Accurate gas concentration prediction is crucial for coal mine safety. In this paper, a Transformer-LSTM multi-step gas concentration prediction model based on the Spatial Information Convergence Module (SIC) is proposed to address multi-step gas concentration prediction in coal mines. Existing methods mainly rely on deep learning models, such as LSTM and GCN-GRU, which they often neglect the impact of gas in surrounding underground roadway areas and are restricted to single - step prediction. To solve these problems, this study presents the SIC-Transformer-LSTM model, which combines the Spatial Information Convergence Module (SIC) and Unified Spatial Attention Allocation (USAA) attention mechanism. The designed prediction model enables a comprehensive analysis of gas distribution in mines by deeply extracting and aggregating gas concentration data from different areas, such that the prediction is improved. Experimental results indicate that, the SIC-Transformer-LSTM model surpasses existing methods in key metrics, such as MSE and RMSE.. It shows higher robustness and generalization ability, especially in complex gas dynamic scenarios. This proposed model offers a novel approach and methodology for intelligent coal mine gas monitoring.

Gas concentration prediction, spatial information aggregation, spatio-temporal features, deep learning, LSTM, Transformer