State of Health Estimation for Lithium-Ion Battery With Label-Free Data Based on Deep Domain Feature Independent Alignment Network

Estimating the state of health (SOH) of batteries is critical for ensuring the battery's safe operation and extending its service life. However, the accuracy and application of SOH estimation methods are limited by insufficient labeled training data and different usage scenarios. To address these issues, a label-free SOH estimation method for lithium-ion batteries is proposed, which is based on a lightweight deep domain feature independent alignment network (DDFIAN) model. In this model, dynamic convolutional layers with a convolutional kernel attention mechanism are designed as feature extractors to capture both domain-invariant and domain-specific features from battery charging data. On this basis, feature space independent alignment (FSIA) methods are utilized to enable unsupervised SOH estimation transfer learning, allowing the DDFIAN model to effectively learn the shared features of the source and target domain battery datasets while preserving the battery-specific information. Finally, the proposed method is validated on three lithium-ion battery datasets under varying temperatures, materials, and operating conditions (charge and discharge strategies). The results demonstrate that the proposed method effectively estimates SOH in multi-condition transfer learning scenarios, using 30% of the early-stage unlabeled charging data from the target domain. The method achieves an average MAE of 0.0098, improving SOH estimation accuracy by 50%, 70% and 90% compared to the GRU-FIAN, DANN, and LSTM models, respectively.