An Optical Imaging Satellite Mission Scheduling Method Based on Hybrid Heuristic Memetic Evolution Algorithm

Abstract

The present invention provides an optical imaging satellite mission scheduling method based on a hybrid heuristic memetic evolution algorithm, which solves the problem of how to accurately and efficiently acquire satellite images. It belongs to the field of engineering. Based on the constructed satellite dataset, task dataset, satellite dynamics model, task decision model and constraint conditions, a satellite scheduling decision model is constructed with the maximization of the satellite's task revenue value as the objective function. A hybrid heuristic memetic evolution algorithm is constructed: an initial solution set of the observation path is obtained through clustering processing and a constraint checking model; the total revenue value of the initial solution is obtained by inputting the initial solution into the satellite scheduling decision model; a neighborhood search algorithm is used to obtain the feasible solution and the total revenue value of the initial solution, and the quality of the feasible solution is obtained from the difference in the total revenue values; combined with a simulated annealing algorithm and an ant colony evolution algorithm for search, after meeting the preset stop condition, an optimal optical imaging satellite mission scheduling scheme is obtained; the present invention realizes the accurate and efficient acquisition of satellite images.

Description

Technical Field

[0001] The present invention relates to the field of engineering technology, and particularly to an optical imaging satellite mission scheduling method based on a hybrid heuristic memetic evolution algorithm. Background Art

[0002] In modern society, the applications of satellite imaging have been widely involved in multiple fields such as agriculture, urban planning, natural resource management, disaster relief, etc. The efficiency and accuracy of satellite image acquisition tasks directly affect the quality and effect of satellite imaging in these applications. By optimizing the scheduling scheme of satellite image acquisition tasks, the utilization efficiency of satellite resources can be maximized, providing more accurate and detailed image data, thereby providing reliable technical support for the applications in related fields.

[0003] In the current technical background of optical imaging satellite scheduling, for the research on imaging satellite mission planning at home and abroad, the main solution strategies can be summarized into four categories: exact solution, heuristic strategy, meta-heuristic strategy, and machine learning algorithm; among them,

[0004] (1) Exact solution

[0005] The exact solution method relies on the construction of a mathematical model and the application of an optimization algorithm to pursue the best solution for satellite scheduling. This type of method can comprehensively consider multiple constraint conditions and seek the optimal solution in large-scale observation tasks. For example, strategies such as linear programming, integer programming, and branch and bound have all shown their effectiveness in the optical imaging satellite scheduling problem.

[0006] (2) Heuristic strategy

[0007] The heuristic algorithm, as a search strategy based on human experience and intuition, simulates human heuristic thinking to guide the search process. By means of iterative search, it provides approximate optimal solutions for those large-scale and high-complexity problems, especially in cases where traditional exact calculation methods are ineffective. Although the heuristic algorithm can find approximate solutions with relatively high quality, such solutions are not always the optimal solutions to the problem, and each heuristic algorithm is usually designed for specific problems and lacks a general algorithm structure.

[0008] (3) Meta-heuristic strategy

[0009] The meta-heuristic algorithm is an integration and improvement of the heuristic algorithm. It aims to improve the solution efficiency by combining the advantages of multiple algorithms, achieve a comprehensive search of the solution space, and balance in terms of time and computing resources. Due to its characteristics such as not requiring a specific problem form and strong robustness, the meta-heuristic algorithm has been widely used in the solution of multi-satellite autonomous mission planning problems.

[0010] (4) Machine learning algorithm

[0011] In recent years, the progress of machine learning technology has provided a new perspective for addressing the imaging satellite scheduling problem. By deeply analyzing historical data and learning from it, machine learning technology can build prediction and optimization models, thereby significantly improving the accuracy and efficiency of scheduling.

[0012] Although the above solutions can improve the accuracy and efficiency of optical imaging satellite mission scheduling in some aspects, the problem of reasonably arranging satellite observation tasks within a certain time to maximize the coverage of all ground targets remains unsolved.

[0013] Considering that this scheduling problem involves many complex factors, such as target location, satellite orbit, observation time window, observation target priority, etc. The combination of these factors will result in a very large state space and high computational complexity, making it difficult to solve through conventional dynamic programming algorithms. And reinforcement learning algorithms need to determine the optimal strategy through a large number of trial and errors. In this problem, only a single satellite agent is considered, without a large number of decision feedback requirements. Therefore, using reinforcement learning algorithms will increase the complexity of problem-solving. In contrast, heuristic algorithms can generate feasible approximate solutions in a relatively short time, guiding the search direction by designing appropriate heuristic functions to achieve a better solution effect, and having good scalability and applicability.

[0014] In the research of combinatorial optimization problems, designing efficient optimization algorithms has always been a core issue. Especially for the satellite scheduling problem focused on in this study, the following aspects are particularly crucial:

[0015] 1. Local optimization ability: In the context of complex constraints, the solution space of the satellite task scheduling problem is particularly narrow, so the importance of local optimization ability is particularly emphasized. This ability is mainly reflected in the algorithm's accurate search and positioning of the optimal solution within a specific area. Algorithms such as hill climbing algorithms, tabu search, and simulated annealing algorithms perform well in this regard. However, these algorithms are prone to falling into local optimal solutions, that is, the so-called "premature" problem.

[0016] 2. Global optimization ability: The exploration ability or global optimization ability of an algorithm refers to the ability to search and optimize in a vast solution space. Avoiding falling into local optima is crucial for the algorithm. In recent years, large neighborhood search algorithms have been widely studied and applied to improve the global optimization ability. Evolutionary algorithms such as genetic algorithms and particle swarm algorithms show strong global optimization ability due to their population evolution and diversity characteristics. However, these algorithms also face problems such as slow early convergence speed, high time complexity, and reduced search efficiency under complex constraints.

[0017] 3. Constraint Complexity: In the present invention, the satellite needs to satisfy a large number of constraint conditions in scheduling, such as time windows, priorities, dependencies, etc. To avoid time conflicts or contradictions and to allocate and utilize resources reasonably, these constraint conditions need to be checked and verified at any time to ensure the rationality and feasibility of the mission plan.

[0018] Regarding the above aspects of local optimization ability, global optimization ability, and constraint complexity, the present invention proposes an optical imaging satellite mission scheduling method based on a hybrid heuristic memetic evolution algorithm, providing an efficient solution means for the satellite conventional mission scheduling problem. Summary of the Invention

[0019] Aiming at the technical problem of how to reasonably schedule an optical imaging satellite during the mission time, arrange corresponding observation tasks, so as to cover all ground target areas to be observed to the greatest extent and accurately and efficiently obtain satellite images, a method for scheduling an optical imaging satellite mission based on a hybrid heuristic memetic evolution algorithm is provided. First, the K-means algorithm is used to cluster multiple ground targets to reduce the solution space complexity of the subsequent algorithm. Secondly, by improving the simulated annealing and ant colony algorithms, the advantages of the two heuristic algorithms are integrated to solve the local optimal solution and the global optimal solution. Through the memetic evolution method, not only the convergence time of the iteration is shortened but also the search ability of the agent is improved, and the quality of the feasible solution is enhanced. During the mission time, the observation tasks of the optical imaging satellite are reasonably arranged to cover all ground targets to the greatest extent, improving the efficiency and accuracy of the satellite image acquisition task. According to the experimental simulation results, the model has good robustness and scalability.

