Free surface flow assessment through a homogeneous earth-fill dam using a feed-forward neural network model

Seepage flow through the body of an earth-fill dam adversely affects dam's stability. Therefore, a better understanding of the seepage phenomenon is required to detect early signs of abnormal behaviour and to plan intervention strategies when needed. In this study, a feed-forward neural network model is used to evaluate the free surface flow through a homogenous earth-fill dam. Data collected from the monitoring system were first analysed using principal component analysis (PCA) to identify the relationships between the selected variables. The model inputs were then pre-processed using Min-Max scaling transformation, and a trial-and-error approach has been used in order to achieve the best neural network architecture. As a next step, the delayed effects of reservoir water level fluctuations on the seepage phenomenon were investigated. Additionally, the effect of the dataset size on the prediction capabilities of the neural network model has also been explored. According to the selected performance criteria, the proposed neural network model was shown to be a powerful tool for predicting piezometric levels in dam's body. Such a result can be used for a continuous simulation of earth dam's behaviour when coupled with a numerical model.