Evolutionary learning of fuzzy models

This paper presents an evolutionary algorithm for generating knowledge bases for fuzzy logic systems. The algorithm dynamically adjusts the focus of the genetic search by dividing the population into three sub-groups, each concerned with a different level of knowledge base optimisation. The algorithm was tested on the identification of two highly non-linear simulated plants. Such a task represents a challenging test for any learning technique and involves two opposite requirements, the exploration of a large high-dimensional search space and the achievement of the best modelling accuracy. The algorithm achieved learning results that compared favourably with those for alternative knowledge base generation methods.