[0020] To achieve the above object, the technical solution of the present invention is as follows:

[0021] An optical imaging satellite mission scheduling method based on a hybrid heuristic memetic evolution algorithm, comprising the following steps:

[0022] Step 1: Based on the optical imaging satellite scheduling mission, construct a satellite data set and a mission data set corresponding to the mission;

[0023] Step 2: According to the longitude and latitude sampling intervals of the satellites in the satellite data set, split the ground target areas to be observed in the mission data set into corresponding numbers of observation sub-areas, and the satellites in the satellite data set observe the observation sub-areas to form an observation path; based on the observation path and the position coordinates of the ground target areas to be observed in three-dimensional space, construct a satellite dynamics model;

[0024] Step 3: Based on the mapping relationship of whether a task is executed by a satellite in the satellite dynamics model, construct a task decision model for the satellite; obtain the constraint conditions for satellite imaging when the satellite executes a task based on the task decision model of the satellite; take maximizing the task benefit value completed by the satellite as the objective function, and construct a satellite scheduling decision model through the constraint conditions;

[0025] Step 4: Use the observation range during a single scan of the satellite in the satellite dataset as the clustering radius, and perform clustering processing on the ground target areas to be observed in the task dataset through the K-means algorithm;

[0026] Step 5: Consist of a constraint check model based on the timeline advancement mechanism from the constraint check conditions in the optical imaging satellite scheduling task; the satellite in the satellite dataset simultaneously observes the ground target areas to be observed in the task dataset after clustering processing, forms a solution set of observation paths, and inputs the solution set of observation paths into the constraint check model in the order of execution time to obtain an initial solution set of observation paths that meets the corresponding constraint check conditions;

[0027] Step 6: Input the initial solution in the initial solution set of observation paths into the satellite scheduling decision model to obtain the total benefit value of the initial solution; use the neighborhood search algorithm to obtain a feasible solution of the initial solution, input the feasible solution into the satellite scheduling decision model to obtain the total benefit value of the feasible solution, and obtain the quality of the feasible solution from the difference between the total benefit values of the initial solution and the feasible solution; if the quality of the feasible solution is less than or equal to 0, output the feasible solution as the optimal feasible solution; if the quality of the feasible solution is greater than 0, obtain the optimal feasible solution through the simulated annealing algorithm, and output all the optimal feasible solutions to obtain a local optimal solution set;

[0028] Step 7: Input the local optimal solution set into the ant colony evolution algorithm for search to obtain the global optimal solution;

[0029] Step 8: Iterate Steps 6 to 7, and after meeting the preset stop condition, obtain the optimal optical imaging satellite task scheduling plan.

[0030] Beneficial effects:

[0031] 1. The present invention solves the problem of satellite scheduling for multiple tasks. By utilizing the local optimization ability of the simulated annealing algorithm and combining the global optimization ability of the ant colony evolution algorithm, it can quickly optimize the scheduling of optical imaging satellite tasks, greatly saving the convergence time of iteration during satellite task planning, improving the satellite task search ability, and enhancing the quality of the optimal feasible solution.

[0032] 2. The constructed satellite dynamics model can achieve large-angle imaging of ground targets by the satellite, improving the observation efficiency.

[0033] 3. The K-means algorithm is used to cluster the ground target areas to be observed within the observation range during a single satellite scan, reducing the complexity of the solution space of the satellite imaging task and improving the imaging efficiency.

[0034] 4. The technical solution disclosed by the present invention has good practicality and applicability. The hybrid heuristic memetic evolution algorithm is used to implement the task scheduling of optical imaging satellites, obtaining specific task planning results. And an evaluation algorithm for the performance of the total revenue value combined with the total number of executed tasks and the average revenue value is designed. Compared with the existing scheduling technologies, the feasibility and effectiveness of the designed hybrid heuristic memetic evolution algorithm are verified, providing a decision-making basis and application support for subsequent task scheduling. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. In the drawings:

[0036] Figure 1 It is a schematic diagram of cooperative observation of multiple agile imaging satellites.

[0037] Figure 2 It is a schematic diagram of cooperative observation of multiple agile imaging satellites.

[0038] Figure 3 It is a schematic diagram of multi-mode imaging of a satellite.

[0039] Figure 4 It is a schematic diagram of three-dimensional coordinates.

[0040] Figure 5 It is a schematic diagram of position update.

[0041] Figure 6 It is the total revenue of the FCFS algorithm.

[0042] Figure 7 It is the total revenue of the asynchronous A2C algorithm.

[0043] Figure 8 It is the total revenue of the GA algorithm.

[0044] Figure 9 It is the total revenue of the hybrid heuristic memetic evolution algorithm.

[0045] Figure 10 It is the amount of tasks executed by the FCFS algorithm for the total revenue.

[0046] Figure 11 It is the amount of tasks executed by the asynchronous A2C algorithm.

[0047] Figure 12 It is the amount of tasks executed by the GA algorithm.

[0048] Figure 13 It is the amount of tasks executed by the hybrid heuristic memetic evolutionary algorithm.

[0049] Figure 14 It is the average revenue value of the FCFS algorithm.

[0050] Figure 15 It is the average revenue value of the asynchronous A2C algorithm.

[0051] Figure 16 It is the average revenue value of the GA algorithm.

[0052] Figure 17 It is the average revenue value of the hybrid heuristic memetic evolutionary algorithm. Specific implementation manners

[0053] It should be noted first that the embodiments of the present invention only disclose the preferred implementation manners and should not be construed as limitations on the implementation of the present invention. The protection scope of the present invention still depends on the content disclosed in the claims.

[0054] The present invention proposes an optical imaging satellite task scheduling method based on a hybrid heuristic memetic evolutionary algorithm, including the following steps:

[0055] Step 1: Based on the optical imaging satellite scheduling tasks, construct a satellite data set and a task data set corresponding to the tasks;

[0056] Step 2: According to the latitude and longitude sampling intervals of the satellites in the satellite data set, split the ground target area to be observed in the task data set into corresponding numbers of observation sub-areas. The satellites in the satellite data set observe the observation sub-areas to form observation paths; based on the position coordinates of the observation paths and the ground target area to be observed in three-dimensional space, construct a satellite dynamics model;

[0057] Step 3: Based on the mapping relationship of whether the tasks are executed by the satellites in the satellite dynamics model, construct a task decision model for the satellites; obtain the constraint conditions for satellite imaging when the satellites execute tasks based on the task decision model of the satellites; take maximizing the satellite task revenue value as the objective function, and construct a satellite scheduling decision model through the constraint conditions;

[0058] Step 4: Use the observation range of the satellites in the satellite data set during one scan as the clustering radius, and perform clustering processing on the ground target area to be observed in the task data set through the K-means algorithm;

[0059] Step 5: Based on the constraint check conditions in the optical imaging satellite scheduling task, a constraint check model based on the timeline advancement mechanism is formed; satellites in the satellite dataset simultaneously observe the ground target areas to be observed in the task dataset after clustering processing to form an observation path solution set, and the observation path solution set is input into the constraint check model in the order of execution time to obtain an initial observation path solution set that meets the corresponding constraint check conditions;

[0060] Step 6: Input the initial solution in the initial observation path solution set into the satellite scheduling decision model to obtain the total revenue value of the initial solution; use the Neighborhood Search Algorithm to obtain a feasible solution of the initial solution, input the feasible solution into the satellite scheduling decision model to obtain the total revenue value of the feasible solution, and obtain the quality of the feasible solution from the difference between the total revenue values of the initial solution and the feasible solution; if the quality of the feasible solution is less than or equal to 0, output the feasible solution as the optimal feasible solution; if the quality of the feasible solution is greater than 0, obtain the optimal feasible solution through the simulated annealing algorithm, and output all the optimal feasible solutions to obtain a local optimal solution set;

[0061] Step 7: Input the local optimal solution set into the ant colony evolution algorithm for search to obtain the global optimal solution;

[0062] Step 8: Iterate Steps 6 to 7, and after meeting the preset stop condition, obtain the optimal optical imaging satellite task scheduling plan.

