基于XGBoost的电动汽车用异步电机全工况及高精度的电流预测方法研究

Prediction Method of Full-condition and High-precision Current of Asynchronous Motors for Electric Vehicles Based on XGBoost

在工程中,使用李斯特(anstaltfur verbrennungskraftmaschinen list,AVL)测功机标定转矩-转速点对应电流值的精度往往过低。使用电驱系统查电流表时,表的精度也对控制结果产生影响。该文研究一种基于机器学习算法的电动汽车用异步电机全工况、高精度电流预测方法,以解决AVL标定电流表精度低的问题。首先,建立异步电机最大转矩的数学模型,分析励磁电流isd及转矩电流isq精度对电磁转矩输出的影响;然后,通过ANSYS模拟实验数据点的精度,提取电流、电压、磁链等数据,使用极端梯度增强算法(eX treme gradient boosting,XGBoost)分析变量重要性,优化模型结构,合理选择模型输入、输出;最后,搭建AVL实验平台,采集实验数据,使用XGBoost算法建模并预测电流,提高电流表精度后,代入系统验证,通过仿真和实验证明该方法的有效性。

In engineering,the accuracy of using AVL dynamometer to calibrate the current value corresponding to the torque-speed point is often too low.At the same time,when using the electric drive system to check the current table(excitation current(isd)),torque current(isq),the accuracy of the current table also affects the control results.Therefore,this paper studied a prediction method of full-condition and high-precision current of asynchronous motors for electric vehicles based on machine learning algorithms to solve the problem of low accuracy of AVL calibration current table.First,the mathematical model of the maximum torque of asynchronous motors was established,and the influence of the accuracy of isd and isq on the output of electromagnetic torque was analyzed.Then,the accuracy of the experimental data points was simulated by ANSYS,the current,voltage,flux and other data was extracted.The eXtreme gradient boosting(XGBoost)algorithm is used to analyze the importance of variables,optimize the model structure,and reasonably select input and output of the model.Finally,after the construction of the AVL experimental platform and the collection of experimental data,the XGBoost algorithm was used for modeling,predicting the current,and improving the accuracy of the current table whose accurate data ware brought into the system for verification.Simulation and experiment prove the validity of the method.