Prediction of seepage flow through earthfill dams using machine learning models

Abstract

In this study, three machine learning models, namely, the Multilayer Perceptron Neural Networks (MLPNN), the Generalized Regression Neural Networks (GRNN) and the Radial Basis Function Neural Networks (RBFNN) were used for predicting seepage flow through an earthfill dam. Moreover, obtained results were compared with those obtained from the standard Multiple Linear Regression (MLR). The three models were developed using piezometer elevations observed at seven different piezometers, in addition to the related reservoir water level and the periodicity for a period of seven years. Obtained results indicated that the GRNN model had substantially better prediction performance than the RBFNN, MLPNN, and the standard MLR models with statistical values of coefficient of correlation R = 0.981, root mean square error RMSE = 0.386 L/s, and a mean absolute error MAE = 0.95 L/s. Moreover, including the periodicity factors improves prediction accuracy of the machine learning models.

Keywords

• Prediction
• Seepage flow
• Earthfill dams
• Machine learning
• Fontaine gazelles dam