Research on the application of transformer model and transfer learning in earthquake P-wave and noise discrimination
-
摘要:
准确可靠地区分地震和噪声信号对于地震危险性分析和地震预警至关重要.然而, 无处不在且复杂的噪声信号使这项任务充满挑战.针对中国和日本数据的差异, 本研究在深度学习模型训练过程中采取了不同的策略来区分地震和噪声信号.首先, 鉴于日本数据丰富, 直接训练一个Transformer模型, 该模型在日本的判别准确率为99.82%.其次, 为缓解数据不平衡, 对中国地震数据采用了随机滑动波形窗进行增强.还使用中国数据对预先训练的日本模型进行了微调, 以更好地适应中国数据集.经过微调后, 模型在中国的判别准确率为99.47%.结果表明, 使用原始波形训练的深度学习模型进行地震事件判别时能够取得很高的准确率.此外, 迁移学习模型在门源6.9级地震和漾濞序列震中得到了良好的验证, 表明迁移学习在台网稀疏地区的应用是有效的, 这为地震学和地震预警提供了一种潜在的方法.
-
关键词:
- 地震预警 /
- 深度学习 /
- 迁移学习 /
- 数据增强 /
- Transformer模型
Abstract:It is crucial for seismic hazard analysis and earthquake early warning (EEW) to discriminate earthquake and noise accurately and reliably. However, ubiquitous and complex noise signals make this a challenging task. Aiming at the differences between Chinese and Japanese datasets, this study adopts different strategies to discriminate earthquake and noise signals during the deep learning model training process. Firstly, because of the abundant data in Japan, we directly trained a Transformer model and the model has a discrimination accuracy of 99.82% in Japan. Secondly, to alleviate the data imbalance, the Chinese seismic data were augmented by a random sliding window of the waveform. We also fine-tuned the pre-trained Japanese model using Chinese data to better fit the Chinese dataset. After fine-tuning, the model has a discrimination accuracy of 99.47% in China. The results demonstrate that the deep learning model trained by the raw waveform can achieve high accuracy in seismic event discrimination. In addition, the transfer learning model has been well tested in the Menyuan M6.9 earthquake and Yangbi sequence earthquakes, which indicates that transfer learning is effective when applied to areas with sparse seismic networks. This provides a potential method for seismology and EEW.
-
-
图 6
Transformer-PN模型在日本测试集上的混淆矩阵
Figure 6.
Confusion matrix of Transformer-PN model on Japanese test set
图 7
Transformer-TL-PN模型在中国测试集上的混淆矩阵
Figure 7.
Confusion matrix of Transformer-TL-PN model on Chinese test set
图 8
正确分类和错误分类地震和噪声信号的例子
Figure 8.
Examples of correctly classified and misclassified earthquake and noise signals
图 9
Transformer-TL-PN模型对地震震级和震中距的判定性能
Figure 9.
The discrimination performance of Transformer-TL-PN model for earthquake magnitude and epicentral distance
图 10
Transformer-TL-PN模型输出为真实P波的概率
Figure 10.
Transformer-TL-PN model output probability of being a real P wave
图 11
Transformer-TL-PN模型在门源地震上的表现
Figure 11.
The performance of Transformer-TL-PN model in Menyuan earthquake
图 13
Transformer-TL-PN模型在漾濞序列震上的表现
Figure 13.
The performance of Transformer-TL-PN model in Yangbi sequence earthquakes
图 15
Transformer-TL-PN模型在连续波形上的表现
Figure 15.
The performance of Transformer-TL-PN model on continuous waveforms
表 1
不同模型在日本测试集上的表现
Table 1.
The performance of different models on the Japanese test set
模型名称 判别类型 召回率(%) 精度(%) F1 (%) 准确率(%) Transformer-PN 地震 99.79 99.86 99.82 99.82 噪声 99.86 99.79 99.82 CNN-PN 地震 97.93 98.12 98.02 98.03 噪声 98.13 97.94 98.03 LSTM-PN 地震 98.53 98.62 98.57 98.58 噪声 98.63 98.53 98.58 
下载: 导出CSV
表 2
不同模型在中国测试集上的表现
Table 2.
