Experiments in fuzzy control using genetic algorithms

Fuzzy controllers are based on a rule set that describe the strategy applied to control a plant. Some rules are obvious but in many cases to choose a set of rules is not a very easy task. Almost all fuzzy consumer products on the market rely on rules supplied by a human expert. Tuning the rules is a lengthy and cumbersome process, and some effort has been made toward the automation of this process. Self-organizing-controllers (SOC) and neural networks are examples. Another approach is found in genetic algorithms, which make it possible to explore a great range of solutions. The advantage is that little has to be known about the plant to be controlled and also about the best set of rules. When applied to servo-systems like positioning systems for NC machine tools, some obvious rules may, or even have to, be used, but a final set may be implemented by the algorithm. In this paper, some experimental efforts to combine the power of genetic algorithms and the practical applications of fuzzy controllers, are described.<
>