Hybrid genetic algorithm and support vector regression for predicting the shear capacity of recycled aggregate concrete beam

The application of recycled aggregate (RA) as a replacement for natural aggregate (NA) in concrete members brings benefits in both cost reduction and environment protection. However, the inferior quality of RA results in significant decrease in the mechanical properties of recycled aggregate concrete (RAC) compared to NA. The shear capacity of RAC is one of the challenges for its application. To address this problem, high accurate calculation of shear capacity of RAC beam and its optimal design become urgent. Here, a data-driven hybrid methodology implemented by support vector regression (SVR) with the optimal parameters from genetic algorithm (GA) optimization approach called SVR-GA is proposed to predict the shear capacity of RAC beams, which achieves the minimal root mean square error (RMSE) of 0.3949 MPa under five-fold cross-validation, yielding the best performance among RF, SVR and ANN, and the RMSEs and MAPEs of which are all lower than that of existing conventional methods. The results indicate that GA is efficient for tuning hyper-parameter of SVR model. The hybrid SVR-GA model can achieve the best prediction accuracy in prediction the shear strength of RAC beam, with the highest correlation coefficient and lowest RMSE. Moreover, parametric sensitivity analysis reveals beam width, shear span ratio and stirrup spacing are the most sensitive variables and the replacement ratio of RA has little effect on shear strength of RAC beam. Furthermore, to meet the targets of lowest cost, minimum CO₂ emissions and maximum shear strength in the design RAC beams, SVR-GA model combining Monte Carlo method was employed for multi-objective optimization discovering the trade-offs among cost, CO₂ emissions and shear strength of RAC beams by the Pareto front. The proposed approach can be used as guidance in the design of RAC beam with higher accuracy and convenience.