AI Well Performance Classification Surrogate Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current well planning methods for hydrocarbon reservoirs are complex and inefficient, especially for large reservoirs or complex geology, as they rely on time-consuming and resource-intensive numerical reservoir simulations, and lack accurate predictive models for well performance.
Innovation Solution
A data processing system that uses probabilistic estimates based on reservoir simulation results, well production rates, and geological properties to classify well performance, incorporating machine learning techniques like artificial neural networks, Bayesian classifiers, and dynamic time warping to predict fluid production rates and pressures, allowing for flexible refinement of input parameters and uncertainty quantification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If numerical reservoir simulation is used to optimize well design and predict well performance, then manufacturing precision of well performance prediction is improved, but productivity deteriorates due to time-consuming and computer resource-intensive simulation runs
Solution Approach 1:
The patent creates a simplified copy or surrogate model of the complex numerical reservoir simulation. This surrogate model captures the essential relationships between well parameters and performance outcomes without requiring full-scale simulation runs, enabling rapid prediction while maintaining acceptable accuracy for well planning decisions
Solution Approach 2:
The patent employs computationally inexpensive models that can be executed quickly and discarded or retrained as needed, replacing the expensive and time-consuming numerical simulations. These simplified models provide sufficient accuracy for preliminary well planning and can be regenerated when new data becomes available
2Productivity
If analytical models are used to estimate well production allocation, then productivity is improved by simplifying the planning process, but manufacturing precision deteriorates when reservoirs are large or geology is complex
Solution Approach 1:
The patent segments the reservoir into distinct regions or zones with similar characteristics, allowing simplified analytical models to be applied to each segment while capturing local geological variations. This segmentation enables the model to handle complex geology without requiring full-scale numerical simulation across the entire reservoir
Solution Approach 2:
The patent dynamically adjusts model parameters based on reservoir characteristics, transitioning between simplified analytical approaches and more sophisticated simulation methods depending on the complexity of the specific reservoir being analyzed. This allows the system to maintain efficiency while adapting precision to match geological complexity
Data Source
AI summary
A heterogeneous classifier based on actual reservoir and well data is developed to qualitatively classify oil well producer performance. Based on the classification a new well is drilled into a producing reservoir, or fluid flows in an existing well are adjusted. The data include perforation interval(s), completion type, and how far or close the perforated zones are located relative to the free water level or gas cap. The data also include geological data, such as major geological bodies like regional faults and fractures. The features may be prioritized before classification. The classifier utilizes four different techniques to apply pattern recognition on reservoir simulation vector data to classify the wells, Three of the classification techniques are supervised learning methods: Bayesian classification, dynamic time warping and neural network. The fourth classification is an unsupervised method, clustering, to automate well grouping into similar categories.