A Creating Method of Fuzzy Inference Rules by Self-Creating Neural Network
Kazuya Kishida, Shinya Fukumoto, Hiromi Miyajima · Journal of Japan Society for Fuzzy Theory and Intelligent Informatics · 1999
There are several fuzzy models using self-organization and vector quantization. It is well known that these models effectively construct fuzzy rules representing the distribution of input data, and are not affected even when the number of input dimensions increases. However most of these models are given the number of fuzzy rules in advance. In this paper, fuzzy rules are created sequentially so as to satisfy an objective value, and the proper number of them is determined finally. That is, the number of neurons which are reference vectors, is determined by using a self-creating neural network first. From the result, fuzzy rules are determined by using the descent method. Then if input-output data are approximated so as to satisfy the objective value, an algorithm terminates. Otherwise an algorithm for increasing the number of neurons is repeated. In order to show the validity of the proposed method, we performed some numerical examples.