An intelligent reinforcement learning framework for congestion aware routing in IoT sensor networks

With the emergence of Internet of Things applications and smart devices, wireless sensor networks (WSNs) have become more crucial for their reliable and scalable communication. Due to dynamic traffic conditions, however, there can be congestion, packet loss, too much delay and unnecessary control overhead because of limited buffer space and unequal forwarding loads. These runtime variations can be difficult to cope with using conventional heuristic routing methods. In this study, an intelligent routing algorithm called Reinforcement Learning-based Congestion-Aware Routing (RLbCAR) is introduced for intelligent routing in IoT sensor networks. RLbCAR considers the routing problem as a reinforcement learning task, where a set of candidate forwarding states, a set of routing actions, and a set of rewards corresponding to the level of congestion are leveraged to determine which nodes are the best next hops. It includes queue backlog, link quality, congestion information, and adaptive control information to enhance the reliability of the routing under different traffic loads. Simulations were run in MATLAB, and an IEEE 802.15.4 network of 30 nodes was used with traffic rates varying from 30 to 120 packets per minute per node. Conventional routing methods were also evaluated, such as CL, HOF, DQN-based, DDQN-based, and multi-agent reinforcement learning methods, and compared with RLbCAR. The results demonstrate the capability of RLbCAR to reach a packet delivery ratio of 97% at 30 packets per minute per node and to keep the packet delivery ratio at 70% under heavy traffic, with a queue loss ratio reduced to 18% at 120 packets per minute per node. The results show that RLbCAR ensures reliable, congestion-adaptive, and computationally efficient routing in a resource-limited IoT sensor network.