基于机器学习的高温后聚丙烯纤维混凝土强度预测

影响高温后聚丙烯纤维混凝土(PFRC)力学性能的因素众多,因此相关试验的周期长,试验量大。如何利用现有试验数据预测高温后聚丙烯纤维混凝土的强度能够有效提高试验效率,为实际工程提供参考。通过研究纤维尺度、纤维掺量和温度对混凝土强度的影响,建立纤维尺度、掺量和温度为因子的回归树(RT)、支持向量机回归(SVR)和BP神经网络三种模型。将聚丙烯纤维混凝土在不同受热温度(20 ℃、200 ℃、400 ℃、600 ℃、800 ℃)下的劈裂抗拉强度和抗压强度试验值与预测值进行比较,结果表明:三种模型均能以较高的精度预测高温后聚丙烯纤维混凝土的劈裂抗拉强度和抗压强度;与实测值相比,三种模型预测值与实测值的相对误差基本控制在15%以内,个别数据出现较大预测误差;比较三种模型的平均绝对误差(MAE)和平均相关系数R²,人工神经网络(ANN)模型的预测结果较好,验证了基于机器学习的高温后聚丙烯纤维混凝土力学性能预测的可靠性。

Strength Prediction of Mechanical Properties of Polypropylene Fiber Reinforced Concrete after High Temperature Based on Machine Learning

There are many factors affecting the mechanical properties of polypropylene fiber reinforced concrete (PFRC) after high temperature. Therefore, the relevant experimental period is long and experimental volume is large. How to use the existing experimental data to predict the strength of PFRC after high temperature effectively improves the test efficiency and provide reference for practical projects. By studying the influences of fiber scale, fiber content and temperature on concrete strength, three models, namely regression tree (RT), support vector regression (SVR) and BP artificial neural network, were established. The experimental values of splitting tensile strength and compressive strength of PFRC at different heating temperatures (20 ℃, 200 ℃, 400 ℃, 600 ℃, 800 ℃) were compared with the predicted values. The results show that three models predict the splitting tensile strength and compressive strength of PFRC at high temperature with high precision. Compared with the measured value, the relative error between the predicted value and the measured value of three models are basically controlled within 15%, except for individual data. By comparing the mean absolute error (U_MAE) and average correlation coefficient R² of three models, the prediction results of artificial neural network (ANN) model are better, which verifies the reliability of machine learning in predicting the mechanical properties of PFRC after high temperature.