SOH Estimation And RUL Prediction Of Vehicle Li-ion Batteries Based On Multi-index Fusion

As the power source and core component of EV,the accurate and effective estimation of the State of Health(SOH)and the Remaining Useful Life(RUL)of the lithium-ion batteries is the key to ensure the safe operation of EV.At present,studies on SOH estimation and RUL prediction of lithium ion batteries are mostly based on laboratory data,and the health indicators considered for lithium ion batteries are relatively simple,which is difficult to apply to real vehicle data with strong real-time performance and complex actual operating conditions.This paper uses real vehicle data to study the SOH estimation and RUL prediction of EV lithium-ion batteries based on multi-index fusion.Specific research work is as follows:(1)The missing values and abnormal values in the real vehicle data were processed,and then the charge and discharge fragments were divided.The key data items were analyzed by using the charging and discharging fragments to reveal the performance decline rule of lithiumion batteries,and then the charging fragments that meet the research content of this paper were selected.(2)According to the segment characteristics of real vehicle data,the calculation methods of capacity,internal resistance and constant current charging time of lithium-ion batteries based on deep charging segments were studied.Firstly,ampere-hour integral method was used to calculate the segment capacity of different charging depths,the depth charging segments that best reflects the capacity decline trend were selected.Then,the equivalent circuit model of lithium-ion batteries based on the depth charging segments was constructed,and the parameters of the model were identified by adaptive forgetting factor recursive least square method.According to the parameters identified,the internal resistance of lithium-ion batteries based on the depth charging segments was calculated.Finally,the multi-stage constant current charging process of lithium-ion batteries was analyzed,and the constant current charging step was selected by using the deep charging segments,and the calculation result of the constant current charging time of lithium-ion batteries based on the deep charging segments was obtained.(3)In view of the inaccuracy of using only a single health indicator to estimate the SOH of lithium-ion batteries at present,a method for estimating the SOH of lithium-ion batteries for the whole period based on the fusion of multiple indicators was proposed.The method fuses deep-charge and discharge fragments of lithium-ion battery SOH based on a single health indicator estimate.Then,feature screening was carried out on the real vehicle data,and the filtered features were used as the input of Light GBM,DNN and CNN models.The results show that Light GBM model has the highest accuracy in estimating the SOH of the multi-index fusion lithium-ion battery in the whole period,and the root-mean-square error of its verification set is only 0.0514,which is at least 0.0353 less than that of the other two models.(4)The RUL of lithium-ion batteries was predicted by using multi-index fusion SOH time series data and Bi GRU-Attention model.First,the Bigru-attention model was constructed,then the model parameters were optimized,and then the prediction effect of the Bigru-attention model was compared with that of LSTM,GRU and Bi GRU models.The results show that Bi GRU-Attention model has the best prediction effect on lithium-ion battery RUL,and the rootmean-square error of its verification set is only 0.0035,which is at least 0.0014 less than that of the other three models.