State of health estimation of lithium-ion batteries based on BKA-FSVR algorithm with feature reconstruction from partial constant current charging interval

State of health (SOH) estimation has great significance to ensure the safe operation of lithium-ion batteries and prolong the service life of batteries. In the application, incomplete charging data, feature redundancy and complex nonlinear mapping limit the accuracy improvement of SOH estimation. To solve the problem, a SOH estimation method is proposed based on black-winged kite algorithm optimization fusion support vector regression (BKA-FSVR) algorithm with feature reconstruction from partial constant current (CC) charging interval. In the partial state of charge interval of the CC charging stage, four charging features of the voltage variation, the average voltage and the maximum and minimum values of dV/dt are selected. The features are denoised and reconstructed with the complete ensemble empirical mode decomposition with adaptive noise and variational mode decomposition method, which improves the correlation coefficients between the features and the SOH, and also reduces the time series complexity of the features. With the reconstructed features, a linear support vector regression model and a Gaussian support vector regression model are fused as a fusion support vector regression model and optimized with black-winged kite algorithm to estimate the SOH. The model is tested and validated using laboratory battery aging datasets and public NASA and NCM battery aging datasets, and also compared with traditional algorithms. The results show that the estimation error of the BKA-FSVR method is minimized. In the laboratory dataset, the errors of MAPE, MAE and RMSE are 0.64649 %, 0.56558 % and 0.75560 % respectively. In the public datasets, the errors are even lower.