基于多台输入的图注意力网络震级估计研究  

Magnitude estimation of graph attention networks based on multi-station inputs

作　　者：余仲黎 朱景宝 李山有[1,2] 宋晋东 YU Zhongli;ZHU Jingbao;LI Shanyou;SONG Jindong(Key Laboratory of Earthquake Engineering and Engineering Vibration,Institute of Engineering Mechanics,China Earthquake Administration,Harbin 150080,China;Key Laboratory of Earthquake Disaster Mitigation,Ministry of Emergency Management,Harbin 150080,China)

机构地区：[1]中国地震局工程力学研究所地震工程与工程振动重点实验室,黑龙江哈尔滨150080 [2]地震灾害防治应急管理部重点实验室,黑龙江哈尔滨150080

出　　处：《地震工程与工程振动》2025年第2期22-32,共11页Earthquake Engineering and Engineering Dynamics

基　　金：中国地震局工程力学研究所基本科研业务费专项(2024C05);国家自然科学基金项目(42304075,51408564)。

摘　　要：震级估计是地震预警中的重要任务之一。准确的震级估计对于地震影响范围的快速判断和地震预警信息的及时发布至关重要。现有的方法通常基于单个台站的加速度时程提取特征信息进行震级估计,再通过多台平均的方法得到结果。文中利用多台输入的图注意力网络算法构建端到端的震级估计模型(GAT_M),作为GAT_M模型输入的是首台P波触发后3 s内的多台站地震加速度时程。本研究利用日本防灾科学技术研究所K-NET强震观测台网的强震资料进行模型训练和测试实验。研究结果表明:在首台P波触发后3 s,震级估计的平均误差和标准差分别为-0.077和0.40,R2为0.72。本研究还分析了震级、时间窗和台站数量对GAT_M模型性能的影响。同时,在首台P波触发后3 s,与传统Pd方法相比,GAT_M模型有更小的震级估计误差,在复杂样本数据的情况下,GAT_M模型有较大的优势且能够更好地应用于地震预警震级估计中。Earthquake magnitude estimation is one of the important tasks in earthquake early warning.Accurate earthquake magnitude estimation is critical to quick judgment of earthquake influence areas and timely release of earthquake warning information.Existing methods usually extract the characteristic information based on the acceleration time history of a single station to estimate the magnitude,and then obtain the result by the multi-station averaging method.In this paper,an end-to-end magnitude estimation model(GAT_M)is constructed using a multi-input graph attention network algorithm.The time history of multi-station seismic acceleration within 3 s after the first P-wave is triggered is input into the GAT_M model.The multi-station seismic acceleration waveforms within 3 s after the first P-wave are used as the input of the GAT_M model.In this study,the strong earthquake data from of the K-NET strong earthquake observation network of Japan Institute of Disaster Prevention Science and Technology were used for model training and test experiments.Within 3 s after the first P-wave triggers,the mean error and standard deviation of magnitude estimation are-0.077 and 0.40 respectively,and R 2 is 0.72.The effects of magnitude,time window and number of stations on the performance of GAT_M model are also analyzed.Simultaneously,within 3 s after the initial P-wave triggers,the GAT_M model demonstrates a reduced magnitude estimation error compared to the traditional Pd method.In the case of complex sample data,the GAT_M model has a greater advantage and can be better applied to magnitude estimation.

关 键 词：图注意力网络 地震监测预警 震级 多台输入 

分 类 号：P315.3[天文地球—地震学] P315.7[天文地球—固体地球物理学] P315.92[天文地球—地球物理学]