A new approach to fuzzy identification for complex systems

In this paper, a simple but effective approach to the identification of fuzzy-rule based models for complex systems with input-output data is presented. The main features of the method are: 1) in the stage of input identification, we neither estimate the parameters of the fuzzy model nor determine the number of the fuzzy rules, which has the advantages of simplicity, flexibility, and reliability as compared with other methods; and 2) in order to achieve the desired identification accuracy with fewer rules, a special fuzzy-neural network (FNN) with a general membership function is used for modeling of systems. Since fuzzy c-means method is utilized to determine the proper structure of the FNN and to set the initial weights in advance, the network can be trained rapidly. Two examples of modeling are shown in this paper.