Slug Forecasting via Machine Learning in Flowlines
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current physics-based models for gas-liquid slugging in oilfield production systems are limited by a lack of understanding of multiphase flow physics, complex scale-up processes, and insufficient field measurements, leading to ineffective capture of actual slugging behavior.
Innovation Solution
A method utilizing machine learning algorithms, specifically a Temporal Fusion Transformer model, to forecast gas-liquid slug flow patterns from real-time and historic field data, enabling accurate prediction of slug frequency, volume, and amplitude, and mitigating actions to reduce negative consequences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If physics-based models are used to understand gas-liquid slugging, then theoretical understanding is provided, but the models fail to capture actual slugging behavior due to lack of understanding of multiphase flow physics and complex scale-up processes
Solution Approach 1:
The patent replaces complex physics-based mechanical models with a data-driven machine learning approach. The system uses historical field measurements and real-time sensor data to train models that predict slugging behavior, substituting theoretical physics calculations with empirical pattern recognition that directly captures actual field behavior without requiring complex multiphase flow physics understanding
Solution Approach 2:
The patent transforms the approach by changing from fixed physics-based parameters to dynamic, data-driven parameters. The system continuously learns from field measurements and adapts its prediction parameters based on actual observed slugging patterns, allowing it to capture real behavior while avoiding the rigidity of theoretical models
2Productivity
If physics-based models are applied in production fields, then flow analysis is provided, but applications are limited due to insufficient tuning from scarce field measurements
Solution Approach 1:
The patent performs preliminary action by extensively training the machine learning models using historical field measurements before deployment. The system accumulates and utilizes available field data in advance to pre-tune and validate models, ensuring they are properly calibrated to actual field conditions before being applied in production environments
Solution Approach 2:
The patent implements continuous feedback mechanisms where real-time field measurements are fed back into the system to validate and refine predictions. The system compares model outputs against actual observed slugging behavior and uses this feedback to continuously improve model accuracy, overcoming the limitation of scarce initial field measurements
3Adaptability or versatility
If empirical models are developed to understand gas-liquid slugging, then theoretical frameworks are created, but actual applicability is hindered by complex scale-up processes
Solution Approach 1:
The patent transitions from theoretical dimensional analysis to practical dimensional prediction by using machine learning models that directly process multi-dimensional field data. The system handles multiple parameters simultaneously (flow rates, pressures, temperatures, compositions) and learns their complex interactions from data, avoiding the simplifying assumptions required in empirical scale-up processes
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
A method for multi-horizon forecasting of gas-liquid slug flow is provided. Field data for a well is obtained. The field data comprises a plurality of features. The plurality of features is correlated across a set of historic data to generate time series data for each of the plurality of features. The time series data is processed by a machine learning model to generate a multi-horizon forecast of a flow pattern for the well, and the output is presented.