This paper proposes a high-precision deterministic provisioning with federated learning training scheme to provide deterministic delay guarantees for federated learning tasks, and improves the training accuracy of machine learning and reduces computing pressure of edge nodes.