Page 52 - 2024年第55卷第1期
P. 52

suspendedsedimenttransportinthelowerreachesisreducedsignificantly,andconditionsofwaterandsediment
                  changeobviously,whichhasasignificantimpactonthereservoirsedimentationprocessandmakestheroleofpeb
                  blebedloadmoreimportantinthetotalsedimenttransport.Therefore ,itisofgreatofsignificancetocarryoutthe
                  lawofpebblebedloadtransportinthelowerreachesoftheJinshaRiverundernewwaterandsedimentconditions,
                  andtoputforwardtheapplicableincipientvelocityandtransportrateofpebblebedload.Basedonthemeasured
                  dataatSanduiziHydrologicalStationfrom2008to2021,theclassicalincipientvelocityformulaandbedloadtrans
                  portrateformulaaretestedandcorrected.Theresultshowsthatcomparedwithclassicalformulae,thecalculation
                  accuracyofthemodifiedformulaissignificantlyimprovedbyintroducingacoefficientwiththeparticlesizeasavari
                  able ,whichinducesdifferentparticlesizeranges.Meanwhile,theresultsofclassicalbedloadtransportratedevi
                  ategreatlyfromthemeasuredvalueswhenthesedimenttransportintensityislow.Byestablishingtheintensityrela
                  tionship ofsedimenttransportand waterflow and introducinghidden parameterstomodified formula , the
                  calculationaccuracyoflowsedimenttransportrateisalsoobviouslyimproved.Thetransportlawofpebblebedload
                  inSanduizireachcanbedmoreeffectivelyreflectedbythemodifiedformulaeofincipientvelocityandbedload
                  transportrate.
                  Keywords:lowerreachesofJinshaRiver;cascadereservoir;pebblebedload;incipientvelocity;pebblebed
                  loadtransportrate

                                                                                    (责任编辑:李福田)



              (上接第 34页)
                               Anintelligentextractionmethodoftunnelconstructionactivity
                                    durationinformationbasedonimprovedHT - LCNN

                                                                                                 1
                                                                                 3
                                                           1
                                                                     1
                                                  2
                                 1
                        XIAOYao,ZHONGDenghua,YUjia,HUYike,XUGuoxin,CHENQiutong
                 (1.StateKeyLaboratoryofHydraulicEngineeringIntelligentConstructionandOperation,TianjinUniversity,Tianjin 300072,China;
                         2.CollegeofWaterResourcesandCivilEngineering,ChinaAgriculturalUniversity,Beijing 100083,China;
                   3.Hanjiang - to - WeiheRiverValleyWaterDiversionProjectConstructionCo.,Ltd.,ShaanxiProvince,Xi’an 710302,China)
                  Abstract:Extractionofactivitydurationinformationisimportantforscheduleanalysisanddecisionoftunnelcon
                  struction.AcommonwaytorecordtheactivitydurationsoftunnelconstructionisdrawingthelinesegmentasGa
                  nttchartintheconstructionlog.However,thetraditionalextractionmethodthatreliesonmanualstatisticsisfa
                  cingadilemmabetweenefficiencyandaccuracy.Tosolvethisproblem,thispaperadoptsalinesegmentdetec
                  tionmethodbasedondeeplearning.Anautomaticextractionmethodoftunnelconstructionactivitydurationinfor
                  mationbasedonimprovedHoughTransform - LineConvolutionalNeuralNetwork(HT - LCNN)isproposed.First,
                  thepreprocessingisbasedonhomographytosolvetheobliquity,rotationandtwistoftheoriginalphotos.Second,
                  theglobalcontextnetwork(GCNet),whichhastheabilityofestablishingthelong - distancedependenceofthe
                  featuregraphandthedependencebetweenchannels ,isintegratedintotheresidualmoduleofHT - LCNNtoim
                  provethemodel ’sattentiontothetargetlinesegmentofGanttchart,reducetheinterferencefrom thelineseg
                  mentofthetableandtext ,andimprovethedetectionaccuracy.Third,thecoordinatesofdetectedlinesareauto
                  maticallyconvertedintoconstructionactivitydurations.Theproposedmethodisusedtoextracttheactivitydura
                  tionsofaTBM diversiontunnelprojectfrom17monthsofconstructionlogs.Thecasestudyshowsthatthedetec
                  tionprecisionAP 5 ,AP 10 andAP 15 are94.7%,95.0% and95.1% respectively,whicharehigherthantheresults
                  ofHT - LCNNandLCNN.Comparedwithmanualextractionresults ,themeanabsoluteerrorofourautomaticex
                  tractionmethodisonly1.82min.Thisstudyprovidesanovelideaforextractionoftunnelconstructionactivitydu
                  rationinformationaccuratelyandefficiently.
                  Keywords:durationoftunnelconstructionactivities;informationintelligentextraction;linesegmentdetection;
                  deepnetworkwithHoughtransform;attentionmechanism

                                                                                    (责任编辑:李福田)



                                                                                                 —  4 7 —
   47   48   49   50   51   52   53   54   55   56   57