Abstract:

To improve the safety of lithium batteries and delay their life degradation, a Gaussian process quantile regression model was proposed for estimating the state of health (SOH) of lithium batteries. The model combined the advantages of quasi-Gaussian process regression with non-stationary time series analysis and quantile regression, which can effectively deal with the nonlinear and time-varying problem of health characteristic data, and had the ability of adaptive adjustment of model parameters, thus improving the accuracy and robustness of SOH estimation. The validity of the proposed model was verified by the NASA battery dataset based on different temperatures. The results show that the mean absolute error (MAE)、root mean square error (RMSE) and mean absolute percentage error (MAPE) of the SOH estimation results of the model are 0.002 8, 0.003 8, and 0.003 4, respectively, and the model runtime is 0.008 1 s. Comparing the results with the Gray Wolf Optimization Guassian process regression model and three typical model estimation in the literature, the accuracy of RMSE is improved by 0.019 9, 0.003 0, 0.019 6, and 0.002 0, respectively, proving that the proposed model is more robust and has high accuracy in estimation results.

Key words: lithium-ion battery, Gaussian process quantile regression, state of health (SOH), non-stationary features, parameter tuning