47th Annual AAPG-SPE Eastern Section Joint Meeting

Datapages, Inc.Print this page

Candidate Well Selection for Re-Fracturing in Tight-Gas Sand Reservoirs Using Artificial Intelligence


Tight-gas sand reservoirs require large hydraulic-fracture stimulation treatments through horizontal wells to be able to recover economic volumes of gas. Characterizing the uncontrollable hydraulic-fracture properties along the horizontal wellbore to identify candidate wells for re-fracturing from limited data remains as a challenging and resource-demanding decision-making problem. In this study, an artificial-intelligence based decision-making protocol is developed to identify wells with relatively low hydraulic-fracture quality in terms of permeability, width and half-length. The methodology is based on fuzzy logic which is used to deal with imprecision and subjectivity through mathematical representations of linguistic vagueness. A fuzzy inference system is developed based on fuzzy if-then rules and fuzzy reasoning. The following key assumption is defined: wells with similar reservoir/hydraulic fracture characteristics and operational conditions are expected to display similar performance characteristics. Similarly, if two wells with similar reservoir and operational characteristics behave differently in terms of production, then the hydraulic-fracture quality must be different. Considering this assumption, we grouped all parameters related to a tight-gas horizontal well into the following categories: reservoir quality, initial conditions, operational constraints and hydraulic-fracture quality. For each category, we developed a rule-based fuzzy-inference system that quantifies the quality of that category (higher quality is better for well performance). Quality indices are calculated by inferring from 27 linguistic rules for 3 input parameters for each category. We also defined a performance index as a function of initial flow rate and cumulative recoveries after 3 and 5 years. Our final fuzzy-inference system for decision making had 4 inputs (reservoir, operational, initial conditions and performance indices) and 1 output (re-fracturing potential index). A total of 81 linguistic rules were developed to quantify the need for re-fracturing. Using available data, all rules are evaluated simultaneously to output the re-fracturing candidacy index for a given well. This approach was successfully validated using case studies from Granite Wash and Williams Fork tight-gas sand formations. These examples showed that the developed system can be used to quickly identify wells in which there is a room for improvement in terms of the hydraulic-fracture quality.