State Of Charge Estimation For Lithium-ion Battery Based On Kalman Filtering Considering Model Error

Battery management system as the core part for electric vehicles plays the central role in ensuring automotive dynamic performance and safety by its energy control strategies and states estimation algorithms.Battery state of charge(SOC)is the most essential and important state among all of the battery states,which serves as not only the indicator of the remaining mileage for electric vehicle,but also the input for various decisions of battery management system.The Kalman Filter(KF)family algorithms are promising for SOC estimation.A sufficiently accurate system model is the precondition for a better performance of the KF-based SOC estimation algorithm.Thus,we manage to globally enhance the SOC estimation accuracy by improving the KF-based algorithm from the battery model error viewpoint,through which the battery model error can be estimated and used for compensating for SOC estimation results.In the research,different types of model error sources are analyzed first.Through qualitative and quantitative analysis,the mechanism that how does the model errors influence the conventional SOC estimation methods and KF-based SOC estimation methods have been studied.On the basis of studying the battery model error,the KF-based algorithm could be improved and the model error observer could be built accordingly.Secondly,the research on battery modeling is conducted from two aspects as model selecting and model parameter identification,in which the equivalent circuit model containing one resistance-capacitance ring is chosen after balancing model complexity against model prediction accuracy.Then three different type of offline parameter identification methods have been studied.The first two methods are based on the least square method and the third is based on genetic algorithm,which is more suitable for parameter identification through various kinds of experimental data.Thirdly,we demonstrate the states output by KF algorithm are less precise with some statistical knowledge-based derivations and discussions if model error occurs.Meanwhile,the necessity of building a model error observer is illustrated.Fourth,in order to jointly estimate the SOC and model error by KF method,the vector characterizing the battery model error is adjoined to the original state vector to form a new state vector,by which a new state space representation is built.After decoupling this joint estimation algorithm,a battery model error observer has been built.Finally,to validate the effectiveness of the proposed method against the battery model error,different error sources such as open circuit voltage drift and voltage sensor drift are injected into battery system as perturbations.The simulation model for verification is established in MATLAB/Simulink.The verified results indicate that the improved SOC estimation algorithm has better robustness and accuracy against the model mismatch compared with the standard KF algorithm.