Satellite measurement and control scheduling method and system based on clustering

Through the construction of a hybrid integer programming model based on clustering, the problems of sequence dependence and over-ordering in satellite measurement and control scheduling are solved, efficient task arrangement is achieved, and the efficiency and quality of large-scale satellite measurement and control scheduling are significantly improved.

CN114781508BActive Publication Date: 2024-11-01NAT UNIV OF DEFENSE TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210407679.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Priority Date
2022-03-28
Filing Date
2022-04-19
Publication Date
2024-11-01
Estimated Expiration
2042-04-19

AI Technical Summary

Technical Problem

The satellite measurement and control scheduling problem is complex and difficult to solve, especially due to the difficulty of task scheduling caused by sequence dependence and over-ordering characteristics, it is difficult for the existing technology to quickly provide large-scale satellite measurement and control solutions.

Method used

Using cluster-based genetic algorithms, a hybrid integer programming model is constructed, and a task arrangement scheme is generated and optimized through technologies such as clustering K-means method and roulette selection to ensure that the task is efficiently executed within the appropriate time window.

Benefits of technology

It effectively solves the sequence dependency problem, improves the efficiency and quality of task arrangement, and can achieve the same optimization effect as traditional methods in a short time, significantly reducing the calculation time.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114781508B_ABST
    Figure CN114781508B_ABST
Patent Text Reader

Abstract

The present invention provides a satellite measurement and control scheduling method and system based on clustering, which divides tasks into K categories according to task characteristics, evaluates the close relationship between tasks, makes similar tasks as close as possible and different tasks as far as possible, and uses a clustering-based genetic algorithm to perform crossover and mutation on two different categories of individuals when the population crosses and mutates, achieving a balance between global search and local search, so that a higher quality solution can be obtained, and the sequence dependency problem is well solved. Experiments show that the method of the present invention only takes about one-third of the time of the knowledge-based genetic algorithm to achieve the same optimization effect.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the field of satellite mission planning, and in particular relates to a clustering-based satellite measurement and control scheduling method and system. Background Art

[0002] In recent years, with the rapid development of space technology, satellites have increasingly played an indispensable role as a space platform with application value in many fields. Satellites are used to complete various types of tasks such as target observation, navigation positioning, communication transmission, etc., depending on the user's task requirements. No matter what kind of task the satellite needs to perform, the satellite ground station needs to transmit the task in the form of instructions through the uplink directly or indirectly (first transmitted to the relay satellite). The process of the ground station uploading tasks and action instructions to the satellite and obtaining the satellite's operating status is called satellite measurement and control. The satellite measurement and control scheduling system achieves effective management of satellite and ground station resources by using efficient scheduling algorithms. With the rapid increase in the number of satellites in orbit and the need for satellite management and control, the effective task scheduling of satellite measurement and control scheduling problems has become more challenging.

[0003] Satellite TT&C scheduling describes the process of reasonably arranging the communication between satellites and ground stations to achieve satellite status monitoring and command injection. It can also be simply described as: finding a suitable combination of TT&C tasks for a series of satellites and a series of satellite ground stations within a mutually visible time window. The main reason why the satellite TT&C scheduling problem is difficult to solve is the characteristics of sequence dependency and oversubscription. Sequence dependency means that the scheduling result of a task will affect subsequent tasks. This makes it difficult to find a feasible solution. Oversubscription means that even if the ground station time window is fully used, some tasks will not be successfully executed. The simultaneous existence of these two characteristics makes the TT&C scheduling problem particularly difficult. The satellite TT&C scheduling problem has been proven to be NP-complete. Summary of the invention

[0004] The technical problem to be solved by the present invention is how to quickly provide a large-scale satellite measurement and control solution, and proposes a satellite measurement and control scheduling method and system based on clustering.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] A satellite measurement and control scheduling method based on clustering includes the following steps:

[0007] Step 1: Obtain schedulable satellite resources, ground station resources, task sets, and visible time window sets;

[0008] Step 2: Construct a mixed integer programming model based on the content obtained in step 1;

[0009] Step 3: Solving the mixed integer programming model;

[0010] Step 4: Output the solved measurement and control solution.

