Nonlinear System Identification With Fuzzy RBF Neural Network Based On C-PSODE Algorithm

Nonlinear system identification is a focus and difficulty in signal processing and control theory..Fuzzy RBF (Radial Basis Function) neural network combines the advantages of the RBF neural network and fuzzy system with powerful data processing ability and nonlinear mapping ability. However, the gradient descent algorithm has many defects, such as slow rate of convergence, local optimum, etc. The main purpose of this thesis is to search hybrid algorithms which have good identification results by combining the intelligent evolutionary algorithms and fuzzy RBF neural network.Aiming at existing problems of the gradient descent algorithm, several intelligent evolutionary algorithms are adopted to train fuzzy RBF neural network. In order to compare the identification effect with different intelligent evolutionary algorithms, GA (Genetic Algorithm), DE (Differential Evolution) and PSO (Particle Swarm Optimization) are used for fuzzy RBF neural network nonlinear systems identification. Simulation results show that PSO has better training effect, followed by DE.To overcome the defects of the differential evolution algorithm, such as search stagnation, an improved differential evolution algorithm (Chaotic Differential Evolution, C-DE) based on the idea of Memetic algorithm is utilized. The chaotic local search and parameter adaptive strategy of differential evolution algorithm are utilized to disturbance population and adjust the control parameters in the improved algorithm, which enhance global search ability and adaptive ability of the algorithm. Simulation results show that identification results of fuzzy RBF neural network based on improved differential evolution algorithm is better than PSO and DE.In order to further improve the identification results, a new C-PSODE hybrid algorithm is proposed by combing PSO, DE and Chaotic search (Chaotic Particle Swarm Optimization&Differential Evolution, C-PSODE). In the hybrid algorithm, DE is used to optimize individual best population of PSO and update the global best. The chaotic search is used to optimize the global best. Not only the global search ability of the algorithm is improved, but also the ability of local search algorithm is enhanced in C-PSODE algorithm. Simulation analysis shows the fuzzy RBF neural network which is trained by C-PSODE algorithm has much better identification results.