Genetic optimization algorithm based on distance similarity in feed satellite-ground networking planning problem

Abstract

The invention discloses a genetic optimization algorithm based on distance similarity in a feed satellite-ground networking planning problem. According to the method, a mixed integer programming model with communication satellite task income maximization is constructed for a satellite-ground networking planning problem considering a feed mode. On the basis of the characteristics of problems, a genetic optimization algorithm based on distance similarity is provided, the attribute characteristics of tasks are considered in the algorithm, and a weighted Euclidean distance calculation method is introduced to determine the similarity between the tasks. In order to obtain more high-quality solutions, different similarity evaluation methods are designed to assist the algorithm to intelligently screen individuals. In addition, the proposed method also adopts a self-adaptive crossover strategy based on a similarity mechanism to guide the algorithm to realize efficient population search. Through simulation experiment analysis, the method can effectively solve the problem of feed satellite-ground networking planning. Compared with other algorithms, the method not only obtains a higher-quality planning scheme, but also has a higher algorithm convergence speed. According to the algorithm, higher and more stable performance is realized, and the data transmission benefit of satellite ground networking is improved. In addition, the similarity evaluation method is also suitable for solving the scheduling problem of other types of communication satellites.

Description

Technical Field

[0001] The invention relates to a genetic optimization algorithm based on distance similarity in a feeding satellite-ground networking planning problem. Background Art

[0002] In recent years, with the continuous development of aerospace technology, satellites are required to communicate with remote ground stations in the fields of information communication, environmental detection, space navigation, etc. Satellite communication technology has become an important part of modern communication systems. Communication satellites have built a comprehensive communication satellite network through multiple resources such as multi-satellite coverage and ground station network construction. The advancement of communication satellite technology has not only solved the complex challenges in various fields, but also witnessed the growing demand for various communication applications in related industries, thus promoting the provision of enhanced and convenient services. In different fields, the successful application of communication satellites is inseparable from the key technology called data return. Data return technology is crucial in satellite communications because it transmits the data collected by the satellite back to the ground station, ensuring high-quality and stable communication services. Since communication satellites operate in different orbits, data return encounters various challenges during the execution of satellite missions. For example, in low orbits, the high speed of communication satellites leads to a reduction in the visible time relative to the ground, and frequent switching of communication links is required, which puts higher requirements on the timeliness of data return. Medium-orbit satellites have a larger ground coverage area than low-orbit satellites, but are more susceptible to signal transmission delay and attenuation problems. Therefore, how to obtain a reasonable return plan is particularly important for the system.

[0003] The communication satellite system can effectively allocate and distribute communication satellites according to the specific requirements of each mission, so that it can complete multiple communication tasks at the same time. In order to meet the mission requirements, the communication satellite system must seriously consider the establishment and stability of the satellite-to-ground communication link. The reliability and stability of these links directly determine the effectiveness of data transmission between the satellite and the ground station, and are the key to the successful upload or download of data. Therefore, the communication satellite system must ensure that the communication links between satellites and between satellites and the ground can be effectively established and maintained to support the smooth execution of various communication tasks. In actual scenarios, satellites can only fly around the earth in a fixed orbit, and the position of the ground station is also fixed, which strictly limits the construction time of the communication link. This means that the total length of time that can be used to build a communication link in a day is limited, and these time segments need to be fully utilized to ensure the timeliness and stability of data transmission tasks. The introduction of the feed switching working mode can effectively solve the problems of low data transmission efficiency and unstable communication connection signals. By being equipped with a dedicated feed antenna, the satellite can transmit additional energy to the satellite, and a new task can be performed immediately after completing a task. Satellites can intelligently switch links according to different needs and conditions, dynamically select the optimal communication antenna, and enable the communication system to adapt to changing communication needs and network conditions, perform communication tasks more efficiently, and provide more reliable data transmission services. However, even with the introduction of a new feed switching mode, there will still be tasks that cannot be performed due to various restrictions and constraints. In this case, how to extend the total time of the constructed communication link and enable high-quality tasks to be performed smoothly has become the optimization direction in satellite ground network planning considering the feed mode. In summary, satellite ground network planning considering the feed mode is to formulate a reasonable satellite network construction plan based on the capabilities of the satellite and the resources of the ground station. Summary of the invention

[0004] The present invention provides a solution to the problem of satellite-to-ground network planning for feeder power. Its purpose is to meet various constraints in the conventional working mode and the feeder switching mode, and to match a given series of visible time windows with satellite communication tasks, to arrange the task sequence and formulate a reasonable work plan, so as to obtain the maximum benefit of the entire task plan. A genetic optimization algorithm based on distance similarity in the problem of satellite-to-ground network planning for feeder power can solve the problems of low data transmission efficiency and unstable communication connection signals in the existing communication satellite network system.

