Satellite task planning method and system based on multiple constraint conditions

Abstract

The invention discloses a satellite task planning method and system based on multi-constraint conditions, and relates to the technical field of reinforcement learning, and the method comprises the steps: receiving task parameters of a plurality of observation tasks, carrying out the multi-constraint analysis of each observation task based on the current satellite orbit data and the task parameters, and obtaining a multi-constraint analysis result; obtaining at least one candidate observation scheme corresponding to each task, wherein the multi-constraint analysis at least comprises orbit visibility analysis and meteorological constraint analysis; based on candidate observation schemes of all observation tasks, a collaborative planning scheme integrating multiple constraint conditions is generated through a collaborative decision model, the collaborative decision model takes maximization of overall task return as a target function, and the target function fuses quality scores of the tasks and inter-task resource conflict penalty terms; and based on the collaborative planning scheme, generating a task instruction sequence for driving the satellite to execute. According to the method, the accuracy and the adaptive capability of satellite task planning in a complex scene are remarkably improved.

Description

Technical Field

[0001] This application relates to the field of reinforcement learning technology, and in particular to a satellite mission planning method and system based on multiple constraints. Background Technology

[0002] With the rapid development of aerospace technology, the number of satellites in orbit continues to grow, and their Earth observation missions are characterized by large scale, high frequency, and high timeliness requirements. As a key link connecting user needs and satellite capabilities, satellite mission planning focuses on efficiently and rationally allocating satellite resources to execute observation missions while meeting various physical constraints.

[0003] In recent years, some studies have begun to explore the application of artificial intelligence technologies such as deep learning to mission planning, such as using deep reinforcement learning for resource allocation. However, traditional satellite mission planning cannot adapt to high-time-efficiency scenarios such as multi-mission collaborative planning, emergency response, and ultra-large-scale mission scheduling. The accuracy of mission planning is insufficient, which seriously affects the practicality and decision-making efficiency of the system.

[0004] Therefore, improving the adaptability and accuracy of satellite mission planning in complex scenarios is a problem that urgently needs to be solved. Summary of the Invention

[0005] The main purpose of this application is to provide a satellite mission planning method and system based on multiple constraints, aiming to solve the technical problem of how to improve the adaptability and accuracy of satellite mission planning in complex scenarios.

[0006] To achieve the above objectives, this application proposes a satellite mission planning method based on multiple constraints, the method comprising: Receive task parameters for multiple observation tasks, wherein the task parameters include at least the target region, expected execution time window, payload type and priority of each task; For each observation task, a multi-constraint analysis is performed based on the current satellite orbit data and the task parameters to obtain at least one candidate observation scheme for each task. The multi-constraint analysis includes at least orbit visibility analysis and meteorological constraint analysis. Based on the candidate observation schemes for all observation tasks, a collaborative planning scheme integrating multiple constraints is generated through a collaborative decision-making model. The collaborative decision-making model takes maximizing the overall task return as its objective function, and the objective function integrates the quality score of each task and the resource conflict penalty term between tasks. Based on the aforementioned collaborative planning scheme, a sequence of mission instructions is generated to drive the satellite's execution.

[0007] In one embodiment, the step of performing multi-constraint analysis on each of the observation tasks based on the current satellite orbit data and the task parameters to obtain at least one candidate observation scheme for each task includes: For each observation task, orbital visibility analysis is performed based on the current satellite orbit data and the target area corresponding to the observation task. The transit time window of the satellite over the target area is calculated, and the observation mode is determined. Based on the multi-source meteorological data acquisition mechanism, real-time meteorological data that is spatiotemporally matched with the transit time window is acquired; Based on the cloud sensitivity rules corresponding to the load type, the cloud coverage in the real-time meteorological data is processed to obtain a meteorological constraint score; Based on the observation mode, the transit time window, and the meteorological constraint score, candidate observation schemes are generated.

[0008] In one embodiment, the steps of performing orbital visibility analysis based on the current satellite orbit data and the target area corresponding to the observation task for each observation task, calculating the satellite's transit time window over the target area, and determining the observation mode include: Input the two lines of orbit data into a simplified perturbation model to calculate the satellite's transit time window over the target area; Within the transit time window, the lateral distance between the target point and the satellite's nadir point trajectory is calculated, wherein the target point is the center point of the target area; Based on the relationship between the lateral distance, the half-width of the satellite payload, and the coverage area of ​​the side-swing, the observation modes are obtained, including normal pushbroom mode, side-swing observation mode, and unobservable mode.

[0009] In one embodiment, the step of obtaining the observation mode based on the relationship between the lateral distance, the half-swath width of the satellite payload, and the lateral glide coverage area, wherein the observation mode includes a normal pushbroom mode, a lateral glide observation mode, and an unobservable mode, further includes: If the observation mode is the side-swing observation mode, the elevation angle at each moment during the passage is calculated based on the simplified perturbation model to form elevation angle time series data; In the elevation angle time series data, all times when the elevation angle is greater than or equal to the minimum observed elevation angle threshold are selected; Merge consecutive time periods from the selected time periods to obtain at least one usable time period; For each available time period, a local coordinate system with the satellite as the origin is established, and the observation vector from the satellite to the target point is calculated, wherein the X-axis of the local coordinate system is along the velocity direction, the Y-axis is perpendicular to the orbital plane, and the Z-axis points to the Earth's center; The lateral swing angle is calculated by the angle between the observed vector and the Z-axis, and the lateral swing direction is determined by the projection of the observed vector onto the Y-axis.

[0010] In one embodiment, the step of acquiring real-time meteorological data that spatiotemporally matches the transit time window based on a multi-source meteorological data acquisition mechanism includes: According to the multi-source meteorological data acquisition mechanism, gridded cloud coverage data is obtained from at least one external meteorological data source. The external meteorological data source includes a primary data source and a backup data source. The multi-source meteorological data acquisition mechanism includes switching to the backup data source if the primary data source fails to acquire the data, and switching to the most recently cached data if both the primary data source and the backup data source fail to acquire the data. Based on the specific time within the transit time window and the specific location of the target area, the gridded cloud coverage data is matched using a spatiotemporal interpolation algorithm to obtain cloud coverage that spatiotemporally matches the transit time window. The spatiotemporal interpolation algorithm includes linear interpolation in the time dimension and bilinear interpolation in the spatial dimension.

[0011] In one embodiment, after the step of generating candidate observation schemes based on the observation mode, the transit time window, and the meteorological constraint score, the method further includes: When a single observation task corresponds to multiple candidate observation schemes, the quality score of each candidate observation scheme is calculated by a multi-factor weighted scoring model based on the observation mode, elevation angle, side swing angle and meteorological constraint score corresponding to each candidate observation scheme. Based on the quality score, the multiple candidate observation schemes are ranked to obtain the ranking result; Based on the sorting results, a predetermined number of schemes are selected from the multiple candidate observation schemes to obtain a set of candidate schemes corresponding to a single observation task.

