DIGITIZING MONITORING IN EARTH-FILL DAMS WITH MACHINE LEARNING FOR PREDICTING SEEPAGE

Seepage presents a significant challenge for earth-fill dams, especially those constructed in the mid-19th century. The original designs of these dams often neglected to account for the impacts of climate change. Effective seepage monitoring and inspection, important for addressing this issue, traditionally rely on manual methods, which are prone to data misinterpretation. Recent advancements in artificial intelligence, particularly machine learning, offer a more reliable solution for monitoring dam seepage. This study utilizes manual inspection data from Tarbela Dam in Pakistan, spanning from 2016 to 2023, to predict seepage using Random Forest (RF) and CatBoost (CB) machine learning algorithms. The results indicate that the coefficient of determination (R²) for RF is 0.980 throughout training, 0.943 validation, and 0.925 testing. For CB, the R² values are 0.981 for training, 0.947 for validation, and 0.967 testing. Additionally, the root mean square error (RMSE) for RF is 0.043 training, 0.065 validation, and 0.069 testing phase. The RMSE for CB is 0.035 through training, 0.063 validation, and 0.113 testing. Findings of analysis demonstrate that CB exhibits superior performance during the testing phase of the ML models. This research provides valuable insights for stakeholders, dam management, and academia, laying the groundwork for the development of real-time seepage monitoring systems for earth-fill dams, incorporating Internet of Things (IoT) technology.