Well Testing Using Artificial Expert Systems: applications And Limitations

Open Access
Author:
Alabbad, Mohammed Ali
Graduate Program:
Energy and Mineral Engineering
Degree:
Doctor of Philosophy
Document Type:
Dissertation
Date of Defense:
May 08, 2012
Committee Members:
  • Turgay Ertekin, Dissertation Advisor
  • Sridhar Anandakrishnan, Committee Member
  • Luis F Ayala H, Committee Member
  • Yilin Wang, Committee Member
  • R Larry Grayson, Special Member
Keywords:
  • well testing
  • neural networks
  • pressure transient
  • predict
Abstract:
Well testing analysis is one of the important tools that a petroleum engineer uses to characterize and analyze the reservoir of concern. The pressure response obtained from disturbing the well by (build up, drawdown, drill stem, injection or falloff test) at a controlled flow rate can describe the reservoir rock-fluid system and will usually provide information that will help in estimating: permeability, damage, reservoir extent and the degree of heterogeneity the system possess. These tests are costly in terms of equipment, labor and most importantly the delay caused by halting the production in part of the field during the test. This study will provide a tool that can help in performing a virtual well test for a reservoir using basic reservoir-production information with partial well test information on the field already available. The ultimate goal is to be able to generate virtual well test data at potential locations to be drilled in future. The importance of this study arises when a considerable amount of reservoir information is not available and, hence, a reservoir model might not be as accurate as it is aimed for. The study starts with a simple scenario with complete homogeneity in terms of uniform thickness, porosity, permeability, a regularly shaped reservoir and constant flow rate wells. Subsequently, the study moves towards more complex domains and field conditions. The fluid system was dry gas and the rock properties were specified according to the scenario complexity. The best network in each scenario was preserved. The artificial neural network tool predicted the pressure transient data over a wide range of selected areas in the reservoir in a short time. When the study moves towards a more complex reservoir system, the proposed ANN network was able to solve the preceding less complex scenario as well and usually with a higher resolution in prediction. At the end of the study, there were 10 different artificial networks each representing a different reservoir and production strategy. These networks were presented in a graphical user interface models to provide an easier communication and better understanding between the models and the end user. There was a good agreement between the expert system and the simulation results which proves the promising applicability of this technique in reservoir engineering and evaluation protocols. The prediction of pressure transient testing by utilizing the developed models will help in a more efficient and better evaluation for the reservoir under study. A preliminary data will be available with minimal cost and no loss of production.