一种基于神经网络的模糊推理方法

对于Zadeh的模糊关系合成法则（简称CRI方法），选择一个适宜蕴涵关系矩阵是至关重要的。通常蕴涵关系矩阵的元素值是通过一个所构造的数学表达式计算的，或是由领域专家凭经验直接给出。然而，有时，特别是对于后一种情形，CRI法不能满足模糊推理最基本的一致性要求。文章构造了一个神经网络用于模糊推理，新方法不仅是CRI方法的推广，而且远比CRI方法易于满足推理的一致性要求。得益于神经网络的长处，新方法具有灵活性、可调性。文章给出了权值的具取值法。

A New Method of Fuzzy Inference Based on a Neural Network

A suitable implication relation matrix is crucial to Zadeh's CRI. The element values of the matrix are usually calculated due to a constructed mathematical expression,or are directly given by experts in application fields. However,sometimes,CRI cannot satisfy consistency principle of fuzzy inference especially in the latter case. In the paper,a new method is presented based on a neural network. It generalizes CRI method and can satisfy consistency principle of fuzzy inference under very weak conditions. Benefited from advantages of the neural network,the inference is flexible and adjustable by the new method. An approach to obtain suitable weights is presented in the paper.