Studies On Adaptive Fuzzy Neural Network Based On Fuzzy Clustering Algorithm

With the development of science and technology, the modern industry systems are becoming more and more complex. The traditional controllers can not satisfy the high performance of the systems. In this background, the intelligent control theory is proposed and it develops very quickly. Fuzzy neural network (FNN) is an active branch in the intelligent control. FNN is composed of the neural network and fuzzy logic system. It is the organic integration of the two parts. FNN compensates for the shortcoming of the pure fuzzy system which lacks of learning capability. Besides, it makes the neural network transparent and interpretable. FNN can deal with the abstract information and it is good at self-learning and self-tuning. So the theory of FNN is very important for the intelligent control.The structure identification of the system is a difficult problem that always exists in the development of FNN, that is how to partition the input and output space and how to generate simple rules from the observation data so that neural network can effectively implement fuzzy input, fuzzy reasoning, propagation in network and interpretation of the finial results. As an unsupervised classifying method, clustering algorithm is able to partition and classify things according to definite requirements and rules. In this paper, we put clustering algorithm into fuzzy neural networks, extract system's characters and optimize input and output space in order to set up the initial rules. After analyzing the shortcomings of fuzzy C-means (FCM) clustering algorithm, we propose an improved FCM clustering algorithm in allusion to the two problems of confirming the number of centers and finding initial cluster centers. According to the results of the improved clustering algorithm, the number of rules and the initial parameters can be obtained. So the initial structure of fuzzy neural network can be defined accordingly.In the learning process of FNN, we use error back-propagation learning algorithm to refine the parameters and sensitivity pruning algorithm to optimize the structure of network with the purpose of adjusting the structure and parameters of FNN self-adaptively and obtaining the best fuzzy rules. Finally, we take function approximation problem as an example to evaluate the performance of the algorithm we proposed. We draw the conclusion that the new algorithm has advantages in adaptation, modeling accuracy and some other ways. It can be used effectively in the problem of fuzzy modeling and some other control problems.