[0011] Furthermore, the mixed integer programming model is:

[0012] The objective function is:

[0013]

[0014] Among them, p t represents the benefit of task t, is a 0-1 decision variable, indicating whether task t is scheduled in the i-th visible time window on antenna a. 1 means scheduled and 0 means no. t represents task, T represents task set, a represents antenna, and A represents antenna set. represents the i-th visible time window in which task t is scheduled on antenna a, and TWta represents the set of time windows in which task t is scheduled on antenna a;

[0015] The constraints are:

[0016]

[0017] Formula 2 indicates that the task needs to start executing the task after the allowed start time; represents the actual measurement and control time length of task t in the i-th visible time window on antenna a, d t It represents the length of measurement and control time required by task t;

[0018]

[0019] Formula 3 indicates that the task needs to be completed before the allowed end time; represents the actual start time of task t scheduled on the ith visible time window on antenna a, represents the end time of task t scheduled in the i-th visible time window on antenna a;

[0020]

[0021] Formula 4 shows that a task can be executed at most once; represents the earliest allowed start time of task t in the ith visible time window on antenna a;

[0022]

[0023] Formula 5 shows that the actual execution time of the task should be the same as the required time; represents the latest allowed end time of task t scheduled in the ith visible time window on antenna a;

[0024]

[0025] Equation 6 shows that the task execution process needs to be within the time range of a ground station; represents the start time of the ith visible time window of task t on antenna a;

[0026]

[0027] Equation 7 shows that each task can only be served by one antenna;

[0028]

[0029] Equation 8 shows that each task can only be executed within a visible time window; represents the end time of the ith visible time window of task t scheduled on antenna a;

[0030]

[0031] Equation 9 shows that each task is executed at most once on all antennas;

[0032]

[0033] Formula 10 shows that each task is executed at most once in the entire time window;

[0034]

[0035] Formula 11 shows that the task is executed at most once in the planning cycle;

[0036]

[0037] Formula 12 shows that the two tasks must meet the interval requirement of task switching time; γ represents the switching time between tasks.

[0038] Furthermore, the method for solving the mixed integer programming model in step 3 is a clustering-based genetic algorithm.

[0039] Further, the clustering-based genetic algorithm is:

[0040] Step 3.1: Initialize genetic algorithm parameters and clustering K-means method parameters;

[0041] Step 3.2: Generate an initial population, where each individual in the population encodes the encoding sequence of all tasks in the task set;

[0042] Step 3.3: When the number of iterations does not reach the maximum number of iterations, execute step 3.3.1; otherwise, execute step 3.4;

[0043] Step 3.3.1: Generate an initial solution of the measurement and control scheme for each individual in the population and calculate the objective function value;

[0044] Step 3.3.2: If the maximum objective function value in the contemporary population is greater than the optimal objective function value, update the optimal objective function value and count count1 = count1 + 1;

[0045] If the maximum objective function value in the current population is greater than the maximum objective function value in the previous population, update the count count2 = count2 + 1;

[0046] If the maximum objective function value in the current population is less than the maximum function value in the previous population multiplied by the proportion per, then update the count count3 = count3 + 1;

[0047] Step 3.3.3: Use roulette wheel to select individuals to generate a new population according to the objective function value;

[0048] Step 3.3.4: Update the new population using a clustering-based crossover and mutation method;

[0049] Step 3.3.5: If count1 is equal to the threshold Thre1, update the parameter K value of the k-means method and re-cluster the tasks in the task set based on the task attributes, and reset the counting parameter count1;

[0050] If count2 is equal to Thre2, the optimal individual corresponding to the optimal objective function value replaces the individual with the smallest objective function value in the new population, and resets the counting parameter count2;

[0051] If count3 is equal to Thre3, local optimization is performed on the individual with the largest objective function value in the new population, a new individual is randomly generated, the individual with the smallest objective function value in the new population is deleted, and the counting parameter count3 is reset;

[0052] Step 3.3.6: Record the maximum objective function value in the new population as the maximum objective function value of the previous generation population;

[0053] Step 3.4: Output the individual corresponding to the optimal objective function value as the measurement and control plan.

[0054] Furthermore, in step 3.3.1: the method for generating the initial solution of the measurement and control scheme for each individual in the population is the task scheduling algorithm, which is specifically:

[0055] All tasks in the task set are tried to be arranged to each task in the order of visible time windows according to the coding sequence. If all time windows have been tried and arranged successfully, the result of successful arrangement is output as the initial solution. Otherwise, continue to try to arrange. The specific arrangement method for each task is:

[0056] 1): Calculate the earliest actual available time of task t eat t and the latest actual available time lat t ;

[0057] 2): If the length of the i-th time window of task t at antenna a is greater than the measurement and control time required by task t, that is, And the actual available time length of task t is greater than the measurement and control time length of task t, that is, (lat t -eat t )≥d t , then go to step 3); otherwise, continue to arrange the next task;

[0058] 3): Eat the earliest actual available time of task t t As the task start time, if the earliest actual available time of task t is eat t Equal to the start time of the ith visible time window of task t on antenna a Go to step 4); otherwise, go to step 5);

[0059] 4): Schedule task t in the i-th visible time window on antenna a Update the remaining time window of the window to a new time window

[0060] 5): Schedule task t in the i-th visible time window on antenna a Update the remaining time window of the window to two new time windows and

[0061] Furthermore, the method for generating the initialization population in step 3.2 is to use a plurality of heuristic initialization methods to generate the initialization population, wherein the initialization population contains individuals generated by various heuristic initialization methods.

