Enhancing the performance of recycled aggregate green concrete via a Bayesian optimization light gradient boosting machine and the nondominated sorting genetic algorithm-III

The application of recycled aggregate concrete (RAC) is important for the sustainable development of construction materials, but few studies of mix proportions have considered how to achieve the required mechanical properties, carbon emissions (CEs), and costs. A hybrid intelligence framework based on Bayesian optimization (BO), a light gradient boosting machine (LGBM) and the nondominated sorting genetic algorithm (NSGA)-III is proposed to predict RAC performance and optimize the mix proportion. The results indicate that (1) BO can be used to effectively search for hyperparameters and identify the optimal solution, effectively completing the hyperparameter optimization search task. (2) The proposed BO-LGBM model exhibits excellent performance. According to the prediction results for 28-day compressive strength (CS), CEs and cost, the goodness of fit is in the range of 0.991-0.997. (3) The BO-LGBM-NSGA-III intelligent mixing algorithm can effectively achieve multiobjective optimization of the mixing proportion for RAC. The percentage of recycled coarse aggregate in the optimized solution was 8.2%, which resulted in a 6.71 % increase in CS, a 25.12% decrease in CEs, and a 2.42% decrease in cost compared with other methods. This finding provides a new approach for the intelligent multiobjective optimization of the RAC mix proportion.