Task-resource unified grid-based representation and planning method for low earth orbit large constellation

Abstract

The application discloses a task-resource unified grid-based representation and planning method for a low-orbit large-scale constellation, comprising the following steps: uniformly encoding the spatial position information of various original observation tasks by using a global discrete grid system; clustering and fusing all original observation tasks contained in the same grid to calculate the grid heat value of each grid; mapping the coverage area of a satellite at different time points and different side swings into a grid sequence to obtain dynamic grid encoding information of the satellite resource load capacity at different time points and different side swings; establishing a task planning model and solving the model based on a constraint-guided heuristic task planning algorithm to obtain an optimal task planning scheme. The application is applied to the field of constellation task planning, and through a unified geographical grid framework, multi-source heterogeneous tasks and dynamic satellite resources are integrally coded and represented, and an efficient solving algorithm is designed, so that the task planning efficiency and quality are simultaneously improved.

Description

Technical Field

[0001] This invention relates to the field of constellation mission planning technology, specifically a unified gridded representation and planning method for missions and resources of large-scale low-Earth orbit constellations based on heatmaps. Background Technology

[0002] Since the successful launch of the first artificial satellite in the last century, space technology has developed rapidly, with an increasingly diverse range of satellite platforms and payloads. Traditional large-capacity, long-life satellites, due to high development costs, difficult maintenance, and long launch cycles, exhibit limitations in meeting the diverse and dynamic mission requirements. In contrast, large-scale Low Earth Orbit (LEO) constellations, with their advantages of low orbit, low cost, wide coverage, short launch cycles, and multiple payloads, have become a focus of current research. Furthermore, breakthroughs in multi-satellite launch and reusable rocket technology have made rapid deployment of LEO constellations possible. These constellations typically carry multiple payloads (high-resolution cameras, multispectral sensors, synthetic aperture radar, etc.), effectively supporting multiple applications such as Earth observation, global communication, and moving target tracking, significantly enhancing the constellation's service capabilities. Against this backdrop, how to achieve efficient utilization of the resources of large-scale LEO constellations and meet the dynamic and diverse mission requirements has become a core challenge in the operation and management of large-scale LEO constellations.

[0003] LEO constellation mission planning is an NP-hard problem, and the quality of the planning result directly determines the overall mission execution effect. Metaheuristic algorithms can provide near-optimal solutions for this type of problem within an acceptable time, and therefore have attracted widespread attention. Early research was mostly based on preset rules and static models, which are difficult to adapt to highly dynamic environments. Some studies have used genetic algorithms for mission planning, which has improved accuracy, but the computational efficiency is low when dealing with large-scale tasks. Another study proposed a decomposition scheduling framework for the 84-satellite Walker constellation and combined it with heuristic methods for solving the problem, but it still suffers from high computational complexity. Other studies have used ant colony algorithms to achieve multi-satellite collaborative scheduling and improve allocation efficiency through orbital resource optimization, but its time overhead is large and it is difficult to meet the requirements of real-time scheduling.

[0004] In recent years, numerous studies have been conducted to improve the real-time performance of satellite mission planning. Some studies have improved genetic algorithms to suit small-scale constellation scheduling problems; others have constructed multi-objective mission planning models for agile satellites and combined genetic algorithms with simulated annealing algorithms for solution; to address the scheduling problem of heterogeneous constellations, some studies have proposed bidding mechanisms and contract network protocols, and subsequent studies have introduced dynamic negotiation mechanisms and distributed negotiation methods to improve the responsiveness to emergency mission planning. However, these studies mainly focus on small-scale mission planning problems, and when facing large-scale, heterogeneous constellation architectures, problems such as low computational efficiency and poor stability still exist.

[0005] On the other hand, existing research has not yet formed a unified framework for task and resource modeling. Current research mainly focuses on designing models for single-type tasks (such as point targets, regional targets, or moving targets), lacking a unified modeling design for multiple types of tasks. This leads to complex conversions between different models and time-consuming operation. Some studies have used grid partitioning to study regional target modeling, but this has not been extended to other task types. Some scholars have implemented a gridded representation of satellite resources based on GeoSOT encoding, but the selected GeoSOT encoding grid size varies with latitude, lacking a unified global grid quantity and size. Other studies have proposed a unified representation method for satellite resources, which simplifies the resource optimization problem to some extent, but it is still insufficient when dealing with multiple types of tasks. Summary of the Invention

[0006] To address the problems of low computational efficiency and inconsistent modeling methods in existing LEO large-scale constellations, which make it difficult to meet the application requirements of large-scale, highly dynamic, and multi-type missions, this invention provides a unified gridded representation and planning method for mission-resources of large-scale LEO constellations. Through a unified geographic grid framework, multi-source heterogeneous missions and dynamic satellite resources are encoded and represented in an integrated manner, and an efficient solution algorithm is designed to achieve simultaneous improvement in mission planning efficiency and quality.

[0007] To achieve the above objectives, this invention provides a unified gridded representation and planning method for mission-resources of large-scale low-Earth orbit constellations, comprising the following steps:

[0008] Step 1: Obtain all raw observation tasks to be executed within the planning cycle of the large-scale low-Earth orbit constellation, and use the global discrete grid system to uniformly encode the spatial location information of all types of raw observation tasks. Step 2: Cluster and fuse all the original observation tasks contained in the same grid, and calculate the grid thermal value of each grid based on the clustering and fusion. Step 3: For each satellite in the large-scale low-Earth orbit constellation, map its coverage area under different times and different lateral swings into a grid sequence to obtain dynamic grid coding information of satellite resource payload capacity under different lateral swings at each time point; Step 4: Based on the gridded representation, a task planning model is established with the goal of maximizing the task completion benefits while meeting task requirements and resource constraints. The model is then solved using a constraint-guided heuristic task planning algorithm to obtain the optimal task planning scheme.

