Lithium polymer battery state-of-charge estimation based on adaptive unscented kalman filter and support vector machine

An accurate algorithm for lithium polymer battery state-of-charge (SOC) estimation is proposed based on adaptive unscented Kalman filters (AUKF) and least-square support vector machines (LSSVM). A novel approach using the moving window method is applied, with AUKF and LSSVM to accurately establish the battery model with limited initial training samples. The effectiveness of the moving window modeling method is validated by both simulations and lithium polymer battery experimental results. The measurement equation of the proposed AUKF method is establishedbythe LSSVM battery model and AUKF has the advantage of adaptively adjusting noise covariance during the estimation process. In addition, the developed LSSVM model is continuously updated online with new samples during the battery operation, in order to minimize the influence of the changes in battery internal characteristics on modeling accuracy and estimation results after a period of operation. Finally, a comparison of accuracy and performance between the AUKF and UKF is made. Simulation and experiment results indicate that the proposed algorithm is capable of predicting lithium battery SOC with a limited number of initial training samples.