Optimization of fuzzy systems by dynamic switching of reasoning methods

The major idea of this dissertation is to use different fuzzy reasoning methods, e.g., aggregation operators and defuzzification methods, for optimization of fuzzy systems. This approach extends the known methods for optimization of fuzzy systems which are based essentially on optimization of the membership functions and rules. In terms of systems theory, the former approach is related to optimization of the structure of the system while the latter is related to optimization of the parameters of the membership functions and rules. The validity of this concept is demonstrated on a number of examples.;The performance of a fuzzy system depends on which reasoning method is chosen. However, the best performing reasoning method depends significantly on the reasoning environment. Hence allowing for dynamic switching of reasoning methods in a fuzzy system as the reasoning situation changes can lead to better performance, even when the choice is only between two different reasoning methods.;The purpose of this dissertation is to construct a generalized framework which dynamically changes the reasoning method to be used in a fuzzy system as the reasoning situation changes. In particular, the Dynamic Switching Fuzzy System (DSFS) model is proposed to dynamically switch and adjust among different reasoning methods. Furthermore, it is shown how parameterized reasoning methods (e.g., BADD defuzzification method) can be tuned by DSFS. Fuzzy meta-rules are used to implement such tuning. Additionally, it is shown that tuning reasoning methods during defuzzification is computationally more efficient than tuning the rules or membership functions.;Finally, practical methods for automatic design and tuning of fuzzy systems are presented and applied to a complex control problem: swing-up control of a two-link robot called the Acrobot. A combination of Genetic Algorithms, Dynamic Switching Fuzzy Systems (DSFS), and Meta-Rule techniques is used to realize a high performance Meta-Rule Enhanced TSK controller for the Acrobot. These methods are integrated; they result in reduced design time and system complexity.