Fast Proxy Models for Probabilistic Evaluation of Downhole Fluid Sampling Operations

Acquisition of fluid samples using a wireline formation tester is an integral part of characterizing and understanding a reservoir. Recent advances in hardware have enabled wireline-based fluid sampling in a wide range of downhole conditions. The array of available tools includes single, multiple, and focused probes and variable-spacing dual packers. The optimal choice among tools depends on the operating environments, and therefore a quantitative evaluation of hardware performance over a wide range of deployment conditions is necessary. With inherent uncertainties in formation and fluid properties, probabilistic evaluations of sampling time are needed. In this paper, we develop a numerical model of the near-wellbore fully miscible flow during fluid sampling. We explore different types of proxy models and show how high-fidelity proxies are built for the fluid sampling process based on a large number of full-scale flow simulations via a two-step filtering approach. The impact of model parameterization and input space sampling on the proxy accuracy is discussed. Through extensive validation, we demonstrate that the average proxy prediction errors are less than 5% for the whole fluid sampling process and over relevant ranges of formation and fluid properties. We then demonstrate how the proxy models employ global sensitivity analysis to enable rapid quantification of uncertainty in sampling time and pinpoint high-impact parameters that contribute substantially to this uncertainty. Finally, we discuss how the proxy models and uncertainty workflows have been implemented in a tool planner for engineers in the field. This paper demonstrates how large-scale proxy modeling can enable probabilistic decision-making workflows by providing access to otherwise prohibitively expensive reservoir simulation-based analysis.