Summary

This project extended the successful CRM (Capacitance Resistance Model) and Reinforcement Learning (RL) workflow previously applied to the Tabasco field to a newer and high-value portion of the West Sak reservoir. The primary goal was to optimize waterflood operations and gain a deeper understanding of well interactions in a setting with greater uncertainty in reservoir properties.

Project Highlights

• Applied the RL+CRM approach to a less mature field
• Implemented a Bayesian CRM model to quantify uncertainty in model parameters and incorporate prior knowledge, such as field geometry
• Used Bayesian inference to identify unexpected injector-producer communication, later confirmed by reservoir engineering analysis
• Provided actionable insights for optimizing waterflood management and well surveillance

Technical Innovation

Key technical advancements included:

• Integration of Bayesian modeling with CRM to capture parameter uncertainty and leverage outside information
• Use of RL algorithms (via Stable Baselines) for waterflood optimization, even in the presence of high model uncertainty
• Bayesian CRM model revealed unexpectedly high-confidence communication between specific injector-producer pairs, improving field understanding

Impact

While the RL component did not yield new operational strategies, the Bayesian CRM model provided critical insights into reservoir connectivity and well interactions. This knowledge enabled reservoir engineers to optimize waterflood operations and surveillance, supporting more effective management of the West Sak field.