Lithium-ion Battery State Estimation Method Based On Improved Kalman Filter

Now lithium- ion battery has been widely used. To know the state of battery, we can get effective reference information to adjust the working mode, expect the charging time, plan the tasks reasonably and maintain or replace the battery in time. Therefore, this thesis divides lithium- ion battery state estimation into three parts: state of charge estimation; state of health estimation; state of health prediction. Now Kalman filter and its derivatives method are widely used in battery state estimation and the research results have shown the effectiveness. But this kind of method also commonly exists the drawbacks. For example, precision is excessively dependent on model, the bad model dynamic characteristic and set noise and the iteration initial value aimlessly. Therefore, this thesis launches a research on improved Kalman filter method which aims to solve batteries' whole life state estimation.Firstly, Kalman filter and the data driven method have defects on lithium- ion battery's state of charge and health estimation respectively. To avoid the weakness of data-driven method, which ignore the internal state of lithium- ion battery and poor universality of Kalman filter's model, the method of least square support vector machine fuses with Kalman filter is proposed in the thesis. Kalman filter has the problem of setting noise variances and iterative initial value aimlessly. To solve the problem of poor accuracy caused by the improper noise variances, this thesis proposes a method which can set noise variances according to the state of s tate equation and measurement equation dynamically. In this thesis we call the method as Varied Variance Kalman filtering fuses with least square support vector machine. Secondly, to predict state of health, the method of Kalman filter fuses with least square support vector machine is presented. The battery degradation tendency combines to the data driven method and solve the problem which Kalman filter's prediction accuracy is highly dependent on data model and data driven method has high uncertainty. Finally, using LabVIEW and Matlab hybrid programming, we complete the lithium- ion battery state estimation software and achieve the function of real time state estimation for lithium- ion batteries in order to provide a support for the following study of battery management system in our team.Finally, the experimental results of NASA and CALCE's lithium- ion battery data sets show the superiority of the method proposed in this thesis compared with other methods and verify the practicality of the software.