Machine Learning Reservoir Model for Pressure Prediction
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 machine learning algorithm is trained using data from multiple hydrocarbon wells to generate a virtual 3D model of the reservoir, predicting changes in reservoir pressure and saturation over time, independent of numerical simulation models, and providing a more accurate digital twin for enhanced production and injection strategies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional numerical simulation methods are used for reservoir modeling, then the process can be performed with established tools, but the prediction accuracy of key hydrocarbon reservoir parameters deteriorates due to inadequate data logging and changes in well dynamics
Solution Approach 1:
The patent replaces traditional numerical simulation methods with a machine learning-based approach. Specifically, it uses a convolutional neural network (CNN) model trained on well log data to predict reservoir parameters, substituting the conventional physics-based numerical simulation with an data-driven intelligent system that can adapt to changing well dynamics and inadequate data conditions
Solution Approach 2:
The patent transforms the reservoir modeling approach by changing from fixed numerical simulation parameters to dynamic machine learning model parameters. The CNN model learns optimal parameters from training data, allowing the system to adapt to varying reservoir conditions and data quality, thereby improving prediction accuracy when traditional methods fail
2Measurement precision
If data logging is increased to improve reservoir parameter predictions, then prediction accuracy improves, but the cost and complexity of data collection increases
Solution Approach 1:
The patent creates a virtual copy of the reservoir through a digital twin model. The CNN-based machine learning model generates a virtual representation of the reservoir that replicates the behavior and properties of the actual reservoir, allowing accurate predictions without requiring extensive physical data collection or complex measurement systems
Solution Approach 2:
The patent introduces a machine learning model as an intermediary between the available well log data and the reservoir parameter predictions. This intermediary layer processes and interprets the limited data, extracting meaningful patterns and predictions without requiring direct measurement of all reservoir parameters, thereby reducing the need for complex data collection systems
Data Source
AI summary
A computer system generates a virtual three-dimensional (3D) model of a hydrocarbon reservoir using a machine learning algorithm. The machine learning algorithm is trained using information obtained from multiple hydrocarbon wells. The virtual 3D model includes a reservoir pressure model of the hydrocarbon reservoir indicating variations in reservoir pressure in accordance with time. A fluid saturation model of the hydrocarbon reservoir indicates variations in reservoir saturation in accordance with time. The computer system executes the machine learning algorithm to determine the variations in the reservoir pressure and the variations in the reservoir saturation with respect to the multiple hydrocarbon wells based on the virtual 3D model. A display device of the computer system generates a graphical representation of the variations in the reservoir pressure and the variations in the reservoir saturation in accordance with time.