This paper presents an unmanned inspection system for urban underground high-voltage cable tunnels, integrating 3D SLAM and digital twin technologies. By fusing LiDAR, IMU, and panoramic camera data, the system enables high-precision 3D modeling and autonomous navigation in GNSS-denied environments. A quadruped robot platform is employed to conduct inspection tasks, supported by robust SLAM algorithms, semantic loop closure detection, and dynamic obstacle avoidance. Field tests in complex tunnel environments demonstrate the system's high modeling accuracy (within 30 cm), strong navigation stability, and real-time responsiveness. The system improves inspection efficiency by 58% compared to manual methods and maintains a closed-loop detection success rate of over 92%. This work validates the practical feasibility of deploying intelligent robots for efficient and reliable tunnel inspection, offering a promising solution for the digital operation and maintenance of urban power infrastructure.