Decentralized Cluster Head Selection in IoV using Federated Deep Reinforcement Learning

The Internet of Vehicles (IoV) is a subset of cyber-physical systems (CPS) that interconnects vehicles and infrastructure through networking to provide safety, convenience, and improved traffic management. IoV promises to mitigate congestion, lower fuel consumption, reduce pollution, and enhance travel efficiency, all while extending internet access to vehicles. Clustering is a technique shown to address challenges in IoV, and cluster head selection plays a significant role in the stability of the network as the cluster head orchestrates intra and inter-cluster communication. When a cluster head possesses a weak link among its members, the risk of link breakage rises, diminishing information sharing. Previous research has offered a number of solutions for cluster head selection in IoV, but current solutions rely on centralized infrastructure and data to determine the optimal cluster head. As such, these solutions suffer from latency and are not applicable in regions where there is limited infrastructure. In addition, the centralization of data raises privacy and security concerns. We propose a novel decentralized approach for cluster head selection enabled by Federated Deep Reinforcement Learning (FDRL). Our solution offers a low latency, distributed learning approach that preserves the private data of each vehicle. The performance of our proposal is examined using the Veins framework that is based on the two simulators OMNET++ and SUMO. Our results achieved low latency communication with high neighbor connectivity, packet delivery, and cluster stability.