[0063] In Step 1, the constructed satellite dataset and task dataset are respectively:

[0064]

[0065] Among them, X is the satellite dataset, x i is the i-th satellite, i is the satellite number, and M is the total number of satellites; Y is the task dataset, y j is the j-th task, j is the task number; N is the total number of tasks to be planned.

[0066] When the ground station receives task requirements from different user terminals, it needs to conduct a preliminary screening of the tasks, that is, clean the geospatial data of the observation target area to judge the feasibility of the tasks. Secondly, task attributes are an important basis for satellite task planning, including task type, task requirements, task timeliness, task priority, etc. Different types of tasks have different characteristics and requirements, so it is necessary to evaluate the importance of tasks according to task attributes to obtain the revenue values for completing each task.

[0067] The tasks of the optical imaging satellite are transmitted by the ground station to the satellite. Reasonable mission planning requires the ground station to process the mapping relationship between a large number of satellites and a large number of tasks. Therefore, in step one, a satellite dataset and a task dataset need to be established.

[0068] In step two, the method for constructing the satellite dynamics model includes:

[0069] Obtain the longitude and latitude of the ground target area in the task dataset; construct a rectangle based on the maximum longitude and maximum latitude; use the longitude and latitude sampling intervals of the satellites in the satellite dataset to split the rectangle into corresponding numbers of observation sub-areas; the satellites in the satellite dataset observe the split observation sub-areas to form an observation path; construct the satellite dynamics model from the observation path and the position coordinates of the ground target area in three-dimensional space.

[0070] To achieve full coverage of a relatively large ground target area, an ordinary optical imaging satellite usually cannot complete this task during a single pass. Therefore, to facilitate the observation and splitting of the ground target area, in step two, the ground target area is divided into a series of cells according to longitude and latitude to ensure that the size of each cell is small enough to be completely covered in a single imaging shot. As Figure 1 and Figure 2 shown, in the embodiments of the present invention, the planned satellite is preferably an agile imaging satellite. An agile imaging satellite (Agile Earth Observation Satellite, AEOS) is a satellite with high maneuverability and multi-functional observation modes. Compared with traditional Earth observation satellites, an agile imaging satellite has a larger ground observation range and stronger observation capabilities. Through attitude control systems such as flywheels and control moment gyroscopes, it can achieve swinging around the roll, pitch, and yaw axes, so that it can perform maneuvers and imaging simultaneously within a certain range, and the maneuvering speed is faster, the angle is larger, and the ability is stronger than that of traditional satellites. This enables the agile imaging satellite to cover a wider ground area in a shorter time and capture more Earth observation data. The agile imaging satellite demonstrates the high-maneuverability attitude adjustment characteristics by virtue of its unique pitch angle and side-sway angle adjustment capabilities. Therefore, it supports diverse imaging modes, including mosaic imaging of the ground target area, multi-angle stereo imaging, and push-broom imaging in the reverse direction. These imaging modes enable the agile imaging satellite to achieve one-time imaging observation of large-area targets, providing great convenience for practical applications. Specifically, as Figure 3 shown, the main decomposition process of the observation sub-areas is as follows: First, determine the maximum longitude and latitude of the ground target area; then, determine the formed rectangle through the four obtained extreme values, that is, divide the rectangle into multiple cells using the longitude and latitude sampling intervals, that is, Figure 3 the red rectangular frame in; finally, the agile imaging satellite observes according to the split observation sub-areas.Figure 3 The dashed line indicated by the arrow is the observation path during satellite observation. Through these imaging modes, agile imaging satellites can obtain rich surface information, provide important support for scientific research and decision-making, and meet the needs of different application fields, such as environmental monitoring, agriculture, urban planning, etc.

[0071] In addition, agile imaging satellites also have a wider observation time window, that is, they can observe within a longer time period. This means that it can better adapt to different observation needs and time constraints, and provide continuous and persistent observation data.

[0072] To more specifically analyze the spatial orbital positions of the satellite and the ground observation targets, a three-dimensional coordinate schematic diagram is drawn as Figure 4 shown. Figure 4 The image in shows the positions of the satellite and the ground target points in three-dimensional space, which are determined by specific x, y, and z coordinate values. By comparing the coordinate values between the satellite and the ground target points, the relative position relationship between them, the coverage area of satellite observation, the spatial distribution between targets, etc. can be evaluated. By calculating the length of the spatial distance, parameters such as the time and energy consumption required for the satellite to move from one observation point to another can be evaluated, providing data support for subsequent mission planning and scheduling.

[0073] In step three mentioned above, the task decision model of the satellite is constructed as:

[0074]

[0075] Among them, is the decision variable of the satellite task. When , then task y i is observed by satellite x i . When , it means to discard this task y i .

[0076] In step three, aiming at the decision relationship of the satellite task scheduling problem, based on the mapping relationship between the task data set and the satellite data set, a 0-1 mixed integer programming decision variable is used to describe the decision relationship existing between the "satellite data set - task data set". The constructed decision model describes the general decision model of the satellite task scheduling problem and serves as the basis for subsequent modeling and algorithm design of the satellite's constraint conditions, objective functions, etc.

[0077] In step three mentioned above, the constraint conditions for satellite imaging include: time window constraint, storage constraint, power constraint, observation angle constraint, task execution uniqueness constraint, and / or avoidance of task conflict constraint; specifically including:

[0078] Time window constraint:

[0079] Among them, is the observable time window for each task to exist, is the execution time of the task; the execution time of the task must fall within the observable time window to be successfully observed.

[0080] Storage constraint:

[0081] Among them, is the mission decision variable of the satellite; Storage(y i |x i ) is the storage space consumed by the satellite when executing the task; Storage total (x i ) is the total storage space of each satellite; N is the total number of tasks to be planned; in the case where the satellite does not download data, the total storage space consumed by the satellite shall not exceed the total storage space provided by the satellite.

[0082] Power constraint:

[0083] Among them, Elec(y i |x i ) is the power consumed by the satellite when executing the task, Elec trf (y i |x i ) is the power consumed by the satellite for attitude adjustment when executing the task; Elec total (x i ) is the total power of each satellite; N is the total number of tasks to be planned; when the satellite executes the task, the power consumed by the satellite for attitude adjustment is positively correlated with the change of the roll angle. In the case where the satellite does not charge in orbit, the total power consumed by the satellite shall not exceed the total power that the satellite itself can provide.

[0084] Observation angle constraint: -90° ≤ θ ≤ 90°;

[0085] Among them, θ is the observation angle of the satellite, that is, the pitch angle of the satellite, which is the angle between the attitude orientation of the satellite and the earth's surface, describing the inclination degree of the satellite relative to the vertical direction of the earth. Taking the orbital plane passing through the earth's center as the 0° angle, the positive value of θ indicates that the satellite is facing above the earth's surface, and the negative value indicates facing below the earth's surface; if the satellite is operating normally in orbit, when it performs imaging observation on ground targets, the above formula is satisfied.

