Author(s): H. Bonakdari; I. Ebtehaj

Linked Author(s): Hossein Bonakdari

Keywords: Rtificial Neural Network (ANN); Extreme Learning Machines (ELM); Sediment transport; Storm sewer; Velocity at limit of deposition

Abstract: In storm channel system design, the system should have the ability to transfer the entire input flow to the system as well as prevent sediment settling. Existing methods of determining the velocity at limit of deposition are minimum velocity or regression-based equations. Because minimum velocity methods fail to consider the effective flow and sediment transfer parameters, while regression-based equations are not flexible in terms of various hydraulic conditions and they do not perform well. Thus, using such equations leads to a lack of designs with optimum, confident coefficients. In this study, Extreme Learning Machines (ELM) are employed to predict velocity at limit of deposition. ELM is a new algorithm for single-hidden layer feed-forward neural network (SLFN) training, which overcomes problems caused by gradient algorithms, for instance low velocity in network training. In this study, dimensional analysis is applied to identify the effective parameters on estimating the velocity at limit of deposition, followed by ELM to predict the parameter values. ELM performance is compared with artificial neural networks (ANN) and regression-based equations. According to the results, using ELM increases the convergence speed to obtain optimum velocity results, and is accordingly more accurate. The results represent the superior performance of ELM compared to existing regression-based equations.

DOI: https://doi.org/10.15142/T3100628160853

Year: 2016