Hybrid Physics-ML Modeling for Continuous Bottomhole Pressure Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for estimating bottomhole pressure (BHP) in subterranean petroleum reservoirs are costly and lack generalizability, as they rely on empirical or mechanistic models that are not applicable to various flow conditions and require manual tuning, making them unsuitable for continuous measurement across large well counts.
Innovation Solution
A hybrid BHP modeling method combining physics-based preprocessing and regularization with machine learning models to estimate BHP from routine production data, using a two-step approach to determine the best physics correlation and residual correction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If permanent downhole pressure gauges are deployed to continuously measure BHP, then measurement precision and reliability are improved, but device complexity and cost increase significantly
Solution Approach 1:
The patent creates a virtual copy of the downhole pressure measurement system through machine learning models that replicate the function of physical downhole gauges. The ML model takes surface measurements (wellhead pressure, flow rate, temperature) and predicts bottomhole pressure, providing continuous BHP data without deploying expensive downhole sensors to every well.
Solution Approach 2:
The patent replaces the mechanical downhole pressure gauge system with an information-based machine learning prediction system. Instead of using physical sensors mechanically deployed in the wellbore, the system uses computational models trained on production data to substitute for the physical measurement apparatus.
2Device complexity
If physics-based multi-phase flow correlations are used to estimate BHP from surface pressure, then device complexity is reduced, but measurement precision and adaptability deteriorate due to limited applicability to various flow conditions
Solution Approach 1:
The patent changes the approach from using fixed physics-based correlations to a machine learning model that automatically adapts to different flow conditions by learning from historical data. The model captures complex non-linear relationships between surface measurements and bottomhole pressure across various flow regimes, well configurations, and operational conditions without requiring manual selection of physics models.
Solution Approach 2:
The machine learning model provides universal applicability across different well types, flow conditions, and operational scenarios. Unlike physics-based correlations that require manual selection based on specific flow regimes, the ML model handles multiple flow patterns (single-phase, two-phase, gas-liquid, liquid-liquid) automatically through its training data, making it universally applicable without sacrificing precision.
3Measurement precision
If manual tuning and selection of physics-based models is performed, then measurement precision may be improved for specific conditions, but productivity and ease of operation deteriorate due to manual intervention requirements
Solution Approach 1:
The machine learning model performs self-service by automatically selecting the appropriate flow regime and applying the correct prediction logic based on the input data characteristics. The model autonomously handles model selection, parameter estimation, and prediction without requiring manual intervention, making the system self-sufficient and highly efficient for large-scale deployment.
Data Source
AI summary
A method of modeling borehole pressure (BHP) for a wellbore, comprising: receiving field data for the wellbore; training a machine learning (ML) classification model to determine a best physics correlation for BHP in the wellbore using a plurality of physics-based models; determining, using the ML classification model, the best physics correlation based on the field data for the wellbore; determining, using the best physics correlation, a BHP estimate based on the field data for the wellbore; training an ML regression model to determine a residual correction using the BHP estimate and the field data for the wellbore; and determining, using the ML regression model, a final BHP using the BHP estimate and the residual correction for the wellbore.