基于格拉姆角场与融合注意力机制优化CNN的变压器绕组故障诊断OA

The state of transformer winding has great influence on the reliability of transformer operation,this paper proposes a transformer winding fault diagnosis method based on Gramian angular field(GAF)and an optimized convolutional neural network(CNN)using a fusion attention mechanism.First,a transformer winding fault simulation experimental platform was established,and the frequency response method was employed to obtain frequency response curves for three types of faults including axial displacement,inter-turn short circuit,and bulging warping across three fault regions,providing data support for subsequent intelligent diagnosis.Next,a frequency response curve image conversion technology based on Gramian angular field was proposed,which transforms frequency response curves into Gramian angular difference field(GADF)and Gramian angular summation field(GASF)images.An analysis was conducted to compare the diagnostic accuracy of different CNNs(VGG,ResNet and DenseNet)for various winding fault types and regions using the attention mechanism to optimize the networks.Finally,the proposed fault diagnosis method was applied to field transformers for analysis and validation.The results show that using GADF and GASF images as inputs for CNNs achieves diagnostic accuracy rates of over 88％for both winding fault types and regions,demonstrating the effectiveness of using GADF and GASF images as inputs.Among them,GADF images yields higher classification accuracy,with the combination of GADF and SE-DenseNet achieving the highest accuracy rates of 98.89％and 97.78％for fault type and region diagnosis respectively.Compared to the non-optimized combination of GADF and DenseNet,using a fusion attention mechanism to optimize the CNN improves the recognition accuracy rates for fault types and regions by 2.22％and 3.34％respectively.