Intelligent Diagnosis of Transformer Winding Deformation Based on Impedance Characteristics

To address the effects of PT/CT measurement errors and operating conditions on online transformer winding deformation diagnosis, an IGWO-CNN-BiLSTM-AE method based on normal impedance feature modeling is proposed. An online impedance observation model and a PT/CT error propagation model are established to analyze impedance calculation errors. Normal pre-fault data are used to learn impedance feature distributions under the same measurement chain. CNN-BiLSTM-AE reconstruction residuals characterize impedance anomalies, while a normal-state three-phase impedance constraint improves physical consistency. IGWO optimizes network hyperparameters, and phase-wise reactance reconstruction errors on the high–medium and high–low voltage sides are compared to localize abnormal sides and phases. Simulation and field tests using fault-recording data from a 220 kV transformer demonstrate the effectiveness of the method. The proposed approach identifies an impedance anomaly on the low-voltage-side B phase, subsequently confirmed as winding deformation by internal inspection, supporting its application to online transformer condition diagnosis.