In this work I am building models using data generated in our earlier paper Modeling "Underlying Physics of Heavy Oil Recovery by Gas Injection: An Experimental Parametric Analysis When Oil Exists in the Form of Oil Based Emulsion
Mohammedalmojtaba Salama, Lixing Lin, Georgeta Istratescu, Tayfun Babadagli University of Alberta
Amin Bademchi Zadeh, Mark Anderson, Chris Patterson Canadian Natural Resources Limited
In this paper, we focus on the interfacial properties, relative volume change, and PVT behavior of CO 2 and CH 4 in (w/ho) emulsions, which is encountered in heavy oil reservoirs during secondary recovery methods such as water or steam injection. We generated the water-in-heavy- oil (w/ho) emulsion using steam at 150 o C for two types of oils (27,000 cP and 4,351 cP). vol.% range from 10-70.
After conducting initial data exploration and visualization I explopred different models to predict IFT and volume expansion given the water content, gas type, and time. A linear model was implemented with Ridge model to control for overfitting, but the obtained score was still low. Using GradientBoostingRegressor I was able to get 99% accuracy on the test data after applying a full pipeline for preprocessing 'scaling' numerical data and 'encoding' categorical variables. Validation using GridSearchCV was performed to optimize the learning rate
My presentation at University of Alberta can be found here
Data is available upon request: Email : [email protected]