Research on Dynamic Task Planning for Multi-Satellite Cooperative Observation Based on MADDPG

Abstract

This paper addresses the challenge of efficient task planning for multi-satellite cooperative observation in dynamic environments and proposes a dynamic task planning method based on an improved multi-agent deep deterministic policy gradient algorithm. The main contributions include the introduction of an attention mechanism to reconstruct agent state representations and enhance collaborative perception among satellites, the development of a hybrid dynamic conflict resolution strategy combining market mechanisms with task priorities, and the integration of prioritized experience replay and soft target updates to improve training stability. The satellite cooperative observation system is formulated as a partially observable Markov decision process, and the centralized training and decentralized execution framework is adopted to achieve autonomous collaborative optimization. Simulation results demonstrate that, compared with baseline methods such as the genetic algorithm, the proposed method increases the task completion rate by 11.1% under dynamic scenarios, reduces the response time to unexpected tasks by 55.9%, and decreases the number of collaborative conflicts by 69.2%. These results indicate that the proposed method provides an effective approach for the intelligent management and control of large-scale satellite constellations.