The performance of different models on the Chinese test set
模型名称 判别类型 召回率(%) 精度(%) F1 (%) 准确率(%) CNN-PN 地震 91.11 92.24 91.68 92.62 噪声 93.83 92.91 93.37 LSTM-PN 地震 92.64 93.22 92.93 93.41 噪声 94.60 93.56 94.08 Transformer-PN 地震 94.44 95.53 94.98 95.56 噪声 96.45 95.58 96.01 Transformer-TL-PN 地震 99.26 99.58 99.42 99.47 噪声 99.67 99.42 99.54
下载: 导出CSV
表 3
漾濞序列震例
Table 3.
Yangbi sequence earthquake cases
发震时刻 经度(°E) 纬度(°N) 震级MS 震源深度/km 2021-05-21 20∶56∶02 99.93 25.63 4.2 8 2021-05-21 21∶21∶25 99.92 25.63 5.6 10 2021-05-21 21∶23∶43 99.97 25.66 4.5 8 2021-05-21 21∶48∶34 99.87 25.67 6.4 8 2021-05-21 21∶53∶47 99.98 25.62 4.1 9 2021-05-21 21∶55∶28 99.89 25.67 5.0 8 2021-05-21 21∶56∶37 99.95 25.64 4.9 8 2021-05-21 22∶02∶00 99.89 25.66 4.1 8 2021-05-21 22∶15∶16 99.96 25.59 4.0 8 2021-05-21 22∶19∶48 99.95 25.60 3.0 8 2021-05-21 22∶30∶27 99.90 25.65 3.1 8 2021-05-21 22∶31∶10 99.97 25.59 5.2 8 2021-05-2122∶59∶37 99.94 25.63 3.5 8 2021-05-21 23∶08∶57 99.98 25.61 3.0 8 2021-05-22 00∶51∶41 99.87 25.70 4.0 8 2021-05-22 00∶53∶31 99.91 25.65 3.2 8 2021-05-22 00∶56∶07 99.91 25.63 3.2 9 2021-05-22 01∶36∶06 99.94 25.62 3.5 9
下载: 导出CSV
-
Allen R V. 1978. Automatic earthquake recognition and timing from single traces. Bulletin of the Seismological Society of America, 68(5): 1521-1532, doi: 10.1785/bssa0680051521.
Brown J R, Beroza G C, Shelly D R. 2008. An autocorrelation method to detect low frequency earthquakes within tremor. Geophysical Research Letters, 35(16): L16305, doi: 10.1029/2008GL034560.
Chai C P, Maceira M, Santos-Villalobos H J, et al. 2020. Using a deep neural network and transfer learning to bridge scales for seismic phase picking. Geophysical Research Letters, 47(16): e2020GL088651, doi: 10.1029/2020GL088651.
Chicco D, Jurman G. 2020. The advantages of the Matthews correlation coefficient (MCC) over F1 score and accuracy in binary classification evaluation. BMC Genomics, 21(1): 6, doi: 10.1186/s12864-019-6413-7.
Cuéllar A, Suárez G, Espinosa-Aranda J M. 2018. A fast earthquake early warning algorithm based on the first 3 s of the P-wave coda. Bulletin of the Seismological Society of America, 108(4): 2068-2079, doi: 10.1785/0120180079.
Ide H, Kurita T. 2017. Improvement of learning for CNN with ReLU activation by sparse regularization. //2017 International Joint Conference on Neural Networks (IJCNN). Anchorage, AK, USA: IEEE, 2684-2691, doi:
10.1109/IJCNN.2017.7966185 .Jia R S, Sun H M, Peng Y J, et al. 2017. Automatic event detection in low SNR microseismic signals based on multi-scale permutation entropy and a support vector machine. Journal of Seismology, 21(4): 735-748, doi: 10.1007/s10950-016-9632-2.
Jozinović D, Lomax A, Štajduhar I, et al. 2022. Transfer learning: improving neural network based prediction of earthquake ground shaking for an area with insufficient training data. Geophysical Journal International, 229(1): 704-718, doi: 10.1093/gji/ggab488.
