The absence of unified standards among smart device brands hinders collaborative management, as it requires dealing with different interfaces and communication protocols of each device, thus complicating the implementation of smart home service management. Moreover, the personalized differences due to lifestyle habits, climate conditions, and other factors also make it difficult for pre-set management rules to meet various requirements. To address these challenges, a spatial-temporal data-driven method for smart home service management is studied in this work. The method involves constructing a temporal knowledge graph for smart homes and utilizing a federated learning-based approach for smart home service management. By recording the state of concept instances in smart home scenarios, the temporal knowledge graph provides temporal data for environmental changes and service statuses. Through federated learning algorithms that incorporate model parameters from different households, personalized model updates and predictions of smart home service statuses are achieved. Experimental results showed that this method can effectively manage smart home devices, accurately meet user demands with satisfied accuracy and rate of convergence.