Study on Adaptive Parameters for GDFNN via Genetic Algorithms

Abstract

In this paper, to solve the difficulty of artificial setting the parameters of generalized dynamic fuzzy neural network (GDFNN), an improved algorithm is proposed based on the genetic algorithm. The variable weighted-average is selected as fitness-function to transform the multi-objective optimization into single-objective optimization. The initial parameters of GDFNN can be set automatically by using genetic algorithm which optimizes the fitting and generalization effect. The efficiency of algorithm is tested by building MISO and MIMO set models. The results of experiments show that the improved algorithm has a better effect on fitting and generalization than artificial deciding parameters without over-fitting, and adapts to different objects, which solves the difficulty of initial parameters setting.