Estimating the health status of lithium-ion batteries using deep learning method based on informer model

Lithium-ion batteries (LIBs) emerge as a promising solution to address the energy crisis and prevent further environmental degradation. However, with the advancement of battery technology, different material systems and batteries with longer cycle lives are being increasingly applied, which in turn presents challenges for accurate prediction of the state of health (SOH) of LIBs. To address this issue, this study proposes an emerging approach that utilizes an attention-based Informer deep learning model. The model selects appropriate health features from charge-discharge process data and uses the model's prediction results as inputs for subsequent predictions, thereby achieving accurate autoregressive forecasting of the SOH. Leveraging the attention mechanism's ability to capture dependencies across long sequence data, the proposed model overcomes the limitations of previous approaches, such as the challenge of SOH prediction for long-cycle-life batteries. Additionally, during the model training process, a transfer learning strategy is employed, including pre-training and freezing the shallow layers of the model. This approach reduces the waste of training resources and addresses the issue of limited battery aging data, enabling accurate SOH prediction for batteries with different material systems. This study also investigates the characteristics of historical cycles and explores the prediction of future cycles, enabling the early anticipation of battery degradation. The effectiveness of transfer learning is analyzed using the Massachusetts Institute of Technology (MIT) and Maryland datasets. Experimental results demonstrate the efficacy of the proposed prediction method in accurately forecasting SOH, even in battery systems with varying characteristics.