Research On Regional Ground Motion Simulation Method Based On Feature Extraction Matching Multi-intensity Measures

Conducting reliable ground motion simulation is one of the most important challenges in earthquake disaster assessment and earthquake engineering.Due to the tectonic and geological diversity,the observed ground-motion records may vary regionally.In this paper,a regional ground motion simulation framework is built based on machine learning and optimization algorithm in order to make the simulated records fit the main characteristics of regional ground motion.Taking the Kanto region of Japan as an example,the proposed regional ground motion simulation framework is applied as follows:1.In the regional ground motion simulation framework built in this paper,the principal component analysis algorithm is used to extract the features of regional database to obtain a set of characteristic ground motion mother waves.Under the given earthquake scenarios,the corresponding target values of simulated sites are given by the regional ground motion prediction equations,i.e.,peak ground acceleration（PGA）,spectral acceleration（S_A）,significant duraion(D_S5-75)and Arias intensity（I_A）.These four ground motion intensity measures as simulation constraints can represent the amplitude,spectrum,duration and cumulative energy of ground motion,respectively.Finally,the multi-objective genetic algorithm is used to determine the combination coefficient corresponding to the superposition of each characteristic ground motion mother waves of in the time domain,and the simulated ground motion records are generated.2.Taking Kanto,Japan as the target region,the proposed simulation framework was applied specifically.The ground motions recorded by Japan’s Kyoshin Network（K-NET）from 2000 to 2020 were collected and the Kanto regional ground motion database was built.Base on the classification scheme for Japan earthquake source categories,the regional database was classified.Then,each record was processed uniformly（i.e.,baseline correction,band-pass filtering）.In addition,the regional database was grouped according to the significant duraion(D_S5-75).And the feature extraction process was applied to each group by the principal component analysis algorithm to obtain the characteristic ground motion mother waves for each group.3.Ground motion prediction equations（GMPEs）for non-spectral ground motion intensity measures in the Kanto region,i.e.,Arias intensity（I_A）,cumulative absolute velocity（CAV）,and significant duraion(D_S5-75and D_S5-95),were developed by support vector regression（SVR）.To verify the rationality and effectiveness of the SVR GMPEs,the performance indices（e.g.,correlation coefficients,slope coefficients）and residuals were analyzed.Furthermore,the SVR GMPEs were also compared with observed data and the previous GMPEs.4.To demonstrate the effect of the regional ground motion simulation framework,specific earthquake scenarios and simulation sites were artificially set.The target duration of each simulation site corresponds to the six groups of the actual record by significant duration,respectively.Therefore,the characteristic ground motion mother waves extracted from each group were used for the calculation.The calculated values of simulated ground motions were matched with the target values of each intensity measures given by the regional GMPEs,thus achieving a comprehensive consideration of the regional ground motion characteristics.