[0012] In one embodiment, the collaborative planning scheme includes multiple target observation schemes. The candidate observation schemes based on all observation tasks are used to generate a collaborative planning scheme that integrates multiple constraints through a collaborative decision model. The collaborative decision model takes maximizing the overall task reward as its objective function. The step of integrating the quality score of each task and the resource conflict penalty term between tasks into the objective function includes: A state space is constructed based on the current satellite status information, the candidate scheme set for each mission, and the mission parameters of each mission. The state space is also used to perform time-series prediction through a long short-term memory network to obtain future environmental information, and to obtain mission relationship information by modeling the constraint relationships between missions through a graph neural network. The state space is input into a deep reinforcement learning decision model, which outputs an initial scheduling scheme and optimizes the strategy to maximize the overall task reward. The initial scheduling scheme includes multiple intermediate observation schemes. The deep reinforcement learning decision model is an Actor-Critic architecture, in which the Actor network outputs actions and the Critic network evaluates the state value. The actions are used to assign the intermediate observation schemes to each observation task. Based on the initial scheduling scheme, multi-objective optimization is performed to generate a Pareto optimal solution set, which includes multiple objective observation schemes.

[0013] Furthermore, to achieve the above objectives, this application also proposes a satellite mission planning system based on multiple constraints, which includes: The task configuration module is used to receive task parameters for multiple observation tasks. The task parameters include at least the target region, expected execution time window, payload type, and priority of each task. The multi-constraint analysis module is used to perform multi-constraint analysis on each of the observation tasks based on the current satellite orbit data and the task parameters, and to obtain at least one candidate observation scheme for each task. The multi-constraint analysis includes at least orbit visibility analysis and meteorological constraint analysis. The scheduling decision module is used to generate a collaborative planning scheme that integrates multiple constraints based on the candidate observation schemes of all observation tasks through a collaborative decision model. The collaborative decision model takes maximizing the overall task return as the objective function, and the objective function integrates the quality score of each task and the resource conflict penalty term between tasks. The results generation module is used to generate a sequence of mission instructions to drive the satellite.

[0014] In one embodiment, the multi-constraint analysis module further includes a visibility analysis unit and a meteorological data integration unit; The visibility analysis unit is used to perform orbit visibility analysis on each observation task based on the current satellite orbit data and the target area corresponding to the observation task. The meteorological data integration unit is used to acquire real-time meteorological data that matches the transit time window in time and space, based on a multi-source meteorological data acquisition mechanism.

[0015] In one embodiment, the satellite mission planning system based on multiple constraints further includes a visualization module for performing three-dimensional visualization demonstration and verification of the collaborative planning scheme. The visualization module includes: The payload simulation unit is used to dynamically calculate and generate a push-broom strip model that fits the Earth's surface based on satellite payload parameters, real-time orbit data, and observation geometric parameters in the collaborative planning scheme. The synchronization control unit is used to keep the simulation display of the push-broom strip model synchronized with the planned observation period, and to keep the attitude of the observation cone synchronized with the planned lateral tilt angle. The interactive demonstration unit is used to display the push-broom strip model and observation cone in a 3D scene in real time, and provides an interactive interface for users to compare and verify multiple schemes.

[0016] One or more technical solutions proposed in this application have at least the following technical effects: The system receives task parameters from multiple observation tasks. These parameters include at least the target region, expected execution time window, payload type, and priority for each task. A unified and integrated task parameter receiving mechanism is adopted, resolving the issues of inconsistent input formats from multiple task sources and cumbersome manual verification. This achieves standardization in task management and improves entry efficiency.

[0017] For each observation mission, multi-constraint analysis is performed based on current satellite orbit data and mission parameters to obtain at least one candidate observation scheme for each mission. The multi-constraint analysis includes at least orbit visibility analysis and meteorological constraint analysis. By adopting a parallel fusion analysis technique for orbit and meteorological data, the problems of ignoring meteorological dynamics or computational redundancy in traditional serial analysis are solved, and the temporal and spatial feasibility and environmental suitability of candidate schemes are simultaneously guaranteed.

[0018] Based on candidate observation schemes for all observation tasks, a collaborative planning scheme integrating multiple constraints is generated through a collaborative decision-making model. The collaborative decision-making model uses maximizing the overall task return as its objective function, which integrates the quality scores of each task with resource conflict penalties between tasks. By employing a collaborative decision-making model aimed at maximizing overall return, this model quantifies and balances the quality gains of each task with the resource conflict costs arising from competition between tasks for the same satellite and the same time slot within a single objective function. This solves the problems of low overall system efficiency and low resource utilization caused by independent optimization of single tasks, achieving a leap from local optima to global optima.

[0019] Based on a collaborative planning scheme, a sequence of mission instructions is generated to drive satellite execution. By employing an automatic generation technology from planning scheme to control instructions, the problems of error-prone and time-consuming manual planning are solved, achieving closed-loop automation from planning to execution. This application, through multi-constraint parallel screening and multi-task collaborative optimization, significantly improves global optimization capabilities and adaptability while ensuring the physical feasibility of the scheme, making it suitable for high-time-efficiency scenarios such as multi-task collaborative planning, emergency response, and ultra-large-scale mission scheduling. Attached Figure Description

[0020] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0021] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of the software architecture of the satellite mission planning system based on multiple constraints in this application; Figure 2 This is a flowchart illustrating the first embodiment of the satellite mission planning method based on multiple constraints of this application; Figure 3 This is a schematic diagram of the first module structure of the satellite mission planning system based on multiple constraints in this application; Figure 4 This is a schematic diagram of the second module structure of the satellite mission planning system based on multiple constraints in this application; Figure 5 This is a schematic diagram illustrating the complete technical flow of an embodiment of this application.

[0023] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0024] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0025] This application provides a satellite mission planning method based on multiple constraints, which can be applied to a satellite mission planning system based on multiple constraints. For ease of explanation, in the following embodiments and implementations, the "satellite mission planning system based on multiple constraints" will be referred to as the "system," and the system software architecture can be referred to... Figure 1The system comprises a data layer, a support layer, a business logic layer, and a presentation layer. The presentation layer, serving as the direct interface between the system and the user, is responsible for presenting information and receiving operational commands. This layer primarily includes task configuration pages, planning result display pages, and a 3D visualization interactive interface. Users submit observation task parameters, set constraints, and view planning schemes and 3D simulation demonstrations through this layer. All user requests are transmitted to lower layers via standardized protocols. The business logic layer is the core processing hub of the system, carrying all key business rules and process controls. This layer integrates task configuration and management modules, visibility analysis modules, intelligent planning and scheduling decision-making modules, and result generation modules. It receives task requests from the presentation layer, coordinates and calls various services and engines in the support layer, and executes the entire process of calculation and logical judgment, from task parsing, multi-constraint analysis, collaborative optimization decision-making to final scheme generation. The support layer provides basic services, general components, and algorithm engine support for the operation of the business logic layer. This layer includes a front-end service framework, a back-end service container, a distributed computing engine, a deep reinforcement learning inference engine, a graph neural network modeling engine, a visualization rendering engine, and data access interfaces. This layer ensures the efficient and stable execution of business logic and encapsulates complex algorithm implementations and system interaction details. The data layer is the cornerstone of the system's data storage and management, responsible for persistently storing all static and dynamic data required for system operation. This layer mainly includes a mission planning database, a satellite orbit and payload parameter library, a historical and real-time meteorological database, and a system log library. It provides unified, secure, and efficient data access services for all the above layers, ensuring data consistency, integrity, and traceability. User commands and data are input from the presentation layer, normalized, and then transmitted to the business logic layer. The business logic layer, based on processing requirements, calls the corresponding services and engines of the support layer and accesses the data layer to obtain or store necessary information. Processing results and status information are returned along the reverse path and finally presented to the user in the presentation layer.

