Constraining the optimization of a fuzzy logic controller using an enhanced genetic algorithm

Abstract

Fuzzy logic controllers (FLCs) are gaining in popularity across a broad array of disciplines because they allow a more human approach to control. Recently, the design of the fuzzy sets and the rule base has been automated by the use of genetic algorithms (GAs) which are powerful search techniques. Though the use of GAs can produce near optimal FLCs, it raises problems such as messy overlapping of fuzzy sets and rules not in agreement with common sense. This paper describes an enhanced genetic algorithm which constrains the optimization of FLCs to produce well-formed fuzzy sets and rules which can be better understood by human beings. To achieve the above, we devised several new genetic operators and used a parallel GA with three populations for optimizing FLCs with 3x3, 5x5, and 7x7 rule bases, and we also used a novel method for creating migrants between the three populations of the parallel GA to increase the chances of optimization. In this paper, we also present the results of applying our GA to designing FLCs for controlling three different plants and compare the performance of these FLC's with their unconstrained counterparts.

On simplifying the automatic design of a fuzzy logic controller

“…We demonstrate the validity of our simplified approach by designing FLCs to control three plant processes with second order transfer functions that we used in our previous work [6]. We assess the performance of the FLCs by simulating step responses and compare them with those obtained in [6] where the membership functions and the rule base were both optimized.…”

“…Based on the experience gained as reported previously in [6] and the results obtained for this work, we present an evaluation of this method of designing FLCs. It has the following merits.…”

“…In order to test the feasibility of our simplified FLC design method, we designed FLCs for controlling unknown plant processes and compared the performance of these FLCs with those designed in [6] where both the rule base and the membership functions were optimized by an EA. To be able to make some generalizations and comparisons we used the same three plant processes with second order transfer functions used in [6].…”

“…To be able to make some generalizations and comparisons we used the same three plant processes with second order transfer functions used in [6]. The three plant processes are labelled Plant A, Plant B and Plant C and are defined as follows…”

“…We used seven fuzzy sets per input/output variable (and hence a 7 by 7 MacVicar-Whelan rule base) of the FLCs in order to compare the FLCs with the best ones obtained in [6]. As discussed above, only one parameter is required to define one fuzzy set and furthermore, since we restrict the parameters of the first and last fuzzy sets to -1.0 and 1.0, only 5 parameters are required per input variable.…”