Dynamic geo‐hydrogeological monitoring‐driven situational awareness for real‐time floor water inrush risk prediction in deep mining

Abstract To address the issues of reliance on static parameters and weak early warning in unmined areas in the risk prediction of water inrush from the bottom slab of deep ultra‐wide working faces, a dynamic mulit‐source monitoring‐driven water inrush risk prediction situation awareness system was constructed, taking the 400 m ultra‐wide working face of Shandong Binhu Coal Mine as the research object. A microseismic–electrical method coupling monitoring system was set up to collect data on microseismic depth, energy, and resistivity of the bottom rock strata, and training labels were created through spatio‐temporal alignment and information entropy superposition. Meanwhile, 10 control factors were selected to construct a dynamic and static fusion feature matrix. The whale optimization algorithm–convolutional neural network model was adopted to optimize the hyperparameters and establish the mapping relationship between features and risks. The results show that when the model advances 370, 550, and 820 m in the working face, the root mean square error gradually decreases to 0.0586, the prediction accuracy of the unmined area reaches 91.54%, and the performance is stable when the data volume fluctuates. This system studies the precise identification of high‐risk areas, guides on‐site measures to ensure safe mining, realizes full‐time and spatial dynamic inversion, and early warning of water inrush risks in deep and ultra‐wide working faces, and provides technical support for the prevention and control of water hazards in deep mining and underground engineering.