首页|Site classification methodology using support vector machine:A study
Site classification methodology using support vector machine:A study
扫码查看
点击上方二维码区域，可以放大扫码查看
原文链接
万方数据
Site classification methodology using support vector machine:A study
The site effect is a crucial factor when analyzing seismic risk and establishing ground motion attenuation re-lationships.A number of countries have introduced building site classification into earthquake-resistant design codes to account for local site effects on ground motion.However,most site classification indicators rely on drilling data,which is often expensive and requires considerable manpower.As a result,the less detailed drilling data may lead to an undetermined site category of numerous stations.In this study,a Support Vector Machine(SVM)algorithm-based site classification model was trained to address this issue using strong ground motion data and site data from KiK-net and K-net.The classification model used the average HVSR curve of the labeled site and the combined inputs,including frequency,peak,"prominence,and"sharpness"extracted from the curve.The SVM classification model has an accuracy of 76.12％on the test set,with recall rates of 82.69％,75％,and 63.64％for sites Ⅰ,Ⅱ,and Ⅲ,respectively.The precision rates are 75.44％,73.77％,and 87.50％,respectively,with F1 scores of 78.90％,74.38％,and 73.68％.For sites without significant peaks in the HVSR curve,the HVSR curve value was used as the characteristic parameter(input),and the SVM-based site classification model was also trained.The accuracy of class Ⅰ and Ⅱ is 75.86％.The results of this study show higher recall and accuracy rates than those obtained using the spectral ratio curve matching method and GRNN method,indicating a better classification performance.Finally,the generalization ability of the model was verified using some basic stations in Xinjiang deployed by the"National Seismic Intensity Rapid Reporting and Early Warning Project".The SVM-based site classification model that employs strong motion data can provide more reliable classification results for sites without detailed borehole information,and the site classification results can serve as a reference for probing ground motion attenuation relationships,ground motion simulation,and seismic fortification considering the site effect.
Site classificationHVSRMachine learningSupport vector machine
Jing Cai、Nan Xi
展开 >
Early Warning and Quick Report Department of China Earthquake Networks Center,Beijing,China
Site classification HVSR Machine learning Support vector machine
万方数据