[0009] Compared with the prior art, the present invention has the following beneficial technical effects: 1. This invention introduces a global discrete grid system (such as the H3 grid) to uniformly encode the spatial locations of various heterogeneous tasks, such as points, regions, and mobile tasks, into standard grid units. It also generates a global task heatmap through thermal value fusion, forming a unified task demand density field. At the same time, it dynamically encodes the time-varying coverage capability of satellites into a time-side tilt angle-grid sequence. This dual-grid representation transforms the complex spatiotemporal matching problem into a resource allocation problem within a unified grid space, greatly simplifying the model construction and data processing process. 2. This invention designs a constraint-guided heuristic task planning algorithm for planning models after gridding representation. By dynamically evaluating the constraint tightness of tasks, it prioritizes the allocation of key tasks with limited optional resources and high degree of conflict, thereby effectively compressing the search space and reducing the complexity of problem solving. Its candidate set generation and update mechanism are deeply adapted to the gridding input, and can quickly generate high-quality planning schemes in scenarios with massive tasks and resources, meeting the application requirements of high dynamic response. 3. The method of this invention does not depend on a specific constellation configuration or mission type. Its gridded representation framework has good versatility and scalability, making it easy to incorporate new mission attributes or satellite constraints. At the same time, the entire process is highly structured, computable, and easy to implement and integrate in software, providing a solid technical foundation for building an efficient and intelligent large-scale constellation mission planning system. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0011] Figure 1 This is a flowchart of the mission-resource unified gridded representation and planning method for large-scale low-Earth orbit constellations in this invention. Figure 2This is a schematic diagram of a large-scale low-Earth orbit constellation mission planning in an embodiment of the present invention; Figure 3 This is a schematic diagram of the IJK three-dimensional coordinate system in an embodiment of the present invention; Figure 4 This is a schematic diagram of dynamic grid coding for satellite resource capabilities in an embodiment of the present invention; Figure 5 This is a schematic diagram comparing the efficiency of different clustering algorithms in an embodiment of the present invention; Figure 6 This is a schematic diagram comparing the number of clusters after different algorithms in an embodiment of the present invention.

[0012] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0014] Furthermore, the technical solutions of the various embodiments of the present invention can be combined with each other, but only if they are feasible for those skilled in the art. If the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0015] like Figure 1 The above illustrates a unified gridded representation and planning method for mission-resources of a large-scale low-Earth orbit constellation disclosed in this embodiment, which mainly includes the following steps: Step 1: Obtain all raw observation tasks to be executed within the planning cycle of the large-scale low-Earth orbit constellation, and use the global discrete grid system to uniformly encode the spatial location information of various raw observation tasks; Step 2: Cluster and fuse all original observation tasks contained in the same grid, and calculate the grid thermal value of each grid based on the clustering and fusion. Step 3: For each satellite in the large-scale low-Earth orbit constellation, map its coverage area under different times and different lateral swings into a grid sequence to obtain dynamic grid coding information of satellite resource payload capacity under different lateral swings at each time point; Step 4: Based on the gridded representation, a task planning model is established with the goal of maximizing the task completion benefits while meeting task requirements and resource constraints. The model is then solved using a constraint-guided heuristic task planning algorithm to obtain the optimal task planning scheme.

[0016] refer to Figure 2 This diagram illustrates the planning of a large-scale LEO constellation mission. The planning of such missions involves multiple satellites, each with different operational states. Furthermore, the problem includes various mission types (point targets, regional targets, and moving targets), each typically possessing different priorities and specific observation constraints. Due to the constraints of satellite operation, different satellites have varying observation capabilities for different missions, necessitating the planning of appropriate satellite resources for each mission. In the context of a large-scale constellation, the sheer number of satellites makes rapid, efficient, and high-quality mission allocation difficult. To address this issue, this embodiment utilizes the geometric consistency and multi-resolution characteristics of a global discrete grid system to construct a mission heatmap and dynamically encode satellite resource capabilities, thereby unifying the representation of heterogeneous missions and multi-dimensional resources. Based on this, a constraint-guided heuristic planning algorithm is designed to achieve rapid and accurate mapping between missions and resources, ultimately achieving efficient allocation of large-scale constellation missions.

[0017] Faced with the diverse, large-scale, and complex requirements of missions in large-scale low-Earth orbit constellations, traditional task-type modeling methods struggle to efficiently match tasks and resources on a unified spatial scale, becoming a bottleneck restricting scheduling efficiency. To address this, this embodiment introduces a global discrete grid system to uniformly encode multiple task types and employs a heatmap fusion method to construct a task heatmap. This aims to integrate the spatiotemporal requirements and priorities of point, regional, and mobile tasks into a situation map that intuitively and quantitatively reflects the global spatial distribution and urgency of tasks. This transforms complex scheduling decisions into responses to heatmap distribution, providing structured input for subsequent satellite resource allocation.

[0018] To efficiently process geospatial data, this embodiment selects a multi-level hexagonal global discrete grid system (H3). Its basic principle is based on an icosahedron, unfolding the Earth and projecting it onto a sphere, then discretizing the globe using hierarchical hexagonal grids. Compared to traditional latitude and longitude grid systems (such as Geohash or GeoSOT), H3 has advantages in cell area uniformity and hierarchical consistency, making it more suitable for the integrated representation of various types of space missions. (Reference) Figure 3 H3 uses the IJK 3D coordinate system to locate elements and maintains consistency across multiple resolutions through parent-child hierarchical relationships. Each mesh is recursively associated with its parent mesh, and child meshes are subdivided by extending the parent coordinate system with three-dimensional coordinates, as shown in Table 1. The vast majority of elements in the system are hexagonal, with a small number of pentagonal elements. Their influence decreases with increasing hierarchy and is generally negligible.

[0019] Table 1. Parent-Child Coding Relationship

[0020] H3 encoding uses a 64-bit integer representation with the following structure: the first bit is reserved; bits 2–5 represent the indexing pattern; bits 6–8 identify the edge or vertex number; bits 9–12 represent the level (0–15); bits 13–19 are the base grid number; and each subsequent 3 bits represent the orientation in a specific IJK coordinate system level, thus achieving multi-level indexing from coarse to fine. H3 encoding allows for the unified conversion of spatial locations from different observation tasks into discrete grid codes, providing a foundation for incorporating temporal attributes and thermal value calculations.

