Explainable ML for Subsurface Fluid Type Likelihood
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current hydrocarbon exploration methods, particularly in subsurface reservoirs, face inaccuracies in determining fluid types due to incomplete surface measurements and the lack of explainable AI in modeling processes, leading to inefficiencies and uncertainties in well-drilling decisions.
Innovation Solution
The implementation of an explainable machine learning algorithm that uses pre-stack seismic data to determine the likelihood of fluid types in subsurface reservoirs, providing a list of influencing features and improving the accuracy of hydrocarbon exploration by incorporating direct hydrocarbon indicators, amplitude versus angle, and amplitude versus offset models, while addressing uncertainties associated with coal and residual gas presence.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional surface measurements are used to determine fluid type, then the process is simple and fast, but the accuracy is low due to incomplete measurements and unknown influencing features
Solution Approach 1:
The patent introduces machine learning algorithms as intermediaries between surface measurements and fluid type determination. These algorithms process seismic data and multiple influencing features (lithology, depth, physical properties) to produce accurate fluid type predictions, acting as a mediator that transforms incomplete surface data into reliable subsurface fluid characterization
Solution Approach 2:
The patent replaces traditional mechanical/physical measurement methods with computational machine learning models. Instead of relying solely on direct physical measurements, the system uses AI algorithms to analyze patterns in seismic data and predict fluid types, substituting physical measurement limitations with computational intelligence
2Measurement precision
If machine learning algorithms are used to improve fluid type determination, then accuracy is improved, but the algorithms function as black boxes without explaining the reasons or features influencing the output
Solution Approach 1:
The patent implements feedback mechanisms where the machine learning model not only predicts fluid types but also identifies and returns the influencing features that contributed to each prediction. This feedback loop provides explanatory information about lithology, depth, and other factors, allowing users to understand the reasoning behind predictions and validate results against geological knowledge
Solution Approach 2:
The patent segments the black box algorithm into interpretable components by identifying and highlighting specific influencing features (lithology, seismic depth, physical properties) that contribute to each prediction. This segmentation breaks down the complex model output into discrete, understandable factors that can be individually analyzed and validated
3Measurement precision
If comprehensive features including coal and residual gas are incorporated, then the determination becomes more accurate, but the complexity of analyzing multiple fluid types and features increases
Solution Approach 1:
The patent creates a universal machine learning model that handles multiple fluid types (oil, gas, water, coal, residual gas) and multiple influencing features within a single integrated system. This multi-functional approach eliminates the need for separate analysis procedures for each fluid type, reducing overall complexity while maintaining comprehensive coverage of subsurface conditions
Data Source
AI summary
A system is described for determining a likelihood of a type of fluid in a subterranean reservoir. The system may include a processor and a non-transitory computer-readable medium that includes instructions executable by the processor to cause the processor to perform various operations. The processor may receive pre-stack seismic data having seismically-acquired data elements for geometric locations in a subterranean reservoir. The processor may determine, using the pre-stack seismic data, input features for each geometric location and may execute a trained model on the input features for determining a likelihood of a type of fluid in the subterranean reservoir and for determining a list of features affecting the likelihood. The processor may subsequently output the likelihood and the list of features.