An Algorithm For Automatically Extracting Fuzzy Rules Based On RBF Neural Network

Fuzzy control has been widely used due to its self-adaptability, robustness and easy implementation. However, fuzzy control rules are usually given by experts according to their experiences, which may not be objective and easy to acquire. It is important to structure the fuzzy rules store in the fuzzy control system. Therefore, researching automatic generation of fuzzy rules has important values in the theory and application.People have studied the combination of fuzzy logic and neural network. For many years, and achieved good results, especially in the hand of industry process modeling and control. The error back propagation neural network (BP) is used widely because its learning scheme is visible and easy to be understood. But the classification capability is lower for those patterns away from sample set or new patterns. Therefore recently radial basis function networks (RBFN) are adopted widely for fuzzy rules drawing and fuzzy inference system modeling. It is because those RBFNs possess of simple structure, good local approximating performance, particular resolvability and function equivalence with fuzzy inference system. However, a fuzzy system that has been trained by using learning algorithms may lose its interpretability or transparency, which is considered to be one of the most important features of fuzzy systems.In this paper, firstly, based on fuzzy control theory and radial basis function networks (RBFN) theory, a structure of RBF networks is proposed, which can expresses the interpretability of fuzzy systems efficiently. Then the learning algorithm of extracting fuzzy rules from this RBF networks is discussed in detail. Lastly, simulation studies are carried out on examples, the results of simulation show that the algorithm of extracting fuzzy rules based on measured data is an effective method.