In wireless sensor networks, tree-based routing can achieve a low control overhead and high responsiveness by eliminating the path search and avoiding the use of extensive broadcast messages. However, existing approaches face difficulty in finding an optimal parent node, owing to conflicting performance metrics such as reliability, latency, and energy efficiency. To strike a balance between these multiple objectives, in this paper, we revisit a classic problem of finding an optimal parent node in a tree topology. Our key idea is to find the best parent node by utilizing empirical data about the network obtained through Q-learning. Specifically, we define a state space, action set, and reward function using multiple cognitive metrics, and then find the best parent node through trial and error. Simulation results demonstrate that the proposed solution can achieve better performance regarding end-to-end delay, packet delivery ratio, and energy consumption compared with existing approaches.
Keywords: Q-learning; multiple objectives; reinforcement learning; tree-based routing; wireless sensor networks (WSNs).