Federated deep reinforcement Q-learning for secure and energy-efficient data routing in internet of things networks

Internet of Things (IoT) enabled wireless sensor networks face severe issues in secure and energy-efficient data routing, which are associated with dynamic topology, limited node energy, non-independent and identically distributed data, and malicious routing. The traditional centralized routing methods have a high communication overhead, a lack of scalability and privacy. In this research, secure routing is developed as a federated multi-agent deep reinforcement Q-learning network to collaboratively optimize energy efficiency, delay, throughput, and routing security. Security is implemented as a reward or penalty for malicious path choice; the abnormal packet drops and unstable links. The potential inference search algorithm is employed to perform cluster formation, and the hybrid human memory golden jackal optimization algorithm is employed to optimize cluster-head selection. federated learning allows learning policies without exchanging raw data, which makes it decentralized, minimizing privacy risks and communication overhead over centralized Deep Reinforcement Learning (DRL). NS3 simulations of 100 nodes indicate that the proposed approach attains 325 kbps throughput, 8 ms delay, 175 J energy consumption and 55% of the residual energy, which is better than existing models in network lifetime and stability. Findings validate the usefulness of federated multi-agent DRL in scalable, secure, and energy-efficient IoT routing.