Deep learning assisted well log inversion for fracture identification

Abstract

Manual fracture identification methods based on cores and image logging pseudopictures are limited by the expense and the amount of data. In this paper, we propose an integrated workflow, which takes the fracture identification as an end-to-end project, to combine the boundary detection and the deep learning classification to recognize fractured zones with accurate locations and reasonable thickness. We first apply the discrete wavelet transform algorithm and a boundary detection method named changing point detection to enhance the fracture sensibility of acoustic logs and segment the whole logging interval into non-overlapping subsections by estimating boundaries. The deep neural network based auto-encoders and the convolutional neural network classifier are then implemented to extract the hidden information from logs and categorize the subsections as the fractured or non-fractured zones. To validate the feasibility of this workflow, we apply it to the logging data from a real well. Compare with the benchmarks provided by the support vector machine , random forest and Adaboost model, the one-dimensional well profile predicted by the proposed changing point detection-deep learning classifier is more consistent with the manual identification result.

From Fluid Flow to Coupled Processes in Fractured Rock: Recent Advances and New Frontiers

“…(2018) developed a stress‐based tomography approach that combines geomechanical modeling of quasi‐static stress variations caused by fractures with an inversion algorithm to reconstruct probable DFN geometries from stress variability profiles obtained from borehole logs. In addition, progress has been made on inversion methods, mainly through the combined inversion of different borehole log data sets (Tian et al., 2021) or by using some borehole data sets to better constrain the inversion of other types of cross‐borehole data sets (Wu et al., 2019).…”

From fluid flow to coupled processes in fractured rock: recent advances and new frontiers

“…Recently, Afshari Moein et al ( 2018) developed a stress-based tomography approach that combines geomechanical modeling of quasi-static stress variations caused by fractures with an inversion algorithm to reconstruct probable DFN geometries from stress variability profiles obtained from borehole logs. In addition, progress has been made on inversion methods, mainly through the combined inversion of different borehole log data sets (Tian et al, 2021) or by using some borehole data sets to better constrain the inversion of other types of cross-borehole data sets (Wu et al, 2019).…”

Automating hyperparameter optimization in geophysics with Optuna: A comparative study

“…Alongside this, the network architecture and its layers will be shown. However, hyperparameter optimization (HPO), a crucial part of this analysis, is often overlooked (Kaur, Pham, et al., 2020; Lähivaara et al., 2019; Kaur, Fomel, et al., 2020; Colombo et al., 2021; Um et al., 2022) or at best, only briefly discussed (Côrte et al., 2020; Tian et al., 2020; Aleardi et al., 2022). This pertains to the justification of why a certain network architecture, depth, learning rate or other hyperparameters were chosen.…”