Research On Lithium Battery State Of Charge Estimation Based On Improved Kalman Filter Algorith

Currently,China’s energy vehicles enter full market expansion period,and lithium-ion batteries have become the main power source of new energy vehicles with their high specific energy,high cost performance,low discharge rate,and no memory effect.Battery management system（BMS）is the core component of new energy electric vehicles,and the lithium-ion battery state of charge（SOC）estimation is the core and difficult point of BMS research,and the accurate SOC estimation value is crucial to guarantee the working performance and energy and safety management of lithium-ion battery pack.In this paper,considering the complex dynamic characteristics of Li-ion batteries,the following studies are carried out for the accurate modeling and SOC estimation of Li-ion power batteries in new energy vehicles:（1）Building a battery testing platform.Firstly,the internal structure and working principle of Li-ion power battery are studied,and then charge and discharge experiments are conducted on Li-ion battery based on the built Xinwei CT-4008-5V12A-TB battery test platform,and Real-time access to battery test data under different operating conditions.It provides training data and test data data for the subsequent modeling and charge state estimation of Li-ion battery.（2）Considering the modeling of lithium-ion batteries and their parameter identification,a second-order RC equivalent circuit model is established and an adaptive identification algorithm for model parameters on separated time scales is used.The parameter identification algorithm consists of two independent modules,one of which is used for the identification of slow dynamics and the other for the identification of fast dynamics.These two modules are executed on different time scales.The proposed separated time-scale adaptive discrimination algorithm is compared and analyzed with the traditional least squares algorithm and stochastic gradient algorithm,and finally the accuracy of the model is verified under different working conditions.（3）To solve the problem of poor robustness in estimating the charging state of lithium batteries using a single extended Kalman filter algorithm,this paper presents a multi-information adaptive extended Kalman filter algorithm,which adjusts the statistical characteristics of the online noise,adds a decaying memory coefficient to the algorithm to distinguish the old from the new data,and expands the residual scalar in the algorithm to a new interest matrix by using the data information of multiple moments.Increase the weight of new observations in correcting the prediction process,and weaken the influence of old data on the current filter value.In addition,considering the change of noise,an iterative estimation of noise is introduced based on the extended Kalman filter to achieve adaptive noise correction.（4）To solve the problem that the time constants of the links caused by the double layer effect are different from those caused by the diffusion effect,a dual-rate sampling adaptive method is proposed.The mixed-rate sampling state space equation is established for different links（R₁-C₁ and R₂-C₂）according to their own time scales.For fast dynamic R₁-C₁ coupling,a smaller sampling rate is set to adapt to the fast state change.It can ensure timely status updates.For slow dynamic R₂-C₂ coupling,setting a larger sampling rate to accommodate slow state changes can avoid incorrect overlap of state update values due to inappropriate sampling rate（5）In order to solve the problem of filter divergence caused by rounding errors because of the limited word length of the computer,the traceless Kalman filter（UKF）method based on the Cholesky decomposition principle is proposed for estimating the lithium battery SOC.It uses the UKF algorithm to update the noise variables,and the traceless Kalman filter does not need to force the nonlinear system to be linearized,avoiding the error caused by the EKF algorithm ignoring the higher order terms.Cholesky decomposition of the covariance matrix is performed to ensure its non-negative determinism,and the covariance decomposition matrix will be iteratively updated instead of the error covariance matrix,which ensures the semi-positive nature of the covariance matrix and the stability of the filter values,thus overcoming the problem of filter divergence.