Prediction of Concrete Dam Deformation through the Combination of Machine Learning Models

Dam safety monitoring is of vital importance, due to the high number fatalities and large economic damage that a failure might imply. This, along with evolution artificial intelligence, has led machine learning techniques being increasingly applied in this field. Many researchers have successfully trained models predict dam behavior, but errors vary depending on method used, meaning optimal model not always same over time. The main goal paper improve precision by combining different models. Our research focuses comparison two successful integration strategies other areas: Stacking Blending. methodology was prediction radial movements an arch-gravity divided into parts. First, we compared usual estimating their hyperparameters, i.e., Random Cross Validation Blocked Validation. This aspect relevant only for importance robust estimates, also because it source data sets used train meta-learners. second topic combination strategies, which types tests were performed. results obtained suggest CV outperforms random approach robustness provides better predictions than generalized linear meta-learners strategy achieved higher accuracy individual most cases.