簡易檢索 / 詳目顯示

回結果列表
研究生： 黃郁洺
Yu-Ming Huang
論文名稱： 用於多測站預警的地震神經網路框架
SENSE:A Seismic Neural Network Framework for Multiple Stations Early Warning
指導教授： 陳冠宇
Kuan-Yu Chen
口試委員: 金台齡
Tai-Lin Chin
陳達毅
Da-Yi Chen
吳逸民
Yih-Min Wu
學位類別： 碩士
Master
系所名稱： 電資學院 - 資訊工程系
Department of Computer Science and Information Engineering
論文出版年： 2023
畢業學年度： 111
語文別： 中文
論文頁數： 147
中文關鍵詞： 地面震動預測 、地震預警系統 、卷積神經網路 、多測站預警
外文關鍵詞： Ground Motion Prediction, Convolutional Neural Network, Earthquake Early Earning, Multiple Stations Early Warning
相關次數： 點閱：237  下載：10 
分享至:
查詢本校圖書館目錄 查詢臺灣博碩士論文知識加值系統 勘誤回報

    • 地震預警在降低地震災害風險方面具有重要作用。然而，目前主要的地震預警方法 多半依賴於單一測站的預警，這限制了預警模型對資料的覆蓋範圍和準確性。因此，如 何整合多個測站的資訊進行預警成為一項具有挑戰性的任務。
    我們提出了一種全新的用於多測站預警的地震神經網路框架，名為 SENSE。 SENSE 模型採用全測站的輸入資料方法和輔助訓練模組，以提升預警效能。首先，我們 會同時將所有測站的地理及三軸波型資訊輸入模型，進行空間訊息的編碼，並利用前/後 期特定地點向量(Early/Late Locality-Specific Embeddings)來捕捉不同測站之間的特 徵差異，這麼做有助於減少額外訊息的丟失，增加模型的穩定性。最後，我們引入了閘 門機制模組(Gate Mechanism)，用以調節波形資訊和地理資訊對預警系統的貢獻程度， 進一步提升模型的效能。
    本論文在台灣和日本數據集上進行一系列詳盡的實驗，實驗結果說明我們所提出的 架構與其他最先進的方法相比皆展現了有競爭力或更好的效果，此外，我們還將 SENSE 模型部署上台灣中央氣象局實際運作的系統上進行測試分析，更進一步證明了模型的可 用性。

    Earthquake early warning systems play a crucial role in reducing seismic disaster risks. However, the current predominant methods for earthquake early warning rely heav- ily on single station-based alerts, limiting the warning models’ coverage and accuracy. Therefore, integrating information from multiple stations to improve early warning capa- bility is challenging.
    We propose a novel framework for a fully station-based earthquake early warning model called a seismic neural network framework for multiple stations early warning (SENSE). The SENSE model improves early warning performance by introducing in- novative data input approaches and auxiliary training modules. First,We will input the geographical and three-axis waveform information of all stations into the model simultane- ously, encoding spatial information. Second, we constructed Early/Late Locality-Specific Embeddings, which capture additional feature variations among stations. Finally, we de- signed a Gate Mechanism module to adjust the contributions of waveform and geograph- ical information to the early warning system.
    Extensive experiments are conducted on datasets from Taiwan and Japan in this study. The results demonstrate that our proposed framework exhibits competitive or even supe- rior performance compared to other state-of-the-art methods. Additionally, we deploy the SENSE model on the operational system of the Taiwan Central Weather Bureau for further testing and analysis, providing additional evidence for the model’s usability.