[0021] In this embodiment, the original observation task objectives include three categories: point targets, regional targets, and moving targets. To achieve standardized processing of task planning, this embodiment designs point target clustering, regional target decomposition, and moving target trajectory discretization methods, respectively, to uniformly convert the task into task units with grid coding, providing a consistent data representation for subsequent fusion and allocation.

[0022] For raw observation tasks with point targets, the H3 system provides... The function can convert any latitude and longitude coordinates. Quickly map to a specified level Mesh encoding below Based on this, all point tasks encoded in the same grid can be aggregated, and their original task numbers, priorities, and other required attributes can be recorded. The clustering result retains the attributes of the original tasks and provides input for the fusion of multiple tasks within the same grid. The specific clustering process for all point tasks encoded in the grid is as follows: Create an empty collection , used to store the final generated point target mesh task; Process the input point target set one by one Each point target in ; For the current target point Extract its latitude and longitude coordinates Call the geoToH3 function of the H3 system and specify the target level. Calculate the level of the coordinate point. The unique coded index of L's H3 grid ; The grid index obtained in the previous step The original index of the target point The relevant attribute information (such as task priority, time window, other parameter resolution requirements, etc.) is encapsulated into a data tuple. And add this tuple to the result set. In, that is .

[0023] After traversing the input set All point targets in the generated, gridded set of point target tasks. .

[0024] For original observation tasks with a target type of regional target, the regional target is usually defined by a polygon boundary, denoted as . (Composed of a sequence of latitude and longitude vertices), the gridded objective of a regional target is at a given level. Next, the H3 mesh encoding set covering the polygonal region is obtained. Considering the hierarchical nature of H3, this embodiment adopts a top-down recursive subdivision strategy, starting from a coarser mesh level and gradually increasing the subdivision level. and Obtain the sub-mesh boundary and sub-mesh set separately, and determine whether to retain the mesh cell by checking whether the mesh cell intersects with the region polygon, until the minimum decomposition level is reached. This yields the grid code set for the region corresponding to the regional target. The specific process of regional target decomposition and coding is as follows: Get the target polygon of the region With minimum decomposition grid level And create an empty set of grids to be processed. In the target polygon of the region Select any point within the H3 system and obtain its grid code at the coarsest level (level 0). and put it into the pending set. middle; Starting from level 0, process layer by layer downwards until the minimum decomposition level of the target is reached. For each level : Create an empty collection This is used to temporarily store the mesh that needs to be processed in the next layer; Traverse the current level's unprocessed set Each grid cell in Call the H3ToGeoBoundary(h) function to obtain the mesh. geographic boundary polygon Check the grid boundaries With the target polygon of the region Whether they intersect (i.e., whether there is an overlapping part), if they do not intersect ( If the grid intersects, ignore this grid and proceed to the next one. If they intersect, perform the following operations: If the current level It is already equal to the target level Then the current grid will be Add to the final result set middle; If the current level The target level has not yet been reached. Then call the H3ToChildren(h) function to get the mesh. Add all next-level subgrids to the temporary set. This allows for more precise judgments in the next cycle; After all meshes at the current level have been processed, Updated to Start the next level The processing; After all nested loops have finished, the data is stored in the collection. All the grids in the diagram constitute the polygon of the covered area. And the hierarchy is The H3 grid encoding set.

[0025] For a primary observation task targeting a moving target, assuming the target's trajectory is obtainable within the planning period, let its position over time be denoted as . To unify continuous trajectories into a sequence of discrete grid cells, a fixed time sampling interval is selected. and within the effective time window of the mobile task Internally generated sampling time sequence: , , .in Indicates the index of the time steps of discrete sampling (the first time step). (secondary sampling) Indicates task The total number of sampling steps (maximum index) within its time window is determined by the sampling interval, where For each sampling time Calculate trajectory position and mapped to hierarchy H3 encoding This yields a timestamped grid-coded sequence. To reduce redundancy and preserve trajectory structure, repeated codes occurring at adjacent time points can be merged. The time frame is retained only once to obtain a compact grid sequence representation of the moving target. Finally, the center points of the grid are connected in chronological order to represent the movement trajectory. This method transforms a continuous trajectory into a structured grid sequence, balancing trajectory shape preservation with computational efficiency, and is suitable for spatiotemporal resource matching.

[0026] After uniformly encoding the spatial location information of various raw observation tasks using H3, these tasks are transformed into a task set with H3 grids as the basic unit. However, a single grid may contain multiple raw observation tasks, each with different requirements such as time, priority, and resolution. To effectively aggregate multiple types of raw observation tasks within the same grid and form a unified thermal value representation, this embodiment proposes a thermal value fusion method based on task attribute similarity. This method aggregates multiple raw observation tasks within the same grid and calculates the corresponding grid thermal values.

[0027] For any heatmap grid Its set of original observation tasks is , Represents a grid The set of original observation task indices included. For any original observation task... It includes the following attributes: Task Number Priority A larger value indicates greater importance. Time window requirements. ,in The earliest start time, Latest end time; resolution requirements This represents the lowest resolution level of the imaging. Furthermore, to reflect the differences in fusion between different task types (point / region / movement), this embodiment also introduces a task type weighting coefficient. These correspond to the weight coefficients of point targets, area targets, and moving targets, respectively, and are used to synthesize the priorities after fusion.

[0028] To quantify the fusion compatibility between tasks, this embodiment defines time requirement similarity and resolution requirement similarity, and constructs an overall fusion similarity, specifically: Time requirement similarity reflects the degree of overlap between the time windows of two tasks; the greater the overlap, the higher the similarity. Their time requirement similarity for: (1) in, Indicates the interval length. , Representing tasks ,Task The earliest start time, , Representing tasks ,Task The latest end time for the task ,Task If the time windows do not overlap, then ; Resolution requirement similarity indicates how close the resolution requirements of two tasks are; the closer the values ​​are, the higher the similarity. For task pairs... Their resolution requirements are similar for: (2) in, , Representing tasks ,Task Resolution requirements; The fusion similarity combines both time and resolution requirements, and is expressed as follows: (3) in, Indicates task pair The fusion similarity, when When, it indicates a task. ,Task They cannot be merged within the current grid.

