Research On State Of Charge Estimation For Li-ion Battereies Management System In Electric Vehicles

Air pollution is getting increasingly serious and the phenomenon of haze is more and more frequent in domestic cities, which seriously affects the production and life of the residents. The development and popularizat ion of electric vehicles can not only improve the energy utilization ratio but also effectively reduce the emissions of urban traffic. As an important part of the elect ric vehicle, the Battery Management System(BMS) plays the role of collecting the batter y information, ensuring the battery work safety and prolonging the battery’s life. The estimation for State of Charge(SOC) of the battery is the key task of BMS, whose accuracy is always the focus and difficult problems in research on BMS. Taking the batt ery management system software as the research object, based on the Li-ion battery electric property and model, aiming at industrial achievement, the paper has designed the distributed battery management system software to analyze and improve Ah method, Extended Kalman Filtering(EKF) algorithm and neural networks for SOC estimation.Ah method has problems on the initial integration value, algorithm model error and time cumulative error although it has been widely used. The project has reduced the model error by modifying the battery capacity model based on temperature weight, reduced the initial integration error by Open Circuit Voltage(OCV) method and reduced the time cumulative error by designing a charge progress to define the full charge state and a discharge progress to define the empty charge state.With the advantages of Ah method and OCV method, EKF algorithm has the ability to converge to the real value. Beginning with Thevenin Model, the project has got certain relationship from temperature and SOC to the model parameter by a series of experiments and then designed an adaptive model EKF algorithm based on the parameter identification result, which has the ability to correct the initial erro in SOC estimation.Both Ah method and EKF algorithm are designed for single battery which means that they cannot deal with the variation in the battery gro up. The project has taken the voltage difference between single batteries, voltage difference in time and current difference in time as input of the Artific ial Neural Network(ANN), taken SOC as output of the ANN and used the Genetic Algorithm(GA) to train the ANN. Compared with GA based on weight value coding method, GA based on weight step coding method has been designed and proved more excellent both on speed and accuracy for ANN training. ANN Normal Expression and ANN Similarity Function have been defined to research the ANN homogeneous and to distinguish the same overall optimal solution with different expression from the part optimal solution. In the experiments, the ANN algorithm can find the the SOC dropping caused by the variation in the battery group.Finally, based on the achievement of BMS software, the project has put forward a series of standards to evaluate SOC estimation algorithm mentioned above and designed a combined optimization solution according to the evaluation result.