Abstract: Designing a deep-space flexible multi-agent probe capable of autonomous task planning is an important direction for future research and development of deep space exploration technology. Multi-agent deep-space probes involve multiple objects, complex constraints and the uncertain deep space environment in mission planning, but traditional mission planning languages may not describe mission planning accurately, intuitively and concisely. In this paper, a knowledge graph is proposed to represent the planning knowledge for a multi-agent deep space probe. The method first carries out knowledge extraction from the deep space probe, then associates the probe with its state and actions by knowledge fusion, and finally mines the potential relationships between agents by knowledge processing. Compared with MA-PDDL, the method proposed in this paper is simpler and more intuitive, which enables the probe to describe its mission planning autonomously, flexibly, and accurately.