Abstract: The application of artificial intelligence technology in mineral exploration is emerging as a cutting-edge field for achieving intelligent fusion of multi-source data and efficient prospecting prediction. By proposing a hybrid deep learning model integrating two-dimensional convolutional neural networks and long short-term memory networks (CNN2D-LSTM), this study addresses the limitations of traditional methods in processing multi-source heterogeneous data and deeply utilizing geological information, thereby enabling intelligent deep fusion of geological, geochemical, and geophysical data. The CNN2D branch efficiently extracts spatial distribution features such as faults and lithological boundaries, while the LSTM branch captures the temporal evolution dependencies of multi-stage hydrothermal activities, constructing an end-to-end prediction workflow. Prospecting prediction was conducted in the Wu’erji area. The results indicate that the predicted area accounts for only 9.21 % of the total area, yet it completely covers known mineral occurrences, and based on this, three prospecting targets with favorable metallogenic potential were delineated. The research findings confirm that CNN2D-LSTM possesses excellent feature extraction and spatiotemporal pattern modeling capabilities in complex geological settings, significantly enhancing the accuracy and efficiency of mineral prediction.