Excitation inrush current identification method based on CNN-BiLSTM

Due to improvements in transformer core material processes, the accuracy of traditional excitation inrush current detection methods has gradually decreased, leading to issues with incorrect operation of transformer differential protection. To address this problem, a CNN-BiLSTM-based excitation inrush current detection model has been proposed. First, various transformer simulation models were constructed using the PSCAD simulation platform, and instantaneous current data were collected from these models to form the dataset for the detection model. Next, the CNN-BiLSTM detection model was developed and trained on the MATLAB platform. Finally, comparative experiments were conducted using different models. The results indicate that the CNN-BiLSTM neural network achieves higher accuracy in identifying excitation inrush currents, as well as fault currents, compared to other fault detection methods. This demonstrates the model’s superior generalization performance and fitting capability.