History Matching With Production Uncertainty Eases Transition Into Prediction

Abstract

Abstract The process of history-matching past reservoir performance involves making reasonable adjustments to key properties in a base geological model, either by ‘trial-and-error’ or using some computer-assisted approach, until observed data are adequately reproduced. Traditionally wells are controlled as individual entities and offtake rates that correspond to their historical allocations are imposed. This approach inherently assumes that there is no uncertainty in allocated well rates, and whilst a good match may be obtained, there is no guarantee that the resultant model will be robust in prediction. It is frequently the case that a hiatus occurs when moving from well-rate controlled history-match mode to Group rate and well pressure (i.e. THP) controlled prediction, caused by wells exhibiting too great a potential – well PI's are typically adjusted to ‘correct’ this problem. This correction step is (a) time consuming (b) downgrades confidence in the relevance of the ‘match’, (c) reduces the predictive capability of the model, especially with regard to infill or new well production forecasting and (d) is different for each history match. In this paper, a methodology is presented and illustrated whereby both static (geological) and dynamic (production allocation) uncertainties are addressed simultaneously during a computer-assisted search for multiple history-matches, which honour observed production performance and move into prediction without ‘correction’. Use of a Bayesian response-surface tool assists both in obtaining multiple matches and may be used to calculate statistically valid confidence intervals around predicted performance.