Estimation of DAS microseismic source mechanisms using unsupervised deep learning

Abstract

Distributed acoustic sensing (DAS) is an increasingly prevalent technology for seismic acquisition. DAS fibers posses many beneficial properties that make them an attractive candidate for microseismic monitoring, where DAS has experienced an important role. A key task in microseismic monitoring is the estimation of the source mechanism that generated the seismic data which provide the information about fracture type, in-situ stress, and whether fractures were generated during hydraulic fracturing or if existing fractures were reactivated. Information of this type is crucial for understanding the success of a hydraulic fracture treatment and for optimizing future treatments. Source mechanism information is encoded in the direct arrivals of the seismic energy emanating from propagating fractures. The source mechanism information is typically extracted from the direct arrivals through a process known as moment tensor inversion. However, because DAS is sensitive to a different portion of the wavefield than geophones or seismometers, moment tensor inversion does not directly transfer. Bearing this in mind, we design a deep learning workflow for extracting source mechanism information from DAS data. The workflow begins with a convolutional auto-encoder which compresses DAS-microseismic data to only those features relevant to moment tensor characterization. To extract the source mechanism information these features contain we further process them using two distinct methods; clustering and a generative adversarial network. Clustering is successful in grouping images with similar source mechanism allowing for inference about the source mechanism type of a given DAS-microseismic event based on the cluster to which it belongs. The generative adversarial network is trained to learn the relationship between the features extracted by the convolutional autoencoder and moment tensor labels. Fully trained, it is able to predict the moment tensor associated with a given DAS-microseismic event.