Anda di halaman 1dari 7

INTERNATIONAL JOURNAL OF COMPUTATIONAL COGNITION (HTTP://WWW.YANGSKY.COM/YANGIJCC.HTM), VOL. 3, NO.

4, DECEMBER 2005 73

Fuzzy Controller: Choosing an Appropriate and


Smallest Rule Set
Seema Chopra, R. Mitra and Vijay Kumar

Abstract— Design of fuzzy controller depends mainly on rule with membership functions for each input, then the number of
base and to derive the desired and small fuzzy rule base rules increases exponentially (assuming all possible rules are
is still based on trial and error. This paper concentrates on used, which is often the case). For example, in a two-input
choosing the rule base and gives some useful results. Firstly,
it has been shown that increasing the rules and number of fuzzy controller with 11 membership functions for each input,
membership functions beyond a certain limit is useless as it the possible rules are 112 =121, and if the number of inputs
increases the complexity of FLC and has almost no effect on are increased, this number will quickly increase. To overcome
output response on the system. Secondly, the authors investigate this problem, the user may want to put constraints on the type
how fuzzy subtractive clustering approach, modelled from a set of fuzzy controller (e.g., membership functions) or limit the
of input/output data; can be applied to the area of control theory
for linear and nonlinear dynamic systems. The objective of using rules. One of the most important problems while designing
this approach is to design a fuzzy logic controller with less fuzzy controller is to derive the desired fuzzy rule base. Trial
number of rules leading to a smaller amount of computational and error has been a natural choice to design fuzzy controller
time. Although the number of rules is automatically determined in this case. The selection of fuzzy if-then rules often relies
by this method, an user-specified parameter ra (the radius on a substantial amount of heuristic observations to express
of influence of cluster center) strongly affects the number of
rules that will be generated. Thirdly, a suggestion to choose proper strategy. Obviously it is difficult for human experts to
the value of ra is also given. To support the work, well-known examine all the input-output data from a complex system to
examples of linear and nonlinear systems are simulated under find the number of proper rules for a fuzzy system. To cope
the Matlab/Simulink environment. Both noisy and non-noisy data with this difficulty, much research effort has been devoted to
have been considered. Simulation shows the effectiveness of the develop alternative design methods.
proposed fuzzy logic controller as compared to existing fuzzy
logic controllers. Copyright ° c 2005 Yang’s Scientific Research Guillaume reviewed many fuzzy inference systems designed
Institute, LLC. All rights reserved. from data and concluded that the blind improvement of the
performance by a numerical index may conflict with the
Index Terms— Fuzzy control, fuzzy subtractive clustering,
extraction of small rule base, radius of cluster center. originality of fuzzy logic: its interpretability [1]. Instead he
suggested three conditions to make a set of induced rules: 1)
fuzzy sets can be interpreted as linguistic labels, 2) the set
I. I NTRODUCTION of rules must be as small as possible and 3) rules should be