[0086] Task execution uniqueness constraint:

[0087] Among them, is the coefficient factor indicating whether the ground target is observed;

[0088] Given the limitations of the satellite's maneuverability, the execution time and the time cost required for attitude adjustment must be comprehensively considered when planning tasks. To optimize the utilization efficiency of satellite resources, it is crucial to ensure the uniqueness of task execution. This means that each task can only be observed by the satellite once, and there is no repeated execution. This constraint is expressed by the above formula.

[0089] Avoid task conflict constraints:

[0090] Among them, is the start time when the second task y b is observed by satellite x i ; is the end time when the first task y a is observed by satellite x i ; is the interval time from the first task y a to the second task y b . a is the departure path node and b is the target path node. After the first task y a is observed, the satellite goes to observe the second task y b ; the end time of the first task observation, plus the conversion interval time, should be less than the start time of the second task observation.

[0091] When a certain satellite is performing tasks and facing different observation targets with overlapping time windows, when planning its tasks, it is necessary to ensure that the execution time windows of any two tasks do not overlap. Therefore, when the first task is executed first and then the second task, it is ensured that the start execution time of the second task is greater than the end time of the first task plus the satellite conversion interval time, that is, the above formula is satisfied.

[0092] When an optical imaging satellite is orbiting in space and observing a specified ground target area, there are restrictions in terms of time, space, etc. Therefore, when scheduling the tasks of an optical imaging satellite, the above several constraint conditions need to be considered comprehensively in various aspects. In addition, during the process of the satellite performing tasks, there are generally also constraint conditions such as communication signals and computing resources. Since the impact of such constraint factors on this research is relatively small, they are not considered.

[0093] In the above-mentioned step three, the constructed satellite scheduling decision model is:

[0094]

[0095] Among them, R(X) is the total revenue value of the satellite scheduling decision model, M is the total number of satellites, is the task decision variable of the satellite; f(y j ) is to execute the jth task yj The revenue value obtained by the satellite later, θ(y j |x i ) is the pitch angle of the i-th satellite x when the j-th task y j is executed, i and is the penalty factor. That is, when the pitch angle is 0°, the satellite imaging quality is the highest, the penalty factor is 0, and it has no impact on the revenue value. When the absolute value of the pitch angle increases, the penalty factor value also increases, so the revenue value of completing the task gradually decreases; when the pitch angle increases to 90°, the penalty factor value is 1, the revenue value of the satellite completing the task drops to 0, and the satellite imaging fails.

[0096] In the research on the task scheduling of optical imaging satellites and the actual application scenario, after each task is completed, there is a certain revenue value, and it is necessary to maximize the total revenue of completing the task. However, during the imaging process of the satellite, the angle size has a great impact on the imaging quality. Therefore, in order to more accurately measure the imaging effect of the satellite and its actual contribution degree to the task, the task revenue value is penalized to a certain extent according to the pitch angle of the satellite during imaging, and the impact of satellite imaging quality on the task revenue is directly quantified through the above formula.

[0097] Steps four to eight are the hybrid heuristic memetic evolution algorithm constructed by the present invention.

[0098] Since the satellite scheduling problem is a dynamic scheduling that changes with time, its solution space also changes with time. When designing a heuristic algorithm, it is necessary to consider solving problems applicable to dynamic problems. The present invention designs a hybrid heuristic memetic evolution algorithm based on improved simulated annealing and ant colony algorithms through steps four to eight.

[0099] In the said step four, the K-means algorithm includes: initializing the centroids, assigning data points to the nearest centroids, and updating the centroid positions.

[0100]

[0101] Among them, J is the objective function, that is, minimizing the distance from the data points within the cluster to the centroids, n is the number of clusters, k represents the k-th cluster, X l represents the data point, C k is the centroid of the k-th cluster, ||X l -C k || 2 represents the square of the Euclidean distance between the data point X l and the cluster center C k ; among them, the calculation formula of the Euclidean distance is:

[0102]

[0103] Among them, dis(Xl , C k ) is the Euclidean distance, X l and C k are two m-dimensional vectors, where m is the number of dimensions of the data feature space, and X lt and C kt are respectively the components of the t-th dimension of X l and C k ;

[0104] The calculation formula for the cluster center C k is as follows:

[0105]

[0106] where C k is the center of the k-th cluster, S k is the number of samples belonging to the k-th cluster, is the sum of the values of the l-th sample data point on the corresponding k-th feature; Through continuous iteration until the center of the cluster no longer changes significantly, the cluster center C k is obtained.

[0107] The K-means algorithm, also known as the K-mean algorithm, is used to partition the clustering segmentation model. Clustering is an unsupervised learning. For this study, when the satellite receives the task of observing ground targets that need to be observed, in order to reduce the complexity of the solution space, this algorithm can be used to group the ground targets to be observed that are relatively close in distance into one cluster, and the cluster center value is calculated by the mean of all points in the cluster. The meaning of the cluster is: the set of data points in all active regions, and the objects in each component cluster are similar. Centroid: In the component cluster, the center of all points (obtained by the mean of all points in the cluster). The clustering radius is determined by the observation range during a single scan of the satellite. The K-means algorithm process is shown in Table 1.

[0108] Table 1

[0109]

[0110] When generating a large number of tasks, after being processed by the K-means algorithm, similar tasks can be observed simultaneously.

[0111] There are complex constraints in the task scheduling process of imaging satellites, and these constraints are one of the main factors contributing to the complexity of the problem. In order to decouple the constraint checking from the scheduling algorithm, the present invention designs a constraint checking algorithm based on the timeline advancement mechanism. This algorithm, as an independent functional module, is used to check whether the scheduling scheme generated during the final problem-solving solution or the intermediate attempt process meets the constraints.

[0112] In the fifth step, the constraint check conditions in the optical imaging satellite scheduling task in the constraint check model based on the timeline advancement mechanism include: the check of cumulative constraints, task attribute constraints, and task correlation constraints; among them,

[0113] 1. The cumulative constraint is that within a planning period, the target statistic is within a preset value;

[0114] In this embodiment, a one-dimensional array Stat k is used to record the cumulative constraint variables at any time in the solution to be checked. Where k represents the kth cumulative constraint, and k 1 represents the total number of items of the cumulative constraint conditions. These physical quantities are non-decreasing functions of time, that is, when t 1 < t 2 , Stat k (t 1 ) ≤ Stat k (t 2 ) always holds. According to the monotonicity of the cumulative constraint, it can be obtained that in the sequence of task scheduling solutions, if there is a subsequence that constitutes a scheduling solution that does not satisfy the cumulative constraint k, the original task scheduling solution also does not satisfy the same cumulative constraint.

[0115] Therefore, the cumulative constraint check idea based on the timeline advancement mechanism is designed as follows: Arrange the tasks in the solution to be checked in the order of execution time, and then add tasks one by one and calculate the corresponding cumulative constraint variables. If the variable value exceeds the required value of the constraint condition after adding a task, the overall solution does not satisfy the cumulative constraint.

[0116] 2. The task attribute constraint is to merge the constraint conditions corresponding to the task attributes through preprocessing;

[0117] For example: The imaging quality constraint and the visibility constraint are merged into a unified constraint condition for evaluation. In addition, constraint conditions such as "imaging tasks cannot be executed in the earth's shadow area" and "imaging tasks cannot be executed in areas with a latitude higher than 60 degrees" can also be merged using similar methods. After merging some constraint conditions, all task attribute constraints can be checked one by one to complete the constraint check process of the task attribute constraints. The time complexity of this process is O(n).

