Nonlinear systems dynamic modeling based on modular neural networks

To model nonlinear systems, an approach which is dynamic modeling based on modular neural networks is proposed. It designs the study systems in a fashion of cooperation of a number of neural networks. Each sub-net has different function, which can distinguish the different influence of approximate effects by different stage dynamic datas so that improve generalization and stability. L-M algorithm training sub-nets which has the fast convergent advantage is adopted. The use of genetic algorithm which can globally optimize in selecting the network structure and initial weights makes the model avoid being trapped in local minima and interrelated. Simulation results show that the approach is more effective than the single network.