基于机器学习和台阵相关性的水力压裂微地震事件自动识别及到时拾取

利用密集台阵对水力压裂微地震进行监测将有助于优化储层压裂、揭示断层活化.为满足密集台阵海量采集数据的处理需求, 本文建立了一种综合运用多种机器学习方法和台阵相关性的、无需人工干预的自动处理流程, 从而能够快速得到高质量的密集台阵震相到时目录.该综合策略包括: (1)利用迁移学习在连续波形中快速检测地震事件; (2)利用U型神经网络PhaseNet自动拾取P波、S波震相; (3)利用三重线性剔除法, 结合密集台阵到时相关性剔除异常到时数据和地震事件; (4)利用K-means和SVM两类机器学习算法, 进一步区分发震时刻接近的多个地震事件, 减小事件漏拾率.通过将该流程应用于四川盆地长宁—昭通页岩气开发区微地震监测数据, 并将自动处理结果与人工拾取结果进行比对发现, 二者在震级测定、定位以及走时成像结果等方面具有很好的一致性, 表明本文处理流程结果精度可达到手动处理精度.本文结果为密集台阵地震监测数据的高效、高精度处理提供了新思路.

Automatic phase detection and arrival picking for microseismic events in hydraulic fracturing based on machine learning and array correlation

Using dense array to monitor microseismicity can help optimize hydraulic fracturing and reveal fault activations. To meet the demand of processing a larger amount of data acquired by dense seismic arrays and obtain high-quality phase picks efficiently, an automatic processing strategy that integrates multiple machine learning methods and array correlation is proposed in this paper. This new automatic processing strategy is carried out without human intervention, which includes the following steps: (1) rapid event detection in the continuous seismic waveforms via transfer learning; (2) applying the U-shaped neural network PhaseNet to automatically pick the P- and S-phases; (3) using a newly developed triple linear elimination method to remove abnormal picks and events; (4) separating multiple adjacent seismic events to avoid missing events. By applying this strategy to the microseismic monitoring dataset acquired at the Changning—Zhaotong shale gas play in the Sichuan Basin, and comparing the results with those through manual picking, we found that the magnitudes, locations and tomographic results from these two approaches agree well. This study provides a new measure to process large quantities of seismic data acquired by dense arrays efficiently and accurately.