Iterative adaptive large neighborhood search method for large-scale communication satellite measurement and control scheduling problem

Abstract

The invention provides an iterative adaptive large neighborhood search method for a large-scale communication satellite measurement and control scheduling problem, which is used for quickly solving the communication satellite measurement and control scheduling problem, and different from the traditional measurement and control problem, the communication satellite measurement and control scheduling problem considers an inter-satellite link to promote the quick completion of a measurement and control task. In the method, a heuristic initial solution is improved by an adaptive large neighborhood search algorithm, and then a frequent pattern mining method is embedded into an iteration mechanism to guide a search process of the algorithm: on one hand, a poorer structure is mined from a low-quality solution to guide a removal process of the adaptive large neighborhood search algorithm; on the other hand, a dominant structure is excavated from a high-quality solution to guide the construction of a new solution; experimental test results of different task scales show that the iterative adaptive large neighborhood search algorithm can effectively process the communication satellite measurement and control scheduling problem, and the performance of the algorithm is superior to that of the three most advanced algorithms.

Description

Technical field:

[0001] The present invention relates to the field of satellite measurement and control technology, and in particular to an iterative adaptive large neighborhood search method for large-scale communication satellite measurement and control scheduling problems. Background technology:

[0002] A network of communication satellites operating at high speed in low orbits forms a low-orbit Internet constellation. With its broadband communication capabilities, low latency, and global coverage of users, the low-orbit Internet constellation has become an indispensable supplement to the ground network. For example, Starlink and OneWeb provide global space broadband access to support global communication missions, including mobile services, emergency communications, and emerging services such as multimedia and the Internet.

[0003] In order to improve the coverage and sustainability of low-orbit Internet constellations, the size of the constellation is constantly expanding. In addition, the joint management of low-orbit Internet constellations and other constellations such as remote sensing constellations is also a general trend. Therefore, a large amount of communication is required between satellites and ground stations to ensure the normal operation of the constellation. In order to ensure that the satellite-to-ground communication requests of the low-orbit Internet constellation are met and improve their efficiency, this paper extends the Satellite Range Scheduling Problem (SRSP, hereinafter referred to as the TT&C problem) to the communication satellite constellation and proposes a large-scale Communication Satellite Range Scheduling Problem (CSRSP).

[0004] According to Luo's research, the TT&C problem is mainly about selecting the appropriate ground antenna and time window for satellite requests to maximize the total profit. However, the scheduling process of the communication satellite TT&C scheduling problem is different from the traditional TT&C problem. From the perspective of the constellation background, the low-orbit Internet constellation has on-board processing and inter-satellite forwarding capabilities. Each satellite has established four inter-satellite links: two links connect adjacent satellites in the same orbital plane, and the other two connect satellites in adjacent orbital planes. Inter-satellite links increase the execution opportunities of TT&C tasks and make up for the disadvantage of insufficient resources. TT&C tasks can transmit satellite data directly to the ground station through the satellite-to-ground link, or forward these data to adjacent satellites through the inter-satellite link. The combination and sequence of satellite-to-ground links in the low-orbit Internet constellation are pre-arranged, and the time period for establishing satellite-to-ground links is defined as arc segments. TT&C tasks must be performed within the scope of these arc segments. In terms of resource types, each satellite and ground antenna is equipped with two types of payloads: millimeter wave band (Q / V band) and microwave band (S band). Therefore, there are two types of arcs as resources that can be allocated to perform TT&C tasks, but only one of the two arcs generated by the same satellite and ground antenna can be used as an effective resource. In terms of task type, there are many types of TT&C tasks, and each type of task has specific resource requirements, which makes the scheduling process complicated.

[0005] The establishment of intersatellite links between satellites gives tasks more execution opportunities and expands the understanding space. At the same time, different types of resources and tasks also increase the complexity of constraints. According to the characteristics of the communication satellite measurement and control scheduling problem, this paper establishes an integer programming model. Then, an iterative adaptive large neighborhood search algorithm (IALNS) is proposed. This algorithm expands the adaptive large neighborhood search algorithm (ALNS) and integrates the idea of ​​iterative local search algorithm (ILS) into ALNS to solve the communication satellite measurement and control scheduling problem in the low-orbit Internet constellation. Summary of the invention:

[0006] In order to solve the above technical problems, the present invention proposes an iterative adaptive large neighborhood search method for large-scale communication satellite measurement and control scheduling problems, which specifically includes the following steps:

[0007] S1. Obtain input data, including arc segment data and mission data; wherein the arc segment data is the result of satellite-to-ground link scheduling, and an arc segment can be used as an executable arc segment of a mission when it satisfies the following conditions at the same time: 1) the arc segment type matches the mission type; 2) the satellite to which the arc segment belongs is the same as the mission satellite or has an intersatellite link with the mission satellite; 3) the mission executable time is within the arc segment range;

[0008] S2. Generate three initial solutions according to the three heuristic rules, and the obtained solutions are scheduling plans, where the scheduling plans include tasks to be executed, execution arcs of the tasks, and execution time;

[0009] S3. Randomly input the three initial solutions into three threads, and run ALNS in parallel for each thread; the running steps of each thread are as follows:

[0010] S3.1. According to the weight of each operator, remove the operator and insert the operator according to the roulette wheel;

[0011] S3.2. If the conditions for activating the iterative mechanism 1 of the frequent pattern mining method are not met, enter S3.3. Otherwise, the ALNS removal process is assisted by the frequent pattern mining method. The specific steps of the iterative mechanism 1 based on the frequent pattern mining method are as follows:

[0012] S3.2.1. Data conversion: convert each partial solution task sequence in the low-quality solution set, each of which is a single task, into a sequence of two adjacent tasks;

[0013] S3.2.2. Use the FP-Growth algorithm to mine frequent patterns. Build an FP tree based on the data set. To mine frequent patterns, the algorithm traverses the paths starting from the leaf nodes and generates a conditional pattern tree. Then, frequent item sets are mined from these paths.

