Surface water represents one of the crucial factors influencing the safety of shallow coal mining within the Western China mining region. The objective of this study was to investigate the infiltration properties of surface water in coal seam mining under gullies and to propose effective preventive measures. We selected the typical working face of the Buliangou coal mine as the study area and conducted a series of physical and mechanical tests on rock samples. Further, we employed the UDEC and COMSOL numerical models to examine the physical and mechanical properties, overburden mining fissures distribution laws, and surface water infiltration characteristics. We noted a good agreement between the measured (223.86 m) and modeled height (230-250 m) of the water-conductive fractured zone. When the working face advanced by 200 m, 250 m, and 300 m, the water-conductive fracture zone connected the surface, and the surface water in the gully infiltrated into the goaf along the fissure, resulting in a stable water volume of 41.9 m3/h, 56.9 m3/h, and 75.0 m3/h, respectively. In the former scenario, when the working face advanced by 200 m, 250 m, and 300 m, an excellent agreement was observed between the numerically derived (41.9 m3/h, 56.9 m3/h, 75.0 m3/h) and field measured (40.0 m3/h, 53.6 m3/h, 70.8 m3/h) stable water volume in the goaf. Consequently, the numerical model accurately predicted the water volume in the goaf from the surface water of the coal seam mining under the gully and can be utilized for further predictions under different advancing distances. Ultimately, we propose preventive measures to curtail the surface water infiltration, effectively barring the influx of surface water into the goaf, precluding the occurrence of a mine water disaster, and ensuring secure mining operations. The research results can provide theoretical guidance for the efficient prevention and control of surface water beneath gullies during the coal seam mining process.

Mine water poses serious threats to mine safety and the sustainable exploitation of resources. Efficient identification of the water source of roof water inflow in coal seams is crucial. Hydrochemical analysis is key for mine water source identification, yet it is challenged by large, complex datasets. Deep learning offers an effective solution with its powerful data representation capabilities. This study develops a water source identification model based on a deep neural network (DNN), using hydrochemical and organic composition indicators including Na⁺, K⁺, Ca²⁺, Mg²⁺, SO₄²⁻, HCO₃⁻, Cl⁻, TDS, TOC, UV254, and dissolved organic matter. Bayesian optimization is applied to tune key hyperparameters of the DNN, such as learning rate, number of neurons in each layer, and training epochs, to achieve an optimal network architecture. The model is validated using 197 water samples collected from three representative coal mines located on the border between Inner Mongolia and Shaanxi Province, China. The proposed model achieves an identification accuracy of 96.31%, outperforming traditional classifiers such as support vector machines and random forests. The results indicate that this method has high accuracy and reliability, and can provide new ideas for quickly and accurately identifying the water source of coal seam roof water inflow.