[0005] To achieve the above object, the technical solution of the present invention is as follows:

[0006] The present invention is based on a mechanism in which similarity exists between the attribute characteristics of a communication satellite mission. In this mechanism, the attribute characteristics of the mission, such as arrival status, duration, and available benefits, will affect the final execution order of these tasks. In order to make full use of the attribute characteristics of the mission, the present invention adopts a distance similarity evaluation method and integrates it into the genetic algorithm framework. The general process of this mechanism is that before arranging the communication satellite mission sequence, all mission information needs to be preprocessed. According to the attribute characteristic information of the task itself, a weighted Euclidean distance method is introduced to calculate the distance between all tasks, thereby obtaining all similarity information between tasks. Task similarity information is the basis of similarity information between individuals in the genetic algorithm. In each iteration, task similarity information needs to be obtained, and a method based on permutation and combination is used to calculate the similarity between individuals in the population. The selection of each generation of individuals in the genetic optimization algorithm is based on individual similarity information. For dominant individuals, similar individuals are selected to mutate into the next generation; for inferior individuals, cross-perturbation processing is performed on the individuals. The iterative process is repeated until the conditions are met.

[0007] The present invention provides a crossover strategy based on similarity mechanism, and designs four different crossover operators, including two traditional crossover operators and two similarity-based crossover operators. In the iterative process of the algorithm, the performance of each crossover strategy is different at different stages, so choosing a suitable crossover strategy is crucial to improving the performance of the algorithm. To solve this problem, the adaptive crossover strategy dynamically scores each crossover operator according to its performance. These scores affect the selection of the next crossover strategy, thereby improving the search efficiency of the algorithm.

[0008] The advantages of the present invention are: on the one hand, the similarity evaluation method can take into account the relationship between communication satellite missions and better evaluate the similarity between individuals in a population, thereby guiding the genetic algorithm to explore more effectively in the search space; on the other hand, the method can use similarity information to enhance the preservation and selection of dominant individuals, so that more similar dominant individuals are concentrated in the population, thereby improving the quality of the solution and the convergence speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 A model diagram for planning the satellite-ground network for power feed;

[0010] Figure 2 Flowchart of the proposed algorithm;

[0011] Figure 3 The task benefits of the four algorithms in large-scale task examples;

[0012] Figure 4 is the average standard deviation of the performance of the four algorithms at different instance sizes;

[0013] Figure 5 To provide the profit results of the proposed algorithm and the algorithm that removes the adaptive crossover strategy in large-scale task examples; DETAILED DESCRIPTION

[0014] The present invention is further described below with reference to the accompanying drawings and examples. The embodiments of the present invention are based on the following problem model:

[0015] like Figure 1 As shown in the figure, in this model, multiple satellite and ground station resources are included. The satellite orbits the earth in a fixed orbit, and only when it reaches the receiving range of the antenna equipment of the ground station can a data transmission link be established to perform the task. This receiving range is also called the visible time window. Each satellite has two types of antennas, which can establish communication links with two types of ground stations respectively, but the antenna of a satellite can only establish a communication link with the antenna of one type of ground station. When one of the time windows meets the task execution conditions, the specific execution time of the task will be determined according to the equipment capabilities of the satellite and the ground station. After completing a task, it takes a period of equipment switching before the next task can be started. Compared with the traditional ground station antenna, the feeding ground station carrying the feeding antenna equipment also has a certain receiving range with the satellite. The receiving range between the feeding ground station and the satellite is called the feeding visible time window. The difference between them is that the feeding antenna allows a new connection to be established while a connection is established between the satellite and other ground stations. This means that there is an overlapping part in the two time windows formed by the adjacent ground station and the feeding ground station respectively. Therefore, when the selected time window cannot meet the execution time of the task, the feeding time window adjacent to or overlapping with the time window can continue to execute the task, thereby meeting the execution conditions of the task.

