A Bayesian Belief Network (BBN) to Predict Bitumen Saturation in the Tengiz Platform-Interior: Implications for Improved Geologic Modeling of Uncertainty
Sean A. Guidry1, Alex Woronow1, Linda Corwin2, Kevin Putney3, and Lyndon Yose1
1 ExxonMobil Upstream Research Company, Houston, TX
2 ExxonMobil Exploration Company, Houston, TX
3 ExxonMobil Development Company, Houston, TX
Bayesian Belief Networks (BBNs) were explored as a means to predict bitumen saturation in Tengiz Field, Kazakhstan, where bitumen has a profound effect on reservoir quality. BBNs are probabilistic networks that make predictions based on expert knowledge, historical data, and causal/correlation relationships. Because of the robust nature of these networks and their ability to function in the face of uncertainty, they appear to be a viable method for reservoir quality predictions. Preliminary results for the platform-interior are encouraging despite two fundamental handicaps: 1) uncertainty in the genetic model for bitumen formation, and 2) limited core plug measurements for calibration. A BBN was designed using coreplug bitumen saturation measurements from eight platform-interior wells at Tengiz. The network incorporated expert knowledge about geological variables recognized to have a relationship with bitumen saturation including: 1) stratigraphic interval, 2) sequence-boundary proximity, and 3) possible porosity. The use of these correlational relationships was necessitated by the complex nature of the problem. The network utilized BBN logic to generate probability distributions, and greatly facilitated exploring/evaluating various genetic models for bitumen, relationships between variables, and implications for reservoir quality. The BBN provides an alternative method of bitumen estimation that readily assesses uncertainty associated with the predictions. Automated population of the Tengiz geologic model with BBN-derived bitumen saturation probability distributions, average bitumen saturation, p50, and standard deviation on a cell by cell basis was relatively straightforward. Incorporation of a probability vector in the geologic model allowed a multitude of realizations to be generated. In summary, BBNs have great potential in terms of improving reservoir quality modeling, predictive diagenesis, alternative scenarios, and incorporating uncertainty into geologic models.