Performance assessment of two adaptive Kalman filters for battery state-of-charge estimation

An accurate state of charge (SOC) is required to improve the reliability, cycle life, safety, and economics of the batteries used in power applications such as electric vehicles and smart grids. The adaptive extended Kalman filter (AEKF) is an advanced technique used to determine the SOC. The first task in estimating the SOC is to choose the initial state covariance (P0) when the process noise covariance (Qk) and the measurement noise covariance (Rk) are simultaneously estimated in the AEKF. The performance of the adaptive methods is also determined by the initial states. This study evaluates the performances of two AEKF approaches, including the Bayesian adaptive estimator (BAE) and the innovation-based adaptive estimator (IAE), which are applied to simultaneously estimate Qk and Rk. These two adaptive filtering methods are implemented on the experimental data of a real lithium-ion battery pack. Their performances, including filtering stability and convergence speed, are compared, and their impact factors are discussed.