Global Database Fuels Machine-Learning Model Predictions

During the previous decade, the oil and fuel trade, each onshore and offshore, has and continues to transition the way in which subsurface knowledge is used to develop and constrain predictive fashions. The basis of any superior analytical effort utilized in any trade is the standard, accessibility, and protection the information set can present to the person making an attempt to coach predictive modeling. Whether the method is hierarchical clustering, random forest, multidimensional engines, or more-advanced multilinear regression evaluation, it should start with extra knowns than the person is making an attempt to outline within the unknowns or predictions.

RFDbase Functions
GeoMark’s RFDbase is a quantity of data and uncooked knowledge from each main petroleum basin on this planet that can be utilized to gasoline a search engine or coaching engine to drive predictions (Fig. 1). It does so by linking the mechanisms representing geology, geochemistry, petrophysics, and engineering elements/inputs to data-science efforts. Covering the wanted end-member variables for each rock and fluid properties, the RFDbase incorporates laboratory-based knowledge that can be utilized in evaluation trying to hyperlink subsurface traits to wellhead efficiency. The geoscience inputs could be constrained whereas the person constrains the manufacturing data (e.g., lateral size and completion measurement) to coach the mannequin.

Fig. 1—Map of RFDbase international protection.

Credit: GeoMark.

Examples
Taking a simplistic method to how RFDbase entry can assist efforts, the schematic in Fig. 2 can be utilized to reference elements of a predictive mannequin to gasoline, drive, after which improve a motorcar.

GeoMark_Fig2.jpg

Fig. 2—Schematic of the evolution of a data-driven mannequin, from uncooked knowledge to the applying of features and algorithms predicting the correlation of properties to the ultimate resolution modeled with uncertainty.

Credit: GeoMark.

Ultimately, the hassle is making an attempt to affiliate identified properties to wellhead efficiency given a number of elements. There could also be a number of x, y, and z dimension values that correlate; nonetheless, ensuring that the information in context has an essential function in guiding the developed mannequin is crucial. Otherwise, one would possibly discover that blueberries and peanut butter could be correlated from a sure perform/regression or engine, however the query must be requested, “should they be?” Fig. 3 presents a correlation within the knowledge that could possibly be match a minimal of three alternative ways.

GeoMark_Fig3.jpg

Fig. 3—Example of two parameters (x and y) that current a number of methods to suit a correlation to extrapolate away from a identified constraint.

Credit: GeoMark.

Which is essentially the most correct? Which gives the perfect method to predicting the minimal, most, mode, distribution, and common of the populations? These correlations can discover extra sturdy methods of making use of all of the statistical options to the best-fit method by enhanced machine studying and knowledge science.

Data Types
GeoMark’s RFDbase incorporates the next crucial items of data. Each knowledge sort is adopted by an inventory of doable methods the information can be utilized and what enter or property to constrain within the modeling efforts.

  • Source rock complete natural carbon and programmed pyrolysis
    • Source rock presence and enrichment (high quality), working petroleum system, a qualitative proxy for the presence of risky fluids, thermal most/maturity, and fluid mobility
  • Fluid properties
    • Includes oil, fuel, and water
    • API, estimated fuel/oil ratio, water salinity and composition (midstream or services functions), fuel compositions, maturity, cost and migration, hydrocarbon high quality, and fluid mobility
  • Rock/petrophysical properties
    • Includes Fourier-transform infrared spectroscopy, X-ray fluorescence, X-ray diffraction, matrix density, bulk density, and complete porosity/storage
    • How a lot fluid-filled storage exists, lithology/mineralogical change, geomechanical modifications associated to brittleness from lithology understanding, chemostratigraphy (upwelling environments, and anoxic vs. dysoxic settings)
  • Pressure/quantity/temperature
    • Specific gravity of fluids, fuel/oil ratio prediction, formation quantity issue constraint, fluid-phase envelope constraint (early fuel breakthrough potential, part change predictions, and drawdown habits modifications

LEAVE A REPLY

Please enter your comment!
Please enter your name here