Abstract:

Underwater sensor networks data transmission faces significant challenges such as high latency and high energy consumption, particularly in multi-hop transmission scenarios where time delay and energy consumption issues become more pronounced. Based on this, a hierarchical reinforcement learning-based underwater routing protocol is proposed, which constructs a two-level decision-making framework comprising high-level and low-level decision layers. By implementing a reward function with variable weights, a spatiotemporal adaptive multi-objective optimization model is achieved. Additionally, a void repair mechanism is introduced, incorporating intelligent state monitoring and void detection to ensure that data can bypass void regions, thereby maintaining network connectivity. Simulation results demonstrate that dynamic Q-learning-based layered acoustic routing (DQLAR) can maintain stable, reliable, and efficient performance in underwater environments with varying node densities, multiple data sources, and dynamic topologies. Dynamic Q-learning-based layered acoustic routing exhibits higher packet delivery rates, lower average end-to-end latency, and significantly reduced energy consumption compared to conventional methods, highlighting its potential for deep-sea long-term monitoring applications.

Key words: underwater sensor networks, routing protocol, void repair