Genetic Adaptation to Optimal Membership Functions for Modelling with B-Splines

This paper proposes a comprehensive approach on automatic optimisation of B-spline based neuro-fuzzy models. The underlying structure of a system decided by the displacements of the membership functions is an important factor for its generalisation ability and output accuracy. Therefore a genetic algorithm is introduced which is used to find a suitable model for an arbritrary given problem. Approximation of benchmark test functions shows that our approach beats the models using other set functions.