地震动峰值速度预测：基于地形特征的随机森林方法

打开文本图片集

中图分类号：P65 文献标识码：A 文章编号：2097-5465（2025）04-0069-07

BAOYu，WANGYanwei，GUIFei

GuilinUniversity ofTechnology，Guilin541004，China

Abstract：PeakGroundVelocity（PGV）isanintensityparameterforassesingthedestructivenesofarthquakes，anditsfast andaccuratepredictioniscrucialforearthquakeearlywarning.Studies haveshownthat topographyhasasignificanteffectothe intensityofgroundshaking.However，traditionalempiricalpredictionmodelsaredificulttoconsidercomplexfactorssuchas topography，whichlimitstheacuracyofpredictionresults.Inthispaper，arandomforest-basedPGVpredictionmodelisproosed byestablishinganonlinearmapping relationship betweePGVandmagnitude，epicenterdistance，andmultiple topographic factors.Themodel wastrainedandvalidatedusing31404 strongsurfaceearthquakerecordsfromtheKiK-netdatabaseinJapan， andcomparedandanalyzed withthewidelyusedBAO8andCYO8 models.TheresultsshowthatthecorelationoftheRFmodelin predicting PGV is improved by about 18% ，the standard deviation of the error isreduced byabout 35% ，and the problems of "overestimationofsmallvalues"and"underestimationoflargevalues"，whicharecommonlyfoundinthetraditionalmodels，are effectively avoided.

Keywords：peak ground velocity；topographic features；random forests；site conditions

来稿日期：2025-05-01 DOI：10.13937/j.cnki.hbdzdxxb.2025.04.009

基金项目：广西自然科学基金（2021GXNSFFA196001）

作者简介：鲍余（2000一），男，浙江宁波人，硕士研究生，研究方向为地震预警。（剩余9263字）