07122-23

Goal
The goal of this project is to develop a Self-Teaching Expert System (SETES) software application for analyzing, predicting, and optimizing natural gas production from extremely tight reservoirs. The project will result in a prototype alpha version of the software tool, ready for testing and evaluation by project collaborators and any interested party.

Performers
Lawrence Berkeley National Laboratory, Berkeley, CA 94720
Texas A&M, College Station, TX 77843
University of Houston, Houston, TX 77204
Anadarko Petroleum Corp., Woodlands, TX 77380
BGI Resources, LLC, Oakland, CA 94606

Background
In a 2007 study, the Potential Gas Committee estimated the technically recoverable, unconventional gas resource in the US to be 293 Tcf. Most of this gas is located in very tight reservoirs that are geologically complex and poorly understood. As a result, development of this resource carries a high degree of uncertainty and economic risk. It is difficult to predict reservoir flow properties from petrophysical surveys, and, once a well is drilled, it is usually necessary to apply costly well stimulation strategies in order to establish economically viable gas flow rates. The proposed project aims to reduce the uncertainties associated with tight gas exploration and production and to aid in the design of well completion strategies that optimize production and recovery from tight gas wells.

The envisioned SETES software tool will integrate information from multiple and diverse sources on a continuous basis and use this information for timely decision making while explicitly addressing uncertainty and risk. The software will incorporate geological, geophysical, fracturing, reservoir, and production data from existing tight gas wells, and it will make recommendations as to well location, orientation, design, and completion strategies for optimizing gas production and recovery while mitigating risk. The SETES will be available to any interested party either as a web application or a computer program that (a) is centrally located on LBNL servers, and is distributed over the web, (b) can use both “public” and “private” databases, (c) protects the confidentiality of data in the “public” database, (d) is used as a web application with only “public” databases, (e) can be easily installed at the user’s facilities, and executed as a computer program (in which case both “private” databases can be invoked, in addition to the “public” ones already included the SETES, (f) continuously updates the databases and refines the underlying decision-making metrics and process (baseline mode), (g) enables the design of appropriate production systems, the operation and management of unconventional (tight) gas resources (UGR), and estimates uncertainties (prediction mode), and (h) allows history matching and parameter identification from UGR data (optimization or history-matching mode).

Note that “public” databases are defined as those that have been made available to SETES by their owners, and the data of which are used by the expert system for self-teaching and expanding the internal knowledge database; although some of the data in these databases may be protected (e.g., exact geographical location, or names of a particular reservoir), the information related to these data forms the basis of the expert system, and is made available to all SETES users. The unprotected data (and the corresponding implications) in public databases are available to all users. “Private” databases are formed by users who are not willing to share their data, and who choose to use SETES by expanding its internal databases to include their own data that do not become part of the public data base.

The project is led by Lawrence Berkeley National Lab (LBNL) and carried out by a team of scientists and engineers from LBNL, Texas A&M, and University of Houston. Tight gas reservoir data, including production, pressure, well completion and stimulation information, and petrophysical data from logs and cores, will be provided by Anadarko Petroleum Corporation and BGI. Anadarko will provide data from tight gas wells in Texas, and BGI will provide data from tight gas wells in Oklahoma. Deliverables for the project will include an alpha release of the SETES software tool and a final report summarizing all technical aspects of the project.

Potential Impacts
Successful development and utilization of the SETES software tool is expected to reduce uncertainties associated with exploration and production of tight gas resources. This should lead to growth in unconventional gas development activity, because it will provide operators with the information needed to make informed decisions on where and how to produce gas from unconventional reservoirs. The resulting increase in development activity is expected to bring previously inaccessible energy resources to market. Increased domestic gas production would result in increased tax revenues, royalties, and regional economic benefits. Improved recovery for individual wells has the added benefit of decreasing the environmental footprint of a field development program.

Accomplishments
Because the various tasks in the project are not the sole responsibility of any single participant, but they represent the collective contributions of the various teams, progress will be discussed (a) by task, and (b) by specific topics (which provide a better description of the efforts and accomplishments in addressing the scientific issues, and of the teams that focus on them). In addition, a Feasibility Assessment has been submitted to RPSEA (January 2010) documenting our accomplishments and defining the path forward.

