基于生成对抗网络和卷积神经网络的高速铁路地震预警干扰信号识别方法  

Interference Signal Identification Method of Earthquake Early Warning for High-Speed Railway Based on Generative Adversarial Network and Convolutional Neural Network

作　　者：宋晋东[1,2] 栾世成 李山有 马强[1,2] 孙文韬 刘赫奕[1,2] 周学影 姚鹍鹏 黄鹏杰 朱景宝 SONG Jindong;LUAN Shicheng;LI Shanyou;MA Qiang;SUN Wentao;LIU Heyi;ZHOU Xueying;YAO Kunpeng;HUANG Pengjie;ZHU Jingbao(Key Laboratory of Earthquake Engineering and Engineering Vibration,Institute of Engineering Mechanics,China Earthquake Administration,Harbin Heilongjiang 150080,China;Key Laboratory of Earthquake Disaster Mitigation,Ministry of Emergency Management,Harbin Heilongjiang 150080,China;Railway Science and Technology Research and Development Center,China Academy of Railway Sciences Corporation Limited,Beijing 100081,China;Ministry of Security Products,HeNan Splendor Science&Technology Co.,Ltd.,Zhengzhou Henan 450012,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 [3]中国铁道科学研究院集团有限公司铁道科学技术研究发展中心,北京100081 [4]河南辉煌科技股份有限公司安防产品部,河南郑州450012

出　　处：《中国铁道科学》2025年第1期225-232,共8页China Railway Science

基　　金：国家重点研发计划项目(2023YFF0725005);国家自然科学基金资助项目(42304074,51408564);中国国家铁路集团有限公司科技研究开发计划课题(K2024G008);中国铁道科学研究院集团有限公司院基金课题(2022YJ149)。

摘　　要：为提升高速铁路地震预警系统中地震事件识别的可靠性,提出基于生成对抗网络(GAN)和卷积神经网络(CNN)的高速铁路地震预警干扰信号识别方法。首先,通过GAN对打夯干扰信号进行数据增强,以实现数据平衡;其次,设计并构建GAN-CNN打夯干扰信号识别模型,并对其进行训练和测试;最后,通过对比试验,验证该模型在干扰信号识别中的有效性和准确性。结果表明:与未使用GAN进行数据增强的情况相比,所提方法识别打夯干扰信号和地震事件信号的准确率分别为99.60%和100%,性能显著提升;此外,GANCNN模型的交并比、准确率、召回率和综合能力评价指标也得到提高。该方法可为高速铁路地震预警干扰信号识别提供参考。In order to improve the reliability of earthquake event recognition in the earthquake early warning system of high-speed railway,a method for identifying interference signals of earthquake early warning for high-speed railway based on generative adversarial network(GAN)and convolutional neural network(CNN)was proposed.Firstly,the data of tamping interference signals was enhanced by GAN to achieve data balance.Secondly,a GAN-CNN tamping interference signal recognition model was designed and constructed,and it was trained and tested.Finally,the effectiveness and accuracy of the model in interference signal recognition were verified by comparative experiments.The results show that compared with the case where data is not enhanced by GAN,the accuracy of the proposed method in identifying tamping interference signals and earthquake event signals is 99.60%and 100%respectively,with significant improvement of performance.In addition,the evaluation indicators of the GAN-CNN model such as intersection-over-union ratio,accuracy,recall rate and comprehensive ability are also improved.This method can provide an important reference for the interference signals identification of earthquake early warning for high-speed railway.

关 键 词：地震预警 高速铁路 卷积神经网络 生成对抗网络 打夯干扰信号 

分 类 号：P315.3[天文地球—地震学]