[0029] In this embodiment, the process of clustering and fusing all original observation tasks contained in the same grid is as follows: For grid All task sets included And calculate the task set. The fusion similarity of all task pairs is calculated, and the task pair with the highest non-zero fusion similarity is selected. Merge to generate new tasks and update the task. Task priority Time window Resolution requirements ; After completing one fusion, a task is adopted. Replace the original task pair and calculate the task The similarity between the merged tasks and the remaining tasks is calculated, and then the task pairs with the highest non-zero similarity are selected for fusion again. This process is repeated until the grid is complete. The maximum fusion similarity of any task pair within the middle is 0 (i.e., fusion cannot continue), resulting in a mesh. The clustering and fusion results are divided into several task fusion clusters and several independent tasks. To characterize the fusion strength, each task fusion cluster is defined. Cumulative fusion similarity This is the accumulation of fusion similarity during the fusion process, i.e. .

[0030] Based on the above fusion cluster division, the grid The calculation process for the grid thermal value is as follows: (4) in, Represents a grid The grid thermal value, Represents a grid The set of task fusion clusters within, Indicates task fusion cluster The fusion task priority (i.e., the sum of weighted priorities of types within the cluster). Represents a grid This method quantifies the similarity between task attributes, enabling the aggregation of multiple tasks within the grid. It preserves the key requirements of the original tasks while generating comprehensive heatmap values ​​that can be used for resource allocation decisions.

[0031] Large-scale LEO constellations typically consist of multiple types of satellites and payloads. To achieve unified representation and efficient scheduling of resource capabilities, this embodiment proposes a dynamic encoding method for satellite resource capabilities based on H3 grids. By standardizing grid modeling of the coverage capabilities of different types of satellites in the constellation, a unified representation of different types of satellite resources and different payload resources is achieved, providing a foundation for rapid matching of resources and tasks.

[0032] satellite Its spatial position is determined by the number of its six orbital elements, including the semi-major axis. eccentricity track inclination Right ascension of ascending node perigee argument , near point angle To calculate the satellite at any point during its mission cycle For the area covered by the Earth, its instantaneous position is first determined using orbital mechanics formulas. First, the mean anomaly angle of the satellite at any given time is calculated. : (5) in, Indicates the initial mean anterior angle. This represents the average angular velocity of the orbital motion. Indicates the reference time; Next, the Kepler equations are solved using a numerical iterative method to obtain the anomalous angle. ,for: (6) Ultimately, based on the near-point angle Obtain the satellite's position vector on the orbital plane. ,for: (7) By rotating the orbital plane, the satellite's position vector is transformed from the orbital plane coordinate system to the geocentric inertial coordinate system, thus obtaining the satellite's position in the geocentric inertial coordinate system. ,for: (8) in, , They represent circumference respectively. shaft and The rotation matrix of the axis; Then, the geocentric inertial coordinate system is transformed into a geocentric Earth-fixed coordinate system to obtain the satellite's position in the geocentric Earth-fixed coordinate system. ,for: (9) in, Represents the Earth's rotation angle, approximately , The angular velocity of Earth's rotation. These are the satellite's position coordinates in the Earth-centered Earth-fixed coordinate system; Then the satellite's radial unit vector for: (10) Based on satellite radial unit vector Calculate the satellite's side angle The unit vector of the observation line of sight below ,for: (11) in, Represents the unit vector normal to the orbital plane; Assume the satellite is at position , line of sight The point of intersection with the Earth's sphere for ,in Let represent the scalar parameter from the satellite's position along the line of sight to the Earth's intersection. satisfy , Expressing the Earth's radius, we get: (12) When the discriminant under the square root in equation (12) is non-negative, it means that the line of sight under that side swing angle intersects the Earth (visible). Side yaw angle is effective. Let... This refers to the effective ground swath width of the satellite payload. Because... At the center point of coverage The surrounding area is approximated using a local tangent plane, approximating the visible area covered by the satellite as having a side length centered at that point. A rectangle. latitude and longitude The latitude and longitude increments of the four corner points of the visible area rectangle can be approximately obtained. , for: (13) (14) This allows us to obtain the satellite's position at a given time. Side swing angle The latitude and longitude coordinates of the four vertices of the rectangle corresponding to the visible area below are: (15) By discretizing the lateral swing angle, the set of coverage areas at different lateral swing angles at any given time can be obtained.

[0033] The position of a satellite in a constellation is determined by its orbit, and its coverage area changes over time. Furthermore, for satellites with agile yaw capabilities, different yaw angles at the same time can create different ground coverage areas. To achieve a unified representation of different resource capabilities, this embodiment proposes a dynamic encoding method for resource capabilities, mapping the coverage areas of satellites at different times and under different yaw angles to an H3 grid sequence, i.e. Figure 4 As shown, the specific implementation process includes the following steps: Step 301, Acquire satellite resources The orbital six-axis information, satellite resource payload swath width Task planning time cycle Time step And the discrete set of side-swing angles, and together input into the satellite resource visible area dynamic coding system; Step 302, based on the step size and task planning time cycle Discretize the time period interval into A point in time; Step 303: Calculate satellite resources according to equation (15) The visible region corresponding to each lateral swing angle in the discrete set of each discrete time point and lateral swing angle is decomposed and encoded in the same way as the region target decomposition and encoding. The encoded information is stored together with the corresponding discrete time point and lateral swing angle to obtain the task planning time period. Satellite resources at different times under different side swings Dynamic mesh coding information of load capacity .

[0034] After the above steps, the satellite's dynamic grid coding information can be obtained. It contains both time-varying coverage information in the time dimension and attitude capability information in the side-swing angle. This method unifies the resource capabilities of heterogeneous satellites into a sequence of "time-side-swing angle-grid" pairs, which not only preserves the spatiotemporal dynamic information of capabilities, but also achieves standardization and structuring with the help of the H3 grid system, laying the foundation for rapid resource retrieval and matching in subsequent mission planning.

