基于切比雪夫图卷积与门控循环单元的风电机组故障诊断方法OA

Addressing the limitations of traditional feedforward neural networks and convolutional neural networks in effectively extracting nonlinear spatial and temporal features from wind turbine operational data,and the current wind turbine fault diagnosis methods that can only perform state monitoring without effective fault localization,this paper proposes a wind turbine fault diagnosis method based on Chebyshev Graph Convolutional Networks and Gated Recurrent Units.First,a graph structure is constructed based on Dynamic Time Warping algorithm;second,Chebyshev Graph Convolutional Networks are used to extract nonlinear spatial correlations from wind turbine operational data;then,Gated Recurrent Units are employed to extract temporal features from the operational data;finally,the wind turbine fault status and fault location are output through fully connected layers and Softmax activation function.Experimental validation shows that this method can not only diagnose potential faults in wind turbines but also effectively determine the specific components where faults occur,achieving an accuracy of 99.33％,with a low false alarm rate of 0.38％and a low missed detection rate of 0.41％.