Multi-Scale Geological Property Modeling via Deep Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing geological property models face challenges in integrating data from different scales, leading to loss of information at higher resolutions and low accuracy at lower resolutions, limiting their scalability and accuracy in predicting rock properties.
Innovation Solution
A scalable geological property model using machine learning algorithms that integrates rock measurement data at multiple scales by training deep neural networks with data samples featuring coordinates, measurement data, and scales, allowing for flexible and accurate prediction of rock properties at selectable scales.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If data is captured at higher resolution, then measurement precision is improved, but data loss occurs when modeled at coarser scales
Solution Approach 1:
The patent implements a multi-scale nested modeling framework where fine-scale geological models are embedded within coarser-scale models. High-resolution measurement data is captured at detailed scales and then nested into progressively coarser scales, allowing the fine-scale information to be preserved and integrated within the hierarchical structure rather than being lost during downsampling. This nested approach enables the model to maintain access to high-resolution data while operating at multiple scales simultaneously.
Solution Approach 2:
The patent introduces a scale dimension to the traditional geological modeling approach. Instead of modeling at a single resolution, the system creates multi-scale models where each scale represents a different level of detail. By adding this dimensional aspect, the system can capture and preserve high-resolution measurements while also providing accurate representations at coarser scales, effectively resolving the contradiction between measurement precision and information loss across scales.
2Device complexity
If data is captured at lower resolution, then device complexity is reduced, but model accuracy deteriorates
Solution Approach 1:
The patent creates a universal multi-scale modeling framework that can operate effectively at any desired scale. The system is designed to be multi-functional, handling everything from fine-scale detailed modeling to coarse-scale regional modeling using the same integrated approach. This universality allows the system to reduce complexity when working at coarser scales while maintaining the capability to access and utilize high-resolution data when needed, thus improving model accuracy without proportionally increasing device complexity.
Solution Approach 2:
The patent implements dynamic scale adjustment capabilities where the modeling resolution can be adaptively changed based on the specific application requirements. The system can dynamically switch between different scales and resolutions, using lower resolution when complexity needs to be reduced and accessing higher resolution when model accuracy is paramount. This dynamic approach allows flexible optimization of the balance between device complexity and model accuracy.
3Adaptability or versatility
If multi-scale data integration is implemented, then adaptability is improved, but computational requirements increase
Solution Approach 1:
The patent segments the geological modeling problem into distinct multi-scale components, organizing data and computations by scale level. This segmentation allows the system to process and integrate data from different scales in a structured manner, improving adaptability to various modeling needs. By dividing the complex multi-scale integration task into manageable segments, the computational requirements are organized and optimized, reducing the overall energy and computational burden compared to handling all scales simultaneously as a monolithic problem.
Data Source
AI summary
A method of predicting rock properties at a selectable scale is provided, including receiving coordinates of locations of respective sample points, receiving measurement data associated with measurements or measurement interpretations for each sample point, receiving for each sample point a scale that indicates the scale used to obtain the measurements and/or measurement interpretations, wherein different scales are received for different sample points. A deep neural network (DNN) is trained by applying the received coordinates, measurement data, and scale associated with each sample point and associating the sample point with a rock property as a function of the coordinates, measurement data, and scale applied for the sample point. The DNN is configured to generate rock property data for a received request point having coordinates and a selectable scale, wherein the rock property data is determined for the request point as a function of the coordinates and the selectable scale.