Feature Extraction and Analysis of Earthquake Motion Using the Gaussian Mixture Model

Abstract

For the purpose of evaluation and analysis of earthquake motion characteristics, the method of characteristics extraction for earthquake motion developed by the authors was applied to the time-frequency analysis of earthquake motion. For analysis of earthquake motion, the Gaussian mixture model was applied to the 99-dimensional feature vector representing the temporal characteristics of a strong motion on the basis of its Husid plot. In this study, the period-dependent feature vector based on the evolutionary power spectrum was defined. Furthermore, the Gaussian mixture model was applied to the period-dependent feature vector. The proposed method was applied to the earthquake motions observed from the Tokaido-oki earthquake that occurred at 23:57 September 5, 2004. The period-dependent feature vector and the Gaussian mixture model by proposed method were included of the characteristics of earthquake motion. For the dispersibility of surface wave found in the feature vector, the element model from the Gaussian mixture model was compared with the dispersion curve calculated from the soil profile.