Multi-Agent DDPG-Based Multipath Routing for Remote Sensing Services in Mega-VLEO Constellations

As an essential component of space-air-ground integrated networks (SAGIN), the construction of very low Earth orbit (VLEO) remote sensing (RS) constellations has seen significant breakthroughs in recent years. The efficient multipath inter-satellite routing strategy have become an urgent requirement for RS data forwarding. However, the continuously expanding constellation scale imposes higher optimization demands on the multipath splitting ratio. To address this problem, this paper proposes a Mega-VLEO Multipath Routing (MVMR) algorithm based on multi-agent deep deterministic policy gradient (MADDPG). After obtaining multiple routing paths, the algorithm estimates the network state across adjacent time slots, with different agents employing the deep deterministic policy gradient architecture to solve the splitting ratio for various region pairs. Throughout the training process, MADDPG leverages global state updates to ultimately maximize the routing performance over the entire cycle. Simulation results demonstrate that, compared to water-filling algorithms, the MVMR improves the average service delay by approximately 14.68% and increases the service completion rate by 5.6%.