The design of self-organizing fuzzy neural networks based on Ga-ecpso and MBP

A novel hybrid learning algorithm which automates the design of the FNNs is proposed in this paper. It is based on two-stage learning process. First, mean shift clustering (MSC) and mean firing strength (MFS) are combined to identify the structure. The MSC is used to generate the initial network structure and parameters of each neuron and MFS refines the initial network to produce the optimal network structure. Next, genetic algorithm enhancing chaotic particle swarm optimization (GA-ECPSO) and modified back-propagation (MBP) are proposed to learn the free parameters. The GA-ECPSO is used to seek the near-optimal parameters solution and MBP continues the learning process until the terminal condition is satisfied. The simulation experiment demonstrates the superior performance of the algorithm.