Application of Bayesian Networks for Predicting the Performance of Gel-Treated Wells in the Arbuckle Formation, Kansas

Abstract

AbstractPolymer-based gels have been used extensively to reduce water production and improve oil productivity in the Arbuckle formation in central Kansas. One of the most important steps required for successful gel-polymer treatments is to identify candidate wells. Traditionally, the selection of wells to be treated has been based on subjective screening guidelines which may result in inconsistent treatment outcomes and poor economic performance. The objective of this study was to develop a method to identify candidate wells in the Arbuckle for gel-polymer treatments using individual well data readily available to an operator.The approach was to use Bayesian networks applied to a data set derived from a number of gel-treated wells in the Arbuckle. Bayesian networks (Belief Networks) are probabilistic graphical models that represent a set of variables and their probabilistic dependencies. For this study two different Bayesian networks, Naïve and Augmented, were trained using pre-and post-treatment well data from 45 wells. The trained networks were then used with pre-treatment data from 14 additional wells to predict the post-treatment well performance economics of those wells. Well performance was evaluated using a discounted cash flow and net present value (NPV) economic analysis.The results indicate that both the Naïve and Augmented Bayesian networks can accurately predict post-treatment well performance economics. For the 14 wells for which predictions were made, the average accuracy in NPV prediction was 77% and 84% respectively for the two Bayesian methods used. The analysis identified completion thickness from the top of Arbuckle and water-oil ratio as the most important parameters for the success of gel treatments in the Arbuckle formation.Compared to other methods of candidate-well selection, which either rely on complicated reservoir models or unreliable anecdotal screening guidelines, Bayesian networks are relatively easy to implement and can accurately predict the post-treatment well performance economics. Results from this study demonstrated that Bayesian networks can be a very powerful tool for candidate-well selection for gel treatments in the Arbuckle formation. The same concept can be applied in other areas where enough data are available to train the Bayesian networks.