A dynamic-weighted kriging-ANN hybrid model for predicting mechanical properties of grouted sands with limited data

Abstract This paper presents a prediction model based on an enhanced data augmentation process and enriched Artificial Neural Network (Kriging-ANN) to predict the strength and porosity of mortar consolidated bodies more accurately. By training the Artificial Neural Network (ANN) model using an enriched dataset, the results demonstrate that the well-trained Kriging-ANN model effectively captures the distribution characteristics of real-world data. Under the constraints of limited experimental data, the accurate prediction of the mechanical strength and porosity of the consolidated bodies validates the effectiveness of the proposed framework. To verify the model’s robustness, a 5-fold cross-validation was conducted, showing low average errors and narrow standard deviations across folds. When compared to traditional Kriging, the dynamically weighted Kriging-generated augmented data result in a significantly lower mean squared error (MSE), mean absolute error (MAE) and mean absolute percentage error (MAPE), during ANN training, MSE, MAE, MAPE decreases by about 22.9%, 14.5%, 20.0% in pressure ranges. For parameter intervals with lower grouting flow rates, decreases by about 80.7%, 53.8%, 14.5%. Moreover, whether under constant-pressure or constant-flow conditions, the optimized algorithm achieves an Coefficient of Determination (R 2 value) close to 1. This study provides an economically efficient and rapid solution for performance evaluation and optimized design of polymer grouting and similar materials.