[0118] 3. The task correlation constraint is to merge the constraint conditions corresponding to the task correlation based on the logical determination between consecutive tasks.

[0119] To ensure that there are no conflicts between tasks in the final task planning scheme, it is necessary to consider task correlation constraints. These constraints are mainly based on the logical determination between two consecutive tasks, such as the determination of the minimum time interval between different types of tasks. Usually, relevant operations are required, such as using parameters like the attitude conversion time calculation function, the minimum pre-order time and the minimum post-order time of tasks to jointly determine. Some correlation constraints can also be considered together, and only the values that meet multiple constraint conditions need to be set. After combining some constraint conditions, we can check the task correlation constraints one by one to complete the constraint check process of task attribute constraints. The time complexity of this process is O(n).

[0120] In step six, the core idea of the Simulated Annealing (SA) algorithm comes from a process in the metallurgy field, that is, heating the metal to a high temperature state and then slowly cooling it to reach a stable state. This process is like the evolution of the system energy over time, and the simulated annealing algorithm exactly uses this process to find the optimal solution. The simulated annealing algorithm is as follows:

[0121] Input the initial solution as the starting point and determine the neighborhood of the initial solution, that is, all possible change ways of the initial solution. This neighborhood can be simple adjacent points or more complex, including combinations of changes of multiple variables, etc.

[0122] Input the initial solution into the satellite scheduling decision model constructed in step three to obtain the benefit value C of the initial solution 0 . Use the neighborhood search algorithm to improve the initial solution to obtain a feasible solution, and calculate the benefit value C of the feasible solution 1 , evaluate the quality of the feasible solution: ΔC = C 0 - C 1 ; if ΔC ≤ 0, the feasible solution is better than the initial solution, and the feasible solution is not dominated by any solution in the initial solution set, then directly accept the feasible solution and output the feasible solution as the optimal feasible solution; if ΔC > 0, the feasible solution is worse than the initial solution, then judge the probability of accepting the feasible solution through the simulated annealing algorithm:

[0123]

[0124] where P(S 1 ) is the probability that the feasible solution S 1 can be accepted when the quality ΔC > 0, T c is the current temperature, and the temperature T c will decay every time it iterates; for example: T n = T c ·ε; in the formula, T nThe temperature for the next iteration, ε is the control parameter for adjusting the temperature decrease rate; To further improve the quality of the feasible solution, a tempering mechanism can be carried out. After the feasible solution is accepted, a certain degree of temperature increase is performed to extend the search time. After the feasible solution is accepted, it is output as the optimal feasible solution, and the search time is extended by increasing the temperature through the search model, where the search model is:

[0125] T n =(T p -T c )·ω+T c ;

[0126] Among them, T p is the temperature of the previous feasible solution, and ω is the control coefficient;

[0127] When the temperature reaches the preset termination condition, the simulated annealing algorithm stops searching and outputs all the optimal feasible solutions to obtain the local optimal solution set.

[0128] The ant colony evolution algorithm is an algorithm for searching the global optimal solution. In this algorithm, ants represent satellites, and tasks represent the path nodes that the ant colony needs to visit; In step 7, the method for obtaining the global optimal solution by inputting the local optimal solution set into the ant colony evolution algorithm includes:

[0129] 1), Based on the local optimal solution, construct the path set L and the speed set V;

[0130]

[0131] Among them, (L 1 ,Y 1 ) is the starting point of the observation path when the satellite executes the task; (L n ,Y n ) is the end point of the observation path when the satellite executes the task; is the initial velocity vector of the observation path when the satellite executes the task; is the final velocity vector of the observation path when the satellite executes the task; Connecting each path node n in the observation path when the satellite executes the task in sequence is a time series of the task execution to be solved;

[0132] 2), The ant position iteration process is as Figure 5 shown, Gbest is the optimal position experienced by all ants; Pbest is the optimal position experienced by this ant; V is the current velocity of this ant. It can be seen from Figure 5 that after iteration, the new position of the ant is closer to the global optimal position. At the beginning of the iteration, the ant colony needs to construct an initial solution, that is, based on the mapping relationship between whether the tasks in the path set and the speed set are executed by the satellite, construct the path decision model of the satellite;

[0133]

[0134] Among them, is the path decision variable of the satellite; according to the path decision variable of the satellite select whether to start from the starting path node a and visit the target path node b at time t, is the set of b that the ant colony starting from a can choose at time t, and σ ab (t) is the heuristic function factor from a to b at time t:

[0135]

[0136] In the formula, c ab (t) is the amount of heuristic information remaining on the ab observation path of the satellite at time t, and τ ab (t) is the amount of pheromone remaining on the ab observation path of the satellite at time t, α is the control parameter of the heuristic information, and β is the control parameter of the pheromone, reflecting the importance degrees of the heuristic information and the pheromone information; at the initial stage of the algorithm iteration, the pheromone τ ab (0) is set to G τ0 which is the starting value of the global pheromone matrix;

[0137] 3) When the satellite has visited all the feasible target path nodes, it ends; if there are still unvisited target path nodes, the satellite continues to complete the construction of the observation path; after all the target path nodes have been visited, a new solution is obtained, and the new solution includes a feasible solution and an infeasible solution. Among them, the feasible solution is: the solution where all the target path nodes are visited and meet the constraint check conditions; the infeasible solution is: the solution where some target path nodes are not visited or do not meet the constraint check conditions.

[0138] If the new solution is a feasible solution, the global pheromone matrix is updated through the first update model; the first update model is as follows:

[0139] Gτ ab (t + 1) = (1 - ρ)Gτ ab (t) + ρGτ 0 ;

[0140] In the formula, Gτ ab (t + 1) is the amount of global pheromone on the ab observation path corresponding to the target path node after update at time t + 1, Gτ ab (t) is the amount of global pheromone on the ab observation path corresponding to the target path node at time t, ρ is the evaporation coefficient of the pheromone, Gτ 0 is the amount of pheromone remaining on the global pheromone by the satellite, and Gτ 0 is equal to Gτ ab (0) which is the starting value of the global pheromone matrix;

[0141] If the new solution is an infeasible solution, the global pheromone matrix is updated by the second update model; the second update model is as follows:

[0142] τ ab (t + 1)=(1 - ρ)τ ab (t)-ρτ 0 ;

[0143] Where τ ab (t + 1) is the amount of pheromone remaining on the ab observation path by the satellite at time t + 1, τ ab (t) is the amount of pheromone remaining on the ab observation path by the satellite at time t, and ρ is the evaporation coefficient of pheromone.

[0144] When the ab observation path corresponding to the target path node at time t is in the infeasible solution, the amount of pheromone on the ab observation path in the pheromone matrix is reduced by the second update model, so that the probability of the observation path corresponding to the target path node being selected is reduced in subsequent iterations;

[0145] When the preset termination condition is reached, the iteration ends, and all satellites in the satellite dataset will gather on the observation path with the highest pheromone concentration, thereby completing the selection of the optimal observation path and obtaining the global optimal solution.

[0146] In order to deeply explore the actual application effect of the hybrid heuristic memetic evolution algorithm in the single satellite scheduling problem of optical imaging, the present invention designs a series of simulation experiments. First, the specific settings and details of the experimental scenario are elaborated in detail. Subsequently, the focus is on the operation results of the algorithm in this specific scenario, and in-depth analysis is carried out to evaluate its solution accuracy and operation efficiency. On this basis, practically significant conclusions are drawn.