Kamigaichi O, Saito M, DoiK, et al. 2009. Earthquake early warning in Japan: Warning the general public and future prospects. Seismological Research Letters, 80(5): 717-726, doi: 10.1785/gssrl.80.5.717.
Kim G, Ku B, Ahn J K, et al. 2022. Graph convolution networks for seismic events classification using raw waveform data from multiple stations. IEEE Geoscience and Remote Sensing Letters, 19: 3004805, doi: 10.1109/lgrs.2021.3127874.
Kingma D P, Ba J. 2017. Adam: A method for stochastic optimization. arXiv preprint arXiv: 1412.6980.
Kong Q K, Wang R J, Walter W R, et al. 2022. Combining deep learning with physics based features in explosion-earthquake discrimination. Geophysical Research Letters, 49(13): e2022GL098645, doi: 10.1029/2022GL098645.
Kuang W H, Yuan C C, Zhang J. 2021. Network-based earthquake magnitude determination via deep learning. Seismological Research Letters, 92(4): 2245-2254, doi: 10.1785/0220200317.
Li Z F, Meier M A, Hauksson E, et al. 2018. Machine learning seismic wave discrimination: application to earthquake early warning. Geophysical Research Letters, 45(10): 4773-4779, doi: 10.1029/2018gl077870.
Lin B H, Jin X, Kang L C, et al. 2021. The research of earthquake magnitude determination based on Convolutional Neural Networks. Chinese Journal of Geophysics (in Chinese), 64(10): 3600-3611, doi: 10.6038/cjg2021O0370.
Liu H Y, Li S Y, Song J D. 2022. Discrimination between earthquake P waves and microtremors via a generative adversarial network. Bulletin of the Seismological Society of America, 112(2): 669-679, doi: 10.1785/0120210231.
Lu J, Behbood V, Hao P, et al. 2015. Transfer learning using computational intelligence: A survey. Knowledge-Based Systems, 80: 14-23, doi: 10.1016/j.knosys.2015.01.010..
Münchmeyer J, Bindi D, Leser U, et al. 2021. Earthquake magnitude and location estimation from real time seismic waveforms with a transformer network. Geophysical Journal International, 226(2): 1086-1104, doi: 10.1093/gji/ggab139.
Ma Q, Jin X, Li S Y, et al. 2013. Automatic P-arrival detection for earthquake early warning. Chinese Journal of Geophysics (in Chinese), 56(7): 2313-2321, doi: 10.6038/cjg20130718.
Maggi A, Ferrazzini V, Hibert C, et al. 2017. Implementation of a multistation approach for automated event classification at Piton de la Fournaise Volcano. Seismological Research Letters, 88(3): 878-891, doi: 10.1785/0220160189.
Meier M A, Ross Z E, Ramachandran A, et al. 2019. Reliable real-time seismic signal/noise discrimination with machine learning. Journal of Geophysical Research: Solid Earth, 124(1): 788-800, doi: 10.1029/2018jb016661.
Mousavi S M, Ellsworth W L, Zhu W Q, et al. 2020. Earthquake transformer—an attentive deep-learning model for simultaneous earthquake detection and phase picking. Nat. Commun., 11(1): 3952, doi: 10.1038/s41467-020-17591-w.
Mousavi S M, Beroza G C. 2022. Deep-learning seismology. Science, 377(6607): eabm4470, doi: 10.1126/science.abm4470.
Mousavi S M, Beroza G C. 2023. Machine learning in earthquake seismology. Annual Review of Earth and Planetary Sciences, 51(1): 105-129, doi: 10.1146/annurev-earth-071822-100323.
Pan Z R, Zhou Y, Miao Z P, et al. 2023. Comparative analysis on instrument intensaity and macroscopic survey intensity of the Menyuan M6.9 earthquake on January, 8, 2022. Plateau Earthquake Research (in Chinese), 35(3): 8-15.
Peng C Y, Jiang P, Ma Q, et al. 2022. Chinese nationwide earthquake early warning system and its performance in the 2022 Lushan M6.1 earthquake. Remote Sensing, 14(17): 4269, doi: 10.3390/RS14174269.
