Machine studying for predicting stochastic fluid and mineral volumes in complicated unconventional reservoirs

Fig. 2. Cross-plot demonstrates strong correlation of modeled kerogen to statistical kerogen for blind wells.


A machine studying workflow can shortly, and precisely, predict mineralogy, porosity and saturation in a number of wells to raised perceive productive layers in unconventional oil reservoirs.

Fred Jenson, Chiranjith Ranganathan, Shi Xiuping, Ted Holden, CGG

Determination of mineralogy is a vital step within the petrophysical evaluation of many kinds of reservoirs. Changes in volumes of minerals point out adjustments in geological deposition, diagenesis, reservoir high quality and brittleness.

Particularly in shale performs, success is determined by deciding on leases, based mostly on identification of hydrocarbons-in-place inside probably productive layers. Potentially productive layers in a shale play are outlined as layers which are sufficiently brittle to answer hydraulic fracture remedies. A layer’s brittleness is set predominately by its relative mineral ratios, with arenites recognized as extra brittle, whereas clay and most calcites are recognized as ductile. This methodology of predicting % brittleness makes use of the ratio of brittle minerals to whole mineral quantity. This technique requires understanding the quantity of clay, arenites, calcite, kerogen and heavy minerals, akin to pyrite, which are current.

Precise dedication of porosity and saturation can be wanted for correct hydrocarbons-in-place calculations. In unconventional reservoirs, the presence of much less frequent minerals, akin to kerogen and pyrite, makes it tougher to calculate mineral and fluid volumes in these complicated lithologies. Mineral volumes and porosity in all these rocks are sometimes decided utilizing core information or estimated with stochastic strategies often known as statistical mineral evaluation, which works fairly properly when utilizing an honest set of wireline logs. Difficulty arises when older boreholes are current within the space of curiosity, and the log curve information required for classy interpretations are usually not obtainable.


The following oil shale case research goals to exhibit how machine studying fashions—aided by Python programming extensions utilizing the machine studying ecosystem of CGG GeoSoftware’s PowerLog petrophysical interpretation expertise—can shortly produce high-quality properly log interpretations, together with mineral volumes in complicated lithologies, for numerous wells. In this context, for correct estimates of rock mineral volumes, we suggest an efficient technique, utilizing a supervised random forest regressor and PowerLog StatMin. This was utilized, utilizing instruments obtainable within the PowerLog machine studying ecosystem and information saved in a database.

The coaching datasets use StatMin, a stochastic evaluation module in PowerLog, to find out volumes of fluids and minerals, utilizing statistical strategies for petrophysical evaluation of inauspicious formations. Porosity, water saturation and complicated mineral volumes are precious outputs typically obtained from StatMin. These are the petrophysical parameters wanted to coach the mannequin.

Fig. 1. Representative show of the enter curves used within the statistical modeling and regressive evaluation.

Methodology. The technique begins by taking a couple of trendy wells with good curve protection and making a stable statistical mannequin utilizing statistical mineral evaluation strategies, akin to StatMin on this case. These wells are then used as management wells to create the machine studying mannequin to foretell the mineral and fluid volumes for different boreholes which have a extra restricted set of curve information. Porosity and saturation values are encoded within the resolved fluid volumes. This workflow additionally might apply with core evaluation to create the recognized mineral distributions. The technique was validated utilizing seven wells from the oil shale discipline:

  • The supervised machine studying mannequin was educated with solely three of the obtainable wells, whereas the remaining 4 wells have been used as blind take a look at wells for establishing the mannequin’s accuracy.
  • To set up a reference, we carried out stochastic modeling on all seven wells, utilizing StatMin. Curves used as function inputs to the regressor have been gamma ray (GR), bulk density (RHOB), neutron porosity (NPHI), deep induction (ILD), and photoelectric seize cross part (PEF).

This article addresses the flexibility to precisely predict StatMin outcomes, utilizing our machine studying technique, even when a photoelectric impact (PE) measurement isn’t obtainable for all predicted wells. Curves used to coach the machine studying regressor are GR, RHOB, NPHI, and ILD, measurements frequent to wells drilled after 1980. Modeled minerals are quartz (sand), calcite (lime), illite (clay), pyrite, and kerogen. Bulk quantity water and bulk quantity hydrocarbons are additionally estimated.

Building the stochastic mannequin. To examine outcomes from the machine studying mannequin with outcomes from the stochastic mannequin in blind wells, the StatMin mannequin was computed for all seven wells within the research, though solely three wells have been concerned in creating the machine studying mannequin. The remaining 4 wells not utilized in coaching comprised the blind properly group. Existing log information for the seven wells are of top of the range. Figure 1 reveals curve information for one of many mannequin wells.

Fig. 2. Cross-plot demonstrates strong correlation of modeled kerogen to statistical kerogen for blind wells.

Fig. 2. Cross-plot demonstrates robust correlation of modeled kerogen to statistical kerogen for blind wells.

