基于随机森林算法的混凝土早期抗裂性预测研究

将随机森林(Random Forest)回归算法应用于混凝土早期抗裂性研究.以松通项目混凝土为例,基于大量文献和工程经验选取了7个主要影响因素的混凝土早期抗裂性指标体系.以原始数据建立训练样本集和测试集,通过计算基于Bootstrap自助重抽样得到的袋外数据(OOB)的模型误判率,确定随机森林预测模型的最优参数,并对影响因素进行重要性排序,利用Pearson相关性图分析各影响因素相关度,然后建立RF训练模型,输出模型训练集和预测集的预测拟合结果.通过RMSE和R~2值分析模型的预测精度,并将预测结果与BP神经网络和小波神经网络模型对比.结果显示,随机森林预测模型误差最小,精度最高,验证了模型的准确性和可靠性.提出的随机森林预测模型为实现混凝土早期抗裂性预测提供了一种有效的方法.

Prediction of Early Crack Resistance of Concrete Based on Random Forest Algorithm

The Random Forest(RF)regression algorithm is applied to the early cracking resistance research of concrete.Taking Songtong Project Concrete as an example,based on a large amount of literatures and engineering experience,seven early impact resistance index systems of concrete are selected.The training sample set and test set from the original data are established,and the model misjudgment rate is calculated based on the out-of-bag data(OOB)from Bootstrap self-resampling to determine the optimal parameters of the Random Forest Prediction Model and make important factors rank.The Pearson correlation diagram is used to analyze the relevance of various factors,and finally the RF training model is established,and the fitting result of the prediction model training set and prediction set are output.The prediction accuracy of the model is analyzed by the values of RMSE and R~2,compared with the BP neural network and wavelet neural network models.The results show that the Random Forest Prediction Model has the smallest error and the highest accuracy,furthermore the accuracy and reliability of the model are verified.The proposed Random Forest Prediction Model provides an effective method for realizing the prediction of early crack resistance of concrete.