Application of Machine Learning in Downhole CO2 Measurement Using Formation Tester

ABSTRACT The reservoir fluid composition measurement can provide great insight for petrophysical interpretation, geochemical analysis, and fluid phase behavior. Laboratory analyses of downhole fluid samples have historically provided much of this required compositional data. The delay in providing this critical information for decision making typically ranges from weeks to months. By the time evaluation occurs, the opportunity to acquire additional samples is lost, and any ambiguities regarding the sampled fluids may not be resolved. Characterizing reservoir fluid in-situ and in real-time by means of wireline formation testers (WFT) can enhance formation evaluation and sampling programs. However, direct single station fluid compositional analysis measurements by WFT have historically been limited to gas components C1-C5 and bulk oil C6+. Downhole quantification of Carbon dioxide (CO2) with formation tester is gaining increasing interest with the different challenges involving monitoring of CO2 content in reservoir fluids. This covers Enhanced oil recovery (EOR) and CCUS monitoring as part of reservoir surveillance among other applications. This paper presents application of downhole fluid analyzer in benchmarking real-time CO2 composition with the utilization of informed machine learning algorithm.