The principal objective of this paper is to develop an artificial expert system capable of instantaneously and accurately predicting complex wells (CW) performance, proposing CW designs, and predicting average reservoir properties, for shale gas wells operating under specified bottom-hole pressure.

Artificial neural networks (ANN) provide the backbone of the expert system. Other methods, such as traditional well-testing, numerical reservoir simulation, and decline curve analysis, have inherent limitations or require significant time and effort expended. ANN methodology has the ability to recognize patterns between various parameters in the presence of large databases and is a powerful tool especially when the existing relationships between the dependent and independent parameters are vague or are not well understood. Thus, it is capable of instantly solving problems that do not have known analytical or numerical solutions. CW are scarce in shale gas reservoirs, thus utilizing real data in ANN training is not possible most of the time. Accordingly, numerical reservoir simulation is used to generate the database necessary for training the expert system.

The expert system developed in this research instantly and accurately performs the tasks below for shale gas CW operating under constant bottom-hole pressure conditions and they are capable of:

  1. Predicting production rates for a given CW design from a given shale gas reservoir.

  2. Proposing a robust CW design capable of producing a given production profile from a given set of reservoir properties.

  3. Predicting unconventional reservoir rock properties corresponding to a given gas production profile from a given CW design.

In addition, results prove that a well-trained ANN is capable of making instantaneous and accurate predictions. Results also increase our confidence in utilizing ANN to solve complex problems in the oil and gas industry. To increase accuracy of the expert system, reintroduction of data combinations as functional links helps reducing prediction error.

CW consume lower amount of water and have a controlled exposure to reservoir, making them an attractive alternative to the common massively hydraulically fractured horizontal wells (MHFHW). Field testing shale gas CW allows us to utilize real data in calibrating and testing the expert system.

Keywords:
Artificial Intelligence, neural network, hydraulic fracturing, cumulative production, machine learning, complex reservoir, lateral, porosity, Prediction error, shale gas
Subjects:
Hydraulic Fracturing, Reservoir Characterization, Unconventional and Complex Reservoirs, Shale gas
Copyright 2017, Society of Petroleum Engineers
You can access this article if you purchase or spend a download.