[0014] S3.2.3. Remove the mined maximum frequent pattern set from the solution;

[0015] S3.3. Use the removal operator to remove the solution, delete the corresponding task, and clear the low-quality solution set;

[0016] S3.4. Use the insertion operator to repair the solution and obtain a new solution, including sorting the unscheduled tasks and inserting them sequentially using the insertion algorithm;

[0017] S3.5. If the benefit value of the new solution is greater than the current solution, it is accepted; otherwise, the new solution is decided according to the simulated annealing criterion;

[0018] S3.6. If the new solution is better than the optimal solution, it will be recorded in the elite solution; if the benefit value of the new solution is more than 1% lower than the current solution, it will be recorded in the low-quality solution, and the elite solution will be continuously replaced and updated during the iteration process;

[0019] S3.6. If the termination condition is met, terminate and output the elite solution set; otherwise, return to S3.1;

[0020] S4. If the termination condition is met, it is terminated; otherwise, iterative mechanism 2 is started;

[0021] S5. After the termination condition is met, the obtained solution is used to solve the communication satellite scheduling problem and is applied to determine the executed tasks, the execution arc of the tasks and the execution time.

[0022] Furthermore, the mission data includes scenarios with different numbers of missions, where the number of missions for each satellite included in the scenarios ranges from 50 to 150, increasing in increments of 10, and all missions are randomly generated based on actual conditions.

[0023] Furthermore, the ALNS is an ALNS algorithm search process assisted by an iterative mechanism.

[0024] Furthermore, the ALNS includes an operator adaptation layer, a neighborhood operator, a new solution, and a preset iterative model processing process.

[0025] Furthermore, the preset iterative model includes recording low-quality solutions, disturbance condition judgment, data conversion, frequent pattern mining, and termination criterion judgment processes.

[0026] Furthermore, the specific steps of the iterative mechanism 2 are as follows: S4.1. The elite solution sets output by the three threads are combined for data conversion, and each solution sequence of a single task in the elite solution set is converted into a sequence of two adjacent tasks;

[0027] S4.2. Use FP-Growth algorithm to mine frequent patterns; construct FP tree based on the data set. In order to mine frequent patterns, the path will be traversed from the leaf node and conditional pattern tree will be generated, and then frequent item sets will be mined from these paths;

[0028] S4.3. Construct two new solutions based on the mined maximum frequent pattern set; first insert the tasks in the frequent pattern, and then use the optimal solution and the randomly selected solution from the elite solution set to guide the insertion process to obtain two new solutions;

[0029] S4.4. The two new solutions obtained are updated with the optimal solution as the initial solution and enter the next round of parallel ALNS together, returning to S3.

[0030] Furthermore, the deleting of the corresponding tasks is specifically to sort the tasks in the solution according to the removal operator and remove them in sequence until a specified number is reached.

[0031] According to another aspect of the present invention, there is also provided an iterative adaptive large neighborhood search system for large-scale communication satellite measurement, control and scheduling problems, the system comprising a plurality of functional modules, and the functional modules implement any of the above-described methods.

[0032] According to another aspect of the present invention, there is further provided an electronic device, the electronic device comprising:

[0033] at least one processor, and a memory communicatively coupled to at least one of the processors;

[0034] The memory stores instructions that can be executed by at least one of the processors, and the instructions are executed by at least one of the processors so that at least one of the processors can execute any of the methods described above.

[0035] According to another aspect of the present invention, a computer-readable storage medium is provided, on which computer instructions are stored, and the computer instructions are used to enable the computer to execute and implement any of the above-mentioned methods. Description of the drawings:

[0036] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0037] Figure 1 This is a flow chart of the iterative adaptive large neighborhood search algorithm in the present invention. Specific implementation method:

[0038] The following is a description of exemplary embodiments of the present application in conjunction with the accompanying drawings, including various details of the embodiments of the present application to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present application. Similarly, for the sake of clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0039] Barbulescu et al. proved that satellite telemetry scheduling on a single antenna corresponds to the traditional scheduling problem - minimizing the number of delayed jobs on a single machine. This conclusion has been extended to general telemetry problems. In order to realize the automation of telemetry scheduling, people have conducted a lot of research. Mathematical programming, metaheuristics and machine learning algorithms are the three main methods to solve this problem.

[0040] Mathematical programming algorithms can ensure the optimal solution under certain assumptions. The difficulty lies in how to use operations research knowledge to construct a linear programming model for the problem. Gooley developed a mixed integer programming model to effectively generate an initial 24-hour satellite measurement and control scheduling plan, achieving more than 91% of satellite request scheduling. Linares introduced a mixed integer linear programming model to optimize antenna-satellite scheduling (one of the types of measurement and control problems that occur in the operation of earth observation satellites) to improve efficiency by balancing time allocation and reducing conflicts.

[0041] However, the computational cost of mathematical programming is often too high to be applied to large-scale practical problems. Therefore, many researchers have begun to study meta-heuristic algorithms and machine learning algorithms to solve these problems.

