Research on coal and gas outburst prediction based on physical information neural network

Abstract

The advancement of AI technologies provides a base for predicting coal and gas outbursts from the mathematical perspective. To enhance prediction precision of coal and gas outbursts, this research proposes a method leveraging a physical information neural network to forecast outburst intensity. The primary factors influencing coal and gas outbursts, such as ground stress, coal firmness coefficient, mining depth, gas pressure, gas content, initial velocity of gas emission, and monotonic outburst characteristics, are integrated into the physical neural network model. This integration achieves a coupling of mechanistic and data-driven models for outburst prediction. Utilizing a typical dataset obtained from a mine, research findings demonstrate that a prediction model based on the physical information neural network surpasses RF, SVM, and BPNN models in terms of generalization capability and accuracy, significantly enhancing prediction accuracy even with limited sample data.