W HILE modern control theory has made a modest inroad incomplete rules. Zadeh suggested linguistic analysis and em-
into practice, fuzzy logic control has been rapidly phasized significant statements for complex systems [2]. This
gaining popularity among practicing engineers. This increased requires that fuzzy rules should be expressed in linguistically
popularity can be attributed to the fact that fuzzy logic pro- meaningful forms and that the size of the rule base should
vides a powerful vehicle that allows engineers to incorporate be small enough to allow modelling of complex systems
human reasoning in the control algorithm. As opposed to the to be easily interpreted by experts. Tang-Kai [4] presents
modern control theory, fuzzy logic design is not based on the a characteristic-point-based fuzzy inference system (CPFIS)
mathematical model of the process. The controller designed for fuzzy modelling aimed to minimize the number of rules.
using fuzzy logic implements human reasoning that has been The aim of the CFPIS is not only a satisfactory and precise
programmed into fuzzy logic language (membership functions, performance, but also to employ as few purely linguistic fuzzy
rules and the rule interpretation). rules as possible by using a minimization-based systematic
When it comes to implementing a fuzzy controller, one often training method. Zhiqiang Gao, Thomas A. Trautzsch and
wants to try to minimize the amount of memory used and the James G. Dawson [3] concluded that Unlike some fuzzy
time that it takes to compute the fuzzy controller outputs given controllers with hundreds and even thousands, of rules running
some inputs. If the inputs to the fuzzy controller are increased on dedicated computer systems, a unique FLC using a small
number of rules and straightforward implementation may be
Manuscript received August 30, 2005; revised September 14, 2005. employed to solve a class of temperature control problems
Seema Chopra, R. Mitra and Vijay Kumar, Department of Electronics with unknown dynamics or variable time delays commonly
and Computer Engineering, Indian Institute of Technology, Roorkee, India.
Emails: seemadec@iitr.ernet.in(S. Chopra), rmtrafec@iitr.ernet.in(R. Mitra), found in industry.
vijecfec@iitr.ernet.in(V. Kumar) Based on the above research, we are motivated to design
Publisher Item Identifier S 1542-5908(05)10407-2/$20.00 a Fuzzy logic controller (FLC), which exhibits good perfor-
Copyright c
°2005 Yang’s Scientific Research Institute, LLC. All
rights reserved. The online version posted on September 15, 2005 at mance with smallest possible rule base. In a conventional
http://www.yangsky.us/ijcc/ijcc34.htm fuzzy logic controller, the number of rules is decided by an
74 INTERNATIONAL JOURNAL OF COMPUTATIONAL COGNITION (HTTP://WWW.YANGSKY.COM/YANGIJCC.HTM), VOL. 3, NO. 4, DECEMBER 2005

expert who is familiar with the system to be modelled. In our 3) A fuzzification interface, which converts controller in-
simulation, however, no expert is available and the number of puts into information that the inference mechanism can
membership functions (MF’s) assigned to each input variable easily use to activate and apply rules.
is chosen empirically, i.e., by examining the desired input- 4) A defuzzification interface, which converts the conclu-
output data [8]. Basically, there are two types of fuzzy logic sions of the inference mechanism in to actual inputs for
controllers. One is Mamdani type, the other is Takagi-Sugeno the process.
(TS) type [12,14]. Compared with Mamdani type FLC, TS For choosing the inputs and outputs of fuzzy controller,
type FLC can have similar performance with much less the controller is to be designed to automate how a human
number of rules. Unfortunately, the physical meaning of rule expert who is successful at this task would control the system.
consequent in TS type FLC is not as straightforward as that First, the expert tells us (the designers of the fuzzy controller)
in Mamdani type. The determination of the parameters for what information the user will use as inputs to the decision-
the rule consequents is extremely difficult, if not impossible. making process. Suppose that for the inverted pendulum [18],
Various TS type FLC design and optimization approaches have the expert says that the user will use
been proposed. To design TS type FLC, the number of fuzzy
values for each linguistic variable, the membership function d
e(t) = r(t) − y(t) and e(t) (1)
for each fuzzy value, the number of rules, the rule antecedents, dt
and the form of consequent function must be determined. as the variables on which decisions are based [10, 11].
In our previous research [13, 15 and 19] for the reduction Certainly, there are many other choices (e.g., the integral of the
of rules for fuzzy controllers we used Fuzzy curve, Fuzzy error e could also be used) but this choice makes good intuitive
subtracting clustering (FSC) and neural networks. It has also sense. Next the controlled variable of fuzzy controller is u(t)
been proved in [13] that after reducing the rules (using the but in this paper the incremental change in controller output
FSC approach) the performance of the reduced rule controller ∆u is taken as output. This will be helpful for common rule
is similar to the original system. Although the number of rules base for any type of fuzzy controller (may be fuzzy PI type,
is automatically determined by this method, the user-specified PD type or PID type controller). Once the fuzzy controller
parameter ra (the radius of influence of cluster center) strongly inputs and outputs are chosen, one must think about what
affects the number of rules that will be generated. A large ra are the membership functions (MFs) are for these input and
generally results in fewer rules, while a small ra can produce output variables. After surveying the literature, for a two input
excessive number of clusters and thus a large number of rules. fuzzy controller MFs are 3, 5, 7, 9 or 11 for each input are
Unfortunately, there is no standard procedure for choosing ra . mostly used. It is a general view that if the number of MFs
In this paper, a method to choose the value of ra is also between the defined ranges is larger, then the possible rules
suggested. increases and the response will be good. In this paper, all
This paper has two main contributions. Firstly, it has been membership functions for controller inputs (i.e., e and ∆e)
shown that by increasing the rules and number of membership and incremental change in controller output (i.e., ∆u) are
functions beyond a certain limit is useless. It only increases defined on the common normalized domain [−1, 1]. Here, the
the complexity of FLC and has no effect on output response 7 membership functions are shown in Fig. 2. Similarly all
on the system. Secondly, the rules are reduced using FSC other MFs when the number is 3, 5, 9 or 11 can be divided
approach with a constant ra (which is decided by a common into the normalized domain.
criterion in all cases) and the results are compared with
larger rule set. A comparison between the clustering based Fuzzy Controller
Fuzzy Logic Controllers (FLC) and conventional Fuzzy Logic
Controllers using simulation of a wide range of linear and Inference
Fuzzification

