Adaptive clustering routing protocol for underwater sensor networks

• Published in: Ad hoc networks 2022-11, Vol.136, p.102953, Article 102953
• Main Authors: Sun, Yao, Zheng, Maochun, Han, Xiao, Li, Shuang, Yin, Jingwei
• Format: Article
• Language: English
• Subjects: Clustering; Reinforcement learning; Routing protocol; Underwater sensor networks
• ISSN: 1570-8705; 1570-8713
• DOI: 10.1016/j.adhoc.2022.102953

In underwater sensor networks, the limited energy of sensor nodes and the difficulty of replacing the power supply means that one faulty node will severely affect the lifetime of the entire network. The design of energy-saving and efficient routing protocols is the key to prolonging the network lifetime. This paper describes an adaptive clustering routing protocol for underwater sensor networks based on multi-agent reinforcement learning. The proposed protocol models the network as a multi-agent system, and allows the nodes to select the global optimal route collaboratively through reinforcement learning. To reduce the probability of hotspots generation, an adaptive cluster head selection algorithm that does not incur any additional communication overhead and does not require consensus from surrounding nodes is proposed, which enables nodes to autonomously decide whether they can act as cluster heads using routing and environment information. Additionally, a biased reward function is designed to feedback the effect of the adaptive cluster head selection algorithm on the routing performance and to encourage the nodes to select the cluster heads as relays. Simulation results show that the proposed adaptive clustering routing protocol achieves higher routing efficiency, lower energy consumption, and longer network lifetime than existing approaches.