The theory of fuzzy sets and fuzzy logic provides a formal framework to represent and process inaccurate and vague knowledge described in linguistic form. The power of fuzzy system resides in the fact that very often a system can not be modeled or controlled by means of conventional techniques but, nevertheless, a human expert can express his knowledge about the system through common linguistic statements. The problem arises, though, when there is no expert who can describe the system, or the available knowledge is insufficient.
Genetic algorithms (GA) are stochastic search methods applicable to the discovery of fuzzy rules. Despite their widely acknowledged power of finding the optimum region, genetic algorithm-like techniques have very low accuracy, i.e., their convergence to the optimum point is slow. In order to improve their convergence, several special features and operators have been proposed.
This research proposes the use of special local optimization techniques inspired by processes that occur at the bacterial genetics level. This algorithm is named Bacterial Evolutionary Algorithm (BEA). The BEA differs from a GA as it uses the bacterial mutation in the place of the ordinary mutation, and the gene transfer operation in the place of the crossover. The gene transfer operation is based on the natural phenomenon called transduction, which consists in the transfer of strands of genes from a bacteriophage to a bacterium. The bacterial mutation is inspired by the phenomenon of acquisition of traits of the bacteriophage by the bacterium. Numerical experiments were done using the BEA for the design of fuzzy systems. In the first session of experiments, the task was the construction of a fuzzy model of a non-linear equation. The second task consisted of building a fuzzy logic controller for a semi-active suspension system for cars. In both cases, the BEA was able to build more efficient systems than the other compared methods.
The results obtained in this research shows the feasibility of using the BEA for the construction of high performance fuzzy systems, in the cases where the necessary knowledge is not available on hand.