Hydrocarbon Digital Twin Calibration for Unmeasurable Variable Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current hydrocarbon system modeling and control methods lack real-time analysis and efficient control capabilities, particularly in predicting and adjusting variables such as gas oil ratio, water cut, and pump leakage, which are difficult to measure directly.
Innovation Solution
The method involves generating a digital twin of the hydrocarbon system using reduced order models (ROMs) instantiated at an operational point, configured to use real-time data, and employing regression or machine learning techniques to estimate and control system variables, including forward, inverse, and calibration ROMs for predictive and corrective actions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional hydrocarbon system modeling methods are used, then system behavior can be analyzed, but real-time prediction and control capabilities are insufficient
Solution Approach 1:
The patent creates a digital twin (a virtual copy) of the hydrocarbon system that mirrors the physical system's behavior. This digital replica is built using reduced-order models derived from comprehensive simulations, enabling real-time prediction and control without requiring complex computational resources of the full-order models.
Solution Approach 2:
The patent replaces traditional mechanical/computational modeling approaches with machine learning techniques. Neural networks and regression algorithms are trained on simulation data to create reduced-order models that can predict system behavior instantaneously, substituting heavy computational mechanics with efficient statistical and ML-based predictions.
2Measurement precision
If comprehensive simulations are performed to capture system behavior, then model accuracy is improved, but computational complexity and time requirements increase
Solution Approach 1:
The patent extracts the essential behavioral characteristics of the hydrocarbon system from comprehensive simulations and encapsulates them in reduced-order models. By separating the critical input-output relationships from the full system complexity, the ROMs achieve sufficient accuracy for real-time applications without requiring the computational resources of complete simulations.
Solution Approach 2:
The patent transforms the system representation by changing parameters from detailed physical models to statistical and machine learning parameters. Regression coefficients and neural network weights are derived from simulation data, converting complex physical parameter relationships into simplified mathematical representations that maintain accuracy while reducing computational burden.
3Measurement precision
If direct measurement of system variables is performed, then accurate data is obtained, but variables such as gas oil ratio, water cut, and pump leakage are difficult to measure directly
Solution Approach 1:
The patent introduces the digital twin as an intermediary that indirectly estimates difficult-to-measure variables. Instead of attempting direct measurement of gas oil ratio, water cut, and pump leakage, the system uses readily available sensor data as inputs to the reduced-order models, which then compute the unmeasurable variables through learned relationships from comprehensive simulations.
Solution Approach 2:
The patent replaces physical measurement instruments with virtual sensing through machine learning models. Neural networks and regression algorithms substitute for physical sensors that would be required to directly measure difficult variables, using patterns learned from simulation data to infer unmeasurable quantities from easily measurable inputs.
Data Source
AI summary
Systems and methods generate and use a digital twin in a hydrocarbon system. The systems and methods can perform the following operations: (1) performing a plurality of simulations in a hyperdimensional space to generate outputs; (2) using the outputs of the plurality of simulations to generate one or more reduced order models (ROMs) using a regression technique or a machine learning technique; (3) generating a digital twin of the hydrocarbon system by instantiating the one or more ROMs at an operational point of the hydrocarbon system, and configuring the digital twin to use real-time data obtained from the hydrocarbon system; and (4) estimating values of one or more variables of the hydrocarbon system in real-time using the digital twin.