Study On Fuzzy RBF Neural Network Based On RPROP-DE Algorithm

Recently, the rapid development of technology has made systems identification increasing complex, and rigorous. Fuzzy radial basis function network systems identification combines the advantages of both artificial neural networks and fuzzy logic, and demonstrated a strong ability of nonlinear mapping. But because the traditional study algorithms of the fuzzy RBF neural network have many drawbacks such as fall in local extremum, premature and bad identification inaccuracy, this thesis researched the topics.Firstly, the frame and identification elements of fuzzy radial basis function neural network (Fuzzy RBF Neural Network, FRBF) is introduced. Aiming at the slow convergence rate, vulnerable to local extremum and low identification precision of genetic algorithm and BP algorithm (The method mostly used in fuzzy RBF neural network), a new way which based on fuzzy radial basis function network of differential evolution algorithm is proposed to make nonlinear systems identification. Differential evolution algorithm is a mightiness global optimization research method. Differential evolution algorithm has a quick convergence rate, easily to realize, highly stable, less domain knowledge required, so it is very suitable for solving complex optimization problems.Then, to overcome the influence from gradient to network weights diversification in fuzzy radial basis function network systems identification, a hybrid algorithm named RPROP-DE is proposed by combinating differential evolution algorithm and resilient back-propagation (RPROP) algorithm to train fuzzy radial basis function neural network, resilient back propagation algorithm is not influenced by network weights diversification, but only decided by the adjustment direction of fuzzy radial basis function neural network weights. Moreover, RPROP-DE algorithm for systems identification makes a perfect identification capability.Lastly, the identification effects for RPROP algorithm and RPROP-DE algorithm are fully compared by simulation results.Experimental results show that:the identification errors and the identification precision based on RPROP-DE algorithm are all better than identification method of DE algorithm optimization.