Heatmap-based unified grid characterization and planning method for task-resource scheduling in large-scale LEO constellations

Large-scale Low Earth Orbit (LEO) constellations, featuring abundant resources, global coverage, and diverse onboard payloads, show great potential for Earth observation applications. However, existing task planning methods still lack a unified representation of tasks and resources, making it difficult to effectively integrate and schedule multi-type tasks with resources—especially in large-scale constellations where timely responses to dynamic task requests are required. To address this issue, this paper proposes an H3-grid-based unified task-resource representation framework. To achieve consistent expression and fusion of point targets, area targets, and moving targets, multi-type tasks are uniformly transformed into heatmaps, and H3-grid-based point-target clustering as well as decomposition algorithms for area and moving targets are designed to enable efficient processing of different task types. Meanwhile, satellite coverage capability is dynamically mapped onto the H3 grid to form a standardized and structured resource capability encoding that contains “time-off-nadir angle-grid cell” information. Building on this unified grid-based model, we further propose a constraint-guided task planning algorithm. Taking grid tasks and the resource capability encoding as inputs, the algorithm explicitly transforms conventional “satellite-time-window” constraints into grid-level feasible execution-mode constraints, and achieves rapid assignment through dynamic constraint tightness evaluation and candidate-set updates, thereby improving both efficiency and solution quality. Finally, experiments are conducted using a simulation platform with a 240-satellite Walker constellation. Results show that the proposed H3-based unified representation can generate grids at the 10⁵ scale within seconds and outperforms the GeoSOT grid representation in terms of areal consistency and latitudinal stability; moreover, CGH-TS outperforms the conventional greedy algorithm and the genetic algorithm in task completion rate, resource balance, and computational efficiency.