Shale Gas Prediction Device Using Interpretable Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI techniques for resource development, such as those using neural networks, are not applicable to shale gas and shale oil mining and lack interpretability in their prediction results, making it difficult to understand the contributing factors.
Innovation Solution
A prediction device and method using machine learning models, specifically heterogeneous mixture learning, to acquire and output feature quantities related to shale gas or shale oil wells, calculating production volumes and sand return amounts, while providing the contribution degree of each feature quantity to the predicted values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If neural network methods are used for resource development predictions, then prediction capability is improved, but interpretability of prediction results deteriorates
Solution Approach 1:
The patent segments the prediction system into multiple independent linear prediction formulas, each handling specific prediction tasks. This segmentation allows each formula to maintain high interpretability while collectively achieving comprehensive prediction coverage, resolving the contradiction between accuracy and interpretability.
Solution Approach 2:
The patent changes the mathematical parameters from complex neural network structures to simple linear equations with explicit coefficients. By transforming the prediction model into linear form with interpretable parameters, the system maintains prediction accuracy while enabling clear understanding of contribution degrees.
2Measurement precision
If complex AI models are used for shale gas and shale oil predictions, then prediction capability is improved, but applicability to specific mining contexts deteriorates
Solution Approach 1:
The patent applies local quality by creating prediction formulas specifically tailored for shale gas and shale oil mining contexts. Each linear prediction formula is designed with parameters and coefficients that reflect the specific characteristics of shale formation, making the model highly adaptable to this specific application while maintaining accuracy.
3Loss of information
If contribution information is provided for each prediction, then interpretability is improved, but device complexity increases
Solution Approach 1:
The patent extracts the contribution information directly from the coefficients of the linear prediction formulas. By taking out the interpretation data as a natural byproduct of the simplified linear model structure, the system provides comprehensive contribution information without adding significant complexity, as the same coefficients serve both prediction and interpretation functions.
Data Source
AI summary
In a prediction device, an acquisition means acquires a feature quantity related to a well of shale gas or shale oil. A prediction means calculates a predicted value of a production volume of the well or a sand return amount of the well, based on the feature quantity, using a machine learning model. An output means outputs the predicted value and a contribution degree of the feature quantity to the predicted value.