Identification of Reservoir Analogues in the Presence of Uncertainty

Abstract

We propose and evaluate a new methodology and tool for the identification of reservoir analogues of a target reservoir, given a database of known reservoirs. In particular, we focus on the common situation where some of the key properties of the target reservoir are unknown. This situation introduces uncertainty into the identification procedure. Therefore, an important step in our methodology is to characterize this uncertainty, so that it can later be considered when estimating production potential and capturing the associated risk.The proposed methodology is composed of a sequence of five steps: Data Preprocessing, Key Parameters Selection, Statistical Learning, Similarity Ranking and Uncertainty Characterization. The first step consists of the analysis and pre-processing of the available database. During Key Parameters Selection, properties with largest impact in the case to be evaluated are identified. In the third step, unknown key properties of the target reservoir are estimated using different statistical learning techniques. In the Ranking step, a similarity function is applied to the database, generating a similarity ranking. Finally, in the Uncertainty Characterization step nonparametric methods are used to estimate probability density/mass functions to characterize the uncertainty in the properties estimated through analogous reservoirs.We perform experiments to assess both the quality of property prediction and the quality of the selected analogues. We report results produced by five different machine learning algorithms using leave-one-out cross-validation. According to our results, Decision Trees was the most effective algorithm for categorical properties, while Regressive Support Vector Machines produced the best results for numerical properties. We also show detailed results for the Casablanca oil field as a target reservoir, which is a mature carbonate reservoir very well known by Repsol.Our methodology allows the identification of analogues for newly discovered reservoirs with limited information. An important additional outcome is the estimation of unknown properties such as recovery factor and production mechanism, and their associated uncertainties, which support the field development plant. IntroductionIn the Oil and Gas E&P Industry, analogous reservoirs are used in many ways to study reservoirs that lack critical information. In this paper, a "target reservoir" is any reservoir with a deficit of critical information, and "analogous reservoirs" are those with the most similar fingerprints to the target reservoir, based on the selected key parameters.

Oil and Gas Reservoirs Parameters Analysis Using Mixed Learning of Bayesian Networks

“…Analogues are also used at mature stages of reservoir development, for example, to find successful cases of increasing oil recovery. There is another way to find reservoir analogues, namely using the similarity function [16]. The difference between the two approaches consists mainly in distributions shapes of reservoir analogues due to more narrow search space made by industry expert [21].…”