An adjusted evolutionary algorithm for the optimization of fuzzy controllers

This paper describes an evolutionary method for automatic generation and optimization of fuzzy controllers (FC). The typical genetic operations mutation and recombination are designed with special respect to the structural characteristics of an FC and permit an effective generation of solutions. The proposed method goes without global parameters for step size adaptation by the use of an individual adaptive mutation mechanism selection and achieves a greater independence of the individuals of a population. The resulting broad dispersion of solutions in the search space leads to a high efficiency in finding good solutions. For the shown control examples the strategy generates controllers out of randomly occupied individuals which show excellent performance. There is high reliability in finding good solutions and usually very few parameters need to be tuned manually.