| As energy and environmental problems become more and more serious,as a low-carbon travel mode of transportation,new energy vehicles are highly valued by all countries.Compared with fuel vehicles,the energy conversion efficiency of electric vehicles has increased by more than twice,but at the same time,the safety of electric vehicles has also attracted much attention.Timely and accurately fault diagnosis and prediction of the abnormal state of electric vehicle batteries is essential to the safe driving of electric vehicles.This paper takes the actual driving data of electric vehicles as the object,and focuses on the analysis and research of the consistency of electric vehicle power battery cells and the state of health.The main research results and contents are as follows:(1)Based on the historical driving data of electric vehicles,an analysis and research on the defects of the consistency failure of the power battery pack of electric vehicles is carried out.The fault segment is extracted by the fault alarm signal,and the working conditions are divided into three main working conditions: driving state,starting state after static charging,and charging state according to information such as current and vehicle speed,and then mining the battery cell consistency difference under different working conditions The voltage range threshold of the fault alarm and the fault characteristics.The study found that the probability of battery cell consistency failures under driving conditions is low,but the average duration is longer,and it may develop into serious failures and derivative failures.The voltage range threshold for fault determination is 0.3.After charging,the battery cell consistency failure under the start-up condition has a short duration and has little potential harm,and its voltage range threshold is 0.26.No fault alarm signal will be sent under charging conditions.(2)Research the feasibility and effect of clustering algorithm for battery cell consistency fault diagnosis,and combine with LS-SVR for fault prediction.Summarizes the cell voltage characteristics when the battery cell consistency failure occurs,and uses the K-Means and DBSCAN clustering algorithms to diagnose the abnormal voltage battery cells.It is found that the DBSCAN clustering algorithm has better versatility and accuracy.Accurately locate abnormal battery cells.On this basis,a fault prediction method based on LS-SVR is proposed.This method can predict the voltage range and cell voltage of up to 10 time points(100 seconds)in the future based on the data of the first 20 time points.Experimental comparison results show that LS-SVR has better prediction accuracy than ordinary SVR.It can predict the voltage range and battery cell voltage value to realize the short-term prediction of battery cell consistency faults and battery cell over/under voltage faults.(3)Based on the principle that the attenuation of the state of health of the battery pack is mainly reflected in the capacity attenuation,it is proposed to calculate the capacity of the battery pack by extracting the constant current charging segment and combining the SVR of the linear kernel function,and then creating the capacity increase curve of each constant current charging segment,Extract the characteristics that can characterize the capacity degradation of the battery pack.The main extracted features include 6 groups of feature variables related to peaks and troughs.Through correlation analysis,only the three features related to troughs are highly correlated with capacity.Combine the BP neural network to build a battery pack health assessment model.Verified by the test set,the RMSE between the estimated value and the actual value is 1.174.After adding temperature as a feature,the RMSE dropped to 0.983.After reconstructing the features with SVR,the RMSE is reduced to 0.341.The experimental results show that the battery pack capacity can be more accurately estimated based on the selected four characteristics and the battery health status can be evaluated accordingly. |
PDF Full Text Request