State of Health Estimation for Li-Ion Batteries Using Machine Learning Algorithms


Creative Commons License

Koc Y.

3rd International Conference on Basic Sciences, Engineering and Technology, ICBASET 2023, Marmaris, Turkey, 27 - 30 April 2023, vol.22, pp.135-141 identifier

  • Publication Type: Conference Paper / Full Text
  • Volume: 22
  • Doi Number: 10.55549/epstem.1339422
  • City: Marmaris
  • Country: Turkey
  • Page Numbers: pp.135-141
  • Keywords: Feature extraction, Li-ion batteries, Machine learning, Regression, State of health estimation
  • Istanbul Technical University Affiliated: Yes

Abstract

As an energy storage system, Li-Ion batteries have many applications from mobile devices to vehicles. No matter what application they are used in, Li-Ion batteries lose performance over time, and this negatively affects the user experience in terms of both comfort and safety. For this reason, it is extremely important to estimate state of health (SOH) of Li-Ion batteries and to use the batteries accordingly. In this study, examinations on the SOH estimation of batteries with different machine learning (ML) methods are included using Constant Current (CC) and Constant Voltage (CV) charge-discharge characteristics of the li-Ion batteries. Moreover, how the estimation performance changes by both short-term and long-term features is observed by using mutual information metric. According to results, the highest accuracy on SOH estimation is achieved when long-term features are used with Bayesian Ridge Regression. When the short-term features are used, the accuracy of Bayesian Ridge Regression is dramatically reduced, and Random Forest Regression provides highest performance.