Temporal-Spatial Graph Neural Network Model for Onsite Peak Ground Acceleration Prediction
-
摘要: 为了探索和建立适用于我国的现地峰值加速度（PGA）预测模型，以及提高现地PGA预测的可靠性，本研究提出了一种基于时空图神经网络的现地PGA预测模型（TSGNN-PGA），并采用中国强震数据对TSGNN-PGA模型进行训练和测试。测试结果表明：P波触发后3s，和Pd-PGA方法相比，TSGNN-PGA模型对于PGA预测有更小的平均绝对误差（MAE）和标准差（STD），以及更大的决定系数，且分别为0.205、0.261和0.688；同时，和Pd-PGA方法相比，在不同的震中距、震级和信噪比范围下，TSGNN-PGA模型对于PGA预测有更小的MAE和STD，这意味着TSGNN-PGA模型对于震中距、震级和信噪比的敏感性更弱，且受影响更小。此外，在漾濞6.4级地震、芦山6.1级地震和积石山6.2级地震中，与Pd-PGA方法相比，TSGNN-PGA模型对于PGA预测表现出更强的鲁棒性。可以推断，TSGNN-PGA模型在一定程度上可以提高我国现地PGA预测的可靠性，且对于地震预警有着重要意义。Abstract: In order to explore and establish the onsite peak ground acceleration (PGA) prediction model suitable for China and improve the reliability of onsite PGA prediction, this paper proposes an onsite PGA prediction model based on Temporal-Spatial Graph Neural Network (TSGNN-PGA), and uses the Chinese strong-ground motion records to train and test the TSGNN-PGA model. The test results show that within 3s after P-wave triggering, compared with the P d -PGA method, the TSGNN-PGA model has smaller mean absolute error (MAE) and standard deviation (STD), as well as larger coefficient of determination for PGA prediction, with values of 0.205, 0.261, and 0.688, respectively; Meanwhile, compared with the P d -PGA method, the TSGNN-PGA model has smaller MAE and STD for PGA prediction within different ranges of epicentral distance, magnitude, and signal-to-noise ratio (SNR). This means that the TSGNN-PGA model is less sensitive to epicentral distance, magnitude, and SNR, and is less affected; In addition, compared with the P d -PGA method, the TSGNN-PGA model shows stronger robustness for PGA prediction in the Yangbi M6.4, Lushan M6.1 and Jishishan M6.2 earthquakes. It can be inferred that the TSGNN-PGA model can improve the reliability of onsite PGA prediction in China to a certain extent, and has important significance for earthquake early warning.1)
1 2https://github.com/Jingbaozhu1996/TSGNN-PGA -
图 1 训练集和测试集的震中和台站分布
Figure 1. Distribution of epicenters and stations in the training and test sets
表 1 TSGNN-PGA模型和Pd-PGA方法预测PGA的MAE、STD和R2对比
Table 1. TSGNN-PGA model and Pd-PGA method for MAE, STD, and R2 of PGA prediction
方法 MAE STD R2 Pd-PGA 0.299 0.355 0.370 TSGNN-PGA 0.205 0.261 0.688 
下载: 导出CSV
表 2 不同震中距下TSGNN-PGA模型和Pd-PGA方法预测PGA的MAE和STD对比
Table 2. TSGNN-PGA model and Pd-PGA method for MAE and STD of PGA prediction within different epicentral distance
方法 震中距≤100 km 震中距>100 km MAE STD MAE STD Pd-PGA 0.284 0.175 0.347 0.266 TSGNN-PGA 0.205 0.117 0.206 0.109
下载: 导出CSV
表 3 不同震级范围下TSGNN-PGA模型和Pd-PGA方法预测PGA的MAE和STD对比
Table 3. TSGNN-PGA model and Pd-PGA method for MAE and STD of PGA prediction within different magnitude
方法 震级≤6 震级>6 MAE STD MAE STD Pd-PGA 0.300 0.354 0.292 0.379 TSGNN-PGA 0.202 0.256 0.305 0.308
下载: 导出CSV
表 4 不同信噪比范围下，TSGNN-PGA模型和Pd-PGA方法预测PGA的MAE和STD对比
Table 4. TSGNN-PGA model and Pd-PGA method for MAE and STD of PGA prediction within different SNR
方法 信噪比≤10 信噪比>10 MAE STD MAE STD Pd-PGA 0.302 0.332 0.294 0.354 TSGNN-PGA 0.201 0.257 0.217 0.268
下载: 导出CSV
表 5 3次破坏性地震的预测PGA的MAE、STD和R2对比
Table 5. MAE, STD, and R2 of PGA prediction for these three destructive earthquakes using TSGNN-PGA model and Pd-PGA method
震例 方法 MAE STD R2 漾濞地震 Pd-PGA 0.354 0.323 0.552 TSGNN-PGA 0.251 0.253 0.755 芦山地震 Pd-PGA 0.350 0.371 −0.118 TSGNN-PGA 0.208 0.245 0.596 积石山地震 Pd-PGA 0.402 0.478 0.430 TSGNN-PGA 0.293 0.295 0.712
下载: 导出CSV
表 6 3次破坏性地震在不同实测PGA范围下的整体绝对误差占比
Table 6. The overall absolute error ratio for these three destructive earthquakes for different observed PGA ranges
方法 PGA≤45.7 Gal PGA>45.7 Gal AE<0.4 AE≥0.4 AE<0.4 AE≥0.4 Pd-PGA 59.47% 40.53% 55.24% 44.76% TSGNN-PGA 81.47% 18.53% 57.69% 42.31%
下载: 导出CSV
-
