Numerical simulation and soft computing approach of pressurized flushing at different water levels

Predicting the sediment inflow to dam reservoirs, followed by implementing flushing operations for dam management, is highly essential. Estimating the sediment inflow to the dam can assist in allocating water resources from the reservoir. Machine learning techniques are efficient tools for such predictions, capable of providing accurate estimates of sediment inflow to dams. This study was conducted in two steps. In the first step, the numerical simulation of flow and sediment flushing in a dam reservoir was performed using the flow-3D model. To do this, the numerical model was calibrated using the results of an actual flushing operation of the study dam. Then, the effects of variables in this study, including the water level at the start of flushing under pressure and the number of active gates during flushing, were studied. In the second step, the sediment inflow to the dam reservoir was predicted using the least squares support vector regression (LSSVR) and multivariate adaptive regression splines (MARS). In this case, the model runtime was 4 year. The Sefidroud dam, one of the most important dams in Iran, was considered as the case study. The results showed that the numerical modeling was capable of determining the initial water level in the reservoir and the appropriate discharge for the dam. The results also demonstrated that LSSVR and MARS could effectively predict sediment outflow from the dam. Moreover, a model including the sediment inflow to the reservoir 1 and 2 days earlier exerted the most suitable prediction results with RMSE, MAE, and NSE of 4.07 kg/m 3 , 0.189 kg/m 3 , and 0.88, respectively. The results of this study indicate the successful application of machine learning models in estimating sediment outflow from dams, which can be utilized for other similar areas.