SOC estimation for batteries using MS-AUKF and neural network

Abstract

In this paper, we present a new method for state of charge (SOC) estimation, which was based on a neural network and the master-slave adaptive unscented Kalman filter algorithm. First, a second-order Thevenin model of batteries was established. In order to improve the fitting accuracy between the open circuit voltage and the SOC of the battery, a neural network model was used to fit the nonlinear relationship, instead of the frequently used polynomial model. To solve the problems that the noise variance is fixed and the estimation accuracy is not high in traditional extended Kalman filter and unscented Kalman filter methods, the master slave-adaptive unscented Kalman filter algorithm was adopted. The main unscented Kalman filter of the algorithm was used to estimate the system state, and the slave unscented Kalman filter was used to estimate the noise variance matrix correspondingly. The update of noise variance of the system model was accompanied by the iteration of the algorithm to get rid of the shortcoming of the traditional Kalman filter algorithm which may cause filter divergence while setting the initial noise variance by experience. Simulation results demonstrate that the master slave-adaptive unscented Kalman filter algorithm shows higher accuracy and faster convergence speed when estimating the SOC of batteries compared with extended Kalman filter and unscented Kalman filter algorithms.