Large-scale satellite measurement and control scheduling method and system based on self-organizing learning algorithm
By applying self-organized learning algorithms and deep reinforcement learning in large-scale satellite measurement and control scheduling, the complexity and solution difficulty of low-orbit satellite measurement and control task scheduling are solved, and efficient task scheduling and resource optimization are achieved.
Patent Information
- Application Number
- CN202411677223.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-22
- Publication Date
- 2025-03-21
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Large-scale low-orbit satellite measurement and control mission scheduling faces complex task modeling and solution difficulties caused by increasing scale, and it is difficult for existing technology to effectively achieve efficient linkage between satellites and ground stations and resource optimization.
Using a method based on self-organized learning algorithm, a large-scale satellite measurement and control scheduling integer planning model is constructed, and genetic algorithm parameters are trained through deep reinforcement learning algorithms to generate the optimal satellite measurement and control scheduling scheme to maintain population diversity, avoid local optimal solutions, and improve search robustness and resilience.
It realizes efficient satellite measurement and control scheduling, improves the ability to maximize task benefits, enhances the processing ability of modeling complex tasks and solves large-scale task scheduling.
Smart Images
Figure CN119669894A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of satellite scheduling, and in particular relates to a large-scale satellite measurement and control scheduling method and system based on a self-organizing learning algorithm. Background Art
[0002] Satellites can be divided into synchronous orbit satellites, medium orbit satellites and low orbit satellites according to their operating altitude. Low orbit satellites are about 500 to 2000 km from the ground. Low orbit satellites have the characteristics of low altitude, low latency, low path loss and low transmission power. At present, there are more and more low-orbit satellites in orbit, and these low-orbit satellites together constitute a new generation of satellite networks. After low-orbit satellites constitute a new generation of networks, there will be more and more satellite mission requirements. However, a large number of low-orbit satellites require the cooperation of ground stations and other measurement and control equipment to jointly complete satellite missions with different requirements. Due to the limitations of ground station measurement and control resources, it is necessary to reasonably arrange these ground station resources to ensure the smooth implementation of satellite telemetry and remote control tasks.
[0003] The low-orbit satellite measurement and control mission is that different user units need to use satellites to complete communication or monitoring requests (the Internet constellation has many tasks). The satellite control department pre-processes the task requirements, including task analysis and decomposition. Then, the task scheduling plan is made through the scheduling algorithm. The purpose of the satellite measurement and control center to make plans is to inject data into the generated satellite requests through the ground station antenna in the form of instructions and send them to the satellite platform and application system. The satellite measurement and control center pre-processes the proposed measurement and control resource applications, analyzes the characteristics of the task requests and the relationship between the tasks, and considers the timing of task injection, constraints and other conditions. The specific process is that the satellite measurement and control center uses the satellite measurement and control scheduling model to formulate a reasonable satellite measurement and control scheduling plan based on the usage status and resource status of the station network, and determines the time window for the satellite task to be executed on the ground station antenna. When the satellite completes the task, the data obtained by the satellite needs to be transmitted to the ground through the ground station antenna. The satellite measurement and control center needs to make a telemetry task plan to receive the satellite data to the ground. At this time, the satellite needs to select a suitable ground station antenna and determine the specific start and end time to receive satellite data. When making a measurement and control plan, the satellite measurement and control center needs to consider the status and capabilities of the ground station antenna and use the measurement and control scheduling algorithm to formulate an appropriate telemetry plan to complete the reception of satellite data. After receiving the satellite data, the satellite measurement and control center organizes the received results and sends the final results to the user demand department in a certain data format.
[0004] However, the scheduling of large-scale satellite tracking and control tasks still faces huge challenges:
[0005] (1) The modeling of low-orbit satellite measurement and control tasks is more complicated. The satellite measurement and control scheduling process is easily affected by the geographical location of the ground station. Due to the limitations of the regional distribution of ground stations and the sharp increase in the number of satellites, it is difficult to establish an effective link between the satellite and the ground station antenna. Automatic tuning is performed for specific problems, and autonomous learning is performed during the process until the algorithm configuration that best suits the current scenario is found. Although many parameterized algorithms provide theoretical worst and average bounds for their performance, their actual performance in practice is highly sensitive to parameter settings. In order to establish satellite communications between satellites and ground stations, establishing communications through intersatellite links is the key to solving satellite measurement and control scheduling problems. However, after using intersatellite link technology, the satellite lacks a visible time window and needs to transmit information through intermediate satellites to achieve the purpose of rapid measurement and control. Mission modeling needs to consider more constraints, making the model more complex.
