This paper proposes an edge computing scheduling method based on temporal knowledge graphs, designed for the efficient allocation of AI algorithms to video monitoring tasks. The scheduling problem is formulated as a four-tuple prediction task (server, algorithm, data, time) and addressed through a novel dual-branch architecture that combines historical periodicity modeling with non-historical dependency analysis. Experiments on a simulated video surveillance dataset demonstrate that our approach significantly outperforms existing temporal knowledge graph methods, achieving improvements of 4.31% in MRR and 6.42% in Hit@1 compared to state-of-the-art methods. The proposed model effectively captures both long-term periodic patterns and short-term variations in scheduling requirements, providing an intelligent solution for edge computing resource allocation in security monitoring scenarios.