Method and system for scheduling large-scale agile constellation multi-type tasks

Abstract

The application provides a large-scale agile constellation multi-type task scheduling method and system, and relates to the technical field of task scheduling. By uniformly converting point target tasks and area target tasks into candidate imaging strips and constructing a unified candidate task set, the problem of difficult unified processing of different types of tasks is effectively solved. By introducing a slidable time window and detailed attitude maneuver constraints, the flexibility of agile satellites can be fully utilized to generate a scheduling scheme that is more in line with actual observation capabilities. The construction of a static heuristic graph and multi-factor heuristic sorting, based on an iterative optimization mechanism of a joint operator, enables the scheduling process to adaptively select an optimization strategy according to the current search state, thereby effectively improving the quality and optimization efficiency of the scheduling solution under complex constraint conditions. The method can provide a more comprehensive, flexible and efficient multi-type task scheduling solution for a large-scale agile constellation.

Description

Technical Field

[0001] This invention relates to the field of task scheduling technology, and more specifically, to a scheduling method and system for large-scale agile constellations with multiple types of tasks. Background Technology

[0002] Existing satellite mission scheduling methods typically model and solve for point-target missions, regional-target missions, or multiple-observation missions separately. For example, for point-target missions, a mission selection and ranking model based on visible time windows is established; for regional-target missions, a coverage optimization model based on grids or coverage is established; and for multiple-observation missions, additional constraints such as revisit intervals or observation counts are established. Because different mission types differ in spatial representation, coverage relationships, resource consumption, and time window constraints, existing methods struggle to uniformly transform point-target, regional-target, and multiple-observation missions into consistent schedulable objects, leading to high complexity in subsequent mission allocation, ranking, and resource constraint processing.

[0003] For agile satellites, attitude maneuvers can enable yaw observation, forward and backward observation, multi-point continuous observation, and multi-strip splicing observation. Therefore, the mission scheduling of agile satellites needs to consider more factors, and the existing satellite mission scheduling methods based on visible time windows are not suitable for scheduling agile satellites. Summary of the Invention

[0004] The problem that this invention aims to solve is that existing satellite mission scheduling methods based on visible time windows are difficult to unify point targets, regional targets, and multiple observation tasks into a consistent schedulable object, and are not suitable for scheduling agile satellites.

[0005] To address the aforementioned problems, in a first aspect, the present invention provides a scheduling method for large-scale agile constellations with multiple task types, comprising: For point target tasks, based on the location of the point target and the operational information of the agile satellite, the sliding time window of the point target on the corresponding agile satellite is calculated, and candidate imaging strips of the point target are generated. For regional target missions, under the visible transit opportunity of agile satellites, the grid set of the covered target area is analyzed to obtain regional target candidate imaging strips and corresponding sliding time windows. Among them, the target area in the regional target mission is discretized and divided into multiple grids. All candidate imaging strips corresponding to the same task are grouped into the same set to form a candidate task set; Based on the candidate task set, a static heuristic graph is constructed, and the task scarcity of each task and the task conflict between tasks in the static heuristic graph are analyzed. Based on the mission observation benefits, the contribution of new regional target coverage, mission scarcity, and insertion compactness, candidate missions are heuristically ranked and inserted into the mission sequence of the corresponding agile satellites to obtain a feasible scheduling solution. Select a joint operator to iteratively optimize the current feasible scheduling solution, perform time-series feasibility verification on the optimized feasible scheduling solution, obtain the current optimal solution and output it as the final scheduling scheme.

[0006] Secondly, the present invention provides a scheduling system for large-scale agile constellations with multiple task types, comprising: The point target task processing module is used to calculate the sliding time window of the point target on the corresponding agile satellite based on the position of the point target and the operation information of the agile satellite, and generate candidate imaging strips for the point target. The regional target task processing module is used to analyze the grid set of the covered target area for regional target tasks under the visible transit opportunity of agile satellites, and obtain regional target candidate imaging strips and corresponding sliding time windows. The target area in the regional target task is discretized and divided into multiple grids. The map analysis module is used to group all candidate imaging strips corresponding to the same task into the same set to form a candidate task set; it is also used to construct a static heuristic map based on the candidate task set and analyze the task scarcity of each task and the task conflict between tasks in the static heuristic map. The scheduling solution generation module is used to heuristically sort candidate tasks based on task observation benefits, contribution of new coverage to regional targets, task scarcity and insertion compactness, and insert the candidate tasks into the task sequence of the corresponding agile satellite to obtain a feasible scheduling solution. The scheduling solution optimization module is used to select joint operators, iteratively optimize the current feasible scheduling solution, perform time-series feasibility verification on the optimized feasible scheduling solution, obtain the current optimal solution and output it as the final scheduling scheme.

[0007] This invention provides a scheduling method and system for large-scale agile constellations with multiple task types. Compared with existing technologies, it has the following advantages: By unifying point target tasks and regional target tasks into candidate imaging strips and constructing a unified candidate task set, this method effectively solves the problem of existing methods struggling to handle different types of tasks uniformly. By introducing sliding time windows and detailed attitude maneuver constraints, this method can fully utilize the flexibility of agile satellites to generate scheduling schemes that better match actual observation capabilities. The construction of static heuristic maps and multi-factor heuristic ranking provide efficient guidance for the initial allocation of large-scale tasks. More importantly, the iterative optimization mechanism based on joint operators enables the scheduling process to adaptively select optimization strategies according to the current search state, thereby effectively improving the quality and optimization efficiency of scheduling solutions under complex constraints. Compared to existing scheduling methods based solely on visible time windows or single task types, this method provides a more comprehensive, flexible, and efficient multi-type task scheduling solution for large-scale agile constellations. Attached Figure Description

[0008] 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 these drawings without creative effort.

[0009] Figure 1 A flowchart illustrating a scheduling method for large-scale agile constellations with multiple task types, provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the operation of the random destruction operator provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the operation of the low-yield destruction operator provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the operation of the high-conflict breaking operator provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the operation of the redundancy destruction operator provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of the operation of the greedy repair operator provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the operation of the random insertion operator provided in an embodiment of the present invention; Figure 8 This is a schematic diagram of the structure of a large-scale agile constellation multi-type task scheduling system provided in an embodiment of the present invention. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions in the embodiments of this application are described clearly and completely. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0011] The technical solution in this application is to solve the above-mentioned technical problems, and the general idea is as follows: The purpose of this invention is to provide a scheduling method for large-scale agile constellations with multiple types of tasks. Through an integrated process of "multi-type task standardization - region coverage discrete mapping - static heuristic graph - context-aware joint operator selection - lightweight feasibility verification", the solution efficiency and solution quality of large-scale task scheduling are improved while ensuring scheduling executability.

[0012] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0013] The scheduling problem for large-scale agile satellite constellations involves both point and regional targets. Due to significant differences in spatial morphology, visibility window, observation duration, attitude pointing, and resource consumption between the two types of targets, unified modeling is difficult and the combined search space expands dramatically. Therefore, a two-stage transformation preprocessing framework is adopted to represent point and regional targets as standard imaging strips, thus transforming the original problem into an imaging strip selection and scheduling optimization problem based on time window constraints.

[0014] The transformed scheduling problem is defined as follows: Given a constellation system consisting of hundreds of agile satellites, under constraints such as satellite visibility window, attitude maneuverability, imaging duration, energy and storage capacity, and regional coverage, the scheduling scheme that maximizes the overall observational benefits is jointly optimized by selecting imaging strips, allocating satellites, and determining the execution timing. For ease of modeling and algorithm design, the following basic assumptions are made: (1) All the satellites in the constellation are isomorphic agile satellites, and each satellite carries the same type of imaging payload.

[0015] (2) The downlink scheduling of observation data and the visibility constraints of ground stations are not considered.

[0016] (3) Once the imaging task has started, it cannot be interrupted.

[0017] (4) Point targets and regional targets are independent of each other and there is no sequential dependency relationship.

[0018] (5) All observed targets are static targets.

[0019] The method provided in this application consists of two parts: a task preprocessing stage and a strip scheduling optimization stage.

[0020] The first part is the mission preprocessing stage, which is used to uniformly convert various types of observation targets, such as point targets and regional targets, into candidate imaging strips. In this stage, based on parameters such as satellite orbit, payload swath width, maximum roll angle, maximum elevation angle, imaging duration, and planned time domain, a sliding observation time window is generated for point targets, and the point targets are represented as short-term degraded imaging strips. Regional targets are discretized into a grid, and candidate imaging strips are generated by combining satellite transit, payload field of view, attitude angle constraints, and imaging duration constraints. Subsequently, the regional candidate strips are deduplicated, pruned, divided into strip families, and subjected to multiple combination screenings, outputting a unified candidate strip set, a strip-grid coverage mapping, and a regional target strip combination scheme.

[0021] The second part is the stripe scheduling optimization stage, which is used to schedule satellite missions for the candidate imaging strips generated in the mission preprocessing stage. This stage uses the candidate imaging strips as a unified scheduling object, and determines whether each strip should be executed, which satellite should execute it, and its execution order in the corresponding satellite mission sequence, under constraints such as visible time window, attitude transition time, power capacity, storage capacity, and area coverage requirements. This stage employs a context-aware adaptive large neighborhood search method for optimization, generating a constellation scheduling scheme that meets the constraints through static heuristic graphs, context-based multi-armed slot machine joint operator selection, neighborhood reconstruction, and temporal feasibility verification.

[0022] like Figure 1 As shown in the embodiment of this application, a scheduling method for large-scale agile constellations with multiple task types is provided, including: S1: For point target missions, based on the location of the point target and the operational information of the agile satellite, calculate the sliding time window of the point target on the corresponding agile satellite and generate candidate imaging strips for the point target.

[0023] Specifically, based on the geographic coordinates of the point target and the orbital parameters of the agile satellite, the visible time period of the satellite over the point target can be determined through geometric calculations, and this time period can be used as a preliminary estimate of a sliding time window. Based on this, one or more candidate imaging strips for the point target can be generated, each strip containing a fixed imaging duration and a corresponding start / end time.

