Abstract
Roof water inrush, induced by upward-developing water-conducting fracture zones (WCFZ) into aquifers, poses significant risks in coal mining. Conventional monitoring methods struggle with accurately capturing the complex 3D spatial structure and dynamic evolution of WCFZs. To address this, we propose an integrated early-warning framework combining microseismic spatial clustering and dynamic multi-indicator risk assessment. Our improved volume-corrected DBSCAN algorithm incorporates the spatial energy influence radius (apparent volume) of microseismic events, substantially enhancing dominant cluster identification. This improvement enables dynamic segmentation of WCFZ development into slow-growth, rapid-growth, and stable stages. We developed a Water-Risk Prediction Index (WRPI) model using twelve key indicators spanning spatial structure, energy characteristics, fault anomalies, and mining progress. A game-theoretic fusion optimizes indicator weighting, generating quantitative scores and four-tiered risk alerts. The model effectively identifies potential surge zones and fault-induced anomalies in real time. Field validation at Wenjiapo Coal Mine's 3# and 6# workfaces confirmed the enhanced accuracy and adaptability of our method, demonstrating its significant potential for intelligent WCFZ recognition and proactive water hazard management under complex geological conditions.