[0026] Figure 2 This is a flowchart illustrating the first embodiment of the satellite mission planning method based on multiple constraints according to this application. In this embodiment, the satellite mission planning method based on multiple constraints includes steps S10 to S40:

[0027] Step S10: Receive task parameters for multiple observation tasks. The task parameters include at least the target area, expected execution time window, payload type, and priority of each task.

[0028] It should be noted that an observation task refers to a specific Earth observation request and is the basic unit of planning. Task parameters define a set of information representing all attributes of an observation task, specifically including: the target area of ​​the task, the expected execution time window, payload type, and priority. The target area of ​​each task can be understood as the latitude and longitude range of the ground area to be observed. The expected execution time window is a time period within which the user hopes the task will be executed. The payload type is the type of satellite sensor required to perform the observation task, such as a visible light camera or Synthetic Aperture Radar (SAR). For example, payload types can include visible light payloads, SAR payloads, hyperspectral payloads, and infrared payloads; the system matches corresponding satellite resources based on the payload type. Priority indicates the importance of the task and is used to determine the order of task execution in case of resource conflicts. For example, it can be divided into five levels: Level 1 represents an urgent task, Level 2 represents high priority, Level 3 represents medium priority, Level 4 represents low priority, and Level 5 represents a normal task.

[0029] It should be noted that the system can automatically verify the completeness and validity of task parameters, specifically including: coordinate verification, time window verification, load type verification, priority verification, and parameter integrity verification. Coordinate verification checks whether the latitude and longitude range is within the valid interval and whether the coordinate format is correct. Time window verification checks whether the time format conforms to the standard and whether the task execution time window is within the system's plannable time range. Load type verification checks whether the load type is in the system's supported load type list. Priority verification checks whether the priority level is within the range of 1-5. Parameter integrity verification checks whether all required parameters are provided. After successful verification, the system can assign a unique identifier to each task, in the format TASK_YYYYMMDD_HHMMSS (a random 6-digit number), such as "TASK_20240101_143025". This identifier ensures the uniqueness and traceability of the task throughout the entire planning process.

[0030] Step S20: For each observation task, perform multi-constraint analysis based on the current satellite orbit data and task parameters to obtain at least one candidate observation scheme for each task. The multi-constraint analysis includes at least orbit visibility analysis and meteorological constraint analysis.

[0031] It should be noted that multi-constraint analysis refers to the process of simultaneously calculating and evaluating multiple constraints during planning. It is understood that the satellite mission planning method based on multi-constraints in this application emphasizes not a single constraint, but a synergistic analysis of constraints. Therefore, multi-constraint analysis here includes at least orbital visibility analysis and meteorological constraint analysis. Orbital visibility analysis can be understood as calculating the physical visibility window within which the satellite can fly over and cover the target area, based on satellite orbital mechanics. Meteorological constraint analysis can be understood as further evaluating whether weather conditions meet the observation requirements of the mission payload type, based on the time window obtained from orbital visibility analysis and combined with real-time or predicted meteorological data. Candidate observation schemes refer to feasible execution opportunities that meet the requirements of multi-constraint analysis and are initially screened for a single mission. It typically includes which satellite, in which precise transit time window, and in what attitude the mission will be executed. It should be noted that a mission may correspond to multiple candidate schemes provided by different satellites or at different times.

[0032] Step S30: Based on the candidate observation schemes of all observation tasks, a collaborative planning scheme integrating multiple constraints is generated through a collaborative decision-making model. The collaborative decision-making model takes maximizing the overall task return as its objective function, and the objective function integrates the quality score of each task and the resource conflict penalty term between tasks.

[0033] It should be noted that the collaborative decision-making model can be understood as a mathematical optimization model, whose input is all candidate observation schemes generated for all tasks in step S20. The collaborative decision-making model can solve the resource competition problem between multiple tasks from a global perspective. The core objective that the collaborative decision-making model needs to maximize is the overall task return. It is a comprehensive quantitative index, which can be composed of two parts: the quality score of each task and the resource conflict penalty term between tasks. The quality score of each task can be understood as a quantitative score of the observation quality of each candidate observation scheme. The higher the score, the better the observation conditions. When two or more tasks are assigned to the same satellite and their execution times are too close, causing the satellite to be unable to complete attitude adjustment or data transmission in time, resource conflicts will occur. The resource conflict penalty term between tasks is used to quantify the cost of such conflicts in the objective function. The collaborative planning scheme is the final output of the collaborative decision-making model after it runs, which can represent a global optimal or suboptimal task allocation table. The collaborative planning scheme selects a target observation scheme for each task and ensures that all selected schemes do not conflict in terms of time and satellite resources, and maximizes the overall task return of the system.

[0034] Step S40: Based on the collaborative planning scheme, generate a sequence of mission instructions to drive the satellite to perform its tasks.

[0035] It should be noted that the mission instruction sequence can be a set of instruction codes that can be directly recognized and executed by the satellite ground control system. For example, based on the collaborative planning scheme obtained in step S30, it can automatically generate a set of instruction codes that can be directly recognized and executed by the satellite ground control system, thereby controlling a series of actions such as satellite platform attitude, payload power on / off, data storage and downlink. For example, the system automatically generates complete mission planning results, including: mission execution schedule, satellite control instruction sequence, multi-mission collaborative execution scheme, alternative schemes, and contingency plans. It supports multiple output formats, such as JSON, XML, and text files, facilitating integration with external systems such as the satellite ground system and mission management system to achieve execution monitoring of the planning results.

[0036] This embodiment employs a unified and integrated task parameter receiving and verification mechanism, resolving the issues of chaotic input formats from multiple task sources and cumbersome manual verification, thus achieving standardization in task management and improved entry efficiency. By utilizing parallel fusion analysis technology of orbital and meteorological data, it overcomes the problems of neglecting meteorological dynamics or computational redundancy in traditional serial analysis, ensuring simultaneous assurance of spatiotemporal feasibility and environmental suitability for candidate solutions. The adoption of a collaborative decision-making model aimed at maximizing overall returns quantifies and balances the quality benefits of each task with the resource conflict costs arising from competition for the same satellite and time within a single objective function, resolving the issues of low overall system efficiency and low resource utilization caused by independent optimization of single tasks, and achieving a leap from local optima to global optima. The use of automatic generation technology from planning schemes to control commands solves the problems of error-prone and time-consuming manual planning results, achieving closed-loop automation from planning to execution. This embodiment, through multi-constraint parallel screening and multi-task collaborative optimization, significantly improves global optimization capabilities and adaptability while ensuring the physical feasibility of the scheme, making it suitable for high-time-efficiency scenarios such as multi-task collaborative planning, emergency response, and ultra-large-scale task scheduling.

[0037] In one implementation, step S20 includes: Step S201: For each observation task, perform orbit visibility analysis based on the current satellite orbit data and the target area corresponding to the observation task, calculate the satellite's transit time window over the target area, and determine the observation mode.

