125 / 2024-02-18 08:30:09

Suspended sediment forecast model based on Microsoft Azure AutoML

Suspended sediment forecast,Microsoft Azure,AutoML,Yangtze River

Defined as the ratio of sediment mass to volume of a sediment-laden mixture, suspended sediment concentration (SSC) characterizes a sediment transport process. Accurate predictions of the SSC are of practical implications in a wide range of hydraulic issues, including the design of the reservoir dead storage, environmental impact assessment, and prediction of aggradation and degradation around bridge piers. However, transport is a complex phenomenon, involving both the flow and sediment motions and their interactions. From the initial detachment of sediment particles to their deposition, every stage is characterized by high nonlinearity and interplay processes, rendering the SSC forecast a challenging task.

The diverse nature of independent parameters required to model sediment problems as well as the complicated nonlinear process of SSC has provided motivation to explore the capability and efficiency of machine learning techniques for SSC modeling.

Notably, however, machine learning techniques for SSC modeling were almost manually designed by experts by trial-and-error process, which means that even experts require substantial resources and time to create well-performing models. To reduce these onerous development costs, a novel idea of automating the entire machine learning (ML) pipeline emerged, i.e., automated machine learning (AutoML). A complete AutoML system can make a dynamic combination of various techniques to form an easy-to-use end-to-end ML pipeline system. In this work, we focus on Microsoft Azure AutoML to help people with little or no ML knowledge to build high-quality SSC forecasting models.

Results show that, for the Qingxichang hydrological station of the Yangtze River, the proposed Microsoft Azure AutoML performed well for forecasting SSC. Among various algorithms, VotingEnsemble is the best which is used for testing. The conclusion shows that the Microsoft Azure AutoML can easily construct an automated machine learning model for accurate SSC forecasting.