[ 0 ,f )
As a result of these difficulties, the PID controllers are rarely
tuned optimally and the engineers need to research for a The disturbance rejection constraint becomes
highly tuning technology.

Proceedings of the Fourth International Conference on Hybrid Intelligent Systems (HIS04)


0-7695-2291-2/04 $ 20.00 IEEE
The optimal tuning proposed in this paper is to find the
d(s) vector c, such that the ITSE performance index, PI (c) is a
minimum using immune algorithm and the constraint
max V (Z , c) G
0.5
is satisfied through real coded
w Z[ 0.f )
immune algorithms.
E(s) Y(s)
R(s) + + -
K(s,c) G(s) C. Immune Algorithm
- In Fig. 2, when an antibody on the surface of a B cell
binds an antigen, that B cell becomes stimulated. The level
of stimulation depends not only on how well the B cells
antibody matches the antigen, but also how it matches other
Fig. 1. Control system with disturbance. B cells in the immune network [8].

w( s )
1  K ( s, c)G ( s ) f
0.5 K
w( jZ ) w( jZ )
max
Z[ 0.f )
1  K ( j Z , c )G ( j Z , c ) K (  j Z , c )G (  j Z , c )
max V (Z , c) 0.5 (3) 1K
Z[ 0.f )
The controller K (s, c) is written as

c2
K ( s, c) c1   c3 s (4) Fig. 2. Dynamic relationship between cells,
s
The vector c of the controller parameter is given by Antibody is actually three-dimensional Y shaped
molecules which consist of two types of protein chain. It also
c >c1 , c2 , c3 @T (5) possesses two paratopes which use to match the antigen. The
stimulation level of the B cell also depends on its affinity
with other B cells in the immune network. This network is
Hence, the condition for disturbance rejection is given as formed by B cells possessing an affinity to other B cells in
max V (Z , c) G .
0.5
the system.
Z[ 0.f )
If the stimulation level rises above a given threshold, the
B cell becomes enlarged and if the stimulation level falls
A. Performance Index for Optimal Controller Design below a given threshold, the B cell die off.
The more neighbours a B cell has an affinity with, the more
The performance index defined as ITSE (Integral of the stimulation it will receive from the network, and vice vers.
Time-Weighted Square of the Error) is written by Against the antigen, the level to which a B cell is
stimulated relates partly to how well its antibody binds the
f
PI t ( E (t )) 2 dt , (6a) antigen. We take into account both the strength of the match
0 between the antibody and the antigen and the B cell objects
m
affinity to the other B cells as well as its enmity. Therefore,
B( s)
b s
j 0
j
m 1
generally the concentration of i-th antibody, which is
E (s) n
. (6b) denoted by G i , is calculated as follows [7-8]:
A( s )
a s i
n 1

dS i (t ) N N
D mik G k (t )  E mi  J i G i (t ) ,
i 0
dt m jiG j (t )  D
j 1 k 1
Because E(s) contains the parameters of the controller (c)
and plant, the value of performance index, PI for a system of (8a)
nth order can be minimized by adjusting the vector c as dG i (t ) 1
, (8b)
follows [7]: dt dS i (t )
1  exp 0.5 
dt
min PI (c)
c
(7) where in Eq. (8), N is the number of antibodies, and D and

Proceedings of the Fourth International Conference on Hybrid Intelligent Systems (HIS04)


0-7695-2291-2/04 $ 20.00 IEEE
E are positive constants. m ji denotes affinities between
where n denotes the population size of immune network. The
antibody j and antibody i (i.e. the degree of interaction), penalty function )(ci ) is discussed in the following.
mi represents affinities between the detected antigens and Let the disturbance rejection constraint be
antibody i, respectively. max(D (Z , ci )) 0.5 . The value of the fitness of each
D. Evaluation Method for Disturbance Rejection Based on individual of memory cell Z j ( j 1,..., m) is determined by
Immune Algorithms the evaluation function, denoted by :(Z j ) as

In this paper, for the constrained optimization tuning,


immune algorithms are considered, i.e., memory cell of :(Z j ) D (Z , ci ) , (10)
immune algorithm to minimize the performance index
PI (c) , and network of immune algorithm to maximize the where m denotes the population size of memory cell.
disturbance rejection constraint D (Z , c) , as depicted in Fig. The penalty for the individual ci is calculated by means
4. Initially, memory cell is started with the controller of the penalty function )(ci ) given by
parameters within the search domain as specified by the
designer. These parameters are transferred then to network,
which is initialized with the variable frequency Z [11]. M 2 if ci is unstable