[0038] It should be noted that Two-Line Orbital Element (TLE) data can be understood as a standard data format describing the orbital position and velocity of a space object at a specific moment. It consists of two lines of code and is the fundamental data for orbit calculations. Simplified General Perturbations (SGP) is a mathematical model for orbit prediction. Using TLE data as input, it can calculate the satellite's position and velocity at any future moment. The transit time window refers to the time interval during which a satellite flies over and is physically able to see a target area on the ground.

[0039] Step S201 includes: inputting two lines of orbit data into a simplified perturbation model to calculate the transit time window of the satellite over the target area; within the transit time window, calculating the lateral distance between the target point and the satellite's nadir trajectory, where the target point is the center point of the target area; and obtaining the observation mode based on the relationship between the lateral distance, the half-width of the satellite payload, and the lateral glide coverage area, including normal pushbroom mode, lateral glide observation mode, and unobservable mode.

[0040] It should be noted that the satellite's nadir trajectory can be understood as the continuously moving trajectory line formed by the vertical projection of the satellite's center of mass onto the Earth's surface. Lateral distance refers to the shortest ground distance between the target point and the satellite's nadir trajectory. The satellite payload's half-swath width refers to half the width of the ground coverage area of ​​the satellite's sensors perpendicular to its flight direction. The lateral roll coverage range can be understood as the farthest ground distance from the nadir trajectory that the satellite can observe by rolling laterally.

[0041] For example, the transit time window of the satellite over the target area can be calculated based on TLE orbit data and the SGP4 model. First, a geometric relationship determination mechanism between the target position and the satellite swath width is established to determine whether lateral observation is needed: the lateral distance between the target point and the satellite's nadir trajectory is calculated. ,like If the width is ≤ half the swath width, select the normal push-broom mode; it is usable throughout the entire transit time window without further judgment. If the width is < half the swath width If the lateral tilt coverage area is less than or equal to the available time period, then the lateral tilt observation mode should be selected. Further determination of the available time period is needed based on the elevation angle relationship. If the area covered by the lateral swing is not observable, it is considered unobservable. The area covered by the lateral swing can be set to the ground area corresponding to ±30°.

[0042] Step S202: Based on the multi-source meteorological data acquisition mechanism, acquire real-time meteorological data that is spatiotemporally matched with the transit time window.

[0043] It should be noted that the multi-source meteorological data acquisition mechanism is a data acquisition mechanism that integrates multiple meteorological data sources. When a data source becomes unavailable, it can automatically switch to a backup source to ensure the continuity and reliability of data services. Spatiotemporal matching can be understood as the acquired meteorological data corresponding to the satellite's transit time window in time and to the target area location in space.

[0044] Step S203: Based on the cloud sensitivity rules corresponding to the load type, process the cloud coverage in the real-time meteorological data to obtain the meteorological constraint score.

[0045] It should be noted that the meteorological constraint score is used to assess the quality of meteorological conditions during the observation period. For example, a sensitivity-differentiated processing mechanism based on cloud cover and payload type can be established. Different cloud quality analysis models can be matched according to different satellite payload types. Specifically, for visible light payloads, which have high cloud sensitivity, a cloud cover threshold of 40% can be set. The corresponding calculation formula for the cloud quality score of visible light payloads is as follows:

[0046] in, Cloud quality score representing visible light load. The score indicates cloud cover. When cloud cover is ≤40%, the score is >0; when cloud cover is >40%, the score is 0, indicating that it is not suitable for observation.

[0047] For SAR payloads, which have low cloud sensitivity and can penetrate clouds, a cloud coverage threshold does not need to be set. The corresponding calculation formula for the cloud quality score of SAR payloads is as follows: in, This represents the cloud quality score of the SAR payload, with 1.0 indicating that it is always a perfect score.

[0048] For hyperspectral payloads, cloud sensitivity is moderate, and the cloud cover threshold can be set to 20%. The corresponding calculation formula for the cloud quality score of hyperspectral payloads is as follows: in, The cloud quality score represents the hyperspectral payload. A score > 0 indicates cloud cover ≤ 20%, while a score > 20% indicates cloud cover > 20%, signifying unsuitability for observation. The system can automatically match the appropriate cloud quality analysis model based on the payload type configured for the task and incorporate the cloud quality score into the subsequent multi-factor comprehensive scoring model. It should be noted that the system can maintain a historical meteorological data cache, providing a historical data foundation for subsequent collaborative decision-making models. For example, cloud data from the most recent 7-30 days can be cached with a time resolution of 1 hour.

[0049] Step S204: Based on the observation mode, transit time window, and meteorological constraint score, generate candidate observation schemes.

[0050] It should be noted that the candidate observation schemes integrate key information such as transit time windows, observation modes, and meteorological constraint scores.

[0051] This embodiment employs a method combining orbital mechanics models with geometric discrimination rules to address the problem of determining the optimal observation attitude solely based on the transit window. This enables precise pattern recognition for each observation opportunity, effectively distinguishing between direct observation, side-swinging, and infeasible scenarios. By utilizing multi-source meteorological data acquisition and spatiotemporal matching techniques, it overcomes the issues of inaccurate cloud cover assessment and data gaps caused by static or single meteorological data sources, achieving dynamic and accurate understanding of meteorological conditions for each transit window. Furthermore, by adopting cloud sensitivity differentiation rules based on payload type, it addresses the problem that uniform meteorological thresholds cannot adapt to the physical characteristics of different payloads, enabling refined and rational evaluation of meteorological constraints for heterogeneous payloads such as optical and radar payloads. Finally, by fusing multi-dimensional information such as observation modes, time windows, and meteorological scores to generate candidate schemes, it solves the problems of isolated constraint evaluation and information fragmentation in traditional methods. This allows for the rapid output of a series of physically feasible, meteorologically suitable, and informationally complete high-quality candidate observation schemes for each task, laying a solid foundation for subsequent collaborative optimization.

[0052] Based on the above embodiments and implementation methods, in one implementation method, step S201 further includes: if the observation mode is a side-swing observation mode, calculating the elevation angle at each moment during the transit based on a simplified perturbation model to form elevation angle time series data; in the elevation angle time series data, filtering out all moments with elevation angles greater than or equal to the minimum observation elevation angle threshold; merging consecutive time periods among the filtered moments to obtain at least one usable time period; for each usable time period, establishing a local coordinate system with the satellite as the origin, and calculating the observation vector pointing from the satellite to the target point, wherein the X-axis of the local coordinate system is along the velocity direction, the Y-axis is perpendicular to the orbital plane, and the Z-axis points to the Earth's center; calculating the side-swing angle by the angle between the observation vector and the Z-axis, and determining the side-swing direction based on the projection of the observation vector onto the Y-axis.

[0053] It should be noted that elevation angle time-series data refers to a series of satellite elevation angle values ​​relative to a ground target point, arranged chronologically during satellite transit, reflecting the changing geometric relationship between the satellite and the target point. The minimum observation elevation angle threshold refers to a preset minimum elevation angle limit for the satellite to ensure observation quality. Consecutive time period merging refers to combining time points in the time series that all meet certain conditions, such as elevation angles greater than the threshold and being sequentially connected, into a single complete time interval. A usable time period refers to a continuous period within the transit time window where the satellite elevation angle consistently meets the minimum requirement and can be used for side-swing observations. It should be noted that a single transit may generate multiple independent usable time periods.

