A comparative evaluation of filters applicable to low-complexity processors in the model-based state-of-charge estimation of lithium-ion batteries

Abstract The widespread adoption of electric vehicles (EVs) has significantly increased demand for lithium-ion batteries (LIBs). The charge and discharge processes of LIBs must be closely monitored by a battery management system (BMS) to ensure safe operation. Accurate state-of-charge (SoC) estimation is crucial for determining remaining driving range, controlling current draw within safe limits, and providing reliable information to users. While advanced techniques such as deep learning offer high accuracy, they require substantial memory and processing power. For cost-sensitive EVs and embedded BMS hardware, resource-efficient methods are essential. This study focuses on ECM-based SoC estimation using terminal current, terminal voltage, and the open-circuit voltage (OCV)-SoC curve. Since ECM-based estimates often exhibit oscillations and abrupt changes, filtering is necessary. This paper compares five filters - Moving Average Filter (MAF), Exponential Filter (EF), Butterworth Filter (BWF), Extended Kalman Filter (EKF), and Particle Filter (PF) - in terms of SoC estimation accuracy and execution time, using battery data from the US06, UDDS, LA92, and two combined drive cycles. Results show that PF with N = 100 particles achieved the highest accuracy across all drive cycles, with MAE below 3% and RMSE below 4%, at an execution time of 12–16 µs per step, comparable to EKF. For severely resource-constrained platforms, EF offered the lowest computational cost at just 4–5 µs per step with minimal memory requirements. BWF, despite accuracy comparable to MAF and EF, required approximately 70 µs per step - up to 15 times longer - making it unsuitable for low-capacity processors.