Application of Artificial Intelligence for Fluid Typing in Exploration Environment using Calibrated Compositional Data
A. Alakeely1, Y. Meridji1, A. Almarzoug1, H. Aljeshi1, and M. Pisharat2
Identifying the type of fluid that will be produced at surface is a significant reservoir characterization challenge that is prone to error and uncertainty in exploration environment. It usually requires rigorous treatment of equation of state coupled with phase envelopes which are usually available in a later stage after drilling the well. The business impact of such errors in fluid identification is remarkable. This can vary from poor completion decisions to incorrect reserves estimation. With the introduction of advanced mud gas logging systems (AMG), quantitative assessment of gas data comparable to PVT analysis is possible in real-time. This facilitates real-time accurate fluid typing. To get the most representative fluid typing results, a frame work has to be established where local production data is mapped to compositional data from PVT through model building techniques. The successful application of this technique has many advantages. It allowed for accurate fluid typing in real-time that provided valuable information for reservoir characterization. This information can impact a spectrum of decisions, starting from rig operations to simulation efforts. In this study, decision trees, which is one form of artificial intelligence, is used to build a model that maps compositional to production data using local data sets. The resulting model is then used as a predictive tool to identify fluid types using AMG data while drilling before any other formation evaluation data, such as wireline logs, becomes available.
AAPG Search and Discovery Article #90188 ©GEO-2014, 11th Middle East Geosciences Conference and Exhibition, 10-12 March 2014, Manama, Bahrain