[S-SS10] 強震動・地震災害

[SSS10-02] Single-station CNN-based earthquake early warning based on the twin algorithm

*Ting Yu Hsu¹、Wen I Liao² (1.National Taiwan University of Science and Technology、2.National Taipei University of Technology)

キーワード：earthquake early warning, CNN, Onsite

Previous research had preliminarily developed a CNN model for predicting PGA. This technique utilizes raw acceleration data within two seconds after P arrival through multi-scale and multi-domain processing as input to the CNN model. The model is trained and validated using seismic data from TSMIP, demonstrating favorable performance in offline verifications across various seismic events. This study aims to practically integrate the CNN model into the Earthworm system in the Central Weather Administration, incorporating a Twin mechanism as the reporting criterion to ensure swift and accurate warnings during real-time operation. The CNN model has been successfully integrated into the Earthworm system in the current study, utilizing equipment provided by the Central Weather Administration for online computation. Throughout this process, improvements have been made to address program issues, enhancing the real-time performance and accuracy of the system. Recent online prediction results for earthquake events will be shown in this study. The results show accurate alert warnings, but the lead time is not as expected probably due to the unsuitable Python algorithm employed.