A Novel Battery Fault Detection Method Based on Self-Supervised Learning and F1-Score Ranking

Electric vehicles (EVs) play a vital role in reducing carbon emissions, yet battery safety concerns remain a critical challenge. Effective fault detection is essential to ensuring battery reliability and performance. In this paper, we propose a novel battery fault detection method based on selfsupervised learning and F1-score ranking. First, a selfsupervised learning framework utilizing a Transformer-based autoencoder is designed to encode and decode battery data, enabling the training of the model and the computation of reconstruction errors. Second, the reconstruction error is evaluated for each sample, and the F1-score is calculated by treating the reconstruction error as the threshold for fault detection. Finally, the optimal threshold corresponding to the highest F1-score is selected as the decision criterion for identifying battery faults. The proposed method is trained and validated on the EVBattery dataset, with experimental results demonstrating its superior accuracy compared to existing approaches.