A new neuro-fuzzy approach for nonlinear system identification based on differential evolution

In this paper, a new neuro-fuzzy approach for complex dynamical systems identification is proposed. The approach combines the merits of fuzzy logic theory, radial basis function neural networks, and differential evolution algorithm. The structure of the proposed algorithm model is a four-layer radial basis function fuzzy neural network (RBFFNN). The differential evolution algorithm is used for network optimization. A parameter called contribution factor is introduced to find out unimportant rules, and delete them. Both the fuzzy network structure and parameter learning can be performed automatically from input-output samples without a priori knowledge. Finally, examples of thermal processes identification are given to illustrate the effectiveness of the proposed approach.