Prediction of Concrete Arch Dam Response Using Locally Estimated Scatterplot Smoothing

In this research, a novel hybrid methodology is proposed for predicting the structural response of high concrete arch dams, combining the Discrete Element Method (DEM) with the Locally Estimated Scatterplot Smoothing (LOESS) technique. A structured calibration strategy is employed during the numerical model preparation to enable the generation of a wide range of reliable output variables for training and prediction. The methodology is then applied to the El Atazar arch dam to demonstrate its capability to forecast displacement and stress responses. The study reveals that using the current air temperature as an input variable is not adequate for representing the thermal behavior of the dam body; instead, the mean air temperature over a specified period yields significantly better results. Additionally, the findings highlight the importance of the loading path and the dam’s initial state in determining its structural response. The developed model shows a strong agreement between predicted and observed data, demonstrating its effectiveness in capturing the nonlinear behavior of high concrete arch dams. Compared to traditional parametric models commonly used for dam deformation analysis, the proposed framework offers greater flexibility in representing nonlinearity while requiring less training data, making it ideal for dams with limited monitoring records, such as older dams or newly operated ones.