Insights on the Thermal Efficiency of SAGD from Data Analytics

The research in this thesis explores three aspects of Steam-Assisted Gravity Drainage (SAGD) oil recovery using public information, with the aim of uncovering new insights about field performance that cannot be deduced from reservoir physics alone. Topics are chosen by the amount and type of public data available, and each analysis spans multiple SAGD fields. A literature review is presented to understand how machine-learning algorithms have been used in SAGD research. Machine learning does not guarantee physically consistent results, therefore particular attention is paid to how analysis results are validated. A description of an ideal SAGD machine learning study is compiled and used to rate all the studies reviewed. The first thesis study uses temperature profiles from observation wells near 13 SAGD well pairs to estimate the volume of gas accumulated at the top of a steam chamber. Together with known or estimated volumes of gas co-injected, produced, and mobilized within the reservoir, a gas material balance is calculated to estimate the unknown volume of gas generated in situ. Heat transfer is also examined in relation to the presence of gas. The second thesis study develops a new Bayesian biclustering method to find and differentiate groups within 328 SAGD well pairs based on their oil production response to steam injection over time. Clusters are described with probability distributions that capture the likelihood of transitioning between discrete steam-to-oil ratio states. Behaviour differences are then described and explained. The third thesis study specifies performance of 1,520 well pairs as a ratio of energy produced to energy injected (EPEI), noting that EPEI ratios become almost constant within the first year of operation. This distinctive behaviour is discussed within the context of five manually defined groups, five case study well pads, and 26 independent parameters. One-by-one parameters are eliminated until only a few remain. Interactions of the remaining parameters are examined with sensitivity analysis. Lastly, corroborating or contradicting findings are reconciled across all three studies – this is the best approach found to verify insights. The results of this research demonstrate that data analytics and machine learning can play a complementary role to physical modelling.

Name thesis:degree_name

Doctor of Philosophy (PhD)

Discipline thesis:degree_discipline

Engineering – Geomatics

Grantor dc:publisher.institution

Schulich School of Engineering

Year dc:date.issued

2020

Author dc:creator

• Pinto, Helen

Advisors dc:contributor.advisor

• Wang, Xin
• Gates, Ian Donald

Committee members dc:contributor.committeemember

• Chen, Shengnan
• Liang, Steve H. L.
• Gao, Yang
• Leung, Juliana Y.

Statement dc:rights

• University of Calgary graduate students retain copyright ownership and moral rights for their thesis. You may use this material in any way that is permitted by the Copyright Act or through licensing that has been assigned to the document. For uses that are not allowable under copyright legislation or licensing, you are required to seek permission.

Language dc:language.iso

eng

DOI

http://dx.doi.org/10.11575/PRISM/37575

OAI identifier oai:identifier

oai:ucalgary.scholaris.ca:1880/111643