[0062] Furthermore, the cluster-based crossover method used in step 3.3.4 means:

[0063] Divide all tasks into K categories according to their attributes using the k-means clustering method;

[0064] During crossover, gene fragments of equal length are selected from two groups of tasks belonging to different categories for crossover, and the category of the task is selected using a roulette wheel based on the objective function value.

[0065] Furthermore, in step 3.3.4, there are two clustering-based mutation methods used, one is mutation within the same category, and the other is mutation within different categories. In each mutation operation, one of the two mutation methods is randomly selected to perform the mutation operation;

[0066] Variation within the same category refers to swapping the positions of two tasks in the same category within an individual.

[0067] Variation within different categories refers to swapping the positions of two tasks in different categories within an individual.

[0068] Furthermore, the clustering method in step 3.1 is based on the attributes of the task for clustering, which include the earliest allowed start time, the latest allowed end time, the benefit, and the required control time length of the task. When the K value is updated and clustering is performed again, the feature of whether the task is successfully arranged is added.

[0069] The present invention also provides a satellite measurement, control and scheduling system based on clustering, comprising the following modules:

[0070] Input module: used to obtain schedulable satellite resources, ground station resources, task sets, and time window sets for satellites to see ground stations;

[0071] Model building module: used to build a mixed integer programming model based on the content obtained in step 1;

[0072] Solving module: used for solving the mixed integer programming model;

[0073] Solution output module: used to output the solved measurement and control solution.

[0074] By adopting the above technical solution, the present invention has the following beneficial effects:

[0075] The present invention discloses a satellite measurement and control scheduling method and system based on clustering, which uses a cluster-based genetic algorithm C-BGA (Cluster-based genetic algorithm) to solve the satellite measurement and control scheduling problem SRSP (satellite range scheduling problem). Tasks are divided into K categories according to task characteristics, and the close relationship between tasks is evaluated, so that similar tasks are as close as possible and different tasks are as far as possible. When the population is crossed and mutated, individuals in two different categories, that is, two dissimilar task categories, are selected for crossover and mutation, so as to achieve a balance between global search and local search, so that a higher quality solution can be obtained, and the sequence dependency problem is well solved. The task scheduling algorithm is used to generate an initial solution of the measurement and control scheme for each individual in the population, the feasibility of task execution is judged, and effective arrangements are made. When initializing the population, a plurality of heuristic initialization methods are used to generate the initialization population, and the generation is not limited to one strategy, so that the solution is diversified. Experiments show that the C-BGA method of the present invention only takes about one-third of the time of KBGA to achieve the same optimization effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 It is a flow chart of the system of the present invention;

[0077] Figure 2 It is the variation curve of K value in clustering method;

[0078] Figure 3 Box plots for L_L-5 and LH-5 examples;

[0079] Figure 4 Results for a medium-sized test set;

[0080] Figure 5 Results on a large test set. DETAILED DESCRIPTION

[0081] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0082] A satellite measurement and control scheduling method based on clustering, such as Figure 1 As shown, the following steps are included:

[0083] Step 1: Obtain schedulable satellite resources, ground station resources, task sets, and time window sets for satellites to see ground stations;

[0084] Step 2: Construct a mixed integer programming model based on the content obtained in step 1.

[0085] In this embodiment, the mixed integer programming model is:

[0086] The objective function is:

[0087]

[0088] Among them, p t represents the benefit of task t, is a 0-1 decision variable, indicating whether task t is scheduled in the i-th visible time window on antenna a. 1 means scheduled and 0 means no. t represents task, T represents task set, a represents antenna, and A represents antenna set. Indicates that task t is scheduled in the ith visible time window on antenna a, TW ta represents the set of time windows in which task t is scheduled on antenna a;

[0089] The constraints are:

[0090]

[0091] Formula 2 indicates that the task needs to start executing the task after the allowed start time; represents the actual measurement and control time length of task t in the i-th visible time window on antenna a, d t It represents the length of measurement and control time required by task t;

[0092]

[0093] Formula 3 indicates that the task needs to be completed before the allowed end time; represents the actual start time of task t scheduled on the ith visible time window on antenna a, represents the end time of task t scheduled in the i-th visible time window on antenna a;

[0094]

[0095] Formula 4 shows that a task can be executed at most once; represents the earliest allowed start time of task t in the ith visible time window on antenna a;

[0096]

[0097] Formula 5 shows that the actual execution time of the task should be the same as the required time; represents the latest allowed end time of task t scheduled in the ith visible time window on antenna a;

