Article: Research on distribution network fault identification based on active transfer learning and autoencoder Journal: International Journal of Environment and Pollution (IJEP) 2026 Vol.76 No.1/2 pp.140 - 157 Abstract: The stable operation of distribution networks is crucial for economic stability and daily life, yet frequent faults arise due to long transmission lines and wide coverage. Conventional fault diagnosis based solely on voltage and current data is limited by load fluctuations and external interference. To address this, this paper introduces time-frequency entropy (TFE) features into an autoencoder model to characterise signal dynamics and energy distribution for improved fault identification. Furthermore, an active transfer learning strategy combined with a self-attention-based autoencoder is developed to enhance cross-domain adaptability between source and target networks. Experiments on the IEEE 33-node distribution network show that the proposed method achieves 99.49% accuracy while significantly reducing training time, demonstrating its effectiveness and practical value for cross-scenario fault diagnosis in distribution networks. Inderscience Publishers - linking academia, business and industry through research

Title: Research on distribution network fault identification based on active transfer learning and autoencoder

Authors: Youzhuo Zheng; Jiang Lin; Kun Zhou; Hengrong Zhang; Kailei Chen

Addresses: Electric Power Research Institute of Guizhou Power Grid Co., Ltd., Guiyang, 550002, China ' Nanjing DeRuan Information Technology Development Co., Ltd., Nanjing, 210000, China ' Electric Power Research Institute of Guizhou Power Grid Co., Ltd., Guiyang, 550002, China ' Electric Power Research Institute of Guizhou Power Grid Co., Ltd., Guiyang, 550002, China ' Electric Power Research Institute of Guizhou Power Grid Co., Ltd., Guiyang, 550002, China

Abstract: The stable operation of distribution networks is crucial for economic stability and daily life, yet frequent faults arise due to long transmission lines and wide coverage. Conventional fault diagnosis based solely on voltage and current data is limited by load fluctuations and external interference. To address this, this paper introduces time-frequency entropy (TFE) features into an autoencoder model to characterise signal dynamics and energy distribution for improved fault identification. Furthermore, an active transfer learning strategy combined with a self-attention-based autoencoder is developed to enhance cross-domain adaptability between source and target networks. Experiments on the IEEE 33-node distribution network show that the proposed method achieves 99.49% accuracy while significantly reducing training time, demonstrating its effectiveness and practical value for cross-scenario fault diagnosis in distribution networks.

Keywords: distribution network; fault identification; active transfer learning; autoencoder; TTE; time-frequency entropy; transformer; cross-domain fault diagnosis.

DOI: 10.1504/IJEP.2026.151766

International Journal of Environment and Pollution, 2026 Vol.76 No.1/2, pp.140 - 157

Received: 03 Apr 2025
Accepted: 24 Oct 2025
Published online: 18 Feb 2026 *

Title: Research on distribution network fault identification based on active transfer learning and autoencoder

Keep up-to-date