[S-SS14] Strong Ground Motion and Earthquake Disaster

[SSS14-P35] Broadband ground motion waveform synthesis utilizing AI-based upsampling technique

*Takahiro Maeda¹, Hiroyuki Fujiwara¹, Kazutoshi Matsuzaki², Hiromitsu Tomozawa², Yuji Mori² (1.National Research Institute for Earth Science and Disaster Resilience, 2.Mizuho Information and Research Institute, Inc.)

Keywords:broadband ground motion waveform synthesis, coupled learning method, artificial intelligence

A method of synthesizing broadband ground motion by adding short-period components to long-period ground motion based on the relationship between long-period and short-period ground motion waveforms extracted from observation records has been proposed (Iwaki and Fujiwara, 2013). In this study, based on the idea of synthesizing short-period components from long-period component, we applied the relationship between the long-period and broadband ground motion waveforms to the artificial intelligence (AI) by using many strong motion data recorded by strong motion networks such as K-NET and KiK-net.
The problem of predicting broadband ground motion from long-period ground motion can be regarded as a problem of upsampling; predicting missing short-period components and synthesizing high-sampling data from low-sampling data lacking short-period component. In this study, we apply the coupled learning method which is one of the machine-learning methods to this problem. We will report the result of the basic study to perform broadband ground motion waveform synthesis.
Acknowledgments: This study was conducted as part of joint research with RIKEN.