By improving the technology in recent decades, there are huge data obtained from oil field studies. These data need to be considered to improve our modeling and simulation of the field. The point is that how we can deal with these huge data? How we can use them to improve our modeling and simulation?It takes more importance when we have different types of data including: logs, RCAL and SCAL, X-ray CT images in macro and micro scales, thin section images, geological facts like faults, active tectonic area, and thin layers that are not distinguishable in logging, rock typing and geomechanical units, geomechanical properties, seismic, UGC and graphic well logs of neighbor wells, production history, and others that may not be noted here. How we can use all of them that are different in their nature, dimension, unit, and scale in a single model? How we can trust the model in subsurface with huge amount of uncertainty? What is the role of distance in an intelligent-based solution in oil-filed?

Machine learning is a technology that can deal with all these questions, and it is an inherent part of the modern computational oil industry. I think that the most important points for developing a well-trained intelligent system for modeling and simulation of the oil filed are as follows:

  • The first and foremost important point in developing every intelligent system in petroleum industry is that the intelligent system is an assistive technology. This point is very important. An intelligent system is not something that can carry out everything instead of human expert. It is just an assistive technology.
  • The second point is that the system is very biased to the initial data; so, pre-processing is very important step in developing an intelligent system. Pre-processing should be taken into account with respect to the nature of the data. However, it may not work well in developing an intelligent system. For example, in resistivity log, the value may be very different in oil layers versus water layers, or in X-ray tomography images, there is very high difference between pore pixel values and those pixels exhibit Anhydrite. Therefore, this step should be carried out based on the nature of the data and the instrument used for data acquisition.
  • The third point is that the reliability of an intelligent system is highly depended on the knowledge of the developer in two areas: the first one in petroleum engineering (or any field that the intelligent system is going to be developed in), and the second one in machine learning. An intelligent system that is developed by an expert that have many experiences in the both scientific and practical sides, is more reliable that those developed by low-experienced persons. Therefore, the expert know that how the data should be divided to different groups for training, and what method (e.g. artificial neural network, fuzzy logic, evolutionary algorithms, committee machine, clustering, classification, decision tree, incremental learning, cascade approach, image processing and analysis, deep learning, etc.) or hybrid of what methods can be used for this specific problem.
  • The forth point is that when the intelligent system will be developed, in many cases, it needs to re-design and re-train and debug for reaching the minimum threshold of our acceptable accuracy in the results.

The fifth point is that developing an intelligent system in petroleum engineering must respect the geological fact. Our modeling results must be double checked with geological facts. For example, we cannot estimate production history of a well based on neighbor wells, when a fault is existed between them or the area has very active tectonic. We must use the effect of this phenomenon in our modeling.
In conclusion, I am inclined to mention that Machine Learning enables one to model reservoir and oil filed, and also estimate different properties. It is an assistive technology under the human expert supervision, and need to be design and train perfectly based on enough and processed data. It is now an inherent part of our modern computational oil industry, and it is progressing rapidly now to play more role in the industry.