首页|Research from University of Michigan-Dearborn Provides New Data on Machine Learn ing (Physics-enhanced machine learning models for streamflow discharge forecasti ng)
Research from University of Michigan-Dearborn Provides New Data on Machine Learn ing (Physics-enhanced machine learning models for streamflow discharge forecasti ng)
扫码查看
点击上方二维码区域,可以放大扫码查看
原文链接
NETL
NSTL
A new study on artificial intelligence is now available. According to news reporting from Dearborn, Michigan, by NewsR x journalists, research stated, "ABSTRACT: Accurate river discharge forecasts fo r short to intermediate time intervals are crucial for decision-making related t o flood mitigation, the seamless operation of inland waterways management, and o ptimal dredging." Funders for this research include Coastal And Hydraulics Laboratory. The news correspondents obtained a quote from the research from University of Mi chigan-Dearborn: "River routing models that are physics based, such as RAPID (‘r outing application for parallel computation of discharge') or its variants, are used to forecast river discharge. These physics-based models make numerous assum ptions, including linear process modeling, accounting for only adjacent river in flows, and requiring brute force calibration of hydrological input parameters. A s a consequence of these assumptions and the missing information that describes the complex dynamics of rivers and their interaction with hydrology and topograp hy, RAPID leads to noisy forecasts that may, at times, substantially deviate fro m the true gauged values. In this article, we propose hybrid river discharge for ecast models that integrate physics-based RAPID simulation model with advanced d ata-driven machine learning (ML) models. They leverage runoff data of the waters hed in the entire basin, consider the physics-based RAPID model, take into accou nt the variability in predictions made by the physics-based model relative to th e true gauged discharge values, and are built on state-of-the-art ML models with different complexities. We deploy two different algorithms to build these hybri d models, namely, delta learning and data augmentation."
University of Michigan-DearbornDearbor nMichiganUnited StatesNorth and Central AmericaCyborgsEmerging Technol ogiesMachine Learning