[0035] The goal of low-Earth orbit (LEO) large-scale constellation mission planning is to maximize mission completion benefits and promote balanced use of satellite resources within the scheduling cycle, while meeting mission requirements and resource constraints. The method in this embodiment addresses a combinatorial optimization problem of multi-satellite, multi-objective allocation and sequencing; therefore, the constructed mission planning model is as follows: (16) (17) (18) in, Let (15) represent the objective function. The first term on the right-hand side of equation (15) is the total task revenue, and the second term is the load balancing penalty term. , Indicates the weighting coefficient. For the number of satellites, Indicates the number of grid cells. Represents decision variables, Represents a grid Assigned to satellite , Represents a grid Unassigned satellite , Represents a grid The grid thermal value, Indicates satellite Normalized task load, Indicates satellite Normalized constellation average load, Indicates satellite Maximum task capacity limit; The task planning model must ensure that the assigned heatmap grid tasks are scheduled for execution within their corresponding visible time windows. Therefore, it must satisfy the task time window constraint, which is: (19) The task planning model must ensure that each transformed heatmap grid task can be executed by at most one satellite to prevent duplicate task allocation. Therefore, it must satisfy the task uniqueness constraint, which is: (20) The mission planning model must ensure that each satellite The total number of assigned tasks cannot exceed its maximum task carrying capacity limit; therefore, satellite resource capacity constraints must be met, namely: (twenty one) The task planning model in this embodiment is a combinatorial optimization problem with visibility and resource constraints. Its solution space grows exponentially with the increase of task size and satellite size, belonging to the NP-hard problem. For the large-scale constellation task planning model constructed in this embodiment, traditional exact algorithms are difficult to obtain feasible solutions within a reasonable time. While metaheuristic algorithms are suitable for such complex problems, their convergence speed is slow and difficult to adapt to the real-time requirements of large-scale constellation task planning. To this end, this embodiment proposes a constraint-guided task planning algorithm (CGH-TS), which uses the constraints of the problem as guidance to narrow the search space and obtain high-quality task planning schemes with lower computational cost. The core idea of ​​CGH-TS is to draw on the classic "Most Constrained First" heuristic strategy. The basic idea is to prioritize tasks with the fewest available resources and the strictest scheduling constraints. These tasks often have limited available satellite resources. If such tasks are not prioritized, they will be difficult to allocate in subsequent scheduling processes, and may even lead to a large number of backtracking adjustments, which may result in a decrease in solution efficiency. The specific process of solving the problem based on the constraint-guided heuristic task planning algorithm in this example is as follows: Step 401, Initialization: Initialize empty allocation scheme Initialize each satellite List of available capabilities status It is used to record the time window occupied, remaining resources, and current attitude information within the task cycle, and to set the unassigned task set. ,in Represents a grid set; For each grid task Construct a set of candidate executable modes in its current state. Candidate elements are defined as executable pattern triples. , indicating at time By satellite With side swing angle Execute grid tasks Then combine Filtering time conflicts and resource constraints, the final result is obtained. ; Step 402, Constraint tightness assessment and task RCL construction: Based on each unassigned task The constraint tightness is calculated based on the size of its candidate set. , and then from Select the one with the tightest constraint The task candidate list consists of [number] tasks. And adopt a randomization strategy from Select the current task being processed ,in This represents a very small constant that prevents division by zero. Step 403, Randomized Greedy Allocation and Dynamic Update: For the selected task traversal Given a candidate set of resources, an evaluation function is used to score the candidates, and the top-scoring candidates are selected. The candidate list consists of 10 candidates. And use roulette to select the final execution mode. and will Add to allocation scheme The process involves marking the time period occupied by the task as occupied, removing the corresponding resources, updating the satellite attitude, and updating the unassigned task candidate set. The evaluation function is as follows:

[0036] in, Represents the evaluation function. Indicates the length of the available time window. Indicates satellite The current load, , , Indicates the weighting coefficient; Step 404, Iteration and Termination: Repeat steps 402 and 403 until... All tasks have been completed, resulting in a feasible programming solution. and repeat. Next, the solution with the optimal objective function value is retained as the optimal task planning scheme. .

[0037] In this embodiment, the time complexity of the constraint-guided heuristic task planning algorithm mainly depends on the following parts: Initializing the candidate set: This requires... Each task is checked separately. One satellite, time complexity is Greedy allocation and dynamic update: In the worst case, for each assigned task, it is necessary to check whether all other unassigned tasks need to be updated in the candidate set. Therefore, the worst-case time complexity of this part is O(log n). ; and repeat simultaneously If the optimal solution is obtained by combining the results, the overall complexity is O(n). In real-world large-scale scenarios, due to the number of tasks... Typically much larger than the number of satellites ,therefore This is the dominant term in terms of complexity. Compared to the exponential complexity of the exact algorithm and the iterative time of the metaheuristic algorithm, this algorithm has a significant advantage in computational efficiency and is suitable for satellite applications requiring large-scale, rapid response.

[0038] To verify the effectiveness of the method in this embodiment for planning large-scale low-Earth orbit constellation missions, three sets of experiments were designed and conducted. All experiments were based on a Walker constellation configuration consisting of 10 orbital planes, each with 24 satellites evenly distributed, totaling 240 satellites. The constellation operates at an altitude of 500 km and an inclination of 60°, carrying a uniformly sized optical imaging payload with an imaging swath of 16 km, enabling revisiting of mid- and low-latitude regions globally. Under this constellation architecture, the three sets of experiments were set as follows: a unified performance test of task requirement representation to verify the effectiveness and operational efficiency of the H3 grid-based unified coding and thermal value fusion method for multiple task types; a unified performance test of resource capability representation to evaluate the accuracy, granularity adaptability, and time-varying feature expression capability of the dynamic coding method for satellite coverage; and a comparison with existing typical algorithms to verify the solution efficiency, task benefits, and resource balance performance of the Constraint-Guided Heuristic Algorithm (CGH-TS) in large-scale scenarios.

[0039] To evaluate the performance of a unified representation method for the requirements of multiple task types, in the eastern longitude... ,north latitude Within a rectangular region, simulation task sets containing point targets, region targets, and moving targets were randomly generated. Each target was assigned a random priority (1–10) and resolution requirements (0.5m, 1m, 2m); the moving targets generated continuous motion trajectories through linear interpolation and discretized into time series points.

