地震時常造成生命與錢財的損失，地震發生當下即時的預警可以使造成的傷害降低，其中現地型預警方法可以在地震發生時縮減預警盲區的範圍。本研究旨在開發一個深度學習模型，透過輸入一段單一測站的初始P波波形，以預測該測站的最大地動加速度(Peak Ground Acceleration, PGA)是否會超過25 Gal。為了實現該目標，本研究提出一個名為ResMP的模型，基於ResNet模型架構進行開發，選用此架構之理由在於其能夠克服模型在放置過多層時所產生的梯度消失問題，進而提高效能。本研究所開發之ResMP模型在測試集上之表現有著相當高的準確率，在各個時間窗的F1-score皆超過90%，優於與之進行比較的傳統Pd閥值方法與CNN方法。本研究亦使用數個實際的地震事件對模型進行評估與比較，結果顯示無論在準確度與速度方面，本研究模型皆優於傳統的Pd閥值方法及CNN方法。此外，本研究亦提出一種將人工智慧模型與傳統地震預警系統互相結合的方法，讓模型得以即時的讀取波形並預測結果，進而將模型的應用範圍擴展到現實世界中，不再限於線下的實驗階段。本研究亦於2023年3月時透過此方法將ResMP模型部署於線上即時地震預警系統中，進行效能評估，結果顯示本研究模型提供了可接受的預測結果，在時效方面則達到平均11.7秒的發報時間。

Earthquakes often cause loss of life and property. Instantaneous earthquake early warning can help reduce the damage caused. Among them, on-site early warning method can narrow down the warning blind zone when an earthquake occurs. This study aims to develop a deep learning model that predicts whether the peak ground acceleration (PGA) at a single station will exceed 25 Gal by inputting the initial P-wave waveform of that station. To achieve this goal, a model called ResMP is proposed, using the ResNet model architecture, which can overcome the problem of gradient vanishing when the model has too many layers, thereby improving efficiency.
The ResMP model developed in this study demonstrates high accuracy on the test set, with F1-scores exceeding 90% in various time windows. It outperforms traditional Pd threshold methods and CNN methods. The model was evaluated and compared using several real earthquake events, and the results showed that it consistently outperformed the traditional Pd threshold methods and CNN methods in terms of accuracy and speed.
Additionally, a method to integrate the artificial intelligence model with traditional earthquake early warning systems is also proposed, enabling real-time waveform reading and prediction. This expands the applicability of the model beyond offline experimental phases to real-world scenarios. In March 2023, the ResMP model was deployed in an online real-time earthquake early warning system. Performance evaluation was conducted. The results demonstrate that the ResMP model provides acceptable prediction results, with an average reporting time of 11.7 seconds.

[1] H. Kanamori, "Real-time seismology and earthquake damage mitigation," Annual Review of Earth and Planetary Sciences, vol. 33, no. 1, pp. 195--214, 2005, doi: 10.1146/annurev.earth.33.092203.122626.
[2] Y.-M. Wu, H.-Y. Yen, L. Zhao, B.-S. Huang, and W.-T. Liang, "Magnitude determination using initial P waves: A single-station approach," Geophysical Research Letters, vol. 33, no. 5, p. L05306, 2006, doi: 10.1029/2005GL025395.
[3] Y.-M. Wu and L. Zhao, "Magnitude estimation using the first three seconds P-wave amplitude in earthquake early warning," Geophysical Research Letters, vol. 33, no. 16, p. L16312, 2006, doi: 10.1029/2006GL026871.
[4] C.-Y. Hsieh, V. W. Chao, and Y. Wu, "An Examination of the Threshold-Based Earthquake Early Warning Approach Using a Low-Cost Seismic Network," Seismological Research Letters, vol. 86, no. 6, pp. 164--167, 2015, doi: 10.1785/0220150073.
[5] S. Zhao et al., "Rapid Earthquake Magnitude Estimation Using Deep Learning," presented at the 2022 International Joint Conference on Neural Networks (IJCNN), 2022.