Results from statistical computation of mineral and fluid volumes from StatMin have been utilized in creating the machine studying predictive fashions. The StatMin course of makes use of error minimization ahead modeling from recognized log responses to minerals and fluids to generate volumes of chosen minerals and fluids. A predominant matrix of parameters is used within the evaluation of the seven oil shale wells, each the mannequin and the blind take a look at boreholes. The “U” curve within the matrix is the photoelectric adsorption coefficient, calculated from the PE curve, and is a vital consider figuring out mineralogy in statistical mineralogy. This PE curve is a contemporary curve and infrequently not obtainable when analyzing older boreholes. It is essential to notice that the machine studying fashions have been constructed with out utilizing the PE curve. The machine studying technique permits analysis of many older wells not possible in StatMin, as a consequence of lack of PE measurement.

Machine studying evaluation. The subsequent step within the course of used the three chosen wells to coach the supervised regressor and predict the minerals and volumes on the remaining wells. We educated the mannequin by deciding on impartial function vectors together with density, gamma ray, deep resistivity, and neutron curves for predicting every mineral sort and fluid quantity output from StatMin. Predicted volumes from every technique have been fairly comparable and really near volumes and fluids within the management wells. A cross-plot compares modeled kerogen to statistical kerogen quantity for the 4 blind wells, Fig. 2.

Fig. 3. Stochastic modeling results compared with machine learning results on a blind test well.

Fig. 3. Stochastic modeling outcomes in contrast with machine studying outcomes on a blind take a look at properly.

Cross-correlation outcomes between modeled mineral and fluid volumes have been pretty much as good as, or higher than, the match between statistical kerogen and machine studying kerogen. Kerogen was chosen as the instance, as it’s a vital mineral when evaluating unconventional reservoirs and is without doubt one of the lower-volume minerals, the place any prediction errors could be magnified. Models for every of the minerals and fluids have been generated and utilized to the blind take a look at wells. The show in Fig. 3 compares StatMin outcomes with the expected outcomes. Visual examination of the log plot demonstrates stable settlement between outcomes from PowerLog StatMin and machine learning-generated outcomes.


It is a typical scenario when evaluating a longtime discipline to have extensively disparate information sorts for every properly. Often a couple of wells within the discipline are trendy and have subtle interpretation, offering a high-quality volumetric evaluation of fluids and minerals. Occasionally, electron seize spectroscopy instruments can be found that precisely measure elemental concentrations and supply stable mineral volumes. Extensive core information are additionally precious in analyzing complicated formations to raised perceive lithological composition.

This case research demonstrates the worth of utilizing high-technology interpretations to construct machine studying fashions with obtainable high-quality log information. We have demonstrated that machine studying fashions can precisely predict mineralogy, porosity and saturation in unconventional reservoirs. This brings clearer understanding of productive intervals, enabling operators to maximise field-wide potential.  


  1. Pendrel, J. and A.I. Marini, “Static models for unconventional reservoirs: A Barnett shale case study,” SEG Summer Workshop Abs., San Diego, California, 2014.
  2. Varga, R., R. Lotti, A. Pachos, T. Holden, I. Marini, E. Spadafora and J. Pendrel, “Seismic inversion in the Barnett shale successfully pinpoints sweet spots to optimize wellbore placement and reduce drilling risks,” SEG Technical Program expanded abstracts: 1-5, 2012.

Fred Jenson has over 40 years of expertise within the oil and gasoline business. His profession has included wireline discipline engineer, gross sales engineer, petrophysical guide, and software program growth administration. Mr. Jenson has labored on discipline research in a lot of the main basins worldwide and has intensive expertise with built-in interpretation strategies. He has spent the final ten years with Jason/CGG GeoSoftware engaged on PowerLog growth and advertising and marketing. Over the final a number of years, he has been working with the PowerLog Ecosystem exploring machine studying, deep studying and different Python-based workflows to reinforce petrophysical interpretations.

Chiranjith Ranganathan is a software program architect with CGG GeoSoftware chargeable for design, growth and enhancements to the machine studying ecosystem of PowerLog. During his 13-year tenure with CGG, he has labored on all kinds of software program issues together with a number of machine studying workflows and purposes to deal with petrophysical challenges. He holds a grasp’s diploma in laptop science from the University of Houston Main Campus.

Shi Xiuping is a petrophysicist with CGG GeoSoftware, based mostly within the China area with accountability for consulting providers in addition to software program gross sales and help. With over ten years of petrophysical and rock physics expertise throughout greater than 50 standard and unconventional reservoir initiatives within the China area, she makes a speciality of petrophysics and rock physics in complicated geological situations, particular logging analysis (picture logging, dipole shear wave logging), geomechanics, and machine studying.

Ted Holden is regional technical supervisor for CGG GeoSoftware in North America and Latin America. Mr. Holden serves as a senior technical advisor to CGG employees and purchasers of CGG GeoSoftware, mentoring advisors and consumer employees in seismic petrophysics, rock physics, seismic inversion and quantitative interpretation. During his 45-year profession, he has suggested consumer corporations on quite a few seismic reservoir initiatives across the globe targeted on exploration and growth of each onshore and offshore, standard and unconventional reservoirs.


Please enter your comment!
Please enter your name here