The Research Of Fuzzy Reasoning Model And Algorithm Based On Neural Network

Reasoning is a form of thinking that obtains new judgments from one or several known judgments. Reasoning is used to obtain the unknown knowledge from the known knowledge, especially the unknown knowledge can't be grasped by sensory experience. Fuzzy reasoning, which is an important branch of reasoning, is an important tool of fuzzy information processing in information science. Therefore fuzzy reasoning has been highly concerned in the fields of computer science and control science.Since 1973, Professor Zadeh proposed an algorithm of fuzzy reasoning based on the theory of fuzzy sets, scholars in this area have proposed a wide variety of fuzzy reasoning algorithms. For example, Professor E.H.Mamdani proposed the Mamdani algorithm. This method has been applied in industrial automation. Recent years, people began to use neural network for fuzzy reasoning, because fuzzy systems have the shortcomings of much manual intervention, slow reasoning and low accuracy. The fuzzy neural network has the advantages of fuzzy theory and neural network, which has the capacity of learning, association, identification and information processing.In order to implement fuzzy reasoning in neural networks, this paper investigates the algorithms of fuzzy reasoning which can be applied to neural networks. So far, the reductive property is the only criterion to illustrate whether a fuzzy reasoning method is good. Therefore, this paper first analyzes four existing fuzzy reasoning methods, and discusses whether they conform to the reductive property. CRI method which Professor Zadeh first proposed does not satisfy the reasoning reductive property; Triple I method proposed by Professor Wang Guojun and the Truth Valued Flow Inference method proposed by Professor Wang Peizhuang only satisfy the reductive property under single rule fuzzy reasoning, while invalid under multiple multidimensional rules fuzzy reasoning; CRIP method proposed by Doctor Xu Weihong satisfy the reductive property of fuzzy reasoning under certain conditions.Because the above-mentioned four methods do not well satisfy the reductive property, I propose a new method in this paper, which make Triple I method satisfy the reductive property under multi-rules fuzzy reasoning. In this paper, I introduce the concept of similarity measure on the base of Triple I method, and utilize similarity measure to endow each rule with the weights. Then I implement the fuzzy reasoning by applying this improved algorithm on the fuzzy reasoning networks which based on the TIS norm.