[0006] (2) The increase in the scale of low-orbit satellite measurement and control tasks increases the difficulty of solving task scheduling. Due to the significant increase in the scale of low-orbit satellite measurement and control tasks, the efficiency of artificial empirical trial and error is too low, and the scale of the solution space for solving satellite measurement and control tasks has increased. With the excellent learning ability of artificial intelligence technology, it is necessary to study the self-learning and self-improvement technology of task planning algorithms to realize feature mining and task scheduling of cross-domain heterogeneous constellation resources and joint tasks, and to automatically realize algorithm evolution. Summary of the invention
[0007] In view of the above-mentioned technical problems, the purpose of the present invention is to provide a large-scale satellite measurement and control scheduling method and system based on a self-organizing learning algorithm. The method constructs a large-scale satellite measurement and control scheduling integer programming model, trains genetic algorithm parameters through a deep reinforcement learning algorithm, and uses the optimal genetic algorithm parameters to generate a satellite measurement and control scheduling plan. It can maintain the diversity of the population, avoid falling into a local optimal solution, and improve the robustness and resilience of the search.
[0008] The technical solution of the present invention is:
[0009] A large-scale satellite measurement and control scheduling method based on a self-organizing learning algorithm comprises the following steps:
[0010] S01: Establish a large-scale satellite measurement and control scheduling model, wherein the optimization goal of the large-scale satellite measurement and control scheduling model is to maximize the mission benefit;
[0011] S02: Generate a measurement and control scheduling plan using a rule-based algorithm;
[0012] S03: When the termination condition is not met, calculate the benefit and update the state at the next moment, calculate the strategy probability, generate an action selection operator based on the strategy probability, use the selection operator to generate a new measurement and control scheduling plan, repeat the process until the termination condition is met and output the final plan.
[0013] In the preferred technical solution, the optimization goal of the large-scale satellite measurement and control scheduling model is:
[0014]
[0015] The constraints are as follows:
[0016] 1) Target uniqueness constraint:
[0017]
[0018] 2) Flow balance constraints, connecting decision variables and
[0019]
[0020] 3) The task sequence of each round starts from the virtual starting task s and ends with the virtual ending task e:
[0021]
[0022]
[0023] 4) Conversion time constraints:
[0024]
[0025] 5) Arc segment time constraints:
[0026]
[0027]
[0028] 6) Task time constraints:
[0029]
[0030]
[0031] 7) Tasks can only be scheduled on executable arcs:
[0032]
[0033] 8) Conflicting arcs cannot be used at the same time, that is, if a task has been executed on an arc, the conflicting arc cannot execute any task:
[0034]
[0035] in, Whether tasks i and j are both executed in arc segment k, is whether the virtual starting task s and task j are both executed in arc segment k, is whether task j and virtual termination task e are both executed in arc segment k, is whether task i is executed in arc segment k, T is the task set, T = {i = 1, ... T,, O is the arc segment set, O = {k = 1, ..., o}, d i is the duration of task i, p i is the benefit of task i, is the visibility of task i in arc segment k, Indicates that task i is visible in arc segment k, otherwise it is 0, es i ,le i is the earliest start time and the latest end time of task i, st k ,et k is the measurement and control start time and measurement and control end time of arc segment k, k0 is the conflicting arc segment of arc segment k, t i ,t j are the observation start time of tasks i and j respectively, and M represents the coefficient.
[0036] In the preferred technical solution, in step S03, a deep reinforcement learning algorithm is used to select a suitable action, and the selected action At is used to generate a new measurement and control scheduling plan. The action includes the crossover operation (Pt-1, At, Oc, α) and mutation operation (Pt-1, At, Om, β) in the genetic algorithm, where Pt-1 is the population at time t-1, Oc is the crossover operator, Om is the mutation operator, α is the crossover probability, and β is the mutation probability.
[0037] The preferred technical solution also includes performing fitness evaluation based on the obtained population Pt, and calculating the reward based on the generated measurement and control scheduling scheme benefit value, and the reward Rt = f t+1 –f t .
[0038] In the preferred technical solution, the softmax Greedy strategy is used to calculate the strategy probability.
