Neural Network Proxy for Geological Property Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Storing secondary data in subsurface formation evaluation is costly due to the large number of properties and the need for data storage across a geological formation, which can be mitigated by using a trained neural network to estimate property values instead of directly storing the data.
Innovation Solution
A trained neural network is initialized with prescribed architecture and trained using secondary data from known locations, allowing it to estimate property values at both known and unknown locations, thereby reducing memory load by deleting the secondary data and storing the neural network representation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If secondary data is stored directly in memory for subsurface formation evaluation, then accurate property modeling is enabled, but memory requirements and storage costs increase significantly
Solution Approach 1:
The patent creates a neural network model that copies the essential patterns and relationships from the original secondary data. Instead of storing the actual data values across all locations, the system stores the trained neural network weights and architecture, which replicate the data's informative content when applied to new locations. This copying approach preserves modeling accuracy while dramatically reducing storage requirements.
Solution Approach 2:
The patent extracts the critical information from the secondary data by training a neural network to capture the underlying patterns and relationships. The neural network learns to map primary data to secondary data properties, effectively separating the essential predictive information from the redundant data storage requirements. This extraction allows the system to retain accurate property estimates without storing the full secondary data dataset.
2Adaptability or versatility
If secondary data is stored for all locations in the geological formation, then complete property information is available, but storage costs increase due to the large number of properties
Solution Approach 1:
The trained neural network serves multiple functions: it can estimate secondary data at any location within the geological formation, handle multiple different property types simultaneously, and adapt to different spatial configurations. This universal model replaces the need for separate storage of secondary data for each location and property, providing versatile property estimation capability while eliminating redundant storage costs.
Solution Approach 2:
The patent transforms the storage approach by changing from storing raw data values to storing neural network parameters (weights and biases). This parameter transformation allows the system to represent complex property relationships with a compact set of numerical parameters, enabling accurate property estimation across all locations and property types without the storage costs associated with storing complete secondary data datasets.
Data Source
AI summary
A neural network trainer trains neural networks to estimate secondary data at locations throughout a geological formation where secondary data is unknown. The neural networks are trained to estimate secondary data using locations in the geological formation as input. Subsequently, the secondary data is deleted from memory using the trained neural network as a proxy representation to reduce memory footprint and allow for estimation of secondary data at locations where it is unknown.