[0040] The choice of the H3 grid system hierarchy is a key parameter affecting the granularity of task representation and computational complexity. To evaluate the impact of different hierarchies on task representation, eight scenarios were set up, and ten sets of point target tasks were randomly generated for each scenario. Conversion and fusion experiments were conducted at the 3rd to 8th levels of H3. The average number of original tasks and the average number of grids at each level after fusion (average of 10 experiments) are shown in Table 2.

[0041] Table 2. Number of tasks and grids at each level

[0042] Experimental results show that the low-level grid (3 levels) has a larger coverage area and generates fewer heatmap grid tasks after fusion. Only 13 heatmap grid tasks are generated for 5000 point targets in the 3-level grid, reducing task complexity, but the coarse-grained representation also leads to a loss of spatial accuracy. Conversely, the high-level grid (8 levels) generates a similar number of tasks as the original, with 4863 heatmap grid tasks for 5000 point targets. While this results in higher accuracy, it increases the computational load.

[0043] Overall, level 6 achieves a good balance between accuracy and efficiency. With 5000 target points, it generates 1751.2 heatmap grids (approximately 35% of the original number of tasks), maintaining high spatial accuracy while keeping subsequent computational scale within a reasonable range. Experimental results show that adjusting the level of the H3 grid can flexibly adapt to application scenarios with different accuracy and efficiency requirements.

[0044] This embodiment tested H3 grid clustering with DBSCAN, hierarchical clustering (AC), and attribute value clustering (AVC) on 10 randomly generated point target datasets. The results are as follows: Figure 5 and Figure 6 As shown. The H3 clustering experiment uses the 6th level of H3 clustering to evaluate the efficiency of point-object clustering in the task modeling phase. Through analysis... Figure 5 It is evident that H3 clustering has a significant efficiency advantage, processing 5000 targets in just 0.8 seconds; while DBSCAN and AC take over 2400 seconds, and AVC even exceeds 5112 seconds. Figure 6 It can be seen that the number of clusters based on H3 clustering is better than the compared clustering methods, while its stability is better than DBSCAN and AVC (whose number of clusters increases almost linearly with the number of target clusters). AC requires a preset number of clusters to control the scale, but the computation time is too long. Experiments show that the H3 clustering method can produce clustering results with stable granularity while maintaining high efficiency, and its overall performance is better than traditional clustering algorithms.

[0045] Ten randomly generated regional target datasets were subjected to grid decomposition. The decomposition results at different levels are shown in Table 3.

[0046] Table 3. Experimental Results of Regional Target Decomposition

[0047] Experimental results show that the low-level (3 levels) computation time is only 0.017 seconds, but the number of grids is only 3.3, making it difficult to accurately depict the details of the regions. The high-level (8 levels) computation time reaches 0.365 seconds with 67,774 grids, and the large amount of grid data increases the complexity of subsequent calculations. At level 5, 100 regions were decomposed into 203.4 grids in 0.014 seconds, achieving a good balance between accuracy and efficiency.

[0048] The trajectory of the moving target was discretized into a temporal grid sequence, and the processing results at different levels are shown in Table 4. At level 5, a total of 245 grids were generated, with a processing time of 0.041 seconds and an average distance deviation of 5.91 km, demonstrating superior overall performance. Lower levels (levels 3 and 4) showed larger deviations of 40.46 km and 16.07 km, respectively, while higher levels (levels 7 and 8) had smaller deviations of 0.84 km and 0.32 km, respectively. However, the number of grids increased by 690% and 1821%, respectively, and the processing time increased by 432% and 1500%, respectively.

[0049] Table 4 Meshization Results of Moving Targets

[0050] Experimental results show that the task representation method based on H3 grid can effectively handle regional targets and moving targets. Among them, level 5 can achieve a good balance between accuracy and efficiency in most scenarios, and can provide a reliable foundation for subsequent task planning.

[0051] To verify the effectiveness of the H3 grid-based dynamic coding method for satellite resource capabilities, the focus was on evaluating its advantage in representing time-varying coverage capabilities compared to traditional grid methods. The experimental satellite's orbital parameters were set as follows: semi-major axis 6871.0 km, eccentricity 0.01, orbital inclination 60°, right ascension of the ascending node 60°, argument of perigee 0.0°, and mean perigee 0.0°, with a satellite payload swath width of 16 km. Dynamic grid coding was performed on the satellite resource coverage area under an experimental period of 24 hours and a step size of 60 seconds. The running time and the number of grids generated at different H3 grid levels are shown in Table 5.

[0052] Table 5. Results of Dynamic Grid Coding Efficiency for Resource Capabilities

[0053] Experimental results show that at lower levels (levels 3 and 4), the processing speed is relatively fast, requiring only 0.057 seconds and 0.084 seconds respectively to complete the encoding of the entire day. At intermediate levels (levels 5 and 6), processing tens of thousands to hundreds of thousands of grids takes a reasonable time (0.313 seconds and 1.382 seconds respectively). Even at the highest level (level 8), processing approximately 12.9 million grids, the total time is only 56.726 seconds, significantly more efficient than traditional window calculation methods. The experimental results demonstrate that the dynamic grid coding method successfully transforms the complex spatiotemporal visibility calculation problem into discrete "grid-time" pairs. By generating encoding sequences through preprocessing, the visible time windows of any grid cell and satellites can be efficiently extracted, providing crucial input for subsequent mission planning and meeting the preprocessing efficiency requirements of large-scale constellation scheduling.

[0054] To quantitatively compare and analyze the differences in resource representation between H3 and GeoSOT grids, strip coverage data from the same satellite in low, mid, and high latitude regions were selected for the experiment. Both grid systems were used for encoding (levels 3 to 8 for H3 and levels 10 to 15 for GeoSOT), and the average area of ​​the grids in each region was calculated. The results are shown in Table 6.

[0055] Table 6 Comparison of results from different grid systems

[0056] Experimental results show that the GeoSOT grid suffers from significant high-latitude area shrinkage. For example, at a 10-level grid, the average grid area in high-latitude regions (807.87 km²) is only 27% of that in low-latitude regions (2987.77 km²), a difference of 3.7 times, and this difference intensifies with increasing grid level. In contrast, the H3 grid maintains a highly consistent cell area across different latitudes, with a maximum difference of no more than 7%. Its hexagonal structure effectively avoids high-latitude deformation; the number of grids mapped to the same satellite coverage area at different latitudes is only related to the actual coverage area. The H3 grid's sensitivity to latitude changes is significantly weaker than that of GeoSOT and other encoding methods, thus ensuring consistency in resource capability assessment globally. The results demonstrate that the H3 grid outperforms GeoSOT-based methods in characterizing satellite resource capabilities.