[0098]

[0099] Equation 6 shows that the task execution process needs to be within the time range of a ground station; represents the start time of the ith visible time window of task t on antenna a;

[0100]

[0101] Equation 7 shows that each task can only be served by one antenna;

[0102]

[0103] Equation 8 shows that each task can only be executed within a visible time window; represents the end time of the ith visible time window of task t scheduled on antenna a;

[0104]

[0105] Equation 9 shows that each task is executed at most once on all antennas;

[0106]

[0107] Formula 10 shows that each task is executed at most once in the entire time window;

[0108]

[0109] Formula 11 shows that the task is executed at most once in the planning cycle;

[0110]

[0111] Formula 12 shows that the two tasks must meet the interval requirement of task switching time; γ represents the switching time between tasks.

[0112] Step 3: Solve the mixed integer programming model.

[0113] In this embodiment, the method for solving the mixed integer programming model is a clustering-based genetic algorithm.

[0114] Clustering based genetic algorithms are:

[0115] Step 3.1: Initialize genetic algorithm parameters and clustering K-means method parameters;

[0116] Step 3.2: Generate an initial population, where each individual in the population encodes the encoding sequence of all tasks in the task set;

[0117] In this embodiment, the method for generating the initialization population is to use a plurality of heuristic initialization methods to generate the initialization population, and the initialization population contains individuals generated by various heuristic initialization methods.

[0118] In this embodiment, the multiple heuristic initialization methods used include four heuristic population initialization methods and one random population initialization method.

[0119] The four heuristic population initializations are called the initialization method based on the earliest allowed start time sorting, the initialization method based on the latest allowed end time sorting, the initialization method based on the task benefit sorting, and the initialization method based on the task duration sorting.

[0120] Initialization based on the earliest allowed start time sorting (EST): Sort the measurement and control tasks in order from the earliest allowed start time to the last, and generate individuals according to the sorted order.

[0121] Initialization based on latest allowed end time sorting (LET): Sort the measurement and control tasks in order from the earliest to the latest allowed end time, and generate individuals according to the sorted order.

[0122] Initialization based on task benefit sorting (TP): Sort the measurement and control tasks in descending order according to their task benefits from large to small, and generate individuals according to the sorted order.

[0123] Initialization based on task duration sorting (TD): Sort the measurement and control tasks in ascending order according to their task durations, and generate individuals according to the sorted order.

[0124] After each heuristic rule is used to generate individuals in the population, it is necessary to make the individuals more diverse by adjusting the partial sequence of the individuals. Combining the heuristic and local random methods can make it easier for each individual to find a good starting position for the search. The four heuristic methods and the random method each occupy 20% of all individuals in the population. Of course, the proportion of the four heuristic methods and the random method to all individuals in the population can also be adjusted as needed. In order to increase the diversity of solutions, the heuristic initialization method can also include other heuristic methods not limited to the four heuristic methods mentioned above.

[0125] Step 3.3: When the number of iterations does not reach the maximum number of iterations, execute step 3.3.1; otherwise, execute step 3.4;

[0126] Step 3.3.1: Generate an initial solution for the measurement and control scheme for each individual in the population and calculate the objective function value.

[0127] In this embodiment, in step 3.3.1: the method for generating an initial solution of the measurement and control scheme for each individual in the population is a task scheduling algorithm, specifically:

[0128] All tasks in the task set are tried to be arranged to each task in the order of visible time windows according to the coding sequence. If all time windows have been tried and arranged successfully, the result of successful arrangement is output as the initial solution. Otherwise, continue to try to arrange. The specific arrangement method for each task is:

[0129] 1): Calculate the earliest actual available time of task t eat t and the latest actual available time lat t ;

[0130] 2): If the length of the i-th time window of task t at antenna a is greater than the measurement and control time required by task t, that is, And the actual available time length of task t is greater than the measurement and control time length of task t, that is, (lat t -eat t )≥d t , then go to step 3); otherwise, continue to arrange the next task;

[0131] 3): Eat the earliest actual available time of task t t As the task start time, if the earliest actual available time of task t is eat t Equal to the start time of the ith visible time window of task t on antenna a Go to step 4); otherwise, go to step 5);

[0132] 4): Schedule task t in the i-th visible time window on antenna a Update the remaining time window of the window to a new time window

[0133] 5): Schedule task t in the i-th visible time window on antenna a Update the remaining time window of the window to two new time windows and

[0134] In this embodiment, all tasks in the task set are judged in turn whether each time window can execute the task. If the time window length exceeds the time required by the measurement and control task, the task is attempted to be scheduled. The present invention adopts a fast judgment method to determine whether a task has the possibility of being scheduled. First, two new variables eat are introduced. t and lat t, respectively represent the earliest actual available time and the latest actual available time, and the calculation method is shown in Formula 13. If the actual available time exceeds the required time of the measurement and control task, it can be successfully scheduled, and the task is scheduled to start execution at the earliest actual available time. After the task is successfully scheduled, the remaining available time window resources need to be updated. If the earliest actual available time of the task is equal to the earliest allowed start time of the task, the remaining time window of the window is cut into a new time window, otherwise, the remaining time window of the window is cut into two new time windows. After the time window is cut, the time window set is updated and the next task is scheduled.

