International Session (Disaster Prevention and Mitigation)

[CS2-29] SYNTHESIS OF DESIGN INPUT GROUND MOTIONS WITH FEATURE IDENTIFICATION BY AUTOENCODER

〇Di Lin¹, MuYao Xiao¹, Riki Honda¹ (1.University of Tokyo)

Keywords:input ground motions, nonlinear structural response, autoencoder

For the design purpose, a tough ground motion is desirable. Many researches focused on the frequency characteristics of ground motion, but ignored the time history. For non-linear response, time domain is also essential; two ground motion with the same amplitude but different time history characteristics can induce quite different response on one structure. Modifying the ground motion without increasing its amplitude but make it more tough for structures should be considered. Since machine learning is a powerful tool to process a large amount of data we propose to use AutoEncoder to identify the features of ground motions, so that they can be exploited for the synthesis of design ground motions.