TrajGNN-RL: Adaptive Mobile Sink Intelligence for Lifetime-Aware Wireless Sensor Networks

Mobile wireless sensor networks require adaptive routing and sink mobility strategies to reduce energy imbalance, congestion, and communication delay. In this study, we propose a new method called TrajGNN-RL that combines the topology learning method based on a graph neural network with the deep reinforcement learning method to plan the trajectory of the mobile sink and route the packet in a coordinated manner. Dynamic communication graphs model the residual energy, link quality, neighborhood connectivity, and traffic conditions, and reinforcement learning optimizes the sink movement and forwarding decisions via a multi-objective approach. The graph changes during network operation to handle topology and traffic changes. The performance of the proposed routing scheme was compared with the performance of the evaluated routing schemes through experimental evaluation using varying numbers of sinks and varying sink mobility. The proposed TrajGNN-RL enhanced the normalized scores of the lifetime, energy efficiency, throughput, PkDel, routing reliability, and delay reduction by 98.34%, 97.86%, 98.12%, 98.57%, 98.83%, and 96.75%, respectively. The relative improvements in the normalized lifetime, energy efficiency, throughput, packet delivery ratio, and delay reduction, respectively, were around 1.47%, 1.55%, 0.68%, 1.19%, and 1.07%, compared to the best corresponding baseline used in the experiments.