Global Heat Flow Mapping With Machine Learning Cosimulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for estimating heat flow and geothermal gradient at a global scale are inaccurate due to reliance on empirical relationships and lack of direct measurements, leading to unreliable boundary conditions for geothermal resource distribution and reservoir temperature predictions.
Innovation Solution
A heat flow modeler that cosimulates observed heat flow data with supervised machine learning outputs, filtering and normalizing data to train optimal models, and using kriging with external drift for interpolation, resulting in a more accurate heat flow map.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If empirical relationships and expert analysis are used to estimate heat flow and geothermal gradient, then the estimation process is simple and fast, but the accuracy and reliability of the results deteriorate
Solution Approach 1:
The patent introduces machine learning models as an intermediary between observed heat flow data and the final heat flow map. The ML models learn complex non-linear relationships from training data and serve as a bridge that captures patterns beyond simple empirical relationships, improving accuracy while the cosimulation framework integrates multiple data sources systematically
Solution Approach 2:
The patent creates a composite modeling approach by combining multiple data sources (observed heat flow data, geological parameters) and multiple modeling techniques (machine learning models, cosimulation methods) into a unified hybrid system. This composite approach leverages the strengths of each component to achieve superior accuracy compared to any single method alone
2Measurement precision
If more observed heat flow data is collected to improve estimation accuracy, then the precision of heat flow maps improves, but the cost and difficulty of data collection increases
Solution Approach 1:
The patent performs preliminary action by collecting and storing geological parameters and observed heat flow data in advance to create comprehensive training datasets. This pre-collected data infrastructure enables the machine learning models to be trained offline, reducing the need for continuous expensive data collection while maintaining high accuracy in heat flow estimations
Solution Approach 2:
The patent creates a virtual copy of the physical measurement system through machine learning models that replicate the relationship between geological parameters and heat flow. Once trained on observed data, the ML models can generate heat flow estimates without requiring additional physical measurements, effectively copying the information extraction capability at lower cost
3Measurement precision
If machine learning models are used to interpolate heat flow values, then the accuracy between observation points improves, but the complexity of the modeling process increases
Solution Approach 1:
The patent implements feedback by using cross-validation and testing on unseen data to evaluate and select the optimal machine learning model. The model selection process uses performance metrics from validation datasets to provide feedback on model quality, ensuring that the complexity added by ML is justified by measurable improvements in interpolation accuracy
Solution Approach 2:
The patent applies parameter changes by tuning hyperparameters of machine learning models and selecting different model architectures to optimize performance. The system explores multiple model configurations and selects parameters that maximize accuracy while managing complexity, adapting the model complexity to the specific characteristics of the dataset
Data Source
AI summary
A heat flow modeler preprocesses geological and heat flow data for an earth formation for inputting into a plurality of supervised learning models. The heat flow modeler trains the plurality of supervised learning models on the preprocessed geological data to estimate heat flow throughout the earth formation. The heat flow modeler interpolates the estimated heat flow values to a set of desired locations in the earth formation and cosimulates the preprocessed heat flow values with the interpolated heat flow values as auxiliary variables to generate a cosimulated heat flow map. A final heat flow map is generated by rasterizing the cosimulated heat flow map.