[0147] Analysis of simulation experiments:

[0148] 1. Experimental scenario design

[0149] ① Task parameter design

[0150] The set of task information can be described by Y s which integrates the visible time window, imaging duration, and corresponding value benefits of the task. In order to simulate the scenario in the real environment as much as possible, the set input data focuses on real imaging requirements. The experimental process starts with receiving these imaging requirements, and then through the task preprocessing process, specific imaging tasks are gradually refined and generated, which in turn serve as the core input data for the imaging satellite scheduling problem.

[0151] ② Task geographical location design

[0152] When generating the geographical locations of tasks, a part of the observation targets focus on the "domestic region", meaning that all tasks will be randomly scattered within the territory of our country; the other is the "US region", representing that tasks are scattered within the US region. Specifically, for tasks in the "domestic region", their longitude range is limited to 73° east longitude to 135° east longitude, and the latitude range is between 3° north latitude and 54° north latitude. The geographical locations of these tasks will be randomly generated according to the principle of uniform distribution. For tasks in the "US region", their longitude will cover the entire longitude range from 70° west longitude to 130° west longitude. Considering the coverage range of the satellite orbit, the latitude is distributed between 60° south latitude and 60° north latitude, and is also randomly generated in a uniform distribution manner.

[0153] To comprehensively evaluate the performance of the algorithm under different task scales and geographical distributions, a task set covering 30 groups with different characteristics was designed. The detailed scales and distributions of each group of tasks are listed in detail. Among them, for the task set in the "domestic region", the geographical locations of the experimental group tasks are relatively concentrated, and their visible time windows also show a certain degree of aggregation, which helps to simulate the actual scenario where the geographical locations of tasks are highly concentrated. For the task set in the "US region", its visible time windows show a more uniform distribution, simulating the extensiveness and diversity of the distribution of external observation tasks.

[0154] The change in task scale directly reflects the overall density of tasks, which enables the experiment to comprehensively test the effectiveness of the algorithm under different task distributions and scales. The original data of the experiment, such as information on imaging quality standards and imaging duration, are all set based on the specific needs of users for satellite images. In the simulation experiment, the method of randomly generating these parameters is adopted.

[0155] ③ Satellite orbit parameter design

[0156] The precise setting of satellite orbit parameters not only determines the flight path of the satellite but also is the key basis for calculating its ephemeris. Ephemeris, as a function describing the variation law of the satellite's position and its instantaneous velocity over time, represents the position of the satellite at any moment, as well as the direction and magnitude of the instantaneous velocity, and its importance is self-evident. Usually, ephemeris data is calculated by combining orbital elements and Kepler's laws. In this experiment, the corresponding ephemeris information is obtained based on the given satellite orbital elements.

[0157] In addition, the satellite used in this simulation experiment is a low Earth orbit circular satellite, and its period for orbiting the Earth once is about 90 minutes. The accuracy of this ephemeris information is crucial for many subsequent calculation processes, including but not limited to the determination of task time windows, the calculation of task pointing angles, and the estimation of attitude conversion times between tasks. Therefore, accurately obtaining the ephemeris information of the satellite is the basis for ensuring the efficient operation of the entire satellite system.

[0158] ④ Satellite Attitude Maneuverability Design

[0159] In this experiment, an agile imaging satellite was designed. Its imaging payload has the ability to adjust the pointing angle within a certain range according to the mission geographical coordinates and the satellite's real-time position. Such a design makes the scheduling of imaging tasks more flexible and variable. The satellite's attitude maneuverability, as the core element in the calculation of imaging task visibility, is the key to determining the task visibility time window.

[0160] Regarding the description of the satellite's attitude, the yaw angle refers to the angle formed by the line connecting the imaging payload and the sub-satellite point when the satellite rotates around the forward direction as the axis; while the pitch angle refers to the angle between the imaging payload and the sub-satellite point when the satellite rotates around the axis perpendicular to the forward direction and parallel to the ground. Further analysis reveals that the yaw angle of the satellite is positively correlated with the offset of the mission position relative to the sub-satellite line, and the pitch angle is positively correlated with the offset of the mission and the sub-satellite point in the satellite's forward direction. Given that the satellite in this experiment is not equipped with advanced functions such as active push-broom or imaging while in motion, the roll angle is set to 0 degrees.

[0161] ⑤ Satellite Capability Parameter Design

[0162] When discussing the scheduling problem of imaging satellites, it is necessary to fully consider their resource capability parameters. These resource capabilities cover key constraint conditions such as the cumulative imaging duration, the satellite's storage capacity (fixed storage), power reserve, attitude adjustment time, and imaging resolution (clarity).

[0163] After designing the satellite and the observation task data parameters, they are imported into the STK simulation software.

[0164] Using STK software simulation, the visibility time window of each imaging task can be accurately calculated, and the pointing angle at any moment within each time window can be determined. It should be noted that in the task scheduling problem of this study, to simplify the task assignment process, that is, each task corresponds to only one time window, which means that each task has only one imaging opportunity.

[0165] So far, the key input parameters required for the imaging satellite scheduling problem and their generation methods have been fully elaborated. Combining the constraint conditions and objective function of the problem, an effective solution can be obtained using the corresponding algorithm.

[0166] 2. Experimental Results and Analysis

[0167] This experiment tested the efficiency of the hybrid heuristic memetic evolution algorithm in solving the imaging satellite task scheduling problem. The FCFS algorithm, asynchronous A2C algorithm, and genetic algorithm were selected as comparison algorithms to comparatively analyze the performance of the proposed algorithm in task scheduling. The following three indicators were used in the experiment to evaluate the performance of the algorithm:

[0168] Total revenue value: The sum of the revenue values of all executed tasks, reflecting the algorithm's ability in terms of maximizing revenue.

[0169] Total number of executed tasks: The number of tasks that the algorithm can execute, reflecting the algorithm's ability in terms of handling the number of tasks.

[0170] Average revenue value: The total revenue value divided by the total number of executed tasks, reflecting the algorithm's ability in terms of average revenue.

[0171] To more intuitively display the performance of each algorithm, the experimental results are visualized as Figures 6 - 9 shown. From the experimental results, the total revenue values under different experimental conditions vary significantly. The following is a detailed analysis of the total revenue values of each experimental group: Among all experimental groups, the experimental group of the hybrid heuristic memetic evolutionary algorithm has the highest total revenue value. This indicates that this algorithm has high efficiency and effectiveness in optimizing task scheduling and can effectively maximize the revenue of tasks. FCFS experimental group: Compared with other experimental groups, the FCFS experimental group has the lowest total revenue value. This is because the FCFS algorithm only schedules tasks in the order of task arrival, without considering task priorities and revenue values, resulting in tasks with lower revenues occupying a large amount of execution resources. The total revenue value of the A3C experimental group is at a medium level. Perhaps because there are still certain limitations and room for optimization when the A3C algorithm adapts to the task scheduling problem. GA experimental group: The total revenue value of the GA experimental group is slightly lower than that of the hybrid heuristic memetic evolutionary experimental group but higher than that of the FCFS and A3C experimental groups. This shows that the genetic algorithm has a certain ability to solve the task scheduling problem, but it is still not as efficient as the hybrid heuristic memetic evolutionary algorithm.

