Page 14 - 2024年第55卷第9期
P. 14

Press,2003.
                [21] ROWEISST,SAULLK.Nonlineardimensionalityreductionbylocallylinearembedding[J].Science,2000,
                      290(5500):2323 - 2326.
                [22] HARTIGANJA,WONGM A.Algorithm as136:ak - meansclusteringalgorithm[J].JournaloftheRoyalSta
                       tisticalSociety ,1979,28(1):100 - 108.



                      Researchonfloodforecastingmethodinmountainoussmallwatershedsbasedon
                          machinelearningforidentifyingrainfalldynamicspatiotemporalfeatures

                                                         3
                                                                        3,4
                                             1,2
                                1,2
                                                                                        5
                                                                                                     6
                   LIUYuanyuan ,LIUYesen ,LIUYang,LIUZhengfeng ,YANGWeitao,HUWencai
                            (1.StateKeyLaboratoryofSimulationandRegulationofWaterCycleinRiverBasin,ChinaInstituteof
                                      WaterResourcesandHydropowerResearch,Beijing 100038,China;
                      2.ResearchCenteronFloodandDroughtDisasterReductionoftheMinistryofWaterResources,Beijing 100038,China;
                         3.MWRGeneralInstituteofWaterResourcesandHydropowerPlanningandDesign,Beijing 100120,China;
                            4.FujianWaterConservancyandHydropowerSurveyandDesignInstitute,Fuzhou 350001,China;
                5.GuangxiZhuangAutonomousRegionWaterConservancyandElectricPowerSurveyandDesignInstituteCo.,Nanning 530023,China;
                                     6.TheYi - Shu - SiRiverBasinAdministration,Xuzhou 221018,China)

                  Abstract:Themountainousregionexperiencesfast - flowingandhighlydestructivefloods,posingchallengesfor
                  accurateandtimelyforecasting.Enhancingtheaccuracyandleadtimeoffloodpredictioninmountainousareasisa
                  pressingissue.Addressingthisconcern ,thispaperproposesaninnovativefloodforecastingmethodbasedonma
                  chinelearningtechnology.Theapproachidentifieshistoricalrainfall - floodeventswiththemostsimilaritytothe
                  currentdynamicspatiotemporalfeaturesofrainfall ,employinga“learnfromthepasttopredictthepresent”strate
                  gy.Theresultsindicatethat,insmallwatershedswithminimalhumaninfluenceandabasinareaofapproximately
                       2
                  600km inmountainousregions,themethodnotonlypredictstheoveralltrendofrainfallbutalsoforecaststheasso
                  ciatedmountainousfloodprocessesunderthisrainfalltrend.Theaverageerrorsforpeakflow ,floodvolume,and
                  peaktimeare8.33%,14.27%,and1hour,respectively,meetingtheaccuracyrequirementsforfloodforecasting.
                  Distinguishedfromtraditionalfloodforecastingmethods ,thisapproachpredictsmountainousfloodsfrom theper
                  spectiveoftheoverallrainfalltrend,providingatargetedstrategyforfloodforecastinginsmallwatershedsinhilly
                  areas.
                  Keywords:artificialintelligence;manifoldlearning;spatiotemporalcharacteristicsofrainfall;floodforecasting
                  insmallwatershedsofmountainousregions


                                                                                    (责任编辑:耿庆斋)



























                                                                                                   0
                                                                                              —   1 1 9 —
   9   10   11   12   13   14   15   16   17   18   19