Pölsterl S. 2020. scikit-survival: a library for time-to-event analysis built on top of scikit-learn. J. Mach. Learn. Res., 21: 1-6.
Raffel C, Shazeer N, Roberts A, et al. 2020. Exploring the limits of transfer learning with a unified text-to-text transformer. J. Mach. Learn. Res., 21: 1-67.
Ross Z E, Meier M A, Hauksson E. 2018. P wave arrival picking and first-motion polarity determination with deep learning. Journal of Geophysical Research: Solid Earth, 123(6): 5120-5129, doi: 10.1029/2017jb015251.
Saad O M, Chen Y F, Trugman D, et al. 2022. Machine learning for fast and reliable source-location estimation in earthquake early warning. IEEE Geoscience and Remote Sensing Letters, 19: 8025705, doi: 10.1109/lgrs.2022.3142714.
Satriano C, Wu Y M, Zollo A, et al. 2011. Earthquake early warning: Concepts, methods and physical grounds. Soil Dynamics and Earthquake Engineering, 31(2): 106-118, doi: 10.1016/j.soildyn.2010.07.007.
Skoumal R J, Brudzinski M R, Currie B S, et al. 2014. Optimizing multi-station earthquake template matching through re-examination of the Youngstown, Ohio, sequence. Earth and Planetary Science Letters, 405: 274-280, doi: 10.1016/j.epsl.2014.08.033.
Song J D, Yu C, Li S Y. 2021. Continuous prediction of onsite PGV for earthquake early warning based on least squares support vector machine. Chinese Journal of Geophysics (in Chinese), 64(2): 555-568, doi: 10.6038/cjg2021O0193.
Song J D, Zhu J B, Li S Y. 2023. MEANet: Magnitude estimation via physics-based features time series, an attention mechanism, and neural networks. Geophysics, 88(1): V33-V43, doi: 10.1190/geo2022-0196.1.
Song J D, Zhu J B, Wei Y X, et al. 2023. Backtracking verification of machine learning earthquake early warning magnitude estimation and on-site threshold alarm for Menyuan M6.9 earthquake in Qinghai on January 8, 2022. Chinese Journal of Geophysics (in Chinese), 66(7): 2903-2919, doi: 10.6038/cjg2022Q0050.
Srivastava N, Hinton G, Krizhevsky A, et al. 2014. Dropout: a simple way to prevent neural networks from overfitting. J. Mach. Learn. Res., 15(1): 1929-1958.
Strauss J A, Allen R M. 2016. Benefits and costs of earthquake early warning. Seismological Research Letters, 87(3): 765-772, doi: 10.1785/0220150149.
Su R H, Yuan D Y, Xie H, et al. 2023. Classified surface rupture characteristics and damage analysis of the 2022 MS6.9 Menyuan earthquake, Qinghai. Acta Seismologica Sinica (in Chinese), 45(5): 797-813.
Talo M, Baloglu U B, Yıldırım Ö, et al. 2019. Application of deep transfer learning for automated brain abnormality classification using MR images. Cognitive Systems Research, 54: 176-188, doi: 10.1016/j.cogsys.2018.12.007.
Vaswani A, Shazeer N, Parmar N, et al. 2017. Attention is all you need. //Proceedings of the 31st International Conference on Neural Information Processing Systems. Long Beach, California, USA: Curran Associates Inc., 6000-6010.
Wang T T, Bian Y J, Zhang Y X, et al. 2023. Classification of earthquakes, explosions and mining-induced earthquakes based on XGBoost algorithm. Computers & Geosciences, 170: 105242, doi: 10.1016/j.cageo.2022.105242.
Weiss K, Khoshgoftaar T M, Wang D D. 2016. A survey of transfer learning. Journal of Big Data, 3(1): 9, doi: 10.1186/s40537-016-0043-6.
Wu Y M, Liang W T, Mittal H, et al. 2016. Performance of a low-cost earthquake early warning system (P-Alert) during the 2016 ML6.4 meinong (Taiwan) earthquake. Seismological Research Letters, 87(5): 1050-1059, doi: 10.1785/0220160058.
