Study on the influence of excavation footage on gas concentration prediction based on the Transformer model

Coal mine gas disasters pose a serious threat to mine safety. Thus, enhancing the accuracy of gas concentration prediction in mines is vital for preventing coal mine accidents effectively. Conventional gas concentration prediction methods typically focus solely on gas concentration variations, overlooking other environmental factors that impact gas outburst amounts. This oversight leads to low prediction accuracy. This study introduces a Transformer-based prediction model, which innovatively integrates mine excavation footage data as an input feature. Unlike traditional approaches, this method more effectively captures gas concentration variations. The experiment utilized gas concentration and excavation footage data from a Hebei Province coal mine, which exhibited diverse data characteristics, for prediction analysis. Comparative experiments were carried out with the traditional long short-term memory and gated recurrent unit models. The results demonstrated that combining excavation footage data with gas concentration data as model inputs yielded lower root mean squared error and mean absolute error values. It also substantially improved the R-Square (R2) for predictions compared to using only gas concentration data. Moreover, the Transformer model exhibited higher prediction accuracy across different datasets than other models, highlighting its superior adaptability and effectiveness. This method is significant for gas concentration prediction and risk warning in coal mine working faces.