| In this thesis, we explore and evaluate two techniques that combine genetic algorithms and fuzzy systems. In particular, the main emphasis of the work will be on designing and evaluating techniques that fuse fuzzy logic and genetic algorithms (GAs), whose territory has not been fully explored. We contend that by employing a combination of techniques, we can yield the advantages and overcome the disadvantages of individual techniques.;The second system we propose is a dynamic parametric genetic algorithm; a genetic algorithm that uses a fuzzy system to control its parameters to improve search performance. The fuzzy system monitors search performance and dynamically controls parameters such as population size or mutation rate. Many genetic algorithm control strategies and heuristics proposed by other researchers can be expressed within the framework of our proposed method. These strategies can be further improved by using our automatic fuzzy system design technique. We have used our methods to design several dynamic parametric genetic algorithms that out-perform conventional genetic algorithms. In addition, intuition from newly obtained control strategies can be extracted from the fuzzy systems, which can lead to a better understanding of the relationship between genetic algorithm performance and parameter settings.;The first system we propose uses genetic algorithms to automatically design fuzzy systems. Unlike other approaches, our approach simultaneously decides the membership functions, number of fuzzy rules, and the consequent part parameters of each fuzzy rule. We also introduce and evaluate techniques that allow our proposed system to take advantage of a priori knowledge. Our results show how the use of a priori can lead to better performance of both the design process and the resulting system. We also show how our proposed system can be guaranteed only to improve on existing solutions. To demonstrate the strength of our technique, we have applied our technique to design fuzzy systems for vehicle control, robot control, and dental age estimation. In each of these applications, fuzzy systems designed using our system out-performed manually designed systems and systems obtained using other methods. |
