Research On Lithium-ion Battery Health Management Based On Data-driven Method

Lithium-ion batteries have become one of the best choices for the power source of new energy vehicles due to their high energy density,long cycle life,and decreasing manufacturing costs year by year.The performance of Lithium-ion batteries continues to decline during use,putting the entire battery system at risk of unexpected failure.Therefore,we need to carry out health management for Lithium-ion batteries.Excellent health management methods are of great significance to the safety,reliability and economy of battery systems.Li-ion battery health management mainly includes two aspects:battery remaining life prediction and abnormal battery screening.The prediction of the remaining life of Lithium-ion batteries is of great significance for their operation and maintenance.The abnormal battery screening function conducts online detection of the battery,and gives an early warning when an abnormality is found,so as to avoid the occurrence of safety problems caused by battery abnormality.The data-driven approach can directly extract battery health characteristics from battery sampling data,establish battery remaining life prediction models and anomaly screening models,and avoid complex electrochemical modeling processes.Therefore,this paper chooses a data-driven approach to study the remaining life prediction and anomaly screening of Lithium-ion batteries.The main work of this paper is as follows:(1)Research on the extraction method of Lithium-ion battery health features.Firstly,this paper studies the aging and failure mechanism of Lithium-ion battery from the structure and working principle,which provides theoretical support for the research on battery health feature extraction method.Then,in view of the quality problems such as data missing,duplication,and abnormality in the process of data collection and transmission,data preprocessing methods such as deduplication and filling,abnormal point elimination,and data interpolation are proposed to provide high-quality data sets for battery health feature extraction.Finally,a battery health feature extraction scheme is designed from two dimensions:statistical characteristics and intrinsic characteristics of Lithium-ion batteries,which are used to characterize the consistency of the battery pack and the aging degree of the battery,respectively.(2)Prediction of the remaining life of Lithium-ion batteries based on a Gaussian process regression model.In this paper,the battery health characteristics are extracted from the real-time operating data and early operating data of the battery,respectively,and two Lithium-ion battery remaining life prediction models are established by using the GPR method.The former provides users with decision support for battery replacement,decommissioning and recycling.The experimental results show that the average R~2index of the GPR model is 0.9844,which achieves the effect of battery life prediction with an average error of less than 3%.The latter provides theoretical support for researchers to optimize the battery’s ability to adapt to specific use environments and adjust battery operating conditions.The experimental results show that the R~2evaluation index of the GPR model is 0.9193,and the prediction accuracy of battery life is less than 8%,reaching the preset target.(3)Abnormal battery screening method based on multi-dimensional indicators and entropy weight method.First of all,this paper proposes to use the entropy weight method to screen abnormal batteries in the battery system,extract multi-dimensional health feature indicators from the sampling data of Lithium-ion batteries,calculate the information entropy of each feature,and determine whether each feature index is in the abnormal evaluation system according to the information entropy value.Calculate the risk coefficient of each battery cell and module.The higher the risk coefficient,the greater the possibility of abnormal battery cells and modules.Then,using the MIT-Stanford public data set to verify the algorithm,the accuracy of the algorithm for abnormal battery screening reached 92.7%,which is much higher than the 85.4%accuracy of the sample entropy abnormal battery screening algorithm based on a single dimension index.Finally,we applied the algorithm to both the energy storage power station data set and the real vehicle data set and successfully screened out abnormal batteries.The algorithm application results of the data set of the energy storage power station and the real vehicle data set show that for the actual operation data set with poor data quality and complex working conditions,the algorithm can accurately locate the abnormal battery during the operation of the battery system.