Adverse-slope stilling basins: machine learning approach to estimation of hydraulic jump features

Abstract

Stilling basins are the most commonly used energy dissipating structure for large dams, due to their high operating reliability under a wide range of inflow conditions. However, unsatisfactory tailwater conditions, primarily low tailwater level, can necessitate the use of additional structural solutions for the stabilization of the occurring hydraulic jump. A possible solution to this issue is the use of the adverse-sloped basins, which use the weight of the water in the jump for jump stabilization. Adverse-slope basin design had attracted the interest of engineers and researchers in the last decade as a simple method for the reduction of the required tailwater depth and basin length. To this date, a number of analytical and empirical methods for the estimation of features of such jumps are developed. However, the applicability of these methods is hindered by their diverse accuracy and/or complexity. Based on the experimental measurements on a stepped spillway model, a multilayer perceptron model (MLP) was trained for the estimation of jump characteristics. An accuracy and sensitivity analyses were also performed for each used method. It was demonstrated that machine learning techniques can potentially provide significantly better results than certain analytical and empirical methods. This allows a simpler, faster, yet accurate analysis of the occurring hydraulic conditions in the early stages of design.