[0024] S2: For regional target missions, under the visible transit opportunity of agile satellites, the grid set of the covered target area is analyzed to obtain regional target candidate imaging strips and corresponding sliding time windows. The target area in the regional target mission is discretized and divided into multiple grids.

[0025] Specifically, the target area can be divided into a uniform square grid, and the coverage can be determined by simply calculating the intersection of the satellite footprint with these grids. Alternatively, several standard-shaped imaging stripes can be predefined, and attempts can be made to place these standard stripes on the target area to generate candidate imaging stripes for the region. This approach may not fully utilize the attitude maneuverability of agile satellites to generate stripes of optimal shape.

[0026] S3: Group all candidate imaging strips corresponding to the same task into the same set to form a candidate task set.

[0027] Specifically, candidate imaging strips for all point target tasks and candidate imaging strips for all region target tasks can be simply aggregated into a single list. This direct aggregation approach may result in a large number of redundant or similar candidate strips in the set, thereby increasing the complexity of subsequent processing.

[0028] S4: Based on the candidate task set, construct a static heuristic graph and analyze the task scarcity of each task and the task conflict between tasks in the static heuristic graph.

[0029] Specifically, a simple undirected graph can be constructed, where nodes represent candidate tasks and edges indicate some kind of relationship between tasks. Task scarcity can be simply obtained by counting the number of candidate imaging strips corresponding to each task. Task conflict can be initially determined by checking whether any two tasks overlap within the same time period on the same satellite. This preliminary conflict assessment may not be accurate enough, failing to fully consider details such as the attitude transition time of agile satellites.

[0030] S5: Based on the mission observation benefits, the contribution of new coverage to regional targets, mission scarcity, and insertion compactness, candidate missions are heuristically sorted and inserted into the mission sequence of the corresponding agile satellites to obtain a feasible scheduling solution.

[0031] Specifically, a simple greedy strategy can be adopted, selecting the highest-priority task each time and inserting it into the first available satellite mission sequence. Mission observation benefit refers to the value or benefit gained from completing a task, such as economic, scientific, or strategic value, and can be allocated according to a preset fixed value. Regional target coverage contribution refers to the incremental contribution to the overall regional target coverage after a candidate imaging strip is executed during scheduling, which can be simply evaluated by calculating the number of grid cells covered by the strip. Insertion compactness refers to the impact of inserting a candidate task into the existing mission sequence of the agile satellite on the compactness or efficiency of the existing sequence; it can be simply defined as the total duration of the mission sequence after insertion. High compactness usually means that the insertion operation has less disturbance to the existing scheduling scheme or can better utilize time resources. This simple heuristic ranking may not fully balance multiple factors, leading to a suboptimal scheduling solution.

[0032] S6: Select a joint operator to iteratively optimize the current feasible scheduling solution, perform time-series feasibility verification on the optimized feasible scheduling solution, obtain the current optimal solution and output it as the final scheduling scheme.

[0033] Specifically, an iterative optimization can be performed using a set of destruction and repair operators. The destruction operator can randomly remove a portion of the scheduled tasks, while the repair operator can randomly insert unscheduled tasks. Timing feasibility verification can check only whether the task time windows overlap, ignoring more complex constraints such as attitude transition time and energy consumption.

[0034] In this optional embodiment, by unifying point target tasks and regional target tasks into candidate imaging strips and constructing a unified candidate task set, the problem of existing methods struggling to uniformly process different types of tasks is effectively solved. By introducing a sliding time window and detailed attitude maneuver constraints, this method can fully utilize the flexibility of agile satellites to generate scheduling schemes that better match actual observation capabilities. The construction of static heuristic maps and multi-factor heuristic sorting provide efficient guidance for the initial allocation of large-scale tasks. More importantly, the iterative optimization mechanism based on joint operators enables the scheduling process to adaptively select optimization strategies according to the current search state, thereby effectively improving the quality and optimization efficiency of scheduling solutions under complex constraints. Compared with existing scheduling methods based only on visible time windows or single task types, this method can provide a more comprehensive, flexible, and efficient multi-type task scheduling solution for large-scale agile constellations.

[0035] The following is a detailed description of each step.

[0036] Phase 1: Task Preprocessing and Unified Imaging Strip Generation S101: Input Task and Satellite Resource Data: Obtain the set of agile satellites, point targets, and regional targets to be scheduled. The satellite set includes information such as satellite name, number of orbital elements, payload swath width, orbital altitude, maximum roll angle, and maximum elevation angle. The point target set includes target number, longitude, and latitude. The regional target set includes regional name and coordinates of regional boundary points. Simultaneously, set the planning start time, planning duration, grid scale, imaging swath width, minimum imaging duration, maximum imaging duration, and candidate strip number control parameters.

[0037] S102: Construct a region polygon based on the boundary points of the target region, discretize the target region according to the preset grid scale, generate the corresponding grid, and assign a globally unique grid number to each grid.

[0038] Specifically, for each regional target, a regional polygon is constructed based on its boundary points, and the regional target is discretized according to a preset grid scale. For grid points falling inside the regional polygon, corresponding grid cells are generated, and each grid cell is assigned a globally unique grid number. Through this step, continuous regional targets are transformed into regional coverage objects composed of several discrete grid cells.

[0039] S103: For each agile satellite and each target area: The agile satellite's orbit is propagated using its orbital six-element data to obtain its spatial position within the mission cycle. Combined with the geographic coordinates of the target area, the relative positional relationship between the agile satellite and the target area is calculated, and the effective observation time range during the agile satellite's transit is determined based on the agile satellite's performance constraints. Within the effective observation time range, the moment when the agile satellite is closest to the target area or has the largest observation elevation angle is selected as the transit peak moment. A local ground coordinate system for satellite transit is established based on the satellite-target area geometric relationship at the transit peak moment. The grid cells or point targets of the target area are projected onto the local ground coordinate system to facilitate the construction of stripes and the calculation of the agile satellite's attitude angles.

[0040] Specifically, a local coordinate system for satellite transit is established: For each satellite and each target region, the visible transit opportunity of the satellite over the target region is calculated based on the satellite orbit information, and a local ground coordinate system is established by selecting the peak transit time. This coordinate system includes an orbital axis along the direction of the satellite's ground trajectory and a transverse axis perpendicular to the direction of the ground trajectory. Subsequently, the positions of regional grid points or point targets are projected into this local coordinate system for subsequent strip geometry construction and attitude angle calculations.

[0041] First, by meticulously dividing and uniquely numbering the target area, a unified and traceable foundation of data is provided for subsequent scheduling and coverage analysis. Next, through precise orbit propagation and relative position calculations, combined with the performance constraints of agile satellites, truly feasible observation periods and optimal observation times are selected. Based on this, a local ground coordinate system is established, and the target is projected into this coordinate system, greatly simplifying the geometric construction of imaging strips and the calculation of satellite attitude angles. This systematic preprocessing workflow provides highly accurate and reliable input data for the subsequent generation of candidate imaging strips for point targets, ensuring that the generated strips meet the actual observation requirements in terms of time, space, and attitude, thus laying a solid foundation for the scheduling of multiple types of missions in large-scale agile constellations. Before generating candidate imaging strips for point targets, the target area can be precisely geometrically defined and discretized, and the observation capabilities of agile satellites can be comprehensively evaluated. This not only ensures the high accuracy and feasibility of the subsequently generated candidate imaging strips, avoiding invalid scheduling due to geometric or temporal mismatches, but also greatly simplifies the complexity of strip construction and attitude angle calculations by establishing a local ground coordinate system, improving the efficiency of the preprocessing stage. This refined preprocessing provides a solid data foundation and computational support for scheduling large-scale agile constellations with multiple task types, thereby improving the robustness and optimization effect of the entire scheduling method.

[0042] S1: For point target missions, based on the point target's location and the agile satellite's operational information, calculate the sliding time window of the point target on the corresponding agile satellite, and generate candidate imaging strips for the point target. This step specifically includes the following:

[0043] S110: For point target tasks, treat them as short-term degradation imaging strips.

[0044] S111: Calculate the sliding time window of the point target on the corresponding satellite based on the point target's position relative to the satellite ground trajectory, the satellite ground velocity, the maximum roll angle, and the maximum pitch angle.

[0045] S112: If the point target satisfies the constraints of roll angle, pitch angle and simulation time range, then generate a candidate imaging strip for the point target. The candidate imaging strip for the point target includes information such as mission number, satellite number, earliest start time, latest end time, fixed imaging duration, roll angle, initial pitch angle, end pitch angle, nominal start time and nominal end time.

[0046] Specifically, establish the geometric relationship between the point target and the satellite ground trajectory: Let the geographic coordinates of the point target be... ,in, The longitude of the target point; The latitude of the target point; The target height is [height].

[0047] The satellite's position at time t is obtained from satellite orbit propagation. , , and Let each represent the three-dimensional coordinates of the satellite's position at time t; and calculate the satellite's nadir point. , and These represent the longitude and geodetic latitude of the nadir point, respectively; the ground distance between the target and the satellite's nadir point is... This distance represents the spatial offset between the satellite's ground trajectory and the target.

[0048] Calculate the lateral offset of the target relative to the satellite trajectory. Establish a local coordinate system for the satellite transit and project the target onto the local orbital coordinate system. , Indicates the distance of the target along the track direction. This indicates the distance along the target's horizontal orbit. The horizontal orbit offset determines whether the satellite needs to side-swing. H represents the satellite's orbital altitude. The required roll angle is represented by time t.

[0049] The observable range is determined based on the satellite's attitude capabilities. Agile satellites have a maximum roll angle constraint. , Indicates the demand roll angle. This represents the maximum roll angle, therefore the objective satisfies... This allows for satellite observation. From this, the permissible lateral coverage area of ​​the satellite can be determined. ,like Then the point target belongs to the candidate observation target of the satellite.

[0050] Calculate the possible starting time range for the point target to slide. The satellite's velocity along its orbit is v. g The target's distance along the orbit is x p Corresponding time offset Considering the imaging duration is T img Then the start time range of the candidate imaging strip corresponding to the point target Where t0 represents the time of the satellite's maximum elevation angle transit. This indicates the imaging start time. This time range is the sliding start time window for the point target. Generate candidate imaging strips for point targets. Since the area of ​​point targets approaches zero, they are represented as degenerate strips. , where P p L represents the center position of the strip, i.e., the point target position. p W represents the strip length. p Indicates the effective width, t s and te These represent the start and end times of the sliding time range, respectively. For point targets... Finally obtained As a set of candidate tasks for the subsequent scheduling phase, among which, This represents the m1th candidate task.