Defuzzification

Reference Input Output


nonlinear processes is presented. Simulation results of well signal r(t)
mechanism
u(t) y(t)
known examples of pendulum, coupled tank and robot arm Process
has also been tested. Due to limitation of space, all the results
are not included here. Rule-base

II. F UZZY L OGIC C ONTROLLER


A block diagram of a fuzzy control system is shown in Fig. 1. Block diagram of fuzzy controller.
Figure 1.The fuzzy controller is composed of the following
four elements:
1) A rule-base (a set of if-then rules), which contains
a fuzzy logic quantification of the expert’s linguistic
description of how to achieve good control. Now the next step is to design the rule base. The incremental
2) An inference mechanism (also called a fuzzy inference change in controller output ∆u for a fuzzy controller is
engine), which emulates the expert’s decision making in determined by the rules of the form:
interpreting and applying knowledge about how best to
control the plant. If e is E and ∆e is ∆E, then ∆u is ∆U .
CHOPRA, MITRA & KUMAR, FUZZY CONTROLLER 75

NB NM NS ZE PS PM PB
1 ∆e/e NB NM NS ZE PS PM PB
NM NB NB NB NM NS NS ZE
NM NB NM NM NM NS ZE PS
NS NB NM NS NS ZE PS PM
-1 -0.5 0 0.5 1 ZE NB NM NS ZE PS PM PB
Fig. 2. MFs for e, ∆e and ∆u. PS NM NS ZE PS PS PM PB
PM NS ZE PS PM PM PM PB
The set of rule bases for computing the output is shown in PB ZE PS PS PM PB PB PB
Fig. 3, which is a fairly standard one. If the number of MFs
for inputs is 7 the corresponding rules are 72 = 49. Similarly Fig. 5. Rule base 3. NB-Negative Big, NM-Negative Medium, NS-Negative
in all other cases the possible rules are 32 =9, 52 =25, 92 =81, Small, NVS-Negative Very Small, ZE-Zero Error, PS-Positive Small, PVS-
Positive Very Small, PM-Positive Medium, PB-Positive Big.
and 112 =121. These are an often used rule base designed with
a two-dimensional phase plane in mind where the FLC drives
the system into the so-called sliding mode. The rule base 1 ∆e/e NB NM NS NVS ZE PVS PS PM PB

for 9 rules, rule base 2 for 25 rules, rule base 3 for 49 rules NB NB NB NB NB NB NM NS NVS ZE

and rule base 4 for 81 rules is shown in Fig. 3 to Fig. 6. NM NB NB NB NB NM NS NVS ZE PVS
NS NB NB NB NM NS NVS ZE PVS PS
NVS NB NB NM NS NVS ZE PVS PS PM

∆e/e
ZE NB NM NS NVS ZE PVS PS PM PB
NB ZE PB PVS NM NS NVS ZE PVS PS PM PB PB
PS NS NVS ZE PVS PS PM PB PB PB

NB NB NB ZE PM NVS ZE PVS PS PM PB PB PB PB
PB ZE PVS PS PM PB PB PB PB PB

ZE NB ZE PB Fig. 6. Rule base 4. NB-Negative Big, NM-Negative Medium, NS-Negative


Small, NVS-Negative Very Small, ZE-Zero Error, PS-Positive Small, PVS-
Positive Very Small, PM-Positive Medium, PB-Positive Big.