Progress by Tasks:

• Task 1.0: Project Management Plan
  This task was completed by 2/5/2009.
• Task 2.0: Technology Status Assessment
  The corresponding technology assessment report was completed on 2/9/2009.
• Task 3.0: Technology Transfer
  This task is on schedule. Three posters were presented at three scientific conferences (TOUGH Symposium 2009, SEG 2009, SPE ATCE 2009). We presented two papers at two conferences (TOUGH Symposium 2009, SPE ATCE 2009), and submitted two paper proposals (accepted) at two more conferences (SPE 2010 Western Regional Conference, and the SPE 2010 International Oil & Gas Conference & Exhibition in China). Two papers are already in review for publication in the SPE Journal and Transport in Porous Media; two more papers are in preparation for submission to SPE Journal and Transport in Porous Media.
• Task 4.0: Development of a Data Abstraction, Reduction and/or Compression (DARC) Technology
  This task was completed on schedule, in the sense of developing effective DARC technology approaches (based on currently available techniques) for particular subsets of the data used by SETES and determining the limits of this approach (i.e., cases and conditions where data compression and reduction using current techniques cannot be achieved unless resources are invested in this fundamental problem – an activity that would involve basic research and is beyond the scope of this project). A DARC framework was developed using geostatistical methods and other analyses discussed in Sections II.1 and II.3.
• Task 5.0: Define Functional Capabilities and Requirements of the SETES Product
  The definition of the functional requirements was completed 6/09/2009. Two different sets of requirements were identified and described: A basic set that defines the capabilities of the first version of the web-based SETES that will be released at the end of this project, in addition to a final (i.e., fuller and more complete) set of capabilities that can be developed if additional resources are made available after the conclusion of this project. In essence, the final version includes all the desirable features of SETES if there were no time and/or funding limitations.
• Task 6.0: Review and Analysis of the “Kernel” Complete Data Set
  This task was completed. The analysis of a subset of (a) geophysical and (b) production data from a superset including data from an entire tight gas field owned by Anadarko (the industrial partner of this project) that helped identify and select the kernel is discussed in Section II.1.
• Task 7.0: Develop Preliminary Decision-Making and Soft-Computing Methodology
  This task is in progress. Although Fuzzy Logic and Neural Networks appear to offer the most promise in addressing the training- and self-teaching aspect, we have not yet arrived at an unequivocal conclusion as to the superiority of one vs. the other. Thus, both techniques are being employed and evaluated in our study, with the expectation that, as increasingly complex issues are addressed, one of the methods may emerge as the clear winner. Thus, the decision as to which to select as the sole method in SETES has not been reached, and it is possible that both will be included in SETES addressing different aspects of the functions of the system.
• Task 8.0: Partial Validation of Approach and Identification of Additional Data and Refinement Requirements
  This task involves the use of SETES and the kernel data set to predict the behavior of the reservoir and wells representing the remaining (excluding the kernel) data set, and is still in progress.

Progress by Scientific Issue:

• 1. Production Data Analysis for Holly Branch Wells
  • 1.1 Work on Developing Basic Knowledge to Describe Production Performance in Variably Fractured Tight Gas Systems
    We have developed high-definition (500,000 to 1,000,000 gridblocks), 3D numerical models to describe the flow and behavior of gas in tight systems with variable fracture (natural and induced) characteristics, flow properties, and sorption characteristics. This study involved representation of individual fractures in addition to the cumulative behavior and performance of different types of secondary fractures, and is the first such study conducted in tight gas systems. The results are included in a paper presented at the SPE 2009 ATCE (currently in review for publication in SPE Journal), and show production curves that follow a distinct and recognizable pattern that appears characteristic of tight gas systems. Comparison to field data that were made available by our industrial partners indicate a very good fit between predictions and production, both in terms of pattern of the production curves and of the corresponding values.

    A challenge that remains is the determination of DARC technology approach to describe the production curves in terms of relatively simple parametric equations that involve a few parameters that can be highly correlated to the properties and conditions of the producing fractured tight-gas system. Such a DARC technology is highly desirable within the SETES because it will greatly reduce the computational effort and the input requirements for the estimation of production predictions by relying on simple equations with few input requirements (instead of having to solve the equations of a numerical model that includes up to 1,000,000 gridblocks). Although it has been possible to accurately represent the numerically-obtained production curves of particular systems over the entire time-frame of interest (ranging from minutes to years) with relatively simple equations, to-date we have been unable to determine parameters of these equations that correlate well with changes in the properties of the system. In other words, we have been able to develop parametric equations that are accurate for particular cases, but are not sufficiently general. Work is in progress to develop parametric equations that are closely related to analytical approximations of the problem of flow through such fractured systems because such equations have the best chance for a good fit. Alternatively, we are beginning to explore another approach involving the numerical solution of the problem by employing a single grid that is to be scaled internally within SETES (with minimal need for inputs by the user) according to the particular properties of the system under investigation.

  • 1.2 Work on Identification of the Kernel Data Set
    We have initially performed production data analysis using simple rate-time relations for 37 wells in the Holly Branch field to estimate the ultimate recovery. For this work we have applied the Arps' hyperbolic and the power-law exponential rate decline relations. Once fit with the data, we extrapolate the rate-time relations to a specified rate or time limit to obtain an estimate of ultimate recovery (EUR). The EUR estimates of these wells range between 0.12 BSCF and 6.83 BSCF.

