Intelligent prediction of compressive strength of self-compacting concrete incorporating silica fume using hybrid IWOA-GPR model
Yang Yu, Guangyin Wang, Ghasan Fahim Huseien, Zhen Zou, Zhenghao Ding, Chunwei Zhang
Abstract
Self-compacting concrete (SCC) has been widely used due to its ability to fill complex molds without the need for mechanical vibration . However, the compressive strength of SCC is influenced by various factors, and traditional prediction methods often fail to provide accurate models. Gaussian process regression (GPR) demonstrates high accuracy when dealing with complex, high-dimensional data; however, the performance of the model is highly dependent on the selection of hyperparameters. To address these issues, this study proposes an improved whale optimization algorithm (IWOA) to optimize the hyperparameters of the GPR prediction model. Cement, water-to-binder ratio, silica fume (nano silica or/and micro silica), coarse aggregate, fine aggregate, superplasticizer , and viscosity-modifying agent are selected as inputs, with the 28-day compressive strength of SCC as the output. A nonlinear convergence factor is employed to enhance the global search capability, thereby improving the prediction accuracy and stability of the model. Finally, experimental data is used to compare the proposed method with various other prediction models, validating the superiority of the proposed approach. The results indicate that the IWOA-GPR model achieved a root mean squared error (RMSE) of 4.80 on the training set and 6.52 on the testing set, with a mean absolute error (MAE) of 3.05 and 4.67, respectively. The coefficient of determination (R²) on the testing set was 0.95, indicating excellent predictive performance. Compared to other models (BPNN, SVM, EDT, RF), the IWOA-GPR model demonstrated significant improvements in RMSE and MAE, with reductions of up to 25.77 % on the training set and 20.90 % on the testing set. Sensitivity analysis identified the water-to-binder ratio as the most critical factor (SHAP value = 8.2), followed by fine aggregate and cement content . These results validate the IWOA-GPR model as a robust tool for SCC strength prediction and mix optimization.