Research On Fault Diagnosis Methods Of Electric Vehicle Power Battery System Based On Big Data

As a key subsystem of electric vehicle,battery system performance is affected by many factors such as driving conditions and environmental conditions.Its external characteristics are characterized by high real-time performance,high nonlinearity,and strong coupling.These lead to the battery faults that occur during working condition are difficult to diagnose in short period of time.According to the characteristics of battery faults,based on National Monitoring and Management Center for NEVS,this paper combines mathematical statistics and machine learning models to statistically analyze the battery fault alarms and in-depth analyze two potential battery faults(voltage outlier and capacity abnormality).The specific research work in this paper includes:(1)Combining fault tree analysis methods,the common fault types in battery system are classified,and the causes for each type are studied.Based on the fault types,the fault trees for GB/T 32960 data items are established.(2)Combining mathematical statistics with electric vehicle big data,the data preprocessing methods for battery research are summarized.Battery fault alarms are statistically analyzed and correlated from three dimensions(fault type dimension,time dimension,and mileage dimension).(3)Combining with the outlier analysis method and EV big data,a 3σ multi-level fault filtering algorithm for large-scale data abnormalities was constructed to effectively filter the fault data of complex multi-order;According to the regulation of battery consistency and characteristics,a battery voltage outlier analysis model was established,and a fault diagnosis method for cell voltage outliers was constructed.(4)Combining the machine learning algorithm(XGBoost),a battery capacity prediction model is constructed;Based on the model,an abnormality detection system for battery capacity is established,and the fault analysis for the abnormal battery capacity of the two input modes(SOC window input,voltage window input)are performed.And the real vehicle data are selected to verify the analysis results.