Oil and Gas Network Modeling Using Combined Long-Term Short-Term Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for modeling oil and gas flow networks struggle to accurately predict responses to changes in control points without iterative updates and are inefficient in handling the 'reservoir effect', which complicates long-term data analysis.
Innovation Solution
A combined modeling approach using a long-term model and a short-term model, where the long-term model accounts for reservoir effects over a longer period, and the short-term model predicts immediate changes, allowing for real-time estimation and prediction of flow rates by disregarding the reservoir effect during shorter time periods.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single long-term model is used to capture reservoir effects, then the model can account for long-term variations, but it cannot accurately predict short-term responses to control changes
Solution Approach 1:
The patent divides the modeling task into two separate models: a long-term model that captures reservoir effects and slow variations, and a short-term model that captures immediate responses to control changes. This segmentation allows each model to be optimized for its specific time scale, resolving the contradiction between long-term accuracy and short-term responsiveness.
Solution Approach 2:
The patent introduces a time-scale dimension by creating models operating at different temporal resolutions. The long-term model operates on extended time scales to capture reservoir dynamics, while the short-term model operates on immediate time scales to capture control responses, allowing the system to adapt to different time horizons.
2Measurement precision
If iterative updates are performed to improve model accuracy, then prediction accuracy improves, but computational efficiency decreases
Solution Approach 1:
By segmenting the modeling into long-term and short-term components, the patent reduces the need for iterative updates. The long-term model captures slow reservoir variations that don't require frequent updates, while the short-term model captures immediate control responses that can be predicted without iteration, thereby improving computational efficiency.
Solution Approach 2:
The patent performs preliminary modeling by separating the long-term reservoir effects from short-term control responses in advance. This preliminary action allows the system to make accurate predictions without requiring iterative updates during operation, as each model component is designed to capture specific aspects of system behavior that can be predicted directly.
3Reliability
If comprehensive data analysis is performed to capture all network behaviors, then model completeness improves, but data processing complexity increases
Solution Approach 1:
The patent segments the data analysis process by creating separate processing streams for long-term reservoir data and short-term control data. This segmentation allows each model to focus on specific types of data and behaviors, reducing the overall complexity of data processing while maintaining comprehensive coverage of network behaviors.
Solution Approach 2:
The patent extracts and separates the reservoir effect component from the overall system behavior, handling it in the long-term model. This extraction allows the short-term model to focus solely on control-related behaviors without being burdened by the complexity of reservoir dynamics, thereby reducing data processing complexity while maintaining model completeness.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
An oil and gas network comprises multiple branched flow paths, such as in multi-zonal wells and/or multibranched wells and/or networks including multiple wells, and multiple control points (20, 32) at different branches, wherein the multiple control points (20, 32) may include multiple valves (20) and/or pumps (32) for controlling the flow rate through respective flow paths of the multiple branched flow network. The network is modelled to model the variation of one or more flow parameter(s) (12) in one or more flow path(s) of the network. A method for this modelling includes: generating a long-term model (8) using a first set of data (10, 12) relating to measurements of the flow parameter(s) and the status of the control points (20, 32) over a first period of time, wherein the long-term model (8) describes the relationship between flow rates, the status (10) of control points (20, 32), and measured flow parameters (12) including pressure and/or temperature; generating a short-term model (16) using a second set of data relating to measurements of the flow parameter(s) (12) and the status (10) of the control points (20, 32) over at least one second period of time, wherein the at least one second period of time is shorter than the first period of time, and wherein the short-term model 16 describes the relationship between the status (10) of control points (20) and flow parameters (12) including pressure and/or temperature; and combining the short-term model (16) with the long-term model (8) by: using the short-term model (16) to determine pressure and/or temperature values (12') that will result from the status (10) of one or more control points (20, 32) or from proposed changes to those control points (20, 32); using the determined pressure and/or temperature values (12') from the short-term model (16) along with the status of, or the proposed changes to, the control points (20, 32) as inputs (10, 12') to the long-term model (8) and then using the long-term model (8) to determine flow rate values that will result from those inputs; and thereby obtaining a combined model (16, 8) allowing for estimation of flow rates in real time as well as prediction of the effects of changes in the status (10) of one or more of the control points (20, 32). A method for training a model of this oil and gas network includes: modelling one or more flow parameter(s) (12) in one or more flow path(s) of the network, the modelling including: generating a model (8) using data relating to measurements of the flow parameter(s) (12) and the status of the control points (20, 32) over a period of time; wherein the model describes the relationship between flow rates (14), the status of control points (20, 32), and measured flow parameters (12) including pressure and/or temperature; and wherein generating the model (8) includes training the model (8) under constraints requiring: (i) that the sum of the modelled flow rates from each branch of the flow network that contribute to a combined flow after branched flow paths join at one or more nodes must be equivalent to the respective measured combined flow rate, where a measurement of the combined flow is available, and (ii) that training of the model (8) is suspended or modified for certain flow paths when the status of the control points (20, 32) is such that those flow paths will have zero flow, and/or that a flow rate for an individual flow path or branch must be or is encouraged to zero when an associated valve (20) is closed and/or if a pump (32) required for non-zero flow rate is inactive.