Evaluation of Machine Learning Techniques for Seepage Analysis of a Hydroelectric Dam in Ghana

2026-08-03T11:46:49+00:00

Seepage within hydroelectric dams is the leading cause of dam failure worldwide. Accurately predicting dam seepage is a primary concern in areas like Ghana, where hydroelectric dams are at risk of failure due to the lack of reliable seepage prediction techniques. The Group Method of Data Handling (GMDH), Multilayer Perceptron (MLP), Tabular Prior-Data Fitted Network (TabPFN), and Extreme Learning Machine (ELM) algorithms were trained and evaluated for their accuracy in predicting dam seepage for a hydroelectric dam in Ghana using 492 data observations collected between 1983 and 2024. Each of these models was trained using data from 1983 to 2016 and tested using data collected between 2016 and 2020. The results of all four models indicated that they have strong predictive capability for dam seepage. Within the linear regression model (GMDH), the determination coefficient (R²) for dam seepage was 0.9813, the root mean square error (RMSE) was 0.13, and the mean absolute error (MAE) was 0.10. Furthermore, the negligible gap between the train and test data for this model (0.004) indicates that the dam seepage at this site is solely linear in function, as predicted from Darcy’s Law. The factors that were most important within each of these models were determined to be the elevation of the reservoir, the average rainfall rate, and the elevation of the tailwater. Additionally, using counterfactual analysis, it was determined that a drawdown of the reservoir of 0.09 to 0.13 meters and an increase in the tailwater elevation of 0.21 to 0.27 meters would reduce the dam seepage to the target of 2.50 units of seepage. This is the first time that counterfactual methods have been applied to the management of seepage within hydroelectric dams in Ghana. Thus, this model has the potential to be replicated in other areas with similar geologic compositions and similar climate conditions to those of Ghana.

International Journal of Hydraulics and Wastewater Treatment Technologies (p-3107-9180)

Evaluation of Machine Learning Techniques for Seepage Analysis of a Hydroelectric Dam in Ghana