ML Flow Rate Prediction for Wells Without Flow Testing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Collecting multiphase flow data for hydrocarbon production wells often requires expensive equipment that disrupts production, limiting the availability of flow testing.
Innovation Solution
A hybrid machine-learning architecture using k-NN models and an artificial neural network predicts flow rate data by analyzing upstream and downstream pressure and temperature data from multiple wells, enabling real-time flow rate estimation without direct testing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If expensive testing equipment is used to collect multiphase flow data, then measurement precision is improved, but device complexity and loss of time increase
Solution Approach 1:
The patent creates a virtual copy of the physical flow testing equipment through machine learning models. Instead of using expensive physical flow testing equipment that requires well shutdowns, the system uses ML models trained on historical data to replicate and predict flow rate measurements, thereby obtaining accurate flow rate data without the physical testing equipment and associated disruptions
Solution Approach 2:
The patent replaces the mechanical/physical flow testing equipment with a computational machine learning system. The ML models process digital data (pressure, temperature, flow regime information) to predict flow rates, substituting the mechanical measurement process with an information-processing approach that eliminates the need for physical testing equipment and well shutdowns
2Measurement precision
If flow testing operations are performed, then measurement precision is improved, but productivity deteriorates due to well shutdown
Solution Approach 1:
The system creates a virtual replica of flow testing through machine learning predictions. By training models on historical flow test data and well characteristics, the system can predict current flow rates without physically testing the well, allowing the well to remain in continuous production mode while still obtaining accurate flow rate information
Solution Approach 2:
The patent performs preliminary training of machine learning models using historical flow test data and well characteristics before actual prediction is needed. This preliminary action enables the system to make accurate flow rate predictions in real-time without requiring physical flow testing equipment or well shutdowns at the time of measurement
3Loss of time
If limited flow testing times are scheduled, then loss of time is reduced, but measurement precision deteriorates due to insufficient data
Solution Approach 1:
The patent replaces time-consuming physical flow testing with rapid machine learning predictions. The ML models process well data (pressure, temperature, flow regime) and historical information to quickly predict flow rates, obtaining accurate measurements instantaneously without the time required for physical testing operations
Data Source
AI summary
A method may include obtaining acquired pressure data for various wells in a geological region of interest. The method may further include obtaining acquired pressure gradient data for the wells. The acquired pressured gradient data may correspond to a pressure difference based on vertical depth at one or more wells among the wells. The method may further include obtaining acquired temperature data regarding the wells. The method may further include determining predicted pressure data for a well in the geological region of interest using a first machine-learning model and the acquired pressure data. The method may further include determining predicted pressure gradient data for the well using a second machine-learning model and the acquired pressure gradient data. The method may further include determining predicted flow rate data for the well using a third machine-learning model, the predicted pressure data, the predicted pressure gradient data, and the acquired temperature data.