Two-stage aging trajectory prediction of LFP lithium-ion battery based on transfer learning with the cycle life prediction

Ziyou Zhou^{^a}, Yonggang Liu^{^a}(), Mingxing You^{^a}, Rui Xiong^{^b}, Xuan Zhou^{^c}

State Key Laboratory of Mechanical Transmission & College of Mechanical and Vehicle Engineering, Chongqing University, Chongqing, 400000, China

Department of Vehicle Engineering, School of Mechanical Engineering, Beijing Institute of Technology, Beijing 100081, China

Department of Electrical and Computer Engineering, Kettering University, Flint, MI-48504, USA

Handling Editor: Professor Fengchun Sun

Show Author Information

HIGHLIGHTS

· The principles for extracting features that are suitable for cycle life prediction from the battery cycling profiles are summarized. Then a novel feature extracted from the battery voltage is found to be effective in achieving accurate cycle life prediction.

· It is found that the cycle-to-cycle evolution of the proposed feature has a high correlation with the battery cycle life for LFP lithium-ion battery, which indicates that the change of battery voltage in charging process is related to the battery cycle life.

· Accurate long-term aging trajectory prediction can be achieved when satisfactory transfer learning is done by finetuning with informative prior information like the battery cycle life which is related to the aging trend of the battery.

· The whole aging process of the batteries is found to be slower when they have longer cycle life even under the same operating condition, which indicates that the difference in aging trajectories indeed has already shown in earlier cycles.

Graphical Abstract

View original image Download original image

Abstract

With the wide application of the LFP lithium-ion batteries, more attention is paid to the battery life and future aging behaviors as the safety and performance of the battery are guaranteed by accurate battery aging monitoring. In recent years, long-term aging trajectory prediction of the lithium-ion battery is always a challenge due to its complex nonlinear aging behaviors especially the aging behaviors in the two aging stages are quite different when the battery experiences the two-stage aging process under fast-charging conditions. Thus, it is harder to achieve accurate long-term aging trajectory prediction of the LFP lithium-ion batteries on the condition of the two-stage aging process. To address it, a novel transfer learning strategy combined with the cycle life prediction technology is presented in this paper. Specifically, a new cycle life prediction method is proposed based on feature extraction and deep learning technology and achieves accurate cycle life prediction. The transfer learning is started by developing a base aging model offline to learn the information of the two-stage aging process. Then, taking the predicted cycle life as its prior information, the Bayesian model migration technology is employed to predict the aging trajectory accurately, and the uncertainty of the aging trajectory is quantified. Two batches of the battery datasets are used for performance evaluation and comparison with two benchmarks. It is novel to combine the cycle life prediction and transfer learning technique to achieve accurate two-stage aging trajectory prediction with only a few data available (first 30%).

Keywords

Battery aging trajectory prediction Data-driven method Feature engineering Cycle life prediction Transfer learning

References

[1]

Zhou D, Li Z, Zhu J, Zhang H, Hou L. State of health monitoring and remaining useful life prediction of lithium-ion batteries based on temporal convolutional network. IEEE Access 2020;8:53307–20. https://doi.org/10.1109/ACCESS.2020.2981261.