Abstract: To address the challenges in medium-voltage (MV) microgrid power generation training, including high costs, spatiotemporal constraints, difficulty in fault characteristic reproduction, and operational errors during emergencies, this study proposes a mixed reality-based teaching system. Utilizing HoloLens2 holographic display and Unity3D platform, the system integrates the following functionalities: virtual modeling of MV power generation vehicles, equipment recognition and principal demonstration, switch state detection based on YOLOX-Tiny algorithm, virtual operation training, on-site operation guidance, knowledge graph construction, and expert remote collaboration. Field tests conducted at Yantai Power Supply Company Live-Work Center demonstrate that the system effectively resolves existing limitations, enhances operational efficiency and safety, while maintaining excellent human-machine interaction.