[0051] S113: After all point target candidate imaging strips for agile satellites are generated, the point target candidate imaging strips corresponding to the same point target mission are screened to obtain fewer than the preset number of point target candidate imaging strips.

[0052] Specifically, candidate observation opportunities for point targets are screened: After candidate imaging strips for point targets from all satellites are generated, a global screening is performed on the candidate observation opportunities corresponding to the same point target. During screening, candidate strips with smaller roll angles, wider sliding time windows, smaller elevation angles, and better time windows are prioritized. A maximum of a preset number of candidate observation opportunities are retained for each point target, thereby reducing the candidate scale in subsequent scheduling phases. Screening candidate imaging strips for point targets ensures that only the most valuable observation opportunities are considered for scheduling, avoiding the impact of suboptimal selection on scheduling efficiency.

[0053] By abstracting point target tasks into short-term degraded imaging strips and accurately calculating their sliding time windows using agile satellite operational information and attitude constraints, this application can generate candidate imaging strips containing complete scheduling information for point target tasks. This ensures that subsequent scheduling algorithms can perform task sorting, conflict detection, and sequence insertion based on accurate and feasible observation parameters when processing point target tasks, thereby improving the accuracy and feasibility of the scheduling scheme and avoiding scheduling failures or inefficiencies caused by incomplete or inaccurate information.

[0054] S2: For regional target missions, under the visible transit opportunity of agile satellites, the grid set of the covered target area is analyzed to obtain regional target candidate imaging strips and corresponding sliding time windows. The target area in the regional target mission is discretized and divided into multiple grids.

[0055] Specifically, candidate imaging strips for regional targets are generated: For regional targets, under the corresponding satellite transit opportunity, the lateral position and start and end positions along the orbit of candidate strips are enumerated based on the local along-orbit / lateral coordinate system. For each candidate geometric strip, the set of grid cells covering its region is calculated, and the required roll angle, starting elevation angle, ending elevation angle, and imaging duration are inversely calculated based on the strip geometry. If the candidate strip satisfies constraints such as maximum roll angle, maximum elevation angle, minimum imaging duration, and minimum number of covering grid cells, it is adopted as a candidate imaging strip for regional targets. This step specifically includes the following:

[0056] S210: For regional target missions, under the corresponding agile satellite transit opportunity, enumerate the lateral position and start and end positions along the track of candidate geometric strips based on the local ground coordinate system.

[0057] S211: For each candidate geometric stripe, calculate the set of grids covering its area, and inversely deduce the required roll angle, initial pitch angle, final pitch angle, and imaging duration based on the geometric relationship of the candidate geometric stripe.

[0058] S212: If a candidate geometric strip satisfies the constraints of maximum roll angle, maximum pitch angle, minimum imaging duration, maximum imaging duration, and minimum number of coverage grids, then it is taken as a candidate imaging strip for the regional target.

[0059] S213: Based on the maximum pitch angle constraint, the continuous feasible range of the strip pitch zero point is inferred, and the continuous feasible range is converted into the earliest and latest start times of the regional target candidate imaging strip, forming a sliding time window.

[0060] Specifically, for regional target A and agile satellite s, the input parameters include ,in, This is the Nth boundary point of the regional target boundary. Satellite and payload parameters include satellite orbital altitude H and satellite ground velocity v. g Maximum roll angle Maximum pitch angle Camera horizontal rail field of view Camera half field of view Let the peak transit time be t0.

[0061] Transform the regional target into the transit local coordinate system. Establish a local ground coordinate system at the peak transit time t0: the X-axis is along the satellite ground trajectory, the Y-axis is the horizontal trajectory, and the Z-axis is the local zenith direction. Project each boundary point of the regional target into the local coordinate system to obtain the j-th boundary point of the regional target. ,in, Let j be the coordinates of the j-th boundary point along the track. Let the x-axis coordinates of the j-th boundary point be used to represent the region target as a two-dimensional polygon. ,in, and Let represent the coordinates along the orbital plane and the coordinates along the horizontal orbital plane of the Nth boundary point, respectively. The orbital plane and horizontal orbital plane ranges of the regional target are as follows: , in, This represents the lower limit of the coordinates of the boundary point along the track. This represents the upper limit of the coordinates of the boundary point along the track. This represents the lower limit of the horizontal coordinates of the boundary point. This indicates the upper limit of the horizontal coordinates of the boundary point.

[0062] The target area is discretized into a grid. Let the grid side length be... A regular mesh is constructed within the bounding rectangle of the region target. The m-th region mesh is represented as... , Let m be the base coordinates along the track of the m-th grid. Given the base horizontal coordinates of the m-th grid, retain the grids that intersect with the region target. For the boundary grid, the area that actually belongs to the regional target is The total area of ​​the target region can be approximately calculated as follows: Each subsequent candidate strip is described by the set of grids it covers.

[0063] Determine the orientation of candidate strips. Generally, the longitudinal direction of the imaging strip is taken as the direction of the satellite's ground trajectory. Let the orientation angle of the candidate strip be θ. Its unit vector along the strip direction is The unit vector of the horizontal strip direction perpendicular to it is For cases where yaw adjustment is not considered, the candidate strip direction angle is taken. , This refers to the orientation angle of the satellite's ground trajectory, meaning the strip direction aligns with the satellite's ground trajectory direction. When considering yaw maneuvers, multiple candidate strip orientation angles can be discretized and generated within acceptable limits. ,in, Let be the direction angle from the walk distance. For any boundary point... Its vertical coordinate in the candidate strip coordinate system and horizontal coordinates They are respectively: .

[0064] The ground width is calculated based on the roll angle and field of view. For candidate roll angles... The roll angle can be generated discretely, and the generated l-th roll angle , , For the long walk, For the discrete maximum number of points. Under the local plane approximation, the intersections of the two edge lines of the camera's field of view with the ground are respectively , and Let these represent the intersections of the two edge lines of sight with the ground. To avoid the influence of conventions regarding positive and negative directions, we can define them as follows: , This represents the lower limit of the horizontal rail position for ground imaging. Given the upper limit of the horizontal rail position for ground imaging, the ground imaging swath width corresponding to this roll angle is: The horizontal rail position of the center line of the strip is... Therefore, the infinitely long imaging band region corresponding to this pose is P represents the boundary point of the target region. Candidate strips are only generated when the imaging band intersects with the target region, i.e., when the condition is met. , For an empty set, the condition can also be written as the interval intersection condition: ,in: , The x-axis coordinates of the target area. The minimum horizontal coordinate of the target area. This represents the maximum horizontal coordinate of the target area.

[0065] Calculate the intersection of the strip and the target area. Find the intersection of the candidate infinite imaging strip and the target area to obtain the intersection region. ,when If the candidate strip is not empty, delete it. For non-empty intersections, calculate the minimum coordinates of the intersection region along the longitudinal direction of the strip. and maximum coordinates , To ensure complete imaging of the boundary, a safety margin can be added at both ends. The minimum corrected coordinates of the intersection region along the longitudinal direction of the strip are obtained. and corrected maximum coordinates , Thus, the candidate strip length is obtained. .

[0066] Construct the geometric boundary of the candidate strip. The candidate strip can be represented as a rectangle or a curvilinear quadrilateral. Under the local planar approximation, its four vertices are: , This forms candidate strip polygons. Polygon is the polygon constructor, whose centerline originates from... and the end point They are respectively: Therefore, each candidate strip can be described by the following geometric parameters. .

[0067] For candidate regional strips that satisfy geometric and attitude constraints, the continuous feasible range of the strip's pitch zero point is inferred from the maximum pitch angle constraint, and this range is converted into the earliest and latest possible start times for the strip. This forms candidate regional imaging strips with sliding time windows, allowing subsequent scheduling phases to select the actual execution time within the time window, rather than using only fixed observation times. Attitude constraints mainly include roll angle constraints, pitch angle constraints, and yaw angle constraints when considering yaw imaging.

[0068] Calculate the duration of strip imaging. Assume the ground scanning velocity during satellite imaging is... Then the duration of strip imaging When the direction of the stripe is consistent with the direction of the satellite's ground trajectory, it can be approximated as... .therefore When considering load start-up time or initial and final safety time, the safe duration for strip imaging is: , This refers to the boot time or initial safe time. This is the shutdown time or the end of the safety period.

[0069] Calculate the set of grid covers for candidate stripes. For each region grid G... m Calculate the overlap area between it and the candidate strips. You can define the continuous coverage ratio. You can also define binary coverage parameters. ,in, The threshold for determining grid coverage is set as follows: The set of grids for candidate stripe coverage is... The coverage rate of this strip for regional targets is [percentage missing]. When using a two-dimensional mesh approximation, it can also be written as .

[0070] Calculate the sliding start time window for the strip. For candidate strips, it is necessary to ensure that the continuous imaging process from the start to the end of the strip satisfies the maximum elevation angle constraint. Select key locations on the strip. , Let represent the coordinates of the Mth key location, where M is the total number of key locations. Key locations must include at least the strip start point, strip center point, and strip end point, and may also include four boundary vertices if necessary. Let the local coordinates of the rth key location be . , The ordinate of the local coordinate system. Let x be the abscissa of the local coordinate system, and let y be the imaging time offset of this position relative to the start time of the strip. Let the start time of the strip be... Then the actual time when this position is imaged is Near the peak of the transit, the position of the satellite's nadir point along its orbit is approximately... Therefore, the relative distance along the orbit between the satellite and this critical location is... The corresponding pitch angle requirement is It needs to meet the following requirements. Define the maximum allowable offset distance along the track. ,but Therefore, the allowable start time interval of the strip at this critical position can be determined. Take the intersection of the time intervals of all key locations. Therefore, the earliest time a strip can start is The latest time a strip can start is When the following conditions are met: At that time, the candidate strip has a feasible sliding start time window. ;when If the condition is met, it indicates that there is no start time at which the entire strip can satisfy the pitch constraint, and the candidate strip should be deleted.

