Early identification of anomalous states in electrical lines based on parameter optimization VMD and Dilated CNN-BiLSTM

Low-voltage electrical lines are critical infrastructure, and early anomaly identification is essential for preventing electrical fires and ensuring system stability. Traditional identification methods relying on temperature sensors suffer from low sensitivity and delayed response, while existing artificial intelligence based approaches face challenges in handling nonstationary signals. This paper proposes a novel hybrid model integrating a parameter-optimization variational mode decomposition (VMD) with a dilated convolutional neural network and bidirectional long short-term (Dilated CNN-BiLSTM) to enhance anomaly identification accuracy and robustness. First, Tent chaotic mapping and an adaptive positive cosine algorithm are employed to refine the northern goshawk optimization (NGO), which is then applied to VMD parameters, effectively suppressing mode mixing to improve VMD decomposition. The power spectrum entropy, alignment entropy and fuzzy entropy of the IMFs with significant correlation post-decomposition are computed to construct the feature information matrix. Second, the obtained feature data are input into the Dilated CNN-BiLSTM. Experimental results indicate that the proposed module achieves 98.6% accuracy across 8 states, and maintains high accuracy under noise interference. This method outperforms the existing methods used for the same purposes, validating the effectiveness and dependability of the proposed hybrid framework, providing an efficient solution for early fault warning in electrical lines.