Page 69 - 2023年第54卷第9期
P. 69
roughset[J].WaterScienceandTechnology,2015,71(10):1507 - 1515.
[19] 郑彦辰,李建柱,荣佑同,等.降雨时空分布量化及其在洪水过程分 类 中 的 应 用 [J].水 利 学 报,2022,
53(5):560 - 573.
[20] WANGW, HUANG Y, WANG Y, et al.Generalized autoencoder: A neural network framework for
dimensionalityreduction [C]??ProceedingsoftheIEEEconferenceoncomputervisionandpatternrecognitionwork
shops.2014.
[21] 刘可新,包为民,阙家骏,等.基于主成分分析的 K均值聚类法在洪水预报中的应用[J].武汉大学学报
(工学版),2015,48(4):447 - 450,458.
[22] 陈建强.基于卷积神经网络的海洋次表层叶绿素遥感反演研究[D].青岛:青岛科技大学,2022.
[23] 刘玉驰,蒋玉峰,王树青,等.基于数据融合及残差卷积自编码器的结构损伤识别方法[J].振动与冲击,
2023,42(4):194 - 203.
[24] 杨亚兰,李晨鑫,秦毓毅,等.基于卷积变 分 自 编 码 器 的 日 负 荷 曲 线 聚 类 研 究 [J?OL].现 代 电 力:1 - 9
[2023 - 03 - 31].https:??doi.org?10.19725?j.cnki.1007 - 2322.2022.0076.
[25] 陈立,魏钰欣,刘斌.基 于 生 成 对 抗 网 络 与 ResUNet的 细 胞 核 图 像 分 割 [J].光 电 子 · 激 光,2023,34
(5):473 - 481.
[26] GIRSHICKR.Fastr - cnn[C]??ProceedingsoftheIEEEinternationalconferenceoncomputervision.2015:1440 -
1448.
ResearchonfloodclassifiedintelligentforecastingmethodbasedonAE - RCNN
3
1,2
4
1,2
1,2
1,2
YUANXimin ,LIDa ,TIANFuchang ,HELixin,WANGXiujie ,GUOLibing
(1.StateKeyLaboratoryofHydraulicEngineeringSimulationandSafety,TianjinUniversity,Tianjin 300350,China;
2.SchoolofCivilEngineering,TianjinUniversity,Tianjin 300350,China;
3.SchoolofWaterConservancyandHydroelectricpower,HebeiUniversityofEngineering,Handan 075000,China;
4.FloodandDroughtDisasterPreventionCenterofNingxiaHuiAutonomousRegion,Yinchuan 750002,China)
Abstract:Hierarchicalfloodforecastingmethodinareaswithcomplexflowgenerationandconfluencecharacteris
ticscanimproveforecastaccuracy.Thispaperproposesahierarchicalintelligentfloodforecastingmethodbasedon
autoencoder(AE)andresidualconvolutionalneuralnetwork(RCNN),usingautoencoderandK - meansclustering
algorithmtorealizefeatureextractionandfloodclassificationofhydrologicaldata ,usingtheRCNNmodeltoim
provetheeffectivetrainingdepthoftheconvolutionalneuralnetwork.TakingtheHuangtaiqiaoHydrologicalStation
intheXiaoqingRiverBasininShandongProvinceasanexample ,theresearchonfloodclassificationintelligent
forecastingwascarriedout.TheresultsshowthattheMAEindex,RMSEindex,andNSEindexoftheAE - RCNN
modelapplyingtheclusteringofdownscaleddataare5.04 ,7.91and0.92,respectively,whicharebetterthanthe
CNNmodel ,RCNNmodel,andrainfallclusteringRCNNmodel.Thismethodcaneffectivelyextractthecharac
teristicsofhydrologicaldataandimprovetheaccuracyoffloodforecasting.
Keywords:hierarchicalintelligentfloodforecasting;AE - RCNN;data - drivenmodel;autoencoder;residual
convolutionalneuralnetwork
(责任编辑:耿庆斋)
0
— 1 7 9 —