Abstract:Accurate prediction of groundwater level dynamics in coal mining areas is essential for providing critical guidance for water inrush forecasting and water disaster prevention under mining-disturbed conditions. Existing methods for groundwater level prediction often suffer from low precision and poor accuracy. making it difficult to meet the safety production requirements of mining areas. In this study, the limestone aquifer of the Carboniferous Taiyuan Formation in the Xieqiao Coal Mine, Huainan, was selected as the research object. Groundwater depth time series data and water inflow records from 2013 to 2024 were utilized. Into deep learning models, including the Long Short-Term Memory (LSTM) network and a Convolutional Neural Network-LSTM (CNN-LSTM) hybrid model, Time series forecasting was conducted to predict the water table level depth at the borehole BuⅢC3-Ⅱ. The results indicate: (1) The C3Ⅱ Group Limestone Aquifer served as a key transitional layer for the Taiyuan formation limestone water, with distinct variations in the response patterns of recharge and discharge observed across boreholes in different sub-areas; (2) The LSTM model with multiple input features demonstrated strong predictive capability, achieving a root mean square error (RMSE) of 0.63 on the test set and a coefficient of determination (R2) of 0.86, indicating excellent prediction accuracy; (3) The CNN-LSTM hybrid model with multiple input features showed higher accuracy compared to single models, achieving a root mean square error (RMSE) of 0.55 on the test set and a coefficient of determination (R2) of 0.9. Both the LSTM model and CNN-LSTM model proposed in this study showed satisfactory predictive performance. However, the hybrid model demonstrated superior predictive capability, The results of this study provide a crucial scientific basis for accurately predicting groundwater level forecasting under mining disturbance and mine water disaster prevention and control.