[0071] Intersect with the original visible window of the satellite. Let the original visible transit window of the satellite for regional targets be . , This is the lower limit of the original visible border crossing window. As the upper limit of the original visible overpass window, in order to ensure that the stripe is imaged within the visible window, the stripe start time should also satisfy the following: , This represents the duration of strip imaging. Therefore, the final sliding start time window is... When the interval is not empty, retain the candidate strips.

[0072] S214: For candidate imaging strips of regional targets corresponding to the same regional target task, deduplication is performed on candidate imaging strips of regional targets with similar coverage grid sets, strip positions, attitude angles and sliding time windows to obtain fewer than a preset number of candidate imaging strips of regional targets.

[0073] Specifically, candidate strip deduplication and pruning: During candidate strip generation, candidate strips with approximately the same coverage grid set, strip geometry, attitude angle, and time window are deduplicated, retaining only the higher-quality candidate strips. Candidate strip quality is comprehensively evaluated based on factors such as the number of coverage grids, coverage efficiency per unit time, roll angle, pitch angle, and imaging duration. When the number of candidate strips for a single transit opportunity or a single area target exceeds a preset threshold, real-time pruning is performed to control the candidate size and avoid candidate set explosion. The deduplication process can be summarized into three steps: limiting the comparison range → determining whether there is near-duplicate → retaining the highest overall quality among duplicate strips. Deduplication is not based on a single similar attribute, but rather on the simultaneous similarity of key attributes such as coverage, geometry, attitude, and time window. Candidate strips are first grouped, and deduplication is only performed between strips that meet the following conditions: corresponding to the same area target; corresponding to the same satellite; corresponding to the same transit opportunity; corresponding to the same imaging mode. Strips generated by different satellites or overpasses, even if covering similar areas, represent different schedulable opportunities and should not be deduplicated. Jaccard similarity is used to calculate the similarity of the coverage grid sets of two strips. When the similarity is greater than a preset similarity threshold, their coverage effects are considered approximately the same. The differences in strip center distance, azimuth angle, relative length, and relative swath width are analyzed. When these differences are all less than their corresponding thresholds, the geometric positions and shapes of the two strips are considered approximately the same. The differences in roll angle and pitch angle are analyzed (the representative pitch angle can be the pitch angle corresponding to the midpoint of the strip's time window). When the differences in roll angle and pitch angle are less than their corresponding thresholds, the attitude requirements of the two strips are considered approximately the same. The intersection-union ratio (IUU) of time windows is analyzed. When it is greater than the corresponding threshold, the sliding time windows of the two strips are considered approximately the same. A constraint on the center difference of time windows can also be added, requiring the center difference to be less than a corresponding threshold, to avoid incorrect deduplication of two strips with a high overlap ratio but significant differences in time position. When two candidate stripes simultaneously meet the above conditions, they are judged to be approximately duplicate stripes. Therefore, coverage similarity is used as the main criterion, and geometric, pose and time window similarity are used as joint confirmation conditions. Only one of the similarities cannot be used to directly remove duplicates.

[0074] For strips deemed approximately duplicates, instead of random deletion, a comprehensive quality score is calculated for each strip, and the strip with the highest score is retained. The effective grid number covered by the candidate strip is calculated; if the grid areas differ, the coverage area is used. Coverage efficiency per unit time is evaluated by dividing the effective grid number covered by the candidate strip by the imaging time, and then normalizing; the more grids or areas covered per unit time, the higher the evaluation. Roll angle is evaluated by dividing the roll angle by the maximum roll angle; a smaller roll angle is better. Elevation angle is evaluated by dividing the average absolute elevation angle within the time window by the maximum elevation angle; a smaller elevation angle is better. Imaging duration is evaluated; shorter imaging duration generally consumes less satellite time, energy, and storage resources. All evaluation values ​​are weighted and fused to obtain a comprehensive quality score. A higher comprehensive quality score indicates that the strip has a greater coverage contribution, higher coverage efficiency, smaller attitude deflection, and shorter imaging duration. Arrange the candidate strips in descending order of their overall quality scores, initialize the retainer set, and process each candidate strip sequentially. If no strip in the current retainer set is approximately identical to a candidate strip, the candidate strip is retained; otherwise, it is deleted. When approximately identical strips have the same overall quality score, compare the following in order: sliding time window width, total attitude deflection, and imaging duration. Prioritize retaining strips with wider time windows, then those with smaller attitude deflections, and finally those with shorter time windows.

[0075] After generating candidate imaging strips for large-scale agile constellation multi-type tasks, the computational burden caused by generating too many redundant or similar strips can be effectively avoided. Deduplicating candidate imaging strips for regional targets significantly reduces the amount of data that the scheduling algorithm needs to process, eliminating computational redundancy caused by numerous similar strips. This makes the subsequent candidate task set more concise, and the process of constructing static heuristic maps and analyzing task scarcity and conflict levels more efficient, thereby accelerating the convergence speed of heuristic sorting and iterative optimization, ultimately improving the overall efficiency and quality of scheduling schemes for large-scale agile constellation multi-type tasks.

[0076] S215: For candidate imaging strips of regional targets within the same target area, strips with similar coverage areas are grouped into the same strip family based on the similarity of the coverage grid sets. Different candidate imaging strips of regional targets within the same strip family can be interchanged at different time windows or different elevation angles. A greedy coverage strategy is employed at the strip family level to generate multiple candidate strip combinations for regional target tasks, thereby improving the scheduling alternative capability when task conflicts occur.

[0077] Specifically, strip families and regional strip combinations are constructed: For candidate strips of the same regional target, strips with highly similar coverage areas are grouped into the same strip family based on the similarity of their coverage grid sets. Different candidate strips within the same strip family can serve as different time windows or different elevation alternatives for the same spatial strip. Subsequently, a greedy coverage strategy is used at the strip family level to generate candidate strip combinations for regional targets. Each time, a strip family that can bring a large contribution to new coverage, high coverage efficiency, and low attitude cost is selected, and the selection is limited to a maximum of one strip per imaging opportunity. For a single regional target, multiple sets of spatially different strip combinations can be output to improve the alternative selection capability in the subsequent scheduling stage when task conflicts occur.

[0078] By dividing candidate imaging strips for regional targets into strip families and generating multiple candidate strip combinations, the scheduling system's flexibility and robustness in handling regional target tasks are significantly improved. When conflicts occur between tasks, the scheduler can utilize the substitutability within strip families and the switching capability between different strip combinations to more effectively find feasible scheduling schemes, thereby reducing the possibility of task abandonment and improving task completion rate and the utilization efficiency of agile satellite resources.

[0079] S216: Output a unified candidate strip set: Merge point target candidate strips and regional target candidate strips into a unified candidate strip set. Each candidate strip is represented by a unified field, including mission number, target type, associated target, satellite number, earliest start time, latest end time, imaging duration, roll angle, elevation angle, coverage grid set, strip family number, resource consumption, and alternative strip information. Simultaneously, output a strip-grid coverage mapping file and a regional target strip combination file as input for the strip scheduling optimization phase.

[0080] This application provides a more refined and flexible candidate imaging strip generation mechanism for regional target missions. By systematically enumerating candidate geometric strips and performing rigorous constraint screening, it ensures that the generated regional target candidate imaging strips are geometrically reasonable and executable within the performance range of agile satellites. Simultaneously, by accurately calculating and forming a sliding time window, the scheduling algorithm's degrees of freedom in the time dimension are significantly increased. This allows for more effective avoidance of task conflicts and improved task insertion success rates in complex scheduling scenarios involving multiple tasks and multiple satellites. This not only optimizes the utilization efficiency of agile satellite resources but also greatly enhances the overall scheduling capability and mission completion quality of large-scale agile constellations for regional target missions.

[0081] Phase 2: Unified Strip Scheduling Optimization S3: Group all candidate imaging strips corresponding to the same task into the same set to form a candidate task set.

[0082] Specifically, the process involves reading candidate strip and mission observation benefit data: reading the candidate strip time window file, strip-grid coverage mapping file, and target benefit file output from the mission preprocessing stage. Fields such as satellite number, mission number, earliest start time, latest end time, imaging duration, and roll angle in the candidate strips are converted into unified fields required by the scheduling algorithm, and the corresponding energy consumption and storage consumption are calculated based on the imaging duration. Construct a unified candidate task set: Extract basic task numbers based on the task numbers of candidate strips, and distinguish between point target candidate strips and regional target candidate strips. For regional target candidate strips, establish a mapping relationship between them and their respective regional targets and coverage grid sets; for point target candidate strips, establish a mapping relationship between them and their point target tasks. Subsequently, all candidate strips corresponding to the same basic task are grouped into the same candidate window set, forming the candidate task set for the scheduling optimization phase.

[0083] S4: Based on the candidate task set, construct a static heuristic graph and analyze the task scarcity of each task and the task conflict between tasks in the static heuristic graph.

[0084] Specifically, a scheduling state representation is established: a scheduling solution state is established to record the currently scheduled strip sequence, remaining energy, remaining storage, unassigned task set, number of regional grid coverage counts, number of grids covered by regional targets, completion status of point targets, and current objective function value for each satellite. For regional targets, the number of grid coverage counts determines whether a grid has been covered; for point targets, the completion count determines whether the point target has completed observation.

[0085] A static heuristic graph is constructed based on a set of candidate tasks. This graph is a task relationship graph built from the preprocessed set of candidate tasks. Each basic task corresponds to a task node, and each node is associated with a set of candidate imaging strips for that basic task. Task scarcity is determined by counting the number of candidate imaging strips for each basic task; fewer strips indicate fewer available observation opportunities and higher scarcity. For any two basic tasks, candidate imaging strips located on the same satellite are selected. The sliding time window, imaging duration, and attitude transition time of each strip are considered to determine whether the two strips can be executed consecutively in different execution sequences. The conflict intensity between the two basic tasks is determined by the proportion of non-consecutive candidate strip pairs among all comparable pairs. When the conflict intensity meets preset edge-building conditions, conflict edges are established between the corresponding task nodes, and the task conflict degree is determined based on the number or weight of the conflict edges associated with each task node. The task scarcity is used to guide the construction of the initial solution and the insertion of unassigned tasks, while the task conflict is used to guide the high-conflict disruptor to select tasks that impose strong constraints on the current scheduling structure.

