Prediction of frost resistance and multiobjective optimisation of low-carbon concrete on the basis of machine learning

Abstract

The use of Portland cement is a major source of carbon dioxide emissions, causing considerable environmental pollution worldwide. Therefore, reducing carbon emissions in concrete is of paramount importance. This study presents a multiobjective optimisation method for low-carbon concrete mix design, combining the use of optimal machine learning models for prediction, with the aim of minimising the cost and carbon emissions of concrete. Several optimisation models, including SVR, CatBoost, RF, XGBoost, DBO-RF and VMD-DBO-RF, were employed to predict the frost resistance of low-carbon concrete. The research findings demonstrate that the VMD-DBO-RF optimisation algorithm achieves high prediction accuracy in terms of RDEM and MLR. For RDEM prediction, R2 = 0.9346, MSE = 0.0508, MAE = 0.1686, and RMSE = 0.2253. For MLR prediction, R2 = 0.9557, MSE = 0.0245, MAE = 0.1042, and RMSE = 0.1565. Furthermore, the model’s interpretability was improved by combining SHAP and PDP analyses.