Acoustic-Aware BDA-QL: A Hybrid Binary Dragonfly and Q-Learning Framework for Adaptive Clustering in Underwater Wireless Sensor Networks

Abstract

Underwater Wireless Sensor Networks (UWSNs) are constrained by limited energy resources, high acoustic propagation delay, and topology variations caused by underwater mobility. This paper proposes a hybrid clustering framework that integrates the Binary Dragonfly Algorithm with Q-learning (BDA-QL) for adaptive cluster-head selection in UWSNs. The proposed method formulates clustering as a binary multi-objective optimization problem considering acoustic-aware energy consumption, end-to-end latency, and cluster load balance. Q-learning is incorporated to dynamically adjust the Dragonfly algorithm coefficients during the optimization stage, while the selected clustering configuration is evaluated under a controlled semicircular mobility model. Simulations were conducted with 100 nodes deployed in a 500 m × 500 m area over 500 simulation rounds and compared against GA, LEACH, C-LEACH, SS-GSO, CDFO-UWSN, and BDA. The results show that BDA-QL preserved the highest number of alive nodes, retaining 59 nodes at the final round, achieved the lowest final latency with 25.43 s, and delivered the highest number of packets, reaching 45,339 packets. BDA-QL provided the strongest overall trade-off across network lifetime, latency, packet delivery, and energy preservation. These findings suggest that reinforcement-learning-based coefficient adaptation can improve the robustness of Dragonfly-based clustering under underwater mobility conditions.