A weight initialization method for fuzzy neural network based on rule partition

The fuzzy neural network has a strong ability in solving pattern recognition, function approximation and control problems, so it plays an increasingly important role in the field of artificial intelligence. Scholars have confirmed that the initial weights have a great impact on the subsequent learning of the fuzzy neural network. However, due to the complexity and uncertainty of the black-box model, the initialization of the model is always a problem. This paper proposes a weight initialization method based on rule partition for the fuzzy neural network to solve this problem. The connection weights between the membership function layer and the rule layer are initialized by classifying weight intervals according to different rules, which could reduce the similarity of fuzzy rules. Finally, the proposed method is verified by some experiments. The results show that the fuzzy neural network initialized by rule partition achieves better generalization performance, which proves the superiority of this method.