A Differential Evolution and Spatial Distribution based Local Search for Training Fuzzy Wavelet Neural Network

Abstract   Many parameter-tuning algorithms have been proposed for training Fuzzy Wavelet Neural Networks (FWNNs). Absence of appropriate structure, convergence to local optima and low speed in learning algorithms are deficiencies of FWNNs in previous studies. In this paper, a Memetic Algorithm (MA) is introduced to train FWNN for addressing aforementioned learning lacks. Differential Evolution (DE) is utilized as the global search. The main contributions of this paper are: (i) Proposing a new fast and effective local search based on spatial distribution (that is named Spatial Distribution Local Search (SDLS)), SDLS can adjust the step size of parameters adaptively toward obtaining the better ones. (ii) Introducing an adaptive selection method to select appropriate individuals from current population for local refinement in MA. (iii) Improving the selection operator in standard DE by an adaptive strategy. In this strategy, worse offspring has a chance to be replaced with its parent to prevent trapping in local optima and controlling the selection pressure. The proposed MA is compared with several training algorithms of FWNNs over some benchmark problems. Experimental results obtained, confirm the effectiveness of the proposed MA for improving the convergence rate and modeling accuracy in comparison to the other training methods.