SOH and RUL prediction of Li-ion batteries based on improved Gaussian process regression

Accurately predicting the state of health (SOH) and remaining useful life (RUL) of Li-ion batteries is the key to Li-ion battery health management. In this paper, a novel GPR-based method for SOH and RUL prediction is proposed. First, five features are extracted from the cyclic charging currents of batteries, and a grey correlation analysis (GRA) shows that these five features are highly correlated with battery capacity. A novel Li-ion battery SOH prediction model is established by improving a basic Gaussian process regression model. Meanwhile, a polynomial regression model is developed to update the feature values in the future. Then the RUL of a battery is predicted by combining the SOH prediction model. Finally, the prediction effect of the proposed model is compared with other models using four Li-ion battery degradation data. The obtained results show that the model proposed in this paper has the highest accuracy. The robustness of the proposed model is verified by random walk battery data.