[0135]

[0136]

[0137] Step 3.3.2: If the maximum objective function value in the contemporary population is greater than the optimal objective function value, the optimal objective function value is updated and counted count1=count1+1; the optimal objective function value in this embodiment is the optimal objective function value in the entire optimization process, and its initial value can be set to 0. Then, in each iteration, the maximum objective function value in each generation of the population is compared with it to obtain the updated optimal value.

[0138] If the maximum objective function value in the current population is greater than the maximum objective function value in the previous population, update the count count2 = count2 + 1;

[0139] If the maximum objective function value in the current population is less than the maximum function value in the previous population multiplied by the proportion per, then update the count count3 = count3 + 1;

[0140] Step 3.3.3: Use roulette wheel to select individuals to generate a new population according to the objective function value;

[0141] In this embodiment, a roulette wheel method is used to select individuals to generate a new population, and then the new population is subjected to crossover and mutation.

[0142] The roulette selection process needs to better reflect the performance differences between individuals in the population. The present invention uses the roulette method to select individuals and task categories based on the objective function value of the individual, making it easier for individuals with better performance and the categories they are in to be selected. The probability formula for selecting individuals using the roulette is as follows:

[0143]

[0144] in, is the probability of individual j, f j is the objective function value of individual j, and P is the population.

[0145] Step 3.3.4: Update the new population using a clustering-based crossover and mutation method.

[0146] In this embodiment, the clustering-based crossover method refers to:

[0147] Divide all tasks into K categories according to their attributes using the k-means clustering method;

[0148] When crossing, we select gene segments of equal length from two groups of tasks belonging to different categories for crossover. Specifically, we select a category according to the roulette wheel, randomly select a task in the category, and use the randomly selected task as the beginning of the gene segment. Then we select another category according to the roulette wheel, and when selecting, we remove the currently selected task category. The probability formula for selecting a category according to the roulette wheel is:

[0149]

[0150] in, is the probability of category k, is task t in category k k The objective function value, f t represents the objective function value of task t, and K is the category set.

[0151] There are two types of clustering-based mutation methods: one is mutation within the same category, and the other is mutation within different categories. In each mutation operation, one of the two mutation methods is randomly selected to perform the mutation operation;

[0152] Variation within the same category refers to swapping the positions of two tasks in the same category within an individual.

[0153] Variation within different categories refers to swapping the positions of two tasks in different categories within an individual.

[0154] The use of these two mutation methods can not only achieve the exchange of execution positions between tasks with similar data features, but also allow the selected task to be exchanged with tasks with larger feature differences. The use of two mutation methods can increase the diversity of mutations, giving the algorithm a better chance to find a better task execution order. By updating individuals in the population through cluster-based crossover and mutation, crossover mutation is performed on individuals from two different categories, that is, two dissimilar task categories, achieving a balance between global search and local search, so that higher quality solutions can be obtained, which well solves the problem of sequence dependency.

[0155] Step 3.3.5: If count1 is equal to the threshold Thre1, update the parameter K value of the k-means method and re-cluster the tasks in the task set based on the task attributes, and reset the counting parameter count1;

[0156] If count2 is equal to Thre2, the optimal individual corresponding to the optimal objective function value replaces the individual with the smallest objective function value in the new population, and resets the counting parameter count2;

[0157] If count3 is equal to Thre3, local optimization is performed on the individual with the largest objective function value in the new population, a new individual is randomly generated, and the individual with the smallest objective function value in the new population is deleted, and the counting parameter count3 is reset;

[0158] Step 3.3.6: Record the maximum objective function value in the new population as the maximum objective function value of the previous generation population;

[0159] The goal of the clustering method is to divide the sample set into multiple subsets based on data features. This method divides the data into multiple groups of discrete or hierarchical structures based on the similarity of data features. The ultimate effect of classification is to make the samples within a group as close as possible, and the samples of a group are not significantly different. The reason why the present invention chooses a clustering algorithm is that on the one hand, it wants to make full use of satellite mission data and use data to drive the optimization process. On the other hand, there are difficult-to-determine classification standards for satellite mission data in planning and scheduling. The clustering method can better realize the classification of tasks according to features. In order to prevent the deviation of the calculated clustering results due to the excessive dimension of some features, the feature data needs to be normalized first. Normalization can make the data unaffected by the dimension and unit, and highlight the content of the data feature itself. The normalized calculation formula is as follows:

[0160]

[0161] Among them, y is the eigenvalue, y max is the maximum value of feature y, y min is the minimum value of feature y.