[0016] Assuming the problem of satellite-to-ground network planning based on feeder, the main model is represented by the following symbols. Wherein, the communication satellite is represented by S, then for each satellite s i , the antenna set is denoted as SA i ; The ground station is represented by G, then for each ground station g j , the antenna set is represented as GA j The time window between the satellite and the ground station is denoted as TW, and the feed time window is denoted as FTW. The communication task set is denoted as T. For each task t k , the task benefit is expressed as p k .

[0017] The specific steps for implementing the present invention are introduced below in conjunction with specific examples.

[0018] 1. A genetic optimization algorithm based on distance similarity in the problem of satellite-to-ground feeding network planning, characterized by comprising:

[0019] Step 1: Modeling and description of the optimization problem. The feed satellite-ground network planning problem is a combinatorial optimization problem that satisfies a series of constraints and selects a suitable time window to arrange the link pairing sequence between the satellite and the ground station in the two working modes of the traditional satellite tracking and control mode and the feed switching mode. After the main symbols are represented, the optimization goal of the problem can be expressed as:

[0020]

[0021] Among them, p k represents the revenue of the task, Indicates whether the mission can be executed within the time window and feed time window between the satellite and the ground station.

[0022] Step 2, algorithm coding. The present invention uses integer coding to generate the link construction task sequence into individual chromosomes. The genes in the chromosome correspond to the task numbers one by one. If there is a gene segment of "31425", it means that the first task of this part of the gene segment is the communication task numbered "3", the second task is the task numbered "1", followed by the tasks numbered "4, 2, 5" in sequence.

[0023] Step three, initialization. For the generation of the initial population before optimization, the present invention adopts three heuristic population initialization rules. Among them, each rule is designed to generate 20% of the population, and the individuals in the remaining 40% are completely randomly generated. Each heuristic population initialization rule adopts different prior knowledge to generate the population according to the problem. Among them, heuristic population initialization rule 1: first randomly generate an individual, and then sort them from small to large according to the earliest allowed execution time of the task; heuristic population initialization rule 2: first randomly generate an individual, and then sort them from small to large according to the latest allowed end time of the task; heuristic population initialization rule 3: first randomly generate an individual, and then sort them from small to large according to the task duration. The method of combining three heuristic population initialization rules with random methods in the present invention takes into account the principle of population diversity and ensures that the population initialization process obtains a high-quality initial solution.

[0024] Step 4, calculate the task similarity. Before the iteration begins, it is necessary to calculate the similarity information of all tasks. The present invention adopts the weighted Euclidean distance method to calculate the similarity of tasks. First, the attribute data of the task is quantified into a spatial vector, and the weight value of each attribute is obtained by the inverse variance weighting method. This weight value will be used to evaluate the importance of the task attribute to the similarity result. The higher the weight value, the greater the degree of change in the task attribute, and the greater the impact on the similarity result. The inverse variance weighting method needs to calculate the mean and variance of each group of attributes of the task, and their calculation formulas are as follows:

[0025]

[0026] in represents the average value of a task attribute state, X i Represents the current data of a task attribute status. x is the variance of the task attribute state x, x1, x2, …, x n Represents n data of a task status x.

[0027] After obtaining the variance, the inverse variance coefficient can be obtained by calculating the variance divided by the mean. The sum of their coefficients is the required weight value. For the convenience of representation, for the attributes selected in the task, 1, 2, ..., m represents all the m attributes of the task. The calculation formula for the inverse variance coefficient and weight is as follows:

[0028]

[0029] where y x is the inverse coefficient of the variance of the task attribute state x, w i is the inverse coefficient of the variance of the task attribute state i.

[0030] By introducing the obtained weight value into the process of calculating the weighted Euclidean distance, the weighted Euclidean distance of the two tasks can be calculated. This distance represents the vector distance between the corresponding attribute states of the two tasks. In order to control the value of this distance within a range, a mathematical transformation of the inverse plus one is performed on the distance result to obtain a similarity between 0 and 1 that is more accurate. The larger the similarity value, the more similar the two tasks are. The formula for weighted Euclidean distance and the calculation formula for task similarity are as follows:

[0031]

[0032] where d (x,y) is the weighted Euclidean distance between task x and task y, s (x,y) is the similarity between task x and task y.

[0033] In order to improve the search efficiency, after calculating the similarity of two tasks, they are stored in a similarity matrix, which can provide necessary information to the individual similarities in the iteration and guide the direction of population optimization.

