General Bayesian Network for Oilfield Hidden Variable Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Oilfield operations face challenges in interpreting data with uncertainties and hidden variables, which can lead to inadequate decision-making and equipment malfunctions, especially due to the difficulty in estimating values for continuous-valued hidden variables using traditional Bayesian Networks.
Innovation Solution
A method employing General Bayesian Networks with a Monte-Carlo approach to model both discrete-valued and continuous-valued variables, where sample vectors are weighted based on observed data to accurately predict non-observed variables, allowing for automated actions and improved operational control.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional Bayesian Networks are used to model oilfield operations, then discrete variables can be effectively analyzed, but continuous-valued hidden variables cannot be accurately estimated
Solution Approach 1:
The patent transforms the Bayesian Network framework by changing the parameter representation from discrete probability distributions to continuous probability density functions. This allows the network to handle continuous-valued variables while maintaining the probabilistic inference capabilities, directly resolving the contradiction between measurement precision and adaptability.
Solution Approach 2:
The generalized Bayesian Network framework unifies the treatment of both discrete and continuous variables within a single probabilistic model. This universal approach enables the system to handle mixed variable types (discrete observed, continuous hidden, and continuous observed) simultaneously, making the network versatile for complex oilfield operations while maintaining accurate estimation capabilities.
2Loss of information
If more sensors are deployed to monitor all variables, then complete information can be obtained, but system complexity and cost increase significantly
Solution Approach 1:
The patent introduces a probabilistic inference mechanism as an intermediary that bridges the gap between incomplete sensor data and complete operational understanding. The Bayesian Network acts as a mediator that mathematically infers hidden variables from observed variables, eliminating the need for direct sensors on all variables while maintaining information completeness.
Solution Approach 2:
The patent replaces the mechanical approach of deploying physical sensors for every variable with a computational inference system. Instead of using more physical measurement devices, the system uses probabilistic algorithms to calculate hidden variables from existing sensor data, substituting mechanical sensing with mathematical reasoning.
3Reliability
If probabilistic models are used to account for data uncertainties, then measurement reliability improves, but computational complexity increases
Solution Approach 1:
The patent changes the computational parameters from discrete probability tables to continuous probability density functions with parametric forms. This transformation maintains the reliability benefits of probabilistic modeling while reducing computational complexity by using compact parametric representations instead of exhaustive probability tables.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Methods for using General Bayesian Networks to automate oilfield operations. In certain aspects, a Monte-Carlo method is used to propagate probability density functions for root- variables to continuous-valued hidden variables reflecting some oilfield operation properties. Evidence in the form of observed properties are used to weight samples used in the Monte-Carlo process thereby propagating the observed values onto other variables. The inferred probability distributions are provided to an oilfield control system or monitoring system.