王红珂,刘啸天,林磊,孙海涛,吕云鹤,张晏玮,薛飞.机器学习在材料服役性能预测中的应用[J].装备环境工程,2022,19(1):11-19. WANG Hong-ke,LIU Xiao-tian,LIN Lei,SUN Hai-tao,LYU Yun-he,ZHANG Yan-wei,XUE Fei.Application of Machine Learning in Predicting Service Performance of Materials[J].Equipment Environmental Engineering,2022,19(1):11-19. 机器学习在材料服役性能预测中的应用 Application of Machine Learning in Predicting Service Performance of Materials 投稿时间：2021-05-26 修订日期：2021-07-25 DOI： 10.7643/issn.1672-9242.2022.01.002 中文关键词 : 数据挖掘 机器学习 服役性能 材料工程 模型预测中图分类号:TG172 文献标识码:A 文章编号:1672-9242(2022)01-0011-09 英文关键词 : data mining machine learning service performance materials engineering model prediction 基金项目 : 国家重点研发计划(2017YFB0702200) WANG Hong-ke Suzhou Nuclear Power Research Institute, Suzhou 215004, China LIU Xiao-tian Suzhou Nuclear Power Research Institute, Suzhou 215004, China LIN Lei Suzhou Nuclear Power Research Institute, Suzhou 215004, China SUN Hai-tao Nuclear and Radiation Safety Center, Beijing 100082, China LYU Yun-he Nuclear and Radiation Safety Center, Beijing 100082, China ZHANG Yan-wei Suzhou Nuclear Power Research Institute, Suzhou 215004, China XUE Fei Suzhou Nuclear Power Research Institute, Suzhou 215004, China 针对材料服役性能预测存在误差大、计算复杂、适用性差等问题，提出了基于数据挖掘的机器学习预测方法。首先阐述了机器学习的应用流程，并总结了常用模型原理及其在材料性能预测中的应用。然后采用多种机器学习模型对RPV钢的辐照性能进行预测，并通过Stacking集成方法提高了模型的预测精度。结果表明，机器学习可用于材料服役性能预测，具有较高的预测精度和可靠性。根据材料服役数据的不同特征选择合适的学习模型，同时进行模型融合和参数优化，可有效提高模型的预测精度及运算速度。 英文摘要 : Aiming at the problems of large error, complex calculation and poor applicability in the prediction service performance of materials, machine learning (ML) based on data mining was proposed. Firstly, the application process of ML is elaborated. Then, the principle of common models and its application in material performance prediction are summarized. Then, various ML models were used to predict the irradiance properties of RPV steel. Furthermore, the prediction accuracy was improved by Stacking integration method. Results show that ML can be used to predict the service performance of materials with high accuracy and reliability. Appropriate models should be selected according to diverse characteristics of materials service data. Model fusion and parameters optimization can improve the prediction accuracy and calculation speed of the ML model effectively. 查看全文 查看/发表评论 下载PDF阅读器