[0039] The present invention also discloses a large-scale satellite measurement, control and dispatching system based on a self-organizing learning algorithm, comprising:
[0040] An inter-satellite link satellite measurement and control scheduling model building module is used to establish a large-scale satellite measurement and control scheduling model, wherein the optimization goal of the large-scale satellite measurement and control scheduling model is to maximize the mission benefit;
[0041] The measurement and control scheduling plan generation module generates a measurement and control scheduling plan using a rule-based algorithm;
[0042] Iterative optimization module, when the termination condition is not met, calculates the benefit and updates the state at the next moment, calculates the strategy probability, and generates an action selection operator based on the strategy probability; uses the selection operator to generate a new measurement and control scheduling plan, and repeats the process until the termination condition is met and the final plan is output.
[0043] In the preferred technical solution, the optimization goal of the large-scale satellite measurement and control scheduling model is:
[0044]
[0045] The constraints are as follows:
[0046] 1) Target uniqueness constraint:
[0047]
[0048] 2) Flow balance constraints, connecting decision variables and
[0049]
[0050] 3) The task sequence of each round starts from the virtual starting task s and ends with the virtual ending task e:
[0051]
[0052]
[0053] 4)Conversion time constraints:
[0054]
[0055] 5) Arc segment time constraints:
[0056]
[0057]
[0058] 6) Task time constraints:
[0059]
[0060]
[0061] 7) Tasks can only be scheduled on executable arcs:
[0062]
[0063] 8) Conflicting arcs cannot be used at the same time, that is, if a task has been executed on an arc, the conflicting arc cannot execute any task:
[0064]
[0065] in, Whether tasks i and j are both executed in arc segment k, is whether the virtual starting task s and task j are both executed in arc segment k, is whether task j and virtual termination task e are both executed in arc segment k, is whether task i is executed in arc segment k, T is the task set, T = {i = 1, ... T,, O is the arc segment set, O = {k = 1, ..., o}, d i is the duration of task i, p i is the benefit of task i, is the visibility of task i in arc segment k, Indicates that task i is visible in arc segment k, otherwise it is 0, es i ,le i is the earliest start time and the latest end time of task i, st k ,et k is the measurement and control start time and measurement and control end time of arc segment k, k0 is the conflicting arc segment of arc segment k, t i ,t j are the observation start time of tasks i and j respectively, and M represents the coefficient.
[0066] In the preferred technical solution, a deep reinforcement learning algorithm is used in the iterative optimization module to select appropriate actions, and the selected action At is used to generate a new measurement and control scheduling plan. The actions include the crossover operation (Pt-1, At, Oc, α) and mutation operation (Pt-1, At, Om, β) in the genetic algorithm, where Pt-1 is the population at time t-1, Oc is the crossover operator, Om is the mutation operator, α is the crossover probability, and β is the mutation probability.
[0067] The preferred technical solution also includes performing fitness evaluation based on the obtained population Pt, and calculating the reward based on the generated measurement and control scheduling scheme benefit value, and the reward Rt = f t+1 –f t .
[0068] The present invention further discloses a computer storage medium on which a computer program is stored. When the computer program is executed, the large-scale satellite measurement, control and scheduling method based on the self-organizing learning algorithm is implemented.
[0069] Compared with the prior art, the present invention has the following beneficial effects:
[0070] This method constructs a large-scale satellite measurement and control scheduling integer programming model, trains the genetic algorithm parameters through a deep reinforcement learning algorithm, and uses the optimal genetic algorithm parameters to generate a satellite measurement and control scheduling plan. It can maintain the diversity of the population, avoid falling into local optimal solutions, and improve the robustness and resilience of the search.
[0071] Efficiency: Use machine learning methods to train crossover and mutation operations to generate the optimal measurement and control scheduling plan. Use the deep reinforcement learning algorithm DRL to accelerate convergence during the optimization process and reduce the number of evaluations. BRIEF DESCRIPTION OF THE DRAWINGS
[0072] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:
[0073] Figure 1 Flow chart of the large-scale satellite measurement and control scheduling method based on the self-organizing learning algorithm of this embodiment;
[0074] Figure 2 This is a block diagram of a large-scale satellite measurement, control and scheduling system based on a self-organizing learning algorithm in this embodiment. DETAILED DESCRIPTION
[0075] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention is further described in detail below in conjunction with specific embodiments and with reference to the accompanying drawings. It should be understood that these descriptions are only exemplary and are not intended to limit the scope of the present invention. In addition, in the following description, the description of well-known structures and technologies is omitted to avoid unnecessary confusion of the concept of the present invention.
