Reservoir Digital Twin via Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional methods for hydrocarbon recovery from oil wells face inefficiencies due to inadequate data logging and changes in hydrocarbon well dynamics, leading to inaccurate predictions of key reservoir parameters.
Innovation Solution
A computer system uses machine learning algorithms trained with normalized reservoir logs, including porosity, petrophysical data, pressure transient test results, and production/injection performance to generate a digital twin of the hydrocarbon reservoir, predicting variations in pressure and saturation over time, enabling more accurate and efficient production strategies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional numerical simulation methods are used for predicting reservoir parameters, then the process follows established conventional approaches, but the prediction efficiency and accuracy deteriorate due to inadequate data logging and changes in well dynamics
Solution Approach 1:
The patent replaces traditional numerical simulation methods with machine learning algorithms. The machine learning model is trained on historical reservoir data including pressure logs, saturation logs, production data, and injection data to predict reservoir parameters such as pressure, saturation, and permeability variations over time, achieving both high accuracy and efficiency without relying on conventional numerical simulation
Solution Approach 2:
The patent transforms the prediction approach by changing from physics-based numerical simulation to data-driven machine learning parameter estimation. The system uses trained machine learning models to directly predict reservoir parameters (pressure, saturation, permeability) based on input features from well logs and production data, fundamentally changing the methodology from traditional simulation
2Reliability
If traditional numerical simulation methods are used, then conventional processes are maintained, but the results become inefficient and less accurate for modern reservoir conditions
Solution Approach 1:
The patent substitutes complex numerical simulation systems with machine learning algorithms that have been trained on historical data. The machine learning approach simplifies the prediction process while improving reliability by learning patterns from actual reservoir behavior data, making the system more adaptable to changing well dynamics and data logging conditions
3Measurement precision
If machine learning algorithms are trained with comprehensive reservoir data including normalized saturation logs, then prediction accuracy improves, but data processing complexity increases
Solution Approach 1:
The patent applies preliminary normalization to saturation logs before training the machine learning model. The system normalizes saturation logs in accordance with time, creating pre-processed training data that improves model accuracy. This preliminary data preparation step simplifies the overall processing by organizing data in an optimal format for machine learning training
Solution Approach 2:
The patent segments the comprehensive reservoir data into distinct categories for targeted processing: pressure logs, saturation logs, production data, injection data, and well log data. Each data type is processed and normalized appropriately before being fed into the machine learning model, making the complex data processing task more manageable and effective
Data Source
AI summary
Methods for training machine learning algorithms for generation of a reservoir digital twin include receiving information obtained from hydrocarbon wells. The information includes porosity logs, petrophysical data, rock typing data, pressure transient test results, vertical production logs, reservoir pressure logs, reservoir saturation logs, production performance, and injection performance. The reservoir saturation logs are normalized in accordance with time. A machine learning algorithm is trained using the reservoir pressure logs, the production performance, and the injection performance to provide variations in reservoir pressure of the hydrocarbon reservoir in accordance with time. The machine learning algorithm is trained to provide variations in reservoir saturation of the hydrocarbon reservoir in accordance with time. The machine learning algorithm is trained using the reservoir saturation logs, the vertical production logs, the production performance, the injection performance, the reservoir pressure logs, and the petrophysical data.