Prediction of Nitrogen Injection Displacement Efficiency for High-Efficiency Methane Emission Reduction in Coal...

Coal mine methane (CH₄) has an extremely high global warming potential, and its capture and utilization have become key pathways for carbon neutrality in the energy sector. Nitrogen injection displacement technology, by enhancing coal seam methane drainage, can effectively curb methane escape into the atmosphere, making it an important means to achieve methane emission reduction in coal mining areas. However, this process is highly complex, and traditional models struggle to accurately predict methane drainage volume (gas purity), limiting the maximization of emission reduction effectiveness assessment and process optimization. To this end, this study constructed a deep learning model (TCN-BiGRU-MHA) integrating a temporal convolutional network (TCN), bidirectional gated recurrent unit (BiGRU), and multi-head attention mechanism (MHA), aiming to achieve dynamic and accurate prediction of methane drainage concentration during nitrogen displacement. By comparing with baseline models such as LSTM and RNN, the model's performance was systematically evaluated across metrics, including MAE, MSE, R ², MAPE, and RMSE. The results show that the TCN-BiGRU-MHA model achieved MAE, MSE, R ², MAPE, and RMSE of 0.00001, 0.00323, 0.9852, 2.6401, and 0.0041, respectively, on the prediction set. Corresponding metrics were 0.00017, 0.001123, 0.90024, 7.0106, and 0.0131 on the test set, demonstrating significantly superior prediction performance compared to the benchmark models. The excellent predictive capability of this model provides core algorithmic support for real-time optimization of nitrogen injection parameters, maximization of methane capture efficiency, and accurate quantification of emission reduction benefits. This research provides a crucial theoretical tool for the coal mining industry to formulate measurable methane emission reduction schemes.