Research On Power Battery Fault Identification And SOH Estimation Based On Data Drive

Against the backdrop of energy shortages,environmental pollution and national efforts to reduce energy consumption and emissions,the total number of new energy vehicles sold each year is climbing.At present,lithium-ion batteries are used by major car manufacturers as a source of power for new energy vehicles,but they have a number of potential safety hazards,and there are often reports of spontaneous combustion in new energy vehicles in the media,which makes people more concerned about the safety of new energy vehicles.In order to reduce or even avoid the safety problems of pure electric vehicles such as spontaneous combustion,this paper focuses on the power battery of pure electric vehicles,based on the historical data of various parameters related to the power battery collected by various sensors at the vehicle end of a vehicle monitoring platform,with the use of K-means++ clustering algorithm to analyse these data,and then identify the voltage inconsistency of the power battery At the same time,the GSCV-XGBoost algorithm is used to estimate the remaining charge capacity of the actual vehicle.By assessing the health status of the battery pack,the estimation results are fed back to the battery management system,and the battery is then reasonably and effectively regulated and managed,thereby extending the cycle life of the battery pack and reducing potential safety hazards.The details of the research in this paper are as follows:(1)The voltage values of each individual unit are dimensionally expanded into a twodimensional voltage feature point set.Based on the K-means clustering algorithm,a power battery voltage inconsistency fault identification method based on dynamic k-value K-means++clustering is proposed by optimising the number of clusters and the initial centre(cluster core)selection.Based on this method,the historical operating data of 105 vehicles(of the same model)driven up to 10,000 km without any abnormal alarm information were analysed,and the outlier factor(OF)threshold for vehicles without alarms was determined to be 0.02 by combining the box plot method.The analysis was conducted on the only vehicle with "poor battery unit consistency" alarm(the same model as the 105 vehicles mentioned above),and it was found that the K-means clustering method was as early as 2 days earlier than the alarm time of the vehicle monitoring platform,while the proposed method could identify the abnormal unit as early as 6 days earlier,which verified that the proposed method could identify the The power battery voltage inconsistency fault was identified earlier.(2)A State Of Health(SOH)method is proposed to correct the health state of the power battery based on real vehicle data using historical vehicle charging data.The charging segment data of 500 vehicles of the same model with a cumulative mileage of 1000 km were extracted from a fixed total voltage range during slow charging(current <20A),and the partial charging capacity was calculated using the ampere-time integration method,and the maximum partial charging capacity within the specified range was counted according to the distribution of the mean value of charging current in different segments and the distribution of the range of charging duration,and the maximum partial charging capacity within each range was The ratio of the current partial charge capacity to the maximum partial charge capacity within the specified range is the corrected SOH.(3)A GSCV-XGBoost-based SOH estimation method for power batteries is proposed using historical vehicle operation data under actual operating conditions.The vehicle historical operation data is divided into segments according to the "discharge+charge" mode,and the probability of vehicle discharge multiplier distribution,charging current,charging time used to charge partial capacity,accumulated driving mileage,battery probe temperature and single cell consistency(outlier OF minimum of single cell voltage characteristic point)are extracted from the discharge and charging segments.A total of 16 capacity decay factors in 6 categories were used as inputs to the GSCV-XGBoost partial charge capacity estimation model,and partial charge capacity was extracted from the charging segment data with a fixed total voltage range as the model output,the original dataset for the model was constructed and the dataset was partitioned,the model was optimized in the training set using Grid Search CV,and in the test set The estimated partial charging capacity was compared with that estimated by XGBoost,GBDT and linear regression,and the model outperformed the other models.The current partial charge capacity estimated by the model is then corrected for the true SOH using a SOH correction method based on real vehicle data.