Deep sparse autoencoder-based fault diagnosis method for voltage transformer secondary circuit in 110 kV substation

Abstract

The secondary circuit of voltage transformers in 110kV substations exhibits multiple fault categories with strongly coupled characteristics. Conventional detection methods that rely on predefined feature templates often fail to capture the intrinsic nature of these faults. To address this issue, this paper proposes a fault diagnosis scheme based on deep dictionary learning and sparse coding theory. The equivalent measurement resistance is first derived and then fed into a three-layer deep dictionary learning framework, which hierarchically extracts discriminative fault features. Experimental results show that the reconstruction correlation coefficients exceed 0.924 for all fault types, and the minimum detection error reaches 0.22%, outperforming KNN, LSTM, and single-layer dictionary learning. The parameter selection principles are systematically analyzed, and the engineering applicability gap between simulation and real-world data is discussed. Furthermore, the proposed method demonstrates robust performance under noisy conditions, maintaining a detection error of only 0.58% at 10dB SNR, which underscores its potential for practical deployment in substation environments.