Skip to main content
Shahab D. Mohaghegh
Professor, Petroleum and Natural Gas Engineering
College of Engineering and Mineral Resources
WVU Home

Surrogate Intelligent Models (SIMs)

Surrogate Intelligent Models - SIMs

The reliance on supercomputing facilities, and long processing times required for such high-performance simulations make them impractical for use in situations where real time information processing is required. Real time decision making, real time optimization and analysis under uncertainty are three of such situations. Indeed, accurate simulation of complex physical processes usually takes much longer than the actual physical time of the process. This is especially true for complex fluid-dynamics and general multi-physics processes. This makes the advanced CFD methods and some of the solutions for PDEs virtually useless in predicting the course of events or making real time decision and optimization of events which are currently in progress. A parallel of similar situation in the oil and gas industry is the smart field initiatives currently being entertained by many majors and national oil companies where real time decision making, real time optimization and real time analysis under uncertainty becomes a vital issue.

If simulators are to be used effectively for real time analysis, they must be capable of processing potential scenarios much faster than their current speed. For mission critical processes such as real time analysis of uncertainties associated with input parameters, real time decision making and real time optimization where comparison of multiple scenarios becomes essential, simulators must approach real time or near real time speeds and be capable of providing results in fraction of a second instead of minutes or hours. The set of techniques that when combined is capable of such performance is based on hybrid intelligent systems and include, but is not limited to, artificial neural networks, genetic algorithms and fuzzy logic. Using intelligent systems it is now possible to build Surrogate Intelligent Models (SIMs) that can mimic functionalities of complex simulators in real time.

Surrogate Reservoir Models - SRMs

Reservoir simulation has become the industry standard for reservoir management. It is now used in all phases of filed development in the oil and gas industry. The routine of simulation studies calls for regular integration of all the static and dynamic measurements into the reservoir model as they become available and enhancing the full field model regularly. The full field reservoir models that have become the major source of information and prediction for decision making are continuously updated and major fields now have several versions of their model. Each new version usually is a major improvement over the previous version. The newer versions have the latest information (geologic, geophysical and petro-physical measurements, interpretations and calculations based on new logs, seismic data, injection and productions, etc.) incorporated in them along with adjustments that usually are the result of single-well or multi-well history matching.

No serious alternative to the conventional reservoir simulation and modeling is in the horizon. It is a well understood technology that usually works well in the hand of experience modelers incorporating reasonably good geological, geophysical, and petro-physical interpretations and measurements with the reasonably sophisticated simulators that are currently available in the market. The reservoir models that are built for an average size field with tens and sometimes hundreds of wells tend to include very large number of grid blocks. As the number of reservoir layers or the thickness of the formations increase the number of cells included in the model approaches several millions. Technologies such as Local Grid Refinements have been developed to dampen the geometric increase of the number of grid blocks required for detail and focused simulation and modeling around the wellbore and locations in the reservoir where more detail is required, but the size of the models remains in the several millions of cells.

As the size of the reservoir models grow the time required for each run increases. Schemes such as grid computing and parallel processing helps to a certain degree but cannot close the large gap that exists between simulation runs and real-time processing. On the other hand with the new push for smart fields (a.k.a. i-fields) in our industry that is a natural growth of smart completions and smart wells, the need for being able to process information in real time becomes more pronounced. Surrogate Reservoir Models - SRMs are the natural solution to address this necessity. Surrogate Reservoir Models are intelligent prototypes of the full field models that can run in fraction of a second rather than in minutes or hours. SRMs mimic the capabilities of a full field model with very high accuracy. They have the advantage of real-time processing. State-of-the-art SRMs can be developed regularly (as new versions of the full field models become available) off-line and can be put online for real-time processing that can guide important decisions.

Real-Time Optimization

In order to perform real-time optimization for any process in the oil field, a comprehensive solution space of the process being optimized is an absolute necessity. A comprehensive solution space is usually developed based on the objective function that accurately represent the process being optimized and that can predict process outcomes. If we assume that:

  1. The process being optimized is identified, and the problem is well defined.
  2. The objective function for the process is available, usually in the form of a computer model, examples of which are:
    • Full field reservoir models for underground fluid movements.
    • Surface facilities models for gathering systems and compressor stations.
  3. The objective function can be used to develop the necessary solution space for the project objective.
  4. The model (objective function) is utilized and a comprehensive solution space is developed.

Then an intelligent and efficient search algorithm can identify the desired optimum from the solution space developed in the step 4 above. None of these steps mentioned so far is a big deal and can be done with today's technology. So where is the "Achilles' heel" of this process?

The "Achilles' heel" is the fact that none of the above processes can be performed in real-time. For an i-filed (smart field) implementation what is needed is the real-time processing of all the information in order to make real-time optimization. Surrogate Intelligent Models can bring real-time processing to i-filed (smart field) of future.

Real-Time Decision Making

The major advantage of all the equipment and sensors that have been developed and are being developed for the use in i-filed (smart field) of future is that for the first time they make it possible for the engineers, geo-scientists and managers to observe and monitor what is happening in the reservoir in real-time. That is a great achievement. But what would you do with that information? How would you use it to your advantage? How are you going to influence the outcome of the process that you are observing or monitoring? The main advantage of being able to see something happening in real-time is to be able to intervene in the process in real-time in order to correct the wrong that is happening or to enhance the outcome of the process.

In order to be able to do such things you need to make accurate decisions in real-time. For making the best decisions in real-time you would need predictive tools (like the full field models) that can provide answers to a variety of possible scenarios in real-time, hence the need for Surrogate Intelligent Models.

Real-Time Analysis Under Uncertainty

We all agree that many of the measurements and interpretations that go into our full filed models are far from being certain. One of the ways to deal with these uncertainties is using Monte Carlo simulation method. In Monte Carlo simulation inputs to the objective function (the full field model in the case of a reservoir simulation process) are presented in the form of probability distribution functions rather than crisp, certain values. The Monte Carlo simulation method requires the objective function to be run hundreds or thousands of times in order to generate probability distribution functions of the objective function's outcome.

Using a full field model, in the case of a reservoir simulation as the objective function, for analysis under uncertain conditions is not practical, especially if the process is being monitored in real time and analysis must be performed in real-time. Again Surrogate Intelligent Models are the answer.

DEVELOPING SURROGATE INTELLIGENT MODELS

The art and science of developing Surrogate Intelligent Models are by no means trivial. The set of techniques that when combined is capable of such performance is based on hybrid intelligent systems and include, but is not limited to, artificial neural networks, genetic algorithms and fuzzy logic. Using intelligent systems it is now possible to build Surrogate Reservoir Models (SRMs) that can mimic functionalities of complex full field models in real time.