A Machine-Learning-Based Software for the Simulation of Regional Characteristic Ground Motion

Abstract

Ground-motion simulations provide input time history data required for designing and assessing structures; however, the simulations conducted by the currently available tools only match the design spectrum without verifying if the statistical characteristics of the spectrum and duration are satisfied. A ground-motion simulation software was developed to resolve these issues. The developed software employs machine learning methods to match the amplitude, spectrum, and duration features of the target region. Principal component analysis is employed to extract features from the actual ground-motion database to detect characteristic ground motions and predict the target acceleration amplitude, response spectrum, and duration, based on the response spectrum and duration prediction equations. The results show that the simulated ground motion can match the amplitude, spectrum, and duration characteristics well. Therefore, the simulated ground motion can provide more reasonable input for the structure. Moreover, the developed software provides visualization functions that enable the user to determine the target area and obtain the amplitude field intuitively.