[0054] For example, multiple independent available time periods can be identified based on elevation angle time series data. Specifically, the elevation angle at each moment during the transit can be calculated based on the SGP4 model to form an elevation angle time series data sequence; a minimum observation elevation angle threshold can be set. The maximum side sway angle can be determined based on the load type and mission requirements; by traversing the time series data of elevation angles, all elevation angles ≥ The continuous time interval is used as the available time period. Each time period must meet the minimum duration requirement, which can be set to 30 seconds. The change in elevation angle within the time period should be smooth to avoid abrupt changes caused by attitude adjustment. For complex crossings with multiple elevation angle peaks, the available time periods near each peak are identified to form multiple independent candidate time periods, providing a complete candidate set for subsequent multi-factor evaluation.

[0055] Furthermore, the magnitude and directionality of the lateral tilt angle can be precisely calculated using a local coordinate system. Specifically, a local coordinate system can be established with the satellite as the origin, where the X-axis is along the velocity direction, the Y-axis is perpendicular to the orbital plane, and the Z-axis points towards the Earth's center. The observation vector can then be calculated. The angle between the lateral angle and the Z-axis is the lateral angle. The calculation formula is: in, The observation vector is directed from the satellite towards the target point. This is the unit vector in the nadir direction. The lateral tilt direction is indicated by the observed vector. The projection on the Y-axis is determined by whether the projection is positive or negative, indicating a rightward swing or a leftward swing.

[0056] In this embodiment, a method is adopted to generate elevation angle time-series data based on the orbital model and perform threshold filtering and time period merging. This solves the problem of traditionally using only a single maximum elevation angle moment to represent the entire transit window while ignoring the actual available observation time. It realizes the extraction of multiple fine, continuous, and available specific observation time periods from the transit window. By constructing observation vectors and performing geometric calculations in the instantaneous local coordinate system of the satellite body, the problems of complex calculations and lack of intuitive physical meaning in the global or Earth fixed coordinate system are solved. This achieves direct, accurate, and stable calculation of side-slip angles and directions, providing reliable input for the generation of satellite attitude control commands. Through in-depth and refined processing of the side-slip observation mode, this implementation not only ensures the geometric quality of the observation but also clarifies the time window and spatial orientation of each action, significantly improving the feasibility of mission planning and control accuracy.

[0057] In one implementation, step S202 includes: acquiring gridded cloud coverage data from at least one external meteorological data source according to a multi-source meteorological data acquisition mechanism, wherein the external meteorological data source includes a primary data source and a backup data source.

[0058] It should be noted that gridded cloud cover data refers to a data format provided by meteorological data sources that divides a geographical area into regular grids, such as latitude and longitude grids, and provides a single cloud cover value for each grid point. This is a common format for numerical weather prediction and satellite inversion data. The multi-source meteorological data acquisition mechanism includes switching to a backup data source if the primary data source fails to acquire the data, and switching to the most recently cached data if both the primary and backup data sources fail.

[0059] For example, both Open-Meteo and Windy API data sources can be integrated, with automatic switching based on availability. The switching logic is as follows: The system first attempts to call the primary data source, the Open-Meteo API. If unavailable, it automatically switches to the backup data source, the Windy API. If both sources are unavailable, the most recently successfully retrieved cached data is used, with a cache validity period that can be set to 24 hours. Switching trigger conditions include, but are not limited to: API response timeout (e.g., response time > 5 seconds), abnormal HTTP status code, incorrect data format or empty data returned, or three consecutive failed requests.

[0060] Step S202 further includes: based on the specific time within the transit time window and the specific location of the target area, matching the gridded cloud coverage data using a spatiotemporal interpolation algorithm to obtain cloud coverage that spatiotemporally matches the transit time window. The spatiotemporal interpolation algorithm includes linear interpolation in the time dimension and bilinear interpolation in the spatial dimension.

[0061] It should be noted that spatiotemporal interpolation is a computational method used to estimate data values ​​at specific time and geographical locations. Its core is to extrapolate values ​​at continuous spatiotemporal points from known data at discrete time and spatial points—i.e., gridded data. For example, spatiotemporal interpolation can be used to obtain cloud cover for each time period. The spatiotemporal interpolation algorithm includes two steps: temporal interpolation and spatial bilinear interpolation. Specifically, in the time dimension, meteorological APIs provide cloud data at the top of the hour. Satellite transit times may fall between the top of the hour, so linear interpolation can be used to calculate cloud cover at non-top-of-the-hour times. The calculation formula is:

[0062] in, For a moment cloud coverage and For adjacent hourly times and Regarding cloud coverage, it should be noted that... < < .

[0063] In the spatial dimension, the meteorological API provides grid point data, such as a 0.1°×0.1° grid. It can be understood that the center point of the target area may lie between grid points. Therefore, the cloud cover at non-grid points can be calculated using a weighted average of four adjacent grid points. The calculation formula is as follows: in, For position cloud coverage The cloud cover of four adjacent grid points, weighted Calculated based on the inverse ratio of the distance between the target point and the grid points.

[0064] This implementation employs an automatic switching mechanism between primary and backup data sources and a local cache degradation mechanism. This addresses the critical vulnerability of relying on a single external data source, which can lead to planning interruptions due to network or service failures, thus achieving high reliability and business continuity in meteorological data acquisition. Furthermore, by using a spatiotemporal matching algorithm that integrates temporal linear interpolation and spatial bilinear interpolation, the inherent spatiotemporal scale mismatch between static grid meteorological data and dynamic satellite transit windows and specific target locations is resolved. This enables refined, point-to-point, and time-specific assessment of the meteorological conditions required for each candidate observation opportunity. This implementation, through the combination of a robust data acquisition chain and a precise spatiotemporal analysis method, significantly improves the usability and accuracy of meteorological constraint analysis, thereby enhancing the decision-making reliability of the entire mission planning system in the face of uncertain external data environments.

[0065] In the second embodiment, after step S204, the method further includes: Step S205: When a single observation task corresponds to multiple candidate observation schemes, the quality score of each candidate observation scheme is calculated through a multi-factor weighted scoring model based on the observation mode, elevation angle, side swing angle and meteorological constraint score corresponding to each candidate observation scheme.

[0066] It should be noted that the multi-factor weighted scoring model combines factors such as the observation mode, elevation angle, side-swing angle, and meteorological constraint score corresponding to each candidate observation scheme, assigning a weight coefficient to each factor. Through weighted calculation, a comprehensive quantitative index is obtained, namely the quality score of each candidate observation scheme, which is used to objectively compare the merits of different candidate observation schemes. For example, a higher score indicates a better expected observation quality for that scheme.

