Fuzzy system identification through hybrid genetic algorithms
Aran Bey Tcholakian, A. Martins, Roberto C. S. Pacheco, Ricardo Miranda Barcia · 2002
Proposes a new hybrid genetic algorithm for fuzzy system learning. The algorithm is based on Baldwin's (1896) effect, with the inclusion of biological principles of learning. Rather than considering mutation as a stochastic event, we take into account the results of biological experiences that seem to indicate an individual capability to choose the best mutation. The proposed adaptive model consists of two levels: (a) an evolutionary or global level, which works on the generation of populations at the genetic code level; and (b) a learning or local level, which works during the lifetime of the agents, with the individuals reacting to environmental stimuli. The method has been applied in well-known learning problems, with strong supremacy over other hybrid genetic approaches, particularly in terms of the expressiveness of the learned fuzzy system.