Abstract: Gas emission is a critical parameter for daily gas prevention in coal mines, and its accurate prediction is vital for ensuring safe coal mine operations. In this study, a hybrid gas emission prediction model based on Temporal Convolutional Network（TCN） and Transformer model（TCN-Transformer-MHSA） is proposed. A dataset is established using six features, including coal seam burial depth, coal seam thickness, and initial gas content, among others. The training and testing results show that the proposed TCN-Transformer-MHSA model achieves favorable predictive performance, with mean absolute error（MAE）, root mean squared error（RMSE）, and coefficient of determination（R²） of 0.073 3, 0.270 7, 0.993 7（training set） and 1.321 8, 1.149 7, 0.897 1（test set）, respectively. Compared with other models including TCN, LSTM, and LSTM-Transformer-MHSA（with R² values of 0.793 9, 0.756 9, and 0.873 1, respectively）, the TCN-Transformer-MHSA model demonstrates the highest prediction accuracy. Furthermore, the TCN model reduces MAE and RMSE by 70.97% and 46.12%, respectively, and improves R² by 17.00% compared to the LSTM model, indicating TCN’s superior capability in extracting temporal features. By integrating the Transformer-MHSA mechanism, the proposed model further reduces MAE and RMSE by 50.78% and 29.85%, respectively, and improves R² by 3.02% compared to the standalone TCN model. In addition, global SHAP feature analysis reveals that initial gas content, and burial depth are the most influential factors affecting the prediction results. Overall, the proposed model significantly improves the accuracy and stability of gas emission predictions, providing reliable theoretical guidance and data support for coal mine gas prevention and control.