Abstract: Aiming at the problems of strong noise interference and significant non-stationarity of sound signals in the process of coal-rock cutting,and the shortcomings of traditional methods in feature extraction and recognition accuracy,a coal rock recognition method combining variational mode decomposition (VMD ) and attention mechanism convolutional long-term and short-term memory network (CNN-LSTM-AM ) is proposed.First,VMD combined with a Shannon entropy-based mode se lection criterion is employed to adaptively decompose and reconstruct raw AE signals,effectively suppressing mechanical,elec trical,and external noise while preserving key acoustic features,thereby achieving denoising and feature enhancement under complex conditions.Second,a CNN-LSTM-AM recognition model is constructed,where convolutional and LSTM layers are uti lized for deep spatial feature extraction and temporal dependency modeling,and an attention mechanism is introduced to strengthen the association between global and local information,enhancing the model′s ability to identify mixed coal-rock media.Finally,a coal-rock cutting AE experimental platform was established to collect sample data under six typical cutting conditions for validation.The experimental results indicate that,for six-class classification,the proposed method attains an ac curacy and precision of 98.33%,along with a recall of 98.37% and an F1-score of 98.33%,markedly surpassing comparative models such as CNN,LSTM,and CNN-LSTM.These findings verify the effectiveness,robustness,and engineering applicability of the proposed method under complex underground conditions.

Key words: coal-rock identification,acoustic emission signal,VMD,CNN-LSTM-AM