[0086] Task scarcity is ; in, For the task The corresponding number of sliding time windows.

[0087] First, construct a task conflict graph. , among which, nodes Represents candidate task units If two candidate task units and If both time window and attitude transition constraints cannot be satisfied simultaneously within the same satellite sequence, conflict edges are added between the corresponding nodes. Based on this, candidate task units are defined. The task conflict level is: ; Among them, if and If a conflict exists, then the conflict value is... Otherwise, conflict value .

[0088] S5: Based on mission observation benefits, contribution of new coverage to regional targets, mission scarcity, and insertion compactness, candidate missions are heuristically ranked and inserted into the mission sequence of the corresponding agile satellites to obtain a feasible scheduling solution. This step specifically includes the following:

[0089] S510: Normalize and weight the mission observation gains, the contribution of new regional target coverage, mission scarcity, and insertion compactness to obtain the dynamic processing priority of the corresponding candidate missions.

[0090] Specifically, the task observation benefit of the point target task is ; in, Let S represent the set of point targets, and S represent the set of agile satellites; Representing decision variables , indicating task Whether to execute on Agile Satellite s: 1 if executed, 0 otherwise; This indicates the target return.

[0091] The task observation benefit of the regional target task is ; in, Represents the set of regional targets. Indicates the first Target revenue for each region Indicates the first Coverage of targets in each region.

[0092] Mission sequence compactness describes the overall relationship between mission execution time, attitude transition time, and idle time within a satellite's entire mission sequence. Insertion compactness describes the impact of inserting a candidate strip at a certain position on the mission sequence compactness. Insertion compactness is determined based on the difference in mission sequence compactness before and after insertion, or based on the remaining effective idle time between the candidate strip and its predecessor and successor missions. The less remaining idle time resulting from the insertion of the candidate strip, or the greater the increment in mission sequence compactness, the higher the insertion compactness of the proposed insertion scheme.

[0093] Assume the current mission sequence of the agile satellite is arranged according to its actual execution time as follows: Where n represents the total number of tasks in the task sequence, and the nth task is the nth task. Task The start time and imaging duration are respectively and The attitude transition time between adjacent tasks is The entire task sequence spans a time period of [time range missing]. , Let be the start time of the nth task. The duration of imaging for the nth task. The time taken for the first task is the start time. The actual time taken for the task includes imaging time and necessary attitude transition time. , Therefore, the idle time within the task sequence is The compactness of a task sequence can be defined as: , in, To prevent extremely small positive numbers with a denominator of zero, a compactness closer to 1 indicates less idle time between adjacent tasks, a more continuous satellite task sequence, and higher effective observation time utilization. A smaller compactness indicates more unusable idle time between tasks. For example, if two tasks are 100 seconds apart, and imaging and attitude transition actually require 80 seconds, then the remaining idle time is 20 seconds, indicating a relatively high compactness; if only 30 seconds are actually used, leaving 70 seconds of idle time, then the sequence is relatively loose.

[0094] Effective idle time before insertion: Insertion compactness is calculated for a specific insertion scheme. Let candidate strip b be inserted between the preceding task u and the succeeding task v. After the preceding task u completes imaging and transitions to the pose required by the succeeding task v, the original idle time is... , This is the start time of the subsequent task v. The start time of the precursor mission u must be deducted from the imaging duration of the precursor mission u. and attitude transition time from the predecessor mission u to the successor mission v The difference in start time between two tasks cannot be used simply.

[0095] Effective idle time after insertion: After inserting candidate stripe b, the idle time between the predecessor task u and candidate stripe b is... ; The start time of candidate band b, Let be the attitude transition time from the preceding task u to the candidate strip b, and let be the idle time between the candidate strip b and the successor task v. , The start time of candidate band b, For the imaging duration of candidate band b, Let be the pose transition time from candidate strip b to successor task v. The remaining idle time after insertion is: If the insertion scheme is feasible, then it should satisfy: .

[0096] Local insertion compactness: Local insertion compactness can be defined based on the remaining idle time after insertion. Where p represents the insertion position. This is the second smallest positive number to prevent the denominator from being 0. The larger this value is, the more fully the original idle time is utilized after the candidate strip is inserted, and the more compact the insertion scheme is. To prevent the value from exceeding the specified range in special cases, it can be truncated to [0,1]. For example, if the original gap is 100 seconds, and Scheme 1 leaves 10 seconds remaining after insertion, the compactness is relatively high; Scheme 2 leaves 40 seconds remaining after insertion, the compactness is relatively low.

[0097] S511: Sort the tasks according to their dynamic processing priority from high to low to obtain the task processing order.

[0098] Specifically, the static priority information of basic tasks is determined based on task observation benefits and task scarcity, and the new coverage contribution and feasible insertion compactness of each candidate strip corresponding to the current scheduling solution are calculated. The task observation benefits, task scarcity, new coverage contribution, and insertion compactness are normalized and weighted to obtain the dynamic processing priority of basic tasks, and basic tasks to be processed are selected from high to low according to the dynamic processing priority.

[0099] S512: Select candidate tasks sequentially from the task processing sorting and attempt to insert them into any insertable position in the corresponding agile satellite's task sequence to obtain multiple insertable schemes.

[0100] S513: Eliminate insertable solutions that do not meet the constraints of visibility time window, attitude transition time, energy capacity and storage capacity. Obtain an insertion score based on the task observation benefits, regional target new coverage contribution and insertion compactness of the task sequence after insertion of the remaining insertable solutions. Then, select the insertable solution with the highest insertion score as the feasible scheduling solution.

[0101] Insert rating : , in, Indicates task priority. Indicates the scarcity of tasks. To add a new number of coverage grids (point target is 0). To encourage sequence compactness after compact insertion, , and These represent the weighted values ​​of the corresponding items.

[0102] S514: After each insertion is completed, update the dynamic processing priority, and then select candidate tasks from the new task processing sorting for insertion, filtering and scoring, until there is no feasible scheduling solution.

[0103] Specifically, for the selected basic task, its candidate strips, corresponding satellites, and insertion locations are traversed, and insertion schemes that do not meet the constraints of visibility time window, attitude transition time, energy capacity, and storage capacity are eliminated. An insertion score is calculated based on the task observation benefits, the number of newly added coverage grids, and the compactness of the task sequence after insertion for the remaining feasible insertion schemes. The candidate strip and insertion location with the highest insertion score are selected for insertion. After each insertion, the task sequence, resource status, and regional grid coverage status are updated, and the dynamic processing priority of the remaining tasks is recalculated until no feasible insertion scheme remains.

[0104] Through the above technical solutions, this application can significantly improve the efficiency and quality of scheduling various types of tasks in large-scale agile constellations. By introducing dynamic priority processing, the scheduling system can more accurately assess the real-time importance and urgency of tasks, avoiding resource waste or omission of critical tasks that may be caused by static priorities. The refined insertion evaluation mechanism combines multi-dimensional indicators (task observation benefits, contribution of new regional target coverage, insertion compactness) and rigorously verifies key constraints such as visible time windows, attitude transition times, energy capacity, and storage capacity, ensuring the physical feasibility of each insertion operation and its positive contribution to the overall scheduling solution. In addition, the iterative process of dynamically updating priorities and reordering after each insertion allows the scheduling system to flexibly adapt to changes in the task environment, continuously optimize scheduling decisions, and gradually build a high-quality, high-efficiency scheduling scheme with better resource utilization. This effectively solves the technical problem of how to efficiently and accurately determine task priorities and ensure the effectiveness of task insertion in complex and ever-changing task scenarios, enabling agile constellations to more fully utilize their observation capabilities.

[0105] S6: Select a joint operator to iteratively optimize the current feasible scheduling solution, perform a time-series feasibility check on the optimized feasible scheduling solution, obtain the current optimal solution, and output it as the final scheduling scheme. A context-based multi-armed slot machine is used to select the joint operator, which includes a destruction operator and a repair operator. The destruction operator includes a random destruction operator, a low-reward destruction operator, a high-conflict destruction operator, and a redundant coverage destruction operator. The repair operator includes a greedy insertion repair operator and a random insertion repair operator.

[0106] Specifically, after each round of destruction-repair neighborhood search, based on the quality relationship between the candidate new solution and the current scheduled solution and the preset acceptance criteria, it is determined whether to use the candidate new solution to update the current scheduled solution; further, the updated current scheduled solution is compared with the global optimal solution. When the updated current scheduled solution is better than the global optimal solution, the global optimal solution is updated, and the number of iterations without a continuously updated global optimal solution is reset to zero; otherwise, the global optimal solution remains unchanged, and the number of iterations without a continuously updated global optimal solution is increased by one. This S6 step specifically includes the following contents.

[0107] S610: Determine the search context state based on the proportion of unassigned tasks in the current feasible scheduling solution and the number of consecutive iterations without improvement. The search context state includes the exploration state, the refining state, and the stagnant state.

[0108] Specifically, the number of unassigned basic tasks is recounted based on the updated current scheduling solution, and the proportion of unassigned tasks is calculated. First, it is determined whether the number of iterations without refreshing the global optimal solution has reached a preset stagnation threshold. If it has, the current search context state is determined to be in a stagnation state. If it has not, the proportion of unassigned tasks is further compared with a preset proportion threshold. When the proportion of unassigned tasks is higher than the proportion threshold, the current search context state is determined to be in an exploration state; otherwise, it is determined to be in a refining state.

[0109] Let the proportion of unassigned tasks in the current solution be . The continuous no-improvement iterative algebra is Simultaneously, define the state transition control hyperparameter: exploration scale threshold. Refining ratio threshold (and satisfy) ), and stagnation tolerance algebra .

[0110] (1) Explore state: when and This indicates that there are many unassigned tasks in the current solution, and the system is in a state of high unassignment rate. At this time, the context will guide the algorithm to focus on global search, giving large-scale destruction operators a higher selection probability.

[0111] (2) Refined state: when and When the solution space has been initially developed, the algorithm enters the local refinement stage, and the context focuses on guiding the local fine-tuning operator to perform deep optimization.

