EKF and SVSF-Based Filters for Robust SOC Estimation of Li-Ion Batteries in Electric Vehicles: A Comparative Study

Electric vehicles are currently the leading solution in the automotive market to reduce greenhouse gas emissions. However, transitioning from traditional internal combustion engines poses many challenges. One of these challenges is the accurate estimation of the battery’s state of charge (SOC) since its energy cannot be directly measured during transient operation. This paper presents a comparative study that examines the robustness of the Extended Kalman Filter (EKF), the Smooth Variable Structure Filter (SVSF), which is a robust sliding mode state estimator, and the SVSF with a varying boundary layer (SVSF-VBL) in estimating the SOC while introducing uncertainties in voltage, current, temperature, and capacity. A comprehensive dataset is generated at McMaster Automotive Resource Center (MARC) for the Samsung INR21700 30T Li-Ion cells. The dataset includes a set of real-world drive cycles and characterization profiles. A fourth-order equivalent circuit model is used, whose parameters were obtained using measurements of the cycled cells through curve fitting of Hybrid Pulse Power Characterization test data (HPPC). Genetic algorithm-based optimization was used to tune the three filters using the measurements from four drive cycles for a wide range of temperatures. The effect of various sources of uncertainty is studied together with the impact of temperature across a wide range. The general comparison shows that the SVSF-VBL outperformed the EKF and the standard SVSF, achieving an average root mean square error (RMSE) of 1.43% of the SOC for all temperature values and various uncertainties and errors injected.