Estimation of seepage through zoned earth dams using soft computing techniques

Abstract Seepage through earth dams should be carefully considered in the design and construction processes of such dams due to its effect on dam safety. In this study, The soft computing techniques, namely, Random Forest (RF), Multi-layer Perceptron (MLP), Random Subspace (RS) and M5P has been adopted to predicate the seepage through zoned earth dams. Al- Kramis earthen dam which located in Algeria was taken as case study. The data of thirteen years spened between (01/2006 to 12/2018) was collected and used. Different effective parameters were employed as input parameters to develop the soft computing models, including the reservoir water level (WL), the upstream right piezometer (h 1 ), the upstream left piezometer (h 2 ), the downstream right piezometer (h 3 ), the downstream left piezometer (h 4 ), while the observed seepage flows (Q) were used as output parameter. The performance of models was evaluated using several statistical indices include Correltion Coefficient (CC), Mean Absolut Error (MAE), Root Mean Square Error (RMSE) and Relative Absolut Error (RAE). The results indicated that the M5P model has the better performance (CC = 1 and 0.998, MAE = 0 and 0.0024, RMSE = 0 and 0.0029, RAE = 0 and 2.011, RRSE = 0 and 2.2545) for both training and validation periods respectively, followed by MLP model. The evaluation also indicated that the worst performance was obtained from the RS model. Besides the statistical assessment of the results, four different graphical methods namely line and scatter plots, Box-Whisker plot, Violin plot and Taylor diagram were adopted in the study to evaluate the model performance graphically. The finding of theis study improved the capability of the soft computing techniques in modeling of the seepage dischange through earth dam.