[0076] Example
[0077] like Figure 1 As shown, a large-scale satellite measurement and control scheduling method based on a self-organizing learning algorithm includes the following steps:
[0078] S01: Establish a large-scale satellite measurement and control scheduling model, wherein the optimization goal of the large-scale satellite measurement and control scheduling model is to maximize the mission benefit;
[0079] S02: Generate a measurement and control scheduling plan using a rule-based algorithm;
[0080] S03: When the termination condition is not met, calculate the benefit and update the state at the next moment, calculate the strategy probability, generate an action selection operator based on the strategy probability, use the selection operator to generate a new measurement and control scheduling plan, repeat the process until the termination condition is met and output the final plan.
[0081] In a preferred embodiment, the optimization goal of the large-scale satellite measurement and control scheduling model is:
[0082]
[0083] The constraints are as follows:
[0084] 1) Target uniqueness constraint:
[0085]
[0086] 2) Flow balance constraints, connecting decision variables and
[0087]
[0088] 3) The task sequence of each round starts from the virtual starting task s and ends with the virtual ending task e:
[0089]
[0090]
[0091] 4)Conversion time constraints:
[0092]
[0093] 5) Arc segment time constraints:
[0094]
[0095]
[0096] 6) Task time constraints:
[0097]
[0098]
[0099] 7) Tasks can only be scheduled on executable arcs:
[0100]
[0101] 8) Conflicting arcs cannot be used at the same time, that is, if a task has been executed on an arc, the conflicting arc cannot execute any task:
[0102]
[0103] in, Whether tasks i and j are both executed in arc segment k, is whether the virtual starting task s and task j are both executed in arc segment k, is whether task j and virtual termination task e are both executed in arc segment k, is whether task i is executed in arc segment k, T is the task set, T = {i = 1, ... T,, O is the arc segment set, O = {k = 1, ..., o}, d i is the duration of task i, p i is the benefit of task i, is the visibility of task i in arc segment k, Indicates that task i is visible in arc segment k, otherwise it is 0, es i ,le i is the earliest start time and the latest end time of task i, st k ,et k is the measurement and control start time and measurement and control end time of arc segment k, k0 is the conflicting arc segment of arc segment k, t i ,t j are the observation start time of tasks i and j respectively, and M represents the coefficient.
[0104] In a preferred embodiment, in step S03, a deep reinforcement learning algorithm is used to select a suitable action, and the selected action At is used to generate a new measurement and control scheduling plan. The action includes the crossover operation (Pt-1, At, Oc, α) and the mutation operation (Pt-1, At, Om, β) in the genetic algorithm, where Pt-1 is the population at time t-1, Oc is the crossover operator, Om is the mutation operator, α is the crossover probability, and β is the mutation probability.
[0105] A preferred embodiment further includes performing fitness evaluation based on the obtained population Pt, and calculating the reward based on the generated measurement and control scheduling scheme benefit value, the reward Rt = f t+1 –f t .
[0106] A preferred embodiment uses the softmax Greedy strategy to calculate the policy probability.
[0107] Another embodiment is a computer storage medium having a computer program stored thereon, wherein when the computer program is executed, the large-scale satellite measurement and control scheduling method based on the self-organizing learning algorithm is implemented. The specific method is consistent with the large-scale satellite measurement and control scheduling method based on the self-organizing learning algorithm described above, and will not be repeated here.
[0108] Another embodiment, such as Figure 2 As shown, a large-scale satellite measurement, control and scheduling system based on a self-organizing learning algorithm includes:
[0109] An inter-satellite link satellite measurement and control scheduling model building module 10 is used to establish a large-scale satellite measurement and control scheduling model, wherein the optimization goal of the large-scale satellite measurement and control scheduling model is to maximize the mission benefit;
[0110] The measurement and control scheduling plan generation module 20 generates a measurement and control scheduling plan using a rule-based algorithm;
[0111] Iterative optimization module 30, when the termination condition is not met, calculates the benefit and updates the state at the next moment, calculates the strategy probability, and generates an action selection operator based on the strategy probability; uses the selection operator to generate a new measurement and control scheduling plan, and repeats the process until the termination condition is met and then outputs the final plan.
