基于生成对抗网络的微地震数据去噪方法  

A Generative Adversarial Network-based Method for Microseismic Data denoising

作　　者：郑晶[1,2] 吴志祥 李德伟[2] 邢立文 ZHENG Jing;WU Zhixiang;LI Dewei;XING Liwen(State Key Laboratory of Coal Resources and Safe Mining,China University of Mining and Technology(Beijing),Beijing 100083,China;College of Geoscience and Surveying Engineering,China University of Mining and Technology(Beijing),Beijing 100083,China)

机构地区：[1]中国矿业大学(北京)煤炭资源与安全开采国家重点实验室,北京100083 [2]中国矿业大学(北京)地球科学与测绘工程学院,北京100083

出　　处：《实验室研究与探索》2021年第5期18-21,共4页Research and Exploration In Laboratory

基　　金：中央高校基本科研业务费专项资金资助(2021JCCXDC02);中国矿业大学煤炭资源与安全开采大学生科技创新计划项目(SKLCRSM20DC01)。

摘　　要：由于微地震事件的振幅通常较小,信噪比低是微地震数据处理,尤其是地面微地震数据处理的最大挑战之一。基于生成对抗网络提出了一种新的降噪网络模型,包括生成器和判别器。构建生成器来生成处理后的数据,网络训练完成后重建去噪后的数据;判别器负责区分真实数据和虚假数据。在单通道时间序列-波形上进行操作,并使用合成数据集对模型进行端到端训练。最后使用实际微地震数据测试该方法,并与其他传统降噪方法进行性能对比。实验表明,该模型能表现出优良的降噪性能。Microseismic event is usually of small magnitude.The biggest challenge of microseismic data processing especially data collected by surface monitoring equipment is the low signal-to-noise ratio.Traditional denoising techniques used in surface microseismic data processing usually operate on the time/frequency or time-frequency domain.The majority of them usually have trouble reconstructing a satisfactory signal at strong noise levels.We develop a new denoising model based on a generative adversarial network(GAN).The network includes two parts:a generator G and a discriminator D.G is built to generate processed data.After training,G will reconstruct the clean data.D is responsible to distinguish between real and fake data.We directly operate at single channel time series,the waveform level,training the model end-to-end using synthetic dataset.The performance of the method is evaluated using synthetic and real microseismic data examples and compared with other traditional denoising methods.The experiments show that the proposed model achieves impressive denoising performance even when the signal is drawn in noise.After denoising,the arrival times of the events are clearer which can help event pickers for later process.

关 键 词：微地震 生成对抗网络 去噪 生成器 判别器 

分 类 号：P315.63[天文地球—地震学]