State of Charge Estimation for EV Batteries Using Support Vector Regression

Accurate and robust state of charge (SOC) estimation for electric vehicle (EV) batteries is an essential key for the advancement of EVs. A battery SOC is highly affected by the driving behavior, ambient temperature, and cell age. This paper proposes a classical machine learning SOC estimation method based on Support Vector Regression (SVR). The model uses experimental testing data for a Lithium-NCA battery under various temperatures and drive cycles, emulating an actual EV application driving conditions. SVR for SOC estimation was rarely visited in previous literature. This paper aims to evaluate the proposed SVR performance for SOC estimation. To demonstrate the potential of the SVR, different SVR kernel performances are analyzed based on prediction accuracy, maximum error standard deviation, runtime, and complexity. Furthermore, the impact of the filtered and the unfiltered data on the model prediction accuracy and time is analyzed to find the best combination of features for the model performance. Finally, the SVR model parameters are optimized using Bayesian optimization method to enhance the model performance.