    目錄 摘要............................. ............. I Abstract........................... ............. II 誌謝.......................................... III 目錄.......................................... IV 圖目錄 . . . . . . . . . . . . . . . . . . . . VII 表目錄......................................... IX 1.緒論........................................ 1 1.1地震監測................................... 1 1.2地震預警................................... 1 2相關研究...................................... 3 2.1 地震預警方法................................ 3 2.1.1 單測站預警方法........................... 3 2.1.2 多測站預警方法........................... 3 2.2 機器學習方法................................ 4 2.2.1 卷積神經網路(Convolutional Neural Network) . . . . . . . . . 4 2.2.2 Transformer 地震警報模型(The Transformer Earthquake Alert ingModel)............................ 6 2.2.3 智慧強震預測模型(Intelligent strong motion prediction) . . . . 10 2.2.4 迅速預測地震地面搖晃強度模型(Rapid prediction of earthquake groundshakingintensity)..................... 11 2.2.5 用於地震事件分類的圖卷積網路模型(Graph Convolution Net- worksforSeismicEventsClassification). . . . . . . . . . . . . 12 2.2.6 用於實時地震強度空間插值的深度學習模型(Deep Learning Model for Spatial Interpolation of Real-Time Seismic Intensity) 13 3實驗方法...................................... 15 3.1全測站地震預警模型............................ 15 3.2 前/後期特定地點向量 (Early/Late Locality-Specific Embeddings) . . . . . 17 3.3 閘門機制(GateMechanism) ........................ 19 3.4 結合特定地點向量與閘門機制之全測站地震預警模型(SENSE) . . . . . . 20 3.5 預警分類器(Discrete)............................ 22 4實驗設計...................................... 23 4.1基礎模型架構................................ 23 4.1.1 卷積變壓器模型(TEAM,Conformer) ............... 23 4.1.2 TEAM預警分類器(TEAM,Discrete) ............... 25 4.2資料集.................................... 26 4.2.1 日本資料集............................. 27 4.2.2 台灣資料集............................. 27 4.3評分方式................................... 28 4.3.1 混淆矩陣衡量指標......................... 30 4.3.2 全測站與部分測站評估....................... 31 4.4實驗設定................................... 31 4.4.1 基礎模型實驗設定......................... 31 4.4.2 實驗訓練流程設定......................... 32 4.4.3 實驗的輸入及輸出設定....................... 32 4.4.4 基於抽取特徵模組的實驗設定................... 33 4.4.5 基於全測站地震預警模型的實驗設定................ 33 4.4.6 SENSE模型的實驗設定...................... 34 4.5實驗結果與分析............................... 34 4.5.1基礎系統.............................. 34 4.5.2 SENSE模型............................ 37 4.5.3 全測站模型之消融研究....................... 38 4.5.4 SENSE模型與近期相關研究之比較................ 39 4.5.5 不同輸入長度及排列組合的影響.................. 41 4.5.6 SENSE模型預警時間分析..................... 42 5討論........................................ 45 5.1重大事件探討................................ 45 5.1.1 編號0號地震事件......................... 45 5.1.2 編號1號地震事件......................... 47 5.2即時地震預警系統.............................. 47 6結論與未來展望.................................. 52 參考文獻........................................ 53 授權書......................................... 57 AbstractinChinese.................................. I AbstractinEnglish .................................. II Acknowledgements.................................. III Contents........................................ IV ListofFigures..................................... VII ListofTables ..................................... X ListofAlgorithms................................... XII 1 Introduction.................................... 1 1.1 SeismicMonitor............................... 1 1.2 EarthquakeEarlyWarning ......................... 1 2 RelatedWork ................................... 4 2.1 EarthquakeEarlyWarning ......................... 4 2.1.1 Single-StationWarning....................... 4 2.1.2 Multi-StationWarning ....................... 5 2.2 MachineLearningMethods........................ . 6 2.2.1 ConvolutionalNeuralNetwork.................. . 6 2.2.2 The Transformer Earthquake Alerting Model . . . . . . . . . . . 7 2.2.3 Intelligentstrongmotionprediction . . . . . . . . . . . . . . . . 13 2.2.4 Rapid prediction of earthquake ground shaking intensity . . . . . 14 2.2.5 Graph Convolution Networks for Seismic Events Classification . 15 2.2.6 Deep Learning Model for Spatial Interpolation of Real-Time Seis- micIntensity ............................ 16 3 ProposedMethods................................. 18 3.1 FullStationEarthquakeEarlyWarningModel . . . . . . . . . . . . . . . 18 3.2 Early/LateLocality-SpecificEmbeddings . . . . . . . . . . . . . . . . . 20 3.3 GateMechanism .............................. 22 3.4 SENSEModel................................ 24 3.5 AlertClassifier(DiscreteModule) ..................... 26 4 Experiment .................................... 28 4.1 Baselinemodelarchitecture......................... 