[0057] To evaluate the overall performance of the Constraint-Guided Task Planning Algorithm (CGH-TS) in solving the large-scale constellation mission planning problem in LEO, tests were conducted to compare and analyze it with the Greedy algorithm and the Genetic Algorithm (GA) from two dimensions: algorithm efficiency and allocation effect. The Greedy algorithm is a simple and fast heuristic algorithm that selects the task-resource pair with the highest current benefit for allocation each time, representing an extreme pursuit of solution speed. The Genetic Algorithm (GA) is a classic metaheuristic algorithm widely used to solve satellite mission planning problems, representing a traditional optimization method that sacrifices some computation time for solution quality. An experiment was conducted to construct a Walker constellation with the following seed satellite orbital elements: semi-major axis 6871.0 km, eccentricity 0.01, orbital inclination 60°, right ascension of ascending node 60°, perigee argument 0.0°, and mean perigee angle 0.0°; payload swath width 16 km; maximum side swing angle 30.0°; and planning period 48 hours. By changing the number of satellites in the orbital plane and the mission scale, three typical test scenarios—small, medium, and large—were defined, as shown in Table 7.

[0058] Table 7 Experimental Scenarios

[0059] The algorithm was run 20 times in each scenario, and the average value was calculated. The results of each algorithm under different scale scenarios and different modeling methods are shown in Table 8. In terms of solution efficiency, CGH-TS showed a significant advantage. In small-scale scenarios, its time (0.46s) was comparable to the greedy algorithm (0.48s). As the scale increased, its efficiency advantage became more pronounced: in medium-scale scenarios, CGH-TS (6.2s) was about 90 times faster than the genetic algorithm (560s); in large-scale scenarios, GA struggled to converge within 30 minutes, while CGH-TS only needed 41.8 seconds to obtain a feasible solution, meeting the need for rapid response in large-scale constellations. Regarding solution quality, CGH-TS maintained excellent allocation results while pursuing high efficiency. Its task completion rate and total reward were significantly better than the greedy algorithm (the reward was about 15.6% higher in large-scale scenarios). Compared with the genetic algorithm, CGH-TS has comparable solution quality at small and medium scales (with a difference in returns of <3%). Although its returns are slightly lower than GA at large scales, its solution efficiency is improved by about 40 times.

[0060] Table 8 Results of different algorithms

[0061] Further comparison of the two modeling methods reveals that the unified grid system has advantages in terms of "efficiency and stability," while traditional modeling may achieve slightly higher solution quality in some scenarios, but it consumes more time. In small-scale scenarios, the completion rate (89.2%) and benefits (995) of traditional modeling are slightly higher than those of the H3 grid system (88.7%, 985); however, as the scale increases, the running time of traditional modeling increases (38.5s for medium-scale and 118s for large-scale), and the overall fluctuation of the indicators is greater. In contrast, the H3 grid system reduces the overhead of candidate evaluation and conflict detection through resource capability encoding of "time-swing angle-grid" and fast indexing, enabling the algorithm to obtain high-quality feasible solutions with less time overhead in medium- and large-scale scenarios.

[0062] Experimental results show that the performance improvement mainly comes from the combined effect of two aspects: on the one hand, unified grid modeling and capability encoding bring about candidate retrieval acceleration and dimensionality reduction effect; on the other hand, the constraint-guided allocation strategy of CGH-TS improves the quality of solution and the ability to avoid resource conflicts, thereby further improving the overall solution efficiency in medium and large-scale scenarios.

[0063] The above description is only a preferred embodiment of the present invention and does not limit the scope of protection of the present invention. All equivalent structural transformations made under the inventive concept of the present invention using the contents of the present invention specification and drawings, or direct / indirect applications in other related technical fields, are included within the scope of protection of the present invention.

Claims

1. A unified gridded representation and planning method for mission-resources of a large-scale low-Earth orbit constellation, characterized in that, Includes the following steps: Step 1: Obtain all raw observation tasks to be executed within the planning cycle of the large-scale low-Earth orbit constellation, and use the global discrete grid system to uniformly encode the spatial location information of all types of raw observation tasks. Step 2: Cluster and fuse all the original observation tasks contained in the same grid, and calculate the grid thermal value of each grid based on the clustering and fusion. Step 3: For each satellite in the large-scale low-Earth orbit constellation, map its coverage area under different times and different lateral swings into a grid sequence to obtain dynamic grid coding information of satellite resource payload capacity under different lateral swings at each time point; Step 4: Based on the gridded representation, a task planning model is established with the goal of maximizing the task completion benefits while meeting task requirements and resource constraints. The model is then solved using a constraint-guided heuristic task planning algorithm to obtain the optimal task planning scheme.

2. The method for unified gridded representation and planning of mission-resources for large-scale low-Earth orbit constellations according to claim 1, characterized in that, The objectives of the original observation task include point targets, area targets, and moving targets; In step 1, the step of uniformly encoding the spatial location information of various original observation tasks using a global discrete grid system specifically includes: For the original observation task with point targets, the latitude and longitude coordinates of the corresponding point targets are directly converted into grid codes of a specified level through the coordinate-to-grid mapping function provided by the global discrete grid system. For the original observation task with the target type being a regional target, a recursive subdivision strategy is adopted. Starting from the coarse-level grid, the relationship between the grid and the polygon corresponding to the regional target is determined. The grids that intersect with it are retained and the subdivision continues to a finer level until the preset minimum decomposition level is reached, thereby obtaining the grid code set of the region corresponding to the regional target. For the original observation task with a moving target type, time sampling is performed at fixed intervals within its effective time window to obtain a series of discrete moments and corresponding trajectory point positions. Each trajectory point position is converted into a grid code of a specified level through a coordinate-to-grid mapping function to obtain a grid code sequence with timestamps, and duplicate codes of adjacent moments are merged.