Immune network maximizes the disturbance rejection )(ci )M 1 max(Z , ci ) if max(D (Z , ci )) 0.5 G (11)
constraint during a fixed number of generations for each
individual of memory cell in immune network. Next, if the 0 if max(D (Z , ci )) 0.5 G .
maximum value will be associated to the corresponding
individual of memory cell. Individuals of memory cell that If the individual ci does not satisfy the stability test
satisfy the disturbance rejection constraint will not be
penalized. In the evaluation of the fitness function of applied to the characteristic equation of the system, then ci
memory, individuals with higher fitness values are selected is an unstable individual and it is penalized with a very large
automatically and those penalized will not survive the positive constant M 2 . Automatically, ci does not survive
evolutionary process.
For the implementation of the immune algorithm, this paper the evolutionary process. If ci satisfies the stability test,
used tournament selection, arithmetic crossover, and but not the disturbance rejection constraint, then it is an
mutation [10]. infeasible individual and is penalized with
M 1 max a(Z , ci ) , where n is a positive constant to be
- Representation
In the immune based representation, the parameters of the adjusted. Otherwise, the individual ci is feasible and is not
controller were coded in floating point and concatenated in penalized.
an individual of immune network. For memory cell of
immune network, an individual consists of only one gene E. Computational Procedure for Optimal Selection of
(frequency ). Parameter
c
The coding of an antibody in an immune network is very
max V (Z , c ) G
0 .5
min PI (c)
Z[ 0.f )
c important because a well designed antibody coding can
Immune max V (Z , c) G
0.5
increase the efficiency of the controller.
algorithm
Z[ 0.f )
Memory cell

C1 2 1 0.5 xxx 0.2 0.1


Fig. 3. Immune algorithm based computational
structure for optimal parameter selection. C2 2 1 0.5 xxx 0.2 0.12

- Fitness Function C3 2 1 0.5 xxx 0.2 0.1


An approach using penalty function [11] is employed to
solve the constrained optimization selection. Fig. 4. Allocation structure for optimal selection of P, I, D
Let the ITSE performance index be PI(c). Then the value gain in locus of antibody of immune algorithm.
of the fitness of each individual of immune network
ci (i 1,.......n) is determined by the evaluation function, As shown in Fig. 4, there are three types antibodies in this
paper: 1) antibody type 1 is encoded to represent only P (c1)
denoted by *(ci ) as
gain in the PID controller; 2) antibody type 2 is encoded to
represent I (c2) gain; 3) antibody is encoded to represent D
*1 (ci )  PI n (ci )  )(c i ) (9) (c3) gains. The value of the k locus of antibody type 1 shows

Proceedings of the Fourth International Conference on Hybrid Intelligent Systems (HIS04)


