Development of a self-organized neuro-fuzzy model for system identification

It has been known that it is difficult to establish a fuzzy logic model with effective fuzzy rules and the associated membership functions. Neural network with its learning capability has been incorporated to make the fuzzy model more adaptive and effective. A self-organized neuro-fuzzy model by integrating the Mamdani fuzzy model and the back-propagation neural network is developed in this paper for system identification. The five-layer network adaptively adjusts the membership functions and dynamically optimizes the fuzzy rules. A benchmark test is applied to validate the model accuracy in nonlinear system identification. Experimental verifications on the dynamics of a composite smart structure and on an acoustics system also demonstrate that the neuro-fuzzy model is superior to the neural network and to an adaptive filter in system identification. The model can be established systematically and is shown to be effective in engineering applications.