A state of charge (SOC) estimation method based on recursive ridge leverage score‐Nystrom‐support vector machine (RRLS‐NY‐SVM) is proposed to improve computational efficiency and robustness while maintaining high estimation accuracy. The NY approximation algorithm is adopted to directly solve the nonlinear mapping function in the original space, greatly reducing the computational complexity compared with least squares SVM (LSSVM). To further enhance stability and robustness, the proposed model combines the RRLS algorithm with the NY‐SVM to optimize sample selection for low‐rank approximation. The proposed SOC estimation method is validated using two types of lithium‐ion batteries under dynamic stress test (DST), federal urban driving schedule (FUDS), Beijing DST (BJDST), and galvanostatic discharge test cycles at various temperatures. Compared with LSSVM and artificial neural network (ANN)–based SOC estimation methods, the proposed SOC estimation method greatly accelerates the model training while maintaining high accuracy. Specifically, NY‐SVM achieves 98.56% and 97.96% faster training than LSSVM and ANN, respectively, while RRLS‐NY‐SVM achieves 68.98% and 55.97% faster training than LSSVM and ANN, respectively. Furthermore, a real‐time SOC estimation model based on RRLS‐NY‐SVM is established to evaluate the real‐time performance, and the results demonstrate the improved SOC prediction accuracy compared with the offline model.
Liu et al. (Thu,) studied this question.