基于积分时滞启发的深度学习水位预测模型

打开文本图片集

中图分类号：TV737 文献标识码：B 文章编号：1001-9235（2025）11-0091-09

Deep Learning-Based Water Level Prediction Model Inspired by Integrator Delay

LUO Wei2， WU Jiahao3， CHEN Lingqiang'， ZHENG Jun+，LEI Xiaohui5， WANG Lizhi²， TAN Min²， XU Jiang

⋅4 （1.SchoolofCivilEngineering，TsinghuaUniversityBeijingOo84，China;2.ChinaEnergyGroupBigDataServiceCo.td， Chengdu61o16，China;3.SchoolofInformatioandElectricalEngineering，HebeiUniversityofEngineering，Handan056038， China;4.ChinaSouth-to-North WaterDiversionGroupWaterNetworkSmartTechnologyCo.，Ltd.，Beijing123O8，China;5.hool of Water Conservancyand Hydroelectric Power，Hebei Universityof Engineering，Handan O56O38，China）

Abstract：Accuratewaterlevelpredictionplaysakeyroleinthesafeoperationofhydropowerstationsandisanimportantguarantee forimprovingpowergenerationeficiency.Thetraditionaldeepleaing methodhaslimitationsindealing withthecomplexdynamics andtime-delaycharacteristicsofhydrologicalsystems.Itisdiiculttocapturethesubtlechangesandlong-termandshort-tem dependenciesinolinearhydrologicalprocesses，anditspredictionaccuracycannotbeguaranteed.Terefore，thisstudyproeda physicallyinspiredhydraulicpredictionmodel（PHM）topredictthewaterlevelinfrontoftesingle-channelpooldamoftheDadu River.Themodeldeeplyexploredtheinteralmechanismof thephysicallyinspiredintegratordelaymodel，miningthelong-term dependenceandsotteabutecorrelationinroloicaldata，eivelyoveromingtprobofteplagodel's insufcientabilitytostimatetiedelaydyamicsndimprovingtheaccuracyofpreditionrsults.Inispaper，esdtal dataof theDaduRiver.Throughalarge numberofexperiments，itisproventhatcomparedwiththeexisting models，PHMhasa significantreductioninthetheekeyindicatorsofmeanabsoluteeror（MAE），rotmeansquareeror（RMSE），andmeanabsolute percentage error （MAPE）. The average reduction of MAE is 43.2% ; the average reduction of MAPE is 42.9% ， and the average reduction of RMSE is 52.3% . The experimental results show that the model is more practical and reliable than the existing models in different scenarios.

Keywords： integratordelay; waterlevel prediction;deep learning; physical inspiration; single-channelreservoir

水位预测是水利水电工程设计、施工与运行管理的关键环节，对防汛、抗旱、供水保障、发电、航运以及优化生态条件发挥重要的作用[1]。（剩余10646字）