Genetic Optimization Of Fuzzy Logic Controller

Fuzzy control is a valuable solution for control of nonlinear, multivariable, and uncertain (structure uncertainty and parameters uncertainty) systems which are usually difficult to be modeled. Fuzzy control has achieved great progress from 70s' and has been successfully applied in many areas. But the systematic design and theoretic analysis of fuzzy controllers have not got the same progress yet. For example, it is difficult to determine the shapes of fuzzy membership functions and to acquire the fuzzy relation between the input and output fuzzy variables.In chapter 1, a brief review of fuzzy control and genetic algorithms are presented. Genetic algorithms are considered to be suitable for designing fuzzy controller.Chapter 2 introduces some research backgrounds of fuzzy controllers, genetic algorithms and methods of optimizing fuzzy controllers by genetic algorithms. A developed concept of rule based fuzzy controllers is proposed, and the differences between rule based fuzzy controllers and domain based fuzzy controllers are also presented. The two models have different structures and manners to partition the input space. Rule based fuzzy controllers are more suitable to be optimized by genetic algorithms than that of domain based fuzzy controllers. The general flow chart of optimizing fuzzy controllers by genetic algorithms is then presented. At last, other researchers' work on optimization of fuzzy controllers/fuzzy classifiers are briefly introduced.The simultaneous evolution of fuzzy controller and fuzzy rules is proposed in chapter 3. There are two basic approaches to evolve fuzzy controllers: Pittsburgh approach and Michigan approach. Both of them have some merit and demerit respectively. The CPM approach presented in this chapter can overcome the shortcomings of the two approaches to some extent. CPM evolves fuzzy controllers in two layers: global and local layer. At global layer, CPM evolves the whole fuzzy controller with traditional Pittsburgh approach. At local layer, the concept of "energy" of fuzzy control system is introduced into CPM. After each control step, CPM finds out the "error of energy", and assigns the "error of energy" to each activated fuzzy rules. The data assigned to a rule is called "strength" of the rule and the strength evaluate the performance of the rule. It can be used by Michigan approach to evolve rules of fuzzy...