Parallel Distributed Two-Level Evolutionary Multiobjective Methodology for Granularity Learning and Membership Functions Tuning in Linguistic Fuzzy Systems

This paper deals with the learning of the membership functions for Mamdani fuzzy systems - the number of labels of the variables and the tuning of them - in order to obtain a set of linguistic fuzzy systems with different trade-offs between accuracy and complexity, through the use of a two-level evolutionary multi-objective algorithm. The presented methodology employs a high level main evolutionary multi-objective heuristic searching the number of labels, and some distributed low level ones, also evolutionary, tuning the membership functions of the candidate variable partitions.