Abstract: To solve the problem of accurately predicting the gas emission volume in the fully mecha⁃ nized mining face of high-gas coal seams，taking the 15# coal seam of Tianchi Coal Mine as the research object，a study on the construction of a time series model for gas emission volume and its engineering appli⁃ cation was carried out. Firstly，based on the geological conditions and mining conditions，an index system for gas emission influence was established，clarifying the key correlation patterns among various influenc⁃ ing factors. Secondly，the recursive feature elimination cross-validation method based on random forest (RFECV) was used to screen the optimal features，and the core prediction features such as the average speed of the coal mining machine，the start-up time，the average advancing distance，and the daily output were determined. Finally，relying on the time series data processing advantages of the bidirectional long short-term memory network (Bi-LSTM)，a gas emission volume prediction model was constructed. The model validation results showed that the prediction values of this Bi-LSTM time series model had a very high fit with the on-site real values，and the overall prediction stability was excellent. In the small sample verification of the 61506 working face，the model's R² reached 0.842 4，demonstrating strong adaptability and robustness，and being able to effectively capture the dynamic time series characteristics of gas emis⁃ sion in high-gas coal seams. The model has been successfully applied in the 15# coal seam of Tianchi Coal Mine，providing precise data support for the optimization of the gas extraction system，adjustment of pro⁃duction processes，and other advanced prevention measures. It also provides a reference method framework for the prediction of gas emission in similar high-gas coal seams.

Key words: high-gas coal seam, gas emission, memory network, time series model, disaster preven? tion and control