[0162] The steps of the clustering method in this embodiment are:

[0163] Input: task feature matrix (each column represents a feature), number of features, number of categories K

[0164] Output: Clustering results

[0165] 1): Randomly select K samples from the feature set as cluster centers (1, 2, ..., K);

[0166] 2): Calculate the Euclidean distance from the data to each center point, and select the nearest center point as the data category;

[0167] 3): Recalculate each category to get a new center point according to the formula |C j | is the number of data belonging to category j, and x represents the data;

[0168] 4): If the center point changes, go to step 2; otherwise, go to step 5;

[0169] 5): Output clustering results.

[0170] In this embodiment, the clustering method is based on the characteristics of the task for clustering, and the characteristics of the task include the earliest allowed start time, the latest allowed end time, the income, and the length of the task required for measurement and control. Each column of the task feature matrix represents a task feature, forming a matrix. When recalculating the cluster, the task execution result of the optimal solution from the beginning of optimization to the present is added to the rightmost column of the feature matrix, that is, when a certain condition is reached and clustering needs to be re-performed, the feature of whether the task is successfully arranged is added, and the arranged tasks are marked as 1, and the unarranged tasks are marked as 0. The clustering method of the present invention divides the task data into K categories according to the task characteristics, evaluates the close relationship between tasks, makes similar tasks as close as possible and different tasks as far as possible, and when the population crosses and mutates, individuals from two different categories, that is, two dissimilar task categories, cross and mutate, so that a higher quality solution can be obtained, and the optimization process of the genetic algorithm is improved. The number of categories of the clustering method changes with the optimization process, and the following will also be proved by experiments that the information provided by clustering for the genetic algorithm is more valuable.

[0171] For the SRSP problem, the number of clusters K divided by clustering should change with the optimization process. A fixed value is prone to invalid search due to inaccurate classification after the optimization process reaches a certain stage. In order to improve the search efficiency and search effect, the K value is updated in this embodiment when count1 is equal to the threshold Thre1, the parameter K value of the k-means method is updated and the tasks in the task set are clustered again based on the task characteristics. The update curve of the K value can be set in advance. In this embodiment, the number of classifications K is selected to change continuously in the interval range of [previous session, next session] in a way that the number of triggering change conditions is reduced by 1. The process of K value change is as follows: Figure 2 As shown in Figure 2, when re-clustering, the feature of whether the task is successfully executed is added.

[0172] Step 3.4: Output the individual corresponding to the optimal objective function value as the measurement and control plan.

[0173] The clustering-based genetic algorithm of the present invention improves the situation that the genetic algorithm has good global search but poor local search performance, and enhances the local search performance of the genetic algorithm.

[0174] Step 4: Output the solved measurement and control solution.

[0175] Experimental verification

[0176] 1) Planning results under different task scales

[0177] 30 instances of different densities are used to verify the planning performance of the algorithm. The results are shown in Table 1. The increase in the scale of the examples is accompanied by an increase in the benefit value. In the scene number expression in the leftmost column of Table 1, the three symbols S on the left represent small scale, M represents medium scale, and L represents large scale; the two symbols L in the middle represent low density and H represents high density; the rightmost number represents the sequence number. When the task scale increases to medium, the growth rate of the benefit begins to slow down. This is because it becomes particularly difficult for limited satellite and ground station resources to find a better task execution plan among many tasks. The C-BGA algorithm has achieved good planning results in all 30 scenarios. In small and medium-scale scenarios, the planning results are C-BGA, Fireworks algorithm FWA (Fireworks algorithm), ALNS / TPF (Tabu-based adaptive large neighborhood search) with taboo strategy, and KBGA (Knowledge based genetic algorithm) from good to bad. In large-scale scenarios, KBGA and ALNS / TPF algorithms are between C-BGA and FWA algorithms. From the overall trend, it can be seen that the C-BGA algorithm is more likely to outperform other comparison algorithms in larger-scale scenarios.

[0178] Table 1 Comparison of benefits under different task scales in 30 scenarios

[0179]

[0180]

[0181] 2) Algorithm stability analysis

[0182] Large-scale instances can more easily reflect the stability of the algorithm. The box plots of LL-5 and LH-5 examples are as follows: Figure 3 As shown in the figure, the stability of the C-BGA algorithm is the best, and it maintains a small volatility under the instance of 1000 tasks. The worst performance is ALNS / TPF, which is most likely to obtain results with a large deviation from the average value. The FWA algorithm and KBGA algorithm are between the C-BGA algorithm and the ALNS / TPF algorithm.

