State-of-Charge Estimation of Li-ion Battery Cell using Support Vector Regression and Gradient Boosting Techniques

Ipek E., Eren M. K., Yılmaz M.

International Aegean Conference on Electrical Machines and Power Electronics (ACEMP) / International Conference on Optimization of Electrical and Electronic Equipment (OPTIM), İstanbul, Türkiye, 27 - 29 Ağustos 2019, ss.604-609 identifier identifier

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Cilt numarası:
  • Doi Numarası: 10.1109/acemp-optim44294.2019.9007188
  • Basıldığı Şehir: İstanbul
  • Basıldığı Ülke: Türkiye
  • Sayfa Sayıları: ss.604-609
  • Anahtar Kelimeler: electric vehicles, li-ion batteries, state of charge estimation, machine learning, support vector regression, gradient boosting
  • İstanbul Teknik Üniversitesi Adresli: Evet

Özet

Increasing demand of li-ion batteries brings the need of high accuracy estimation and control of SOC. Different conventional approaches exist to estimate SOC such as open circuit voltage measurement, coulomb counting, electrical model or electrochemical model. Development of data science brings machine learning techniques into SOC estimation of Li-ion batteries. There are number of works which are presented to prove application of machine learning techniques in li-ion battery state estimation. In this paper, two different machine learning algorithms are implemented to estimate SOC of Li-Iron-Phosphate battery cell experimental test data. Support Vector Regression (SVR) and XGBoost are used to estimate SOC. SVR is Support Vector Machine (SVM) based regression method which is used frequently in data science applications. Also, XGBoost is a novel approach for gradient boosting technique which has parallel computation and decreased training time. Radial Basis Function (RBF) kernel of SVR is used to estimate SOC and evaluated to improve results in this study. SVR and XGBoost are compared in terms of ease of implementation, performance, accuracy and duration. Between 97%-99% coefficient of determination is achieved during the estimations by adapting different parameters.