Vector characteristics of electromagnetic radiation and anomalous source response patterns during coal excavation

Coal and gas outburst represents one of the most catastrophic dynamic disasters in deep coal mining. Accurate identification of its precursors and precise localization of hazard sources are essential for effective early warning and safe mining operations. In this study, a novel six-dimensional feature fusion framework is proposed, based on a self-developed triaxial electromagnetic radiation (EMR) vector monitoring system. Continuous online EMR monitoring was carried out during excavation at the 2615 working face of Juji Coal Mine. Within successive ten-minute time windows, EMR amplitude peaks were extracted, and the following key parameters were calculated: horizontal deviation, vertical deviation, amplitude deviation, and their corresponding permutation entropies. The resulting feature vectors were classified using the K-nearest neighbor (K-NN) algorithm against six prototypical event vectors to enable rapid identification of outburst-related events. Identified events were categorized as either coal-seam internal fracturing or interface disturbance, and localized via two single-sensor strategies: for internal fracturing, the stress-peak location derived from numerical simulation was used; for interface disturbance, localization was achieved through deviation-vector extrapolation to plane intersection. The average localization errors were approximately 1.34 m (0.073 rad) for internal fracturing and 2.18 m (0.005 rad) for interface disturbance. The proposed "single-sensor + numerical-model" framework is easy to deploy in underground settings and supports rapid spatial localization of high-risk outburst events, offering a practical and efficient technical solution for mine safety monitoring and early warning.