Application Research Of Cluster Analysis On Micro-tremor's H/V Spectra

Site conditions have an important influence on ground motion and the determination of site characteristic parameters is the primary task of the study on site condition.Micro-tremor's H/V spectra method is widely used to evaluate the dynamic characteristics of the site because of its economy.However,limited by the incomplete theoretical assumptions,in complex excitation sources site,H/V curve not completely consistent with the dynamic response or transfer function.It's difficulty for Micro-tremor's H/V spectra to effectively eliminate the influence of excitation sources,and establish the relationship between H/V spectra and field characteristic parameters.Based on the efficient performance of cluster analysis in unsupervised learning sample features and automatic classification,this paper studies the H/V spectrum ratio extracted to represent site information,establishes the relationship between its predominant frequency and site parameters.The main contents and achievements of this dissertation are as follows:1.The clustering effectiveness of K-means,hierarchical clustering and FCM clustering on multi-peak spectral curves was compared.In-class and between-class difference of Micro-tremor's H/V spectra ratio after clustering and their relationship with excitation sources and site features were analyzed.Site correlation H/V spectral ratio extraction method based on K-means was constructed.The H/V spectra ratio clustering analysis of 20 complex sites show that the clustering results are not affected by the initial value selection and more stable.Comparison with site transfer function shows that the precision of the predominant frequency determined by clustering analysis method has been significantly improved in 19 fields.2.Through Micro-tremor observation and H/V spectral ratio cluster analysis of the thin overburden site in the desert and thick overburden site in the southeast coast.The applicability of site characteristic parameter test method based on Micro-tremor single point spectral ratio predominant frequency is tested.The range of applicable frequency band is revised and extended,the method of site classification based on H/V curve predominant frequency of microtremor is presented.The results show that in desert region,the predominant frequency of H/V spectral curve of ground motion and microtremor has high consistency,when use the method to determine the site characteristic parameters based on the single point spectral ratio of microtremor tocalculate the thickness of sediments and site period,deep overburden sites such as the eastern coast will be underestimated and shallow overburden sites such as the desert will be overestimated.The method was modified with the results of 46 microtremor tests,and the fitting relationship between the predominant frequency and the site characteristic parameters with the interval distribution of 0.58~12.5Hz in frequency domain was obtained.Discrimination of 235 sites based on H/V curve predominant frequency,the accuracy is 87%,which is far higher than the 28% accuracy rate of the technical procedures of site micro-tremor measurement.3.The difference between equivalent shear wave velocity,overburden thickness and site classification was analyzed.The distance function was constructed to cluster 79H/V spectral ratios.A new method of site classification based on the similarity of H/V spectral ratio is established for region and single site respectively.The results show that for the H/V spectral ratio of these 79 sites,distance function used to classify site class,error is 0.When classify equivalent shear wave velocity of the site,error is 9%.Cluster analysis was used to classify Karamay site characteristic parameters and compared with traditional fitting method,results show that the two results are consistent in the calculation of overburden thickness.When classifying equivalent shear waves and site classify,the clustering method takes into account the characteristics such as the prominent frequency and peak of H/V spectral ratio.The characteristic parameters of four different sites were classified using the single site characteristic parameters classification method,and results showed that the overburden thickness of only one site was classified incorrectly.