| As the core part of power battery systems of electric vehicles,battery management system(BMS)is responsible for managing safe and efficient operation of lithium-ion batteries(LIBs).The State of Charge(SOC)and capacity,as the monitored key indicators of BMS,their accurate estimation is an important prerequisite for ensuring normal charging and discharging of batteries,efficient energy management and remaining mileage prediction.Therefore,it has important theoretical significance and application value to carry out research on estimation methods of the SOC and capacity of LIBs.In view of the main factors that affect state estimation accuracy,such as selection of equivalent circuit model,parameter identification method,filter method and noise estimatior for SOC estimation and feature construction in the charging and discharging curves for capacity estimation,after comprehensive analysis of existing SOC and capacity estimation methods,in this paper the following research work have been carried out:(1)Aiming at the problem that initial parameters should be given for online parameter identification method,a seeking optimization method is proposed to identify the parameters of equivalent circuit model(ECM).This method uses a seeking optimization algorithm(SOA)to perform a global optimization of the parameters of ECM,and uses optimization results as initial parameter values of forgetting factor recursive least square method to update the model parameters online in real time.The effectiveness of the algorithm is verified through the test data of dynamic stress test(DST)and federal urban driving schedule(FUDS)standard conditions.The results show that the parameter identification method of ECM optimized by SOA has relatively high accuracy.(2)Aiming at the problem that the fixed-length error innovation sequences(EIS)are used to update the noise covariance,which results in the low SOC estimation accuracy based on the adaptive extended Kalman filter algorithm(AEKF),a Changing Window AEKF(CW-AEKF is reduced to IAEKF,where ‘I’ is the abbreviation of ‘improved’)is adopted for SOC estimation.In this method,a changing window noise estimator,assuming that the EIS obeys Gaussian distribution,identifies the moment when the amplitude distribution of the EIS changes through the maximum likelihood function,and selects the EIS after the amplitude distribution changes to update noise covariance,is proposed.The changing window noise estimator and AEKF algorithm are combined to estimate the SOC online.The effectiveness of the algorithm is verified through the test data of DST and FUDS standard conditions.The results show that,compared with the AEKF algorithm,the IAEKF algorithm has higher SOC estimation accuracy.By adjusting the maximum window length and threshold parameters,the SOC estimation accuracy of the battery can be further improved.(3)Aiming at the problem that the IAEKF method performs Taylor expansion on the nonlinear measurement equation,only the linear term after expansion is taken,which leads to the problem that only the first-order approximation accuracy can be achieved,a Changing Window Adaptive Unscented Kalman Filter(IAUKF,the principle of abbreviation is the same as IAEKF)is proposed for estimating SOC.This method takes into account that the adaptive unscented Kalman filter(AUKF)algorithm approximates the probability density distribution of the nonlinear function,and has second-order approximation accuracy.The changing window noise estimator and AUKF are combined together,where the time when the distribution of the EIS changes is identified,and the EIS after the distribution change are selected to update the noise covariance,and then the SOC is estimated.The effectiveness of the algorithm is verified through the test data of DST and FUDS standard conditions.The results show that,compared with the IAEKF algorithm,the IAUKF algorithm has higher SOC estimation accuracy.(4)Aiming at the problem that the difference of constant voltage(CV)charging current curve is not significant,resulting in difficulty to directly extract features based on the such curve,a capacity estimation method based on CV charging curve combined with the reconstructed health features is proposed.The CV charging capacity,which has significanty difference,is chosen as one of the health feature parameters through analysis.Considering that CV charging capacity can’t reflect the capacity compresively,the relationship model between the CV charging capacity and the constant current(CC)charging capacity with data obtained offline is established,and the accuracy of the model is verified.When estimating the capacity online,the CV charging capacity is used as the input of relation model to estimate the CC charging capacity,and the CV charging capacity and the estimated CC charging capacity estimates are used as the joint features to predict the capacity based on the support vector machine(SVM).The effectiveness of the algorithm is verified through the performance degradation experimental data provided by National Aeronautics and Space Administration(NASA).The results show that the capacity estimation accuracy can be improved based on CV charging curve with reconstructed health features.(5)Aiming at the problem that only the random discharge data are given,which makes it impossible to estimate the capacity based on charging curve,a capacity estimation method based on random discharge curve is proposed.This method analyzes and mines information from random discharge curves,constructs the mean and standard deviation of random discharge capacity,two health feature parameters reflecting capacity changes,and uses principal component analysis to analyze the correlation of health feature,eliminating the redundancy of feature parameters.Based on the principal component and capacity data of part of the test batteries,the SOA is used to globally optimize the hyper-parameters of SVM and train the model.When estimating the capacity,the trained model is used to predict the capacity of other batteries.The effectiveness of the algorithm is verified through performance degradation test data provided by NASA.The results show that when only random discharge data are given,the health features can be extracted from the random discharge curve to predict the capacity of LIBs.In this paper,LIBs are used as the objects to carry out research on SOC and capacity estimation methods.By selecting equivalent circuit model,optimizing the parameter identification process,imporving noise estimator adopted in AEKF and AUKF methods,the SOC estimation accuracy of the LIBs is improved.By extracting and reconstructing the health features through the charging and discharging curves respectively,and optimizing the SVM,the capacity estimation accuracy of LIBs is improved.This paper provides more reliable basic data for the safe and efficient use of LIBs by improve the SOC and capacity estimation accuracy. |
