Performance and robustness evaluation of EKF, UKF, and adaptive Kalman filter for battery state of charge estimation

Knowing how much charge left in a lithium-ion battery is called the State of Charge (SoC) is very important for electric vehicles and energy storage systems. But getting this estimate right is not easy, especially for lithium iron phosphate (LFP) batteries, which have a flat voltage curve that makes traditional methods less effective. In this study, the battery is modelled using a Pi-equivalent circuit in Simscape, and three advanced Kalman filter techniques are used to estimate SoC, Extended Kalman Filter (EKF), Unscented Kalman Filter (UKF), and Adaptive Kalman Filter (AKF). These filters help combine the battery model with real-time data to get more accurate SoC predictions. The EKF works by simplifying the battery’s behaviour into linear parts, while the UKF uses special sample points to better handle the battery’s non-linear nature. The AKF goes a step further by adjusting itself to changes in the battery’s environment, such as temperature or aging. The performance of each method is tested and compared using simulation and experimental data. Results show that the UKF gives the most accurate results overall, while the AKF offers better flexibility under changing conditions. This research helps improve battery monitoring systems for safer and more efficient use in electric vehicles.The performance of each method is evaluated through both simulation and experimental data. The results show that the UKF achieves the best performance with an RMSE of 0.0961, MAE of 0.0769, correlation coefficient (R) of 0.995, and R 2 of 0.990. These findings demonstrate that the UKF provides the most accurate SoC estimation.