28 4.1.1 TEAM(Conformer)......................... 28 4.1.2 TEAMDiscrete........................... 31 4.2 Dataset ................................... 32 4.2.1 JapaneseDataset .......................... 33 4.2.2 TaiwanDataset........................... 34 4.3 Evaluation.................................. 34 4.3.1 ConfusionMatrixMetrics ..................... 35 4.3.2 Fullandpartialstationevaluation ................. 38 4.4 Experimentsettings............................. 38 4.4.1 Experimentaltrainingprocess ................... 38 4.4.2 Experimentalinputandoutputsettings. . . . . . . . . . . . . . . 39 4.4.3 Experimental settings based on the extracted feature module . . . 40 4.4.4 Experimental Settings for the Full-Station Earthquake Early Warn- ingModel.............................. 40 4.4.5 ExperimentalSettingsforSENSE ................. 41 4.5 Experimentalresultsandanalysis...................... 41 4.5.1 Baselinesystem........................... 41 4.5.2 SENSEModel ........................... 44 4.5.3 Ablationstudyofthefullstationmodel . . . . . . . . . . . . . . 47 4.5.4 Comparison of the SENSE Model with Recent Related Studies . . 48 4.5.5 The Impact of Different Input Lengths and Combinations . . . . . 50 4.5.6 SENSEModelLeadingtimeAnalysis............... 51 5 Discussion..................................... 54 5.1 CaseStudy ................................. 54 5.1.1 EarthquakeEventNo.0 ...................... 54 5.1.2 EarthquakeEventNo.1 ...................... 55 5.2 Real-timeEarthquakeEarlyWarningSystem . . . . . . . . . . . . . . . 57 6 Conclusion..................................... 62 References....................................... 63 ............................................. 67

    [1] A. Lomax, A. Michelini, and D. Jozinović, “An investigation of rapid earthquake characterization using single-station waveforms and a convolutional neural net- work,” Seismological Research Letters, vol. 90, no. 2A, pp. 517–529, 2019.
    [2] J. Song, J. Zhu, Y. Wang, and S. Li, “On-site alert-level earthquake early warning using machine-learning-based prediction equations,” Geophysical Journal Interna- tional, vol. 231, no. 2, pp. 786–800, 2022.
    [3] M. Kriegerowski, G. M. Petersen, H. Vasyura-Bathke, and M. Ohrnberger, “A deep convolutional neural network for localization of clustered earthquakes based on mul- tistation full waveforms,” Seismological Research Letters, vol. 90, no. 2A, pp. 510– 516, 2019.
    [4] S.M.MousaviandG.C.Beroza,“Bayesian-deep-learningestimationofearthquake location from single-station observations,” arXiv preprint arXiv:1912.01144, 2019.
    [5] S. M. Mousavi, W. L. Ellsworth, W. Zhu, L. Y. Chuang, and G. C. Beroza, “Earth- quake transformer—an attentive deep-learning model for simultaneous earthquake detection and phase picking,” Nature communications, vol. 11, no. 1, p. 3952, 2020.
    [6] 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, 2021.
    [7] M.A.Bilal,Y.Ji,Y.Wang,M.P.Akhter,andM.Yaqub,“Earlyearthquakedetection using batch normalization graph convolutional neural network (bngcnn),” Applied Sciences, vol. 12, no. 15, p. 7548, 2022.
    [8] S.M.MousaviandG.C.Beroza,“Amachine-learningapproachforearthquakemag- nitude estimation,” Geophysical Research Letters, vol. 47, no. 1, p. e2019GL085976, 2020.
    [9] D. A. Reynolds et al., “Gaussian mixture models.,” Encyclopedia of biometrics, vol. 741, no. 659-663, 2009.
    [10] M. A. Hearst, S. T. Dumais, E. Osuna, J. Platt, and B. Scholkopf, “Support vector machines,” IEEE Intelligent Systems and their applications, vol. 13, no. 4, pp. 18– 28, 1998.
    [11] I.Rishetal.,“Anempiricalstudyofthenaivebayesclassifier,”inIJCAI2001work- shop on empirical methods in artificial intelligence, vol. 3, pp. 41–46, 2001.
    [12] R. Tous, L. Alvarado, B. Otero, L. Cruz, and O. Rojas, “Deep neural networks for earthquake detection and source region estimation in north-central venezuela,” Bul- letin of the Seismological Society of America, vol. 110, no. 5, pp. 2519–2529, 2020.
    [13] K. O’Shea and R. Nash, “An introduction to convolutional neural networks,” arXiv preprint arXiv:1511.08458, 2015.
    [14] W. Zaremba, I. Sutskever, and O. Vinyals, “Recurrent neural network regulariza- tion,” arXiv preprint arXiv:1409.2329, 2014.
    [15] A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,” Advances in neural information pro- cessing systems, vol. 30, 2017.
    [16] O. M. Saad, Y. Chen, A. Savvaidis, S. Fomel, and Y. Chen, “Real-time earthquake detection and magnitude estimation using vision transformer,” Journal of Geophys- ical Research: Solid Earth, vol. 127, no. 5, p. e2021JB023657, 2022.
