Development of a self-organized neuro-fuzzy model by using genetic algorithm for system identification

In neuro-fuzzy applications, it is known that the selections in neural network structure, fuzzy logic membership functions, and fuzzy logic rules are very challenging as they are sensitive to modeling accuracy. A neuro-fuzzy model with genetic algorithm is developed for system identification, where fuzzy logic is to tune the membership functions by three-phase learning and genetic algorithm is to search the optimal parameters of the model. The weight/bias in artificial neural network, the center/width of membership function, and the fuzzy logic rules can all be determined. Performance verification of system identification by a benchmark nonlinear difference equation shows that the neuro-fuzzy model with genetic algorithm is most effective in modeling accuracy.