[0042] Metaheuristic algorithms are guided by experience or intuition and guide the search process through neighborhood operators or evolution operators. Chen introduced a genetic algorithm with population perturbation and elimination to optimize multi-satellite measurement and control scheduling, which effectively improved profits and task completion rates. Wang proposed a multi-objective differential evolution algorithm based on space partitioning and adaptive selection strategy to solve multi-objective measurement and control problems. Machine learning algorithms aim to identify and extract hidden patterns from large data sets and are becoming increasingly popular in measurement and control problems. Ou combined deep reinforcement learning with heuristics to effectively solve the measurement and control problem. Liang integrated a mask processing mechanism in deep reinforcement learning for fast rescheduling of measurement and control problems, and demonstrated optimization performance and computational efficiency through experimental results.

[0043] Most of the research on measurement and control problems is focused on remote sensing satellites, while communication satellite constellations have received less attention. Wu considered the problem of scheduling communication tasks on intersatellite links and proved that the problem is NP-complete. Gu combined the ground measurement and control resources of the constellation with the intersatellite links and studied how to comprehensively utilize the intersatellite links and ground resources in the field of measurement and control. Although these studies include intersatellite links, they do not consider multiple types of payloads and tasks. Therefore, the present invention uniformly defines the measurement and control problems involving multiple types of payloads and tasks in intersatellite links as communication satellite measurement and control scheduling problems based on the characteristics of communication satellite constellations. In short, any measurement and control problem with a communication satellite constellation as a background or taking into account certain characteristics of a communication satellite constellation can be regarded as a communication satellite measurement and control scheduling problem and can be solved by the algorithm of the present invention.

[0044] As the algorithm in this paper, one of the characteristics of ALNS is the strategy of enhancing solutions through continuous iteration, that is, destroying part of the existing solution and then reconstructing the existing solution, with the aim of finding an improved solution in each iteration. Usually, a set of destruction and repair operators are used, and the selection of operators is adaptively determined as the algorithm iterates. In the field of satellite scheduling, Liu et al. and He et al. respectively extended and applied ALNS to single and multiple agile satellite scheduling problems. Liu et al. proposed a data-driven parallel ALNS algorithm for the inter-satellite link scheduling problem in a low-orbit communication satellite constellation. The motivation for designing an iterative mechanism in ALNS in the present invention comes from the idea of ​​ILS, which has been successfully applied to satellite scheduling problems, by providing sufficient disturbances to help the algorithm get rid of local optimality and reduce invalid searches. The iterative mechanism will start when the algorithm falls into a local optimality. The frequent pattern mining method is used in the iterative mechanism to overcome the limitations of blind search and enable the algorithm to explore a larger solution space in a directional and well-founded manner.

[0045] When a satellite orbits the earth, it will pass over the ground station many times. When the ground antenna and the satellite beam overlap each other, a satellite-to-ground link can be established between the ground antenna and the satellite. The time period for establishing the satellite-to-ground link is defined as an arc segment. In the communication satellite measurement and control scheduling problem, both the satellite and the ground antenna are equipped with two types of payloads: millimeter wave band (Q / V band) and microwave band (S band). Therefore, when the conditions for establishing a satellite-to-ground link are met, two different types of arc segments will be generated at the same time. The arc segment generated by the Q / V band is called the feeding arc segment, and the arc segment generated by the S band is called the measurement and control arc segment, such as Figure 1 As shown. Only one of the feed arc and the tracking and control arc generated simultaneously by the same satellite and the same ground antenna can be used as a valid resource, so the two are in conflict with each other. The combination and order of the satellite-to-ground links are pre-arranged, which means that the arcs are pre-generated and can be allocated for the execution of tracking and control tasks. The tasks generated by the satellite need to be executed on the satellite arcs that have intersatellite links with the satellite or with it. In short, the communication satellite tracking and control scheduling problem is to assign different types of tracking and control tasks to specific types of arcs without violating constraints and determine the execution order and time of the tasks.

[0046] In order to study the characteristics of large-scale complex communication satellite measurement and control scheduling problems, this paper introduces some reasonable assumptions to regulate this problem:

[0047] 1) The transmission of satellite data is limited to a single-layer intersatellite link, that is, the data of a satellite can only be transmitted to the satellite with which it has an intersatellite link, and no secondary transmission is allowed.

[0048] 2) The task conversion time is a fixed constant, and these constants are the maximum time required to ensure the normal operation of the equipment.

[0049] 3) Adequate amounts of power, storage, and other resources are available to accomplish the mission, regardless of unexpected circumstances.

[0050] 4) Temporary tasks are not considered. After the task starts executing, there will be no interruption or resource preemption.

[0051] According to the problem description and the above assumptions, an integer programming model is proposed. The model and related variables are defined as follows.

[0052] Variable Definition

[0053] R is the set of tasks, R = {i = 1, ..., |r}

[0054] O is a set of arc segments, O = {k = 1,...,|o}

[0055] d i Duration of task i

[0056] p i Profit of task i

[0057] The executability of task i on arc segment k. If task i can be executed on arc segment k, then otherwise

[0058] es i ,le i The earliest start time and the latest end time of task i

[0059] st k ,et k The measurement and control start time and the measurement and control end time of arc segment k

[0060] T Task transition time

[0061] k0 The conflicting arc of arc k (i.e., k and k0 are obtained by using the same antenna and the same satellite at the same time with different payloads)

[0062] M is a predetermined integer

[0063] Decision variables:

[0064] If tasks i and j are executed in arc k and j is executed immediately after i = 1, otherwise = 0 If task i is executed in arc k = 1, otherwise = 0

[0065] t i The start time of task i

[0066] Objective function:

[0067]

[0068] constraint:

[0069]

[0070] The objective function (1) represents the maximum total benefit of the scheduling task.

