Bed load sediment transport estimation in a clean pipe using multilayer perceptron with different training algorithms

Abstract

Due to the presence of solid matter in the flow passing through sewer pipes, determining the minimum velocity that prevents sediment deposition is essential. In this study, the Multilayer Perceptron (MLP) network optimized with three different training algorithms, including variable learning rate (MLP-GDX), resilient back-propagation (MLP-RP) and Levenberg-Marquardt (MLP-LM) is studied in terms of ability to estimate sediment transport in a clean pipe. The results indicate that for all algorithms, model ANN(d) that uses volumetric sediment concentration (C V ), median relative size of particles (d/D), ratio of median diameter particle size to hydraulic radius (d/R) and overall sediment friction factor (λ s ) as input parameters, is more accurate than the other models. In predicting Fr, the results of MLP-LM (R 2 = 0.98, RMSE = 0.02 and MAPE = 5.13) are better than MLP-GDX (R 2 = 0.96, RMSE = 0.03 and MAPE = 5.9) and MLP-RP (R 2 = 0.95, RMSE = 0.26 and MAPE = 5.74). A comparison of the model selected in this study with existing equations of sediment transport in sewer pipes also indicates that ANN(d)-LM (RMSE = 0.025 & MAPE = 5.78) perform better than existing equations.