Short-term early warning identification of coal-gas compound dynamic disasters under fault influence

Under deep mining conditions, coal-gas compound dynamic disasters in fault-affected zones show complex precursor signals, making short-term early warning difficult. To address the challenge of short-term early warning and identification of coal-gas compund dynamic disasters under fault influence, this study conducts acoustic emission(AE) event stream modeling and short-term warning identification based on AE data from physical simulation experiments under at normal-fault dip angles of 45°, 60°, and 75°. After removing manual knocking interference and local abnormal fluctuations, a coal-fracture acoustic emission event stream model was constructed. Samples were then generated using a unified anchor-point sliding window, and energy enhancement in the next three consecutive acoustic emission events was used as the early warning label. Different window settings and models were compared to evaluate their warning performance. The results show that the 11-3 sliding window performs best, providing a suitable balance between historical information and warning timeliness. The CARIME-BiLSTM model achieved an accuracy of 0.965 5, precision of 0.8, recall of 1, AUC of 1, and F₁ score of 0.888 9 on the test set, with no missed warnings and only one false alarm. These results indicate that the proposed method can effectively identify future short-term energy enhancement before coal-gas compound dynamic disasters in fault-affected zones. This study provides a basis for acoustic emission precursor recognition and short-term early warning of coal-gas compound dynamic disasters under fault influence.