RUL Prediction of Lithium-Ion Batteries Using VMD and a Multiscale TCN

Accurate prediction of the remaining useful life (RUL) of lithium-ion batteries mitigates overuse risks, provides decision-making support for the secondary utilization of retired batteries, and enhances their second-life value. To address the impact of nonlinear characteristics caused by noise and capacity recovery phenomena on the accuracy of RUL prediction, this paper proposes a joint method based on variational mode decomposition (VMD) and a multiscale temporal convolutional network (TCN). First, the raw battery capacity data was decomposed using VMD, which separated the nonlinear degradation characteristics into high-frequency components and the primary degradation trend into low-frequency components. Second, for the high-frequency components, rolling iterative forecasting with a multiscale TCN was employed to capture short-term capacity variations. For the low-frequency components, features extracted from operational data were fed into the multiscale TCN to capture long-term capacity trends. Finally, the predictions from both components were integrated to reconstruct the capacity forecast. Validation results on NASA datasets demonstrate that the proposed method achieves a minimum root mean square error (RMSE) of 0.0104 and a corresponding mean absolute error (MAE) of 0.0084 for capacity prediction. The RUL prediction error is largely confined to within one cycle, indicating the method's excellent prediction capability.