| According to the research trend of the state of health(SOH)estimation of lithium-ion batteries at home and abroad,we found that the data-driven methods have become a research hotspot in this field.The data-driven method of feature engineering based on the degradation mechanism of battery can avoid the black box problem of traditional data-driven methods.At the same time,most of the research objects in the existing literature are single cells,and there are few studies on electric vehicle power battery packs.Therefore,this paper takes lithium-ion single cells under different working conditions and power battery packs under real vehicle conditions as the research objects to carry out the research on the state of health estimation of lithium batteries based on data-driven approach.The following are the innovations and achievements of the study:1.In this study,we analyzed the degradation mechanism of lithium-ion batteries,and based on this,we carried out data exploration and feature engineering.On this basis,we proposed the key feature of equal charge interval electric work value that can be applied to both single cell and power battery pack,which also covers the information of voltage and current changes during battery charging,and has more characteristic information than that of equal charge capacity HI.To preserve more useful feature information in the looping data,we also selected a series of auxiliary features.The multicollinearity problem caused by these features is solved by elastic network with regularization technology.2.Through the elastic network modeling analysis of the lithium-ion cell data under different working conditions,we found that the better the state consistency of the charging curve in the battery experimental cycle,the smaller the prediction error.Therefore,we adopt the consistency evaluation method of extracting the information entropy of the charging curve,calculating the local outlier factor of the information entropy in terms of battery,and then performing K-means clustering analysis based on the LOF.Through the comparative analysis of modeling and prediction,the role of charge curve consistency evaluation in improving the estimation accuracy is verified.3.When modeling the electric vehicle power battery packs,we found that the capacity calibration error of the battery working cycle was too large.This error may be caused by superimposed factors such as excessive data uplink time,large current charging,and fluctuations in the current at the later stage of charging in the power battery pack dataset.In order to control the error of power battery capacity calibration,we designed a 10 step cycle noise reduction method based on boxplot’s idea.This method not only controls the error,but also takes into account the diversity of feature sources,and finally achieves good prediction performance in the scenario of large data error.By modeling and analyzing the health state of lithium-ion single cells and power batteries,we verified the validity,generalization ability and interpretability of the battery health state feature HI and elastic network model methods proposed in this paper,and achieved the research purpose.Finally,it summarizes the shortcomings of this study and prospects for future research. |
