Gas outbursts in coal seams represent a severe and formidable hazard, posing a significant threat to the safety of coal mining operations. The advanced early warning is a crucial preventive measure against outbursts. Acoustic emission (AE) and electromagnetic radiation (EMR) are advanced monitoring and early warning techniques for gas outbursts. However, during the mining operations, interference signals from AE and EMR may arise. Due to the impact of these interference signals, the use of statistical indicators and time-frequency feature analysis may lead to false alarms and missed detections in outburst warnings. The advancement of deep learning offers new methods for intelligent identification of gas outburst risks. This article proposes an outburst warning method for detecting outburst precursor signals and conducting comprehensive index analysis based on deep learning techniques for AE and EMR. First, reconstruct the signal using wavelet packet decomposition and then process the resulting signal with the diffusion-semi-supervised classification algorithm, employing partially labeled signals to train the model for intelligent identification of outburst precursor risk indicators of AE and EMR. By analyzing the prominent risk precursor signals of AE and EMR, establish a gas outburst risk analysis method based on Bayesian networks, thereby achieving early warning of gas outbursts. The findings suggest that the method in question, which employs a training dataset comprising 60% manually annotated data, is proficient in precisely identifying to outburst precursor signals of AE and EMR, and is adept at identifying a range of precursor signals. It provides a basis for distinguished multi-level early warning. The research outcomes significantly enhance the reliability of AE and EMR monitoring signals, offering effective monitoring and early warning for gas outbursts in coal seams, gas power manifestations, and abnormal gas.
