Semester

Summer

Date of Graduation

2014

Document Type

Thesis

Degree Type

MS

College

Statler College of Engineering and Mineral Resources

Department

Lane Department of Computer Science and Electrical Engineering

Committee Chair

Shahab D. Mohaghegh

Committee Co-Chair

Sam Ameri

Committee Member

Kashayar Aminian

Committee Member

Ebrahim Fathi

Committee Member

Ali Takbiri

Abstract

Nowadays, due to advancements in data acquisition technologies in oil and gas industry more data are available for generating reservoir simulation models. This leads to high fidelity reservoir simulation models which are highly complex and computationally expensive. The conventional reservoir management studies require hundreds realizations of the simulation models. As S. Gencer (2007) described, the reservoir simulation trend is towards "more: more users, more models, more cells, more wells, more cases, more data and more integration" [1]. In order to enhance the reservoir model descriptions, more computational power would have to be designed and engineered to keep up with our modeling needs; hence, creating an unsustainable cyclical process. Therefore, even with the advancements in the computational powers, the industry cannot take advantage of the full potential of these full-field reservoir simulation models.;Many studies have tried to create alternative methods in order to replicate the performance of full-field reservoir simulation models and at the same time decrease the cost of operation. Traditional proxy models, such as statistical based approaches, are examples of these studies. The degree of success, particularly practical aspects, for these approaches remains to be argued.;As an alternative to traditional proxy modeling methods, the objective of this study is to investigate the feasibility of use of a fast intelligent approximation of the numerical simulation model. This replica will accurately reproduce dynamic reservoir properties of complex full-field numerical simulation models in matter of seconds. A Grid-based Surrogate Reservoir Model (GSRM) is developed based on data-driven and Artificial Intelligence techniques. This technology is able to learn from the provided examples of the reservoir simulation model. The robustness of this technology is validated by testing it on non-seen instances. Finally the trained and validated GSRM will produce the results of full-field simulation models accurately and in a very short time (seconds).;This concept will be proven by building a GSRM of CO2 injection--EOR numerical model of SACROC field, Scurry County, Texas. The SACROC model (CMG GEM) in use was previously generated and history matched by the Petroleum Engineering & Analytics Research Lab - PEARL - at West Virginia University, it is based on a comprehensive geological study that includes 3D seismic surveys and well logs; in order to generate the GSRM this model is to be ran using multiple injection scenarios that will create an appropriate solution space so we can comprehend and grasp its behavior using artificial intelligence.

Share

COinS