Lithium-ion battery RUL prediction based on optimized VMD-SSA-PatchTST algorithm

Accurate prediction of lithium-ion batteries' remaining useful life (RUL) is critical for system reliability and safety. This study proposes a novel forecasting framework that fuses modal decomposition with the advanced PatchTST model. Initially, the Spearman correlation coefficient is employed to identify features strongly associated with battery capacity. The Variational Mode Decomposition (VMD) method is then used to break down the raw capacity sequence into a set of intrinsic mode functions. To enhance decomposition quality, the Whale Optimization Algorithm (WOA) optimizes the number of modes K and penalty factor α by minimizing mean envelope entropy. The selected features and decomposed components are subsequently input into a PatchTST network, whose hyperparameters are tuned via the Sparrow Search Algorithm (SSA), to predict battery RUL. Experimental validation on the NASA Battery dataset and NASA Randomized Battery Usage Dataset demonstrates that the proposed WOA-VMD-SSA-PatchTST model consistently outperforms baseline models, including CNN, GRU and PatchTST, achieving superior prediction accuracy and robustness.