Reservoir prescriptive management combining electric resistivity tomography and machine learning

Abstract

In this paper, I introduce a comprehensive workflow aimed at optimizing oil production and CO2 geological storage. I show that the same methodology can be applied to different categories of problems: a) real-time reservoir fluid mapping for predicting and delaying water breakthrough time as far as possible in oil production; b) real-time CO2 mapping for maximizing the sweep efficiency and storage capacity of CO2 in geological formations. Despite their intrinsic differences, these types of problems share common aspects, issues and possible solutions. In both cases, various geophysical techniques can be applied, including Electric Resistivity Tomography (briefly ERT) for accurate fluid mapping and monitoring. This method is highly effective and sensitive for detecting the type of fluid and for estimating saturation in the geological formations. The robustness and the accuracy of the ERT models increase if densely spaced electrodes layouts are permanently deployed into the production and injection wells. In the first part of the paper, I discuss how in both scenarios of oil production and CO2 storage, we can apply time-lapse borehole ERT method for mapping fluids in the reservoir. Next, I discuss how to apply various techniques of time-series analysis for predicting the evolution of the fluids distribution over time. Finally, using Q-Learning, that is a specific Reinforcement Learning method, I discuss how we can optimize the decisional workflow using our models about past, real-time and predicted fluids displacement. The result is the definition of a "best policy" addressed to both problems of optimized oil production and safe CO2 geological storage. In the second part of the paper, I show benefits and limitations of my approach with the support of synthetic tests.

Keywords

• reservoir monitoring
• water breakthrough
• electric tomography
• machine learning