摘要
泥沙预报是开展水库泥沙实时调度的前提,而水沙作用机理和演进规律的复杂性又导致开展高效、精准的泥沙预报较为困难。基于微软在2018年发布的Azure AutoML自动化机器学习技术,进行了泥沙预报模型构建与应用的探索。选取三峡水库泥沙重要控制站——寸滩、清溪场、万县、黄陵庙站构建了含沙量预报模型,并从模型构建与评估、预报精度、输入因子重要性等角度开展了分析。研究结果表明:Azure AutoML技术可便捷地进行自动化机器学习模型的构建,基于该技术建立的预见期为1~3 d的模型针对沙峰消退阶段和含沙量较小阶段预报效果较好;预见期为1~2 d的模型可以对沙峰开展较为准确的预报;寸滩、清溪场站含沙量主要受到上游来沙的影响,而万县、黄陵庙站的含沙量自相关性较强。
Sediment forecast is the premise of real-time operation of reservoir sediment,and the complexity of water-sediment action mechanism and evolution law makes it difficult to carry out efficient and accurate sediment forecast.Based on the Azure AutoML automatic machine learning technology released by Microsoft in 2018,the construction and application of sediment prediction model were explored.The important sediment control stations along the Three Gorges Reservoir,Cuntan,Qingxichang,Wanxian and Huanglingmiao Station were selected to construct a sediment concentration prediction model,and the analysis was carried out from the perspectives of model construction and evaluation,prediction accuracy and importance of input factors.The results showed that the Azure AutoML technology can be used to construct the automatic machine learning model conveniently.The model constructed by this technology with a forecast period of 1~3 days has better prediction effect for the sediment peak regression stage and the small sediment concentration stage.While the proposed model with a forecast period of 1~2 days can carry out more accurate prediction of sediment peaks.The sediment concentration of Cuntan Station and Qingxichang Station is mainly affected by the upstream sand,while the sediment concentration of Wanxian and Huanglingmiao stations has strong autocorrelation.
作者
曹辉
陈柯兵
董炳江
CAO Hui;CHEN Kebing;DONG Bingjiang(State Key Laboratory of Hydraulics and Mountain River Engineering,Sichuan University,Chengdu 610065,China;Water Resources Research Center,China Yangtze Power Company Limited,Yichang 443002,China;Bureau of Hydrology,Changjiang Water Resources Commission,Wuhan 430010,China)
出处
《人民长江》
北大核心
2023年第4期94-100,共7页
Yangtze River
基金
国家自然科学基金联合基金项目(U2040218)
三峡后续工作项目“三峡水库区间支流水沙变化”(21303)
国家重点研发计划项目(2019YFC0409000)。
关键词
泥沙预报
沙峰传播
含沙量
Azure
AutoML
自动化机器学习
三峡水库
sediment forecast
sediment peak spreading
sediment concentration
Azure AutoML
automatic machine learning
Three Gorges Reservoir
