Cluster satellites' component-level collaborative observation enables on-demand stitching of the observation chain. By responding directly to dynamic targets and environmental changes, this capability represents a key trend in meeting future complex observation requirements. Mission planning is critical to realizing this collaboration. Existing methods typically employ subsystem-level models and reinforcement learning algorithms to plan missions under deterministic operational flows. However, realizing on-demand stitching requires mission planning to address the challenge of nested space sparsity optimization while accurately reflecting component-level characteristics. To address this, this paper utilizes multi-granularity digital twin models to achieve component-level on-demand modeling. We introduce the logical dimension from systems engineering to decouple nested space sparsity. Following the self-similar logical steps of synthesis, analysis, and assessment, the optimization problem is transformed into a set of high-cohesion, low-coupling sub-problems, thereby guiding the reinforcement learning process. By switching computational models based on the specific requirements of logical dimensional reinforcement learning, we established the multi-granularity digital twin logical dimensional reinforcement learning method to realize on-demand stitching of the observation chain. To validate this capability, this paper designed typical cluster satellite observation scenarios corrected by real telemetry parameters. Using the number of confirmed unknown moving targets as a performance indicator, we tested the ability of our method and deterministic planning methods to respond to complex demands under dynamic environmental conditions. Furthermore, sparsity and feature analyses were conducted to verify the rationality of the proposed approach in optimizing nested space sparsity. The results demonstrate that the proposed method successfully achieves on-demand stitching of the observation chain for cluster satellites. This approach provides an effective pathway for adapting to future complex observation requirements and serves as an exemplar for applying systems engineering to guide machine learning in solving complex problems.
扫码关注我们
求助内容:
应助结果提醒方式:
