Multi-level Federated Learning Mechanism with Reinforcement Learning Optimizing in Smart City

While taking account into data privacy protection, federated learning can mine local data knowledge and gather data value, which has been widely concerned by the Smart city and Internet of Things. At present, a large amount of data is generated by the massive edge network in the smart city, but the resources of the edge side are limited. How to reduce the communication overhead between the edge and the centralized cloud server, improve the convergence speed of data model, and avoid resource waste caused by synchronized blocking of federated learning has become the core issue for the integration of federated learning and the Internet of Things in the smart city. For this reason, this paper designs a multi-level federated learning mechanism in the smart city, and uses reinforcement learning agents to select nodes to offset the influence of the non-IID data that is not independent and identically distributed. At the same time, asynchronous non-blocking updating method is used to perform model aggregation and updating of federated learning to release the resources of faster devices and improving the efficiency and stability of federated learning. Finally, simulation results show that the proposed method can improve the efficiency of federated learning tasks in edge network scenarios with a lot of devices in the smart city.