Application of Multi-Sensor Data Fusion and Machine Learning for Early Warning of Cambrian Limestone Water Hazards

The issue of water disasters in the mining floor is extremely severe. Despite significant progress in the on-site monitoring and identification of water inrush channels, research on the spatial development characteristics of cracks and the temporal evolution patterns remains insufficient, resulting in the incomplete development of microseismic-based water disaster early warning theory and practice. Based on this, the present study first derives the expressions for the diameter and length of water inrush channels according to the damage characteristics of microseismic events and the glazed porcelain shape features of the channels. A theoretical model for the correlation between microseismic-water inrush volume is established, and the relationship between microseismic and water level is revealed. Analysis of field monitoring data further indicates that when high-energy microseismic features (such as single high-energy events and higher daily cumulative energy) are detected, the aquifer water level begins to decline, followed by high water inrush events. Therefore, a decrease in water level accompanied by high-energy microseismic features can serve as an important early warning marker for water disasters. Finally, advanced machine learning methods are applied, in which the optimal index combination for floor water inrush prediction is obtained through the genetic algorithm, and the weights of each index are determined by integrating the analytic hierarchy process with the random forest model. Field engineering verification demonstrates that the integrated early warning system performs significantly better than any single monitoring indicator, and all high-water-inrush events are successfully predicted within four days.