[0034] Step 5: Calculate individual similarity. After the iteration starts, before each generation of the algorithm starts to traverse the population, it is necessary to calculate the individual similarity of all individuals in the population. The present invention uses an individual similarity calculation method based on permutations and combinations. First, expand the sequences of the two tasks, calculate the sum of the corresponding task similarities one by one, and finally divide by the number of task sequences to get the similarity between the two individuals. For example, two individuals A and B, their task sequences are A = {a1, a2, ..., a n} and B={b1,b2,...,b n}, then the similarity between A and B is calculated as follows:

[0035]

[0036] Among them, India S(A,B) is the similarity between individual A and individual B, T s is the task similarity matrix. S(A,B) The larger the value, the greater the similarity between individuals, and the more similar the individuals are.

[0037] Step 6. Select individuals for similarity evaluation. The algorithm has two ways to select individuals in each generation. One method is to use the roulette method to select an individual as a benchmark individual before traversing the population. When the benchmark individual is selected, it will serve as the basis for similarity evaluation with other individuals. The advantage of using roulette is that the benchmark individual not only has a greater probability of becoming the optimal individual, but also has a certain probability of becoming the optimal potential individual, which means that there is greater diversity. Another method is to simply traverse the population to obtain the current individual. The advantage of this selection method is that it reduces the complexity of the selection, while also being able to explore all individuals and obtain all comparison opportunities. In the initial iteration, the average value of the individual similarities is selected as the initial similarity judgment threshold, and the threshold is gradually lowered as the population iterates. The discrete function formula for lowering the threshold is as follows:

[0038]

[0039] Where x represents the number of iterations of the population, m represents the maximum number of iterations, f(x) represents the maximum number of iterations of the selection threshold of the population in each iteration, and ave x It represents the average similarity value of individuals in the xth generation, std x It represents the average similarity standard deviation of individuals in the xth generation.

[0040] At each iteration, the similarity between the benchmark individual and the current individual is compared with the similarity threshold. If the similarity is greater than the threshold, the current individual is subjected to mutation perturbation and selected to enter the next population; if the similarity is less than the threshold, the current individual is subjected to large-scale cross perturbation and selected to enter the next population.

[0041] Step seven, adaptive crossover strategy. The present invention adopts an adaptive crossover strategy, which is executed when the algorithm iterates to a certain threshold. A total of two types of crossover operators are designed, one of which includes a variety of random crossover operators, and the other is a crossover operator based on a similarity mechanism. Each type has two operators, a total of four operators. The following is a brief description of these four operators: Random crossover operator 1: randomly select two chromosome fragments from individual chromosomes, exchange the positions of the two chromosome fragments to form a new chromosome; Random crossover operator 2: randomly select a chromosome fragment from an individual chromosome, disrupt the original order and reinsert it into the corresponding position of the individual in relative order; Similarity crossover operator 1: randomly select a chromosome fragment α from the first 50% of the individual chromosome, and then randomly select several different chromosome fragments from the last 50% to find the chromosome with the smallest similarity to α, and exchange the positions of the two chromosome fragments to form a new chromosome. Similarity crossover operator 2: First, obtain the unscheduled task chromosome fragments from the task scheduling algorithm used for decoding, find the eliminated chromosome α with the highest benefit, and then randomly select several different chromosome fragments from the first 50% of the part, find the chromosome with the smallest similarity to α, and exchange the positions of the two chromosome fragments to form a new chromosome.

[0042] Based on the above four operators, the algorithm can make adaptive selections based on the search performance of the operator and the results of the offspring found. Before starting the iteration, each operator is set to an equal score, and the weight of the crossover operator is calculated based on the score. The weight calculation is calculated by the proportion of the operator score to the total operator score. Based on the obtained weight value, the roulette method is used to determine the crossover operator to be used. The population then evolves in the same way as the selected operator. After the crossover evolution, the algorithm updates the score of the crossover operator used according to the fitness after the crossover, and determines whether to update the evolved population. After the score is updated a certain number of times, the weight of the operator needs to be updated. Then the next operator selection will be completed based on the new weight value. The formula for updating the weight is as follows:

[0043]

[0044] The score i represents the score of the i-th crossover operator, w irepresents the weight of the i-th crossover operator, M represents the number of crossover operators, and in this paper M = 4. When the optimization generation reaches a certain threshold, each operator updates the operator weight according to the changed score.

