Probabilistic Random Forest for Hydrocarbon Production Uncertainty
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current geostatistical machine learning algorithms fail to account for uncertainty in input variables and regression, leading to inaccurate hydrocarbon production predictions and assessment of prediction uncertainty in subsurface assets.
Innovation Solution
A system and method that incorporate uncertainty in the training and testing phases of machine learning algorithms using probabilistic random forest regression, which accounts for geological and engineering parameter uncertainties, represented by probability density functions, to predict hydrocarbon production and its corresponding uncertainty.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If current geostatistical machine learning algorithms are used for hydrocarbon production prediction, then the prediction process is simple and fast, but the accuracy of production predictions and uncertainty assessment is poor due to failure to account for input variable uncertainty
Solution Approach 1:
The patent transforms the deterministic machine learning approach into a probabilistic framework by changing the fundamental parameter representation. Instead of using fixed input values, the system uses probability density functions to represent input parameters, allowing the algorithm to account for uncertainty in geological and engineering variables while maintaining predictive capability
Solution Approach 2:
The patent introduces probabilistic random forest regression as an intermediary computational framework that bridges the gap between simple deterministic predictions and complex uncertainty analysis. This intermediary approach processes probability distributions through the random forest algorithm to produce both predictions and uncertainty assessments in a unified manner
2Reliability
If uncertainty is incorporated in training and testing phases using probabilistic random forest regression, then the accuracy and reliability of hydrocarbon production predictions improve, but the computational complexity and processing requirements increase
Solution Approach 1:
The patent implements uncertainty propagation selectively through the random forest regression process, focusing computational effort on the critical training and testing phases where uncertainty assessment is most valuable. The approach applies probabilistic processing to the essential predictive stages while avoiding unnecessary computational overhead in other system operations
3Measurement precision
If probability density functions are used to represent geological and engineering parameter uncertainties, then the assessment of prediction uncertainty becomes accurate and robust, but the data processing and model training become more complex
Solution Approach 1:
The patent changes the representation of input parameters from deterministic values to probability density functions, fundamentally altering how geological and engineering parameters are processed. This parameter transformation enables the system to capture and propagate uncertainty through the prediction model, producing more accurate uncertainty assessments despite increased processing complexity
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Methods and systems for predicting hydrocarbon production and production uncertainty are disclosed. Exemplary implementations may: obtain training data, the training data including (i) training production data, (ii) training engineering parameters, and (ill) a training set of geological parameters and corresponding training geological parameter uncertainty values; obtain an initial production model; generate a trained production model by training the initial production model; store the trained production model; obtain a target set of geological parameters and corresponding target geological parameter uncertainty values and target engineering parameters; apply the trained production model to generate a set of production values and corresponding production uncertainty values; generate a representation using visual effects to depict at least a portion of the set of production values and corresponding production uncertainty values as a function of position within the subsurface volume of interest; and display the representation.