Design of a Recurrent Functional Neural Fuzzy Network Using Modified Differential Evolution

In this paper, a recurrent functional neural fuzzy network (RFNFN) with modified differential evolution (MDE) method is proposed to solve the prediction problems. The proposed RFNFN model adopts a functional link neural network (FLNN) to the consequent part of the fuzzy rules. FLNN uses orthogonal polynomials and linearly independent functions to form a functional expansion. Thus, the consequent part is a nonlinear combination of input variables. This model also adds feedback connections in the membership function layer to memorize past information for solving temporal problems. Moreover, an efficient learning algorithm, called modified differential evolution (MDE), is proposed to speed up the learning curve and to improve the prediction accuracy. Finally, the RFNFN model is applied to prediction problems of the chaotic time series and the forecast of the sunspot number. The simulation results show that the RFNFN model has a superior performance than DE1 and DE2 methods.