[0067] For example, for a single mission, multiple transit periods from multiple satellites can be evaluated to establish a multi-factor comprehensive scoring model. This model integrates factors such as elevation angle, side-swing angle, and payload sensitivity to score each candidate transit period. Payload sensitivity may include the impact of cloud cover. Specifically, for a single mission, the system can iterate through the visibility analysis in step S201 to obtain all observable satellites for that mission, and then score the quality of each transit period for each satellite, forming a set of all candidate transit periods for that mission. For example, the set of candidate transit periods may include transit periods 1 and 2 for satellite A, transit periods 1 and 2 for satellite B, etc. Then, all candidate transit periods are uniformly scored and ranked. Specifically, the multi-factor comprehensive scoring formula is:

[0068] in, The corresponding weighting coefficient for the elevation angle can be set to 0.4; The corresponding weighting coefficient for the lateral sway angle can be set to 0.3; The corresponding weighting coefficient for load sensitivity can be set to 0.3. The weights must satisfy... The weights can be dynamically adjusted based on task type and decision preferences; for example, for emergency tasks, the weights can be increased. Weighting can be improved in routine tasks. Weighting. Specifically, the quantification formulas for each factor are as follows:

[0069] in, This indicates a high-angle rating. The optimal elevation angle for the current time period. The minimum observation elevation angle threshold can be between 30° and 45°. The larger the elevation angle, the higher the score. The value range can be set to [0,1].

[0070] in, Indicates the cost of the lateral swing angle. The lateral swing angle ranges from 0° to 45°. The larger the lateral swing angle, the higher the cost. The value range is [0,1].

[0071] Load sensitivity can be scored by matching the appropriate cloud quality analysis model according to the load type. include: in, Sensitivity score for visible light load. Sensitivity score for SAR payload. Sensitivity score for hyperspectral load.

[0072] Step S206: Based on the quality score, sort the multiple candidate observation schemes to obtain the sorting results.

[0073] For example, the system can automatically calculate the quality score of all candidate time periods, sort them from high to low, and output the Top-K preferred solution, where K can be 5-10.

[0074] Step S207: Based on the sorting results, a predetermined number of schemes are selected from multiple candidate observation schemes to obtain a set of candidate schemes corresponding to a single observation task.

[0075] For example, the candidate scheme set includes Top-K preferred schemes, each of which contains complete information such as satellite identifier, time period information, quality score, elevation angle, side-swing angle, and cloud coverage, providing a candidate set for multi-task collaborative scheduling. The time period information may include the observation start time and observation end time.

[0076] In this embodiment, a multi-factor weighted scoring model is used to comprehensively and quantitatively evaluate each candidate observation scheme. This solves the problems of bias and subjectivity in evaluation caused by relying on a single indicator or human experience in traditional methods, and achieves a scientific, objective, and unified measurement of the observation quality of the schemes. By employing a mechanism that sorts and selects a predetermined number of optimal schemes based on quality scores, the problem of information overload leading to computational combinatorial explosion when faced with a large number of feasible schemes is solved. This significantly reduces the complexity of subsequent multi-task collaborative planning while retaining high-quality alternative schemes. This implementation provides high-quality, standardized input for subsequent global collaborative decision-making by refining the scoring and optimization of candidate schemes for individual tasks, thus balancing planning efficiency and solution optimization overall.

[0077] Based on the second embodiment described above, it can be understood that the collaborative planning scheme includes multiple target observation schemes. Therefore, in one implementation, step S30 includes: Step S301: Construct a state space based on the current satellite status information, the candidate scheme set for each mission, and the mission parameters of each mission. The state space is also used to perform time-series prediction through a long short-term memory network to obtain future environmental information, and to model the constraint relationships between missions through a graph neural network to obtain mission relationship information.

[0078] It should be noted that, for multi-task planning scenarios, the system adopts a Deep Reinforcement Learning (DRL) scheduling framework. This framework unifies all multi-constraints, such as orbital constraints, meteorological constraints, payload constraints, attitude constraints, time constraints, and resource constraints, into an end-to-end optimization problem, thereby achieving multi-constraint collaborative optimization and globally optimal planning. In the deep reinforcement learning model, the state space can be a multi-dimensional vector, used to describe the information set of all key states of the entire satellite mission planning system at a given moment.

[0079] It should be noted that an initial task relationship graph with tasks as nodes can be constructed based on the task parameters of each task and the temporal and resource relationships of candidate observation schemes. A graph neural network (GNN) is then used to learn this initial task relationship graph, outputting a task compatibility matrix representing the conflict, dependency, or synergistic relationships between tasks. This task compatibility matrix is ​​incorporated into the state space. Historical meteorological data and satellite state data are collected to construct a fixed-length historical time series. A long short-term memory (LSTM) network is used to process this historical time series to predict future meteorological conditions and resource availability trends. The prediction results are then incorporated into the state space as forward-looking information.

[0080] For example, a Markov Decision Process (MDP) model for task scheduling can be constructed based on the current satellite status information, the candidate scheme set for each task, and the task parameters of each task. This model can include task information, satellite status, candidate time period information, future environmental information predicted by LSTM, a task relationship graph modeled by GNN, the current scheduling status, and time information. Among these, task information includes task identifier, priority, target area, payload type, and task execution time window; satellite status includes satellite identifier, current attitude, available time period, and orbital parameters; candidate time period information includes satellite identifier, time period information, quality score, elevation angle, side-slip angle, and cloud coverage; future environmental information predicted by LSTM includes predictions of cloud evolution trends and satellite resource availability for the next 6-12 hours; the task relationship graph modeled by GNN can include a task compatibility matrix, constraint propagation information, and conflict / dependency / cooperation relationships between tasks; the current scheduling status includes a list of assigned tasks, occupied time periods, remaining candidate schemes, and resource occupancy; and time information includes the current system time, task execution time window, and prediction time range.

[0081] LSTM can be used for time-series forecasting to predict cloud evolution trends and satellite resource availability over the next 6-12 hours. The specific steps are as follows: First, acquire historical meteorological data and cached historical orbital data, including historical transit time windows and historical elevation angle time-series data. The data can be formatted as a time series, with features including timestamps, cloud cover, payload sensitivity scores, and satellite resource availability. The LSTM network is configured with 2-3 layers and 64-128 hidden units. The input is historical time-series data with a time window length of 24-72 hours. The output is environmental forecast information for the next 6-12 hours, including cloud cover, payload sensitivity scores, and satellite resource availability for each future time period. The forecast time resolution is 1 hour, including point prediction values ​​and confidence intervals. The future environmental information predicted by LSTM is combined with current real-time meteorological data to form complete time-series data, including both current and future forecast data. This complete time-series data is integrated into the DRL state space to enhance the foresight of decision-making.

[0082] Deep learning and modeling of multi-task, multi-constraint relationships can be achieved using Generative Neural Networks (GNNs). An initial structure of the task relationship graph can be constructed based on task configuration data and visibility analysis data. Task configuration data can include task identifiers, priorities, target regions, payload types, and task execution time windows; visibility analysis data can include transit time windows. In the initial structure of the task relationship graph, nodes represent tasks, and edges represent relationships between tasks. Based on the spatial proximity of target regions, the overlap of time windows, and the compatibility of payload types, potential conflict, dependency, and cooperative relationships between tasks are initially identified. The GNN network is then input into this task relationship graph and utilizes the multi-constraint information from the preceding steps, such as orbital constraints, time constraints, and resource constraints. Through graph convolutional layers and attention mechanisms, it automatically learns complex constraint relationships between tasks, including conflict, dependency, cooperative relationships, and resource competition. Conflict relationships can include resource conflicts or attitude adjustment time conflicts within the same satellite and time period; dependencies can be understood as task B must be executed after task A is completed; cooperative relationships include tasks in adjacent regions being able to execute consecutively to reduce attitude adjustments; and resource competition relationships are multiple tasks competing for the optimal time slot on the same satellite. The GNN network outputs a task compatibility matrix and constraint propagation information, such as a task conflict matrix, dependency graph, collaboration graph, and constraint propagation path, to guide multi-constraint collaborative scheduling decisions.

