Compressive strength resistance coefficient of sustainable concrete in sulfate environments: Hybrid machine learning model and experimental verification

Abstract

The compressive strength corrosion resistance coefficient of sustainable concrete (SC-SCRC) is a crucial indicator for evaluating the mechanical performance degradation of SC after erosion. Therefore, this study collected 326 sets of SC-SCRC data from 20 literature sources and established an optimized machine learning algorithm-based SC-SCRC prediction model, evaluating its performance. To validate the model's accuracy, thirteen different SC mix ratios were prepared for SC-SCRC testing. The results demonstrate that the grey wolf optimization supported vector regression (GWO-SVR) model provides predictions that better align with the actual values. The model exhibits lower standard deviation and mean of model residuals. The performance evaluation indicators of the GWO-SVR model (R=0.972; MSE=1.46e-3; MAE=2.9e-2; MAPE=3.1e-2; RMSE=3.9e-2) are also outstanding. Furthermore, Shapley Additive Explanations reveal that the dry-wet cycle count and the water-binder ratio are the two most critical influencing factors on SC-SCRC, with both showing negative correlations. A graphical user interface (GUI) based on the GWO-SVR model has been developed to enable efficient predictions of SC-SCRC. Test results indicate that the error between the experimental values and the predicted values is less than 6%, demonstrating the model's high accuracy and generalizability. This work lays the foundation for optimizing the performance of SC and predicting SC-SCRC. The findings provide valuable guidance for engineering projects in areas with high concentrations of sulfate ions.