[0112] The following is a detailed description of the workflow of a large-scale satellite measurement and control scheduling system based on a self-organizing learning algorithm using an optimal example:
[0113] To address the problem of lack of standardized mathematical models and reinforcement learning optimization algorithms for inter-satellite link satellite measurement and control scheduling, 1) an integer programming model for inter-satellite link satellite measurement and control scheduling was established, and 2) a self-organizing learning algorithm was provided.
[0114] 1. Establishment of large-scale satellite measurement and control scheduling model based on self-organizing learning algorithm:
[0115] Task input:
[0116] The variables and symbols involved in the large-scale satellite measurement and control scheduling model are shown in Table 1.
[0117] Table 1 Symbols and variables for large-scale satellite measurement and control scheduling problem
[0118]
[0119] Task Output
[0120]
[0121]
[0122] t i : Integer variable, indicating the observation start time of task i.
[0123] Task Constraints
[0124] 1) Target uniqueness constraint:
[0125]
[0126] 2) Flow balance constraints, connecting decision variables and
[0127]
[0128] 3) The task sequence of each round starts from the virtual starting task s and ends with the virtual ending task e:
[0129]
[0130]
[0131] 4) Conversion time constraints:
[0132]
[0133] 5) Arc segment time constraints:
[0134]
[0135]
[0136] 6) Task time constraints:
[0137]
[0138]
[0139] 7) Tasks can only be scheduled on executable arcs:
[0140]
[0141] 8) Conflicting arcs cannot be used at the same time, that is, if a task has been executed on an arc, the conflicting arc cannot execute any task:
[0142]
[0143] Optimization goals
[0144] Maximize mission benefits
[0145] The algorithm of the self-organizing learning algorithm that combines machine learning with genetic algorithm:
[0146] First, initialize the reinforcement learning parameters and experimental parameters. Then, use the rule-based algorithm to generate a scheduling plan. When the termination condition is not met, calculate the benefit and update the state at the next moment. Use the softmax Greedy strategy to calculate the policy probability and generate an action selection operator based on the policy probability. Use the selected operator to generate a new plan. Repeat this process until the termination condition is met and output the final plan.
[0147] Preferably, the deep reinforcement learning algorithm DRL model is modeled as a Markov decision process, which is a mathematical model of sequential decision-making, used to simulate the stochastic strategies and rewards that can be achieved by intelligent agents in an environment where the system state has Markov properties, mainly including state characteristics, rewards, actions and other parts.
[0148] State features: The state representation needs to provide enough information so that the neural network can decide which action is more appropriate at the current step. In the proposed algorithm, the current state needs to encode the information of the current solution and the difference in the benefits of NP solutions.
[0149] Reward: Reward is an evaluation of the performance of the agent after taking an action. Reward can reflect whether the agent's decision is correct and the impact of the decision. The reward calculation is determined based on the generated task measurement and control solution benefit value. The reward value calculation formula used in this article is shown below.
[0150] Rt=f t+1 –f t
[0151] Action: The action definitions of this model are divided into two categories, namely selection crossover and mutation operations.
