An algorithm of extracting fuzzy rules directly from numerical examples by using FNN

In this paper, a general method to obtain fuzzy rules directly from numerical data is proposed. The membership function of antecedent part and consequent part is determined by calculating the variance and expectation of the examples. The extraction of fuzzy rules is done by using a fuzzy neural network. The algorithm has two parts. The first part is to determine the optimal number of fuzzy rules. The second part is to improve the accuracy of the inference system. The structure of the FNN is different from previous methods. In the rule layer, there are three counters in each node in order to calculate the maximum degree and the probability of the rule. The output of the node and the weight connecting to the next layer depend on the contents of these counters. From these counters, we can obtain the useful fuzzy rules. The backpropagation learning method is used to improve the accuracy of the system. Finally, through a simulation example, the effectiveness of the proposed algorithm is verified.