Uncertainty Analysis and History Matching on Grid Responses from a Reduced-basis Approach

Abstract

Numerical models of the reservoirs usually depend on a set of uncertain parameters. To adjust these parameters and build representative models, the dynamic behavior of many reservoir models needs to be investigated, what can be very time consuming. A way to handle this difficulty consists in building proxy models to approximate the responses of interest from a limited number of simulated values. These models, also called response surfaces, can then replace the calls to the fluid-flow simulator. In reservoir engineering, many simulation outputs depend on time or space. However, the response surfaces only approximate scalars, what can be a strong limitation to properly study uncertainties: building a proxy model per time or space takes time and neglects possible correlations. An alternative methodology is thus considered here. It consists in building response surfaces for new variables deduced from a reduced-basis approach, what makes it possible to approximate grid outputs from a limited number of response surfaces. A more systematic uncertainty analysis for grid outputs as well as new tools for history matching can then be envisaged. In this paper, we propose to investigate the potential of the approach for seismic attributes, considering a synthetic Steam Assisted Gravity Drainage (SAGD) produced field.