Real-time prediction of earthquake ground motions using deep learning

2023 Fiscal Year Final Research Report

• Project/Area Number: 21K18791
• Research Category: Grant-in-Aid for Challenging Research (Exploratory)
• Allocation Type: Multi-year Fund
• Review Section: Medium-sized Section 25:Social systems engineering, safety engineering, disaster prevention engineering, and related fields
• Research Institution: Kyoto University
• Principal Investigator: Goto Hiroyuki 京都大学, 防災研究所, 教授 (70452323)
• Project Period (FY): 2021-07-09 – 2024-03-31
• Keywords: 深層学習 / 地震動

Outline of Final Research Achievements

This research consists of a deep learning model for real-time prediction of ground motions in different regions during an earthquake. The objective is to estimate ground motions from data acquired at seismic stations. Various models were considered based on actual data, but they could not represent the phase characteristics of the ground motions due to insufficient information in the actual data. To compensate for the lack of data, the elastic wave equation was introduced in the framework of PINNs. As a result, the time history data could be reproduced adequately even in the vicinity of the source fault. On the other hand, PINNs lack real-time performance due to computational costs that must be solved.

Free Research Field

地震工学

Academic Significance and Societal Importance of the Research Achievements

データ数の不足を物理法則である波動方程式で補うことによって，深層学習により地震動の予測が可能になる可能性を示したと言える．この結果は震源近傍の地震動を周囲の観測記録から再構成できる可能性も示しているため，例えば2024年能登半島地震のように震源断層直上の地震動の空間分布を適切に表したい場合にも活用できる可能性がある．