Research on Switchgear Partial Discharge Diagnosis Based on Transformer Time-Frequency Feature Analysis

In response to the current issue of insufficient information extraction for switchgear partial discharge fault diagnosis, which mainly relies on manual experience judgment and has a long cycle, this paper proposes a Transformer-based time-frequency feature switchgear fault diagnosis model. The aim is to effectively extract and integrate time-domain and frequency-domain features from pulse data sequences to achieve high-precision fault diagnosis. The data transformation module in the TFFormer model converts the original pulse data into timedomain and frequency-domain representations, enabling the model to fully capture the characteristics of the signal. The feature extraction and fusion module uses re-projection and layer normalization techniques to map the features into a unified space, and enhances the model’s understanding of the sequence with learnable position encoding. The classification module adopts a soft-max structure optimized by cross-entropy, combined with the sample distribution perception capability of the Transformer Encoder, to achieve accurate diagnosis of fault types. An experimental model was built for verification, and the TFFormer has achieved excellent performance on multiple switchgear fault datasets, verifying the effectiveness and reliability of the model in practical applications.