Comparing different versions of differential evolution for training Fuzzy Wavelet Neural Network

Some derivative free methods have been introduced for training Fuzzy Wavelet Neural Network (FWNN). Among them, Evolutionary Algorithms (EA) are more attractive because of their training ability. In this paper, we review eight basic different versions of Differential Evolution (DE) and then compare their power in FWNN training using a nonparametric statistical test. We choose DE among EAs because of its lower computation and better solutions. Approximation of a piecewise function, time series prediction, and two problems about identification of dynamic plants are used as benchmarks in simulation results.