Lithium-ion battery state of charge estimation based on square-root unscented Kalman filter

This study represents a method for estimating the state of charge (SOC) of lithium-ion batteries using radial basis function (RBF) networks and square-root unscented Kalman filter (KF). The RBF network is trained offline by sampled data from the battery in the charging process. This type of neural network finds the non-linear relation which is required in the state-space equations. The state variables include the battery terminal voltage and the SOC, at the previous sample and the present sample, respectively. The proposed method is tested experimentally on a lithium-ion battery with 1.2 Ah capacity to estimate the actual SOC of the battery. The experimental results of the proposed method show some advantages, which include: (i) it is not very sensitive to determine, precisely, the measurement and process noise covariance matrices such as Kalman filter and (ii). It contains lower noise on the output, in comparison with Adaptive extended Kalman filter (EKF).