基于机理模型与深度学习的小样本洪水预测

打开文本图片集

中图分类号：TV122;TP18 文献标志码：A doi：10.3969/j.issn.1000-1379.2026.02.010引用格式：，，，等.基于机理模型与深度学习的小样本洪水预测[J].人民黄河，2026，48（2）：58-64.

Small Sample Flood Prediction Based on Mechanistic Model and Deep Learning

LI Faxian1，LI Hang¹，LYU Mou1，CHEN Yunpeng²，DONG Shen1 （1.School of Environmental and MunicipalEngineering，Qingdao Universityof Technology，Qingdao 266520，China; 2.Qingdao Yanbo Data Information Technology Co.，Ltd.，Qingdao ，China）

Abstract：Toaddresstlmitatiosofechanisticodelsunerhrologicallyatascareondisndteinsufcetgenealtiopabilityofneuraletorks，isstudyproosedamecanisticodel-deelaingramewrktoaeflodpredictiouacyist，a high-precision hydrodynamic model wasconstructedusingHEC-HMSand HEC-RAStosimulate hydrologicaldataoutputs，overcomingthe aplicatiootsfldelsiaissgoatabslttdod simulateddataasfedintomultiplenuralnetworkarchitectures（AN，RNN，LSTM，ransforer）forwaterlevelpredictionoeling. Thisaproachleveragedhig-qualitytrainingdatageneratedbythemechanistcmodeltoexpandtheneuralnetworks’samplespacewhile utilizingdatadrienodelstmpeateforteomputatioallitatiosofhaistcodelsiomplexliardacpr. ExperimentalresultsemonsratehatteTansforerodelwitdataugmetatioacevestialpredictioperforance，witcocient of determination （ R2 ）of 0.970 and a mean squared eror（MSE）of O.O17，significantlyoutperforming traditional neural network models.Prediction time is reduced by 97.64% compared to iterative computations in mechanistic models.Flood inundation analysis integrating TransformerandRASMapperimprovestieliessbynearlyO-foldovrsnglemechanistcmodels.Thesudycofisthatthecollbative frameworkserhnistcdelddepleageivelyaesprdictioneliabilitndeli，engoelodeling paradigm for flood risk management in data-deficient regions.

Key Words： HEC-RAS model; inundation analysis;deep learning；flood prediction

0 引言

洪水灾害作为全球常见的自然灾害，其危害性体现在对人类生命安全、社会经济和城市发展等多重威胁[1-2]。（剩余12058字）