Abstract: Considering that gas concentration data is a discontinuous sequence, if the original discontinuous sequence is directly input into the model for gas concentration prediction, it will lead to weak capture ability of the model for potential periodicity, trends, and abnormal features within the sequence, affecting the accuracy and stability of the model’s prediction. Therefore, a coal mine gas anomaly warning method based on advanced detection technology is proposed. By using Rayleigh wave advanced detection technology to obtain geological structure information of coal mines, fully considering geological characteristics such as coal seams, rock layers, and faults, areas that may produce high concentrations of gas are identified as key monitoring areas, and gas sensors are uniformly arranged based on safety distance and ventilation conditions. Collecting gas concentration data in coal mines through gas sensors, decomposing the obtained gas concentration data using discrete wavelet transform, obtaining its high and low frequency components, and inputting them into the LSTM-Attention mechanism model. LSTM fully captures the temporal relationship of gas concentration data and integrates the attention mechanism into the LSTM model. The tanh function is used to process the feature information output by the LSTM model, enhancing the ability to extract key information, complete concentration surge prediction, and finally complete anomaly warning by setting a threshold. The experimental results show that the maximum accuracy of the proposed method for predicting gas concentration anomalies is 97.0%, which is significantly higher than existing methods, indicating that its coal mine gas anomaly warning effect is better and more suitable for practical scenarios.