基于DE-Xgboost的U71Mn钢粗糙度预测模型

Roughness Prediction Model of U71Mn Steel Based on DE-Xgboost

摘要:

U71Mn高锰钢为我国铁轨主要原材料，当铣削参数配置不合理时易导致金属表面马氏体粗大造成加工硬化，难以满足使用要求。针对此问题使用M-V5CN铣削U71Mn高锰钢获取了1000组切削数据集，建立了基于Xgboost算法的表面粗糙度预测模型，作为非线性模型其训练参数众多为最大化Xgboost模型性能,提出一种改进的混合编码DE算法进行模型超参数优化。模型建立完成后，经测试较未经优化的Xgboost最大误差下降7.4%，平均绝对误差下降11.7%，方差降低6.4%,且较主流DNN、GA-SVM模型性能提升明显可以更有效承担U71Mn高锰钢粗糙度预测任务。

Abstract:

U71Mn high manganese steel is the main raw material of railway track in China. When the milling parameters are not reasonable, it is easy to cause large martensite on the metal surface and work hardening, which is difficult to meet the requirements of use. In order to solve this problem, 1000 sets of cutting data sets were obtained by milling U71Mn high manganese steel with m-v5cn, and a surface roughness prediction model based on xgboost algorithm was established. As a nonlinear model, many of its training parameters were to maximize the performance of xgboost model. An improved hybrid coding DE algorithm was proposed to optimize the model parameters. After the establishment of the model, the maximum error of xgboost decreases by 7.4%, the average absolute error decreases by 11.7%, and the variance decreases by 7.4%. Compared with the mainstream DNN and GA-SVM models, the performance of xgboost model is significantly improved, and it can effectively undertake the task of roughness prediction of U71Mn high manganese steel.