机器学习算法用于自密实混凝土性能设计的研究进展  

Research Progress of the Use of Machine Learning Algorithm in Performance Design of Self-compacting Concrete

作　　者：龙武剑[1,2] 罗盛禹 程博远 冯甘霖 李利孝 LONG Wujian;LUO Shengyu;CHENG Boyuan;FENG Ganlin;LI Lixiao(College of Civil and Transportation Engineering,Shenzhen University,Shenzhen 518060,Guangdong,China;Guangdong Provincial Key Laboratory of Durability for Marine Civil Engineering,Shenzhen 518060,Guangdong,China)

机构地区：[1]深圳大学土木与交通工程学院,广东深圳518060 [2]广东省滨海土木工程耐久性重点实验室,广东深圳518060

出　　处：《材料导报》2024年第11期104-113,共10页Materials Reports

基　　金：国家自然科学基金-山东联合基金(U2006223);深圳市科技计划项目(JCYJ20190808151011502);广东省重点领域研发计划项目(2019B111107003)。

摘　　要：为了厘清自密实混凝土材料各组分与其工作性能、力学性能、耐久性能间的复杂耦合影响机制,近年来,机器学习方法被越来越多地应用于自密实混凝土配合比设计与优化以及性能分析方面的研究。机器学习方法具有在复杂数据集中学习数据间的内在规律或映射关系的能力,在自密实混凝土设计领域被认为是构建混凝土原材料配合比与性能间映射关系模型的一种具有广阔前景的技术手段。然而,机器学习方法由于其依赖的数据集难以满足以及算法架构的可解释性较差等因素限制,使得基于机器学习在自密实混凝土性能设计领域面临着一系列挑战。本文系统总结梳理当前机器学习在自密实混凝土各项性能设计的应用情况,重点分析数据驱动的机器学习算法应用于自密实混凝土设计领域时面临的主要技术难点:高维度与小样本数据的难题、性能多目标优化难题、模型复杂而缺乏可解释性难题;归纳总结机器学习在自密实混凝土材料性能设计领域应用的发展趋势及未来发展方向。In order to clarify the complex coupling mechanisms between the components of self-compacting concrete materials and working, mechanical and durability performances, machine learning methods have been increasingly applied to the design and optimization of self-compacting concrete mixes and performance analysis in recent years. Machine learning methods have the ability to learn intrinsic patterns or mapping relationships between data in complex data sets, and are considered a promising technical tool in the field of self-compacting concrete design for mode- ling mapping relationships between concrete raw material mixes and performances. However, machine learning methods face a number of challenges in the field of self-compacting concrete performance design based on machine learning due to the limitations of its reliance on unsatisfiable data sets and poor interpretability of algorithm architectures. This paper systematically summarizes and compares the current applications of machine learning in the design of self-compacting concrete performances, focusing on the main technical difficulties faced by data-driven machine learning algorithms when applied to the field of self-compacting concrete design:the challenges of high dimensionality and small sample data, the challenges of multi-objective optimization of performances, and the challenges of complex and uninterpretable models. The paper also summarizes the development trends and future directions of machine learning applications in the field of self-compacting concrete material performances design.

关 键 词：机器学习 自密实混凝土 配合比设计 性能预测 

分 类 号：TU528[建筑科学—建筑技术科学]