    Based on the similar production profiles and estimated ultimate recoveries, we have identified 12 wells (of the original 37) in the Holly Branch field which we are using as the kernel data set. Ongoing work involves performing model based production data analysis with the inclusion of pressure data to confirm/validate the results obtained from the simple rate-time analysis. This analysis will also be used to obtain reservoir and well properties that cannot be obtained from simple rate-time analysis.

  • 1.3 Fast Optimization Methods for Parameter Estimation
    In the “Optimization Mode”, SETES is expected to provide parameter estimates when provided with well test data. The solution of this inverse problem includes a standard optimization technique that is well established. We are in the process of developing a fast optimization code that is based on the history-matching production codes of LBNL codes that are used for the solution of industrial-strength problems (such as the determination of fracture properties at the Yucca Mountain site of the proposed High-Level Nuclear Waste Repository).
  • 1.4 Uncertainty Analysis and Quantification
    Mathematical methods for determining the cumulative effects of parametric uncertainty in the estimation and behavior of the various components of the flow, geophysical and geomechanical problems associated with production from tight gas systems have been developed. Implementation and testing of these methods into the DARC framework and into the overall workflow of the SETES are under way.
• 2. Geomechanical Analysis
  Best practices in well stimulation, appropriate methods and techniques, and the underlying logic and justification, are being systematically analyzed. Hierarchical catalogues and a decision flow-chart (relating the flow of information and field observations and conditions to the well stimulation method and the specifics of its application) are being developed. Additionally, stress field analysis is being conducted, and appropriate methods are investigated to relate the stress distribution (a larger-scale/regional condition) to well observations and well stimulation decisions.
• 3. Geophysical Analysis and Development of Geophysical Tools
  • 3.1 2D Tests
    Some tests of the 2D tools developed using synthetic data (presented in previous reports) have been performed using the Knowles section from the Anadarko data set. These tools include several interpolation and simulation techniques: pseudo-wells generation tool, full 2D conditional simulation tool, 2D co-kriging tool, low frequency estimation+high frequency conditional simulation tool, and an image-guided conditional simulation tool (indicator approach). Seismic attributes such as acoustic impedance between these three vertical wells are not yet at disposal and they would be of great help to improve and constrain these interpolations/simulations. This set of tools has to be still extensively tested using other real data sets, such as one available from LBL to monitor injection that has both kinds of information at disposal. We have been exploring the open source program SEISLAB (version 3.0) that provides a series of functions for manipulation of well logs and seismic attributes constructed under MATLAB. This program can identify and discriminate different facies, and allows implementation rock physics models to relate porosity and permeability. SGEMS (Stanford University) provides a free set of 2D and 3D geostatistical tools including multipoint geostatistics, and can be very useful for reservoir modeling.
  • 3.2 3D Tools
    Expansion and adaptation of the 2D tools to the 3D case is under research and development. Again the interpolation can be constrained using seismic attributes exhaustively available on the complete 3D reservoir model. We expect that the combined use of compression techniques, machine learning, and interpolation/simulation techniques on the attribute space will produce plausible 3D geological images that are constrained to the observed data. Interpolation on the attribute space is a new technique that involves working in a multidimensional coordinates system that incorporates the geographical coordinates of the data points and the attributes sampled at these points. The aim is to easily incorporate attributes to constrain the interpolation/simulation. We plan to compare the results with those produced by conventional geostatistical techniques using SGEMS (Stanford University). We plan to introduce beta versions of these tools in April 2010.
  • 3.3 Preliminary Results on the Anadarko Data Set
    The Holly Branch data was provided by Anadarko in order to test and train our interpolation methodology. The available data is located on the eastern part of a syncline structure where the main prospects of the area are the Bossier Sand and the Gas Shale. We have at disposal 33 logs including porosity, density, VP, VS, Gamma Ray, neutron and SP. Initially we use both kinds of well-logs and seismic data as data kernel to construct the prior reservoir porosity and permeability models. We have performed preliminary tests using a section including boreholes Knowles A-2, A-1 that intersect the sand and gas shale formations. No seismic attributes are needed to perform this interpolation.
• 4. Self-Learning Expert System Development
  • 4.1 Self-learning expert system for tight gas production: System integration
    The effort in 2009 focused on developing a methodology integrating all data sources using all forms of available information and corresponding uncertainty description to develop an optimization approach for the successive design (location, geometry, stimulation) of wells that maximize gas recovery in a tight gas field. The overall workflow corresponding to the proposed optimization scheme (including data flow, decision-making actions and working groups responsible for development of corresponding modules) is shown below.

