基于改进的K最近邻分类器的风机故障诊断

Fault Diagnosis for Wind Turbine Based on Improved K-nearest Neighbors Classifier

为提高风机故障的预警诊断准确度,提出了一种基于改进的K最近邻分类器的故障诊断方法。通过引入核函数主元分析,计算各特征向量的贡献度,对欧式距离进行加权,弥补传统K最近邻分类器同贡献权重分配的缺陷。样本训练时,依据各特征向量的贡献数值分配权重。该方法被用于风机故障诊断。实验结果表明该方法增强了诊断准确度,便于工程应用。

To improve the early diagnosis accuracy of wind turbine faults, a fault diagnosis method based on improved K-nearest neighbors （IKNN） classifier was proposed. By means of introducing kernel principal component （KPC） analysis, calculating contribution degrees of the various feature vectors, adding weight for Euclidean distance, the same contribution weight allocation defect of the traditional K - nearest neighbors classifier was remedied. When training sample, the weights were distributed according to the contribution numerical values of the various feature vectors. This method was used to wind turbine fault diagnosis. The experimental results indicated that the method could enhance accuracy, and was convenient for engineering application.