A New Time Series of Gas Outburst Prediction Model and Application in Coal Mine and Through-Coal-Seam Tunnel Mining Face

To address the limitations of existing gas anomaly prediction models in mining environments, this paper proposes a novel time series forecasting model named MSA-BiLG, which combines a multi-scale attention mechanism with a BiLSTM-GRU neural architecture. The model is designed to capture both local and global spatiotemporal dependencies in gas concentration data. To validate its effectiveness, we apply the model to three real-world datasets collected from a through-coal-seam tunnel in Sichuan, a coal mining face in Heilongjiang, and a mining project in Jilin, totaling over 460,000 data points. Experimental results show that MSA-BiLG achieves RMSEs of 0.0075, 0.0099, and 0.0007 and MAEs of 0.0072, 0.0076, and 0.0004 on the DSzQ-3, A52301, and N11902 datasets, respectively, outperforming baseline models such as GRU-LSTM, CNN, MLP, and SARIMA. The model demonstrates superior convergence, robustness, and generalization across varying data scales. This study provides a reliable and scalable solution for intelligent gas monitoring and early warning in high-risk mining and tunnel construction scenarios.