Abstract: With the shallow mineral resources are becoming increasingly depleted, it has become an inevitable choice to develop deep mineral resources for safeguarding national resource security. The method of intelligent mineral prospecting prediction based on three-dimensional geological modeling and deep learning technologies is proposed, using the AuIV ore belt in the Seri gold deposit of Qinghai Province as a case study. Based on the GoCAD software platform, this study integrates multi-source geological data, including positioning, inclination, lithology, and geochemistry, from 34 boreholes in the study area to construct a detailed three-dimensional geological model, providing a visual and quantitative data foundation for deep ore body spatial analysis. To address the challenges of deep ore body identification, such as complex geological structures and hidden mineralization information, this study innovatively developed a convolutional neural network prediction framework suitable for three-dimensional geological environments. During the data processing phase, the raw borehole data was cleaned, the ore body boundary was optimized using kriging interpolation, and gridding was performed to generate 210 sets of enhanced samples for model training and testing. An intelligent deep mineral prospecting prediction model was constructed based on 3 D convolutional neural networks, achieved significant results on the test datasets, with a prediction accuracy of 78.3 %, a precision of 82.1 %, and an AUC value of 0.82, demonstrating good generalization and reliability. The model was applied to deep-level location prediction in the Seri gold deposit, successfully identifying three high-probability mineralization targets. Two of these targets were located within a secondary fault structure-controlled area at depths of 500–650 meters. The prediction error was kept within 12 %, which is consistent with the verification results of known borehole data, demonstrating that this method has high accuracy in spatially locating deep concealed ore bodies. This achievement not only verifies the feasibility of integrating 3 D geological modeling with artificial intelligence technology, but also demonstrates its significant advantages in improving the efficiency of deep resource prediction.