Abstract: The height of water conducting fractured zone is a basic parameter for the study of water preserved coal mining.Accurately revealing the development law of water conducting fracture zone is helpful to realize water preserved coal mining.The three dimensional seismic technology has the characteristics of horizontal continuity and high vertical data resolution.To a certain extent,the data obtained by three dimensional seismic technology can compensate for the scarcity of the data obtained by drilling.In order to study the scale and the law of evolution of water conducting fracture zone at different mining stages,the 30101 working face and the 30102 working face in a coal mine in northern Shaanxi were taken as example,the neural network method of depth feeder with leakage as monitoring data was used to detect the development height of water conducting fracture by coal mining in the study area.First of all,combined with drilling fluid data,the deep learning of seismic properties,which were selected by analyzing the drilling fluid data,were carried out.Then,the fracture model was established by merging various attribute information,and the three dimensional spatial range,morphological characteristics,vertical rock failure degree and development height of water conducting fracture zone were determined by the model.Finally,the laws of development and closure of the water conductive fracture were inferred through analyzing the mining time different between the two working faces.The research shows that the maximum height of fracture zone in the 30101 working face is 120 m,and that of the 30102 working face is 133 m.Fractures with high angle are developed in water conducting fracture zone along the vertical and parallel working face direction.The fracture development in the water conducting fracture zone increases first and then decreases,the fracture closure in the upper part of the fracture zone is better,and the fracture closure in the lower part of the fracture zone and the edge of the working face is worse.The number of fractures in 18 months after mining is reduced by 21% compared with that in 4 months after mining,and the number of fractures in separation layer is reduced by 50%.Through this study,the deep learning based seismic attribute fusion technology is an effective detection technology,which is intended to detect the law of damage by rock formation caused by coal mining,and its accuracy can meet production needs.