| Lithium ion battery has many advantages,such as high energy density,long cycle life and good stability.It has been widely used in electric vehicles and power grid energy storage.In order to ensure the safe and reliable operation of the battery system,it is very important to estimate the state of Health(soh)of lithium-ion battery accurately and quickly.Lithium ion battery is a complex non-linear dynamic system,and its health state is difficult to measure directly in the actual working condition,It can only evaluate the SOH index indirectly by reflecting the external characteristic parameters of the battery aging process.The method based on a single aging characteristic or model is difficult to guarantee the reliability of the results.Therefore,this thesis proposes a data-driven XGBoost and Kalman filter(KF)combined multi feature joint SOH estimation method.Based on the data,PCA algorithm is used to reconstruct a variety of battery aging features,and XGBoost online estimation model is constructed based on the reconstructed feature data.XGBoost model is modified in real time by introducing time domain Kalman filter,and finally the joint optimal estimation of lithium-ion battery SOH is realized.In this thesis,NASA battery data is used to build and verify the model.The results show that the accuracy and reliability of this method is high.It has a certain reference value for building the health management system of energy storage lithium battery.The specific work is as follows:(1)After analyzing and sorting out the data of the external characteristics of the battery,a variety of characteristic parameters which can fully describe the aging process of the battery are extracted for SOH estimation.According to the external characteristics of different aging degree of battery data,the characteristic parameters of indirect mapping battery SOH are extracted from the working voltage,current and internal resistance data.The quantitative analysis shows that the extracted features are highly correlated with SOH,which can provide quantitative basis for health status assessment and model building.(2)Several methods are used to realize SOH estimation.According to the test results,the evaluation indexes of the model are calculated,and the effect of SOH estimation based on different models is studied.Based on NASA battery data,through the training and fitting of data to establish the nonlinear mapping relationship between eigenvector and SOH,the SOH estimation method based on XGBoost,gbdt,random forest,neural network different data-driven models is realized.The ohmic resistance of lithium-ion battery is taken as the target parameter to characterize the SOH of lithiumion battery.The estimation method of SOH based on Kalman filter is realized by tracking and identifying ohmic resistance through Kalman filter.After the verification and analysis of the results: the data-driven SOH estimation method shows good accuracy for the non-linear battery system which can not establish accurate mechanism model,and XGBoost estimation model has higher estimation accuracy and adaptability compared with other data-driven models;the Kalman filter SOH estimation method is simple and practical,which can reduce the dependence of observation data,and has a certain smooth filtering on the estimation results wave effect.(3)XGBoost and Kalman filter are combined to construct the joint estimation SOH model of XGB-AKF.The adaptive KF algorithm based on sage＿Husa dynamic tracking noise is introduced to correct and filter the XGBoost estimation results in real time.The Kalman filter can correct the error fluctuation of XGBoost model according to the initial state determined by the battery and balance the state equation based on the decay trend of time series,and the adaptive noise algorithm is added to improve the ability of real-time noise estimation.The results show that this method reduces the error fluctuation caused by the over dependence of XGBoost on historical data,and improves the accuracy and robustness of the estimation model. |
PDF Full Text Request