Date of Graduation


Document Type


Degree Type



Statler College of Engineering and Mineral Resources


Petroleum and Natural Gas Engineering

Committee Chair

Shahab Mohaghegh

Committee Member

Kashy Aminian

Committee Member

Sam Ameri

Committee Member

Mehrdad Zamirian


Numerical simulation and data-driven modeling are two current approaches in engineering reservoir modeling. Numerical reservoir simulation attempts to match past production history by modifying reservoir properties of the model. After multiple computationally intensive trial and error efforts, accurate history matches are identified. These history matches are used by project management for production forecasting purposes. Data-driven reservoir modeling utilizes measured data and is, therefore, free of assumptions that are often included in numerical reservoir simulations. Artificial intelligence and machine learning algorithms are technologies implemented in the development of a data-driven reservoir model with efforts to learn fluid flow through porous media from the datasets provided to the system. Training, calibration, and validation datasets ensure the success during the teaching process.

Models, such as oil, gas, and water production, reservoir pressure and water saturation, are trained, calibrated, verified to ensure the success in the teaching process. After appropriate hyper-parameter tuning, well-trained models are tested on blind datasets. This leads to the model being deployed on new datasets to again test the model’s performance during forecasting. The accuracy is based upon the model achieving a similar result to what numerical simulation found on the same dataset.

The objective of this thesis is to use a 22-year dataset from a reservoir model generated by a numerical simulator that undergoes many complexities to approach the reality that takes place in the industry through the use of operational constraints, workover events requiring shut-in, random bottom hole pressure (“BHP”) trend, etc. Should the Top Down Model (“TDM”) be able to accurately learn the relationships between input attributes and outputs from the data, a similar procedure can then be applied to real field data in the future. The TDM’s capabilities will attempt to be proved when the blind validation’s in-time results match the data from the numerical simulation reservoir model. This will be done by excluding the data from the training, calibration, and validation datasets used to create the TDM. This thesis’s results should demonstrate that data-driven modeling is capable of history matching and forecasting the complexities of a reservoir.