微地震数据去噪方法综述

随着常规油气藏资源不断枯竭，非常规油气藏的勘探开发已逐渐成为一种必然趋势，从而使得微地震监测技术快速发展。微地震事件的发生持续时间短、声波频率高，使得实际采集到的微地震数据信噪比较低。本文首先简要介绍了微地震监测技术能够在非常规油气藏开发中保证高效的增产，以及微地震噪声压制在微地震监测技术数据处理流程中是关键一步，直接影响着后续微地震研究的准确性和可靠性；并对地面微地震监测数据中的噪声源进行分析，归纳了地面微地震监测中常见噪声：强脉冲干扰、工业交流电干扰、钻井干扰、声波干扰、规则干扰等，分析了其各自的基本特点。然后，概述了地面微地震数据去噪方法已取得的成果，按频率、传播方向、空间分布区域等特性进行分类，分析各种去噪方法在实际应用过程中针对的噪声类型，以及在去噪过程中对有效信号造成的影响等。最后，基于深度学习具有更强的复杂函数表征能力，分析了3种典型深度学习模型的结构及特点；结合在其他相关领域数据去噪问题的成功应用，深度学习可以解决目前微地震数据噪声压制存在的问题，可以作为微地震数据去噪的一种新方法；考虑到目前微地震数据样本数量可能影响深度学习在微地震监测中大规模的应用，本文提出用生成式对抗网络来构建微地震数据样本库以解决该问题，并将其用于后续深度学习过程中的模型训练。

Review of Microseismic Data Denoising Methods

With the continuous depletion of conventional oil and gas resources, the exploration and development of unconventional oil and gas reservoirs have gradually become an inevitable trend, which makes micro-seismic monitoring technology develop rapidly. The duration of microseismic events is relatively short,and the frequency of sound waves is relatively high, which result in a low signal-to-noise ratio of the microseismic data collected by actual acquisition. Firstly, the authors briefly introduce in this paper the possibility of guaranteeing high productivity in unconventional oil and gas development by microseismic monitoring technology, and the importance of microseismic noise suppression in data processing, as it directly affects the accuracy and reliability of subsequent microseismic research. Then, the noise sources and several common noise types in the ground microseismic monitoring data are summarized,such as strong pulse interference, 50 Hz industrial alternating current interference, drilling interference, acoustic interference, regular interference, and their basic characteristics are analyzed. The results of denoising methods in the ground microseismic data are classified according to the characteristics of frequency, direction of propagation,and spatial distribution area,also the noise types and the influences of various denoising methods on effective signals are discussed. Finally, based on the stronger representation abilities of deep learning complex functions, the structure and characteristics of three typical deep learning models are analyzed. In combination with the successful application of data denoising in other related fields,deep learning can solve the problem of noise suppression of current microseismic data, and can be used as a new method for microseismic data denoising. Considering that the current number of microseismic data samples may affect the large-scale application of deep learning in microseismic monitoring, the authors propose to build a microseismic data sample library by generating countermeasure network to solve this problem,and which can be used in model training in the subsequent deep learning processe.