[6] T.-Y. Hsu and C.-W. Huang, "Onsite early prediction of PGA using CNN with multi-scale and multi-domain P-waves as input," Frontiers in Earth Science, vol. 9, p. 626908, 2021, doi: 10.3389/feart.2021.626908.
[7] Y.-J. Chiang, T.-L. Chin, and D.-Y. Chen, "Neural network-based strong motion prediction for on-site earthquake early warning," Sensors, vol. 22, no. 3, p. 704, 2022.
[8] Y. Nakamura, "On the urgent earthquake detection and alarm system (UrEDAS)," in Proc. of the 9th World Conference on Earthquake Engineering, 1988, vol. 7, pp. 673-678.
[9] S. Horiuchi, H. Negishi, K. Abe, A. Kamimura, and Y. Fujinawa, "An Automatic Processing System for Broadcasting Earthquake Alarms," Bulletin of the Seismological Society of America, vol. 95, no. 2, pp. 708-718, 2005, doi: 10.1785/0120030133.
[10] J. M. Espinosa Aranda et al., "Mexico City seismic alert system," Seismological Research Letters, vol. 66, pp. 42-53, 1995, doi: 10.1785/gssrl.66.6.42.
[11] J. M. Espinosa-Aranda et al., "Evolution of the Mexican seismic alert system (SASMEX)," Seismological research letters, vol. 80, no. 5, pp. 694-706, 2009, doi: 10.1785/gssrl.80.5.694.
[12] R. M. Allen and H. Kanamori, "The Potential for Earthquake Early Warning in Southern California," Science, vol. 300, no. 5620, pp. 786-789, 2003, doi: doi:10.1126/science.1080912.
[13] R. M. Allen, H. Brown, M. Hellweg, O. Khainovski, P. Lombard, and D. Neuhauser, "Real-time earthquake detection and hazard assessment by ElarmS across California," Geophysical Research Letters, vol. 36, no. 5, p. L00B08, 2009, doi: 10.1029/2008GL036766.
[14] A. Zollo et al., "Earthquake early warning system in southern Italy: Methodologies and performance evaluation," Geophysical Research Letters, vol. 36, no. 5, p. L00B07, 2009, doi: 10.1029/2008GL036689.
[15] Y.-M. Wu, T.-C. Shin, C.-C. Chen, Y.-B. Tsai, W. H. Lee, and T. L. Teng, "Taiwan rapid earthquake information release system," Seismological Research Letters, vol. 68, no. 6, pp. 931-943, 1997, doi: 10.1785/gssrl.68.6.931.
[16] Y.-M. Wu et al., "Development of an integrated earthquake early warning system in Taiwan-Case for the Hualien area earthquakes," Terrestrial, Atmospheric and Oceanic Sciences, vol. 10, no. 4, pp. 719--736, 1999, doi: 10.3319/TAO.1999.10.4.719(T).
[17] Y.-M. Wu, T.-C. Shin, and Y.-B. Tsai, "Quick and reliable determination of magnitude for seismic early warning," Bulletin of the Seismological Society of America, vol. 88, no. 5, pp. 1254-1259, 1998, doi: 10.1785/BSSA0880051254.
[18] Y.-M. Wu and T.-l. Teng, "A virtual subnetwork approach to earthquake early warning," Bulletin of the Seismological Society of America, vol. 92, no. 5, pp. 2008-2018, 2002, doi: 10.1785/0120010217.
[19] N. C. Hsiao, Y. M. Wu, T. C. Shin, L. Zhao, and T. L. Teng, "Development of earthquake early warning system in Taiwan," Geophysical research letters, vol. 36, no. 5, p. L00B02, 2009, doi: 10.1029/2008GL036596.
[20] C. E. Johnson, A. Bittenbinder, B. Bogaert, L. Dietz, and W. Kohler, "Earthworm: A flexible approach to seismic network processing," IRIS Newsletter, vol. 14, no. 2, pp. 1--4, 1995.
