一种模糊逻辑推理神经网络的结构及算法设计

建立了一种基于模糊逻辑推理的神经网络．由样本获取的初始规则确定规则层神经元个数，并确立模糊化层与规则层之间的连接．利用黄金分割法确定模糊化层隶属度函数的初始中心和宽度；根据初始规则的结论确定清晰化层的初始权值；针对网络结构提出了改进的BP算法．仿真实例表明，网络结构合理。具有较好的非线性映射能力，改进的BP算法适合于此网络，与另一种模糊神经网络相比较具有较快的训练速度和较好的泛化能力．

Structure and Algorithm Design of a Fuzzy Logic Inference Neural Network

A fuzzy neural network is proposed according to the fuzzy logic inference.Initial rules are got from samples and every rule becomes a neuron of the rule layer.How fuzzification layer and rule layer connect with each other is determined by the rules.The initial centers and widths of membership functions of the fuzzification layer are decided based on the golden partition method.And the initial weights of defuzzification layer are determined according to the conclusions of the initial rules.For the structure of the fuzzy neural network an improved back propagation(BP) algorithm is presented.The results of simulation show the nonlinear mapping ability of the fuzzy neural network.Compared with the existed results,the fuzzy neural network has faster training speed and better generalization ability.