3. The method for unified gridded representation and planning of mission-resources for large-scale low-Earth orbit constellations according to claim 1, characterized in that, In step 2, the process of clustering and fusing all the original observation tasks contained in the same grid specifically involves: For grid All task sets included And calculate the task set. The fusion similarity of all task pairs is calculated, and the task pair with the highest non-zero fusion similarity is selected. Merge to generate new tasks and update the task. Task priority, time window, and resolution requirements; After completing one fusion, a task is adopted. Replace the original task pair and calculate the task The similarity between the merged tasks and the remaining tasks is calculated, and then the task pairs with the highest non-zero similarity are selected for fusion again. This process is repeated until the grid is complete. The maximum fusion similarity of any task pair within the middle is 0, resulting in a mesh. The clustering and fusion results are used to form several task fusion clusters and several independent tasks.

4. The method for unified gridded representation and planning of mission-resources for large-scale low-Earth orbit constellations according to claim 3, characterized in that, For task pairs The calculation process for its fusion similarity includes: Computational tasks With the task Time requirement similarity ,for: in, Indicates the interval length. , Representing tasks ,Task The earliest start time, , Representing tasks ,Task The latest end time; Computational tasks With the task similarity of resolution requirements ,for: in, , Representing tasks ,Task Resolution requirements; Based on time demand similarity Similarity to resolution requirements Computational task pairs The fusion similarity is: in, Indicates task pair The fusion similarity.

5. The method for unified gridded representation and planning of mission-resources for large-scale low-Earth orbit constellations according to claim 3 or 4, characterized in that, The calculation process for the grid thermal value is as follows: in, Represents a grid The grid thermal value, Represents a grid The set of task fusion clusters within, Indicates task fusion cluster The priority of fusion tasks Indicates task fusion cluster The cumulative fusion similarity during the fusion process. Represents a grid A collection of independent tasks within, Indicates task Task priority.

6. The method for unified gridded representation and planning of mission-resources for large-scale low-Earth orbit constellations according to claim 3, characterized in that, Step 3 is as follows: Step 301, Acquire satellite resources The orbital six-axis information, satellite resource payload swath width Task planning time cycle Time step And the discrete set of lateral swing angles; Step 302, based on the step size and task planning time cycle Discretize the time period interval into A point in time; Step 303, Calculate satellite resources For each discrete time point and the corresponding visible region under each lateral swing angle in the discrete set of lateral swing angles, the visible region is decomposed and encoded using a global discrete grid system. The encoded information is then stored together with the corresponding discrete time point and lateral swing angle to obtain the task planning time period. Satellite resources at different times under different side swings Dynamic mesh encoding information for load capacity.

7. The method for unified gridded representation and planning of mission-resources for large-scale low-Earth orbit constellations according to claim 6, characterized in that, In step 303, the calculation process for the visible area includes: Calculate the radial unit vector of the satellite ,for: in, This represents the satellite's position vector in the Earth-centered, Earth-fixed coordinate system. Based on satellite radial unit vector Calculate the satellite's side angle The unit vector of the observation line of sight below ,for: in, Represents the unit vector normal to the orbital plane; Assume the satellite is at position , line of sight The point of intersection with the Earth's sphere for ,in Let represent the scalar parameter from the satellite's position along the line of sight to the Earth's intersection. satisfy , Expressing the Earth's radius, we get: At the intersection The surrounding area is approximated using a local tangent plane, approximating the visible region as a region centered on the intersection point. Centered on, with side length as Let the rectangles intersect at points. The latitude is The latitude and longitude increments of the four corner points of the visible area rectangle are then calculated. , for: Thus, the satellite's time point is obtained. Side swing angle The visible area below corresponds to the latitude and longitude of the four vertices of the rectangle. .

8. The method for unified gridded representation and planning of mission-resources for large-scale low-Earth orbit constellations according to any one of claims 1 to 4, characterized in that, The task planning model is specifically as follows: in, Describe the objective function. , Indicates the weighting coefficient. For the number of satellites, Indicates the number of grid cells. Represents decision variables, Represents a grid Assigned to satellite , Represents a grid Unassigned satellite , Represents a grid The grid thermal value, Indicates satellite Normalized task load, Indicates satellite Normalized constellation average load, Indicates satellite The maximum task capacity limit.

9. The method for unified gridded representation and planning of mission-resources for large-scale low-Earth orbit constellations according to claim 8, characterized in that, The specific process of solving the constraint-guided heuristic task planning algorithm is as follows: Step 401, Initialization: Initialize empty allocation scheme Initialize each satellite List of available capabilities status It is used to record the time window occupied, remaining resources, and current attitude information within the task cycle, and to set the unassigned task set. ,in Represents a grid set; For each grid task Construct a set of candidate executable modes in its current state. Candidate elements are defined as executable pattern triples. , indicating at time By satellite With side swing angle Execute grid tasks Then combine Filtering time conflicts and resource constraints, the final result is obtained. ; Step 402, Constraint tightness assessment and task RCL construction: Based on each unassigned task The constraint tightness is calculated based on the size of its candidate set. , and then from Select the one with the tightest constraint The task candidate list consists of [number] tasks. And adopt a randomization strategy from Select the current task being processed ,in This represents a very small constant that prevents division by zero. Step 403, Randomized Greedy Allocation and Dynamic Update: For the selected task traversal Given a candidate set of resources, an evaluation function is used to score the candidates, and the top-scoring candidates are selected. The candidate list consists of 10 candidates. And use roulette wheel to select the final execution line mode. and will Add to allocation scheme The system also marks the time period occupied by the task as occupied, removes the corresponding resources, updates the satellite attitude, and updates the unassigned task candidate set. Step 404, Iteration and Termination: Repeat steps 402 and 403 until... All tasks have been completed, resulting in a feasible programming solution. and repeat. Next, the solution with the optimal objective function value is retained as the optimal task planning scheme. .

10. The method for unified gridded representation and planning of mission-resources for large-scale low-Earth orbit constellations according to claim 9, characterized in that, The evaluation function is specifically as follows: in, Represents the evaluation function. Indicates the length of the available time window. Indicates satellite The current load, , , This represents the weighting coefficient.

Images