Simultaneous structure and parameter design of fuzzy systems by hybridizing multi-group genetic algorithm and particle swarm optimization

Article type: Research Article

Authors: Juang, Chia-Feng

Affiliations: Department of Electrical Engineering, National Chung-Hsing University, Taichung, 402 Taiwan, R.O.C. E-mail: [email protected]

Abstract: A new approach that automates the design of Fuzzy systems by hybridizing Multi-group Genetic Algorithm and Particle Swarm Optimization, called F-MGAPSO, is proposed in this paper. By F-MGAPSO, we aim to simultaneously design the number of fuzzy rules and free parameters in a fuzzy system. In the initial population of the conceived GA model, the number of rules encoded in each individual is randomly assigned, and the individuals with equal number of rules constitute the same group. These groups will compete against with each other, and the superior ones will gradually prevail over inferiors. Evolution of population consists of three major operations: group enhancement, variable-length individual crossover and mutation. Group enhancement is to enhance elites in each group by a local-neighborhood version of particle swarm optimization. By performing variable-length individual crossover and mutation operations on elites of the same or different groups, we create offsprings, and non-elites in the old population are replaced by these newly bred ones. Performance of F-MGAPSO is verified through simulations and comparisons with other types of genetic algorithms.

Keywords: Flexible partition, structure/parameter learning, TSK-type fuzzy rule, fuzzy control, rapid thermal processing control

Journal: Journal of Intelligent & Fuzzy Systems, vol. 17, no. 2, pp. 83-93, 2006