[0083] Step S302: Input the state space into the deep reinforcement learning decision model, output the initial scheduling scheme, and optimize the strategy to maximize the overall task reward. The initial scheduling scheme includes multiple intermediate observation schemes. The deep reinforcement learning decision model is an Actor-Critic architecture, in which the Actor network outputs actions, the Critic network evaluates the state value, and the actions are used to assign intermediate observation schemes to each observation task.

[0084] For example, an Actor-Critic architecture can be adopted. The Actor network outputs actions based on the current state, allocating optimal solutions, i.e., intermediate observation solutions, to the task. These intermediate observation solutions can include satellite identifiers and time period information. The Critic network evaluates the value of the state and maximizes the long-term cumulative reward through policy gradient optimization. The objective function is:

[0085] in, To score the time period quality of task i; Let i be the priority weight of task i; is the conflict penalty value for task i, based on the conflict matrix output by the GNN.

[0086] Step S303: Perform multi-objective optimization based on the initial scheduling scheme to generate a Pareto optimal solution set, which includes multiple objective observation schemes.

[0087] For example, the initial scheduling scheme can be input into a multi-objective optimization neural network, which outputs multiple optimization objective values, such as task completion rate, average observation quality, and resource utilization. Algorithms like NSGA-II are then used to generate a Pareto optimal solution set, such as 5-10 alternative schemes, for decision-makers to select based on current needs. The system can collect execution feedback during actual operation, such as task success, actual observation quality, and actual cloud cover, compare this feedback with LSTM prediction data, and update the LSTM prediction model and DRL decision model to achieve continuous optimization.

[0088] This implementation employs a comprehensive state-space construction method that integrates real-time state, LSTM future environment prediction, and GNN task relationship modeling. This addresses the short-sightedness and conflict issues in traditional planning models caused by incomplete information, lack of foresight, and insufficient understanding of complex relationships, providing a comprehensive, accurate, and predictive information foundation for intelligent decision-making. By using a deep reinforcement learning model based on an Actor-Critic architecture for end-to-end decision optimization, it overcomes the difficulty of finding high-quality global solutions in complex, multi-constraint environments using traditional rule-based or simple heuristic algorithms. Furthermore, the introduction of subsequent multi-objective optimization to generate a Pareto-optimal solution set solves the problem that a single objective function cannot simultaneously accommodate various decision preferences and the flexibility of business scenarios. Through the organic combination of predictive modeling, deep reinforcement learning, and multi-objective optimization, the global optimality, robustness, and practicality of satellite mission planning schemes in large-scale, complex scenarios are significantly improved.

[0089] Based on the same inventive concept, such as Figure 3 As shown, this application also provides a satellite mission planning system based on multiple constraints, which includes: Task configuration module 10 is used to receive task parameters for multiple observation tasks. The task parameters include at least the target area, expected execution time window, payload type and priority of each task. The multi-constraint analysis module 20 is used to perform multi-constraint analysis on each of the observation tasks based on the current satellite orbit data and the task parameters, and to obtain at least one candidate observation scheme for each task. The multi-constraint analysis includes at least orbit visibility analysis and meteorological constraint analysis. The scheduling decision module 30 is used to generate a collaborative planning scheme that integrates multiple constraints based on the candidate observation schemes of all observation tasks through a collaborative decision model. The collaborative decision model takes maximizing the overall task return as the objective function, and the objective function integrates the quality score of each task and the resource conflict penalty term between tasks. Result generation module 40 is used to generate a sequence of mission instructions for driving the satellite to perform tasks.

[0090] The multi-constraint analysis module 20 also includes a visibility analysis unit and a meteorological data integration unit. The visibility analysis unit is used to perform orbit visibility analysis for each observation task based on the current satellite orbit data and the target area corresponding to the observation task. The meteorological data integration unit is used to acquire real-time meteorological data that matches the transit time window in time and space according to the multi-source meteorological data acquisition mechanism.

[0091] It should be noted that, in one implementation method, such as Figure 4 As shown, the satellite mission planning system based on multiple constraints also includes a visualization module 50, used for three-dimensional visualization demonstration and verification of the collaborative planning scheme. The visualization module includes: a payload simulation unit, used to dynamically calculate and generate a pushbroom strip model conforming to the Earth's curvature based on satellite payload parameters, real-time orbit data, and observation geometric parameters in the collaborative planning scheme; a synchronization control unit, used to keep the simulation display of the pushbroom strip model synchronized with the planned observation period, and to keep the attitude of the observation cone synchronized with the planned side-swing angle; and an interactive demonstration unit, used to display the pushbroom strip model and the observation cone in real-time in a three-dimensional scene, and to provide an interactive interface for users to compare and verify multiple schemes. For example, the visualization module 50 can be implemented using the Cesium 3D engine. It can establish a satellite payload parameter library, dynamically calculate the ground coverage swath based on real-time orbital altitude and field of view, and use the Spherical Linear Interpolation (SLERP) algorithm to generate strips conforming to the Earth's curvature, eliminating ghosting and distortion, thereby achieving realistic payload parameterized pushbroom simulation. The pushbroom simulation is precisely synchronized with the observation period, and the attitude and side-swing angle of the cone are synchronized in real-time, supporting seamless switching between multiple time periods. Through various 3D demonstration effects such as dynamic tracking of the view frustum, real-time generation of curved surface strips, synchronous demonstration of lateral sway posture, and interactive comparison of multiple solutions, decision-makers are provided with a WYSIWYG planning experience. The complete technical process is as follows: Figure 5 As shown.

[0092] It should be noted that the satellite mission planning system based on multiple constraints provided in this application embodiment and the satellite mission planning method based on multiple constraints provided in this application embodiment are based on the same application concept. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned satellite mission planning method based on multiple constraints, and the repeated parts will not be described again.

[0093] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0094] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application. Therefore, the scope of this application should not be limited to the above embodiments. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A satellite mission planning method based on multiple constraints, characterized in that, The method includes: Receive task parameters for multiple observation tasks, wherein the task parameters include at least the target region, expected execution time window, payload type and priority of each task; For each observation task, a multi-constraint analysis is performed based on the current satellite orbit data and the task parameters to obtain at least one candidate observation scheme for each task. The multi-constraint analysis includes at least orbit visibility analysis and meteorological constraint analysis. Based on the candidate observation schemes for all observation tasks, a collaborative planning scheme integrating multiple constraints is generated through a collaborative decision-making model. The collaborative decision-making model takes maximizing the overall task return as its objective function, and the objective function integrates the quality score of each task and the resource conflict penalty term between tasks. Based on the aforementioned collaborative planning scheme, a sequence of mission instructions is generated to drive the satellite's execution.

