基于核主元分析与模糊神经网络的汽轮发电机振动故障诊断方法

Kernel Principal Component Analysis and Fuzzy Neural Network for Turbo-generator Vibration Fault Diagnosis

针对现有汽轮发电机振动故障诊断运算量大、时间长的问题,本文提出基于核主元分析与模糊神经网络的汽轮发电机振动故障诊断方法。首先采用核主元分析并经矩阵变换和降维来提取故障的主要特征值,其次将提取(降维后)的数据作为Takagi-Sugeno模糊神经网络输入数据,最后在Matlab中建立Takagi-Sugeno型自适应模糊神经网络进行训练测试。该方法用较少的数据代表原数据的最大信息量,并且仿真与标准的模糊神经网络、BP神经网络进行性能对比,最后仿真结果表明该方法的有效性,并且具有诊断速度快、收敛迅速和故障诊断效率高等特点。

Aiming at reducing the requiring of vast computation and significant time dviring vibration fault diagnosis for turbogenerator, an approach is proposed based on Kernel Principal Component Analysis （KPCA） and fuzzy neutral network. First fault feature values were extracted through KPCA, which included matrix transformation and dimensional reduction, then were then imported into MATLAB. Afterwards, a self-adaptive Takagi-Sugeno fuzzy neutral network was established for simulation. The predicted results of proposed approach showed agreement with the standard fuzzy neutral network and BP neutral network. The proposed approach, which only requires a small amount of original data, can accelerate the diagnostic speed and improve the efficiency of vibration fault diagnosis.