FDNet: Knowledge and Data Fusion-Driven Deep Neural Network for Coal Burst Prediction.

Area of Science:

• Mining Engineering
• Geophysics
• Artificial Intelligence

Background:

• Coal burst prediction is critical for mine safety.
• Existing methods often struggle with complex seismic data.
• Integrating expert knowledge with data-driven approaches is needed.

Purpose of the Study:

• To develop FDNet, a deep neural network for accurate coal burst prediction.
• To fuse explicit features from physical models with implicit features from seismic data.
• To improve the reliability of coal burst prediction in underground mines.

Main Methods:

• Developed FDNet, a knowledge and data fusion-driven deep neural network.
• Implemented an expert knowledge indicator selection using a subset search strategy.
• Utilized deep convolutional neural networks for microseismic data feature extraction.
• Employed an attention mechanism for deep fusion of extracted features.

Main Results:

• FDNet demonstrated superior performance in engineering experiments at Gaojiapu Coal Mine.
• Prediction accuracy improved by 5% compared to state-of-the-art data-driven methods.
• Prediction accuracy improved by 16% compared to knowledge-driven methods.

Conclusions:

• FDNet effectively integrates expert knowledge and deep learning for coal burst prediction.
• The proposed fusion strategy significantly enhances prediction accuracy.
• FDNet offers a promising advancement for improving coal mine safety.