PB ZE PB PB
Fig. 3. Rule base 1. NB-Negative Big, NM-Negative Medium, NS-Negative
Small, NVS-Negative Very Small, ZE-Zero Error, PS-Positive Small, PVS-
Positive Very Small, PM-Positive Medium, PB-Positive Big. III. F UZZY L OGIC C ONTROLLER U SING FSC A PPROACH
To identify the Fuzzy controller using FSC approach, some
data is needed, i.e., a set of two-dimensional input vectors
X = {X1 , X2 , . . . , Xn } where X = {e and ∆e} and the
associated set of one-dimensional output vectors as Y =
∆e/e NB NM ZE PM PB {Y1 , . . . , Yn } where Y = {u} is required. In the present work
this data has been generated by sampling input variables e
NB NB NB NB NM ZE and ∆e uniformly at the step size of 0.1 from the above
Fuzzy controllers, and computing the value of {u} for each
NM NB NB NM ZE PM sampled point. The Fuzzy toolbox of Matlab and programming
in Matlab has been used to generate the data points. In all
ZE NB NM ZE PM PB the above fuzzy controllers the membership functions for
controller inputs (i.e., e and ∆e) and incremental change in
PM NM ZE PM PB PB controller output (i.e., ∆u) are all defined on the common
normalized domain [-1, 1]. The number of data points thus
PB ZE PM PB PB PB generated in is 441 each case. The next task is to extract a
smaller set of rules using FSC approach to do the same.
Fig. 4. Rule base 2. NB-Negative Big, NM-Negative Medium, NS-Negative To extract rules from data, we first separate the data into
Small, NVS-Negative Very Small, ZE-Zero Error, PS-Positive Small, PVS-
Positive Very Small, PM-Positive Medium, PB-Positive Big. groups according to their respective classes. Here it is a
collection of 441 data points {x1 , x2 ,..., xn } in 3 dimensional
space (because X has 2 columns and Y has 1 column) .
Without loss of generality, it is assumed that the data points
have been normalized in each dimension so that they are
76 INTERNATIONAL JOURNAL OF COMPUTATIONAL COGNITION (HTTP://WWW.YANGSKY.COM/YANGIJCC.HTM), VOL. 3, NO. 4, DECEMBER 2005

bounded by a unit hypercube. Each data point is considered as of membership functions beyond 49 rules is futile. It increases
a possible cluster center and defines a measure of the potential the complexity of FLC and has no effect on output response
of data point xi as on the system. The results using 9 rules, 25 rules, 49 rules and
n
81 rules are presented in support of this. Secondly, the rules
X 2 are reduced to 8 rules using FSC approach. A comparison
Pi = e−αkxi −xj k (2)
j=1
between the clustering based Fuzzy Logic Controllers (FLC)
and conventional Fuzzy Logic Controllers using unit step input
where in both cases is presented.
4 In order to demonstrate the effectiveness and better com-
α= . (3)
ra2 parison of the presented approach, the computational time is
calculated using the Process Explorer —Sysinternals software.
k · k denotes the Euclidean distance, and ra is a positive
It has been shown clearly in the Table I that the value of
constant. Thus, the measure of the potential for a data point is
computational time reduces for each system in case of 8 rules
a function of its distances to all other data points. A data point
then 49 and 81 rules even using same simulation time.
with many neighboring data points will have a high potential
The general control scheme of the system comprising of a
value.
Fuzzy PI type Controller is shown in Fig. 7. The change in
The constant ra is effectively the radius defining a neigh-
error is defined as
borhood; data points outside this radius have little influence
on the potential. The parameter ra strongly affects the number
∆e(k) = e(k) − e(k − 1) (4)
of clusters that will be generated. Now to choose the radius
ra we always keep in mind that it has a value between 0 and where e(k) is the error at the kth sample. Here the input and
1 and it specifies the size of the cluster in each of the data output gains are Ge , G∆e and Gu . The operation of PI type
dimensions. Specifying a smaller cluster radius will usually FLC can be described by
yield more (smaller) clusters in the data. We suggest that the
radius of influence in the first data dimension is half the width u(k) = u(k − 1) + ∆u(k) (5)
of the data space (range of data) and the range of influence
in the second data dimension is one quarter the width of the In (5), ∆u is the incremental change in controller output,
data space and so on [13,15] is a good choice. When data is which is determined by the rules of the form
generated after sampling the inputs variables to calculate the
If e is E and ∆e is ∆E, then ∆u is ∆U .
output variable the range of output lies between -1 and 1 and
exact range depends on the shape and number of membership Here it is emphasized that this accumulation (5) of controller
functions. When e, ∆e and ∆u have 9 membership functions output takes place outside the FLC and is not reflected in the
then the output is in the range of −0.85 to 0.85. Hence the rules themselves. On the other hand, if the output of the FLC
value of ra for the entire 3 data dimension is [1 0.5 0.2125]. is u (not ∆u) and there is no accumulation of controller output
Similarly, we choose value in all other cases when the rules then Fig 7 is converted to a PD type FLC [9, 11].
are 121, 49 and 25.
After the potential of every data point has been computed, ∆eN
we select the data point with the highest potential as the first - ∆e
Z-1 G∆e ∆uN ∆u +
+ u
cluster center as discussed in [5]. The process of acquiring R + e e eN
Control
Gu
Rule-Base
new cluster center and revising potentials repeats until the Ge +
-
remaining potential of all data points falls below some fraction y Z-1
of the potential of the first cluster center P1∗ . Hence, after the
whole process eight clusters are found in each case. y Process

