ESTIMATING THE SEQUENT DEPTH OF A MOVING HYDRAULIC JUMP: A NEURAL NETWORK APPROACH

摘要

Moving hydraulic jump could be regarded as a special case in unsteady flow, which changes the flow regime and generates a hydraulic discontinuity along the channel. The knowledge of the flow behavior within a reach that retains this phenomenon is quite essential for river flow routing and the management of flow distribution in a network. However, in spite of its practical importance, experimental data are scarce and the application of numerical simulation due to the presence of hydraulic discontinuity is, also, rather complicated. In such circumstances, compiling experimental data and analyzing them by using artificial intelligence (AI) aid to distinguish the parameters that influence the phenomenon. In this research, variety of moving hydraulic jump conditions was produced by generating different hydrographs at the upstream end of a rectangular flume. The compiled experimental data were used for training and testing a multilayer percepetron (MLP) network. This network was capable to estimate the subcritical flow depth at the downstream side of the jump, based on the given discharge and the associated supercritical flow depth. The number of hidden layer(s) processing elements (Pes) of the network and momentum rate (one of the training rule's parameters) were optimized by means of the genetic algorithm (GA). The results indicated that the trained network could be regarded as a useful mean to estimate one of the flow parameters in such complex flow condition. Therefore employing the artificial neural networks (ANNs) to evaluate more experimental and field data, seems to build up the require knowledge to provide a simple yet reliable approach for mixed flow regime.