Fuzzy neural network model for a class of nonlinear systems

For a class of nonlinear systems that the entire operation region can be divided into several operating regions and the property of each subsystem is linear, a modified fuzzy neural network (FNN) is designed as identifier. The prior knowledge of the system is considered adequately in the FNN model. The local model network of FNN is used to approximate the subsystem of each zone according to ARMAX model. The output of FNN model is the interpolation of local model network outputs with the output of the fuzzy decision-making layer. A new cost function is used when training FNN model. It can reduce over-fit of FNN model.