0-7695-2291-2/04 $ 20.00 IEEE
P gain allocated to route 1. That is, the value of the first concentration is calculated by affinity. So, V k is represented
locus of antibody type 1 means that P gain allocated to route by
1 is obtained by route 2 [9-10].
On the other hand, the n locus of antibody 2 represents I (c2) sum of antibodies with same affinity as mMk
gain for tuning of the PID controller with disturbance Vk = . (13)
rejection function. Here, the objective function can be sum of antibodies
written as follows. This algorithm is implemented by the
following procedures. Using equation (13), a immune system can control the
Given the plant with transfer function G (s ) controller with concentration and the variety of antibodies in the
fixed structure and transfer function C ( s, c) , and the lymphocyte population. If antibody obtains a higher affinity
against an antigen, the antibody stimulates. However, an
weighting function W (s ) , determine the error signal E (s)
excessive higher concentration of an antibody is suppressed.
and the disturbance rejection constraint a(Z , c) . Through this function, an immune system can maintain the
disturbance rejection constraint a(Z , c) . Specify the diversity of searching directions and a local minimum. That
lower and upper bounds of the controller parameters. Set up is, for each individual ci of the network, calculate
immune network and memory cell parameters: crossover max(a(Z , ci )) using memory cell, and initialize the gene of
probability, mutation provability, population size, and each individual Z j ( j 1,..., m) in the population.
maximum number of generations.
It is more convenient to describe the method in the form of [Step 6] Stimulation of Antibody: To capture the unknown
an algorithm. antigen, new lymphocytes are produced in the bone marrow
in place of the antibody eliminated in step 5. This procedure
[Step 1] Initialization and recognition of antigen: The can generate a diversity of antibodies by a genetic
immune system recognizes the invasion of an antigen, which reproduction operator such as mutation or crossover. These
corresponds to input data or disturbances in the optimization genetic operators are expected to be more efficient than the
problem. That is, initialize the populations of network generation of antibodies. If the maximum number of
ci (i 1,..., n) and memory cell Z j ( j 1,..., m) , and set the generations of memory cell is reached, stop and return the
generation number of network to g1 1 , where g1 denotes fitness of the best individual max(a(Z , ci )) to network;
the number of generations for network. otherwise, set g 2 g 2  1 and go to step 3.
[Step 2] Product of antibody from memory cell: The immune
system produces the antibodies that were effective to kill the 5. Simulations and Discussions
antigen in the past. This is implemented by recalling a past
successful solution from memory cell. The simulation results are shown as Fig. 7-10. Fig. 8
represents response to average values on parameter learning
For each individual ci of the network population, calculate
of immune network on parameters, Pm=0.35, Pc=0.1 to 0.5.
the maximum value of a(Z , ci ) using memory cell. If no Also, Fig. 9 illustrates response to average values on
individuals of the network satisfy the constraint parameter learning of immune network by Pm=0.5, Pc=0.42
max(a (Z , ci )) 0.5  G , then a feasible solution is assumed to to 0.47 and Fig. 10 is response to average values on
parameter learning of immune network of Pc=0.29, Pm=0.1
be nonexistent and the algorithm stops. In this case, a new to 0.5.
controller structure has to be assumed. On the other hand, Figs. 6-9 show response to disturbance
[Step 3] Calculation for searching an optimal solution: rejection depending parameter variation.
Calculate the fitness value for each individual ci of
network by using (9) and (10).
[Step 4] Differentiation of lymphocyte: The B - lymphocyte
cell, the antibody that matched the antigen, is dispersed to
the memory cells in order to respond to the next invasion
quickly. That is, select individuals using tournament
selection and apply genetic operators (crossover and
mutation) to the individuals of network.
[Step 5] Stimulation and suppression of antibody: The
expected value K k of the stimulation of the antibody is
given by
mMk
Kk (12)
Vk Fig. 5. Block diagram of Simulink for simulation.
where V k is the concentration of the antibodies. The

Proceedings of the Fourth International Conference on Hybrid Intelligent Systems (HIS04)


0-7695-2291-2/04 $ 20.00 IEEE
6. Conclusions 1.4
Step Response(disturbance : 0.1sin(t)

IA
GA

1.2

Up to now, the PID controller has been used to operate the 1

process loops including motor control. However, achieving y(t) 0.8

an optimal PID gain is very difficult for the control loop with 0.6

disturbances. Since the gain of the PID controller has to be 0.4

tuned manually by trial and error. Tuning of the PID 0.2

controller may not cover a plant with complex dynamics,


0

such as large dead time, inverse response, and a highly 0 5 10 15 20

time
25 30 35 40 45 50

nonlinear characteristic without any control experience. Fig. 8. Comparison of response between genetic algorithm
For tuning an optimal controller that can actually be operated and immune algorithm to step input. (Pm=0.02 0.04)
on the motor control loop, this paper focuses on tuning of
PID controller with disturbance rejection. For
Table 1. Parameter obtained by immune.
this purpose, we suggest an immune algorithm
based tuning method for the PID controller. Kp Td Ti a b c d e f
Parameters P, I, and D encoded in antibody are Pc=0.10 5.8454 1.6732 19.865 348.57 271.61 177.66 314.14 67.779 359.92
randomly allocated during selection processes Pc=0.15 7.4582 0.05142 26.412 358.43 266.33 243.17 314.14 67.779 359.92
Pc=0.20 16.201 0.02176 27.842 237.77 244.69 279.14 314.14 67.779 359.92
to obtain an optimal gain for plant. The object Pc=0.25 7.322 0.00958 18.72 300.68 258.62 224.93 314.14 67.779 359.92
function can be minimized by gain selection for Pc=0.30 21.339 20.489 33.092 185.22 250.58 180.56 302.75 102.6 306.81
Pc=0.35 5.7534 15.607 26.966 305.97 235.63 211.14 300.52 77.984 281.62
control, and the variety gain is obtained as Pc=0.40 16.026 22.067 27.995 211.55 283.94 198.09 268.08 97.794 128.67
shown in Table 1. The suggested controller can Pc=0.45 8.2225 4.9614 24.898 294.16 240.15 148.32 311.11 68.524 352.67
also be used effectively in the control system as Pc=0.50 6.9902 29.222 38.304 280.48 277.11 253.82 265.11 103.09 139.72
seen from Figs. 6 9.

