Evolution of Uncertainty Methods in Decline Curve Analysis

摘要

Decline curve analysis has dominated reserve estimation techniques for decades. Ironically, the introduction and expansion of probabilistic calculation techniques largely by-passed these kind of calculations, but not for lack of effort. Much thought and writing has been dedicated to the unique issue of defining uncertainty in future production trends. Despite two decades of slow evolution, the resulting methods have been partial and prohibitive. Methods like scenario analysis, Delphi forecasting, bootstrapping and Bayesian networks have been proposed and even sometimes utilized. Though they have demonstrated some success, they all suffer from limitations in difficulty and objectivity and aggregation. They have also all utilized smooth model fits and projections. In tandem, other kinds and ways of uncertainty analysis have been applied in the time-domain or dependent sequences using decision trees, Bayesian networks and, more recently, Markov chains. Newly developed machine-learning techniques combine these two separate streams of uncertainty analysis into a framework that promises to address both the quantification and aggregation of uncertainty in an objective way. What seems like a black-box method is instead a natural and automated extension of well-accepted techniques, and the few published studies of the technique suggest that it may be robust, particularly over the scale of years.