| Popis: |
Log and core data has been collected in the mature supergiant fields of the Middle East over many decades with variable data quality and vintages. These data could be affected by artifacts created by borehole conditions, different logging tools, water- or oil-based mud systems, and different processing parameters. The data are frequently lacking traceability to borehole/experiment conditions required to apply appropriate and consistent corrections for further modeling workflows. As a result, petrophysicists must perform substantial log, core quality control (QC) and editing prior to interpreting logs or using core data. The purpose of this paper is to discuss machine-learning (ML) applications grouped in the solution to automatically QC and process core and log data for hundreds of wells. The goal was to detect poor data and outliers to apply corrections that mimic human-led interpretation for at least 80 to 90% of the data processed. For several data sets and cases, ML was successfully used to various aspects of log editing. (Akkurt et al. 2018; Liang et al. 2019; Mawlod et al. 2019; Singh et al. 2020). However, as a result of our 2020 study, we discovered numerous lessons that are addressed in this work: Geological and geospatial information are potentially as valuable as well data and cannot be overlooked by any ML algorithm working with well data at the field scale.ML algorithms are more efficient than human-led data processing and can produce more accurate and consistent outcomes. However, the parameter selection and QC of outcomes require an expert assessment. Thus, ML-based applications cannot be "black boxes" and must include a user-interaction toolbox for efficient workflow control.To deliver a more efficient solution in terms of performance over large data sets, the full capacity of cloud-based technologies should be utilized, thereby enabling the parallelization of operations while simultaneously constructing and analyzing multiple solution scenarios. In the current work, we developed an ML application to QC core and well log data to perform log editing (artifact correction) and missing log prediction using other available data. The ML application is guided by petrophysical domain expertise and advanced data-driven algorithms to perform complex data homogenization and predictions for hundreds of wells using a cloud-based environment. The application may communicate with a wellbore platform project and read data from separate files in the log ASCII standard (LAS) and digital log interchange standard (DLIS) formats. The methodology is as follows: Data liberation from wellbore platform projectIntegration of geospatial and geological data with available core and log data in a single data frameOutlier detection of petrophysical logs and core dataFeatures selection for log editingClustering–prediction–validation of training data sets using a variable number of clusters to select the best model for a targeted variable predictionResults transfer to a wellbore platform project The proposed ML application integrates geological and geospatial information to provide a quality homogenized data set to be used in rock-typing, permeability, and saturation modeling. The ML application will significantly save time and effort to eliminate repetitive human tasks. The cloud-based implementation, when combined with existing petrophysical platforms, enables the highest performance and data exchange between the different software platforms. Interactive and user-friendly dashboards will provide the geoscientists with complete control over each step of the ML data-driven workflow. |
