Abstract
The lithium-ion batteries widely used in e-transport applications are accompanied by a battery management system(BMS) for state-of-charge(SoC) estimation. In this view, various ensemble-based machine-learning methods have been adopted. However, accurate and speedy estimation of SoC constitutes a critical task, considering the non-linear battery characteristics, the dynamic nature of operating conditions, and the temperature the battery is subjected to. To circumvent the limitations of the currently employed ensemble methods, this article proposes a gradient-boosted support vector regression(GB-SVR)ensemble method. By principle, GB-SVR performs iterative progression towards the minimized loss function, regularized by an error-tolerance value. Comprehensive validation of the proposed methodology has been carried out on four data sets of distinct battery chemistry, capacity, temperature, and dynamic driving profiles. The proposed ensemble approach is found to capture the dynamics efficiently with respect to computational efficiency and accuracy.
Funding Information
  • Office of the Principal Scientific Advisor, to the Government of India, through the Project Sustainable E-Mobility Theme of Delhi Research Implementation and Innovation

This publication has 34 references indexed in Scilit: