| 引用本文: | 朱景宝,宋晋东,李山有.地震烈度仪对卷积神经网络模型地震预警震级估计的影响[J].哈尔滨工业大学学报,2024,56(6):81.DOI:10.11918/202306043 |
| ZHU Jingbao,SONG Jindong,LI Shanyou.Influence of low-cost sensors on earthquake early warning magnitude estimation using convolutional neural network model[J].Journal of Harbin Institute of Technology,2024,56(6):81.DOI:10.11918/202306043 |
|
| 摘要: |
| 为探索地震烈度仪对卷积神经网络模型地震预警震级估计的影响,以2022年中国发生的5次破坏性地震（MS≥5.8）为例,将地震数据应用到卷积神经网络震级估计模型中,分析引入烈度仪地震数据后的震级估计。结果表明:在P波到达后3 s,烈度仪和强震仪的单台震级估计误差主要分布在±1震级单位范围内；对于震中距100 km以内的数据,在P波到达后10 s内,与强震仪相比,烈度仪震级估计误差均值更接近0；对于信噪比小于20的数据,强震仪震级估计误差均值比烈度仪的震级估计误差均值更接近0,且烈度仪有更大震级估计误差的不确定度。此外,对于这5次地震,与强震仪相比,烈度仪的数量更多、分布更密,引入烈度仪地震数据后,卷积神经网络模型更快地获得鲁棒的震级估计。研究结果为地震烈度仪在卷积神经网络震级估计模型中的适用性提供了依据,也为地震预警系统震级估计提供参考。 |
| 关键词: 地震预警 卷积神经网络 震级估计 地震烈度仪 破坏性地震 |
| DOI：10.11918/202306043 |
| 分类号:P315 |
| 文献标识码:A |
| 基金项目:国家自然科学基金(U9,4,51408564)；黑龙江省自然科学基金(LH2021E119)；中国铁道科学研究院集团有限公司科研项目(2022YJ149)；地震科技星火计划项目(XH23027YB)；国家重点研发计划(2018YFC1504003) |
|
| Influence of low-cost sensors on earthquake early warning magnitude estimation using convolutional neural network model |
|
ZHU Jingbao1,2,SONG Jindong1,2,LI Shanyou1,2
|
|
(1.Key Laboratory of Earthquake Engineering and Engineering Vibration (Institute of Engineering Mechanics, China Earthquake Administration), Harbin 150080, China; 2. Key Laboratory of Earthquake Disaster Mitigation, Ministry of Emergency Management, Harbin 150080, China)
|
| Abstract: |
| To explore the impact of low-cost sensors on the earthquake early warning (EEW) magnitude estimation of convolutional neural network (CNN) model, taking five destructive earthquakes (MS≥5.8) that occurred in China in 2022 as examples, seismic data was applied to the CNN model, and the magnitude estimation results after incorporating the data recorded by low-cost sensors were analyzed. The results show that within 3 s after the P-wave arrival, based on single station, the magnitude estimation error of the low-cost sensors and the strong-motion instruments is mainly distributed in the range of ±1 magnitude unit. For the seismic records with epicentral distance less than 100 km, within 10 s after the P wave arrival, the mean value of magnitude estimation error of the low-cost sensor is closer to 0 than that of the strong-motion instrument. For the seismic records with signal noise ratio less than 20, the mean value of magnitude estimation error of strong-motion instrument is closer to 0 than that of low-cost sensor, and the low-cost sensor has greater uncertainty of magnitude estimation error. Additionally, for these 5 earthquakes, compared with the strong-motion instrument, the low-cost sensor has a larger quantity and denser distribution. the CNN model obtains robust magnitude estimation faster when considering the data recorded by low-cost sensors. The results provide a basis for the applicability of low-cost sensors in CNN magnitude estimation models, and serve as a reference for magnitude estimation in EEW systems. |
| Key words: earthquake early warning convolutional neural network magnitude estimation low-cost sensors destructive earthquakes |
