Usage
  • 3 views
  • 2 downloads

Machine learning and computer vision applications for hydraulic fracture monitoring

  • Author / Creator
    Nustes Andrade, Jorge
  • The spatiotemporal distribution of hydraulic fracturing-induced microseismicity is complicated and depends on various mechanical and diffusional parameters. Hydraulic fracture modeling can aid in understanding fluid-induced microseismicity. Nevertheless, the interaction of several physical processes occurring within and around the fracture adds complexity in developing real-time models for microseismic prediction. This study introduces three methodologies to forecast the microseismic cloud size, which engineers can use to improve the treatment's effectiveness during pumping. We compare a random forest model trained with statistical features derived continuously from the injection monitoring data, a physics-based approach based on diffusivity estimates from the microseismic observations, and a convolutional neural network (CNN) trained in real-time with the engineering curves to predict the future microseismic cloud size. The prediction accuracy of all methods varies depending on the microseismic behavior. We postulate that predictive models could be improved by including more physics into the input data. Distributed acoustic sensing (DAS) is increasingly used in hydraulic fracturing operations. The low-frequency band of DAS (LFDAS) contains high-resolution information of the far-field strain perturbations that can be used to constrain fracture geometry. Nevertheless, locating the time and depth intersection of a propagating fracture with an offset monitoring well in the same pad (frac-hit) is mainly made using simple cumulative strain maps, which can be subjective and inefficient. We introduce a computer vision workflow based on image matching to automate the detection of frac-hits in LFDAS data. Furthermore, we developed a method to remove the frac-hit strain signal using affine image transformations and warping. The workflow is applied to a real LFDAS dataset from Western Canada, and results exceed the detection performance achieved with cumulative strain maps.

  • Subjects / Keywords
  • Graduation date
    Fall 2021
  • Type of Item
    Thesis
  • Degree
    Master of Science
  • DOI
    https://doi.org/10.7939/r3-90xj-r378
  • License
    This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.