Simple Models for Unconventional Hydrocarbon Recovery

Graduate student: Brian Lee
Principal Investigator: Larry Lake

The goal of the project is to develop a probabilistic graphical model that, when given some properties of the reservoir, produces a predicted value of recovery factor. The fundamental mechanics behind this process is described as Bayes’ updating applied to a chain of variables, each with its own probability distribution.

The variable network structurally contains assumptions of correlation and independence, and each variable’s conditional probability distribution is trained using reservoir property data from databases. Various network models will be tested for accuracy and efficiency. The product of this research is an expert system that takes advantage of available data to predict recovery factors of reservoirs at a low computational cost.

CPGE CPARM Simple models for unconventional hydrocarbon recovery

 

CPGE CPARM Simple models for unconventional hydrocarbon recovery