Flexible health prognosis of battery nonlinear aging using temporal transfer learning

The growing market for electric vehicles and portable electronics requires the reliability assessment of Li-ion batteries, especially in terms of nonlinear capacity degradation. For this purpose, a flexible health prognostic approach for nonlinear battery aging is proposed by leveraging temporal transfer learning. In contrast to existing prediction methods, our approach provides an unsupervised segmentation approach that characterizes the temporal distribution of nonlinear aging trajectories, which is proven to be associated with the loss of active materials. Subsequently, the temporal domain features are aligned with adaptively adjusted weight alongside the capacity decay trend, followed by the domain loss calculation and nonlinear aging prediction. Three battery datasets based on different cell types and conditions undertake the validation of model adaptability. The experiments demonstrate outstanding performance in accurate health prognostic when collaboratively learning short-term incremental capacity and temporal distribution. The proposed approach utilizes temporal transfer learning for health prognosis and promises to be flexibly extended to other battery management systems.