A new ground-motion simulation procedure based on feature extraction matching multiple intensity measures

Abstract

A new ground-motion simulation procedure based on feature extraction matching multiple intensity measures is proposed. The new simulation procedure is developed with the aim of generating a suite of acceleration time series at a single site that captures the multiple characteristics of ground motions. Building a regional ground-motion database in the data-driven simulation procedure is the first step of this work. Then, the principal component analysis algorithm is used to extract the main features of the regional database. Moreover, after dimension reduction, the regional eigenquakes database is generated. Multiple intensity measures (i.e., peak ground acceleration, spectral acceleration, significant duration and Arias intensity) are selected to construct a target vector-IM. A suite of target multiple IMs is obtained by sampling from the multivariate normal distribution of the target vector-IM, which is used as constraint conditions to control the goodness-of-fit. Finally, the superposition coefficients of the eigenquakes are determined using the multi-objective genetic algorithm, and the simulated records are generated by superposition in the time domain. Taking Kanto, Japan, as the target region, the proposed simulation procedure is specifically applied. These results indicate that the simulation procedure is able to achieve the coordinated matching of the amplitude, spectrum, duration and cumulative energy of ground motions. This new simulation procedure can provide reasonable input for structural dynamic analysis.