Page 28 - 2024年第55卷第1期
P. 28
2022,26(8):1636 - 1649.
[22] 张辉,戈宝军,韩斌,等.基于 GAF - CapsNet的 电 机 轴 承 故 障 诊 断 方 法 [J].电 工 技 术 学 报,2023,38
(10):2675 - 2685.
[23] 陈之毅,黄鹏飞.基于概率神经网络的地铁车站易损性分析[J].同济大学学报(自然科学版),2021,49
(6):791 - 798.
[24] 杨顺,常晁瑜,余思汗,等.一种适用于黄土斜坡地震稳定性快速评估的综合判别方法 - 以宁夏西吉县为
例[J].地震工程学报,2022,44(1):54 - 61.
[25] 王琪琪,汤井田,张良,等.利 用 多 层 感 知 机 的 地 震 数 据 去 噪 [J].石 油 地 球 物 理 勘 探,2020,55(2):
272 - 281.
[26] 陈鹏,吴一凡,蔡爽,等.基于自编码压缩与多尺度 特 征 提 取 的 抽 水 蓄 能 机 组 劣 化 趋 势 评 估 与 预 测 [J].
水利学报,2022,53(6):747 - 756.
[27] 张君,余佳,任炳昱,等.考 虑 高 寒 低 温 影 响 的 高 心 墙 堆 石 坝 仓 面 施 工 仿 真 模 型 研 究 [J].水 利 学 报,
2022,53(2):200 - 211.
[28] SABOURS,FROSST N, HINTON G E.Matrix capsules with EM routing[C]??Proceedings ofthe 6th
InternationalConferenceonLearningRepresentations ,2018.
[29] SABOURS,FROSSTN,HINTONGE,etal.DynamicroutingbetweenCapsules[C]??Proceedingsofthe31st
ConferenceonNeuralInformationProcessingSystems ,California:CurranAssociates,2017.
[30] 梁辉,赵文光,郭胜山,等.高拱坝- 地基体系整体稳定概率地震风险分析[J].水利学报,2021,52(3):
310 - 322.
[31] 费康,刘汉龙.ABAQUS的二次开发及在土石坝静、动力分析中的应用[J].岩土力学,2010,31(3):881 - 890.
ResearchonimprovedMLPmodelforseismicfragilityanalysisof
rockfilldam byembeddingspatialfeaturesoftime - frequency
1
1
2
1
1
1
WANGXiaoling,LIQingmeng,LIUZongxian,YUJia,YUHongling,WANGHaodong
(1.StateKeyLaboratoryofHydraulicEngineeringIntelligentConstructionandOperation,TianjinUniversity,Tianjin 300072,China;
2.YalongRiverHydropowerDevelopmentCompany,Ltd,Chengdu 610051,China))
Abstract:Inviewofthefactthatthepresentgroundmotionindexessuchaspeakaccelerationandpeakvelocity
usedinseismicfragilityanalysisdonotfullyreflectthecomplexspatialfeaturesoftime - frequencyofgroundmo
tion,andthecurrentseismicdemandmodelbasedontheassumptionoflinearfunctioninlogarithmicspaceisdiffi
culttorevealthecomplexnonlinearrelationshipbetweengroundmotionindexandseismicresponse,animproved
multilayerperceptron(MLP)modelforseismicfragilityofearth - rockfilldamsbyembeddingspatialfeaturesoftime -
frequencyisproposed.Themodelmakesuseoftheadvantagethatcapsulenetwork (CapsNet)canfullycaptureand
characterizethespatiallocationdistributionoftargetfeatures , extractthedeepfeaturesreflectingthespatial
featuresoftime - frequencydistributionofgroundmotionfrom theWavelettime - frequencydiagram ofgroundmo
tion ,andmergethemwithtraditionalfeaturesinthewayoffeaturesplicingtoconstructthegroundmotionfusion
index.Furthermore,thesuperparameterssuchasthenumberofneuronsandlearningrateofMLPareoptimizedby
TreestructuredParzenEstimator(TPE),andtheseismicdemandmodelofearth - rockfilldambasedonTPE - MLP
isproposedtoreflectthecomplexnonlinearrelationshipbetweengroundmotionfusionindexandseismicresponse ,
soastorealizethereliableanalysisofseismicfragilityofearth - rockfilldam.Thecasestudyshowsthat,compared
withthetraditionalgroundmotionindex ,theMAEoftheearth - rockfilldam seismicdemandmodelbasedonthe
groundmotionspatialfeaturesoftime - frequencyfusionindexisreducedby40.5%,indicatingthereliabilityand
superiorityoftheproposedmodel.
Keywords:earth - rockfilldam;seismicfragility;spatialfeaturesoftime - frequencyofgroundmotion;seismic
demandmodel ;CapsNet;MLP;wavelettransform
(责任编辑:李福田)
— 2 3 —