[0045] Step 8: Algorithm decoding. Decoding is the process of generating a task execution plan. In the problem of power feed satellite-ground network planning, the benefit value must be calculated based on the task execution plan while satisfying the power feed switching constraints. When a chromosome sequence is given, the corresponding task execution plan generation will be completed one by one in order from front to back. The decoding process is completed when all tasks have been scheduled. Specifically, first a task searches a time window to see if it can be executed in sequence, and two new variables aest are introduced before scheduling. i and alet i , respectively represent the earliest actual available time and the latest actual end time. The calculation method is shown in the formula.

[0046] aest k =max{est k , tws j}

[0047] alet k = min{let k ,twe j}

[0048] If the length of a time window exceeds the time required by the task, the task will try to schedule in the current time window. If the actual available time range exceeds the time range required by the task, it can be successfully scheduled. After the task is successfully scheduled, it needs to be inserted into the time window according to the earliest actual time and actual end time of the task, and the time window needs to be trimmed. After the trimming process is completed, the time window set will be updated.

[0049] All the steps of the present invention are as follows: Figure 2 Get a clear presentation;

[0050] 2. The following will be combined with the attached figure of a genetic optimization algorithm based on distance similarity in a feed satellite-to-ground network planning problem to analyze the experimental results in detail:

[0051] In order to further analyze the effect of the present invention, a test example of the algorithm of the present invention according to the communication satellite mission is given here based on the problem of feed satellite-ground networking planning. In this example, the relevant parameters under the scenario are set as follows: the time period of the scenario is 24 hours, and the number of tasks is 100 to 1000. The time required for the measurement and control task follows a normal distribution with a mean of 55 and a standard deviation of 45. The task profit follows a uniform distribution with a mean of 10 and a standard deviation of 9. According to the number of instances, it is divided into three types: small-scale, medium-scale and large-scale. The small-scale instance is set between 100 and 300, the medium-scale instance is set between 400 and 600, and the large-scale instance is set between 700 and 1000. At the same time, all algorithms are set to terminate after evaluating the fitness 5000 times, the population size of the algorithm is set to 10, the fragment length is set to 2, and the maximum length of the buffer pool is set to 20. The probability of the adaptive crossover strategy is set to 0.8, the threshold is 20, and the score is 50, 30, 10, and 5 according to the performance bonus. The parameter settings of the comparison algorithm are consistent with the settings of the present invention. To ensure fairness, the algorithm was run 10 times independently for each parameter setting.

[0052] The experimental results are as follows Figure 3 As shown in the figure, the algorithm results of 500-1 and 1000-1 run 10 times. It can be seen from the results that as the task scale increases, the profit gap between the algorithm proposed in the present invention and other comparison algorithms also increases. Under large-scale examples, the algorithm of the present invention obtains the maximum detection profit in most cases. There is a clear gap between the algorithm and other algorithms, which are genetic optimization algorithm based on distance similarity, artificial bee colony algorithm, knowledge-based genetic algorithm, and improved adaptive large neighborhood search algorithm from best to worst.

[0053] Figure 4 The average standard deviation of the performance of different instance sizes is shown. It can be seen that the algorithm proposed in the present invention is also very good in terms of stability. It not only has a low standard deviation in most cases, but also has very stable search volatility. The experimental results show that better optimization performance can be obtained compared to the comparison algorithm.

[0054] The present invention also tests whether the adaptive crossover strategy based on similarity improvement can play a role in the proposed algorithm. An example with a task scale of 1000 is selected to compare the proposed algorithm with the algorithm without the adaptive crossover strategy. The average result of the benefit is as follows Figure 5 As shown, it can be seen that the algorithm containing the adaptive crossover strategy can always obtain higher profits. Therefore, this shows that the adaptive crossover strategy plays a vital role in the iterative search process of.

[0055] It can be seen from the above experimental results that the algorithm can effectively solve the scheduling problem in large-scale mission scenarios. Heuristic rules, similarity mechanism, and adaptive cross optimization search effectively improve the search efficiency of the genetic algorithm, and the scheduling results are more ideal. The similarity-based genetic optimization algorithm proposed in the present invention has achieved good optimization performance in solving practical engineering problems including the studied feeder satellite-ground networking planning problem, and provides a new idea for solving optimization problems in the future.