    Optimization scheme workflow

    

    Particular emphasis is being placed on the use of self-teaching technologies (the ones being considered now include both Fuzzy Logic and Neural Networks – we have been unable to determine unique advantages of either; see discussion in Section 1) in the analysis of multi-variate and extremely heterogeneous systems. Such systems would require investigation of a very large number of possibilities to determine optimal designs (such as the placement, orientation and length of wells in heterogeneous tight-gas reservoirs), if conventional approaches are used. Such a conventional optimization approach is practically impossible because literally thousands of predictions (samplings) would be necessary before trends that could be used for optimization began to appear. Consequently, we are investigating very powerful optimization techniques associated with Fuzzy Logic and Neural Networks that can yield results within short time frames and without the need for large computational infrastructure.

• 5. Challenges and Bottlenecks
  After discussions among the various participants in this project, the following challenges and bottlenecks have been identified:
  • Development of universally acceptable protocols for data forms for data exchange among the project participants.
  • Standardization of data types and forms for entry into the expert system. A basic requirement is that development of data in the type and format required by the expert system should not place an undue burden on the SETES end user (who would be expected to enter such data to use the system).
  • Neither of the more promising self-teaching methods (Fuzzy Logic and neural Networks) shows any distinct advantage over the other, and we have not yet reached a final decision as to which to select.
  • The replacement of the numerical simulation approach by simple parametric equations that predict production from fractured tight-gas systems has not yet been successful because the parameters do not correlate well to differences in the various properties and conditions.
  • Reliance on graduate students causes problems and temporary setbacks because of graduations of older students and the delays in appropriately preparing and training new students. Because of such disruptions, some straightforward studies have not yet been completed (and universities have been undercharging). These delays may necessitate a request for 6-month, no-cost extension of the project.

Project Start: January 1, 2009
Project End: The original project end was intended to be December 31, 2010. However, because of delays caused by graduate student turnaround (and the time needed to find and train appropriate replacements, in addition to the fact that student recruitment can only occur at very specific times in the academic calendar), it appears that a 6-month delay is likely. To that end, we are preparing a request for a no-cost project extension, with the new end date being June 30, 2011.

DOE Contribution: $1,774,840
Performer Contribution: $1,794,750

Contact Information:
RPSEA – Charlotte Schroeder (cschroeder@rpsea.org or 281-690-5506)
NETL - Virginia Weyland (Virginia.Weyland@netl.doe.gov or 281-494-2517)
LBNL - George Moridis (gjmoridis@lbl.gov or 510-486-4746)

Publications:
Final Project Report [PDF-5.79MB]  User's Guide

Mattar, L., Gault, B., Morad, K., Clarkson, C., Ilk, D., Blasingame, T.A.: "Production Analysis and Forecasting of Shale Gas Reservoirs: Case History-based Approach," paper SPE 119897 presented at the 2008 Shale Gas Production Conference, Irving, TX, USA, 16-17 November 2008.

Johnson, N.L., Currie, S.M., Ilk, D., Blasingame, T.A.: "A Simple Methodology for Direct Estimation of Gas-in-place and Reserves Using Rate-Time Data," paper SPE 123298 presented at the 2009 SPE Rocky Mountain Petroleum Technology Conference, Denver, CO, 14-16 April 2009.

Freeman, C.M., G.J. Moridis, D. Ilk and T. Blasingame, “A Numerical Study of Microscale Flow Behavior in Tight Gas and Shale Gas Reservoir Systems”, Proceedings, TOUGH Symposium 2009, Lawrence Berkeley National Laboratory, 14-16 Sept. 2009 (LBNL-2790E, 2009) – In review for publication in Transport in Porous Media.

Freeman, C.F., Ilk, D., Moridis, G.J., and Blasingame, T.A.: ”A Numerical Study of Performance for Tight Gas and Shale Gas Reservoir Systems" paper SPE 124961 presented at the 2009 SPE Annual Technical Conference and Exhibition, New Orleans, LA, USA, 4-7 October 2009 – In review for publication in SPE Reservoir Evaluation and Engineering.

Moridis, G.J. and Blasingame, T.A.: "Analysis of Mechanisms of Flow in Fractured Tight Gas Reservoirs" paper SPE 125027 presented at the 2009 SPE Annual Technical Conference and Exhibition, New Orleans, LA, USA, 4-7 October 2009.

Ilk, D., Rushing, J.A., and Blasingame, T.A.: "Decline-Curve Analysis for HP/HT Gas Wells: Theory and Applications" paper SPE 125031 presented at the 2009 SPE Annual Technical Conference and Exhibition, New Orleans, LA, USA, 4-7 October 2009.

Bhattacharya, S., Kumar,A. and Nikolaou, M.: “An Integrated Approach to Tight Gas Production”, commitment for presentation at the AIChE Annual Meeting, November 2009

Bhattacharya, S., Kumar,A. and Nikolaou, M.: “A Self-Learning System for Tight Gas Production”, commitment for presentation at the Intelligent Energy Conference and Exhibition, Utrecht, The Netherlands, 23 - 25 Mar 2010.