Deep Neural Networks for Detection and Location of Microseismic Events and Velocity Model Inversion from Microseismic Data Acquired by Distributed Acoustic Sensing Array

Abstract

Fiber-optic cables have recently gained popularity for use as Distributed Acoustic Sensing (DAS) arrays for borehole microseismic monitoring due to their physical robustness as well as high spatial and temporal resolutions. As a result, the sensors record large amounts of data, making it very difficult to process in real-/semi-real-time using the conventional processing routines. We present a novel approach, based on deep learning, for handling the large amounts of DAS data in real-/semi-real-time. The proposed neural network was trained on synthetic microseismic data contaminated with real-ambient noise from field data and was validated using field DAS microseismic data obtained from a hydraulic fracturing operation. The results indicate that the trained network is capable of detecting and locating microseismic events from DAS data and simultaneously update the velocity model to a high degree of precision. The mean absolute errors in the event locations and the velocity model parameters are 2.04, 0.72, 2.76, 4.19 and 0.97 percent for distance (x), depth (z), P-wave velocity, S-wave velocity and density, respectively. In addition to automation and computational efficiency, deep learning reduces human expert data handling during processing, thus preserving data integrity leading to more accurate and reproducible results.

Microseismic Monitoring and Analysis Using Cutting-Edge Technology: A Key Enabler for Reservoir Characterization

“…Due to their computational efficiency, the models can be used in the field to process the data in real-time during its acquisition, thereby scaling down the amount of data to be stored while providing necessary information that could help optimize the field operations. Huot and Biondi [50], Wamriew et al [51], and Huot et al [52] emphasized that without the complete automation of microseismic data processing, large volumes of collected data could be wasted due to human processing limitations.…”

“…In this study, we demonstrate the use of two cutting-edge technologies: distributed acoustic sensing (DAS) and deep learning for microseismic monitoring and analysis. Building on the work by Wamriew et al [51], we investigated the possibility of improved microseismic event detectability and location: (i) given a well-known velocity model; (ii) using different neural network architectures; and (iii) reducing the number of output parameters. In addition, the output detections by the neural networks were verified using the conventional STA/LTA method.…”

“…This makes generating reflection images from the data recorded in this geometry much more difficult than from the data recorded with sources and receivers at the same elevation. For our purposes, however, there is no need for high-level processing techniques since a neural network is capable of learning the properties of the seismic waveforms by itself to a high level of precision [51]. To refine the wavefield of the DAS data and to simplify the task of searching for seismic events in the data, spectral processing of the data can be considered necessary and sufficient [56].…”

“…A single 1-D anisotropic velocity model with three layers was used in the forward model. The Swave velocity was taken from the FORGE velocity estimates by Zhang and Pankaw [61] and Wamriew et al [51], while the P-wave velocities and the densities were estimated using the Castagna [62] and Gardner [63] equations, respectively. Relatively high Thomsen anisotropic parameters [64] of ϵ = 0.51, γ = 0.36, and σ = 0.25, were chosen since previous studies have revealed that a neural network, trained on high anisotropic parameters, would generalize well when presented with waveforms from lower anisotropic models [48].…”

Earthquake Distribution and Crustal Seismic Wave Velocity Based on Machine Learning

“…The output layer is the last layer in the network. According to the specific problem, the category can be output in the classification problem or a certain value can be output in the regression problem [18]. Typically, the output layer of the classification problem is preceded by Softmax activation functions that convert probabilities into categories.…”