Mitigating the Effect of Noise Uncertainty on the Online State-of-Charge Estimation of Li-Ion Battery Cells

Real-time battery state-of-charge (SOC) estimation is critical in many applications. The extended Kalman filter (EKF) has been successfully deployed in SOC estimation allowing real-time SOC monitoring. However, modeling inaccuracies, measurement faults, and wrong initialization can cause the estimation algorithm to diverge. The precise knowledge of statistical information about process and measurements noise is crucial for accurate system modeling and estimation. This paper presents a novel SOC estimation approach based on maximum-likelihood estimation (MLE). The process and measurements models are transformed to an error state propagation system where the innovation covariance is utilized to maximize the likelihood of the multivariate innovation distribution with respect to process and measurement covariances. The MLE formulation allows the estimation of the process and measurement noise covariance magnitudes, which are used to obtain an optimal SOC estimate. The proposed method is validated experimentally using a number of Li-ion battery cells under various testing conditions. The estimation performance is compared with that of the conventional EKF technique as well as previously published results based on autocovariance least-squares measurements noise estimation. The results indicate an enhanced performance for the new algorithm over the traditional EKF across all conducted tests.