A hybrid-driven method for predicting the remaining useful life of lithium-ion batteries

With the rapid development of the new energy vehicle industry, lithium-ion batteries (LIBs) have become widely used, therefore, an accurate prediction of its remaining useful life (RUL) is essential. However, LIBs exhibit a capacity regeneration (CR) phenomenon during degradation, resulting in a volatile and nonlinear capacity degradation curve. This challenges the prediction model's adaptability and accuracy in predicting the battery's RUL. To address this challenge, we propose a method that combines sequence decomposition with deep learning to predict the RUL of LIBs. First, the battery capacity sequence is adaptively decomposed using time-varying filtered empirical mode decomposition. The resulting components are reconstructed into high-frequency and low-frequency sequences based on the over-zero rate, significantly reducing the time series complexity and mitigating the impact of the CR on predictions. Second, we designed the beluga whale optimization algorithm to optimize the combined ordered neurons long short-term memory and convolutional neural network, as well as the AdaBoost-based relevance vector machine, for predicting the low-frequency and high-frequency components, respectively. This approach aims to enhance prediction accuracy. Finally, the predictions for the low-frequency and high-frequency components are combined to yield the final prediction result. To test the model's generalization and robustness, we conducted experiments on the NASA and CALCE datasets. We evaluated the model using metrics such as root mean squared error, mean absolute error, absolute error, prediction interval coverage probability, and prediction interval normalized average width, and the results demonstrated superior performance compared to other models.