Data Driven-Based Fault Diagnosis Of Power Battery For Pure Electric Vehicles

Faced with energy crisis,environment and traffic pollution,worldwide attention is paid to promoting the development of new energy vehicles.As a key component of pure electric vehicles（EVs）,the power battery is one of the main failure sources of EVs under complex working conditions,and its performance has big influence on the safety of the electric vehicle.Based on the data of the National Monitoring and Management Center for New Energy Vehicles（NMMC-NEV）,this paper proposes data driven-based fault diagnosis of power battery for pure EVs,which takes lithium ion battery as the research object and provides the technical means for big data application of new energy vehicles.Data validity and availability are the prerequisites for data analysis and power system fault diagnosis in this paper.Firstly,aiming at the validity of voltage data for EVs in NMMCNEV,missing value handling mechanism is proposed to deal with continuous voltage and discontinuous voltage value loss.Fault threshold judgment based on the basic method of safe battery cell voltage can quickly analyze fault pre-alarm,but this method lacks the ability of diagnostic analysis.In the early stage of power battery system application,the voltage data generally obeyed the normal distribution.According to the continuous sample data,abnormal cells can be selected and the long-term warning and the system defect analysis can be realized by Pauta criterion.With the aging of power battery system,voltage data gradually deviate from the normal distribution,so a more effective data-driven approach should be taken.Thus a fault diagnosis of power battery system strategy is suggested to optimize fault diagnosis processes and resources.Secondly,the real-time voltage data of the non-normal distribution is used to diagnose and evaluate the abnormal battery cell using the Entropy-based weight and the dynamic BIAS method.（1）Based on the Shannon entropy theory,a modified Entropy-based weight method is proposed to evaluate each cell by calculating weight of each moment index.（2）Based on the moving average,the change of each battery cell at each moment is reflected by the dynamic BIAS at that moment,and a ten-minute dynamic BIAS concept is proposed.Then,the battery cell fault diagnosis method can be established by the evaluation index standard deviation of dynamic BIAS.Finally,in terms of the historical data of non-normal distribution voltage,the deviations of each cell voltage proportion under the time series are taken as the processing targets.The95 th percentile is used as the cell evaluation threshold of the same battery pack.By the comparison of the cumulative value of the proportion deviation in different battery pack,a battery consistency mechanism based on statistical diagnosis can be obtained.