Enhancing Horizontal Displacement Prediction of Shoring Systems in Deep Urban Excavations Using Artificial Neural Networks

This study aims to accurately predict the horizontal displacement of shoring systems in deep urban excavations, where numerous risks threaten the stability of surrounding structures. The methodology integrates Finite Element Modeling (FEM) to simulate soil behavior and excavation conditions, providing a detailed assessment of displacement under varying construction scenarios. The results from FEM serve as input for an Artificial Neural Network (ANN), which optimizes prediction accuracy by managing the nonlinear relationships between displacement factors. ANN predictions are then validated against actual field monitoring data, demonstrating significantly improved accuracy with low forecast errors and stable predictive capacity. By combining FEM and ANN, this research enhances predictive capabilities, offering a robust and feasible solution for optimizing shoring system design in deep excavations. Experimental results show that the proposed model achieves Mean Squared Error (MSE) = 0.0658, Root Mean Squared Error (RMSE) = 0.2566, and Mean Absolute Error (MAE) = 0.1835, proving its superiority over traditional methods. Additionally, feature importance analysis highlights the Y-coordinate and excavation depth as the most influential factors in displacement prediction, providing valuable insights into engineering applications. The findings also reveal that ANN is highly sensitive to input variations, particularly in FEM parameters such as elastic modulus and excavation depth. To improve generalization, future research should expand to various soil types and excavation conditions, ensuring broader applicability. Looking ahead, integrating deep learning techniques such as LSTM or CNN could further enhance real-time prediction and safety monitoring, ensuring more efficient and reliable excavation management in urban construction projects.