[0172] The total number of executed tasks is as Figures 10 - 13 shown. As Figures 10 - 13 can be seen, there are also certain differences in the total number of executed tasks under different experimental conditions. The experimental group of the hybrid heuristic memetic evolutionary algorithm performs well in terms of the number of executed tasks, can handle more tasks and the scheduling curve is very stable, indicating that the algorithm has a good optimization strategy and a stable model, and can improve the execution efficiency while ensuring task quality. The FCFS experimental group executes the fewest tasks. The number of tasks executed by the A3C experimental group is initially similar to that of the FCFS experimental group and is superior to the FCFS experimental group after subsequent iterations. This indicates that it has a certain optimization ability. The GA experimental group performs well in terms of the number of executed tasks but is slightly lower than the HHMEA experimental group, indicating that the genetic algorithm has a certain ability in task scheduling but still needs further optimization to improve the execution efficiency.

[0173] The average revenue value is as Figures 14 - 17 shown, and the average revenue values under different experimental conditions are also important indicators for evaluating the algorithm performance. Figures 14 - 17The experimental results show that the four algorithms perform similarly in terms of the average revenue value.

[0174] Experimental conclusion:

[0175] Through this experiment, an in-depth analysis and comparison were conducted on the performance of the hybrid heuristic memetic evolution algorithm in the single-satellite task scheduling problem. The experimental results clearly show that the hybrid heuristic memetic evolution algorithm has achieved good results in three key indicators: the total revenue value, the total number of executed tasks, and the average revenue value. The hybrid heuristic memetic evolution algorithm combines the ideas of heuristic search and memetic evolution. This combination enables the algorithm to be more flexible and efficient during the task scheduling process. Heuristic search allows the algorithm to quickly find potential high-quality solutions based on the specific nature and constraints of the problem, while memetic evolution continuously optimizes the quality of the solutions by simulating natural selection and genetic mechanisms, thus realizing the intelligence and automation of the search process. Secondly, the hybrid heuristic memetic evolution algorithm realizes a comprehensive exploration and in-depth mining in the search space through the effective combination of local search and global search. Local search can find better solutions in the vicinity of the current solution, while global search can break out of the limitations of the current solution and explore a wider search space, thus avoiding being trapped in local optimal solutions. This combination not only improves the search efficiency of the algorithm but also ensures the search quality of the algorithm.

[0176] In the specific experimental results, by comparing with the FCFS algorithm, the A3C algorithm, and the GA algorithm, the hybrid heuristic memetic evolution algorithm has achieved significant advantages in the total revenue value, indicating that in the same task scheduling scenario, the designed hybrid heuristic memetic evolution algorithm can find and execute more high-revenue tasks, thus realizing the maximization of resource utilization. At the same time, in terms of the total number of executed tasks, the hybrid heuristic memetic evolution algorithm also performs excellently, demonstrating its high efficiency in processing a large number of tasks. In terms of the average revenue value, the average revenue values obtained by the hybrid heuristic memetic evolution algorithm and other algorithms are similar, indicating the overall performance stability of the algorithm.

[0177] Beneficial effects:

[0178] 1. The present invention solves the problem of satellite scheduling for multiple tasks. By utilizing the local optimization ability of the simulated annealing algorithm and combining it with the global optimization ability of the ant colony evolution algorithm, it can quickly optimize the scheduling of optical imaging satellite tasks, greatly saving the convergence time of iteration during satellite task planning, improving the satellite task search ability, and enhancing the quality of the optimal feasible solution.

[0179] 2. The constructed satellite dynamics model can achieve large-angle imaging of ground targets by the satellite, improving the observation efficiency.

[0180] 3. The K-means algorithm is used to cluster the ground target areas to be observed within the observation range during a single satellite scan, reducing the complexity of the solution space of the satellite imaging task and improving the imaging efficiency.

[0181] 4. The technical solution disclosed by the present invention has good practicality and applicability. The hybrid heuristic memetic evolution algorithm is used to implement the optical imaging satellite task scheduling, obtaining specific task planning results, and a total revenue value is designed. The performance of the algorithm is evaluated by combining the total number of executed tasks and the average revenue value. Compared with the existing scheduling technologies, the feasibility and effectiveness of the designed hybrid heuristic memetic evolution algorithm are verified, providing a decision-making basis and application support for subsequent task scheduling.

[0182] Finally, it should also be pointed out that any unit or individual using or implementing the technical solution of the present invention is an infringement of the present invention. Without the permission of the applicant, no unit or individual can implement this patent alone. And any unit or individual inspired by the present invention or implemented after simple adjustment should also be considered within the protection scope of this patent.

Claims

1. An optical imaging satellite task scheduling method based on a hybrid heuristic memetic evolutionary algorithm, characterized in that: The following steps are involved: Step 1: Based on the optical imaging satellite scheduling task, construct the satellite data set and task data set corresponding to the task; Step 2: According to the latitude and longitude sampling intervals of the satellites in the satellite data set, the ground target area to be observed in the mission data set is divided into a corresponding number of observation sub-areas, and the satellites in the satellite data set observe the observation sub-areas to form an observation path; The satellite dynamics model is constructed based on the observation path and the position coordinates of the ground target area to be observed in three-dimensional space; Step 3: Based on the mapping relationship between whether the task is executed by the satellite in the satellite dynamics model, a satellite task decision model is constructed; Based on the satellite mission decision model, the satellite imaging constraints are obtained when the satellite performs its mission. Taking maximizing the satellite mission completion benefit as the objective function, a satellite scheduling decision model is constructed through constraint conditions; Step 4: Use the observation range of the satellite in the satellite data set during one scan as the clustering radius, and use the K-means algorithm to cluster the ground target area to be observed in the mission data set; Step 5: The constraint check conditions in the optical imaging satellite scheduling task form a constraint check model based on the timeline advancement mechanism; the satellites in the satellite data set simultaneously observe the ground target area to be observed in the task data set after clustering processing to form an observation path solution set, and the observation path solution set is input into the constraint check model in the order of execution time to obtain the initial observation path solution set that meets the corresponding constraint check conditions; Step 6: Input the initial solution in the initial observation path solution set into the satellite scheduling decision model to obtain the total benefit value of the initial solution; use the neighborhood search algorithm to obtain the feasible solution of the initial solution, input the feasible solution into the satellite scheduling decision model to obtain the total benefit value of the feasible solution, and obtain the quality of the feasible solution by the difference between the total benefit values ​​of the initial solution and the feasible solution; if the quality of the feasible solution is less than or equal to 0, output the feasible solution as the optimal feasible solution; if the quality of the feasible solution is greater than 0, obtain the optimal feasible solution through the simulated annealing algorithm, and output all the optimal feasible solutions to obtain the local optimal solution set; Step 7: Input the local optimal solution set into the ant colony evolution algorithm to search for the global optimal solution; Step 8: Iterate steps 6 to 7, and after the preset stop condition is met, the optimal solution for optical imaging satellite mission scheduling is obtained.

2. The method according to claim 1, characterized in that In step 1, the satellite data set and mission data set constructed are: Among them, X is the satellite dataset, x i is the i-th satellite, i is the satellite number, M is the total number of satellites; Y is the mission data set, y j is the jth task, j is the task number; N is the total number of tasks to be planned.

3. The method according to claim 2, characterized in that In the step 2, the method for constructing the satellite dynamics model includes: Get the longitude and latitude of the ground target area in the mission dataset; Construct a rectangle based on the maximum longitude and latitude; The rectangle is split into a corresponding number of observation sub-areas using the latitude and longitude sampling intervals of the satellites in the satellite dataset; The satellites in the satellite data set observe the split observation sub-areas to form observation paths; The satellite dynamics model is constructed based on the observation path and the position coordinates of the ground target area in three-dimensional space.