To extract the rules, firstly data is separated into groups


according to their respective classes. Subtractive clustering Fig. 7. Fuzzy PI type controller.
is then applied to the input space of each group of data
individually for identifying each class of data [6, 7, 20-22].
Hence the total rules are 8 in each case. Now it has been seen
The proposed FLC is applied to a PT2 system (inverted
that using subtractive clustering and by choosing the value of
pendulum) [18] which is given by the following differential
ra by the above mentioned method 81, 49, and 25 rules can be
equation:
converted into 8 rules. In the next section, the implementation
and results using the FSC approach and without clustering is 1 2D
presented. ẍ + ẋ + x = V y (6)
wo2 wo
where
IV. S IMULATION R ESULTS √
2√
wo = 50, D =
2, V = 1. (7)
The proposed controller has been tested on a variety of 5
processes. Two sets of simulation results are presented next. The PT2 system models the behavior of a two mass system
Firstly it has been shown that increasing the rules and number where y as an input variable and x as an output variable, for
CHOPRA, MITRA & KUMAR, FUZZY CONTROLLER 77

0.7
example, spring damper combinations or revolution controls
for electric motors. This system is controlled by FPDC in all
0.6
cases (using 9 rules, 25 rules, 49 rules and 81 rules) and results
are compared in Fig. 8.
0.5
It is observed that when the rules are increased from 49 rules
to 81 rules then the unit step response is almost same as using
0.4
49 rules. And when the rules are 9 and 25 then the response

Output
is distinctly different. Hence when rules are less (may be upto
25) it is desirable to increase the rules for a proper response 0.3

but after increasing the rules to a certain limit (49 rules in this
case) there is no need to go beyond since the response does 0.2
not change appreciably.
Now the next objective is to justify whether the Fuzzy 0.1
FPDC
controller with reduced number of rules (obtained by FSC) can HFPDC
provide the same level of performance as that of the system 0
0 0.5 1 1.5 2 2.5
with 49 rules and 81 rules. To demonstrate the effectiveness of Time (sec.)
the FSC approach, the results are reported for system with 49
rules and 81 rules and system with FSC minimized rule base Fig. 9. Unit step response of inverted pendulum with 81 rules (FPDC) and
(8 rules). In case of Fuzzy PI type and PD type controller, 8 rules (HFPDC).
the system with 49 rules and 81 rules is denoted by FPIC
and FPDC and system with 8 rules is denoted by HFPIC and
HFPDC. In this paper, it is emphasize that a reduced rule 1
base system is called satisfactory only with respect to the G2 (s) = .e−0.1S (8)
s+1
closeness of its step response to the large rule base system. In
Fig 9, it has been shown that the unit step response of inverted A 2% random noise is also added in the system in case of
pendulum system with 81 rules (FPDC) is very close to 8 rules FPDC system. Fuzzy controller with 49 and 81 rules again
(HFPDC). gives an identical performance in case of FPIC as shown in
Fig 10. The response characteristics of FPDC (with noisy data)
and FPIC (with non-noisy data) with 49 rules and 8 rules using
0.7 FSC approach are also shown in the Fig 11 and 12. Here again
the performance of 8 rules (with FSC approach) is close to the
0.6 original system (49 rules).