Claims

1. A genetic optimization algorithm based on distance similarity in the problem of satellite-to-ground feeding network planning, characterized in that: include: Step 1: Modeling and description of the optimization problem. The feed satellite-ground network planning problem is a combinatorial optimization problem that satisfies a series of constraints and selects a suitable time window to arrange the link pairing sequence between the satellite and the ground station in the two working modes of the traditional satellite tracking and control mode and the feed switching mode. After the main symbols are represented, the optimization goal of the problem can be expressed as: Among them, p k represents the revenue of the task, Indicates whether the mission can be executed within the time window and feed time window between the satellite and the ground station. Step 2, algorithm coding. The present invention uses integer coding to generate the link construction task sequence into individual chromosomes. The genes in the chromosome correspond to the task numbers one by one. If there is a gene segment of "31425", it means that the first task of this part of the gene segment is the communication task numbered "3", the second task is the task numbered "1", followed by the tasks numbered "4, 2, 5" in sequence. Step three, initialization. For the generation of the initial population before optimization, the present invention adopts three heuristic population initialization rules. Among them, each rule is designed to generate 20% of the population, and the individuals in the remaining 40% are completely randomly generated. Each heuristic population initialization rule adopts different prior knowledge to generate the population according to the problem. Among them, heuristic population initialization rule 1: first randomly generate an individual, and then sort them from small to large according to the earliest allowed execution time of the task; heuristic population initialization rule 2: first randomly generate an individual, and then sort them from small to large according to the latest allowed end time of the task; heuristic population initialization rule 3: first randomly generate an individual, and then sort them from small to large according to the task duration. The method of combining three heuristic population initialization rules with random methods in the present invention takes into account the principle of population diversity and ensures that the population initialization process obtains a high-quality initial solution. Step 4, calculate the task similarity. Before the iteration begins, it is necessary to calculate the similarity information of all tasks. The present invention uses the weighted Euclidean distance method to calculate the similarity of tasks. First, the attribute data of the task is quantified into a spatial vector, and each attribute weight value is obtained by the inverse variance weighting method. This weight value will be used to evaluate the importance of the task attribute to the similarity result. The higher the weight value, the greater the degree of change in the task attribute, and the greater the impact on the similarity result. The inverse variance weighting method requires calculating the mean and variance of each set of attributes of the task. Their calculation formulas are as follows: in represents the average value of a task attribute state, X i Represents the current data of a task attribute status. x is the variance of the task attribute state x, x1, x2, …, x n Represents n data of a task status x. After obtaining the variance, the inverse variance coefficient can be obtained by calculating the variance divided by the mean. The sum of their coefficients is the required weight value. For the convenience of representation, for the attributes selected in the task, 1, 2, ..., m represents all the m attributes of the task. The calculation formula for the inverse variance coefficient and weight is as follows: where y x is the inverse coefficient of the variance of the task attribute state x, w i is the inverse coefficient of the variance of the task attribute state i. By introducing the obtained weight value into the process of calculating the weighted Euclidean distance, the weighted Euclidean distance of the two tasks can be calculated. This distance represents the vector distance between the corresponding attribute states of the two tasks. In order to control the value of this distance within a range, a mathematical transformation of the inverse plus one is performed on the distance result to obtain a similarity between 0 and 1 that is more accurate. The larger the similarity value, the more similar the two tasks are. The formula for weighted Euclidean distance and the calculation formula for task similarity are as follows: where d (x,y) is the weighted Euclidean distance between task x and task y, s (x,y) is the similarity between task x and task y. In order to improve the search efficiency, after calculating the similarity of two tasks, they are stored in a similarity matrix, which can provide necessary information to the individual similarities in the iteration and guide the direction of population optimization. Step 5: Calculate individual similarity. After the iteration starts, before each generation of the algorithm starts to traverse the population, it is necessary to calculate the individual similarity of all individuals in the population. The present invention uses an individual similarity calculation method based on permutations and combinations. First, the sequences of the two tasks are expanded, and the corresponding task similarities are calculated one by one. Their sum values ​​are finally divided by the number of task sequences to obtain the similarity between the two individuals. For example, two individuals A and B have task sequences A = {a1, a2, ..., a n } and B={b1,b2,...,b n }, then the similarity between A and B is calculated as follows: Among them, India S(A,B) is the similarity between individual A and individual B, T s is the task similarity matrix. S(A,B) The larger the value, the greater the similarity between individuals, and the more similar the individuals are. Step 6: Select individuals for similarity evaluation. The algorithm has two ways to select individuals