A robust battery state-of-charge estimation method for embedded hybrid energy system

An optimized state of charge (SOC) estimation method is critical for energy control strategy in hybrid energy system. For an embedded system, the executed algorithm should be less time consuming and also robust on measurement noise from sensors. Moreover, the estimation method should also be insensitive to initial SOC for the purpose of avoiding battery relaxing time in real application. The proposed method in this paper combines adaptive unscented Kalman filter (AUKF) and multivariate adaptive regression splines (MARS) to meet the above demands of embedded hybrid energy system. Samples which consist of battery current, terminal voltage and temperature are used to for MARS model training. The effectiveness and robustness of the proposed method is validated by experimental test. Also, the proposed method is compared with least squares support vector machine (LSSVM) based method in estimated accuracy and time consumption. Experiment results indicate that the proposed method is less time consuming as well as good accuracy is guaranteed.