Abstract:

A kind of fuzzy rough neural network (FRNN) with five layers is proposed based on fuzzy rough membership functions. A switch function is used to describe the link among neurons, thus both the optimization of structure and parameters learning are transformed to a pure function optimization problem. A hybrid intelligent optimization algorithm (HIOA) is proposed to optimize the structure and parameters of FRNN and the fitness function considers both accuracy of the model and succinctness of FRNN simultaneousy. The typical experiment results show that FRNN is suitable for modeling nonlinear systems and it is better than the traditional neural network and its optimization method on both accuracy and generalization.