基于机器学习的河网糙率反演

《基于机器学习的河网糙率反演》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、浙江大学建筑工程学院博士学位论文基于机器学习的河网糙率反演姓名：张潮申请学位级别：博士专业：市政工程指导教师：张土乔;毛根海20080101和最优性检验，从而构建起完整的河网糙率反演问题的解决框架。5、Bayesian方法能实现对解分布情况的检验很大程度上取决于它的遍历性，所以计算时间过长是它的缺点。将机器学习引入似然函数的计算中，用BP神经网络训练代替大量重复的正演，提出一种新的BP—Bayesian方法，大大提高了计算效率。6、通过一个工程算例，验证了将基于机器学习的河网糙率直接反演方法应用于复杂大型河网糙率反演的可行性。关键词：河网，糙率反演，直接反演方法，机器

2、学习，数据挖掘，BP神经网络。GA-RBF方法，BP-Bayesian方法，后验分布InversionofroughnessparameterofrivernetworkbasedonmachinelearningAbstractHydraulictechnologyhasbeenmaturesince1970s，butparameterincalculation,suchasroughnessparameterishardtodefineautomaticallyinproject，SOthatautomaticinversionofroughnessparamet

3、erdemandspromptsolution．Presentresearchismainlybased011optimizationmethodandthereexistthreeproblems：Conventionaloptimizationmethodseasilygetsstuckataloc擅doptimum；LackingofresearchOnoptimallayoutofobservestationsandlackingofinversionofroughnessparameterinrivernetworkthroughthetimesequenc

4、eobservationsonfewhydrologystations；Lackingoftestofinversionsolutionandevaluationonuniquenessandoptimization．弛eauthorcIompletedawholeframeworkofinversionofroughnessparameterinfivernetworkbasedonmachinelearningafterresearching011thewholeprocessfromthecalculationofrou露-lnessinversiontothe

5、testofinv饿ionsolution．Theachievernentsofthedissertationareasfollows：1．Anewdirectinversionofroughnessparameterinrivernetworkisproposedbymeansofdataminingbasedonmachinelearningtheery．ItCaninverseroughnessparametersofaUthechannelsinthewholerivernetworksinceitinheritedthetraditionalideaofdi

6、rectinversionandisintroducedintothelatestdevelopmentofartificialIntelligence．Throughafewcalculations，itisfumedouttobehighlyaccuracytomeettheactualdemandofproject．2．Thedissertationexploredthetechniquesofdatecollectingandinversionofloughnessparameterinallthechannelsbyusingthetimesequenceo

7、bservationsOnfewobsel"vestationsinthefivernetwork。ThecalculationshowedthattimesequenceobservationsonfewobservationpointsCallbeusedtoinversetheroughnessparametersofallthechannelsonlybyfullutilizationof两orinformation．Meanwhile，reasonablequantitiesandinstallationpositionof．3。Abstr