Yang H F, Wang D, Guo R M, et al. 2022. Rapid report of the 8 January 2022 MS6.9 Menyuan earthquake, Qinghai, China. Earthquake Research Advances, 2(1): 100113, doi: 10.1016/j.eqrea.2022.100113.
Zhang H C, Jin X, Wei Y X, et al. 2016. An earthquake early warning system in Fujian, China. Bulletin of the Seismological Society of America, 106(2): 755-765, doi: 10.1785/0120150143.
Zhang H C, Melgar D, Sahakian V, et al. 2022. Learning source, path and site effects: CNN-based on-site intensity prediction for earthquake early warning. Geophysical Journal International, 231(3): 2186-2204, doi: 10.1093/gji/ggac325.
Zheng Z, Lin B H, Yu W H, et al. 2024. Predicting instrumental seismic intensity using deep learning and physical features. Chinese Journal of Geophysics (in Chinese), 67 (7): 2712-2728, doi: 10.6038/cjg2023R0521.
Zhu J, Fang L H, Miao F J, et al. 2023a. Deep learning and transfer learning of earthquake and quarry-blast discrimination: applications to southern California and eastern Kentucky. Geophysical Journal International, 236(2): 979-993, doi: 10.1093/gji/ggad463.
Zhu J, Li Z F, Fang L H. 2023b. USTC-Pickers: a Unified Set of seismic phase pickers Transfer learned for China. Earthquake Science, 36(2): 95-112, doi: 10.1016/j.eqs.2023.03.001.
Zhu J B, Li S Y, Ma Q, et al. 2022a. Support vector machine-based rapid magnitude estimation using transfer learning for the Sichuan-Yunnan Region, China. Bulletin of the Seismological Society of America, 112(2): 894-904, doi: 10.1785/0120210232.
Zhu J B, Li S Y, Song J D. 2022b. Hybrid deep-learning network for rapid on-site peak ground velocity prediction. IEEE Transactions on Geoscience and Remote Sensing, 60: 5925712, doi: 10.1109/TGRS.2022.3230829.
Zhu J B, Li S Y, Wei Y X, et al. 2023. On-site instrumental seismic intensity prediction for China via recurrent neural network and transfer learning. Journal of Asian Earth Sciences, 248: 105610, doi: 10.1016/j.jseaes.2023.105610.
Zhu W Q, Beroza G C. 2019. PhaseNet: A deep-neural-network-based seismic arrival-time picking method. Geophysical Journal International, 216(1): 261-273, doi: 10.1093/gji/ggy423.
林彬华, 金星, 康兰池等. 2021. 基于卷积神经网络的地震震级测定研究. 地球物理学报, 64(10): 3600-3611, doi: 10.6038/cjg2021O0370.
马强, 金星, 李山有等. 2013. 用于地震预警的P波震相到时自动拾取. 地球物理学报, 56(7): 2313-2321, doi: 10.6038/cjg20130718.
潘章容, 周扬, 苗在鹏等. 2023. 2022年1月8日青海门源6.9级地震仪器烈度与宏观烈度对比分析. 高原地震, 35(3): 8-15.
宋晋东, 余聪, 李山有. 2021. 地震预警现地PGV连续预测的最小二乘支持向量机模型. 地球物理学报, 64(2): 555-568, doi: 10.6038/cjg2021O0193.
宋晋东, 朱景宝, 韦永祥等. 2023. 2022年1月8日青海门源6.9级地震机器学习地震预警震级估计与现地阈值报警的回溯验证. 地球物理学报, 66(7): 2903-2919, doi: 10.6038/cjg2022Q0050.
苏瑞欢, 袁道阳, 谢虹等. 2023. 2022年青海门源MS6.9地震地表破裂特征分类及震害分析. 地震学报, 45(5): 797-813.
郑周, 林彬华, 于伟恒等. 2024. 利用深度学习与物理特征预测仪器地震烈度. 地球物理学报, 67(7): 2712-2728, doi: 10.6038/cjg2023R0521.
-
图(15)
表(3)
计量
- 文章访问数:
- PDF下载数:
- 施引文献: 0