A Swarm Intelligent Collaboration Approach for Ultra Task Scheduling in Mega-Constellation

Abstract

The integrated space-terrestrial network is playing an increasingly crucial role and is expected to become an indispensable component of future communication systems. As the demand for user terminals surges, task congestion issues are becoming more prominent, gradually emerging as a key factor affecting service quality. Most existing task scheduling strategies rely on the shortest forwarding path, with few considering factors such as network load conditions. Moreover, when using multi-satellite coordination to solve this problem, existing methods face drawbacks like local optimization, data leakage, and high communication costs, which result in weak inter-satellite task load-balancing capabilities. To address these issues, this paper proposes the ASCIS, an Adaptive Swarm Collaborative Intelligent Scheduling scheme. It is based on the satellite edge computing operating model, which converts service scenarios as sequential decision problems to implement quantitative optimization. ASCIS utilizes single-node multi-level fusion network to encode service status data, achieves accurate prediction of objective rewards through multi-satellite cluster cooperation, and ultimately incorporates a deep reinforcement learning framework to formulate effective task scheduling strategies, thereby effectively improving network service quality. We also propose an integrated mechanism that combines swarm learning and interactive joint training. By sharing critical gradient and variable, it reduces data transmission volume to lower communication costs. This mechanism also protects user privacy by avoiding the transmission of raw data, and enable broad-perspective data analysis through regional information fusion. Simulation experiments demonstrate that our proposed scheme has significant gains in metrics such as load distribution, task latency, resource utilization, and communication cost. Compared with other state-of-the-art schemes, ASCIS has significant performance advantages in promoting load-balancing.