Subsurface Proxy Modeling for Faster Unconventional Well Planning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Unconventional hydrocarbon extraction faces challenges in accurately modeling subsurface geologic structures and optimizing well performance due to uncertainties in geology and hydraulic fracture geometry, leading to long cycle times in development planning.
Innovation Solution
A computer-implemented method using machine learning to generate inverse proxy models based on subsurface process data, which iteratively selects and refines a subset of previously generated subsurface models to optimize operations, leveraging a physics simulator and machine learning to accelerate development planning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional subsurface modeling methods are used to accurately represent geologic structures and hydraulic fracture geometry, then model accuracy is improved, but development planning cycle time increases significantly
Solution Approach 1:
The system pre-generates a comprehensive library of subsurface models with varying geologic and completion parameters before actual development planning begins. This preliminary model library is created offline and stored for rapid retrieval during planning, eliminating the need to generate models from scratch during time-critical planning phases.
Solution Approach 2:
Instead of creating new detailed subsurface models during planning, the system uses machine learning to generate simplified proxy models that replicate the behavior of complex pre-generated models. These proxy models capture essential subsurface responses without requiring full computational model regeneration, enabling rapid evaluation of multiple scenarios.
2Reliability
If comprehensive subsurface models with detailed geologic and completion parameters are created, then well performance prediction accuracy is improved, but computational complexity and processing time increase
Solution Approach 1:
The system introduces machine learning proxy models as intermediaries between complex physics-based subsurface models and development planning applications. These proxy models are trained on comprehensive model data to capture complex geologic and completion parameter relationships, then used for rapid prediction without requiring full computational model execution.
Solution Approach 2:
The system transforms complex subsurface modeling problems by changing parameters from continuous physical variables to discrete model library categories. By organizing models according to key geologic and completion parameters, the system enables efficient parameter-based retrieval and comparison without requiring full computational analysis of each scenario.
3Reliability
If multiple subsurface scenarios are evaluated to account for geologic uncertainties, then decision reliability is improved, but the number of models to analyze increases, extending planning time
Solution Approach 1:
The system pre-organizes a diverse library of subsurface models representing various geologic scenarios and completion configurations before planning begins. This pre-categorized model library allows planners to quickly select and evaluate appropriate scenarios based on specific field conditions without generating or analyzing models during the planning process.
Solution Approach 2:
The system evaluates a curated subset of pre-generated models that are most relevant to the specific planning scenario rather than analyzing all possible models. By using machine learning to identify and select the most pertinent models from the comprehensive library, the system achieves sufficient decision reliability without requiring exhaustive analysis of every possible scenario.
Data Source
AI summary
Methods for analyzing subsurface process data in order to perform one or more subsurface operations in a subsurface are provided. Generating subsurface models is typically a long and laborious process in which subsurface process data is analyzed in order to generate the subsurface models. In contrast, work in generating the subsurface models may be front-loaded by first using a physics simulator in order to generate a training set of subsurface forward models, and then performing machine learning using the training set to generate one or more proxy models, such as a forward proxy model and an inverse proxy model. The machine learning may be constrained using physics-based rules to better converge on the proxy models. In this way, the already-trained inverse proxy model may input the subsurface process data in order to generate potential inverse models, which may then be used to perform subsurface operations in the subsurface.