Abstract: A hybrid model combining single and composite features is proposed to address the problem of the state-of-health(SOH) assessment and remaining useful life (RUL) prediction of lithium-ion batteries. This method first uses the EEMD-LSTM-LightGBM model to effectively predict battery capacity regeneration by analyzing capacity degradation curves, and the goal that the error of RUL is less than 2 cycles and root mean square error(RMSE) SOH prediction is as low as 1.2%. Secondly, a gradient boosted desicion tree model based on composite features is developed, where feature selection is performed through correlation analysis and principal component analysis, and hyperparameters are optimized, resulting in an SOH prediction accuracy with an RMSE of 0.014. Next, based on these algorithms, a lithium-ion battery SOH prediction platform is developed on a local server, integrating both single and multi-feature models. It provides a user-friendly interface that visually displays key indicators such as capacity degradation curves and prediction errors. The experimental results show that through innovative feature extraction and algorithm integration the proposed method, significantly improves the accuracy of lithium-ion battery SOH assessment and RUL prediction, which can be used as a reference for the development and application of related platforms.

Key words: lithium-ion battery, state of health, remaining useful life, machine learning, long short-term memory(LSTM) network, light gradient boosting machine