A novel ant colony-optimized extreme gradient boosting machine for estimating compressive strength of recycled aggregate concrete

摘要

Utilization of recycled aggregates generated from demolition waste for concrete production is a viable option for reducing the environmental impact of the construction industry on the environment. Compressive strength (CS) is a crucial parameter of recycled aggregate concrete (RAC), which is required in the design and construction phases of concrete structures. This study proposes and verifies a novel integration of ant colony optimization (ACO) and extreme gradient boosting machine (XGBoost) as an integrated data-driven approach for estimating the CS of RAC. A large-scale dataset, including 1100 samples, is collected from previous experimental works to construct the integrated approach, named ACO-XGBoost. Contents of cement, silica fume, fly ash, water, natural aggregates, recycled aggregates, and curing age are employed as influencing factors. Experimental results, consisting of 20 independent runs, point out that ACO-XGBoost achieves outstanding prediction accuracy with a root mean square error of 4.98, a mean absolute percentage error of 8.95, and a coefficient of determination of 0.93. The newly proposed method has gained a roughly 32 improvement in terms of prediction accuracy compared to benchmark approaches. In addition, an asymmetric loss function is employed in the training phase of XGBoost to decrease the number of overestimated cases by roughly 11. Hence, ACO-XGBoost is a promising decision support tool for designing RAC mixes.