Objective Accurately predicting mine water inflow before mining can provide directive guidance for preventing potential water hazards and ensuring safe production.

Methods To enhance the prediction accuracy and stability of water inflow in open-pit metal mines, for which atmospheric precipitation acts as the primary recharge source of water, this study developed a prediction model that coupled a bidirectional recurrent neural network (Bi-RNN) and the Groundwater Modeling System (GMS) software. Specifically, based on historical forecasted precipitation data provided by the Global Forecast System (GFS), the fluctuation pattern of differences between predicted forecasted and actual precipitation was analyzed. After being corrected using the Bi-RNN, the forecasted precipitation data were input into GMS for prediction. The coupling model was employed to predict mine water inflow in the northern and southern mining areas in the study area. Concurrently, the mine water inflow in the mining areas was also predicted using both the traditional large-well method and the recharge modulus large-well method. Finally, the prediction results based on the three methods were compared.

Results and Conclusions  The results indicate that the coupling model, the traditional large-well method, and the recharge modulus large-well method yielded mine water inflow of 294 m³/d, 276.651 to 940.613 m³/d, and 287.241 m³/d, respectively for the northern mining area and 1160 m³/d, 3330.107 to 5090.944 m³/d, and 1108.575 m³/d, respectively for the northern mining areas. These results suggest that the proposed coupling model, a prediction method combining multiple data sources, has achieved certain results and enjoys certain advantages in predicting mine water inflow. This model provides a new philosophy and technical support for predicting mine water inflow, exhibiting high theoretical value and great potential for practical application.