Abstract: As shallow mineral resources become increasingly depleted, the development of deep mineral resources has become an inevitable choice for safeguarding national resource security. Taking the AuIV ore belt of the Seri Gold Deposit as a typical case, this study systematically explored the application of integrating 3D geological modeling with deep learning technology in deep prospecting prediction. Based on GoCAD software, it integrated multi-source geological data from 34 boreholes, including borehole locations, inclination surveys, lithological data, and geochemical data, and constructed a fine-scale 3D geological model, providing a visualized and quantitative data foundation for spatial analysis of deep orebodies. In response to the challenges faced in deep orebody identification, such as complex geological structures and concealed mineralization information, the study innovatively developed a convolutional neural network prediction framework suitable for 3D geological environments. In the data processing stage, the original borehole data were cleaned, which was followed by the optimization of orebody boundaries with the Kriging interpolation method and gridding, with a total of 210 sets of augmented samples generated for model training and testing. An intelligent deep prospecting prediction model was constructed based on a 3D convolutional neural network. It achieves remarkable results on the test set with a prediction accuracy of 78.3 %, a precision of 82.1 %, and an AUC value of 0.82, demonstrating good generalization ability and reliability. The model was applied to conduct location prediction in the deep part of the Seri Gold Deposit. It identifies 3 prospecting targets with high mineralization probability, of which 2 are located in secondary fault-controlled areas at depths of 500–650 m, with prediction errors controlled within 12 %, showing good agreement with verification results from known borehole data. This indicates that the method has high accuracy in spatially locating deep concealed orebodies. The results not only verify the feasibility of integrating 3D geological modeling with artificial intelligence technology, but also demonstrate its significant advantages in improving the efficiency of deep resource prediction.