[0071] Constraint (2) represents the target uniqueness constraint.

[0072] Constraint (3) is a flow balance constraint and connects the decision variables and

[0073] Constraints (4) and (5) indicate that the task sequence of each arc starts from a virtual start task s and ends with a virtual end task e.

[0074] Constraint (6) represents the transition time constraint.

[0075] Constraints (7) and (8) represent arc segment time constraints.

[0076] Constraints (9) and (10) represent task time constraints.

[0077] Constraint (11) indicates that tasks can only be scheduled onto executable arcs.

[0078] Constraint (12) indicates that conflicting arcs cannot be used at the same time, that is, if a task has been executed on an arc, no task can be executed on its conflicting arc.

[0079] Constraint (13) represents the value range of the decision variable.

[0080] In the above algorithm model, there is a preprocessing process.

[0081] Preprocessing process:

[0082] The preprocessing process includes calculating the executable arc of each task. An arc can be used as an executable arc of a task if it meets the following conditions at the same time:

[0083] 1) The arc type matches the task type.

[0084] 2) The satellite to which this arc belongs is the same as the mission satellite or there is an intersatellite link between the satellite and the mission satellite.

[0085] 3) The task executable time is within the arc segment range.

[0086] In order to explain this preprocessing process simply and clearly, the present invention provides a simple example including 8 arcs and 8 tasks. If all three conditions are met, the arc is an executable arc of the task. For example, the type of arc A1 is the same as the type of task T1, the satellites of arc A1 and task T1 are both S1, and the time range of task T1 is included in the time range of arc A1. In summary, A1 meets all the necessary conditions to become an executable arc of T1.

[0087] In order to quickly construct the initial solution, the present invention adopts three heuristic rules: maximum priority, earliest start time and random construction. For each task, its executable arc is traversed, and if the task meets all the constraints, the task is inserted into the current scheduling scheme and the traversal is terminated.

[0088] The adaptive large neighborhood search algorithm is specifically:

[0089] ALNS has strong adaptability and self-calibration capabilities, and has shown strong performance in many combinatorial optimization problems. ALNS uses a pair of heuristic operators to explore the solution space by destroying and repairing the current solution respectively. This paper designs six removal operators and four insertion operators, which define how each task is arranged in the schedule.

[0090] The adaptive large neighborhood search algorithm includes: a removal operator and an insertion operator process. The adaptive large neighborhood search algorithm searches for a better solution by performing an effective search in the solution space, and its core includes destruction and repair. First, the solution is destroyed by removing the operator, and then the solution is repaired by inserting the operator to obtain a new solution. During the algorithm search process, if the new solution obtained has not been improved for many times, the first iteration mechanism is enabled to assist the removal process.

[0091] The removal operator is as follows: In order to allow new insertions and thus obtain a new scheduling solution, the removal operator must first be used to remove some tasks to destroy the current solution. The number of removed tasks is determined by the removal rate r, which represents the percentage of the number of tasks removed from the current scheduling solution to the total number of scheduled tasks.

[0092] Random removal: Randomly select tasks and remove them until the specified number is reached. This operator is the simplest removal heuristic and has a certain degree of blindness, but it helps to obtain solutions with different structures and can help the algorithm escape from the local optimum.

[0093] Priority removal: Sort candidate tasks in ascending order of priority. Tasks with lower priorities will be removed first until the specified number of removals is reached.

[0094] Execution opportunity removal: A task may have multiple executable arcs and therefore multiple execution opportunities within the scheduling scope. Tasks with more execution opportunities are removed first to ensure that tasks with less flexibility are retained. If the number of executable arcs for tasks is the same, the tasks with lower priority are removed first.

[0095] Conflict removal: The average overlap time between a task and other unscheduled tasks is defined as the task's conflict degree. Tasks with greater conflict degrees are more likely to hinder the insertion of other tasks. Therefore, tasks with greater conflict degrees are removed first.

[0096] Priority removal per unit time: Sort tasks from high to low according to the ratio of priority to their duration. Tasks with low priority and long execution time will be removed first.

[0097] Arc Priority Removal: According to Pisinger, it is more effective to remove tasks with geographical or temporal similarity. The execution time of tasks scheduled on the same arc is similar. To ensure the temporal similarity of the removed tasks, the operator selects the arc with high priority tasks and removes all tasks in the selected arc.

[0098] Specifically, the insertion operator is as follows: the insertion operator sorts the unscheduled tasks and the deleted tasks by different heuristic methods, and inserts the tasks sequentially by using an effective fast insertion algorithm.

[0099] Priority insertion: According to He, introducing randomness can diversify the insertion process: the priority is considered as a heuristic value p. By using p*(1+r) as the heuristic value (where r is a random value in the range [0,1]), the random component is added without ignoring the heuristic.

[0100] Execution opportunity insertion: This operator is used to preferentially insert tasks with fewer execution opportunities. Candidate tasks are sorted in ascending order based on the number of their executable arcs.

[0101] Conflict Insertion: This operator is similar to Conflict Removal in that tasks with less conflicts are more likely to be inserted into the current solution.

[0102] Unit time priority insertion: This operator sorts candidate tasks in descending order according to the ratio of their priority to their duration.

[0103] After sorting the candidate tasks by the above heuristic method, the fast insertion algorithm, i.e., Algorithm 1, is used to quickly insert tasks. This method traverses the executable arcs of each task until the insertion position is found. If the conflicting arc of the current arc has executed the task, the task cannot be inserted into the current arc. Otherwise, it traverses all the tasks executed in the current arc and calculates the slack time of each task based on the latest end time. If the task satisfies the constraints and the subsequent task after insertion does not violate the constraints, it can be successfully inserted.

