Research On Battery State Of Health Prediction Model Based On Real Vehicle Data

The prediction of battery state of health(SOH)has always been a hot research topic in the field of electric vehicles.Accurate battery state of health prediction can ensure the safe and reliable operation of the power battery system,optimize the operation of the power battery system,and provide a basis for the energy management and safety management of electric vehicles.However,most of the current battery state of health prediction models rely on battery test experiments,which are not necessarily suitable for battery state of health prediction during the actual operation of electric vehicles.Therefore,based on the real vehicle data recorded by the new energy vehicle big data platform,this paper uses data mining techniques to study the battery state of health prediction model under actual operating conditions.First,the abnormal data in the real vehicle data set is processed,and then the battery state of health prediction model is established based on two different dimensions,time series and ampere-hour throughput,some works have been done and summarized as follows:(1)Abnormal data processing in real vehicle data.Considering the influence of factors such as electromagnetic interference,equipment quality and network instability,it is inevitable that abnormal data will appear in the real vehicle data set.In order to improve the efficiency of data mining,K-Means clustering algorithm is used to process the abnormal data in the real vehicle data set.Statistical analysis of the data set after removing abnormal data,combined with the reference range of normal data,shows the effectiveness of the method.(2)Battery state of health prediction model based on time series.According to the charging data segment extracted from the real vehicle data set,the capacity data sequence is obtained through battery capacity calculation,capacity correction and data screening.And use the proposed adaptive Kalman filter model to filter the noise of the capacity data sequence.After obtaining the battery state of health sequence distributed by time according to the capacity definition method,the battery state of health models under three different function curves are compared and analyzed.Combined with the Arrhenius formula of the battery aging semiempirical model,a power function curve is used to establish a battery state of health prediction model based on time series.The prediction error results show that the established prediction model can predict the attenuation relationship of the battery state of health with time,and the effect is better than the average method.(3)Battery state of health prediction model based on ampere-hour throughput.According to the objective phenomenon that the attenuation degree of the battery state of health is largely related to the cumulative usage of the vehicle,the rain flow counting method is used to decompose the battery load history into several battery charge-discharge cycles.Considering the battery charge-discharge cycle,the battery ampere-hour throughput is classified and counted.Due to the different attenuation rates of batteries in different SOC cycle intervals,a battery state of health prediction model based on ampere-hour throughput is proposed by using different influence coefficients.The prediction results of the models under different scale training sets show that when the proportion of training data reaches or exceeds 50%,the proposed battery state of health prediction model based on ampere-hour throughput can predict the battery state of health within 1% error range.