Extreme Learning Machines in Predicting the Velocity Distribution in Compound Narrow Channels

Due to the importance of velocity distribution in open channels, this matter has long been of interest to scientists and engineers. Studies show that the maximum velocity in open channels often occurs below the free surface, something known as the dip phenomenon. This phenomenon can serve as a measure in river hydraulics and sediment transport studies. However, existing relationships for estimating the dip phenomenon has advantages and disadvantages, with the biggest problem being poor performance in various hydraulic conditions. To overcome such problems and to achieve high modeling speed and good performance, Extreme Learning Machines (ELM), a new soft computing technique for velocity field modeling is used in this study. This method is a novel algorithm for feedforward neural networks and has only one hidden layer. Field-measured datasets from a narrow compound open channel in France are employed to model the velocity field. The results indicate that ELM performed adequately (MAPE = 3.03; RMSE = 0.03) in velocity field prediction. However, it was found that the ELM offers higher performance than the existing methods.