A Multi-Attention Mechanisms Based Transfer Learning Framework for State of Health Estimation of Lithium-ion Battery

Accurate state of health (SOH) estimation is essential for ensuring the stable and safe operation of lithium-ion batteries. However, the adaptability of the estimation method for batteries with different formulations remains challenging. In this paper, partial charging segments are collected and processed. A pre-training convolutional neural network (CNN), which combines attention mechanisms for heightening the estimation performance, is proposed for integrating the inputs and extracting the hidden features automatically. Experiments are performed to show that the proposed method could reduce the estimation error by 80.9%, 41.3% and 25.6% for LCO, LFP and NCA respectively. Moreover, to reduce the computation burden between different kinds of batteries, a transfer learning (TL) strategy is utilized by fine-tuning the dense layers. The transfer learning results show that the estimation root mean square error (RMSE) of LFP and NCA are only 1.3% and 2.6%, respectively.