Micro-genetic algorithms in the optimisation of neuro-fuzzy controllers

Abstract

The neuro-fuzzy network is a combination of fuzzy logic and neural nets which benefits from both approaches. A backpropagation algorithm applied to such a network may converge towards a local optimum. The authors apply the micro-genetic algorithm to optimise the architecture of the neuro-fuzzy network and to ensure its convergence towards the global optimum. This algorithm accomplishes crude approximation of the network architecture near a global optimum, towards which its direct convergence is afterwards brought about by backpropagation.