基于机器学习的混凝土自生收缩预测算法与解释

Prediction Algorithm and Interpretation for AutogenousShrinkage of Concrete Based on Machine Learning

目的 研究混凝土自生收缩的多因素作用机理,建立适用于混凝土自生收缩预测的机器学习模型,增强机器学习算法的可解释性。方法 将水灰比、水胶比等14个指标作为输入变量,混凝土自生收缩值作为输出变量进行预测;采用BPNN、SVM、RF及XGBoost算法建立混凝土自生收缩预测模型,基于判定系数R2、均方根误差RMSE及平均绝对误差MAE,选取最适用于混凝土自生收缩的预测模型;采用SHAP法解释输入变量对输出变量的贡献程度、相关性及各输入变量的作用机理。结果 相较于其他算法而言,XGBoost算法可以有效预测混凝土的自生收缩,此时得到的R2、RMSE及MAE分别为0.956、0.055及0.026。结论 骨灰比是影响混凝土自生收缩的关键变量;骨灰比、高吸水树脂掺量等指标与混凝土自生收缩呈现负相关;时间与硅灰掺量等指标与混凝土自生收缩呈现正相关;采用SHAP法可以有效解决机器学习模型存在的黑盒问题,提高模型的可解释性。

In order to study the working mechanism of concrete autogenous shrinkage with multiple factors,a machine learning model suitable for predicting concrete autogenous shrinkage is established to enhance the interpretability of machine learning algorithms.14 indexes such as water-cement ratio and water-binder ratio were used as input variables while concrete autogenous shrinkage was used as output variable;BPNN,SVM,RF,and XGBoost algorithms were used to establish the prediction model of concrete autogenous shrinkage,and the best prediction model was obtained by comparing the coefficient of determination(R 2),root means square error(RMSE)and mean absolute error(MAE);the SHAP method was used to explain the degree of contribution of the input variables to the output variables,the correlation and the mechanism of each input variable.The results show that the XGBoost algorithm can effectively predict the autogenous shrinkage of concrete compared with the remaining three algorithms,the mean values of R 2,RMSE,and MAE are 0.956,0.055,and 0.026,respectively;the aggregate-cementratio is the critical variable affecting the autogenous shrinkage of concrete,and variables such as aggregate-cement ratio,the content of highly absorbent resin have a negative correlation with autogenous shrinkage of concrete while time and silica fume content have a positive correlation.The SHAP method can effectively solve the black box problem of the machine learning model and improve the interpretability of the model.