A genetic based fuzzy-neural networks design for system identification

In this paper, we use a modified genetic algorithm (MGA) to construct a fuzzy neural network (FNN), spontaneously, which can approximate a nonlinear function as well as possible. With the specific structure of the chromosome, the special mutation operation and the adequate fitness function, the proposed method with MGA produces a FNN with minimum structure of neural network, smaller number of rules, suitable placement of the premise´s fuzzy sets and proper location of the consequent singletons. Finally, an example is illustrated to show the effectiveness of the proposed method on the nonlinear function approximation.