[0104] The iterative mechanism design process includes a frequent pattern mining process, and the frequent pattern mining process includes a data conversion process, FP-Growth, and a frequent pattern mining disturbance process.

[0105] The design of the iteration mechanism is the core component of the algorithm. By mining relevant information from historical solutions, the algorithm decision is made more intelligent. Therefore, the present invention embeds the frequent pattern mining method in the iteration mechanism to improve the quality of the solution.

[0106] Frequent pattern mining process

[0107] The frequent pattern mining method can guide the removal of solutions in ALNS by mining low-quality structures from low-quality partial solutions, and can also guide the construction of new solutions by mining better structures from high-quality solutions. The implementation process of this method will be described in detail below.

[0108] 1. Transform Data

[0109] For the communication satellite tracking and control scheduling problem, the execution order of tasks greatly affects the scheduling results. Therefore, the combination of adjacent tasks is defined as the frequent pattern of the problem, named "knowledge".

[0110] 2.FP-Growth

[0111] The FP-Growth algorithm can reduce the costs associated with generating and testing candidate item sets, thereby accelerating the frequent pattern mining process. Therefore, compared with other frequent pattern mining algorithms such as Apriori, this algorithm is more suitable for large data sets.

[0112] The FP-Growth algorithm mines frequent patterns by storing the dataset in a tree structure called a frequent pattern tree. First, the frequency of each item is calculated. Subsequently, the support is calculated, which is the ratio of the number of item occurrences to the total number of items in the dataset. Items with a support greater than the minimum support s are retained. When creating the header table, the items are sorted in descending order of frequency of occurrence. Then, an FP tree is constructed based on the dataset. To mine frequent patterns, the algorithm traverses the paths starting from the leaf nodes and generates a conditional pattern tree, and then mines frequent item sets from these paths.

[0113] 3. Frequent pattern mining disturbance

[0114] If the benefit value of the new solution is more than 1% lower than that of the current solution, it is considered a low-quality solution. Tasks that exist in the new solution but not in the current solution may be the cause of the low-quality solution, and these tasks constitute low-quality partial solutions. Such low-quality partial solutions will be recorded during the local search process.

[0115] If ALNS fails to improve the solution in M1 iterations and the number of recorded partial solutions reaches m, the frequent pattern mining method described above is used to mine low-quality structures from the low-quality partial solutions. These low-quality structures are then considered as knowledge to guide the removal process. The removal ratio (denoted by R) is increased to make the search neighborhood larger. If there is a low-quality structure in the solution, it is first removed and then the removal operator is used. After each mining process, the recorded partial solutions are cleared.

[0116] 4. Frequent pattern mining builds new solutions

[0117] The frequent pattern mining method can also be used to mine other useful information, such as specific patterns that frequently appear in high-quality solutions. After each round of parallel ALNS search, the frequent pattern mining method is applied to mine subsequences of adjacent tasks that appear frequently in high-quality solutions. According to the mined patterns, two new solutions are constructed. First, the tasks in the frequent pattern are inserted, and then the optimal solution and the solution randomly selected from the elite solution set are used to guide the insertion process to obtain two new solutions. All insertion processes use the fast insertion algorithm (Algorithm 1). Finally, the two new solutions enter the next round of parallel ALNS together with the optimal solution.

[0118] 5. Data Validation

[0119] During the data verification process, 11 scenarios with different numbers of missions were constructed. As shown in Table 1, the number of missions for each satellite in these scenarios ranged from 50 to 150, increasing by 10. All missions were randomly generated based on actual conditions. The specific steps for mission generation are as follows:

[0120] Step 1: Read the arc segment data and initialize the problem parameters.

[0121] Step 2 generates n missions for each satellite.

[0122] Step 2-1 randomly generates an integer from 1 to 10 as the priority pi of task i.

[0123] In step 2-2, a random integer between 40 and 50 is generated as the duration ti of task i.

[0124] Steps 2-3 randomly generate 0 or 1 as the task type.

[0125] Steps 2-4 randomly select an arc k of the satellite. stk and etk represent the start and end time of arc k, respectively. Generate an integer in the interval [stk,etk-ti] as the earliest start time of task i, that is, esi. Generate an integer in the interval [esi+ti+10,esi+ti+30] as the latest end time lei of task i.

[0126] Step 2-5 returns to execute steps 2-1 to 2-4 in sequence to generate the next task, and repeats the process until the number of generated tasks reaches n.

[0127] Table 1 Experimental settings

[0128]

[0129] 6. Parameter settings

[0130] In IALNS, the number of solutions m for frequent pattern mining is set to 10; the minimum support s is set to 0.2; the removal rates r and R are set to 0.1 and 0.3 respectively; the annealing coefficient c is set to 0.95; the maximum number of generations of ALNS is set to 500; the fragment size of each generation of ALNS is set to 30; the number of unimproved iterations with perturbation enabled is set to 100; the maximum number of unimproved iterations is set to 1000; the parameters of operator weight update and score increment refer to the settings of Liu. The parameters of the comparison algorithm that do not exist in the algorithm of this paper are consistent with the settings in the literature.

[0131] 7. Analysis of experimental results

[0132] Scheduling results in different scenarios

[0133] The present invention selects several advanced algorithms for solving satellite scheduling problems (including measurement and control and satellite imaging scheduling problems) as comparison algorithms to demonstrate the performance of the proposed IALNS. The comparison algorithms include Cluster-based Genetic Algorithm (C-BGA), Frequent Pattern-Based Parallel Search (FPBPS) and Adaptive Large Neighbourhood Search (ALNS). The k-means clustering method is combined with the genetic algorithm.

