نوع مقاله : مقاله پژوهشی
عنوان مقاله English
نویسندگان English
Studying and developing water resource systems requires an accurate prediction of river suspended sediment load. Since sediment load measurement is costly and time-consuming, the simulation methods based on the sediment load time series are usually restricted to sample numbers. This study used two different methods to forecast the monthly suspended sediment load. In the first method, the Support Vector Machine (SVM) algorithm was used to forecast sediment load based on a monthly time series of suspended sediment data. In this method, the hyperparameter coefficients of the SVM model were optimized using the Simulated Annealing (SA) algorithm. Additionally, this model was evaluated under single-input-single-output (SISO) and multiple-input-single-output (MISO) patterns. The suspended sediment load and flow data of the Yalfan River from 2007 to 2015 were the SISO’s inputs and outputs, respectively. Moreover, the MISO’s second input was a fixed number series. In the second method, the sediment rating curve and the FAO method forecast it. A %9.93-point increase in the Standard Error (SE) index was obtained by normalizing the SVM-SA’s input data. Switching the input patterns from single input-single output (SISO) to multiple input-single output (MISO) resulted in a %1.03 improvement in SE. By optimizing the coefficients of the discharge-sediment variables in the rating curve and FAO method using the General Reduced Gradient (GRG) technique, the SE index improved by %10.13 and %73.19, respectively. The results showed that the SVM-SA model based on the normalized MISO approach is the best method for suspended sediment load forecasting.
کلیدواژهها English