Advanced voltage abnormality detection in real-vehicle battery systems using self-organizing map neural networks and adaptive threshold

The safe operation with electric vehicles depends on stable battery systems. This study introduces an innovative voltage abnormality diagnostic method based on self-organizing map neural networks, offering a novel approach to enhancing battery safety. Using one year of real-world vehicle data, the study investigates the evolution patterns of battery system inconsistency and examines correlations between various driving parameters. To amplify voltage abnormal fluctuations, the first type of curve integral is employed, enabling precise identification of abnormal cell voltages through self-organizing map neural networks. The effect of different output grid configurations (2 × 2, 4 × 4, and 8 × 8) reveals that the 2 × 2 grid achieves the best clustering results for abnormal cell detection. Additionally, an adaptive thresholding method is proposed to enhance the accuracy of abnormal cell identification. The results shown that clustered cell voltages follow a normal distribution, enabling robust identification of abnormal cells using Lajda's criterion with a dynamically adaptive computational window of 5. By analyzing the clustering results and risk coefficients of cell voltages over long-term operation, the method effectively identifies abnormal cells in real-world battery systems. This study provides an effective tool for voltage abnormality detection while offering theoretical and practical support for battery safety.