基于机器学习算法的混凝土28 d抗压强度预测

The Prediction of 28 d Compressive Strength of Concrete Based on Machine Learning Algorithms

为较准确预测混凝土28 d的抗压强度,以水泥、粉煤灰、砂、碎石和水5种原材料的单位用量作为输入,混凝土28 d抗压强度作为输出,构建了一个拥有249组样本的数据库。运用决策树、支持向量机、高斯过程回归、集成学习、神经网络5种机器学习算法进行超参数优化,构建了混凝土28 d抗压强度预测模型,并对其精度进行了评估。对比5种机器学习算法结果可知:高斯过程回归模型是预测混凝土28 d抗压强度的最优预测模型;基于构建的高斯过程回归模型对无岳高速WYTJ-07标段隧道工程自制的5种花岗岩混凝土28 d抗压强度进行了预测,预测值的最大相对误差为8.95%,证明该模型预测精度良好,可靠性高。

In order to accurately predict the 28 d compressive strength of concrete,a database with 249 groups of samples was constructed with the unit dosage of cement,fly ash,sand,gravel and water as the input and the 28 d compressive strength of concrete as the output.5 machine learning algorithms(namely decision tree,support vector machine,Gaussian process regression,ensemble learning and neural network)were used to optimize the parameters.The prediction models of 28 d compressive strength of concrete were constructed and their accuracies were evaluated.The results of the 5 machine learning algorithms show that the Gaussian process regression model is the optimal model for predicting 28 d compressive strength of concrete.Based on the constructed Gaussian process regression model,the 28 d compressive strength of five kinds of granite concrete prepared by the project department of WYTJ-07 section of Wuyue Expressway tunnel project is predicted.The maximum relative error between the predicted and actual values is 8.95%,proving that the constructed model has good prediction accuracy and high reliability.