Seismic Attribute Analysis Using Synthetic Pseudo-Well Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for training machine learning systems to analyze seismic attributes and predict reservoir properties are time-consuming and costly, requiring physical surveys and measurements, and struggle with interpreting complex multidimensional FAVO responses.
Innovation Solution
The use of simulated reservoir property profiles and seismic attributes for pseudo-wells, combined with deep learning models like CNNs and LSTMs, to learn correlations between seismic attributes and reservoir properties, allowing for more accurate and efficient predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If physical surveys and measurements are used to train machine learning systems, then the training data is accurate, but the process is time-consuming and costly
Solution Approach 1:
The patent creates synthetic seismic data that replicates the characteristics of actual seismic data from physical surveys. This synthetic data is generated through computer simulations that model seismic wave propagation through geological formations, allowing the machine learning system to be trained on realistic data without requiring actual physical surveys. The synthetic data includes FAVO responses, seismic attributes, and associated reservoir properties, providing accurate training data that would otherwise require time-consuming field work.
2Ease of operation
If conventional methods are used to analyze seismic data, then the process is simple, but the interpretation of complex multidimensional FAVO responses is difficult
Solution Approach 1:
The patent introduces machine learning models as an intermediary between the complex seismic data and the desired reservoir property predictions. The model receives multidimensional FAVO responses and seismic attributes as input, processes them through learned patterns and relationships, and outputs predictions for reservoir properties such as fluid content, porosity, density, and seismic velocity. This intermediary approach allows the system to handle the complexity of multidimensional data while providing simple, actionable predictions.
3Productivity
If machine learning models are trained with limited data, then the training process is faster, but the prediction accuracy is reduced
Solution Approach 1:
The patent generates large volumes of synthetic training data through computer simulations, effectively copying and replicating seismic data characteristics. This allows the machine learning model to be trained on extensive datasets that would be impossible to obtain through physical surveys alone. The synthetic data includes varied geological conditions, rock types, fluid saturations, and seismic parameters, providing the model with comprehensive training examples that improve prediction accuracy without requiring proportional increases in physical survey time.
Data Source
Figure 1~2
Figure 3~4
Figure 5
AI summary
Systems and methods are disclosed that include generating reservoir property profiles corresponding to reservoir properties for pseudo wells based on reservoir data, generating seismic attributes for the pseudo wells, and training a machine learning model by comparing the reservoir property profiles against the seismic attributes. In this manner, the machine learning model may be used to predict reservoir properties for use with seismic exploration above a region of a subsurface that contains structural or stratigraphic features conducive to a presence, migration, or accumulation of hydrocarbons.