[0183] 3) Optimize time analysis

[0184] This paper compares the time required for the clustering-based genetic algorithm C_BGA algorithm to achieve the same average optimization result as KBGA. The results are as follows Figure 4 and Figure 5 As shown, Figure 4 and Figure 5 The results are respectively for medium-scale and large-scale optimization, but the results obtained through experiments are consistent. The 100% column in the figure represents the time taken by the KBGA algorithm to reach the average performance of each instance, and the black column represents the percentage of the C-BGA algorithm based on clustering of the present invention in the reference algorithm time. It can be seen from the results that the C-BGA algorithm only needs about one-third of the time of the KBGA algorithm to achieve the same optimization effect. In other words, a solution of good quality can be found in a very short time after the population is initialized.

[0185] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A satellite measurement and control scheduling method based on clustering, characterized in that: The following steps are involved: Step 1: Obtain schedulable satellite resources, ground station resources, task sets, and visible time window sets; Step 2: Construct a mixed integer programming model based on the content obtained in step 1; Step 3: Solving the mixed integer programming model; Step 4: Output the obtained measurement and control scheme; The mixed integer programming model is: The objective function is: (1) in, Representation Task of income, is a 0-1 decision variable, indicating the task Is it arranged on the antenna? Previous Visible time windows, 1 for scheduled, 0 for not, Indicates the task, T represents a task set, Indicates the antenna, represents the antenna set, Representation Task Arrange in antenna Previous Visible time window, Representation Task Arrange in antenna The set of time windows on ; The constraints are: (2) Formula 2 indicates that the task needs to start executing the task after the allowed start time; Indicates the task Arrange on antenna Previous The actual measurement and control time length in a visible time window, Indicates the task Required length of measurement and control time; (3) Formula 3 indicates that the task needs to be completed before the allowed end time; Indicates the task Arrange on antenna Previous The actual start time of the visible time window, Indicates the task Arrange on antenna Previous The end time of the visible time window; (4) Formula 4 shows that a task can be executed at most once; Representation Task Arrange in antenna Previous The earliest allowed start time in a visible time window; (5) Formula 5 shows that the actual execution time of the task should be the same as the required time; Representation Task Arrange in antenna Previous The latest allowed end time in the visible time window; (6) Equation 6 shows that the task execution process needs to be within the time range of a ground station; Indicates the task Arrange on antenna Previous The start time of the visible time window; (7) Equation 7 shows that each task can only be served by one antenna; (8) Equation 8 shows that each task can only be executed within a visible time window; Indicates the task Arrange on antenna Previous The end time of the visible time window; (9) Equation 9 shows that each task is executed at most once on all antennas; (10) Formula 10 shows that each task is executed at most once in the entire time window; (11) Equation 11 shows that the task is executed at most once in the planning cycle; (12) Formula 12 shows that the two tasks must meet the interval requirement of task switching time; Indicates the transition time between tasks.

2. The method according to claim 1, characterized in that The method for solving the mixed integer programming model in step 3 is a clustering-based genetic algorithm.

3. The method according to claim 2, characterized in that The clustering-based genetic algorithm is: Step 3.1: Initialize genetic algorithm parameters and clustering K-means method parameters; Step 3.2: Generate an initial population, where each individual in the population encodes the encoding sequence of all tasks in the task set; Step 3.3: When the number of iterations does not reach the maximum number of iterations, execute step 3.3.1; Otherwise, proceed to step 3.4; Step 3.3.1: Generate an initial solution of the measurement and control scheme for each individual in the population and calculate the objective function value; Step 3.3.2: If the maximum objective function value in the contemporary population is greater than the optimal objective function value, update the optimal objective function value and count ; If the maximum objective function value in the current population is greater than the maximum objective function value in the previous population, the count is updated ; If the maximum objective function value in the current population is less than the maximum function value in the previous population multiplied by the proportion , then update the count ; Step 3.3.3: Use roulette wheel to select individuals to generate a new population according to the objective function value; Step 3.3.4: Update the new population using a clustering-based crossover and mutation method; Step 3.3.5: If Equal to threshold , then update the parameter K value of the k-means method and re-cluster the tasks in the task set based on the task attributes, and reset the counting parameter ; if equal , then the optimal individual corresponding to the optimal objective function value replaces the individual with the smallest objective function value in the new population, and resets the counting parameter ; if equal , then perform local optimization on the individual with the largest objective function value in the new population, randomly generate a new individual and delete the individual with the smallest objective function value in the new population, and reset the counting parameter ; Step 3.3.6: Record the maximum objective function value in the new population as the maximum objective function value of the previous generation population; Step 3.4: Output the individual corresponding to the optimal objective function value as the measurement and control plan.