[0134] Table 2 shows the average objective function values ​​and CPU running time of the five runs of the algorithms, where the best values ​​and running times of these algorithms are marked in bold. It can be seen from the table that IALNS outperforms other algorithms in most cases. However, IALNS is slightly inferior to FPBPS in the C1 scenario with the smallest task scale. Compared with other algorithms, C-BGA applied to the low-orbit Internet constellation system performs the worst in small-scale scenarios (C1-C4), has the longest running time, and the worst objective function value. The possible reason is that the clustering process of C-BGA is more time-consuming and the search neighborhood is more random than other algorithms. C-BGA performs better in larger-scale scenarios. In addition, as the number of tasks increases, the performance difference between IALNS and other comparison algorithms becomes large, especially compared with ALNS. Overall, the proposed IALNS has more advantages than other algorithms in solving the communication satellite measurement and control scheduling problem.

[0135] Table 2 Comparison results of IALNS, C-BGA, FPBPS, ALNS

[0136]

[0137] IALNS validity test:

[0138] Effectiveness of IALNS components

[0139] IALNS consists of two parts: ALNS and an iterative mechanism based on frequent pattern mining. The effectiveness of each component of the algorithm will be analyzed. The specific results are shown in Table 3, which lists the average values ​​of the initial solution generated by the greedy heuristic algorithm, the solution improved by ALNS, and the final solution after further improvement by the iterative mechanism. In the table, △1% and △2% represent the percentage of improvement obtained using ALNS and the iterative mechanism, respectively. In addition, the last column shows the improvement of IALNS compared to the initial solution (expressed in △%). It can be seen that the ALNS algorithm can improve the initial solution generated by the greedy heuristic algorithm by 19.54%, and then further optimize it through the iterative mechanism, and the benefit value is increased by an average of 1.82%. The IALNS algorithm can improve the initial solution by an average of 21.74%. In addition, the present invention also observes that in scenarios with larger task scales, the improvement ratio is greater, which means that the larger the task scale, the more effective ALNS and the iterative mechanism based on frequent pattern mining are.

[0140] Table 3 Detailed results of IALNS

[0141]

[0142] Effectiveness of ALNS adaptive layer and heuristic operator

[0143] Several removal and insertion heuristic operators for the ALNS framework are proposed. To demonstrate the effectiveness of the adaptive layer and each operator, the present invention considers several configurations, including removing the adaptive layer (corresponding to iterative large neighborhood search (ILNS)) and other configurations where one operator is removed. The present invention uses the scenario C11 with the largest number of tasks for testing.

[0144] The summary of these experiments is shown in the table. R1-R6 represent random removal, priority removal, execution opportunity removal, conflict removal, priority to execution time ratio removal, and priority arc removal, respectively. I1-I4 represent priority insertion, minimum execution opportunity insertion, minimum conflict insertion, and priority to execution time ratio insertion, respectively. -avg.△% represents the average percentage decrease in the benefit value under different configurations compared with the IALNS algorithm. avg.t(s) represents the average running time under different configurations.

[0145] In the first six configurations, the present invention studies the impact of the removal operator. The seventh to tenth configurations represent the impact of the insertion operator. The last configuration shows the impact of the adaptive layer. These test results show that the adaptive layer and all the removal and insertion operators perform well and play an important role in improving the performance of IALNS. The percentage drop in configuration 1 is the largest, which shows that the random removal operator is the most effective.

[0146] Table 4 Detailed configuration and experimental results

[0147]

[0148] Use Friedman test to compare algorithm performance

[0149] According to previous studies (cite{Joaquin2011,OJW2023}), the Friedman test can test the significant differences between various results at a significance level of 0.05, thereby verifying the effectiveness of the algorithm proposed in this invention. Table 5 shows the statistical results of the Friedman test of the four comparison algorithms based on the profit value of the scheduled tasks. Among the four algorithms, IALNS ranks first, followed by FPBPS. The results show that the proposed IALNS algorithm is superior to other algorithms in terms of the index values ​​evaluated by the Friedman test, and can effectively handle the communication satellite measurement and control scheduling problem.

[0150] Table 5 Ranking of four compared algorithms

[0151]

[0152] Algorithm stability analysis

[0153] Scenarios with larger task scales can better reflect the stability of the algorithm. The box plots for scenarios C10 and C11 are shown in the figure. The stability of the algorithm can be estimated by the length of the box, the difference between percentiles, the upper and lower whiskers, and whether there are outliers. It can be seen that none of these algorithms have outliers, and IALNS and C-BGA both show good stability in C10 and C11.

[0154] The present invention has the following technical effects:

[0155] Compared with other heuristic algorithms, the adaptability of the Adaptive Large Neighbourhood Search Algorithm (ALNS) and the iterative mechanism of the Iterative Local Search (ILS) algorithm can effectively prevent the algorithm from falling into the local optimum. However, the traditional iterative mechanism lacks guidance and directionality, and it is difficult to find a better solution through perturbation, especially for the extremely complex communication satellite measurement and control scheduling problem. Therefore, it is necessary to use the information of historical solutions to guide the algorithm perturbation process.

[0156] Two iterative mechanisms are embedded in parallel ALNS to provide sufficient perturbations to prevent the algorithm from falling into local optimality. Both iterative mechanisms utilize frequent pattern mining methods: one mechanism identifies specific structures that frequently appear in low-quality partial solutions to assist the ALNS removal process. The other mechanism mines better structures from the elite solutions obtained by all threads to guide the construction of new solutions, which are regarded as the initial solutions for the next round of ALNS.

