Machine Learning-Based Full-Cycle Evaluation Method for Gas Extraction Performance in Coal Mines.

Area of Science:

• Mining Engineering
• Geotechnical Engineering
• Computational Science

Background:

• Coal mining depth increases, leading to critical gas-related disasters.
• Current gas extraction lacks scientific evaluation, dynamic monitoring, and timely effectiveness assessment.
• These limitations hinder safe and efficient coal resource extraction.

Purpose of the Study:

• Establish a comprehensive, full-cycle evaluation framework for coal mine gas extraction.
• Develop predictive models for preliminary scheme evaluation and dynamic monitoring.
• Provide scientific decision-making support for enhanced coal mine safety and intelligent management.

Main Methods:

• Bayesian optimization-based Random Forest Regression (BO-RFR) for preliminary scheme evaluation.
• Deep neural networks and convolutional autoencoders (DNN-CAE) for dynamic prediction and reconstruction of residual gas pressure fields.
• Integration of these models into a full-cycle evaluation framework from design to implementation.

Main Results:

• BO-RFR model achieved a residual gas pressure prediction error below 0.02 MPa.
• DNN-CAE dynamic evaluation model demonstrated high accuracy with MSE of 2.73 × 10^-5 and MAE of 0.00493.
• Field tests confirmed the accuracy and reliability of the proposed methods in practical applications.

Conclusions:

• The developed framework and models offer effective solutions for coal mine gas control challenges.
• Precise control and intelligent management of gas extraction processes are enabled.
• Significant practical value is demonstrated for enhancing coal mine production safety.