EKF-Based State of Charge Estimation and Its Embedded Implementation

Abstract

: In battery management systems, accurate estimation of the State of Charge (SOC) of the battery pack is essential. The Coulomb counting method is a commonly used approach for SOC estimation, but it is susceptible to initial SOC errors and current accumulation errors, leading to estimation drift. This paper establishes a second-order RC battery model, employs recursive least squares for model parameter identification, and implements SOC estimation based on the extended Kalman filter. The proposed EKF-SOC estimation algorithm is deployed on the STM32F446 embedded platform. By leveraging the CMSIS-DSP math library for optimized matrix operations, the measured single estimation time is 34 µ s with a CPU utilization of only 0.034%, demonstrating the real-time feasibility of the algorithm on resource-constrained platforms. To address potential numerical divergence issues in single-precision floating-point operations on embedded systems, the Joseph form is adopted for covariance update, together with forced symmetrization and a minimum threshold protection for diagonal entries, effectively ensuring numerical stability. Experiments are conducted under FUDS and DST operating conditions to validate the estimation performance under initial SOC errors and random current biases, respectively. The results show that the EKF-SOC algorithm can effectively correct initial SOC deviations and current accumulation errors, with MAE and RMSE both controlled within 2%. The error covariance updates normally throughout the process without anomalies or divergence, confirming the satisfactory stability of the algorithm.