Deep reinforcement learning-driven routing algorithms for high-performance wireless sensor networks

Abstract

Wireless sensor networks (WSNs) underpin a wide range of mission-critical applications, yet conventional routing protocols still struggle to balance energy efficiency, reliability, delay, and scalability under dynamic traffic and energy conditions. This study investigates deep reinforcement learning (DRL)-driven routing algorithms for high-performance WSNs and evaluates their effectiveness against classical and shallow RL-based schemes. We design two DRL routing variants DRL-DQN and DRL-AC in which each node acts as an autonomous agent that selects next hops based on a multidimensional state vector encoding residual energy, link quality, queue occupancy, hop distance, local traffic rate, and, for energy-harvesting (EH) scenarios, short-term energy arrival statistics. A multi-objective reward function jointly optimizes packet delivery, end-to-end delay, energy consumption, and load balancing. The algorithms are implemented in a discrete-event simulator and compared with LEACH-type clustering, energy-aware shortest path routing, and Q-learning-based routing across multiple topologies, traffic loads, node densities, and sink placements, including EH and heterogeneous traffic configurations. Results over extensive Monte Carlo runs show that the DRL variants significantly extend network lifetime (up to ~65% over LEACH-type and ~28% over Q-learning), increase packet delivery ratio, reduce average delay, lower energy per delivered bit, and improve fairness of residual energy distribution. DRL-AC consistently provides the best trade-off among these metrics, highlighting the benefits of actor-critic architectures for multi-objective routing. Generalization experiments on unseen topologies confirm that the learned policies retain most of their performance gains without loss of stability. These findings demonstrate that DRL-driven routing offers a practical, scalable, and adaptable solution for next-generation high-performance WSN deployments.