Abstract: Based on the deep study of fuzzy neural networks and hierarchical genetic algorithm, an algorithm is proposed to optimize fuzzy neural network. In the proposed algorithm, the hierarchical coding is adopted to each chromosome, and an improved crossover operater is proposed, so it can evolve both the fuzzy neural network's topology and weighting parameters. Furthermore, a two-objective function is used as fitness fuction to evaluate the structure complexity and the performance of the fuzzy neural networks, and we can confirm the proportion between the complexity and the performance by changing the value of the parameter for a given problem, then we can obtain the near-optimal fuzzy neural network architecture for the problem. Numerical simulations showed the effectiveness of the proposed algorithm.

Key words: hierarchical genetic algorithm, fuzzy neural network, optimize