Genetic algorithm-assisted data-driven model for boundary shear distribution and stage-discharge: Compound open channel flows

Abstract

In this paper, a unified method has been proposed for the percentage shear force carried by floodplains using fewer non-dimensional parameters such as the floodplain's percentage area, the ratio of Manning’s roughness and the depth ratio. This new data-driven dynamic model for percentage shear force is obtained using a genetic algorithm (GA) program, a well-documented machine-learning software, which can examine the existing relationship among the variables and explore the influencing factors. GA facilitated a unified relationship between apparent shear force and the chosen parameters. The new proposed model is simple and accurate compared to the previous models, which were either complex or distinct for different configurations. The efficiency of the new model shows that the most predicted test case results are under the 5% error cap. The cohesive expression derived for smooth and roughened compound channels is the most significant advantage of the GA-based data-driven model, providing a simple and easy-use formula for engineers to apply for broad applications. Compared with the other available approaches, the model proposed provides the most accurate discharge prediction for various unique datasets.