Spe 59767

This paper summarizes the development of a methodology for the restimulation candidate selection in tight gas sands. The methodology incorporates virtual intelligence techniques (artificial neural networks, genetic algorithms and fuzzy logic) to achieve this objective. Artificial neural networks are used to develop a representative model of the completion and hydraulic fracturing process in a specific field. Genetic algorithms are used as a search and optimization tool to identify the missed incremental production based on the neural network model. Finally fuzzy logic is used to capture the unique field experiences of the engineers as well as detrimental parameters (if such parameters are indeed present) and incorporate them in the decision making process. Approximate reasoning approach is used at the decision making level to identify the restimulation candidates. Once the methodology is introduced, it is applied to an actual tight sand field in the Rocky Mountain region and the results are presented. Statement of the Problem In 1996, the Gas Research Institute (GRI) performed a scoping study to investigate the potential for natural gas production enhancement via restimulation in the United States (lower 48 onshore). The results indicated that the potential was substantial (over a Tcf in five years). Particularly in tight sand formations of the Rocky Mountains, Mid-Continent and South Texas regions. However, it was also determined that industry’s current experience with restimulation is mixed, and that considerable effort is required in candidate selection, problem diagnosis, and treatment selection/design/ implementation for a restimulation program to be successful. Given a lack of both specialized (restimulation) technology and “spare” engineering manpower to focus on restimulation, GRI initiated a subsequent R&D project in 1998 with several objectives. Those objectives are to 1) develop efficient, costeffective, reliable methodologies to identify wells with high restimulation potential, 2) identify and investigate various mechanisms leading to well underperformance, and 3) develop and test restimulation techniques tailored to each cause of to well underperformance. Addressing the first of the project objectives, an integrated methodology has been developed to select high-potential restimulation candidates in a reliable, cost-effective manner. The technique involves several steps. First, sophisticated statistical approaches are utilized to identify both obvious and subtle differences in well performances, and provide initial insights into potential candidate wells. Secondly, virtual intelligence techniques (a hybrid of artificial neural networks, genetic algorithms, and fuzzy logic) are used to recognize patterns in well performances as they relate to both geologic/reservoir conditions and completion/stimulation operations. With this information, controllable well performance “drivers” can be identified, and this information can in turn be used to select candidate wells, identify possible causes of well underperformance, and begin the treatment selection process. Third, engineering methods such as typecurves are used to high-grade potential restimulation candidates by providing a (relative) indication of reservoir quality and completion efficiency, and hence restimulation potential. Finally, high-potential candidates are individually screened for mechanical integrity, reservoir pressure and other important historical information that may not be uncovered in the previous steps. Lastly, low-cost candidate verification tests are performed to ensure candidate selection potential. This paper is a review of the second technique used in this study, namely the use of virtual intelligence for identifying restimulation candidates. A brief introduction on the tools used in this study will be presented followed by a description SPE 59767 Development of an Intelligent Systems Approach for Restimulation Candidate Selection Shahab Mohaghegh, West Virginia University, Scott Reeves, Advanced Resources International, David Hill, Gas Research Institute 2 MOHAGHEGH, REEVES, HILL SPE 59767 of how techniques were implemented in achieving the project objectives. Application of the methodology to a field in Rock Mountains concludes the paper. Introduction to Virtual intelligence Virtual intelligence may be defined as a collection of new analytic tools that attempts to imitate life. Virtual intelligence techniques exhibit an ability to learn and deal with new situations. Artificial neural networks, evolutionary computing and fuzzy logic are among the paradigms that are classified as virtual intelligence. These techniques possess one or more attributes of "reason", such as generalization, discovery, association and abstraction. In the last decade virtual intelligence has matured to a set of analytic tools that facilitate solving problems that were previously difficult or impossible to solve. The trend now seems to be the integration of these tools together, as well as with conventional tools such as statistical analysis, to build sophisticated systems that can solve challenging problems. These tools are now used in many different disciplines and have found their way into commercial products. Virtual intelligence is used in areas such as medical diagnosis, credit card fraud detection, bank loan approval, smart household appliances, subway systems, automatic transmissions, financial portfolio management, robot navigation systems, and many more. In the oil and gas industry these tools have been used to solve problems related to pressure transient analysis, well log interpretation, and reservoir characterization, among other areas. Artificial Neural Network Artificial neural networks are probably the best known of the techniques that have been used in this study. Neural networks, a biologically inspired computing scheme, are an analog, adaptive, distributive, and highly parallel system that has been used in many disciplines and has proven to have potential in solving problems that require pattern recognition. The main interest in neural network has its roots in the recognition that the brain processes information in a different manner than conventional digital computers. Computers are extremely fast and precise at executing sequences of instructions that have been formulated for them (algorithms). A human information processing system is composed of neurons switching at speeds about a million times slower than computer gates. Yet, humans are more efficient than computers at computationally complex tasks such as speech understanding and other pattern recognition problems. Artificial neural systems, or neural networks, are physical cellular systems, which can acquire, store, and utilize experiential knowledge. The knowledge is in the form of stable states or mapping embedded in networks that can be recalled in response to the presentation of cues. Unlike a digital, sequential computer with a central processor that can address an array of memory locations, neural networks store knowledge in the overall state of the network after it has reached some equilibrium condition (stable state). In other words, knowledge in a neural network is not stored in a particular location. Knowledge is stored both in the way processing elements are connected, and in the importance of each input to the processing element (embedded mapping). Pattern recognition has proven to be one of neural nets' strong points. The essence of pattern recognition is the concurrent processing of a body of information, all of which are available at the same time. The parallel-distributed information processing characteristics of neural networks accommodate this necessity. The science of pattern recognition is concerned with three major issues; 1) The appropriate description of objects, physical or conceptual, in terms of representation space, 2) The specification of an interpretation space, and 3) The mapping from representation space into interpretation space. Another important characteristic of neural networks is their adaptability. Neural networks do not use algorithmic processes. They respond (like humans) to things learned by experience. Therefore, it is necessary to expose the network to sufficient examples, so it can learn and adjust its links and connections between different neurons. Neural networks can be programmed to train, store, recognize, and associatively retrieve patterns or database entries; to solve combinatorial optimization problems; to filter noise from measurement data; and to control ill defined problems; in summary, they estimate sampled functions when we do not know the form of the functions. As such, they are well suited for modeling a complex problem, like the accurate prediction of gas production from tight sand wells, where the factors that influence production are many an varied (e.g., reservoir properties, completion/stimulation procedures, etc.)