汽轮发电机多故障诊断的SOM神经网络方法

Applications of SOM Neural Network in Multiple-faults Diagnosis of Turbogenerator Set

汽轮发电机组的振动故障具有多样性的特点,经常出现多种故障同时发生的情况。传统的BP神经网络方法可对单一故障有效诊断,若要对多故障进行诊断,则需对各种多故障样本进行学习,使输入空间在训练过程中被样本空间完全覆盖,将大大增加样本空间及学习训练负担,同时网络归纳、联想能力随之大幅度下降,诊断难以实施。因此,将自组织特征映射 (SOM)神经网络应用于汽轮发电机组的振动多故障诊断,用单一故障样本对网络进行训练,根据输出神经元在输出层的位置对多故障进行判断。经实例分析证明,该方法可对多故障进行有效诊断。

The turbogenerator vibration faults have the character of variety. Many faults often occur synchronously. The traditional BP neural network can diagnose the single fault effectively. If we diagnose the multiple faults by using the BP neural network, we must train all samples of multiple faults, which is will increase the number of training samples and the burden of learning greatly. So the diagnosis can not be performed easily. This paper introduces a method based on SOM neural network, which is studied by using the single sample and diagnosing the multiple faults according to the position of output nerve cell. By analyzing the examples, the method is proved to be available for diagnosing the multiple faults of Turbogenerator set.