Fig. 9. Performance comparison of disturbance


rejection by genetic algorithm and immune algorithm
Fig. 6. Comparison between genetic algorithm and immune
algorithm to step input. (Pc=0.30 0.45) (ISE criteria).

Step response
1.4
disturbance(sin0.1)(IA)
no disturbance(IA)
disturbance(sin0.1)(GA)
1.2 GA dis(sin0.1) no disturbance(GA)

GA no dis

IM dis(sin0.1)
0.8
IM no disturbance
)
(t
y
0.6

0.4

0.2

0
0 5 10 15 20 25 30 35 40 45 50
time

Fig. 10. Performance comparison of disturbance


rejection by genetic algorithm and immune algorithm
Fig. 7. Comparison between genetic algorithm and immune (ISE criteria).
algorithm to step input. (Pm=0.02 0.04)
References

[1] D. avid Lindsley, Boiler Control Systems, McGrawill,


1991.
[2] S. Matsummura, Adaptive control for the steam t

Proceedings of the Fourth International Conference on Hybrid Intelligent Systems (HIS04)


0-7695-2291-2/04 $ 20.00 IEEE
Table 2. Parameter obtained by immune .
` Kp Td A. Ti a b c d e f
Pc=0.28 7.5935 0.020934 32.17 235.34 162.8 157.75 314.14 67.779 359.92
Pc=0.29 11.086 17.803 29.326 218.24 301.42 198.53 298.1 143.31 226.88
Pc=0.30 21.339 20.489 33.092 185.22 250.58 180.56 302.75 102.6 306.81
Pc=0.31 10.028 21.443 26.728 167.39 303.41 131.84 358.29 275.84 318.73
Pc=0.32 6.9831 0.66115 30.341 153.25 279.98 103.38 358.37 275.84 318.73
Pc=0.33 12.288 12.176 24.555 258.26 253.67 278.87 276.81 63.038 294.76

emperature of thermal power plants, Proceedings the 1993


IEEE on Control applications, pp. 1105-1109, Sept.
1998.
[3] Wen Tan, Horacio, multivariable robust controller
design for a boiler system, IEEE Trans. On control
systems technology, Vol. 10, no. 5, pp. 735-742, Setp.
2002.
[4] P. Anderson, J. Doudna, Dynamic models for fossil
fueled steam units in power system studies, IEEE
Trans. On power systems. Vol. 6, no. 2, pp. 735-761,
May 1991.
[5] Chen-Kuo Weng, Asok Ray, Modeling of power plant
dynamics and uncertainties for robust control
synthesis, Appl. Math. Modeling, Vol. 20, pp. 501-512,
May, 1996.
[6] Chen-Huo Weng, Asok Ray, Robust wide range control
of steam electric power plants, IEEE Trans. On control
systems technology, vol. 5, no. 1, pp. 74-88, Jan. 1997.
[7] J. D. Farmer, N. H. Packard and A. S. Perelson, The
immune system, adaptation, and machine learning,
Physica. D, no. 22, pp. 187-204, 1986.
[8] Kazuyuki Mori and Makoto Tsukiyama, Immune
algorithm with searching diversity and its application to
resource allocation problem, Trans. JIEE, Vol. 113-C,
no. 10, 93.
[9] Dong Hwa Kim, Intelligent Tuning of the 2-DOF PID
Controller On the DCS for Steam Temperature Control
of Thermal Power Plant, IEEE Industrial Application
Society. I&CPS 2002, May 5-8, 2002, Savannah, GA,
USA.
[10] Dong Hwa Kim, Intelligent tuning of a PID controller
using an immune algorithm, Trans. KIEE, Vol. 51-D,
no.1, pp. 2002.
[11] Reato A. Krohling and Joost P. Rey, Design of
Optimal Disturbance Rejection PID Controllers Using
Genetic Algorithms, IEEE Trans. Evolutionary and
computation. Vol. 5, no. 1, Feb. 2001.
[12] Dong Hwa Kim, Parameter tuning of fuzzy neural
networks by immune algorithm, May 12-16, 2002,
2002 IEEE international conference on fuzzy systems,
Honolulu, Hawaii, 2002.
[13] Yong-Yan Cao, Stability analysis and synthesis of
nonlinear time-delay systems via linear Takagi-Sugeno
fuzzy models, Fuzzy sets and systems 124, pp.
213-229, 2001.

Proceedings of the Fourth International Conference on Hybrid Intelligent Systems (HIS04)


0-7695-2291-2/04 $ 20.00 IEEE

Вам также может понравиться