Study On SOC Estimation Method Of Battery Management System For Electric Vehicles

The state of charge （SOC） is the key of the battery management system ofElectric vehicles, an accurate estimate of SOC is not only of great significance toimprove the performance of vehicles and battery efficiency, but ensure the safety ofdriving and reduce the cost of operation.In this paper, the research is about the SOC estimation algorithm for LiFePO₄battery. By comparison and analysis of the current battery SOC estimation method,the Kalman Filter is chosen for estimation. After a brief analysis of the workingprinciple, structure and characteristics of LiFePO₄battery, the voltage feature,resistance feature and cycle feature of the battery are discussed. According to theeffects of the charge and discharge rate, temperature, self-discharge and aging onthe estimation of LiFePO₄battery, some appropriate method of correction is used.By analysis of some common characteristic of the equivalent circuit model, themodel of PNGV circuit is selected. According to the impact factors on SOCestimation, the parameters of the battery model are obtained by considering thedifferent SOC and current direction. According to the voltage response of thebattery charge and discharge, the formula of the battery model parameters isdesigned. The function of each parameter and the SOC is obtained by usingleast-square method. The simulation model is established in Matlab, the resultsshow that the selected circuit model can imitate the dynamic characteristics of thebattery effectively to a certain degree of accuracy.According to the basic principle of the Kalman Filter, the Unscented KalmanFilter is chosen for SOC estimation of LiFePO₄battery. The equation of state of thefilter is obtained by calculation of Ah method and the equation of observation ofthe filter is obtained by battery model. The estimation procession based on the UKFis designed. The simulation experiment is carried in the case of the constant-currentpulse discharge and dynamic simulation conditions. Compared the results with thetheoretical values, it is seen that the method can correct the wrong initial value andhave a rapid convergence rate.In order to reduce the estimated error while the current changes frequently orheavily, the algorithm is improved. The divergence of estimated value can bereduced by adjusting the gain of filter and estimating the noise of filter. The simulation results show that the improved algorithm can quickly approaching realvalue in the dynamic condition, and the estimation accuracy is increased obviously.