Degradation-Aligned Self-Supervised Learning for State of Health Estimation of Lithium-Ion Batteries under Label Sparsity

To address the issue of sparse label data in lithium battery management, researchers proposed a degenerate alignment self-supervised learning framework based on convolutional neural networks and gated recurrent units (CNN-GRU). This framework utilizes cyclic sorting targets for pre-training on unlabeled data, followed by fine-tuning on sparse labeled data. Experiments show that this method can achieve robust estimation of battery health status (SOH) with only 1% of unevenly distributed labeled data; the average absolute error (MAE) for tested batteries is 1.718%, and the root mean square error (RMSE) is 2.329%. The study also analyzed the impact of label distribution on battery inter-cell robustness, aiming to address the practical need for insufficient label efficiency in battery management.