Well Dynamics Digital Twin for Low-Frequency Broadband Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Physics-based models struggle to accurately replicate the low-frequency aspects of broadband and high-resolution well dynamics behavior, limiting the effectiveness of predicting frequency aspects of broadband well dynamics performance in green energy generation systems.
Innovation Solution
A digital twin-based approach is developed to integrate multi-physics, multi-scale, probabilistic simulations of an as-built system, enabled by digital thread, and a digital twin-based approach is developed to integrate multi-physics, multi-scale, probabilistic simulations of an as-built system, using digital thread, best available models, sensor information, and input data to emulate activities and/or performance over the life of a physical system, such as a well system, applying an artificial neural network and Levenberg-Marquardt algorithm in a physics constrained machine learning workflow to predict broadband well dynamics performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a physics-based model is used to predict well dynamics behavior, then the model provides a good degree of accuracy for general dynamics, but it fails to accurately replicate low-frequency aspects of broadband and high-resolution well dynamics behavior
Solution Approach 1:
The patent combines physics-based models with machine learning models to create a hybrid prediction system. The physics-based model captures general dynamics behavior while the machine learning model specifically addresses low-frequency broadband aspects that the physics model cannot accurately replicate. This merging allows the system to maintain the interpretability and physical consistency of physics-based models while gaining the ability to capture complex broadband dynamics through data-driven approaches.
2Productivity
If a lumped-parameter model is used to predict broadband dynamics performance, then the model is computationally efficient, but it limits the effectiveness of predicting frequency aspects of broadband well dynamics
Solution Approach 1:
The patent segments the prediction task into different frequency bands and uses appropriate modeling approaches for each segment. The lumped-parameter model handles high-frequency components where it remains effective, while machine learning models handle low-frequency broadband components. This segmentation allows the system to maintain computational efficiency for high-frequency predictions while improving accuracy for low-frequency broadband aspects through specialized machine learning models.
Data Source
AI summary
A method of managing a well system includes: obtaining, based on a predetermined monitoring criterion, well dynamics behavior data of the well system; obtaining modeled well dynamics behavior data for the well system using a physics-based model; decomposing each of the well dynamics behavior data and the modeled well dynamics behavior data into a plurality of frequency band components based on a plurality of predetermined frequency partitions; training a physics constrained machine learning model using one or more machine learning algorithms based on the plurality of frequency band components of the decomposed well dynamics behavior data and the decomposed modeled well dynamics behavior data as input data; obtaining new well dynamics behavior data of the well system; outputting predicted well dynamics behavior data based on the new well dynamics behavior data using the physics-based model and the trained physics constrained machine learning model.