The Fuzzy Expert Exploration Tool for the Delaware Basin: Development, testing and applications S.M. Schrader a, * , R.S. Balch b , T. Ruan b a Department of Petroleum Engineering, Montana Tech, Petroleum Building 208, 1300 W. Park Street, Butte, MT 59701, United States b Petroleum Recovery Research Center, New Mexico Tech, United States article info Keywords: Expert systems Neural networks Petroleum reserves abstract The Delaware Fuzzy Expert Exploration Tool (FEE Tool) is an expert system designed to reduce explora- tion risk for the Lower Brushy Canyon formation of the Delaware Basin. The components of the Delaware FEE Tool include a knowledge base containing sets of rules developed through expert interviews, an answer base of numerical inputs to these rules, an inference engine that uses fuzzy logic to evaluate the rules with answer base or user-provided data, and a user interface where the user can work with input data and interpret the tool’s results. For each of 60,478,40-acre locations in the New Mexico portion of the Delaware Basin, the FEE Tool out- put includes a scaled quality estimate in the set {0, 1}, with a value of 0.65 or greater, indicating a low risk prospect. In testing, the quality estimates were found to be significantly higher at locations where recent successful wells were located. The Delaware FEE Tool was also used as a reserve estimation tool by relating the FEE Tool estimate at a known producing well to its total expected production. Then the FEE Tool estimates at undrilled locations were used to calculate a reserve estimate. Using this approach, the probable regional reserves were esti- mated to fall between 278 and 432 million bbls. Ó 2008 Elsevier Ltd. All rights reserved. 1. Introduction Soft computing techniques including expert systems, fuzzy lo- gic and neural networks, are powerful tools with many applica- tions in the petroleum engineering field. An expert, or knowledge-based system, is a computer program that allows the user to apply knowledge collected from experts to solve a problem. An early example of the use of expert systems in petroleum engineering was the MUD system; an expert system developed to help users’ select appropriate drilling muds (Kahn & McDermott, 1993). The MUD system stored expert knowledge in a series of rules contained in a knowledge base. A second example of a knowledge-based expert system is the RELPERM software (Ali & Fawcett, 1996). This system contains expert-derived rules that help the user to acquire relative permeability models for use in res- ervoir simulators while minimizing the need for costly laboratory studies. Another example is the development of an expert system to aid diagnosing formation damage mechanisms and designing stimulation treatments (Xiong, Robinson, & Foh, 2001). In this sys- tem the knowledge base was developed through the use of inter- views, literature reviews and field examples. Fuzzy logic is a type of logic in which an element can have partial membership in a set. In classical or predicate logic, an ele- ment is either a member or not a member of a set. For instance, it may be reasonable to describe a well as having been either com- pleted or not completed; however it may not be reasonable to de- scribe a formation rock as porous or not porous. In the latter case, linguistic terms such as slightly porous or highly porous might be used to describe the rock. Fuzzy logic provides the mathematical tools to work with these types of descriptions. Many expert sys- tems, such as the formation damage system (Xiong et al., 2001), use fuzzy logic to store and evaluate expert rules. When the infer- ence engine (the process in which the rules are evaluated) is based on fuzzy logic, the system may be termed a fuzzy expert system. The FEE Tool is an example of a fuzzy expert system, as is a program termed MULTSYS (Garrouch, Lababidi, & Ebrahim, 2004), a web-based fuzzy expert system designed to aid in well completion. Neural networks are a type of soft computing in which, after exposure to data, the machine ‘‘learns” to recognize patterns. A neural network consists of a network of artificial neurons de- signed to mimic the biological neurons in the human brain. In most applications, the neural network is trained by providing it with data sets, including the input data and the desired out- puts. The difference between the neural network output and 0957-4174/$ - see front matter Ó 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2008.08.004 * Corresponding author. Tel.: +1 432 552 2217. E-mail address: sschrader@mtech.edu (S.M. Schrader). Expert Systems with Applications 36 (2009) 6859–6865 Contents lists available at ScienceDirect Expert Systems with Applications journal homepage: www.elsevier.com/locate/eswa