[0112] (3) Stagnation: when At any time, regardless Regardless of the value, it is determined that the search has fallen into a local optimum. At this point, a stagnation state is triggered, a strong perturbation mechanism is executed to escape the local optimum trap, and the value is reset. .

[0113] S611: Under the current search context, compute the value estimate for each joint operator.

[0114] For each context state Maintain joint operations for (Destruction Operator) × Repair Operator )of Value table. In each iteration, an action is selected using UpperConfidenceBound (UCB) based on the current context.

[0115] S612: Determine the uncertainty exploration term based on the cumulative number of executions of each joint operator.

[0116] Joint Action The uncertainty exploration term can be calculated as ; in, Let be the exploration coefficient under state c. This represents the total number of times the current state has occurred and the combined action has been executed. This indicates the number of times the current combined action has been executed. This exploration item has the following characteristics: The smaller the value, the larger the exploration item, indicating that the joint action has not been fully explored; The larger the value, the smaller the number of exploration items, indicating that the effects are already well understood; As the number of steps increases, joint actions that have not been selected for a long time can still receive some exploration compensation. The uncertainty exploration term is not a direct evaluation of the solution quality, but rather an evaluation of "whether the current understanding of this joint action is sufficient".

[0117] S613: Add the value estimate and the uncertainty exploration term to obtain the upper confidence bound selection value for each joint operator.

[0118] For each search context state, the value estimate and cumulative execution count of each destruction-repair joint operator are maintained. The value estimate characterizes the historical search benefit of the corresponding joint operator in the given search context state. In the current search context state, the value estimate of each joint operator is used as a historical utilization term, and an uncertainty exploration term is calculated based on the cumulative execution count. The value estimate is then added to the uncertainty exploration term to obtain the upper confidence bound selection value of the joint operator. ; in, Indicates joint action Value estimation under state c, i.e., historical performance; This indicates the number of times the joint action has been executed in state c. The joint operator with the largest upper confidence bound selection value is selected to perform a neighborhood search. After the joint operator is executed, the immediate reward is determined based on the change in the quality of the new solution, and then... , The empirical decay factor, The value estimate of the joint operator is updated for the immediate reward obtained from the current action, so that the updated value estimate can be used to calculate the joint operator selection value in subsequent searches under the same search context.

[0119] S614: Select the joint operator with the largest upper confidence bound selection value to perform the current round of neighborhood reconstruction, wherein, according to the destruction operator in the selected joint operator, several scheduled candidate imaging strips are removed from the current feasible scheduling solution, and the unassigned tasks are re-inserted according to the repair operator in the selected joint operator.

[0120] Specifically, joint operator selection , This represents the set of joint operators, i.e., the joint action of selecting the largest destruction-repair value to perform the neighborhood search in this round.

[0121] The number of stripes to be removed in this round is determined based on the number of scheduled stripes in the current scheduling solution and the preset destruction ratio, and the stripes to be removed are determined according to the selected destruction operator. For example... Figure 2 As shown, the task sequence originally contained four tasks. For the random disruption operator, a corresponding number of stripes are randomly selected from the currently scheduled stripe set; for example... Figure 3 As shown, the task sequence originally contains four tasks and their corresponding rewards, assuming rewards of 1, 2, 3, and 4 respectively. For low-reward destructive operators, the priority for removal is determined based on the marginal reward per unit time of each scheduled stripe, from low to high. Figure 3 Task 1, which yields a profit of 1, is removed from the middle; for example... Figure 4 As shown, the task sequence originally contains four tasks and their corresponding conflicts, assuming conflicts are 1, 2, 3, and 4. For high-conflict disruptive operators, the cumulative conflict level between the basic task corresponding to each scheduled strip and the currently unassigned task is calculated based on the static heuristic graph. Strips with higher cumulative conflict levels are prioritized. Figure 4 Task 4, which yields a profit of 4, is removed from the middle; for example... Figure 5 As shown, for the redundancy coverage destruction operator, stripes with lower exclusive coverage contributions or higher repetition rates are prioritized based on their exclusive coverage contribution and degree of overlap. To increase search diversity, a specified number of stripes can be randomly selected from the high-priority removal candidate set determined according to relevant indicators. The selected stripes are then removed from their corresponding satellite mission sequences, and the energy, storage, regional grid coverage, and unassigned mission status are updated synchronously.

[0122] Unassigned tasks are re-inserted based on the selected repair operator. For example... Figure 6 As shown, the greedy insertion repair operator traverses the candidate strips and insertable positions of unassigned tasks, prioritizing feasible insertion schemes that offer higher returns, greater contribution to new coverage, lower resource consumption, and a more compact task sequence. For example... Figure 7As shown, the random insertion repair operator increases search diversity by randomly shuffling the order of unassigned tasks and candidate stripes and selecting the first insertion scheme that satisfies the feasibility constraint.

[0123] For each feasible insertion scheme h, calculate the benefit evaluation, new coverage evaluation, resource consumption evaluation, and task sequence compactness evaluation separately. All indicators should be normalized to [0,1] before weighted fusion.

[0124] The marginal benefit of the insertion scheme is the difference in benefits before and after the insertion, which is then normalized. If different candidate strips for the same basic task have the same benefits, this indicator is mainly used for comparison between different basic tasks.

[0125] The evaluation of the contribution of new coverage is based on the set of grids in the newly covered area. For point targets, this item is not included in the scoring, and the weights of other indicators are renormalized to avoid point targets being inherently at a disadvantage due to the lack of grid coverage attributes.

[0126] Assume the energy consumed by inserting the strip is E. h Storage capacity M h The corresponding satellite's current remaining energy and remaining storage capacity are as follows: and Define normalized resource consumption: , in, As a weight for energy consumption, As a weight for storage capacity, , To prevent the denominator from being zero, the resource consumption evaluation is as follows: The less energy and storage consumed, The larger.

[0127] Task sequence compactness evaluation. Suppose a candidate strip b is inserted between the predecessor task u and the successor task v. The effective idle time before insertion is... The two valid idle times after insertion are: and The local compactness after insertion is This value is limited to the range [0,1]. The larger the value, the less idle time remains after insertion, and the better the candidate strips fill the existing task gaps.

[0128] If the insertion is at the beginning or end of the sequence, it can be based on the increase in the total time span of the task sequence. Evaluation: h represents a task added at the beginning or end of the task sequence. To increase the time span of the task sequence after task h, Indicates adding or inserting. For the original task sequence The time span, i.e. The smaller the span increase, the higher the compactness evaluation. The four evaluation indicators are weighted to obtain a comprehensive insertion score. Greedy insertion selection involves choosing the option with the highest comprehensive score from all feasible options formed by unassigned tasks, candidate strips, satellites, and insertion positions.

[0129] S615: During the process of removing and inserting candidate imaging strips, perform a time-series feasibility check on the mission sequence of the affected agile satellites. If the check passes, complete the domain reconstruction.

[0130] Specifically, a timing feasibility check is performed: During strip insertion, removal, or adjustment, the timing feasibility of the mission sequence for the affected satellites is checked. Based on the earliest start time, latest start time, imaging duration, and attitude transition time of adjacent missions in the mission sequence, forward and backward propagation are performed to calculate the actual executable start time for each strip. If the actual start time of any strip exceeds its latest executable start time, or if the attitude transition time requirement cannot be met between adjacent strips, the insertion or adjustment scheme is deemed infeasible; otherwise, the scheme is deemed to meet the timing feasibility requirement.

[0131] S616: Perform local optimization on the mission sequence of each agile satellite. For a newly generated scheduling solution, if the objective function value is better than the current solution, accept the newly generated scheduling solution; otherwise, accept the newly generated scheduling solution with a certain probability according to the simulated annealing criterion.

[0132] Specifically, local optimization and solution acceptance: After completing neighborhood reconstruction, local optimization is performed on the task sequence of each satellite, attempting to improve scheduling benefits or release subsequent insertion space through adjacent task swapping, local rearrangement, or insertion operations. For newly generated scheduling solutions, if their objective function value is better than the current solution, the solution is accepted; if their objective function value is worse than the current solution, it is accepted with a certain probability according to the simulated annealing criterion, thereby avoiding the search from getting trapped in local optima too early.

[0133] The objective function is: ; This formula indicates that the objective function is to maximize the comprehensive observation benefits of the star cluster. The preset constraints of the objective function are: Constraint 1: ; Constraint 2: ; Constraint 3: ; Constraint 4: ; Constraint 5: ; Constraint 6: ; Constraint 7: ; Constraint 8: ; Constraint 9: ; Constraint 10: ; Constraint 11: ; Where F represents the overall observational benefit of the constellation; T represents the set of missions. ; Indicates task Visible time window on Agile Satellites This indicates the lower limit of the visible time window. Indicates the upper limit of the visible time window. Indicates task Duration of the observation window Indicates task Observation start time; Indicates task Observation start time, Indicates two adjacent tasks and The transition time Indicates the satellite's attitude stabilization time. Indicates two adjacent tasks and Side swing angle, Indicates two adjacent tasks and pitch angle, This represents the angular velocity of the agile satellite. Indicates task Energy consumed per unit of time This indicates the upper limit of the energy of the agile satellite s. Indicates task Storage consumed per unit of time, This indicates the upper limit of the storage capacity of the agile satellite s. This represents the minimum coverage limit for regional targets. This represents the aggregate of discrete grids after the region target is segmented; Represents grid coverage variable Regional targets grid Whether it has been overwritten; if it has been overwritten, the value is 1; otherwise, the value is 0. Indicates coverage parameters If the task Covered by grid If the value is 1, then the value is 1; otherwise, the value is 0. Indicates that the agile satellite s performs the mission Resolution at that time Indicates that the agile satellite s performs the mission The minimum resolution limit at that time.

