Joint Optimization of AUV Trajectory, Service Caching, and Task Offloading in Marine Internet of Things

This article addresses the global coordination challenge of task offloading, service caching, and autonomous underwater vehicle (AUV) trajectory optimization in Marine Internet of Things (M-IoT) systems. To this end, we propose a joint alternating optimization framework driven by block coordinate descent (BCD). Specifically, a multilayer heterogeneous network architecture is constructed, comprising underwater sensor nodes (USNs), AUVs, and surface buoy nodes (BNs). To handle the strong coupling among caching, mobility, and offloading, we adopt a BCD-inspired alternating framework: an improved bat algorithm (IBA) is tailored for discrete service caching, proximal policy optimization (PPO)-based deep reinforcement learning is leveraged for adaptive trajectory planning, and a Lagrangian multiplier method is used to derive efficient offloading decisions under resource constraints. Extensive simulations show consistent gains over representative baselines in delay, energy consumption, cache hit ratio, and overall cost, while maintaining stable empirical convergence.