During coal mining extracting face operations, disturbance induced in aquifers can readily lead to increased water inflow at the working face, threatening mine safety. Timely advanced water inflow prediction and evaluation of results can effectively prevent water hazard incidents. Consequently, building upon a summary and analysis of current water inflow prediction methods, this study investigates a multi-feature prediction method for working face water inflow based on deep learning models. The main procedures are as follows: By integrating water inflow data characteristics and on-site conditions, "dual aquifer water levels + maximum height of microseismic events" were selected as the feature dataset. A multi-feature prediction model based on VMD-iCHOA-GRU was established for working face water inflow. Comparative models were also configured. Furthermore, validation tests were conducted under scenarios involving different feature combinations, single-target conditions, and diverse extracting face datasets to verify the multi-feature model's efficacy. Results demonstrate that the VMD-iCHOA-GRU model achieved MAE, RMSE, and MAPE values of 53.56 m3/d, 62.98 m3/d, and 3.1% respectively, outperforming all comparative models with highest prediction accuracy. The multi-feature model consistently surpassed single-factor, single-feature, and dual-feature models in accuracy. Notably, features exhibiting stronger correlation with research targets demonstrated greater advantages in enhancing prediction accuracy. Additionally, the model maintained robust performance across different extracting face datasets, confirming its relative stability and extensibility. This research provides novel methodologies for working face water inflow prediction, offering significant scientific guidance and reference value for mine water prevention and control, thereby safeguarding mine production safety.