To address the problems of insufficient equipment, high operational risks, single interaction mode, and unbalanced distribution of educational resources in traditional training within the rail transit field, this study constructs a three-layer virtual simulation training system framework of “immersive scenarios-multimodal interaction-cloud data intelligence” based on embodied cognition and situational learning theories. The system framework restores rail transit equipment and scenarios at a 1:1 scale through three-dimensional modeling, integrates multimodal interaction methods including hardware sensing, mobile AR, voice/gesture recognition, and 5G remote holography, and realizes resource sharing and dynamic updates relying on cloud servers. It breaks through the temporal and spatial constraints of traditional training, not only reduces teaching costs, but also strengthens students’ practical abilities and safety awareness through the operational experience in the training space. This framework effectively makes up for the deficiencies of existing teaching resources, provides an innovative solution for the cultivation of professional talents in the rail transit sector, and thus boasts significant teaching application value and promising promotion prospects.