Research on coal mine gas emission prediction based on an improved Prophet-GRU model

Accurate gas emission forecasting is vital for mine safety, yet complex coupling between geological conditions and mining activities hinders precision. To address the coexistence of linear trends and nonlinear fluctuations, this study proposes an improved Prophet-gated recurrent unit (GRU) hybrid framework. The Prophet component is optimized via a first-order difference operator for trend refinement and a 3D-feature-space reconstruction mechanism (integrating trend, seasonality, and residuals) to isolate outliers via Euclidean distance thresholding. Operational covariates, specifically mine ventilation volume and advance footage, are integrated to account for dynamic mining impacts. The resulting residual sequences, representing high-order nonlinear dynamics, are subsequently modeled using a GRU for deep nonlinear compensation. This architecture utilizes GRU's update and reset gates to capture long-range dependencies within residuals. Validation using 182 days of field data from Donghuantuo Mine, Hebei, shows the hybrid model achieves an MAE of 6.18 m 3 /min, RMSE of 7.85 m 3 /min, and R 2 of 0.8514. This represents a 30.7% improvement over the standalone Prophet model. Furthermore, 5-fold cross-validation confirms strong generalization, with validation RMSE consistently stabilizing within the 0.15–0.25 range across partitions. This framework provides a robust reference for proactive gas disaster early warning and mine ventilation safety management.