Flow distribution in a compound channel using an artificial neural network

Abstract

In compound channels, much of the hydraulic resistance may be assigned by channel and floodplain geometry and bed roughness. Understanding the distribution of flow in the subsections of a compound river channel is a tedious work due to complex momentum transfer at the junction. The zonal discharges are mostly dependent upon the hydraulic resistance of the corresponding subsections. Hence, different experimentations in laboratory channels are required to be investigated by analyzing the dependence of momentum transfer on individual flow capacities of the subsections. So, experiments are performed in non-homogeneous roughness beds of asymmetric compound channels to examine the flow behavior. Total 272 numbers of experimental data sets comprising wide ranges of width ratio, relative flow depth, aspect ratio and roughness ratio with the present experimental data series are used for both training and validation of the model. Previous models can provide good results only for specific ranges of independent parameters whereas the back propagation of artificial neural network (BPNN) model is capable of performing well for the global ranges of independent parameters. This is because BPNN is able to perform the nonlinear mapping between the dependent and independent variables during the training. The efficacy of the models is verified with the standard statistical error analysis for the data sets. The BPNN model is found to perform well as compared to other researcher’s models.