4. The method according to claim 3, characterized in that In step 3.3.1: The method for generating the initial solution of the measurement and control scheme for each individual in the population is the task scheduling algorithm, which is as follows: All tasks in the task set are tried to be arranged to each task in the order of visible time windows according to the coding sequence. If all time windows have been tried and arranged successfully, the result of successful arrangement is output as the initial solution. Otherwise, continue to try to arrange. The specific arrangement method for each task is: 1): Computational tasks Earliest actual available time and the latest actual available time ; 2): If the task On the antenna No. The time window length is greater than the task The required measurement and control time length, i.e. , and the task The actual available time is longer than the task The length of the measurement and control time, that is , then go to step 3); otherwise, continue to arrange the next task; 3): The earliest actual available time of task t As the task start time, if the earliest actual available time of task t is Equal to task Arrange in antenna Previous The start time of the visible time window , go to step 4); otherwise, go to step 5); 4): Task Arrange on antenna Previous Visible time windows , update the remaining time window of the window to a new time window ; 5): Task Arrange on antenna Previous Visible time windows , update the remaining time window of the window to two new time windows and .

5. The method according to claim 3, characterized in that: The method for generating the initialization population in step 3.2 is to use a plurality of heuristic initialization methods to generate the initialization population, wherein the initialization population contains individuals generated by various heuristic initialization methods.

6. The method according to claim 3, characterized in that The cluster-based crossover method used in step 3.3.4 means: Divide all tasks into K categories according to their attributes using the k-means clustering method; When crossing over, gene fragments of equal length are selected from two groups of tasks belonging to different categories for crossover, and the categories of tasks are selected according to roulette wheel.

7. The method according to claim 6, characterized in that There are two clustering-based mutation methods used in step 3.3.

4. One is mutation within the same category, and the other is mutation between different categories. In each mutation operation, one of the two mutation methods is randomly selected to perform the mutation operation. Variation within the same category refers to swapping the positions of two tasks within the same category within an individual; Variation within different categories refers to swapping the positions of two tasks in different categories within an individual.

8. The method according to claim 3, characterized in that The clustering-based method in step 3.1 performs clustering based on the characteristics of the task, which include the earliest allowed start time, the latest allowed end time, the benefit, and the required measurement and control time length of the task. When the K value is updated and clustering is performed again, the feature of whether the task is successfully scheduled is added.

9. A satellite measurement, control and scheduling system based on clustering, characterized in that: Includes the following modules: Input module: used to obtain schedulable satellite resources, ground station resources, task sets, and time window sets for satellites to see ground stations; Model building module: used to build a mixed integer programming model based on the content obtained in step 1; The mixed integer programming model is: The objective function is: (1) in, Representation Task of income, is a 0-1 decision variable, indicating the task Is it arranged on the antenna? Previous Visible time windows, 1 for scheduled, 0 for not, Indicates the task, T represents a task set, Indicates the antenna, represents the antenna set, Representation Task Arrange in antenna Previous Visible time window, Representation Task Arrange in antenna The set of time windows on ; The constraints are: (2) Formula 2 indicates that the task needs to start executing the task after the allowed start time; Indicates the task Arrange on antenna Previous The actual measurement and control time length in a visible time window, Indicates the task Required length of measurement and control time; (3) Formula 3 indicates that the task needs to be completed before the allowed end time; Indicates the task Arrange on antenna Previous The actual start time of the visible time window, Indicates the task Arrange on antenna Previous The end time of the visible time window; (4) Formula 4 shows that a task can be executed at most once; Indicates the task Arrange on antenna Previous The earliest allowed start time in a visible time window; (5) Formula 5 shows that the actual execution time of the task should be the same as the required time; Representation Task Arrange in antenna Previous The latest allowed end time in the visible time window; (6) Equation 6 shows that the task execution process needs to be within the time range of a ground station; Indicates the task Arrange on antenna Previous The start time of the visible time window; (7) Equation 7 shows that each task can only be served by one antenna; (8) Equation 8 shows that each task can only be executed within a visible time window; Representation Task Arrange in antenna Previous The end time of the visible time window; (9) Equation 9 shows that each task is executed at most once on all antennas; (10) Formula 10 shows that each task is executed at most once in the entire time window; (11) Equation 11 shows that the task is executed at most once in the planning cycle; (12) Formula 12 shows that the two tasks must meet the interval requirement of task switching time; Indicates the transition time between tasks; Solving module: used for solving the mixed integer programming model; Solution output module: used to output the solved measurement and control solution.