| description abstract | One of the significant hydrological processes affecting the sustainability of river engineering is sedimentation. Sedimentation has a major impact on reservoir and dam operations. This study conducted an experiment using a rainfall simulator with varying intensity of rainfall (1–3 L/min) with slopes from 0° to 5°, leading to the generation of runoff and sediment loads (SLs). Precipitation and runoff data from the rainfall simulator were used to develop a sediment load model via hybrid machine learning approaches. Predictive abilities of a robust phase space reconstruction integrated with support vector machine and firefly algorithm (PSR-SVM-FFA) were investigated for estimating sediment load. The accuracy of the PSR-SVM-FFA was assessed versus that of integrated support vector machine and firefly algorithm (SVM-FFA) and conventional support vector machine (SVM) models. In phase space reconstruction (PSR), the delay time constant and embedding dimension were determined to select optimal parameters in the SVM-FFA model. Four performance measures, namely RMS error (RMSE), mean absolute percentage error (MAPE), Willmott index (WI), and bias, were employed to evaluate the performance of the proposed models. The results revealed that prominent values of WI were 0.942, 0.955, and 0.966 for the SVM, SVM-FFA, and PSR-SVM-FFA methods, respectively, for a slope of 4°. PSR-SVM-FFA had better performance than SVM-FFA and conventional SVM for each slope. Based on analysis of the obtained results, it is evident that PSR-SVM-FFA can estimate SL more accurately. | |
