A Magnetizing Inrush Identification Approach Based on Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise and Bidirectional Long Short-Term Memory Networks

Magnetizing inrush resulting from no-load transformer energization is a major cause of transformer differential protection maloperation. To address this challenge, this paper presents a hybrid identification method combining Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN) and a Bidirectional Long Short-Term Memory (BiLSTM) neural network. First, transformer fault and inrush scenarios are modeled in MATLAB to generate simulation data. ICEEMDAN is then applied to decompose the signals, enabling effective extraction of high-frequency harmonic features associated with inrush current. The decomposed signals, together with other transformer fault data, are subsequently fed into a BiLSTM network for classification. Comparative experiments with CNN, BiLSTM, and EMD-BiLSTM models demonstrate that the proposed ICEEMDAN-BiLSTM approach achieves superior identification accuracy, validating its effectiveness and robustness.