一种改进模糊神经网络及其在故障诊断中的应用

构造了一种改进模糊神经网络模型，该网络由四层神经元构成，第二层为模糊化支，文中给出了该层的通用隶属函数表达式。计及否定规则的影响，提出了扩展通用隶属函数的概念，并结合汽轮发电机组振动故障的不同特征征兆，给出了其具体表达式。在该网络的第二层与第四层间建立了部分直接连接关系，根据不同征兆对故障诊断结果的重要度不同，赋予了部分连接的优先权值，阐述了建立部分连接的依据和优先权值的确定方法，给出了网络的具体学习算法，并从单故障和多故障识别两个角度，比较了该模型与某改进BP网络的诊断结果，证明了该模型具有较强的故障识别能力。

An Improved Fuzzy Neural Network and Its Application in Fault Diagnosis

An improved fuzzy neural network(IFNN) is established in this paper. The IFNN consists of four layers. The second layer is fuzzy layer, whose general membership function is defined. Considering the effect of denying rules, a concept of extended general membership function (EGMF) is brought forward. Combined with the different symptom feature of turbo generator set's vibration faults, the expressions of EGMF are presented in detail. A partial direct connection is built between the second layer and the fourth layer of a IFNN. According to the importance of different symptoms to fault diagnosis, the prior wheight values of the partial connection are evaluated, and the foundation for the partial connection and the method of determining the prior weight value are set developed. The learning algorithm of the IFNN is given at length. The comparing an improved BP network with the IFNN for a single fault and multiple faults, proved that the IFNN has a good ability of fault recognition.