Expert Systems for Gas Production Prediction from Hydraulically Fractured Horizontal Wells Completed in Shale Gas Reservoirs and Establishing Equivalencies Between Different Hydraulic Fracture Representations
Siripatrachai, Nithiwat; Bodipat, Kanin; Ertekin, Turgay
To unlock natural gas from shale reservoirs, horizontal wells coupled with hydraulic fracturing are implemented. In this paper, two hydraulic fracture representations (discrete transverse fracture representation and crushed zone representation) are considered. Service companies design multi-stage transverse hydraulic fractures. In each stage, a massive volume of pressurized fluid is injected into the reservoir to create fractures which serve as high permeability pathways for natural gas to flow to the wellbore. In reservoir simulation studies, each fracture can be represented by a transverse fracture plane of high conductivity. However, microseismic field data show that multi-stage hydraulic fracturing can result in a stimulated reservoir volume which is referred as "crushed zone". The "crushed zone" has relatively higher permeability and smaller fracture spacing compared to the unstimulated zone. In the numerical model, crushed zone is represented by an elliptic zone of higher permeability and smaller fracture spacing around the wellbore.
Conducting simulation runs for optimization of a design can be time consuming and the optimal design may never be achieved. Additionally, we are interested in the equivalency of the two representations described above. In this study, the equivalency between the two hydraulic fracture representations is achieved when the average gas production rates from two representations are in agreement within a margin of ±10%. Establishing the equivalency between the two representations can prove to be an arduous task. In this work, artificial neural network(ANN) technology is utilized to make the sought equivalency more easily attainable. ANN is widely accepted for its ability to instantly provide simulation results for complex problems. Accordingly, a properly structured and trained ANN can serve as a powerful tool in reservoir simulation and overcome the aforementioned challenges. We use numerical reservoir models to generate production profiles for hydraulically fractured horizontal well to train the network which eventually establishes the sought equivalency between two representations. The search process can start with any of the two hydraulic fracture representations. Given reservoir properties and one hydraulically fractured horizontal well modeled with one representation, the developed ANNs can instantly predict gas production rates as well as establish its equivalent hydraulic fracture representation.
AAPG Search and Discovery Article #90163©2013AAPG 2013 Annual Convention and Exhibition, Pittsburgh, Pennsylvania, May 19-22, 2013