Parallel Production Simulation for Mature Hydrocarbon Fields
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Mature hydrocarbon fields present a challenge in forecasting production quantities due to the large Vapnik-Chervonenkis dimension of existing modeling approaches, which leads to unreliable forecasts and inefficient resource allocation, as current methods either overcomplicate the models with too many parameters or oversimplify the physics and geology of the reservoir.
Innovation Solution
A production simulator is developed that balances complexity and simplicity by using Vapnik statistical learning theory to ensure accurate forecasting, matching history data, and verifying the Vapnik condition, allowing for reliable simulation of production scenarios and optimal parameter selection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If meshed models with more than 100,000 elements are used to model reservoir behavior, then the model can capture complex physics and geology interactions, but the VC-dimension becomes very large leading to unreliable forecasts and high computational complexity
Solution Approach 1:
The patent segments the reservoir model into discrete grid cells with specific properties (permeability, porosity, saturation) and applies physical laws to each cell. This segmentation allows the model to capture complex local interactions while maintaining a manageable overall structure that can be calibrated with available history data, resolving the contradiction between model detail and forecast reliability.
Solution Approach 2:
The patent transforms the high-dimensional meshed model into a lower-dimensional representation by changing key parameters to be calibrated from history data. This parameter transformation reduces the VC-dimension while preserving the essential physics, enabling reliable forecasts without requiring excessive model complexity.
2Device complexity
If over-simplified models like decline curves are used, then the model complexity is reduced, but the model cannot properly capture relevant physics and geology leading to large empirical risk
Solution Approach 1:
The patent introduces an intermediary calibration layer between the simple model structure and the complex physical reality. By calibrating model parameters against history data, the simple model structure mediates between computational simplicity and physical accuracy, capturing essential reservoir behavior without requiring complex meshed models.
Solution Approach 2:
The patent creates a simplified copy of the reservoir system that replicates key historical production patterns through parameter calibration. This copied model maintains the essential physics and geology relationships while using a computationally simple structure, achieving reliable forecasts without high model complexity.
3Measurement precision
If complex meshed models are used to achieve good history match, then the model can fit past data well, but there are billions of ways to match the past leaving large uncertainty on forecast accuracy
Solution Approach 1:
The patent performs preliminary calibration of model parameters against history data before making forecasts. This preliminary action constrains the solution space to models that accurately reproduce past behavior, eliminating billions of spurious matches and reducing forecast uncertainty while maintaining history match accuracy.
Solution Approach 2:
The patent uses history data as feedback to calibrate and constrain model parameters. This feedback mechanism ensures that only models consistent with observed past production are considered, reducing the number of valid solutions from billions to a manageable set that provides reliable forecasts with quantified uncertainty.
Data Source
Figure 1
Figure 2~3
Figure 4~5
AI summary
Embodiments for parallel production simulation of an oil field are disclosed. A plurality of production parameter data sets corresponding respectively to a plurality of scenarios are received and aligned in a memory of a vectorized computation hardware. In SIMT-based architecture, each well of each scenario is associated with a thread of a plurality of threads. In a SIMD-based architecture, each well of each scenario is associated with a respective lane of a multi-lane data pipeline.