0.5 1.4

0.4 1.2
Output

0.3 1

0.2 0.8
Output

9 rules
0.1 25 rules 0.6
49 rules
81 rules
0 0.4
0 0.5 1 1.5 2 2.5
Time (sec.)
9 rules
0.2 25 rules
Fig. 8. Unit step response of inverted pendulum system with FPDC. 49 rules
81 rules
0
0 2 4 6 8 10 12
Time (sec.)

Fig. 10. Unit Step Response of G1 (s) = e−0.1s /(s + 1) with FPIC.
Finally, the proposed fuzzy logic controller is applied on a
well-known example of first order time delay process. This
form of transfer function is typically used to approximate
process control systems (8). The process plant is taken as first
order system with time delay and transfer function is given To summarize, firstly the response using 9, 25, 49 and 81
by: rules are shown and it has been proved that increasing the
78 INTERNATIONAL JOURNAL OF COMPUTATIONAL COGNITION (HTTP://WWW.YANGSKY.COM/YANGIJCC.HTM), VOL. 3, NO. 4, DECEMBER 2005

0.7 1.4

0.6 1.2

0.5 1

0.4 0.8
Output

Output
0.3 0.6

0.2 0.4

0.1 0.2
FPDC FPIC
HFPDC HFPIC
0 0
0 5 10 15 0 5 10 15 20 25
Time (sec.) Time (sec.)

Fig. 11. Unit step response of G1 (s) = e−0.1s/(s+1) including noisy data Fig. 12. Unit Step Response of G1 (s) = e−0.1s /(s + 1) without noisy
with FPDC (49 rules) and HFPDC (8 rules). data with FPIC (49 rules) and HFPIC (8 rules).

