Two-Stage Genetic Algorithm Offline Parameter Optimization of Adaptive Extended Kalman Filter for Robust Battery State-Of-Charge Estimation

Accurately estimating battery state of charge (SOC) in electric vehicle applications (EVs) is crucial to ensure a safe and reliable vehicle operation. However, robust SOC estimation under all possible operating conditions is challenging due to varying load conditions, varying and non-linear battery impedance, sensor inaccuracies, among others. Meanwhile, battery management systems (BMS) are trending toward more compact designs to enhance reliability by reducing wiring and boosting energy density. Hence, minimizing the memory footprint of SOC estimation algorithms is a key challenge, as their design and tuning remain a time-consuming and costly process for the industry. This paper introduces an Adaptive Extended Kalman Filter (AEKF) algorithm with a two-stage genetic algorithm (GA) for parameter optimization. The first stage role is to find the equivalent circuit parameters’ optimal values in a non-SOC-dependent manner. The second GA optimizes the initial AEKF model tuning parameters. To mitigate the randomness of the GA, an algorithm is designed to automatically determine the optimum set of parameters with minimal user intervention. Finally, to avoid calibrating the AEKF to a Coulomb counter, the obtained parameters were tested locally and using an online tool to ensure the robustness of the estimator. The described algorithm achieves a low root mean square error (RMSE) of 0.7% to 2% across various positive and negative temperatures under several drive conditions. With this tool, the AEKF can be rapidly tuned with minimal user effort, providing fast and robust SOC estimation suitable for automotive applications.