关键词: 周期性轮巡任务, 多星协同观测, 多目标优化, 改进NSGA-III, 多层级编码

Abstract: In sudden disaster monitoring scenarios, the distribution of affected areas exhibits spatiotemporal correlation characteristics. Consequently, there is an urgent need to complete polling observations of all associated targets within a short time window to support the rapid assessment of the overall situational status. Meanwhile, facing the dynamic evolution of disasters, efficient periodic observations must be performed to ensure information timeliness and continuity under the constraints of limited satellite resources. This complex requirement, featuring both spatiotemporal correlation and periodic revisit characteristics, is defined as the periodic polling observation requirement, which poses significant challenges to existing remote sensing satellite cooperative scheduling technologies. To address this, this paper designs evaluation metrics including task timeout degree, polling time span, and satellite load balance degree, and constructs a multi-objective optimization scheduling model. In the model solving stage, a multi-level encoding structure and a round alignment mechanism are designed. Furthermore, an improved NSGA-III algorithm integrating the multi-level encoding and round alignment mechanism is proposed. The algorithm adopts an initialization strategy that combines ideal reference point guidance with random time interval selection to generate high-quality initial solutions that satisfy constraints. Additionally, multiple crossover and mutation operators, conflict detection and repair strategies, and a historical elite archive mechanism are designed for iterative optimization. Extensive simulation results demonstrate that, compared with multi-objective optimization algorithms such as AFL-NSGA-II, NSGA-III, MOEA/D, and ODEA-ARA, the proposed method can stably converge to a high-quality Pareto front across scenarios with different task scales, verifying its effectiveness and robustness.

Key words: Periodic polling observation tasks, Multi-satellite cooperative observation, Multi-objective optimization, HERA-NSGA-III, Multi-level encoding