New Development of Strong Motion Evaluation and Realtime Prediction by Deep Learning

2021 Fiscal Year Final Research Report

• Project/Area Number: 19K22002
• Research Category: Grant-in-Aid for Challenging Research (Exploratory)
• Allocation Type: Multi-year Fund
• Review Section: Medium-sized Section 23:Architecture, building engineering, and related fields
• Research Institution: Tohoku University
• Principal Investigator: OHNO SUSUMU 東北大学, 災害科学国際研究所, 准教授 (40361141)
• Project Period (FY): 2019-06-28 – 2022-03-31
• Keywords: 深層学習 / 強震動評価 / 建物応答 / 即時予測

Outline of Final Research Achievements

In order to mainly improve the evaluation accuracy of seismic motion, which is less accurate than that of the building side, this study challenged the prediction of strong ground motion, real-time and spatial evaluation of seismic motion, and building response prediction using deep learning. We used a large number of strong-motion records including large-amplitude records of the Tohoku region, Japan.
Although it should be noted that deep learning is highly dependent on the number and distribution of data, we showed that deep learning can model complex phenomena and contribute to evaluation accuracy improvement. Although it takes time to learn, neural network can be applied immediately and is effective for earthquake early warning and immediate response for earthquake disasters.

Free Research Field

地震工学

Academic Significance and Societal Importance of the Research Achievements

深層学習による物理シミュレーションの代理モデル作成は，複雑な現象をモデル化できること，学習に時間はかかるが即時に適用可能な点が大きな特徴である。本研究では，単に震度予測にとどまらず，建物では高次モードまで含めた多成分の応答波形予測が可能であること，地震動では事前予測としてスペクトルの高精度評価，早期地震警報としてスペクトルの即時予測，地震直後ではスペクトルの面的評価がそれぞれ可能であることを示した。早期地震警報や発災時対応に有効と思われる。