4. The method according to claim 3, characterized in that In step 3, the satellite mission decision model is constructed as follows: in, is the satellite's mission decision variable, when When j Satellite x i Perform observations when , it means abandoning the task y j .

5. The method according to claim 4, characterized in that In step 3, the satellite imaging constraints include: time window constraints, storage constraints, power constraints, observation angle constraints, task execution uniqueness constraints and / or task conflict avoidance constraints; specifically include: Time window constraints: in, is the observable time window for each task, is the execution time of the task; Storage constraints: in, is the satellite's mission decision variable; Storage(y j |x i ) is the storage space consumed by the satellite when performing the mission; Storage total (x i ) is the total storage space of each satellite; N is the total number of tasks to be planned; Power Constraints: Among them, Elec(y i |x i ) is the amount of electricity consumed by the satellite when performing its mission, Elec trf (y j |x i ) is the amount of electricity consumed by the satellite to adjust its attitude during mission execution; Elec total (x i ) is the total power of each satellite; N is the total number of tasks to be planned; Observation angle constraint: -90°≤θ≤90°; Among them, θ is the satellite's observation angle, that is, the satellite's pitch angle, which is the angle between the satellite's attitude and the earth's surface, and describes the satellite's inclination relative to the earth's vertical direction. The orbital plane passing through the center of the earth is taken as an angle of 0°. A positive value of θ indicates that the satellite is facing upwards from the earth's surface, and a negative value indicates that it is facing downwards from the earth's surface. Task execution uniqueness constraint: in, is the coefficient factor indicating whether the ground target is observed; Avoid conflicting task constraints: in, For the second task y b Satellite x i The start time of the observation, For the first task y a Satellite x i The end time of the observation, For the first task y a To the second task y b The converted interval time, a is the starting path node, and b is the target path node.

6. The method according to claim 5, characterized in that In step 3, the satellite scheduling decision model constructed is: Among them, R(X) is the total revenue value of the satellite scheduling decision model, M is the total number of satellites, is the satellite's mission decision variable; f(y j ) is to execute the jth task y j The benefit value obtained by the subsequent satellite, θ(y j |x i ) is the jth task y j When executed, the i-th satellite x i The pitch angle, is the penalty factor, that is, when the pitch angle is 0°, the penalty factor is 0, and there is no effect on the benefit value; when the pitch angle increases to 90°, the penalty factor value is 1, the benefit value of the satellite completing the mission drops to 0, and the satellite imaging fails.

7. The method according to claim 6, characterized in that In step 4, the K-means algorithm includes: Among them, J is the objective function, which is to minimize the distance from the data point in the cluster to the centroid, n is the number of clusters, k represents the kth cluster, and X l represents a data point, C k is the centroid of the kth cluster, ||X l -C k || 2 Represents data point X l With cluster center C k The square of the Euclidean distance between them; the calculation formula of the Euclidean distance is: Among them, dis(X l , C k ) is the Euclidean distance, X l and C k are two m-dimensional vectors, where m is the dimension of the data feature space, X lt and C kt They are X l and C k The component of the tth dimension of ; Cluster center C k The calculation formula is: Among them, C k is the center of the kth cluster, S k is the number of samples belonging to the kth cluster, is the sum of the values ​​of the lth sample data point on the corresponding kth feature; through continuous iteration, until the center of the cluster no longer changes significantly, the cluster center C is obtained k .

8. The method according to claim 1, characterized in that In the step 5, the constraint check conditions in the optical imaging satellite scheduling task in the constraint check model based on the timeline advancement mechanism include: checking of cumulative constraints, task attribute constraints and task correlation constraints; wherein, Cumulative constraints require that the target statistic be within a preset value within a planning cycle; Task attribute constraints are to merge the constraints corresponding to task attributes through preprocessing; The task dependency constraint is a constraint condition corresponding to the task dependency merged based on the logical determination between consecutive tasks.

9. The method according to claim 8, characterized in that In step 6, the simulated annealing algorithm is: Where P(S1) is the probability that the feasible solution S1 can be accepted when the quality ΔC>0, T c is the current temperature. Each iteration, the temperature T c Will decay; T n =T c ·ε; where T n is the temperature of the next iteration, ε is the control parameter for adjusting the temperature drop rate; After the feasible solution is accepted, it is output as the optimal feasible solution, and the search time is extended by heating up the search model, where the search model is: T n =(T p -T c )·ω+T c ; Among them, T p is the temperature of the last feasible solution, ω is the control coefficient; When the temperature reaches the preset termination condition, the simulated annealing algorithm stops searching and outputs all the optimal feasible solutions to obtain the local optimal solution set.

10. The method according to claim 9, characterized in that In step 7, the method of inputting the local optimal solution set into the ant colony evolution algorithm for searching to obtain the global optimal solution includes: 1) Based on the local optimal solution, construct the path set L and speed set V; Among them, (L1, Y1) is the starting point of the observation path when the satellite performs the mission; (L n ,Y n ) is the end point of the observation path of the satellite when performing its mission; is the initial velocity vector of the observation path when the satellite performs its mission; is the final velocity vector of the observation path when the satellite performs the mission; the path nodes n in the observation path when the satellite performs the mission are connected in sequence, which is a task execution time sequence to be determined; 2) Based on the mapping relationship between the path set and whether the speed set task is executed by the satellite, a satellite path decision model is constructed; in, is the satellite's path decision variable; according to the satellite's path decision variable Choose whether to start from the starting path node a and visit the target path node b at time t, is the set of b that the ant colony can choose from a at time t, σ ab (t) is the heuristic function factor from a to b at time t: In the formula, c ab (t) is the amount of heuristic information remaining on the observation path ab by the satellite at time t, τ ab (t) is the amount of pheromone remaining on the observation path ab by the satellite at time t, α is the control parameter of the heuristic information, and β is the control parameter of the pheromone; 3) When the satellite has visited all feasible target path nodes, the process ends; if there are still unvisited target path nodes, the satellite continues to complete the observation path construction; when all target path nodes are visited, a new solution is obtained, which includes a feasible solution and an infeasible solution, wherein a feasible solution is a solution in which all target path nodes are visited and meet the constraint check conditions; an infeasible solution is a solution in which some target path nodes are not visited or do not meet the constraint check conditions; If the new solution is a feasible solution, the global pheromone matrix is ​​updated through the first update model; the first update model is as follows: Gt ab (t+1)=(1-ρ)Gτ ab (t)+ρGτ0; In the formula, Gτ ab (t+1) is the amount of global pheromone on the ab observation path corresponding to the target path node after the update at time t+1, Gτ ab (t) is the amount of global pheromone on the ab observation path corresponding to the target path node at time t, ρ is the evaporation coefficient of pheromone, Gτ0 is the amount of global pheromone remaining in the satellite, and Gτ0 is equal to Gτ ab (0) is the initial value of the global pheromone matrix; If the new solution is an infeasible solution, the global pheromone matrix is ​​updated through the second update model; the second update model is as follows: t ab (t+1)=(1-ρ)τ ab (t)-rt0; Among them, τ ab (t+1) is the amount of pheromone remaining on the ab observation path at time t+1, τ ab (t) is the amount of pheromone remaining on the observation path ab by the satellite at time t, and ρ is the evaporation coefficient of the pheromone.

Images