[21] D. Chen, N.-C. Hsiao, and Y. Wu, "The Earthworm Based Earthquake Alarm Reporting System in Taiwan," Bulletin of the Seismological Society of America, vol. 105, no. 2A, pp. 568--579, 2015, doi: 10.1785/0120140147.
[22] R. V. Allen, "Automatic earthquake recognition and timing from single traces," Bulletin of the Seismological Society of America, vol. 68, no. 5, pp. 1521-1532, 1978, doi: 10.1785/bssa0680051521.
[23] R. Allen, "Automatic phase pickers: Their present use and future prospects," Bulletin of the Seismological Society of America, vol. 72, no. 6B, pp. S225-S242, 1982, doi: 10.1785/bssa07206b0225.
[24] W. Zhu and G. C. Beroza, "PhaseNet: a deep-neural-network-based seismic arrival-time picking method," Geophysical Journal International, vol. 216, no. 1, pp. 261-273, 2018, doi: 10.1093/gji/ggy423.
[25] S. M. Mousavi, W. L. Ellsworth, W. Zhu, L. Y. Chuang, and G. C. Beroza, "Earthquake transformer—an attentive deep-learning model for simultaneous earthquake detection and phase picking," Nature communications, vol. 11, no. 1, p. 3952, 2020, doi: 10.1038/s41467-020-17591-w.
[26] W. Y. Liao, E. J. Lee, D. Mu, P. Chen, and R. J. Rau, "ARRU phase picker: Attention recurrent‐residual U‐Net for picking seismic P‐and S‐phase arrivals," Seismological Society of America, vol. 92, no. 4, pp. 2410-2428, 2021, doi: 10.1785/0220200382.
[27] C. A. Cornell, "Engineering seismic risk analysis," Bulletin of the seismological society of America, vol. 58, no. 5, pp. 1583-1606, 1968, doi: 10.1785/BSSA0580051583.
[28] P.-S. Lin and C.-T. Lee, "Ground-Motion Attenuation Relationships for Subduction-Zone Earthquakes in Northeastern Taiwan," Bulletin of the Seismological Society of America, vol. 98, no. 1, pp. 220-240, 2008, doi: 10.1785/0120060002.
[29] C. T. Lee, B. S. Hsieh, C. H. Sung, and P. S. Lin, "Regional Arias intensity attenuation relationship for Taiwan considering Vs 30," Bulletin of the Seismological Society of America, vol. 102, no. 1, pp. 129-142, 2012, doi: 10.1785/0120100268.
[30] J. Münchmeyer, D. Bindi, U. Leser, and F. Tilmann, "The transformer earthquake alerting model: a new versatile approach to earthquake early warning," Geophysical Journal International, vol. 225, no. 1, pp. 646-656, 2020, doi: 10.1093/gji/ggaa609.
[31] Y.-M. Wu and H. Kanamori, "Experiment on an Onsite Early Warning Method for the Taiwan Early Warning System," Bulletin of the Seismological Society of America, vol. 95, no. 1, pp. 347-353, 2005, doi: 10.1785/0120040097.
[32] Y.-M. Wu and H. Kanamori, "Rapid assessment of damage potential of earthquakes in Taiwan from the beginning of P waves," Bulletin of the Seismological Society of America, vol. 95, no. 3, pp. 1181--1185, 2005, doi: 10.1785/0120040193.
[33] K. He, X. Zhang, S. Ren, and J. Sun, "Deep Residual Learning for Image Recognition," presented at the 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016.
[34] K. He, X. Zhang, S. Ren, and J. Sun, "Identity Mappings in Deep Residual Networks," presented at the Computer Vision -- ECCV 2016, 2016.
[35] F. J. Hernandez and F. A. Martinez-Torres, "A new library for interfacing Python with Earthworm: PyEarthWorm," presented at the AGU Fall Meeting Abstracts, 2018.
[36] S. Van Der Walt, S. C. Colbert, and G. Varoquaux, "The NumPy array: a structure for efficient numerical computation," Computing in science & engineering, vol. 13, no. 2, pp. 22--30, 2011, doi: 10.1109/MCSE.2011.37.