基于机器学习方法再生混凝土装配式节点剪切性能预测

Prediction of Shear Performance of Recycled Aggregate Concrete Prefabricated Joints Based on Machine Learning Methods

摘要: 本研究针对建筑废弃物资源化中再生混凝土(RAC)与超高性能混凝土(UHPC)组合构件的极限抗剪承载能力，通过正交试验与有限元方法开展研究。以RAC强度、配筋率、UHPC强度、剪切键个数为4个关键因素(各4水平)，设计16组试验方案，采用ABAQUS软件建模分析，获取16组极限荷载数据。基于此数据集，运用偏最小二乘回归(PLS)构建承载力预测模型，经mapminmax标准化预处理、特征工程扩展(4维基础特征增至7维增强特征)及留一交叉验证优化(最优潜变量数为2)后，模型的决定系数R²达0.9159，较基础模型提升8.5%。变量重要性分析显示，RAC强度及其平方项(VIP值分别为1.6784、1.6861)是影响承载力的关键因素。残差分析表明，预测误差均值−0.214 kN，多集中于±20 kN内且呈随机分布。实验证实在本研究的参数范围内，所构建的PLS模型展现出对UHPC构件承载力的高精度预测能力，为工程设计提供初步的理论参考。

Abstract: This study focuses on the ultimate shear capacity of composite members made of recycled aggregate concrete (RAC) and ultra-high performance concrete (UHPC) in the context of construction waste recycling. Research was conducted using orthogonal experiments and finite element methods. Four key factors, namely RAC strength, stirrup ratio, UHPC strength, and number of shear keys, were selected, with each factor set at 4 levels. A total of 16 experimental schemes were designed, and ABAQUS software was used for modeling and analysis to obtain ultimate load data for 16 groups of specimens. Based on this dataset, a partial least squares regression (PLS) model was established for predicting bearing capacity. After data preprocessing via mapminmax normalization, feature engineering expansion (increasing from 4-dimensional basic features to 7-dimensional enhanced features), and optimization through leave-one-out cross-validation (with the optimal number of latent variables determined as 2), the coefficient of determination (R²) of the model reached 0.9159, representing an 8.5% improvement compared to the basic model. Variable importance analysis showed that RAC strength and its square term (with VIP values of 1.6784 and 1.6861, respectively) were the key factors affecting the bearing capacity. Residual analysis indicated that the average prediction error was −0.214 kN, and most errors were concentrated within the range of ±20 kN with a random distribution. This study confirms that within the parameter range investigated in this paper, the constructed PLS model exhibits high-precision predictive capability for the bearing capacity of UHPC members, thereby providing preliminary theoretical references for engineering design.