Hybrid data-driven method with TADBO-SVMD for lithium-ion battery remaining useful life prediction

Abstract The precise estimation of the remaining usable life (RUL) of lithium-ion batteries is essential for guaranteeing battery safety and enhancing comprehensive life cycle management. Nevertheless, current methodologies exhibit constraints in addressing the non-linearity and non-stationarity of capacity data, together with the capacity regeneration (CR) issue. Furthermore, conventional signal decomposition techniques depend on manual parameter adjustment, which is susceptible to mode mixing or inadequate decomposition. This research presents a hybrid data-driven approach utilizing Tactical Adaptation Dum Beetle Optimizer (TADBO)- sequential variational mode decomposition (SVMD) and deep learning to overcome these difficulties. The primary parameters of the SVMD are adaptively optimized by the TADBO to mitigate noise interference and precisely extract deterioration signals. Secondly, a dual-channel prediction architecture is designed: DMLSTM-FESANet is used to fuse time-frequency domain features to model the long-term degradation trend, and TCN-BimLSTM-AM is combined to capture CR fluctuations, enabling efficient modeling of multi-scale degradation patterns. Experiments on NASA, CALCE, and MIT datasets show that the proposed method significantly outperforms mainstream models, with the average RMSE as low as 0.63%, 1.47%, and 0.17% respectively, representing a maximum improvement of 93.5% compared to the comparative methods. In addition, this method remains robust under extreme operating conditions such as fast charging (3.6 C), verifying its generalization ability across battery systems and complex scenarios. This study provides a new high-precision and adaptive RUL prediction paradigm for lithium battery health management, which has important application value for the safe operation and maintenance of electric vehicles and energy storage systems.