Simulation of daily suspended sediment load using an improved model of support vector machine and genetic algorithms and particle swarm

Predicting suspended sediment load is one of the most important issues in relation to watershed basins which must be in a way that decision-makers make correct decisions about the hydrology of the basin, as well as the related hydraulic structures, based on the predicted values. The present study evaluated the model of support vector machine for predicting suspended sediment load in two earth dams in Iran. According to the fact that support vector model is one of the regression methods, it has parameters with unknown value and the optimization algorithms which have been used in this study including genetic algorithm and particle swarm, have enhanced the support vector model and after completing the model, the pointed method is used to simulate the sediment load. Two earth dams namely Royan and Veynakeh were used as a case study, which were in Semnan province in Iran. Also, we have compared the outcomes of the improved support vector machine method with the M5Tree model and also the multivariate spline model. The results for the Royan station showed that the RMSE error value for the SVR-PSO3 model decreased to 14 up to 75%. In addition, MAE based on SVR-PSO3 was reduced to 12 up to 73% compared to other M5T, SVR-GA, and MARS models. In addition, for Veynakeh station, for example, the SVM-PSO model has been able to reduce the RMSE rate by 19 up to 76%. Also, the evaluation about input parameters for simulated suspended sediment load showed that, the flow rates just like inputs can have more exact results to simulate suspended sediment load and the use of inputs with 1-day delay precipitation reduced the accuracy of the models. Therefore, it has been suggested to use SVM-PSO model as a suitable tool for hydrological simulations in addition to the sedimentation basin.