Page 44 - 《水资源与水工程学报》2025年第1期
P. 44

!36 "!1 #                         & ' ( ) & * + , -                               Vol.36No.1
               2025 $ 2 %               JournalofWaterResources&WaterEngineering                 Feb.,2025

            DOI:10.11705/j.issn.1672-643X.2025.01.05


                      's Reg-Crossformer fOW#äŒ4µ¦



                                                    ,GH, IJK
                                            ( E9a0 í00G, ¡Á FU 211100)
                > ?: Ö³EJKAB@A, ^+´¯Z[š 1961—2015 T{âÿ‚Jíê, eL£ç Crossformer iº, b
                •–F™Ú‰'p´, 'ڃÜëÏÎ@<íê, ¢Ý4©ª‹Òë~|Î{so。–#¯®/Ü^5úQÞ{
                Reg-Crossformer iº, ðñ³EJKÿ‚J, Æ}Éóàuiº‚+,Z[òó>Ã{È'。®«S¤:Reg-
                Crossformer iº’6¨ Crossformer iº, I’stí(R) ž² {Ñtí(NSE) µT–é¯ 7.46%ž 21.63%, 3
                €"Ð6(RMSE) ÊÓ¯ 15.25%; ¦‹Òiº‚6Ãq4,Reg-Crossformer iº¦Á(³´ü†¬3»ƒ2†p
                €0iº(SVM) \æ~0iº(LSTMž Informer), S0#¢»{ij{«ží·Ñ。Reg-Crossformer iº
                ìàðñ³EJK‚J–—¯}.,{€, ‚ôÎJKZ]^œ„\æ~0iº¦Z[ij}KՀr
                ‡ˆq1´î。
                1@A: ÿ‚JÏÎ@<ðñ; æ~0;Reg-Crossformer; ³EJK
                BC<DE:P333   FGHIJ:A    FKLE:1672643X(2025)01004007

                         DailyrunoffpredictionbasedonReg-Crossformermodel

                                              WANGKunyu,LIUXiangyang
                                    (SchoolofMathematics,HohaiUniversity,Nanjing211100,China)
                Abstract:TakingtheWeiheRiverBasinastheresearchbackground,thedailyrunoffdataofXianyang
                stationfrom1961to2015areselectedasthedatainputfortheCrossformermodel.Focusingonthemulti
                dimensionaltimeseriesdata ,thispaperadoptsthetwostageattention(TSA)mechanismtobettercapture
                thecorrelationbetweendifferentdimensions.Inaddition,theReg-Crossformermodelincorporating
                multisourcecovariatesisproposedtofurtherenhancetheadaptabilityofthemodeltocomplexhydrological
                conditions.TheresultsofdailyrunoffpredictionintheWeiheRiverBasinshowthatcomparedwiththeo
                riginalCrossformermodel ,theproposedmodelimprovesthecorrelationcoefficient(R)andNashefficien
                cycoefficient (NSE)by7.46% and21.63% respectively;reducestherootmeansquareerror(RMSE)
                by15.25%.Inthecomparativeexperimentsofdifferentmodels,Reg-Crossformeroutperformsthecon
                ventionalmachinelearningmodel (SVM)anddeeplearningmodels(LSTMandInformer)acrossvarious
                evaluationindicators,demonstratingsuperiorsimulationperformanceandstability.Reg-Crossformer
                modeloffersanewapproachfortheaccuratepredictionofrunoffintheWeiheRiverBasin,andprovides
                valuableinsightsintofutureapplicationofwaterresourcesmanagementanddeeplearningmodelsinhydro
                logicalsimulation.
                Keywords:dailyrunofftimeseriesprediction;deeplearning;Reg-Crossformer;theWeiheRiverBasin
                                                               m‡ˆ?W‰Š        [1] 。xï, ðñøùú¼、 ›œ¼
            1 ./01
                                                               ÆDžNv$%{‹-89, JKt†út$Î$
                 JKZ[ðñ”Z]^œ„{sîJª:                           +,, IZ[LDšrµñÏ1fgÑ、 Ü^úûå
            µ, ‚ƒp0Z]^›¡¢b•、 þœ„žSl                          /:c>{;6       [24] 。Z[ðñ}K{;6VW/

                !"#$:20240305; %&#$:20240818
                '()*: !"wx#0%»?@(3(41830110)
                +,-.: )²g(1999—), Æ, ¡ÁFUN, ÈPABS, VWXYÏÎ@<ðñ€Ô{AB
                qr+,: 5Ô¯(1976—), K, Ç~8ûN, OP, QR, ÈTSTU, VWXYíêµ¶、 p€0、 áȸ¹:€r
                        {AB。
   39   40   41   42   43   44   45   46   47   48   49