Prediction Method of Lithium-Ion Battery Life for LSTM Network Based on Model Decomposition and Bayesian Optimization

Minimizing risks associated with batteries and guaranteeing operational stability highly depend on the accurate prediction of the remaining useful life (RUL) of lithium - ion batteries. In this study, A method for predicting the lifespan of lithium - ion batteries is presented. In particular, an LSTM network is integrated with modal decomposition and Bayesian optimization. The battery capacity data is split into multiple high - and low - frequency components by using the adaptive noise fully integrated empirical mode decomposition (CEEMDAN) algorithm. As a result, the noise interference in the predictions can be reduced. After that, The hyperparameters of the LSTM - based forecasting model are fine - tuned with the Bayesian optimization method. After that, this optimized neural network is used to predict each decomposed component. After that, all the component - level predictions are combined to get the final RUL estimation. In order to assess the efficiency of the proposed approach, Public lithium battery datasets from the University of Maryland's Center for Advanced Life Cycle Engineering (CALCE) are used for experiments. It is confirmed through comparison with both single - model and hybrid approaches that the CEEMDAN - BO - LSTM model can provide accurate RUL predictions. The mean absolute error, square error mean root, and the mean absolute percentage error are kept under 0.029, 0.036, and 4%. In order, This emphasizes the strong forecasting ability of the model regarding lithium battery remaining useful life (RUL).