Time series prediction based on fuzzy-membership cerebellar model neural network with differential evolution

In this paper, a fuzzy-membership cerebellar model neural network with differential evolution (DEFM-CMNN) is proposed, which improves the prediction accuracy of the cerebellar model neural network based on fuzzy membership (FM-CMNN) for prediction of time series. Firstly, use the linear quantization function of FM-CMNN input space to quantize the time series into nonlinear state variables. Secondly, the mutation, crossover and selection operations of the differential evolution (DE) algorithm are used to learn the connection weight of the network. Meanwhile, the adaptive mutation operator is introduced to enhance the diversity of individuals in the early stages of evolution. Finally, the effectiveness of the proposed DEFM-CMNN is demonstrated in experiments for time series prediction. The results show that this model has better real-time performance and higher prediction accuracy.