陈苏, 丁毅, 孙浩等, 2023. 物理驱动深度学习波动数值模拟方法及应用. 力学学报, 55(1): 272−282.Chen S., Ding Y., Sun H., et al., 2023. Methods and applications of physical information deep learning in wave numerical simulation. Chinese Journal of Theoretical and Applied Mechanics, 55(1): 272−282. (in Chinese) 代友林, 刘洋, 梁远玲等, 2022. 芦山6.1级地震与泸县6.0级地震人员伤亡对比分析. 震灾防御技术, 17(4): 674−681.Dai Y. L., Liu Y., Liang Y. L., et al., 2022. Comparative analysis of casualties between Lushan M6.1 earthquake and Luxian M6.0 earthquake. Technology for Earthquake Disaster Prevention, 17(4): 674−681. (in Chinese) 李亮, 李山有, 纪忠华等, 2018. 仪器烈度计算方法研究. 震灾防御技术, 13(4): 801−809.Li L., Li S. Y., Ji Z. H., et al., 2018. Study of the computational method of instrumental seismic intensity. Technology for Earthquake Disaster Prevention, 13(4): 801−809. (in Chinese) 李山有, 金星, 马强等, 2004. 地震预警系统与智能应急控制系统研究. 世界地震工程, 20(4): 21−26.Li S. Y., Jin X., Ma Q., et al., 2004. Study on earthquake early warning system and intellegent emergency controling system. World Earthquake Engineering, 20(4): 21−26. (in Chinese) 李小军, 孙晓燕, 荣棉水等, 2023. 云南漾濞6.4级地震房屋震害特征受地震动方向性和场地条件的影响. 应用基础与工程科学学报, 31(3): 650−662. doi: 10.16058/j.issn.1005-0930.2023.03.010Li X. J., Sun X. Y., Rong M. S., et al., 2023. Characteristics of building seismic damages in MS6.4 Yangbi earthquake affected by ground motion directivity and site condition. Journal of Basic Science and Engineering, 31(3): 650−662. (in Chinese) doi: 10.16058/j.issn.1005-0930.2023.03.010 刘辰, 李小军, 景冰冰等, 2019. 地震预警PGV-Pd关系参数的距离分段特征. 地球物理学报, 62(4): 1413−1426.Liu C., Li X. J., Jing B. B., et al., 2019. The distance segmentation characters of PGV-Pd relationship parameters for earthquake early warning. Chinese Journal of Geophysics, 62(4): 1413−1426. (in Chinese) 刘平, 倪晓霞, 2025. 基于知识元与贝叶斯网络的地震次生地质灾害情景演化分析. 震灾防御技术, 20(2): 254−267.Liu P., Ni X. X., 2025. Evolutionary analysis of earthquake secondary geological disasters scenario based on knowledge element and Bayesian network. Technology for Earthquake Disaster Prevention, 20(2): 254−267. (in Chinese) 马强, 2008. 地震预警技术研究及应用. 哈尔滨: 中国地震局工程力学研究所.Ma Q., 2008. Study and application on earthquake early warning. Harbin: Institute of Engineering Mechanics, China Earthquake Administration. (in Chinese) 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.Ma Q., Jin X., Li S. Y., et al., 2013. Automatic P-arrival detection for earthquake early warning. Chinese Journal of Geophysics, 56(7): 2313−2321. (in Chinese) 彭朝勇, 杨建思, 薛兵等, 2013. 基于汶川主震及余震的预警参数与震级相关性研究. 地球物理学报, 56(10): 3404−3415.Peng C. Y., Yang J. S., Xue B., et al., 2013. Research on correlation between early-warning parameters and magnitude for the Wenchuan Earthquake and its aftershocks. Chinese Journal of Geophysics, 56(10): 3404−3415. (in Chinese) 宋晋东, 2013. 高速铁路运行控制用地震动参数及单台地震预警技术研究. 哈尔滨: 中国地震局工程力学研究所.Song J. D., 2013. Research on seismic ground motion indices for operation control and single station earthquake early warning applied for high-speed railway. Harbin: Institute of Engineering Mechanics, China Earthquake Administration. (in Chinese) 宋晋东, 教聪聪, 李山有等, 2018. 基于地震P波双参数阈值的高速铁路Ⅰ级地震警报预测方法. 中国铁道科学, 39(1): 138−144.Song J. D., Jiao C. C., Li S. Y., et al., 2018. Prediction method of first-level earthquake warning for high speed railway based on two-parameter threshold of seismic P-wave. China Railway Science, 39(1): 138−144. (in Chinese) 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.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, 64(2): 555−568. (in Chinese) 苏闻浩, 刘启方, 2024. 基于深度神经网络的地表地震动幅值预测研究. 震灾防御技术, 19(2): 387−396.Su W. H., Liu Q. F., 2024. Study on the prediction of ground motion amplitude based on deep neural network. Technology for Earthquake Disaster Prevention, 19(2): 387−396. (in Chinese) 王运生, 赵波, 吉锋等, 2024. 2023年甘肃积石山MS 6.2级地震震害异常的启示. 