Machine Learning Prediction of Subsurface Rock Thermal Properties
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for predicting thermal properties of subsurface rock formations are limited by the wide variation in mineral composition and fluid saturation, failing to accurately account for mixed fluid environments, which is crucial for applications like thermal oil recovery and geothermal evaluations.
Innovation Solution
A machine learning model trained on a dataset of petrophysical properties of rock samples with varying lithologies, fluid saturations, and salinity is used to predict thermal properties, incorporating measurements of resistivity, porosity, permeability, and other properties to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional theoretical models are used to calculate thermal conductivity from mineral composition, then the calculation process is simple, but the prediction accuracy is insufficient due to wide variation in thermal properties and inability to account for mixed fluid environments
Solution Approach 1:
The patent replaces traditional mechanical/theoretical calculation models with a machine learning model that uses electrical resistivity measurements to predict thermal properties. This substitution allows the system to capture complex non-linear relationships between petrophysical properties and thermal conductivity without requiring explicit theoretical formulations, thereby improving prediction accuracy while managing complexity through data-driven approaches.
Solution Approach 2:
The patent transforms the prediction approach by changing from direct thermal conductivity calculation based on mineral composition to an indirect prediction method using electrical resistivity as a proxy parameter. The machine learning model learns the relationship between resistivity and thermal properties, allowing accurate prediction of thermal conductivity through measurement of electrical properties, which are easier to obtain and more sensitive to fluid saturation variations.
2Adaptability or versatility
If existing methods consider only single fluid type saturation, then the model is simpler, but it fails to accurately represent real subsurface conditions with mixed fluid environments
Solution Approach 1:
The patent creates a universal prediction model that handles multiple fluid types (oil, water, gas) and their mixtures simultaneously. The machine learning model is trained on diverse datasets containing various fluid saturations and compositions, enabling it to generalize across different subsurface conditions. This multi-functional approach allows the same model to accurately predict thermal properties regardless of the specific fluid mixture present, improving both adaptability and reliability.
Data Source
AI summary
Methods and systems are provided for predicting thermal properties of a subsurface rock formation. A training dataset is derived from petrophysical properties of a plurality of formation rock samples and thermal properties of the plurality of formation rock samples. The training dataset is used to train a machine learning model that predicts label data representing the predefined set of thermal properties given input data representing the predefined set of petrophysical properties of an arbitrary formation rock sample. The machine learning model can be validated and deployed for use in predicting thermal properties of subsurface rock formations.