Probabilistic reasoning in intelligent systems: networks of plausible inference
Probabilistic reasoning in intelligent systems: networks of plausible inference
Information Theory, Inference & Learning Algorithms
Information Theory, Inference & Learning Algorithms
| Hi-index | 12.05 |
The exploration for oil and gas requires real-time petrophysical expertise to interpret measurement data acquired in boreholes and to recommend further steps. High time pressure and the far reaching nature of these decisions, as well as the limited opportunity to gain in depth petrophysical experience suggests that a decision support system that can aid the petrophysicist will be very useful. In this paper we describe a Bayesian approach for obtaining compositional estimates that combines expert knowledge with information obtained from measurements. We define a prior model for the compositional volume fractions and observation models for each of the measurement tools. Both prior and observation models are based on domain expertise. These models are combined in a joint probability model. To deal with the nonlinearities in the model, Bayesian inference is implemented by using the hybrid Monte Carlo algorithm. In the system, tool measurement values can entered and the posterior probability distribution of the compositional fractions can be obtained by applying Bayes' rule. Bayesian inference is also used for optimal tool selection, using conditional entropy to select the most informative tool to obtain better estimates of the reservoir. Reliability and consistency of the method is demonstrated by inference on synthetically generated data.