Predicting apparent shear stress in prismatic compound open channels using artificial neural networks

The apparent shear stress acting on the vertical interface between main channel and floodplain (with and without vegetation) in prismatic compound open channels was studied using artificial neural networks (ANNs). Apparent shear stress is commonly used to quantify the transverse momentum exchange between sub-areas of the cross section, and its influencing factors include channel geometry, boundary roughness conditions and hydraulic properties of flow. In particular, if vegetations exist, the eco-characteristics of vegetation can be crucial. To mathematically describe the effect of channel symmetry, we present a new expression of width ratio through which the symmetrical and asymmetrical cross section can be distinguished. The effect of vegetation is considered using degree of submersion and the porosity (volume ratio occupied by water). Dimensional analysis was conducted to determine the mathematical formula of apparent shear stress, and seven non-dimensional parameters were selected as the influencing factors. In total, 260 sets of data (including our new experimental series conducted in a compound channel with vegetated floodplain) were used for training and testing a three-layer, feed-forward neural network with Levenberg-Marquardt (LM) as the selected training algorithm. Also, the effects of main influencing factors on the apparent shear stress were investigated.