    [17] T.-L. Chin, K.-Y. Chen, D.-Y. Chen, and T.-H. Wang, “An attention-based hypocen- ter estimator for earthquake localization,” IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1–10, 2021.
    [18] T. Perol, M. Gharbi, and M. Denolle, “Convolutional neural network for earthquake detection and location,” Science Advances, vol. 4, no. 2, p. e1700578, 2018.
    [19] K. Irie, A. Zeyer, R. Schlüter, and H. Ney, “Language modeling with deep trans- formers,” arXiv preprint arXiv:1905.04226, 2019.
    [20] Q. Wang, B. Li, T. Xiao, J. Zhu, C. Li, D. F. Wong, and L. S. Chao, “Learning deep transformer models for machine translation,” arXiv preprint arXiv:1906.01787, 2019.
    [21] B. Zhang, I. Titov, and R. Sennrich, “Improving deep transformer with depth-scaled initialization and merged attention,” arXiv preprint arXiv:1908.11365, 2019.
    [22] I. Shahin, A. B. Nassif, and S. Hamsa, “Novel cascaded gaussian mixture model- deep neural network classifier for speaker identification in emotional talking envi- ronments,” Neural Computing and Applications, vol. 32, pp. 2575–2587, 2020.
    [23] Y.-J.Chiang,T.-L.Chin,andD.-Y.Chen,“Neuralnetwork-basedstrongmotionpre- diction for on-site earthquake early warning,” Sensors, vol. 22, no. 3, p. 704, 2022.
    [24] D. Jozinović, A. Lomax, I. Štajduhar, and A. Michelini, “Rapid prediction of earth- quake ground shaking intensity using raw waveform data and a convolutional neural network,” Geophysical Journal International, vol. 222, no. 2, pp. 1379–1389, 2020.
    [25] G. Kim, B. Ku, J.-K. Ahn, and H. Ko, “Graph convolution networks for seismic events classification using raw waveform data from multiple stations,” IEEE Geo- science and Remote Sensing Letters, vol. 19, pp. 1–5, 2021.
    [26] R.Otake,J.Kurima,H.Goto,andS.Sawada,“Deeplearningmodelforspatialinter- polation of real-time seismic intensity,” Seismological Society of America, vol. 91, no. 6, pp. 3433–3443, 2020.
    [27] C.-T. Lee and B.-R. Tsai, “Mapping vs30 in taiwan,” TAO: Terrestrial, Atmospheric and Oceanic Sciences, vol. 19, no. 6, p. 6, 2008.
    [28] J. Semblat and A. Pecker, “Waves and vibrations in soils: earthquakes, traffic, shocks, construction works (i. press, ed.),” 2009.
    [29] S. Bloemheuvel, J. v. d. Hoogen, D. Jozinović, A. Michelini, and M. Atzmueller, “Multivariate time series regression with graph neural networks,” arXiv preprint arXiv:2201.00818, 2022.
    [30] M. P. van den Ende and J.-P. Ampuero, “Automated seismic source characterization using deep graph neural networks,” Geophysical Research Letters, vol. 47, no. 17, p. e2020GL088690, 2020.
    [31] C. M. Bishop, “Mixture density networks,” 1994.
    [32] A. Gulati, J. Qin, C.-C. Chiu, N. Parmar, Y. Zhang, J. Yu, W. Han, S. Wang, Z. Zhang, Y. Wu, et al., “Conformer: Convolution-augmented transformer for speech recognition,” arXiv preprint arXiv:2005.08100, 2020.
    [33] Y. Okada, K. Kasahara, S. Hori, K. Obara, S. Sekiguchi, H. Fujiwara, and A. Ya- mamoto, “Recent progress of seismic observation networks in japan—hi-net, f-net, k-net and kik-net—,” Earth, Planets and Space, vol. 56, no. 8, pp. xv–xxviii, 2004.
    [34] D. J. Wald, V. Quitoriano, T. H. Heaton, and H. Kanamori, “Relationships between peak ground acceleration, peak ground velocity, and modified mercalli intensity in california,” Earthquake spectra, vol. 15, no. 3, pp. 557–564, 1999.
    [35] D.-Y.Chen,N.-C.Hsiao,andY.-M.Wu,“Theearthwormbasedearthquakealarmre- porting system in taiwan,” Bulletin of the Seismological Society of America, vol. 105, no. 2A, pp. 568–579, 2015.

    QR CODE