Research on Power Transmission Line Fault Diagnosis Based on Transformer-BiGRU

Efficient and accurate identification and diagnosis are of paramount importance to the development of modern power systems. This paper proposes a fault diagnosis method that integrates Transformer and BiGRU models. The Transformer model processes input sequences in parallel, employing self-attention mechanisms to prioritize the identification of critical features. Bidirectional GRU (BiGRU) processes the data sequentially from both the past to the future and future to the past, comprehensively capturing bidirectional temporal dependencies within the fault data. It addresses a limitation of the Transformer in this context, thereby compensating for its deficiency. By means of feature engineering, this paper successfully extracts 11-dimensional time-domain features, which are consequently fed into the Transformer-BiGRU model for the purposes of fault identification and classification. Experimental verification indicates that this model possesses accurate fault detection and classification abilities. The overall fault recognition accuracy amounts to 98.84%, and the recognition accuracy for five types of faults reaches 100%. In comparison with three traditional machine learning methods, namely CNN, LSTM, and SVM, it exhibits more superior performance.