Modeling And Online State Of Charge Estimation Of Lithium-ion Batteries Under Complex Conditions

Electric vehicle can facilitate the development of energy conservation and environmental protection.It has become the current research hotspot.Lithium-ion batteries have long service life and high energy density,and are widely used in electric vehicles.Battery management systems are an important part of electric vehicles.The State Of Charge(SOC)prevents lithium-ion batteries from overcharging and over-discharging,guides battery equalization,improves battery efficiency and estimates remaining mileage.It can only be estimated by combining battery model and estimation algorithm.However,in the vehicle environment,under normal circumstances,the internal parameters of the lithium-ion battery are time-varying with temperature,battery aging and working conditions.In the estimation process,under abnormal circumstances,various abnormal factors such as sensor noise,model parameter decay,error accumulation,initialization error and sensor drift are also faced.The time-varying of the internal parameters and abnormal factors can seriously affect the accuracy of the SOC estimation.Therefore,this paper mainly studies how to model the lithium-ion battery and online SOC estimation under the time-varying parameters and the abnormal factors.The main contents are as follows:Firstly,this paper studies and analyzes the time-varying characteristics of equivalent resistance,SOC-OCV curve,battery capacity and coulombic efficiency through the builtup high performance battery charge and discharge test platform,which provides a basis for precise modeling lithium-ion batteries under complex conditions.Then considering the battery combination model,this paper studies and analyzes various abnormal factors under complex conditions,such as sensor noise,battery model decay,initialization error,etc.,and takes state equation and observation equation framework to analyze different requirements for accurate online SOC estimation under different anomalies.It provides the foundation for the robust estimation algorithm of SOC under various anomalies.Secondly,under normal circumstances,based on the time-varying analysis of the model parameters,this paper designed a parallel algorithm framework for online model parameter identification and online state estimation.In order to better describe the dynamic characteristics of lithium-ion batteries,a novel AutoRegressive eXogenous(ARX)model is used,and it has better scalability and smoothness.In order to better track the dynamic parameters,this paper proposed the adaptive differential recursive least squares to identify the model parameters online.On the one hand,it can ignore the difference part of the Open Circuit Voltage(OCV).And on the other hand,the adaptive multiple forgetting factors can not only solve the multiple time scale problems of parameter variation,but also reduce the computational complexity of online model identification.Further,the model is updated in real time through online identification parameters,and the SOC is estimated online by using Extended Kalman Filter(EKF)under the latest model.The parallel algorithm can ensure the estimation accuracy of the SOC under complex conditions.It is verified by experiments that under the complicated conditions,compared with the conventional recursive least squares online model and the offline model based on the evolutionary algorithm,the proposed algorithm can better fit the parameter time-varying with higher SOC estimation accuracy,especially in the severely changing conditions.Finally,under abnormal circumstances,based on the previous analysis results of relations between various anomalies and online SOC estimation errors,this paper proposed a novel multi-strategy probabilities based fusion method.It can combine the dynamic tracking capability of Proportional-Integral Observer(PIO)and noise tolerance capability of Extend Kalman Filter(EKF).The different abnormal issues,such as internal resistance change,capacity attenuation,data saturation and initialization error,may have a serious impact on the estimates.While the proposed method can guarantee high accuracy estimation and robustness under these issues.It tracks the changes of residual sequence dynamically,and the weighting coefficient can be real-time adjusted according to Bayesian analysis.Through experimental verification,under various abnormal factors,the proposed algorithm can adaptively adjust the optimal fusion mode according to the current state,ensuring high accuracy of less than two percentage points.