[0134] The constraints listed above are as follows: Constraint 1 represents the mission uniqueness constraint: the same mission can only be observed once by a single satellite; Constraint 2 represents the visibility time window constraint: the mission start and end times must fall within the corresponding satellite's visibility time window for that mission; Constraints 3 and 4 represent attitude transition time constraints: when the same satellite performs two consecutive missions, the attitude maneuver transition time requirement must be met between adjacent missions; Constraint 5 represents the energy constraint: the total energy consumed by the satellite in performing the mission must not exceed its energy capacity limit; Constraint 6 represents the storage constraint: the amount of data generated by the satellite in performing the mission must not exceed its storage capacity limit; Constraints 7 and 8 represent the area coverage constraint: the overall coverage rate constituted by all its missions must meet the minimum mission requirements; Constraint 9 represents the grid coverage determination constraint: only the area grids actually covered by the candidate strips that have been scheduled for execution can be included in the covered grid set of the area target; Constraint 10 represents the coverage unique counting constraint: even if the same area grid is repeatedly covered by multiple candidate strips, its coverage contribution is only counted once; Constraint 11 represents the resolution constraint: the imaging resolution of the satellite when performing the mission should meet the mission observation requirements.

[0135] By explicitly defining the objective function and pre-defined constraints, this paper provides a framework for the quantitative evaluation and rigorous verification of scheduling multiple mission types for large-scale agile constellations. After generating an initial feasible scheduling solution, the objective function and constraints are used in an iterative optimization process. During the iterative optimization process, after each adjustment to the scheduling solution (e.g., removing a mission by breaking an operator and then reinserting it by repairing an operator), the merits of the new solution are evaluated according to the objective function, and temporal feasibility is verified according to the pre-defined constraints. Only solutions that satisfy all constraints and whose objective function values ​​are improved (or accepted with a certain probability according to the simulated annealing criterion) are retained. This mechanism ensures that the final output scheduling scheme is not only operationally feasible but also achieves optimal or near-optimal results in terms of overall observational benefits. This application ensures that the generated scheduling scheme for multiple mission types in large-scale agile constellations is not only theoretically feasible but also strictly adheres to the performance limitations and mission requirements of agile satellites in actual operation. This significantly improves the reliability and efficiency of the scheduling scheme, avoids mission failures due to resource overruns or operational conflicts, and thus maximizes the overall observational benefits and mission completion rate of the constellation.

[0136] S617: Update the value estimate and cumulative execution count of the selected joint operator based on the current round of neighborhood reconstruction results, providing a basis for the calculation of the upper confidence bound selection value in the future.

[0137] Specifically, the joint operator score is updated: the reward for the selected joint action is updated based on the neighborhood reconstruction result of the current round. A higher reward is given when a new solution refreshes the global optimum; a slightly higher reward is given when a new solution is better than the current solution but does not refresh the global optimum; a lower reward is given when a poor solution is accepted; and a zero reward is given when a new solution is rejected. Subsequently, a weight decay method is used to update the evaluation value of the joint action in the current search context, and the number of times it is selected is recorded to provide a basis for subsequent upper confidence bound selection. This update method allows the evaluation value to simultaneously reflect long-term search experience and current iteration performance. When a joint action consistently produces high-quality solutions, its evaluation value gradually increases; when a joint action has a poor long-term effect, its evaluation value gradually decreases with decay.

[0138] S618: When the maximum number of iterations is reached or the number of consecutive no improvements reaches the preset number of improvements, the current solution is taken as the optimal solution, and terminal post-processing is performed before the final agile satellite constellation scheduling scheme is output.

[0139] Specifically, terminal insertion and redundancy cleanup: When the maximum number of iterations is reached or the number of consecutive times without improvement reaches a preset threshold, terminal post-processing is performed on the current optimal solution. First, stripes with low contribution or high redundancy in the regional coverage are cleaned up; then, unassigned tasks are sorted according to task scarcity, and attempts are made to insert tasks with feasible observation opportunities into the satellite sequence; finally, feasibility verification and redundancy cleanup are performed again to obtain the final scheduling scheme.

[0140] Output the final constellation scheduling scheme. The scheduling scheme includes the candidate strip number to be executed, the corresponding target number, the executing satellite, the actual start time, the actual end time, the imaging duration, the roll angle, the target type, the area coverage information, and the scheduling benefits. For area targets, the area coverage rate is also output; for point targets, the mission completion status is output. This forms an executable scheduling scheme that satisfies the constraints of the visible time window, attitude transition, energy, storage, and area coverage.

[0141] In this embodiment, the task completion level and search stagnation level of the current scheduling solution are jointly constructed into a dynamic search context, which is updated after each iteration. The updated search context state is used to index the value estimate and cumulative execution count of each damage-repair joint action in the corresponding state, so that the selection of subsequent joint actions can be adaptively adjusted according to the search stage.

[0142] Through the above technical solution, this application effectively addresses the problem that relying solely on heuristic methods is insufficient to obtain the globally optimal solution in the scheduling of multiple types of tasks in large-scale agile constellations. By introducing an iterative optimization mechanism based on joint operators, the scheduling process can adaptively adjust destruction and repair strategies according to the current search state, thereby achieving a dynamic balance between exploring a broad solution space and utilizing known effective strategies. This significantly enhances the global search capability of the scheduling algorithm, enabling it to effectively escape local optima and find higher-quality scheduling solutions. Simultaneously, combining temporal feasibility verification and local optimization ensures the practical executability and refinement of the scheduling scheme. Ultimately, this application provides a more optimized, robust, and efficient task scheduling scheme for large-scale agile constellations, greatly improving the task completion efficiency and resource utilization of the constellations.

[0143] like Figure 8 As shown in the embodiment of this application, a scheduling system for large-scale agile constellations with multiple task types is provided, comprising: The point target task processing module 10 is used to calculate the sliding time window of the point target on the corresponding agile satellite based on the position of the point target and the operation information of the agile satellite, and generate candidate imaging strips of the point target; The regional target task processing module 20 is used to analyze the grid set of the covered target area under the visible transit opportunity of the agile satellite for the regional target task, and obtain the regional target candidate imaging strip and the corresponding sliding time window. The target area in the regional target task is divided into multiple grids after discretization processing. The atlas analysis module 30 is used to group all candidate imaging strips corresponding to the same task into the same set to form a candidate task set; it is also used to construct a static heuristic atlas based on the candidate task set and analyze the task scarcity of each task and the task conflict between tasks in the static heuristic atlas. The scheduling solution generation module 40 is used to heuristically sort candidate tasks based on task observation benefits, regional target new coverage contribution, task scarcity and insertion compactness, and insert candidate tasks into the task sequence of the corresponding agile satellite to obtain a feasible scheduling solution. The scheduling solution optimization module 50 is used to select joint operators, iteratively optimize the current feasible scheduling solution, perform time-series feasibility verification on the optimized feasible scheduling solution, obtain the current optimal solution and output it as the final scheduling scheme.

[0144] In this embodiment, the beneficial effects of the scheduling system for large-scale agile constellations with multiple types of tasks are similar to those of the scheduling method for large-scale agile constellations with multiple types of tasks described above, and will not be repeated here.

[0145] In summary, compared with existing technologies, it has the following beneficial effects: 1. Compared with existing methods such as independent scheduling of point targets, separate optimization of region coverage, or ordinary adaptive large neighborhood search, this invention has the following advantages: 2. Improve the unified scheduling capability for multiple types of tasks: By standardizing the expression of candidate tasks, point objectives and regional objectives can be jointly optimized under the same scheduling framework, reducing the fragmentation caused by separate modeling and separate solving.

[0146] 3. Improve the accuracy of regional coverage benefit assessment: Through grid-strip coverage mapping and unique coverage counting, it is possible to distinguish between new coverage, duplicate coverage and exclusive coverage, reduce redundant observations and improve the quality of regional target coverage.

[0147] 4. Improve search efficiency in large-scale scenarios: Static heuristic graphs improve the quality of initial solutions, lightweight feasibility checks reduce full constraint checks, and context-aware operator selection reduces invalid neighborhood reconstruction, thereby improving response speed in large-scale task scenarios.

[0148] 5. Enhanced algorithm stability and ability to escape local optima: The contextual multi-armed slot machine mechanism can dynamically adjust the selection of joint operators according to the search state, and adopt different search strategies in the exploration, refinement and stagnation phases to improve the stability of solution quality.

[0149] 6. Improve project executability: The scheduling results simultaneously meet constraints such as time window, attitude transformation, energy, storage, coverage and resolution. The output task sequence, execution time and resource usage information can be directly used for subsequent task planning and execution.

[0150] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0151] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A scheduling method for large-scale agile constellations with multiple task types, characterized in that, include: For point target tasks, based on the location of the point target and the operational information of the agile satellite, the sliding time window of the point target on the corresponding agile satellite is calculated, and candidate imaging strips of the point target are generated. For regional target missions, under the visible transit opportunity of agile satellites, the grid set of the covered target area is analyzed to obtain regional target candidate imaging strips and corresponding sliding time windows; In the regional target task, the target region is discretized and divided into multiple grids; All candidate imaging strips corresponding to the same task are grouped into the same set to form a candidate task set; Based on the candidate task set, a static heuristic graph is constructed, and the task scarcity of each task and the task conflict between tasks in the static heuristic graph are analyzed. Based on the mission observation benefits, the contribution of new regional target coverage, mission scarcity, and insertion compactness, candidate missions are heuristically ranked and inserted into the mission sequence of the corresponding agile satellites to obtain a feasible scheduling solution. Mission observation benefits refer to the benefits that can be obtained by completing the mission; regional target coverage contribution refers to the increase in the overall coverage of regional targets after executing candidate imaging strips; insertion compactness refers to the degree of impact of inserting a candidate mission into the agile satellite mission sequence on the compactness of the mission sequence; Select a joint operator to iteratively optimize the current feasible scheduling solution, perform time-series feasibility verification on the optimized feasible scheduling solution, obtain the current optimal solution and output it as the final scheduling scheme.

2. The scheduling method for large-scale agile constellations with multiple task types as described in claim 1, characterized in that, Before execution, this method also includes: Construct a region polygon based on the boundary points of the target region, and discretize the target region according to the preset grid scale to generate the corresponding grid, and assign a globally unique grid number to each grid. For each agile satellite and each target area: The orbital propagation of the agile satellite is performed based on the six elements of the agile satellite's orbit to obtain the spatial position of the agile satellite during the mission cycle. By combining the geographic coordinates of the target area, the relative positional relationship between the agile satellite and the target area is calculated, and the effective observation time range during the agile satellite's transit is determined based on the performance constraints of the agile satellite. Within the effective observation time range, the moment when the agile satellite is closest to the target area or when the observation elevation angle is largest is selected as the transit peak moment, and the satellite-target area geometric relationship at the transit peak moment is used to establish a local ground coordinate system for satellite transit. Projecting the grid cells or point targets of the target area onto the local ground coordinate system facilitates the construction of strips and the calculation of the attitude angles of agile satellites.

