Quantitative Conformance Assessment in CO2 Storage Reservoirs under Geological Uncertainties Using Convolutional Neural Network Classifiers

Responsible use of geological reservoirs for storage purposes requires that operators demonstrate that their assets can be managed safely or, in other words, in conformance with the intended plan and targets. Smart monitoring plans provide sufficient evidence of the reservoir behavior to improve the understanding of the system and support decision making regarding subsequent development, operational and monitoring activities. In previous work we introduced a model-based workflow to objectively quantify the usefulness of monitoring within the context of conformance verification in CO2 storage under geological uncertainties, to support the design of effective monitoring strategies. Now we investigate the use of convolutional neural networks to render conformance classification more practical and swift within the workflow. The approach was applied to a case study based on a real storage aquifer and showed to be suitable for conformance classification based on time-series pressure measurements and 2D time-lapse images of the CO2 plume. The results obtained indicated that both types of data can, in time, provide sufficient evidence for accurately inferring the chances of future migration of CO2 to undesired areas of the reservoir. These promising results confirm the suitability of machine learning techniques to further improve workflows for quantitative conformance assessment under uncertainties.