2. The method as described in claim 1, characterized in that, The step of performing multi-constraint analysis on each observation task based on the current satellite orbit data and the task parameters to obtain at least one candidate observation scheme for each task includes: For each observation task, orbital visibility analysis is performed based on the current satellite orbit data and the target area corresponding to the observation task. The transit time window of the satellite over the target area is calculated, and the observation mode is determined. Based on the multi-source meteorological data acquisition mechanism, real-time meteorological data that is spatiotemporally matched with the transit time window is acquired; Based on the cloud sensitivity rules corresponding to the load type, the cloud coverage in the real-time meteorological data is processed to obtain a meteorological constraint score; Based on the observation mode, the transit time window, and the meteorological constraint score, candidate observation schemes are generated.

3. The method as described in claim 2, characterized in that, The steps of performing orbital visibility analysis based on the current satellite orbit data and the target area corresponding to the observation task for each observation task, calculating the satellite's transit time window over the target area, and determining the observation mode include: Input the two lines of orbit data into a simplified perturbation model to calculate the satellite's transit time window over the target area; Within the transit time window, the lateral distance between the target point and the satellite's nadir point trajectory is calculated, wherein the target point is the center point of the target area; Based on the relationship between the lateral distance, the half-width of the satellite payload, and the coverage area of ​​the side-swing, the observation modes are obtained, including normal pushbroom mode, side-swing observation mode, and unobservable mode.

4. The method as described in claim 3, characterized in that, The observation mode is obtained based on the relationship between the lateral distance, the half-swath width of the satellite payload, and the side-swing coverage area. The observation mode includes a normal pushbroom mode, a side-swing observation mode, and an unobservable step, and also includes: If the observation mode is the side-swing observation mode, the elevation angle at each moment during the passage is calculated based on the simplified perturbation model to form elevation angle time series data; In the elevation angle time series data, all times when the elevation angle is greater than or equal to the minimum observed elevation angle threshold are selected; Merge consecutive time periods from the selected time periods to obtain at least one usable time period; For each available time period, a local coordinate system with the satellite as the origin is established, and the observation vector from the satellite to the target point is calculated, wherein the X-axis of the local coordinate system is along the velocity direction, the Y-axis is perpendicular to the orbital plane, and the Z-axis points to the Earth's center; The lateral swing angle is calculated by the angle between the observed vector and the Z-axis, and the lateral swing direction is determined by the projection of the observed vector onto the Y-axis.

5. The method as described in claim 2, characterized in that, The step of acquiring real-time meteorological data that spatiotemporally matches the transit time window based on the multi-source meteorological data acquisition mechanism includes: According to the multi-source meteorological data acquisition mechanism, gridded cloud coverage data is obtained from at least one external meteorological data source. The external meteorological data source includes a primary data source and a backup data source. The multi-source meteorological data acquisition mechanism includes switching to the backup data source if the primary data source fails to acquire the data, and switching to the most recently cached data if both the primary data source and the backup data source fail to acquire the data. Based on the specific time within the transit time window and the specific location of the target area, the gridded cloud coverage data is matched using a spatiotemporal interpolation algorithm to obtain cloud coverage that spatiotemporally matches the transit time window. The spatiotemporal interpolation algorithm includes linear interpolation in the time dimension and bilinear interpolation in the spatial dimension.

6. The method as described in claim 2, characterized in that, After the step of generating candidate observation schemes based on the observation mode, the transit time window, and the meteorological constraint score, the method further includes: When a single observation task corresponds to multiple candidate observation schemes, the quality score of each candidate observation scheme is calculated by a multi-factor weighted scoring model based on the observation mode, elevation angle, side swing angle and meteorological constraint score corresponding to each candidate observation scheme. Based on the quality score, the multiple candidate observation schemes are ranked to obtain the ranking result; Based on the sorting results, a predetermined number of schemes are selected from the multiple candidate observation schemes to obtain a set of candidate schemes corresponding to a single observation task.

7. The method as described in claim 6, characterized in that, The collaborative planning scheme includes multiple target observation schemes. Based on the candidate observation schemes for all observation tasks, a collaborative decision model generates a collaborative planning scheme that integrates multiple constraints. The collaborative decision model takes maximizing the overall task reward as its objective function. The step of integrating the quality score of each task and the resource conflict penalty term between tasks into the objective function includes: A state space is constructed based on the current satellite status information, the candidate scheme set for each mission, and the mission parameters of each mission. The state space is also used to perform time-series prediction through a long short-term memory network to obtain future environmental information, and to obtain mission relationship information by modeling the constraint relationships between missions through a graph neural network. The state space is input into a deep reinforcement learning decision model, which outputs an initial scheduling scheme and optimizes the strategy to maximize the overall task reward. The initial scheduling scheme includes multiple intermediate observation schemes. The deep reinforcement learning decision model is an Actor-Critic architecture, in which the Actor network outputs actions and the Critic network evaluates the state value. The actions are used to assign the intermediate observation schemes to each observation task. Based on the initial scheduling scheme, multi-objective optimization is performed to generate a Pareto optimal solution set, which includes multiple objective observation schemes.

8. A satellite mission planning system based on multiple constraints, characterized in that, The satellite mission planning system based on multiple constraints includes: The task configuration module is used to receive task parameters for multiple observation tasks. The task parameters include at least the target region, expected execution time window, payload type, and priority of each task. The multi-constraint analysis module is used to perform multi-constraint analysis on each of the observation tasks based on the current satellite orbit data and the task parameters, and to obtain at least one candidate observation scheme for each task. The multi-constraint analysis includes at least orbit visibility analysis and meteorological constraint analysis. The scheduling decision module is used to generate a collaborative planning scheme that integrates multiple constraints based on the candidate observation schemes of all observation tasks through a collaborative decision model. The collaborative decision model takes maximizing the overall task return as the objective function, and the objective function integrates the quality score of each task and the resource conflict penalty term between tasks. The results generation module is used to generate a sequence of mission instructions to drive the satellite.

9. The satellite mission planning system based on multiple constraints as described in claim 8, characterized in that, The multi-constraint analysis module also includes a visibility analysis unit and a meteorological data integration unit; The visibility analysis unit is used to perform orbit visibility analysis on each observation task based on the current satellite orbit data and the target area corresponding to the observation task. The meteorological data integration unit is used to acquire real-time meteorological data that matches the transit time window in time and space, based on a multi-source meteorological data acquisition mechanism.

10. The satellite mission planning system based on multiple constraints as described in claim 8, characterized in that, The satellite mission planning system based on multiple constraints also includes a visualization module for three-dimensional visualization demonstration and verification of the collaborative planning scheme. The visualization module includes: The payload simulation unit is used to dynamically calculate and generate a push-broom strip model that fits the Earth's surface based on satellite payload parameters, real-time orbit data, and observation geometric parameters in the collaborative planning scheme. The synchronization control unit is used to keep the simulation display of the push-broom strip model synchronized with the planned observation period, and to keep the attitude of the observation cone synchronized with the planned lateral tilt angle. The interactive demonstration unit is used to display the push-broom strip model and observation cone in a 3D scene in real time, and provides an interactive interface for users to compare and verify multiple schemes.

Images