Applications of Artificial Expert Systems in the Diagnosis and Analysis of Unexpected Spatial and Temporal Changes in Reservoir Production Behavior

Open Access
Author:
Bu Khamseen, Nader
Graduate Program:
Energy and Mineral Engineering
Degree:
Doctor of Philosophy
Document Type:
Dissertation
Date of Defense:
May 29, 2014
Committee Members:
  • Turgay Ertekin, Dissertation Advisor
  • Turgay Ertekin, Committee Chair
  • Paul Griffin, Committee Member
  • Michael Adebola Adewumi, Committee Member
  • Sarma V Pisupati, Committee Member
Keywords:
  • Artificial neural network
  • reservoir simulation
  • artificial expert systems
  • history matching
  • prediction
  • unexpected behavior.
Abstract:
In reservoir simulation, the history matching process can easily become a time sink. Since conventional history matching involves manual input, most guidelines suggest a simplified model be used, and only the parameters that influence the outcome and those with the highest uncertainty be changed (Carlson, 2003). However, even with this simplification, the process of history matching still consumes a considerable amount of time. Moreover, as the complexity of the reservoir increases, the time required to perform a successful history match increases as well. After achieving a satisfactory history match, one is set to perform forecasting studies. Forecasting is the ultimate objective of a simulation study. In later times, if the prediction is not in agreement with the observed data, history matching parameters need to be re-tuned. After this tune-up process, predictions should be close enough to the observed data. When observed data deviates from predictions, reservoir engineers then have the daunting task of identifying the causes of such deviation in a rather short period of time to ensure that necessary preventive measures can be implemented promptly. Expert systems can be used to assist reservoir engineers with refining and re-evaluating their history matching parameters. In this proposed study, an artificial expert system is developed, which in turn provides the reservoir engineer with a good set of starting points to pick the history matching parameters for a newly created reservoir simulation model. The same expert system can also help in fine-tuning the parameters if the model is already history matched. For complex models, the expert system can use new production data to improve the history matching parameters. Once the prediction process is under way, the expert system can be expanded into a suit of diagnostic tools to detect changes in reservoir responses that might occur over time. These problematic changes in responses can be caused by the alteration of the area around the well or from other geomechanical changes that take place within the reservoir. These developed tools are triggered when the well production profile behaves unexpectedly. The first tool looks into the possibility of a developed skin around the well may have caused the decline in production. Another tool evaluates whether a set of perforations have been plugged. If the reservoir has hydraulically fractured wells, an expert tool is used to analyze the fracture effective permeability and length to see if they match the specifications of the fracture job. Two more tools are used to assess two geomechanical features of the reservoir. The first one can help figure out if the production changes are caused by the reservoir compaction and the second looks into the possibility of tarmat breakage at the base of the hydrocarbon column. The last tool looks to identify areas of the reservoir where there is a possibility of having natural fractures. Coupled with the engineer’s expertise, the expert system can be extended to more complex scenarios. It is worth noting that each expert system is explicitly developed for a specific reservoir and is not interchangeable with other reservoirs. The last part of this research involves developing graphical user interfaces (GUIs) that provide user-friendly interfaces for the engineers to input and edit the data and generate numerical and graphical results. It will also enable the engineers to validate the results with the numerical reservoir simulator. In this research, the assisted history matching expert system has shown its ability to bring the reservoir properties used by the reservoir simulation model closer to their original values. In addition, it has helped in reducing the uncertainty ranges for the different parameters used in the history matching process. The diagnostic expert systems have also shown the strength of the artificial intelligence protocol applied in these systems. For the forward solution part, all diagnostic expert systems show excellent results and thus can be safely utilized as proxies to the reservoir simulator. The backward solution part is more difficult to achieve. However, except the tarmat breakdown expert system, all diagnostic expert systems have shown satisfactory accuracy as identified by the user. The tarmat breakdown expert system, performed very well when only one area of the tarmat layer is broken whereas it struggled when another area of the tarmat broke at a later time.