Predicting daily pore water pressure in embankment dam: Empowering Machine Learning-based modeling

Abstract

Dam safety assessment is important to implement the appropriate measures to avoid a dam break disaster as part of the water reservoirs management process. Prediction-based approaches are valuable to compare the actual measurements with the simulated values to proactively detect anomalies. However, the application of the conventional hydrostatic seasonal time (HST) has some limitations related to an instantaneous response of the dam to environmental factors, which can lead to inaccurate prediction and interpretation, especially for daily measurements. Besides, the generalization ability (GA) of these models is not analyzed enough despite its crucial importance in selecting the appropriate models. In this study, the multiple linear regression (MLR), artificial neural network (ANN), support vector regression (SVR), and adaptive boosting (AdaBoost) models with nonlinear autoregressive exogenous (NARX) inputs are proposed to incorporate the response delay of the dam to the hydraulic load. Thus, these models were evaluated and compared with the HST model for predicting the daily pore water pressure in an embankment dam. Moreover, we proposed a classification method of the models into four categories, namely perfect, excellent, good, and poor according to the GA. Results show that, except for the AdaBoost, the other ML models outperformed the traditional statistical approach (HST) in terms of prediction accuracy as well as the GA. Overall, the study results provide new insights in enhancing the monitoring processes and dam safeties by detecting the anomalies early through the comparison of the measurements and simulated results produced by the best-fitted models from the confidence interval (CI) perspective.

Keywords: Dam safety; Heimer dam; Hydrostatic seasonal time; Machine learning models; Pore water pressure.