A machine learning enabled IoT framework for intelligent resource management in 6 G enabled smart cities

The Sixth-Generation (6 G) networks and Internet of Things (IoT) are anticipated to introduce smart city applications that will reshape the urban lifestyle with reliable, low latency and high connectivity communications. Ensuring the smooth operation of the smart city applications along with efficient 6 G-IoT resource management in heterogeneous and dynamic environments is a major challenge. The static and centralized resource management schemes are not suitable for 6 G-IoT due to the large scale, complexity and dynamic nature of the network. This paper proposes a scheme that leverages the benefits of the Federated Deep Reinforcement Learning (FDRL) and proactive Digital Twin (DT) for efficient resource management of 6 G-IoT in a manner that is intelligent, autonomous and secure. A five-layer resource management architecture consisting of IoT sensing layer, 6 G communication layer, edge intelligence layer, digital twin layer and management layer is proposed in this paper. The proposed 6 G-IoT resource management scheme utilizes FDRL for edge nodes to learn the globally optimal joint communication and computation resource allocation policies through collaborative learning without sharing the underlying data, thereby resolving the privacy and scalability issues. The proactive DT is utilized as a “sandbox” to do simulations of possible resource allocation actions to predict the outcomes of the actions and to preconfigure the required resources before they are really needed in the network to prevent any potential Quality of Service (QoS) violations. In this work, we demonstrate the validity of proposed framework and performance gain in smart city communication through our detailed co-simulation validation over three representative scenarios: Urban Mobility (UD), Public Event (PE) and Emergency Response (ER). The proposed framework is validated against three state-of-the-art solutions namely Static Slicing (SS), Centralized Deep Reinforcement Learning (CDRL) and Isolated Edge Deep Reinforcement Learning (IEDRL). The simulation results confirm the feasibility of proposed framework and significantly outperform existing solutions in terms of various performance metrics. Specifically, for the Urban Mobility (UD) scenario, proposed framework ensures URLLC for 98.7% of the time during the peak hour which is 65% better than the existing solutions; Digital Twin of the network has 73% service convergence time improvement and 45% SLA violations reduction. Moreover, the reliability of the emergency services provided by proposed framework is over 99% within 2.1 min after occurrence of the infrastructure failure. Furthermore, proposed framework achieves 49% energy efficiency improvement of network resources by the intelligent resource orchestration which supports the sustainable urban operations. All the research objectives of this project have been fulfilled: - The system architecture is verified. - The performance of the proposed FDRL algorithm is verified to achieve cross-domain optimization. - The Digital Twin can be used for measurable proactive maintenance. - The performance of the proposed system is verified by experiments and the proposed system shows great improvement. - The sustainability and privacy advantages of the proposed system are quantified. This work advances both the theoretical and practical sides of distributed AI for network functions. It also provides a useful practice reference for building efficient, reliable and sustainable smart cities in the 6 G era.