This paper proposes a terahertz fingerprint recognition and clustered federated learning collaboration method for anti-tampering protection of power equipment in power-supplied distribution areas, aiming to enhance the security management of massive distributed power devices within such areas. First, specialized manufacturing techniques are used to create terahertz tags for power equipment, providing a foundation for physical anti-counterfeiting. Then, a dual-branch multimodal convolutional neural network is introduced to detect abnormal tampering of device fingerprints. Finally, a clustered federated learning training method is designed to address data-sharing challenges caused by privacy concerns in anti-tampering recognition models, enabling distributed joint modeling and efficient collaborative training of terahertz fingerprint data across multiple devices in power-supplied distribution areas. Experimental results demonstrate that compared with traditional methods, the proposed approach improves recognition accuracy by 212% over the histogram similarity algorithm. Additionally, it significantly outperforms conventional image recognition algorithms in terms of fingerprint recognition accuracy, training efficiency, and ensuring data usability while maintaining privacy. Finally, the fingerprint recognition accuracy reached more than 90%, thereby provides a novel technical pathway for security monitoring and attack prevention of power equipment.