Produced Water Oil Estimation Using Physics-Driven Soft Sensors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for monitoring oil content in produced water in oil fields are inadequate, failing to accurately measure large volumes and relying on sampling that does not account for complex interplay of flow regimes and chemical interactions, leading to inefficiencies in crude oil production.
Innovation Solution
A multi-target time-series based hybrid physics and machine learning model is developed to predict oil content in produced water, incorporating feature engineering, preprocessing, natural language processing, and machine learning techniques to generate actionable insights for maintaining crude oil quality within specified thresholds.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional sampling methods are used to measure oil content in produced water, then measurement cost is reduced, but measurement precision and reliability deteriorate due to inability to capture complex interplay of flow regimes and chemical interactions
Solution Approach 1:
The patent introduces soft sensors as intermediary computational models that predict oil content by processing data from multiple existing process sensors. These soft sensors act as mediators between the complex separation process and the measurement need, capturing the interplay of flow regimes and chemical interactions through machine learning models trained on historical data, thereby achieving high measurement precision without direct complex measurement devices
Solution Approach 2:
The machine learning framework integrates multiple functions into a unified system: it processes data from various process sensors, performs predictions of oil content, identifies root causes of poor separation, and provides actionable insights. This multi-functional approach eliminates the need for separate specialized measurement devices for each function, reducing overall device complexity while maintaining high measurement precision
2Productivity
If conventional monitoring methods are used, then operational simplicity is maintained, but productivity deteriorates due to inability to provide real-time predictive insights
Solution Approach 1:
The system performs preliminary actions by training machine learning models on historical data to establish predictive relationships before real-time operation. The soft sensors are pre-trained to capture the complex interplay of process variables, enabling real-time predictions without requiring complex real-time computations during operation. This preliminary training phase enhances productivity by providing immediate predictive insights while keeping the real-time system relatively simple
Solution Approach 2:
The system implements feedback mechanisms where prediction results and root cause analyses are continuously fed back to operators and control systems. This feedback loop enables proactive adjustments to maintain optimal separation performance, enhancing productivity. The feedback is delivered through a user-friendly interface that translates complex model outputs into actionable insights, maintaining operational simplicity despite the sophisticated underlying models
3Measurement precision
If comprehensive data analysis is performed to capture complex interplay of process variables, then measurement precision improves, but loss of time in data processing increases
Solution Approach 1:
The machine learning models perform comprehensive data analysis in advance during the training phase, capturing the complex interplay of flow regimes, chemicals, and emulsions. Once trained, the models can make rapid predictions during real-time operation without requiring extensive computation for each new measurement. This preliminary action transfers the computational burden from real-time operation to the training phase, maintaining both high precision and fast response times
Solution Approach 2:
The soft sensors create computational copies of the physical separation process by training machine learning models on historical process data. These digital twins or copies replicate the complex relationships between process variables and oil content, enabling rapid predictions without requiring real-time analysis of all underlying physical phenomena. The copying approach maintains measurement precision while dramatically reducing data processing time during operation
Data Source
AI summary
A method to perform oil in produced water analysis allows measuring the large volume of oil in produced water reliably. In the method, a time-series and physics based machine learning model of a gas oil separation plant is generated, advisory actionable items for maintaining a crude oil quality within a pre-determined threshold are generated based on machine learning model coefficients and outputs of soft sensors, and then the advisory actionable items are presented to a user.