Well Test Physics Model Selection Using Siamese Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Manually identifying and parameterizing physics models for pressure transient analysis (PTA) in well test analysis is a tedious task, and existing machine learning approaches face challenges in accurately selecting and parameterizing suitable models for diverse well/reservoir/boundary combinations.
Innovation Solution
A two-step machine learning approach using Siamese neural networks to identify a class of physics models and estimate model parameters, involving a physics model identification module and a parameter estimation module, to automate the selection and parameterization process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual identification and parameterization of physics models is performed, then accuracy of well test analysis can be achieved, but time consumption and labor intensity increase significantly
Solution Approach 1:
The system pre-processes PTA data by calculating pressure derivatives and organizing data into structured formats before model identification. This preliminary preparation enables the machine learning model to quickly identify appropriate physics models without manual data processing during the analysis phase, resolving the contradiction between accuracy and time consumption.
Solution Approach 2:
A machine learning model acts as an intermediary between raw PTA data and physics model selection. The model automatically identifies appropriate physics models and estimates parameters by learning from training data, eliminating the need for manual model identification while maintaining accuracy. This intermediary system resolves the contradiction by automating the complex decision-making process.
2Extent of automation
If existing machine learning approaches are used for model selection, then automation is improved, but accuracy in selecting suitable models for diverse well/reservoir/boundary combinations deteriorates
Solution Approach 1:
The system transforms PTA data into multiple derived parameters including pressure derivatives, dimensionless time, and flow rate changes. These transformed parameters serve as features for the machine learning model, enabling it to accurately distinguish between different physics models for diverse well/reservoir/boundary combinations while maintaining full automation.
Solution Approach 2:
The approach segments the model selection process into distinct phases: data pre-processing, feature extraction, model identification, and parameter estimation. Each phase is handled by specialized components, with the machine learning model focusing specifically on identifying the appropriate physics model class. This segmentation improves accuracy by dedicating specific computational tasks to specialized modules.
3Adaptability or versatility
If comprehensive physics models are used to cover diverse well/reservoir/boundary combinations, then adaptability improves, but complexity of the analysis process increases
Solution Approach 1:
Instead of attempting to create a single comprehensive physics model that handles all well/reservoir/boundary combinations, the system inverts the approach by using the machine learning model to identify which specific physics model class is appropriate for each dataset. This inversion simplifies the overall system by allowing standard physics models to be selected and applied based on automatic identification rather than requiring a unified complex model.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method involves obtaining query pressure transient analysis (PTA) data from a well associated with a reservoir, and obtaining a selected class of physics models from a multitude of classes of physics models using a first machine learning model operating on the query PTA data. A physics model in at least one of the multitude of classes of physics models includes a well model and a reservoir model. The well model and the reservoir model are parameterized with model parameters having model parameter values. The method further involves obtaining a multitude of model parameter value estimates to form a parameterized query physics model of the selected class of physics models, using a second machine learning model operating on the query PTA data; and providing the parameterized query physics model to a user.