Transmission line fault diagnosis based on SENet-ResNext-LSTM

Abstract

Abstract A proposed model for diagnosing faults in transmission lines, integrating attention mechanisms, the ResNext network, and the Long Short-Term Memory (LSTM) network addresses issues such as diminished precision, constrained adaptability, and reliance on conventional manual fault detection methods. Initially, waveforms captured from transmission line faults undergo wavelet transformation filtration. These waveforms are subsequently partitioned into two segments for feature extraction. One segment undergoes processing by the ResNext module, equipped with an attention mechanism, thereby reducing computational complexity. In contrast, the other segment is input to the LSTM network to ascertain the dependence on fault data sequences. Ultimately, features derived from both branches are combined and input to the Softmax layer for fault classification. Experimental results indicate that our proposed approach surpasses existing deep learning methods in accuracy metrics for diagnosing faults in transmission lines.