[0157] Considering the satellite's on-board processing and inter-satellite forwarding capabilities as well as the matching of different types of resources and tasks, the communication constellation tracking and control scheduling problem based on the low-orbit Internet constellation is elaborated in detail. This paper conducts an in-depth discussion on this issue, which will help other researchers to conduct subsequent research on this issue.

[0158] The idea of ​​ILS is integrated into ALNS, and two iterative mechanisms are designed using the frequent pattern mining method, which mines the knowledge of the problem and intelligently guides the search process. This fast and simple algorithm framework can effectively solve the communication satellite scheduling problem in the low-orbit Internet constellation.

[0159] Based on the above content, the present invention provides an iterative adaptive large neighborhood search method for large-scale communication satellite measurement and control scheduling problems, which specifically includes the following steps:

[0160] S1. Obtain input data, including arc segment data and mission data; wherein the arc segment data is the result of satellite-to-ground link scheduling, and an arc segment can be used as an executable arc segment of a mission when it satisfies the following conditions at the same time: 1) the arc segment type matches the mission type; 2) the satellite to which the arc segment belongs is the same as the mission satellite or has an intersatellite link with the mission satellite; 3) the mission executable time is within the arc segment range;

[0161] S2. Generate three initial solutions according to the three heuristic rules, and the obtained solutions are scheduling plans, where the scheduling plans include tasks to be executed, execution arcs of the tasks, and execution time;

[0162] S3. Randomly input the three initial solutions into three threads, and run ALNS in parallel for each thread; the running steps of each thread are as follows:

[0163] S3.1. According to the weight of each operator, remove the operator and insert the operator according to the roulette wheel;

[0164] S3.2. If the conditions for activating the iterative mechanism 1 of the frequent pattern mining method are not met, enter S3.3. Otherwise, construct a new solution based on the frequent pattern mining method. The specific steps of the iterative mechanism 1 of the frequent pattern mining method are as follows:

[0165] S3.2.1. Data conversion: convert each partial solution task sequence in the low-quality solution set, each of which is a single task, into a sequence of two adjacent tasks;

[0166] S3.2.2. Use the FP-Growth algorithm to mine frequent patterns. Build an FP tree based on the data set. To mine frequent patterns, the algorithm traverses the paths starting from the leaf nodes and generates a conditional pattern tree. Then, frequent item sets are mined from these paths.

[0167] S3.2.3. Remove the mined maximum frequent pattern set from the solution;

[0168] S3.3. Use the removal operator to remove the solution, delete the corresponding task, and clear the low-quality solution set;

[0169] S3.4. Use the insertion operator to repair the solution and obtain a new solution, including sorting the unscheduled tasks and inserting them sequentially using the insertion algorithm;

[0170] S3.5. If the benefit value of the new solution is greater than the current solution, it is accepted; otherwise, the new solution is decided according to the simulated annealing criterion;

[0171] S3.6. If the new solution is better than the optimal solution, it will be recorded in the elite solution; if the benefit value of the new solution is more than 1% lower than the current solution, it will be recorded in the low-quality solution, and the elite solution will be continuously replaced and updated during the iteration process;

[0172] S3.6. If the termination condition is met, terminate and output the elite solution set; otherwise, return to S3.1;

[0173] S4. If the termination condition is met, it is terminated; otherwise, iterative mechanism 2 is started;

[0174] S5. After the termination condition is met, the obtained solution is used to solve the communication satellite scheduling problem and is applied to determine the executed tasks, the execution arc of the tasks and the execution time.

[0175] Furthermore, the mission data includes scenarios with different numbers of missions, where the number of missions for each satellite included in the scenarios ranges from 50 to 150, increasing in increments of 10, and all missions are randomly generated based on actual conditions.

[0176] Furthermore, the ALNS is an ALNS algorithm search process assisted by an iterative mechanism.

[0177] Furthermore, the ALNS includes an operator adaptation layer, a neighborhood operator, a new solution, and a preset iterative model processing process.

[0178] Furthermore, the preset iterative model includes recording low-quality solutions, disturbance condition judgment, data conversion, frequent pattern mining, and termination criterion judgment processes.

[0179] Furthermore, the specific steps of the iterative mechanism 2 are as follows: S4.1. The elite solution sets output by the three threads are combined for data conversion, and each solution sequence of a single task in the elite solution set is converted into a sequence of two adjacent tasks;

[0180] S4.2. Use FP-Growth algorithm to mine frequent patterns; construct FP tree based on the data set. In order to mine frequent patterns, the path will be traversed from the leaf node and conditional pattern tree will be generated, and then frequent item sets will be mined from these paths;

[0181] S4.3. Construct two new solutions based on the mined maximum frequent pattern set; first insert the tasks in the frequent pattern, and then use the optimal solution and the randomly selected solution from the elite solution set to guide the insertion process to obtain two new solutions;

[0182] S4.4. The two new solutions obtained are updated with the optimal solution as the initial solution and enter the next round of parallel ALNS together, returning to S3.

[0183] Furthermore, the deleting of the corresponding tasks is specifically to sort the tasks in the solution according to the removal operator and remove them in sequence until a specified number is reached.

[0184] According to another aspect of the present invention, there is also provided an iterative adaptive large neighborhood search system for large-scale communication satellite measurement, control and scheduling problems, the system comprising a plurality of functional modules, and the functional modules implement any of the above-described methods.