3. The scheduling method for large-scale agile constellations with multiple task types as described in claim 1, characterized in that, The step of calculating the sliding time window of the point target on the corresponding agile satellite based on the position of the point target and the operational information of the agile satellite, and generating candidate imaging strips for the point target, includes: For point target tasks, they are treated as short-term degraded imaging strips; Based on the point target's position relative to the satellite's ground trajectory, the satellite's ground velocity, maximum roll angle, and maximum pitch angle, calculate the sliding time window of the point target on the corresponding satellite; If the point target satisfies the constraints of roll angle, pitch angle and simulation time range, then a candidate imaging strip for the point target is generated. The candidate imaging strip for the point target includes the mission number, satellite number, earliest start time, latest end time, fixed imaging duration, roll angle, initial pitch angle, end pitch angle, nominal start time and nominal end time.

4. The scheduling method for large-scale agile constellations with multiple task types as described in claim 1, characterized in that, For regional target missions, under the visible transit opportunity of agile satellites, the analysis of the grid set of the covered target area to obtain regional target candidate imaging strips and corresponding sliding time windows includes: For regional target missions, under the corresponding agile satellite transit opportunity, the lateral position and start and end positions along the track of candidate geometric strips are enumerated based on the local ground coordinate system. For each candidate geometric strip, calculate the set of grids covering its area, and inversely deduce the required roll angle, initial pitch angle, final pitch angle, and imaging duration based on the geometric relationship of the candidate geometric strip; If a candidate geometric strip satisfies the constraints of maximum roll angle, maximum pitch angle, minimum imaging duration, maximum imaging duration, and minimum number of coverage grids, then it is considered as a candidate imaging strip for the regional target. Based on the maximum pitch angle constraint, the continuous feasible range of the strip pitch zero point is deduced, and the continuous feasible range is converted into the earliest and latest start times of the regional target candidate imaging strip, forming a sliding time window.

5. The scheduling method for large-scale agile constellations with multiple task types as described in claim 1, characterized in that, Also includes: After all point target candidate imaging strips for agile satellites are generated, point target candidate imaging strips corresponding to the same point target mission are filtered to obtain fewer than the preset number of point target candidate imaging strips. For candidate imaging strips of regional targets corresponding to the same regional target task, duplicate candidate imaging strips of regional targets with similar coverage grid sets, strip positions, attitude angles and sliding time windows are deduplicated to obtain fewer than a preset number of candidate imaging strips of regional targets.

6. The scheduling method for large-scale agile constellations with multiple task types as described in claim 1, characterized in that, Before grouping all candidate imaging strips corresponding to the same task into the same set to form a candidate task set, the process also includes: For candidate imaging strips of regional targets in the same target area, candidate imaging strips of regional targets with similar coverage areas are grouped into the same strip family based on the similarity of the coverage grid sets; Different target candidate imaging strips within the same strip family can be replaced at different time windows or different elevation angles; At the stripe family level, a greedy coverage strategy is used to generate multiple candidate stripe combinations for regional target tasks, in order to improve the scheduling alternative capability when task conflicts occur.

7. The scheduling method for large-scale agile constellations with multiple task types as described in claim 1, characterized in that, The process involves heuristically ranking candidate tasks based on mission observation benefits, contribution from new regional target coverage, mission scarcity, and insertion compactness, and then inserting these candidate tasks into the corresponding agile satellite's mission sequence to obtain a feasible scheduling solution, including: The observation benefits of the task, the contribution of new coverage of regional targets, the scarcity of the task and the compactness of insertion are normalized and weighted to obtain the dynamic processing priority of the corresponding candidate tasks. The tasks are sorted from highest to lowest according to their dynamic processing priority to obtain the task processing order; Candidate tasks are selected sequentially from the task processing sorting and attempted to be inserted into any insertable position in the corresponding agile satellite's task sequence, resulting in multiple insertable schemes; Insertable solutions that do not meet the constraints of visible time window, attitude transition time, energy capacity and storage capacity are eliminated. Based on the task observation benefits, regional target new coverage contribution and insertion compactness of the task sequence after insertion of the remaining insertable solutions, an insertion score is obtained, and the insertable solution with the highest insertion score is taken as the feasible scheduling solution. After each insertion, the dynamic processing priority is updated, and then candidate tasks in the new task processing sort are selected for insertion, filtering and scoring, until no feasible scheduling solution exists.

8. The scheduling method for large-scale agile constellations with multiple task types as described in claim 1, characterized in that, The joint operator includes a destruction operator and a repair operator. The destruction operator includes a random destruction operator, a low-yield destruction operator, a high-conflict destruction operator, and a redundant coverage destruction operator. The repair operator includes a greedy insertion repair operator and a random insertion repair operator. The selection of the joint operator iteratively optimizes the current feasible scheduling solution, performs time-series feasibility verification on the optimized feasible scheduling solution, obtains the current optimal solution, and outputs it as the final scheduling scheme, including: Based on the proportion of unassigned tasks in the current feasible scheduling solution and the number of consecutive iterations without improvement, the search context state is determined, which includes the exploratory state, the refining state, and the stagnant state. Given the current search context, compute the value estimate for each joint operator; The uncertainty exploration term is determined based on the cumulative number of executions of each joint operator; The value estimate and the uncertainty exploration term are added together to obtain the upper confidence bound selection value for each joint operator; The joint operator with the largest upper confidence bound selection value is selected to perform the current round of neighborhood reconstruction. In this process, several scheduled candidate imaging strips are removed from the current feasible scheduling solution according to the destruction operator in the selected joint operator, and unassigned tasks are re-inserted according to the repair operator in the selected joint operator. During the process of removing and inserting candidate imaging strips, the timing feasibility of the mission sequence of the affected agile satellite is verified. If the verification is successful, the domain reconstruction is completed. For each agile satellite, perform local optimization on the mission sequence. For a newly generated scheduling solution, if the objective function value is better than the current solution, accept the newly generated scheduling solution; otherwise, accept the newly generated scheduling solution with a certain probability according to the simulated annealing criterion. The value estimate and cumulative execution count of the selected joint operator are updated based on the current round of neighborhood reconstruction results, providing a basis for the calculation of the upper confidence bound selection value in the future. When the maximum number of iterations is reached or the number of consecutive no improvements reaches the preset number of improvements, the current solution is taken as the optimal solution, and terminal post-processing is performed before the final agile satellite constellation scheduling scheme is output.

9. The scheduling method for large-scale agile constellations with multiple task types as described in claim 8, characterized in that, The task observation benefit of the point target task is ； in, Let S represent the set of point targets, and S represent the set of agile satellites; Representing decision variables , indicating task Whether to execute on Agile Satellite s: 1 if executed, 0 otherwise; Indicates the target return at a given point; The task observation benefit of the regional target task is ； in, Represents the set of regional targets. Indicates the first Target revenue for each region Indicates the first Coverage of targets in each region; The objective function is: ； The preset constraints of the objective function are: Constraint 1: ; Constraint 2: ; Constraint 3: ; Constraint 4: ; Constraint 5: ; Constraint 6: ; Constraint 7: ; Constraint 8: ; Constraint 9: ; Constraint 10: ; Constraint 11: ; Where F represents the overall observational benefit of the constellation; T represents the set of missions. ; Indicates task Visible time window on Agile Satellites This indicates the lower limit of the visible time window. Indicates the upper limit of the visible time window. Indicates task Duration of the observation window Indicates task Observation start time; Indicates task Observation start time, Indicates two adjacent tasks and The transition time Indicates the satellite's attitude stabilization time. Indicates two adjacent tasks and Side swing angle, Indicates two adjacent tasks and pitch angle, This represents the angular velocity of the agile satellite. Indicates task Energy consumed per unit of time This indicates the upper limit of the energy of the agile satellite s. Indicates task Storage consumed per unit of time, This indicates the upper limit of the storage capacity of the agile satellite s. This represents the minimum coverage limit for regional targets. This represents the aggregate of discrete grids after the region target is segmented; Represents grid coverage variable Regional targets grid Whether it has been overwritten; Indicates coverage parameters If the task Covered by grid If the value is 1, then the value is 1; otherwise, the value is 0. Indicates that the agile satellite s performs the mission Resolution at that time Indicates that the agile satellite s performs the mission The minimum resolution limit at that time.

10. A scheduling system for large-scale agile constellations with multiple task types, characterized in that, include: The point target task processing module is used to calculate the sliding time window of the point target on the corresponding agile satellite based on the position of the point target and the operation information of the agile satellite, and generate candidate imaging strips for the point target. The regional target task processing module is used to analyze the grid set of the covered target area for regional target tasks under the visible transit opportunity of agile satellites, and obtain regional target candidate imaging strips and corresponding sliding time windows. The target area in the regional target task is discretized and divided into multiple grids. The map analysis module is used to group all candidate imaging strips corresponding to the same task into the same set to form a candidate task set; it is also used to construct a static heuristic map based on the candidate task set and analyze the task scarcity of each task and the task conflict between tasks in the static heuristic map. The scheduling solution generation module is used to heuristically sort candidate tasks based on task observation benefits, regional target coverage contribution, task scarcity, and insertion compactness, and insert the candidate tasks into the corresponding agile satellite task sequence to obtain a feasible scheduling solution. Here, task observation benefits refer to the benefits that can be obtained by completing the task; regional target coverage contribution refers to the increase in the overall coverage of regional targets after executing the candidate imaging strip; and insertion compactness refers to the degree of impact of inserting a candidate task into the agile satellite task sequence on the compactness of the task sequence. The scheduling solution optimization module is used to select joint operators, iteratively optimize the current feasible scheduling solution, perform time-series feasibility verification on the optimized feasible scheduling solution, obtain the current optimal solution and output it as the final scheduling scheme.

Images