| In recent years,hydraulic fracturing technology has been widely used in the development of oil and natural gas fields.Hydraulic fracturing has become one of the necessary measures for the exploitation of unconventional oil and gas with huge reserves.Deploying geophones on the ground or in the borehole to receive the microseismic signals from rock ruptures,microseismic monitoring is an effective way to evaluate the result of hydraulic fracturing by studying the source location and mechanism inversion of the recorded microseismic events.One of the basic objectives of microseismic monitoring is to obtain the source location of microseismic events based on source location methods in seismology.With the advantage of noise suppression,the migration stacking location methods based on waveform information exhibit the good performance of noise suppression and have been widely used in surface microseismic monitoring that has a relatively lower signal-to-noise ratio(SNR).By directly stacking waveforms according to the travel time curve of underground imaging points,the diffraction stacking(DS)location method has high calculation efficiency and great potential in microseismic real-time monitoring.Rock fractures during hydraulic fracturing contain a large number of shear components.For microseismic events caused by shear ruptures,the surface records may have polarity reversal due to the impact of their source mechanisms.In this situation,for the DS location method,directly stacking waveforms cannot achieve the focused source image at the real focal position,but obtain the symmetric distribution centered on the true source position at the source origin time.We call this symmetric distribution the quasi-radiation pattern because it is similar to the source radiation pattern.Therefore,in practical applications,the DS location method will have the problem of inaccurate location when the surface records have polarity reversal.To solve the above-mentioned problem,the thesis first proposes a location method named diffraction stacking interferometric imaging(DSII),in which the spatial interferometric imaging condition is applied to correct the quasi-radiation pattern.The spatial interferometric imaging algorithm at each time step includes the following steps:(1)For a given spatial interferometric radius,extracting a 3D cube of DS image centered on an arbitrary imaging point;(2)Within the extracted 3D cube,multiplying imaging values of two points symmetrical about the center point;(3)Summing productions of all centrosymmetric point pairs,and taking the absolute value of the summation as the interferometric imaging value of the center points;(4)Repeating steps(1)-(3)for all imaging points to obtain the interferometric source image.The interferometric imaging condition can focus the quasi-radiation pattern from the DS image onto its center,that is,the real source position.Therefore,the proposed DSII method can accurately locate microseismic events.Although the spatial interferometric imaging method can accurately locate microseismic events by correcting the DS source image,the interferometric source images have artifacts but do not perfectly focus on the true source The thesis proposes a microseismic location method based on deep learning image recognition,in which a 4D diffraction stacking source imaging is taken as the input and a modified 3D U-Net is applied to fit the input into a Gaussian distribution centered on the true source.The quasi-radiation pattern in the DS source image is mainly dependent on the focal mechanism and has good consistency in the different work areas.So,we can use the synthetic data for network training and apply the trained network to the prediction of field data.The prediction achieves a better-focused source image than that from the DSII because the network fits the output into the Gaussian distribution.The velocity model has an impact on the absolute location from the migrationbased location method.The accurate velocity inversion requires a high SNR of the microseismic data.Therefore,we propose a new idea of microseismic data denoising based on DS source imaging:Due to the strong coherence of seismic signal,the DS source imaging focuses the signal energy as a quasi-radiation pattern around the true source and disperses the noise energy to the underground imaging space;The signal waveform can be recovered by using demigration from the energy of the quasi radiation pattern back to the surface receivers.The proposed DSII location method can not only accurately locate microseismic events,but also retain a good performance on noise suppression from the traditional DS method,and it has high computational efficiency as well.Besides,the DSII method is also suitable for complex velocity models and has good performances in cases of velocity perturbation or sparse observation.The proposed location method based on deep learning further optimizes source images.The trained network has a good generalization and can be applied to different velocity models and observation systems.The DS source imaging before network input provides the noise suppression of this method.During the prediction,the output source image is sensitive to the characteristic of the quasiradiation pattern.Thus,we can obtain a well-focused source image for an event with low SNR,only if the diffraction stacking source image has an obvious quasi-radiation pattern.The finally proposed denoising method for microseismic data can not only effectively remove random noise interference and regular linear interference,but also complete the missing traces(such as the eliminated bad traces).Recovered waveforms of microseismic signals lay a foundation for the subsequent velocity model inversion. |