[0185] According to another aspect of the present invention, there is further provided an electronic device, the electronic device comprising:

[0186] at least one processor, and a memory communicatively coupled to at least one of the processors;

[0187] The memory stores instructions that can be executed by at least one of the processors, and the instructions are executed by at least one of the processors so that at least one of the processors can execute any of the methods described above.

[0188] According to another aspect of the present invention, a computer-readable storage medium is provided, on which computer instructions are stored, and the computer instructions are used to enable the computer to execute and implement any of the above-mentioned methods.

[0189] The above specific implementations do not constitute a limitation on the protection scope of this application. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of this application should be included in the protection scope of this application.

[0190] In summary, although the basic structure, principle and method of the present invention are specifically described through the above examples, it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention.

Claims

1. An iterative adaptive large neighborhood search method for large-scale communication satellite measurement and control scheduling problems, specifically comprising the following steps: S1. Obtain input data, which includes segment data and task data; wherein, The arc data is the result of satellite-to-ground link scheduling. When an arc satisfies the following conditions at the same time, it can be used as an executable arc of the task: 1) the arc type matches the task type; 2) The satellite to which the arc belongs is the same as the mission satellite or has an intersatellite link with the mission satellite; 3) The mission executable time is within the arc range; S2. Generate three initial solutions according to the three heuristic rules, and the obtained solutions are scheduling plans, where the scheduling plans include tasks to be executed, execution arcs of the tasks, and execution time; S3. Randomly input the three initial solutions into three threads, and run ALNS in parallel for each thread; the running steps of each thread are as follows: S3.

1. According to the weight of each operator, remove the operator and insert the operator according to the roulette wheel; S3.

2. If the conditions for activating the iterative mechanism 1 of the frequent pattern mining method are not met, enter S3.

3. Otherwise, the ALNS removal process is assisted by the frequent pattern mining method. The specific steps of the iterative mechanism 1 based on the frequent pattern mining method are as follows: S3.2.

1. Data conversion: convert each partial solution task sequence in the low-quality solution set, each of which is a single task, into a sequence of two adjacent tasks; S3.2.

2. Use the FP-Growth algorithm to mine frequent patterns. Build an FP tree based on the data set. To mine frequent patterns, the algorithm traverses the paths starting from the leaf nodes and generates a conditional pattern tree. Then, frequent item sets are mined from these paths. S3.2.

3. Remove the mined maximum frequent pattern set from the solution; S3.

3. Use the removal operator to remove the solution, delete the corresponding task, and clear the low-quality solution set; S3.

4. Use the insertion operator to repair the solution and obtain a new solution, including sorting the unscheduled tasks and inserting them sequentially using the insertion algorithm; S3.

5. If the benefit value of the new solution is greater than the current solution, it is accepted; otherwise, the new solution is decided according to the simulated annealing criterion; S3.

6. If the new solution is better than the optimal solution, it will be recorded in the elite solution; if the benefit value of the new solution is more than 1% lower than the current solution, it will be recorded in the low-quality solution, and the elite solution will be continuously replaced and updated during the iteration process; S3.

6. If the termination condition is met, terminate and output the elite solution set; Otherwise, return to S3.1; S4. If the termination condition is met, it is terminated; otherwise, iterative mechanism 2 is started; S5. After the termination condition is met, the obtained solution is used to solve the communication satellite scheduling problem and is applied to determine the executed tasks, the execution arc of the tasks and the execution time.

2. The method according to claim 1, characterized in that: The mission data includes scenarios with different numbers of missions. The number of missions for each satellite included in the scenarios ranges from 50 to 150, increasing in increments of 10. All missions are randomly generated based on actual conditions.

3. The method according to claim 1, characterized in that: The ALNS algorithm searches through an iterative mechanism to assist the ALNS algorithm.

4. The method according to claim 1, characterized in that: The ALNS includes an operator adaptation layer, a neighborhood operator, a new solution, and a preset iterative model processing process.

5. The method according to claim 4, characterized in that: The preset iterative model includes the processes of recording low-quality solutions, judging disturbance conditions, converting data, mining frequent patterns, and judging termination criteria.

6. The method according to claim 1, characterized in that: The specific steps of the iterative mechanism 2 are as follows: S4.

1. The elite solution sets output by the three threads are combined for data conversion, and each solution sequence of a single task in the elite solution set is converted into a sequence of two adjacent tasks; S4.

2. Use FP-Growth algorithm to mine frequent patterns; construct FP tree based on the data set. In order to mine frequent patterns, the path will be traversed from the leaf node and conditional pattern tree will be generated, and then frequent item sets will be mined from these paths; S4.

3. Construct two new solutions based on the mined maximum frequent pattern set; first insert the tasks in the frequent pattern, and then use the optimal solution and the randomly selected solution from the elite solution set to guide the insertion process to obtain two new solutions; S4.

4. The two new solutions obtained are updated with the optimal solution as the initial solution and enter the next round of parallel ALNS together, returning to S3.

7. The method according to claim 1, characterized in that: The deleting of the corresponding tasks specifically involves sorting the tasks in the solution according to the removal operator and removing them in sequence until a specified number is reached.

8. An iterative adaptive large neighborhood search system for large-scale communication satellite measurement and control scheduling problems, characterized by: The system includes multiple functional modules, and the functional modules implement the method according to any one of claims 1-7.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor, and a memory communicatively coupled to at least one of the processors; The memory stores instructions executable by at least one of the processors, and the instructions are executed by at least one of the processors so that the at least one processor can execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the method according to any one of claims 1 to 7.

Images