成都理工大学学报(自然科学版), 51(1): 1−8.Wang Y. S., Zhao B., Ji F., et al., 2024. Preliminary insights into the hazards triggered by the 2023 Jishishan MS 6.2 earthquake in Gansu province. Journal of Chengdu University of Technology (Science & Technology Edition), 51(1): 1−8. (in Chinese) 张红才, 2013. 地震预警系统关键技术研究. 哈尔滨: 中国地震局工程力学研究所.Zhang H. C., 2013. Study of key technologies in earthquake early warning system. Harbin: Institute of Engineering Mechanics, China Earthquake Administration. (in Chinese) Allen R. M., Stogaitis M., 2022. Global growth of earthquake early warning. Science, 375(6582): 717−718. doi: 10.1126/science.abl5435 Chen S., Long Z. Y., Luan S. K., et al., 2025. Physical-guided coupling neural network approach for seismic wave propagation. Earthquake Engineering and Resilience, 4(2): 167−177. doi: 10.1002/eer2.70005 Ioffe S. , Szegedy C. , 2015. Batch normalization: accelerating deep network training by reducing internal covariate shift. In: Proceedings of the 32nd International Conference on International Conference on Machine Learning. Lille: JMLR. org, 448−456. Jozinović D., Lomax A., Štajduhar I., et al., 2020. Rapid prediction of earthquake ground shaking intensity using raw waveform data and a convolutional neural network. Geophysical Journal International, 222(2): 1379−1389. doi: 10.1093/gji/ggaa233 Kipf T. N. , Welling M. , 2017. Semi-supervised classification with graph convolutional networks. In: 5th International Conference on Learning Representations. Toulon: OpenReview. net. Kiranyaz S. , Ince T. , Hamila R. , et al. , 2015. Convolutional neural networks for patient-specific ECG classification. In: 2015 37th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC). Milan: IEEE, 2608−2611. Murphy S., Nielsen S., 2009. Estimating earthquake magnitude with early arrivals: a test using dynamic and kinematic models. Bulletin of the Seismological Society of America, 99(1): 1−23. doi: 10.1785/0120070246 Nagi J. , Ducatelle F. , Di Caro G. A. , et al. , 2011. Max-pooling convolutional neural networks for vision-based hand gesture recognition. In: 2011 IEEE International Conference on Signal and Image Processing Applications (ICSIPA). Kuala Lumpur: IEEE, 342−347. Peng C. Y., Yang J. S., Chen Y., et al., 2015. Application of a threshold-based earthquake early warning method to the MW 6.6 Lushan Earthquake, Sichuan, China. Seismological Research Letters, 86(3): 841−847. doi: 10.1785/0220140053 Peng C. Y., Ma Q., Jiang P., et al., 2020. Performance of a hybrid demonstration earthquake early warning system in the Sichuan–Yunnan border region. Seismological Research Letters, 91(2A): 835−846. doi: 10.1785/0220190101 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 Srivastava N., Hinton G., Krizhevsky A., et al., 2014. Dropout: a simple way to prevent neural networks from overfitting. The Journal of Machine Learning Research, 15(1): 1929−1958. Wang C. Y., Huang T. C., Wu Y. M., 2022. Using LSTM neural networks for onsite earthquake early warning. Seismological Research Letters, 93(2A): 814−826. doi: 10.1785/0220210197 Wen W. P., Xu T. F., Hu J., et al., 2025. Seismic damage recognition of structural and non-structural components based on convolutional neural networks. Journal of Building Engineering, 102: 112012. doi: 10.1016/j.jobe.2025.112012 Wu Y. M., Kanamori H., 2005. Rapid assessment of damage potential of earthquakes in Taiwan from the beginning of P waves. Bulletin of the Seismological Society of America, 95(3): 1181−1185. doi: 10.1785/0120040193 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 Zhu J. B., Li S. Y., Song J. D., 2022. Magnitude estimation for earthquake early warning with multiple parameter inputs and a support vector machine. Seismological Research Letters, 93(1): 126−136. doi: 10.1785/0220210144 -
点击查看大图
图(7) / 表(6)
计量
- 文章访问数: 33
- HTML全文浏览量: 5
- PDF下载量: 2
- 被引次数: 0
