Partitioning strategy for investigating the prediction capability of bed load transport under varied hydraulic conditions: Application of robust GWO-kernel-based ELM approach

Abstract

In this study, modeling capability of bed load sediment transport was investigated under varied hydraulic conditions through Grey Wolf Optimization based Kernel Extreme Learning Machine (GWO-KELM) and Standard Gradient Descent based Gaussian Process Regression (SGD-GPR) models. 966 sets of data, covering wide range of gravel-bed rivers and streams were utilized to test the developed methodology. The partitioning strategy was used in which, the available data was categorized into different quantitative intervals based on the hydraulic and sediment characteristics. Some specific quantitative intervals of the bed load transport rate, flow discharge, median particle size of bed material, the Shear Reynolds number and the ratio of hydraulic radius to median particle size were examined. The results demonstrated the great consistent performance of the proposed hybrid GWO-KELM model under different hydraulic conditions.