An Ultra-Early Prediction Framework for Battery Capacity Degradation Trajectories

Speaker：Pengyu Wang

Location: SIST 1A-200

Time：    23rd Sept. 2026 15:30

Host:        Prof. Hengzhao Yang

Abstract:

To support lithium-ion battery health management and safe operation, this paper proposes an ultra-early prediction framework for battery capacity degradation trajectories using visualized single-cycle data. This framework is composed of three phases: image construction, knot prediction, and trajectory prediction. First, the voltage, current, and capacity curves extracted from a single cycle at the ultra-early stage are converted into a three-channel image. Then, a convolutional neural network (CNN) model composed of three AlexNet blocks is employed to predict multiple knots on the trajectory, each consisting of a specific state-of-health (SOH) threshold and the corresponding cycle index. Finally, piecewise cubic Hermite interpolating polynomial (PCHIP) interpolation is implemented to construct the complete trajectory based on the predicted knots. The effectiveness of the proposed framework is demonstrated using the Severson dataset with 124 battery cells. For the 29 cells in the test set, the average trajectory prediction error is 46.7 cycles in terms of mean absolute error (MAE), 51.3 cycles in terms of root mean square error (RMSE), and 7.6% in terms of mean absolute percentage error (MAPE), respectively.

Biography:

Mr. Pengyu Wang received his B.S. degree in microelectronics science and engineering, in Summer 2023, from Jiangnan University, Wuxi, China. He is currently working toward the M.S. degree with the School of Information Science and Technology, ShanghaiTech University, Shanghai, China. His research interests include capacity degradation trajectory prediction of lithium batteries.