Research On State Of Charge Online Estimation And State Of Health Prediction For Lithium-ion Battery

A reliable and efficient battery management system can fully utilize the advantages of lithium-ion battery and ensure the lithium-ion bat-tery operate in a safe range.The state of charge estimation and state of health prediction are the key technologies of battery management system,however,it is a challenge to obtain accurate state of charge estimation and state of health prediction results due to the influence of internal chemical re-actions,external environment and complex working conditions of lithium-ion battery.Therefore,in this paper we focus on the state of charge es-timation and state of health prediction methods for lithium-ion battery as follows:First,to improve the accuracy of lithium-ion battery state of charge estimation result,a method based on adaptive extended Kalman particle filtering is proposed in this paper.For the second-order RC equivalent cir-cuit model of the lithium-ion battery,the parameters are identified online using a variable forgetting factor recursive least squares method,which in-troduces the open window theory and enables the forgetting factor to change adaptively according to the error.Based on the online parameter identifi-cation results,the lithium-ion battery’s state of charge is estimated using the adaptive extended Kalman particle filtering algorithm,which can adapt to a wider range of noise than the adaptive extended Kalman filtering,and ensure the diversity of particles while improving the particle degradation phenomenon compared with the particle filtering.Second,to improve the accuracy of lithium-ion battery state of health estimation result,a method based on improved Gaussian process regression is proposed in this paper.The current,voltage,temperature and the incre-mental capacity curve are taken as measurable variables,and comprehen-sive health features are extracted from them.Pearson correlation coefficient is used to filter the extracted health features,and principal component anal-ysis is used to fuse and optimize the filtered health features,thus reducing the health feature complexity.The fused and optimized health features are used as input and the lithium-ion battery state of health is used as output to achieve the prediction of lithium-ion battery state of health based on the improved Gaussian process regression algorithm.The improved Gaussian process regression algorithm introduces the particle swarm optimization al-gorithm for hyperparameter seeking,which improves the problem that the hyperparameters of Gaussian process regression algorithm are prone to fall into local optimal solutions.Finally,the accuracy of the proposed method is verified based on the lithium-ion battery dataset from NASA AMES prognostics data reposi-tory and the Center for Advanced Life Cycle Engineering Battery Research Group at the University of Maryland.There are 49 figures,16 tables,and 69 citations in this thesis.