RL-Tree: A Reinforcement Learning-Based Adaptive and Secure Routing Protocol for Wireless Sensor Networks

Abstract

In the field of wireless sensor networks (WSNs), this study proposes RL-Tree, a reinforcement learning (RL)-based adaptive and secure routing protocol. The protocol enables nodes to dynamically select optimal parent nodes by applying a Q-learning algorithm with a multi-objective reward function combining energy efficiency, transmission delay, and link security. To enhance data reliability under non-Gaussian noise, an adaptive filter integrating a variable scale factor and the Half Quadratic Criterion (HQC) is designed. The experimental platform was implemented on low-power MCUs to simulate a real WSN environment. Performance was benchmarked against RPL, AODV, LEACH, and QELAR. Results demonstrate that RL-Tree reduces average node energy consumption by 30% and achieves a data transmission delay of 0.07 seconds, outperforming baseline protocols. Integrated security mechanisms—including identity verification, encryption, and traffic monitoring—further improve network resilience under attack scenarios.