Automatic learning of fuzzy partitions in human Central Nervous System modeling using genetic algorithms

摘要

The Fuzzy Inductive Reasoning (FIR) methodology, based on fuzzy logic, is sensible to the variations of the number and shape of the fuzzy sets during the discretization and modeling processes. Until now, some heuristics have been used to determine a system variable partition. The main goal of this paper is to present a genetic algorithm (GA) in the context of FIR methodology, that allows to determine in an automatic way a fuzzy partition. The GA can be viewed as a pre-process of FIR methodology that allows the modeler not to rely on heuristics for the definition of a system variable partition. Two different AG cost functions have been implemented in this paper. This new approach is used to obtain accurate models for the five controllers that compose the human central nervous system (CNS). The results are compared and discussed with those obtained using the Simulated Annealing approach for the same problem, as well as with those obtained by other inductive methodologies for the same application.