Using an Extreme Learning Machine and an Extended Kalman Filter, State of Charge Estimation for Lithium-Ion Batteries

When discussing battery management systems (BMSs), one of their most crucial tasks is undoubtedly detecting the state of charge (SoC) of the battery. Background of electrochemical properties of lithium-ion batteries (LIBs), accurately estimating the SoC can be challenging. This study proposes an approach that combines the use of a machine learning technique, the Extreme Learning Machine (ELM), with an extended Kalman filter (EKF) to improve the accuracy of SoC estimation. First, we develop an accurate model of LIBs using ELM. EKF approach with the trained ELM model to estimate the SoC. To validate our approach, compare it with a traditional EKF methodology, the Equivalent Circuit Model (ECM)-EKF, using the Federal Urban Driving Schedule (FUDS) movement over three warming cycles. Results achieved show that the ELM can more accurately monitor voltage compared to the ECM, and the ELM-EKF-based SoC estimation provides more precise voltage estimates compared to ECM-EKF.