基于TAMamba-BiGRU的晶闸管变流器换相失败诊断方法OA

To address the problem that commutation failure features of thyristor converters were easily submerged under strong noise interference,leading to degraded diagnostic accuracy,a thyristor converter commutation failure diagnosis model integrating time aware Mamba and bidirectional gated recurrent unit(TAMamba-BiGRU)was proposed.Firstly,one-dimensional data was transformed into two-dimensional feature image using the Gramian angular difference field(GADF).Subsequently,a dual-branch architecture was constructed:the data in the one-dimensional branch were linearly projected to a higher-dimensional space to form high-dimensional sequences,while image features were extracted by a residual network(ResNet)in the two-dimensional branch and further transformed into sequence representations.The two sequences were fused through cross attention and fed into TAMamba module to capture long-range sequence dependencies and rapid dynamic variations,followed by further bidirectional temporal refinement using the BiGRU.Finally,the output features were fed into a fully connected layer to achieve commutation failure fault classification.The results demonstrated that a diagnostic accuracy of 91.52%was still achieved by the TAMamba-BiGRU model under a strong noise condition with a signal-to-noise ratio of 2dB.The model was shown to be effective for commutation failure diagnosis in thyristor converters.