Prediction of the SOH and cycle life of fast-charging lithium-ion batteries based on a machine learning framework

Accurately assessing the state of health (SOH) of lithium-ion batteries is vital for enhancing longevity, optimizing utilization, and ensuring safety. Although data-driven approaches like sliding window-based neural networks show effectiveness in SOH prediction, they often fail to accurately model the nonlinear 'slow-then-fast' capacity degradation trajectory observed in fast-charging lithium-ion batteries. This study presents a novel machine learning framework that integrates cycle life matching via a gated recurrent unit (GRU) network with a sliding window-based long short-term memory (LSTM) model for accurate prediction of state of health (SOH) and cycle life in fast-charging lithium-ion batteries. Evaluation results demonstrate that using only the first 400 cycles of data, the proposed method achieves an average root mean square percentage error (RMSPE) of 1.3389 % and mean absolute percentage error (MAPE) of 1.1879 % for SOH prediction, with an average relative error of 2.0816 % for cycle life prediction. These findings highlight the efficacy of combining GRU-based cycle life matching with sliding window-LSTM in modeling nonlinear degradation behavior, providing a high-precision solution for real-time health monitoring in battery management systems (BMS). • Feature selection based on batteries with similar life spans. • Sliding-window based data modeling enables future health feature predictions. • State of health and cycle life are predicted by a machine learning framework.