Answers are grounded in this paper and cite their sources.
Power transformers are critical components in transmission and distribution systems, and their operational condition is paramount to grid safety and stability. Current transformer winding monitoring methods, due to their need for extra sensors or signal injection, fail to meet new intelligent power grid’s true online monitoring demand. To address the limitations of invasive monitoring, this paper proposes a non-invasive diagnostic framework utilizing substation fault-recorded waveform data and a hybrid neural network. The model integrates Squeeze-and-Excitation (SE), Bidirectional Gated Recurrent Unit (BiGRU), and ProbSparse Attention (PA) mechanisms. The approach extracts port ΔU-I signals to reconstruct Lissajous figures, applying a preprocessing pipeline of wavelet denoising and load normalization to ensure feature stability under operational fluctuations. To ensure robust training, an 800-sample hybrid dataset was established by combining field-circuit simulations in Ansys Electronics Desktop 2022 R1 with 10 kV experimental tests to cover diverse fault types and realistic interference. Results show the model achieves a diagnostic accuracy of 97.5%, significantly outperforming traditional architectures. The practical feasibility was further validated through two engineering examples, where diagnostic outcomes perfectly aligned with actual field maintenance reports.
No takes yet. Share an insight, caveat, or question.
Zou et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: