Online State of Charge Estimation Framework using Hybrid Equivalent Circuit Model and Neural Network

Abstract

State of charge (SOC) estimation is a mandatory requirement for almost all battery management systems. A correct and quick estimation of SOC is vital to avoid overcharging and discharging of cells and helps keep the temperature in check. Several approaches have been used to estimate SOC, such as the Open-Circuit Voltage (OCV) look-up method, the Coulomb Counting (CC) method, model-based methods, and data-driven approaches that leverage artificial intelligence. This paper proposes a new approach for online SOC estimation for dynamically discharged Li-ion batteries using a neural network trained on synthetic data generated from a pulse discharge test. The approach requires the user to discharge the Li-ion cell for a few seconds before estimating the SOC. The results with the proposed framework were found to be in good agreement with the benchmark extended Kalman Filter (EKF) approach and converged instantaneously as opposed to the EKF approach, which requires hundreds of iterations for convergence.