Asymmetric hydraulic jump prediction based on interpretable soft voting ensemble learning

Against the backdrop of carbon peaking and carbon neutrality, some water conservancy projects have begun adopting combined energy dissipation arrangements featuring both spillway tunnels and spillways. However, due to variations in inflow characteristics across different spillway structures, the flow distribution within stilling basin becomes non-uniform and turbulent. This paper addresses the issues of non-uniform flow distribution and turbulent flow patterns within stilling basin caused by asymmetric inflow in combined spillway tunnel and spillway channel energy dissipation arrangements. It proposes an energy dissipation scheme involving the installation of a twisted bucket lip at the spillway channel outlet to generate an asymmetric hydraulic jump. Numerical simulations yielded 140 sets of data of asymmetric hydraulic jump length, with experimental validation confirming the reliability of the simulated results. Simultaneously, multiple machine learning models were employed for length of hydraulic jump prediction. Results indicate that extreme gradient boosting, categorical gradient boosting, and light gradient boosting machine delivered optimal performance. To further enhance the prediction accuracy, this study constructed an ensemble model based on soft voting, integrating the strengths of all three approaches to improve predictive precision. The interpretability analysis revealed that the width ratio (β) exerted the strongest influence on the hydraulic jump length, while the segment ratio (α) showed a negative correlation with the length of the hydraulic jump, and the Froude number (Fr) exhibited a positive correlation. This ensemble model provides an effective tool for accurately predicting the asymmetric hydraulic jump under low Froude number conditions, demonstrating significant engineering application value.