Parallel Proxy Model for Oil Reservoir Production Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Oil reservoir numerical simulation is computationally expensive and time-consuming, especially for complex reservoirs with high-dimensional problems, limiting the efficiency of production optimization due to the inability to utilize parallel calculation resources effectively.
Innovation Solution
A parallel proxy model based machine learning method that utilizes sampling points to construct a Kriging proxy model, allowing for parallel evaluation of candidate solutions and reducing optimization time by leveraging multiple computational resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If serial dynamic sampling with proxy model is used for optimization, then computational cost per iteration is reduced, but optimization time increases because only one candidate point can be evaluated per iteration
Solution Approach 1:
The patent divides the candidate points into multiple batches or groups that can be evaluated in parallel. Instead of evaluating one candidate point sequentially, the method segments the evaluation process into multiple parallel streams, allowing simultaneous assessment of multiple candidates while maintaining the proxy model's computational efficiency.
Solution Approach 2:
The patent performs preliminary parallel evaluation of multiple candidate points before selecting the best one for actual simulation. By pre-evaluating multiple candidates in parallel using the proxy model, the method prepares a shortlist of promising candidates that can then be efficiently selected, reducing the need for sequential iterations.
2Measurement precision
If traditional numerical simulation is used for production optimization, then accurate evaluation of production plans is achieved, but computation time becomes excessively long for complex reservoirs requiring thousands of iterations
Solution Approach 1:
The patent creates a proxy model that serves as a computational copy or approximation of the full numerical simulation. This proxy model replicates the essential behavior of the reservoir system but with significantly reduced computational complexity, allowing rapid evaluation of candidate production plans without requiring expensive full-scale simulations for every iteration.
Solution Approach 2:
The patent introduces a proxy model as an intermediary between the optimization algorithm and the actual numerical simulation. This intermediary layer filters and pre-evaluates candidate solutions, allowing the system to quickly eliminate poor candidates before submitting them to the full numerical simulation, thus reducing the total number of expensive simulation runs required.
3Productivity
If existing parallel sampling methods like Constant Liar or Kriging Believer are used, then parallel computation resources can be utilized, but these methods cannot handle high-dimensional problems typical in oilfield production optimization
Solution Approach 1:
The patent transforms the high-dimensional optimization problem into a lower-dimensional search space by applying dimensionality reduction techniques or by projecting the search onto a manifold of lower dimension. This allows parallel sampling methods to effectively explore the solution space despite the original problem's high dimensionality, maintaining both parallel computation benefits and applicability to oilfield problems.
Solution Approach 2:
The patent modifies the parameters or representation of the optimization problem to make it suitable for parallel sampling. This may involve changing how candidate points are generated or represented, adjusting the dimensionality or structure of the search space, or transforming the objective function in a way that enables effective parallel exploration while preserving the essential characteristics of the high-dimensional oilfield optimization problem.
Data Source
AI summary
The present disclosure relates to a parallel proxy model based machine learning method for oil reservoir production. With the proposed method, multiple optimized candidate solutions can be obtained within an iteration, and then a matrix laboratory (e.g., MATLAB) is used to call numerical simulation software Eclipse in parallel to conduct actual evaluation on the candidate solutions simultaneously, so that optimization time of complex problems can be greatly reduced. With the method of the present disclosure, the efficiency of solving an oilfield production optimization problem can be speeded up to a greater extent than in the art, and the final optimization effect can be improved. Moreover, the method of the present disclosure may further be used for well pattern optimization, history matching, and so on, apart from adjusting schedules of the producers and injectors in the oilfield.