Fuzzy information granulation for capacity efficient prediction in lithium-ion battery

For lithium-ion cell health diagnosis, machine learning techniques have been widely used but still leave something to be desired. Specifically, Gaussian process regression (GPR) suffers from the exponential increase in computational load when training large sample data, which is not suitable for online deployment. In this study, the fuzzy information granulation (FIG) technique is first combined with the GPR model for short-term lookahead state of health (SOH) estimation and long-term remaining useful life (RUL) prediction. The FIG technique divides the original data into fuzzy particles, which eliminates its volatility and uncertainty to improve prediction accuracy and reduce computational complexity. In the experiment, the laboratory data set and three public datasets under fast charging and constant current and constant voltage conditions are used to verify the effectiveness and robustness. Compared with the single GPR, the proposed method improves the prediction accuracy by 73.04 %, and the corresponding calculation time is reduced from 34.33s to 0.86s.