In recent years, deep carbonate fault-controlled reservoirs have gradually become one of the hot spots and key areas of oil and gas exploration in China, and significant breakthroughs have been made in Yingxiongling area of Qaidam Basin. However, the geological conditions in Qaidam Basin are extremely complex. The severe surface environment in high-altitude areas, the subsurface high-steep structure and the widely distributed plastic salt rock with rapid lateral variations all bring great challenges to the prediction of favorable reservoirs in pre-salt carbonate. Besides, the researches on seismic prediction technology of fault-controlled fracture-cavity reservoirs in the Qaidam Basin are relatively limited and have not go far enough. Hence, in order to explore the geological genesis and corresponding seismic prediction techniques, this paper combines artificial intelligence with seismic multi-attribute analysis, optimizing and fusing the most sensitive attributes. Ultimately, the fusion attribute of the AI fracture recognition and the RMS amplitude variation is adopted as the prediction result of the fracture-cavity body. This seismic prediction workflow established for fault-controlled fracture-cavity reservoirs has achieved precise characterization of pre-salt carbonate fracture-cavity bodies. Based on the analysis of prediction result, it is come to a conclusion that the oil and gas migration and accumulation of fracture-cavity bodies are controlled by the main source-connecting strike-slip faults, and the high-quality reservoirs are mainly distributed nearby. This provides a direction for the subsequent exploration and well deployment in Yingxiongling area, supporting the objective of high production from sparse wells.