Lithium-ion batteries lifetime early prediction using domain adversarial learning

—Early prediction of the battery lifetime plays an important role in the safety of battery usage. However, existing methods face challenges stemming from a limited variety of training data. In this study, to address this data scarcity issue, a transfer learning approach for battery lifetime prediction is proposed, utilizing data from different datasets to train the prediction model. Firstly, a deep learning model is developed for lifetime prediction, incorporating a feature extractor, a lifetime predictor, and a domain classifier. Convolutional neural networks with attention mechanism is used in the feature extractor for comprehensive feature extraction. Secondly, a domain adversarial learning strategy is implemented to train the model, encouraging to extract features that are domain independence. The strategy guides the feature extractor to yield domain-invariant features crucial for knowledge transfer. Finally, the effectiveness of the proposed method is validated using publicly available datasets. Experimental findings demonstrate that the root mean square errors decrease by 68.1 % and 17.9 % on two datasets, respectively. It underscores that the model's proficiency in predicting battery lifetime without reliance on labeled data from the target dataset.