number of rules after a certain limit is useless. Secondly, data R EFERENCES


set is generated by the FLC with 49 and 81 rules by choosing
the value of ra as specified. Using the same data Fuzzy [1] S. Guillaume, “Designing fuzzy inference systems from data: an
interpretability-oriented review,” IEEE Trans. Fuzzy System, vol. 9, pp.
controller is designed using FSC approach and it gives 8 rules 426-443, June 2001.
in both cases. The overall performance of the proposed Fuzzy [2] L. A. Zadeh, “Outline of a new approach to the analysis of complex sys-
Controllers (with 8 rules) is compared with those of Fuzzy tems and decision processes,” IEEE Trans. System., Man, Cybernetics,
vol. SMC-3 pp. 28-24, Jan, 1973.
Controller with 49 rules and 81 rules. Response characteristics [3] Zhiqiang Gao, Thomas A. Trautzsch and James G. Dawson, “A Sta-
of the identified system in both cases (FPDC and FPIC) are ble Self-Tuning Fuzzy Logic Control System for Industrial Tempera-
very close to the original one. ture Regulation”. URL: http://academic.csuohio.edu/aerl/papers/IAS200-
FLC-Tempctrl.pdf
TABLE I [4] Tang-Kai, “A Characteristic-point-based fuzzy inference system aimed
to minimize the number of rules,” IEEE Trans. Fuzzy System, vol. 12,
No. 2, pp. 250-273, April 2004.
Sr. No. System With 8 rules With 49 rules With 81 rules
1 Inverted pendulum 422 ms 2 sec 47 ms 3 sec 406 ms
[5] S.L. Chiu, “Fuzzy model identification based on cluster estimation”,
2 -0.1s
e /(s+1) with 687 ms 5 sec 297 ms 8 sec 953 ms Journal of Intelligent and Fuzzy System, Vol.2, pp. 267-278, 1994
noise [6] S. Chiu, “Extracting fuzzy rules from data for function approximation
3 e-0.1s /(s+1) with 859 ms 7 sec 672 ms 9 sec 625 ms and pattern classification,” to appear as Chapter 9 in Fuzzy Set Methods
FPIC in Information Engineering: A Guided Tour of Applications, ed. D.
Dubois, H. Prade, and R. Yager, John Wiley, 1997.
[7] S. Chiu, “An Efficient Method for Extracting Fuzzy Classification Rules
from High Dimensional Data” J. Advanced Computational Intelligence,
Vol. 1, No. 1, 1997.
V. C ONCLUSION [8] F. Klawonn, R. Kruse, “Constructing a fuzzy controller from data”,
This paper presents two valuable results regarding rule Fuzzy sets and Systems Vol. 85, 1997 pp 177-193.
[9] Kuhu Pal, Rajani K. Mudi, and Nikhil R. Pal, “A New Scheme for
base of fuzzy controller. Firstly it has established that rules Fuzzy Rule-Based System Identification and Its Application to Self-
should be in a certain limit. Increasing the rules beyond that Tuning Fuzzy Controllers”, IEEE transactions on systems, man, and
limit is ineffective. Secondly, the rules can be reduced using cybernetics—part b: cybernetics, vol. 32, no. 4, August 2002.
[10] Petr Pivonka, “Comparative analysis of Fuzzy PI/PD/PID
FSC approach and gives the similar performance as by the controller based on classical PID controller approach”. URL:
larger rule set. But in FSC approach to choose the value of http://www.feec.vutbr.cz/∼pivonka/
radius of influence is based on hit and trial. In this paper, [11] R.K.Mudi and N.R.Pal, A robust self-tuning scheme for PI and PD type
fuzzy controllers, IEEE trans. Fuzzy System., 7(1), 1999, 2-16.
an idea to choose the value of ra is also given. Based on
[12] James Carvajal, Guanrong Chen, Haluk Ogmen, “Fuzzy PID controller:
that value of ra the rules are reduced in all cases. The rules Design, performance evaluation, and stability analysis,” Information
are reduced to 8 from 81, 49 and 25. Simulation results Sciences, 123, 2000, 249- 270.
show that by choosing the value of ra in this way gives [13] Seema Chopra, R. Mitra and Vijay Kumar, “Identification Of Rules
Using Subtractive Clustering With Application To Fuzzy Controllers,”
us a closed performance. In addition, well-known examples Proc. of the Third International Conference on Machine Learning and
of inverted pendulum and a delayed process with noisy and Cybernetics, 2004, 4125-4131.
non-noisy data are simulated and comparison of results has [14] D. Driankov, H. Hellendorn, and M. Reinfrank, An Introduction to Fuzzy
Control. New York: Springer-Verlag, 1993.
shown effectiveness of a proposed fuzzy controller using FSC [15] Seema Chopra, R. Mitra and Vijay Kumar, “Identification of Self-
approach. Tuning Fuzzy PI type controllers with reduced rule set,” Proc. of the
IEEE International Conference on Networking, Sensing and Control,
(Accepted) Conference is in March, 2005.
CHOPRA, MITRA & KUMAR, FUZZY CONTROLLER 79

[16] K. Ogata, Modern Control Engineering. Englewood Cliffs, NJ: Prentice-


Hall, 1970.
[17] M. Gopal, Control Systems Principles and Design. India: Tata McGraw-
Hill, 1993.
[18] Kevin M.Passino and Stephen Yurkovich, “Fuzzy Control”, Addison
Wesley Longman, Inc., California, 1998.
[19] Seema Chopra, R Mitra, Vijay Kumar, “Analysis and Design of Fuzzy
models using Fuzzy Curves”, Proceedings of the Second national
Conference on Intelligent systems & networks, Page 65-71, Feb, 25-26,
2005.
[20] Dae-Won Kim, Ki Young Lee, Doheon Lee, and Kwang H. Lee, “A
kernel-based subtractive clustering method,” Pattern Recognition Letters
26 (2005) 879–891.
[21] Plamen Angelov and Richard Buswell, “Evolving Rule- based Models:
A Tool for Intelligent Adaptation”, Germany: Springer-Verlag, 2002.
[22] P. Angelov and R. Buswell, “Identification of evolving fuzzy rule-based
models,” IEEE Trans. Fuzzy Syst., vol. 10, pp. 667-677, Oct. 2002.

Anda mungkin juga menyukai