in each generation. One method is to use the roulette wheel method to select an individual as a benchmark individual before traversing the population. When the benchmark individual is selected, it will serve as the basis for evaluating the similarity with other individuals. The advantage of using a roulette wheel is that the benchmark individual not only has a greater probability of becoming the optimal individual, but also has a certain probability of becoming the optimal potential individual, which means that there is greater diversity. Another method is to simply traverse the population to obtain the current individual. The advantage of this selection method is that it reduces the complexity of the selection, while also being able to explore all individuals and obtain all comparison opportunities. In the initial iteration, the average value of the individual similarities is selected as the initial similarity judgment threshold, and the threshold is gradually lowered as the population iterates. The discrete function formula for lowering the threshold is as follows: Where x represents the number of iterations of the population, m represents the maximum number of iterations, f(x) represents the maximum number of iterations of the selection threshold of the population in each iteration, and ave x It represents the average similarity value of individuals in the xth generation, std x It represents the average similarity standard deviation of individuals in the xth generation. At each iteration, the similarity between the benchmark individual and the current individual is compared with the similarity threshold. If the similarity is greater than the threshold, the current individual is subjected to mutation perturbation and selected to enter the next population; if the similarity is less than the threshold, the current individual is subjected to large-scale cross perturbation and selected to enter the next population. Step seven, adaptive crossover strategy. The present invention adopts an adaptive crossover strategy, which is executed when the algorithm iterates to a certain threshold. A total of two types of crossover operators are designed, one of which includes a variety of random crossover operators, and the other is a crossover operator based on a similarity mechanism. Each type has two operators, a total of four operators. The following is a brief description of these four operators: Random crossover operator 1: randomly select two chromosome fragments from individual chromosomes, exchange the positions of the two chromosome fragments to form a new chromosome; Random crossover operator 2: randomly select a chromosome fragment from an individual chromosome, disrupt the original order and reinsert it into the corresponding position of the individual in relative order; Similarity crossover operator 1: randomly select a chromosome fragment α from the first 50% of the individual chromosome, and then randomly select several different chromosome fragments from the last 50% to find the chromosome with the smallest similarity to α, and exchange the positions of the two chromosome fragments to form a new chromosome. Similarity crossover operator 2: First, obtain the unscheduled task chromosome fragments from the task scheduling algorithm used for decoding, find the eliminated chromosome α with the highest benefit, and then randomly select several different chromosome fragments from the first 50% of the part, find the chromosome with the smallest similarity to α, and exchange the positions of the two chromosome fragments to form a new chromosome. Based on the above four operators, the algorithm can make adaptive selections based on the search performance of the operator and the results of the offspring found. Before starting the iteration, each operator is set to an equal score, and the weight of the crossover operator is calculated based on the score. The weight calculation is calculated by the proportion of the operator score to the total operator score. Based on the obtained weight value, the roulette method is used to determine the crossover operator to be used. The population then evolves in the same way as the selected operator. After the crossover evolution, the algorithm updates the score of the crossover operator used according to the fitness after the crossover, and determines whether to update the evolved population. After the score is updated a certain number of times, the weight of the operator needs to be updated. Then the next operator selection will be completed based on the new weight value. The formula for updating the weight is as follows: The score i represents the score of the i-th crossover operator, w i represents the weight of the i-th crossover operator, M represents the number of crossover operators, and in this paper M = 4. When the optimization generation reaches a certain threshold, each operator updates the operator weight according to the changed score. Step 8: Algorithm decoding. Decoding is the process of generating a task execution plan. In the problem of power feed satellite-ground network planning, the benefit value must be calculated based on the task execution plan while satisfying the power feed switching constraints. When a chromosome sequence is given, the corresponding task execution plan generation will be completed one by one in order from front to back. The decoding process is completed when all tasks have been scheduled. Specifically, first a task searches a time window to see if it can be executed in sequence, and two new variables aest are introduced before scheduling. i and alet i , respectively represent the earliest actual available time and the latest actual end time. The calculation method is shown in the formula. east k =max{is k ,tws j } ale k =min{let k ,two j } If the length of a time window exceeds the time required by the task, the task will try to schedule in the current time window. If the actual available time range exceeds the time range required by the task, it can be successfully scheduled. After the task is successfully scheduled, it needs to be inserted into the time window according to the earliest actual time and actual end time of the task, and the time window needs to be trimmed. After the trimming process is completed, the time window set will be updated.

Images