Abstract:

Accidents involving sudden water inrushes during coal mining processes often result in significant casualties and property damage. A quantifiable risk prediction model is proposed using a multi-dimensional integrated analysis approach based on the nonlinear features of small-sample drilling geological data， and from the perspective of geological structures and hydrogeological conditions， with Wu Qiang's "three-maps method" as the foundation of research. The triangle fuzzy number quantitative characterization of the empirical comparison matrix is used to reasonably allocate subjective and objective weights of the main control factors through cooperative game theory. The established weights of the five main controlling factors are as follows： aquifer thickness （0.217）， aquifuge thickness （0.209）， core recovery rate （0.251）， permeability coefficient （0.137）， and sandstone lithology coefficient （0.186）. Then the PSO-SVM-GIS is used to fully explore the data relationships of water-conducting channels and to achieve data upscaling and spatial information processing using collaborative kriging interpolation. Finally， by coupling the representation of the necessary and sufficient conditions of risk， the quantitative results are visually represented qualitatively， achieving the goal of accurately predicting disaster risk using small-sample data. The Yingpanhao Coal Mine is selected as a case study for empirical analysis， employing the vulnerability index method to integrate thematic maps of the controlling factors within ArcGIS， weighted by their respective proportions. The working face is systematically categorized into zones according to the risk of water inrush， delineating areas as safe， relatively safe， transitional， moderately hazardous， and hazardous. This classification demonstrates a strong correlation with empirical observations， thereby facilitating the precise prediction of disaster risks through the utilization of a limited dataset. The results show that the optimized model of the three-graph method based on multi-dimensional integrated analysis has good accuracy and generalization ability， and accurately characterizes the nonlinear dynamic process of sudden water inrushes controlled by multiple factors， with small data volume and extremely complex formation mechanism， achieving precise judgment of high-risk areas for water inrushes. This provides technical support for formulating targeted preventive measures， which is significant in guiding the prediction and prevention of roof water damage and subsidence disasters， and thus ensures the safe production of coal mines in advance.

Key words: water inrush, subjective and objective dual factor weighting, multivariate integrated analysis, risk prediction