Author ORCID Identifier

https://orcid.org/0000-0003-4647-3455

Semester

Spring

Date of Graduation

2023

Document Type

Dissertation

Degree Type

PhD

College

Statler College of Engineering and Mineral Resources

Department

Petroleum and Natural Gas Engineering

Committee Chair

Shahab D. Mohaghegh

Committee Member

Sam Ameri

Committee Member

Kashy Aminian

Committee Member

Mehrdad Zamirian

Committee Member

Dengliang Gao

Committee Member

Fareed AlHashmi

Abstract

With the rise of high-performance computers, numerical reservoir simulators became popular among engineers to evaluate reservoirs and develop the fields. However, this technology is still unable to fully model the reservoirs with commingled production and highly complex geology, especially when it comes to uncertainty qualification and sensitivity analysis where hundreds of runs are required. This dissertation aims to provide a successful case study of the history matching of a complex reservoir in the North Sea (Volve field).

The proposed model relies only on the measured field variables such as well, formation, completion characteristics, production rates, and operational conditions while it stays away from interpretation and assumption. More than eight years of data from the Volve field was used to generate a comprehensive dataset, and then key parameters were extracted using fuzzy pattern recognition. A system of fully coupled artificial neural networks (feed-forward and LSTM networks) was used to train, calibrate and validate the model. The Artificial neural network enables us to extract hidden patterns in the field by learning from historical data.

The model successfully history-matched the well-head pressure, well-head temperature, and production rates of all the wells through a completely automated process. The forecasting capability of the model has been verified through blind validation in time and space using data that the model has not seen before.

In contrast to the numerical simulator, which is only a reservoir model, this technology is a coupled reservoir and well-bore model which is able to learn the fluid motion behavior in a complex porous media with fewer resources and higher speed. The efficiency of this approach makes it a suitable tool for uncertainty quantification when a large number of runs is required. The combination of artificial intelligence and domain expertise makes this technology more reliable and closer to reality by staying loyal to field measurements.

Download

Share

COinS