[0152] The algorithm flow is as follows:
[0153] Input: population size Np, deep reinforcement learning algorithm DRL, crossover operator Oc, mutation operator Om, crossover probability α, mutation probability β;
[0154] Output: measurement and control scheduling plan;
[0155] Initialize algorithm parameters;
[0156] Set t←0;
[0157] Generate initial state S0;
[0158] P0←Generate the initial population (Np) using rules and / or DRL;
[0159] while Do when the algorithm termination condition is not reached
[0160] At←Use DRL to select the appropriate action (St);
[0161] Use the selected action At to generate a new measurement and control scheduling scheme through cross-mutation;
[0162] Pt←Update population (Pt-1,At);
[0163] Cross operation (Pt-1, At, Oc, α); / / Cross operation of measurement and control scheme
[0164] Mutation operation (Pt-1, At, Om, β); / / Mutation operation of measurement and control scheme
[0165] Ft←Fitness evaluation (Pt);
[0166] Rt←calculate reward; record state transfer;
[0167] If the DRL training conditions are met then / / Determine whether to train or test, set True for training and False for testing
[0168] Train DRL model parameters;
[0169] Else
[0170] Import the trained DRL model;
[0171] end if
[0172] end while
[0173] The initialized model parameters are as follows:
[0174] Parameter name Parameter value setting Initialization Scheme NP 100 Crossover rate α 0.9 Mutation rate β 0.2 Interval Algebra C 5 Training strategy ε 0.1 Optimizer Adam Learning Rate 0.0001 Discount Factor 0.99 Batch size 32 Activation Function Relu Number of hidden layers 4
[0175] The key to the design of the algorithm framework is to determine the combination of deep reinforcement learning methods and evolutionary algorithms, which will affect the role of deep reinforcement learning methods in the algorithm search process. There are various ways to combine deep reinforcement learning methods with evolutionary algorithms. Information can be used in multiple search and evolution links in the evolutionary algorithm framework to allow the agent to decide on appropriate actions to participate in the population evolution process, or task execution plans can be directly obtained based on the selected actions. According to the role of DRL and evolutionary algorithms in learning evolutionary algorithms, there are two main ways to combine algorithms: one is that DRL generates solutions directly for evolutionary algorithm search optimization, and the other is that the DRL method helps the heuristic algorithm to use more effective population evolution operations to search the solution space. This algorithm uses DRL to select appropriate crossover and mutation operations to generate satellite measurement and control scheduling plans. Crossover mainly includes single-point crossover and multi-point crossover. Compared with crossover, the probability of mutation is low and the change range of individuals is small. Mutation has the function of global search. This method can accelerate the convergence of the algorithm.
[0176] It should be understood that the above specific embodiments of the present invention are only used to illustrate or explain the principles of the present invention, and do not constitute a limitation of the present invention. Therefore, any modifications, equivalent substitutions, improvements, etc. made without departing from the spirit and scope of the present invention should be included in the protection scope of the present invention. In addition, the appended claims of the present invention are intended to cover all changes and modifications that fall within the scope and boundaries of the appended claims, or the equivalent forms of such scope and boundaries.
Claims
1. A large-scale satellite measurement and control scheduling method based on a self-organizing learning algorithm, characterized in that: The following steps are involved: S01: Establish a large-scale satellite measurement and control scheduling model, wherein the optimization goal of the large-scale satellite measurement and control scheduling model is to maximize the mission benefit; S02: Generate a measurement and control scheduling plan using a rule-based algorithm; S03: When the termination condition is not met, calculate the benefit and update the state at the next moment, calculate the strategy probability, generate an action selection operator based on the strategy probability, use the selection operator to generate a new measurement and control scheduling plan, repeat the process until the termination condition is met and output the final plan.
2. The large-scale satellite measurement and control scheduling method based on the self-organizing learning algorithm according to claim 1 is characterized in that: The optimization goal of the large-scale satellite measurement and control scheduling model is: The constraints are as follows: 1) Target uniqueness constraint: 2) Flow balance constraints, connecting decision variables and 3) The task sequence of each round starts from the virtual starting task s and ends with the virtual ending task e: 4)Conversion time constraints: 5) Arc segment time constraints: 6) Task time constraints: 7) Tasks can only be scheduled on executable arcs: 8) Conflicting arcs cannot be used at the same time, that is, if a task has been executed on an arc, the conflicting arc cannot execute any task: in, Whether tasks i and j are both executed in arc segment k, is whether the virtual starting task s and task j are both executed in arc segment k, is whether task j and virtual termination task e are both executed in arc segment k, is whether task i is executed in arc segment k, T is the task set, T = {i = 1, ..., |T|, O is the arc segment set, O = {k = 1, ..., |o|}, d i is the duration of task i, p i is the benefit of task i, is the visibility of task i in arc k, Indicates that task i is visible in arc segment k, otherwise it is 0, es i ,le i is the earliest start time and the latest end time of task i, st k ,et k is the measurement and control start time and measurement and control end time of arc segment k, k0 is the conflicting arc segment of arc segment k, t i ,t j are the observation start time of tasks i and j respectively, and M represents the coefficient.
3. The large-scale satellite measurement and control scheduling method based on the self-organizing learning algorithm according to claim 1 is characterized in that: In step S03, a deep reinforcement learning algorithm is used to select a suitable action, and the selected action At is used to generate a new measurement and control scheduling plan. The action includes the crossover operation (Pt-1, At, Oc, α) and mutation operation (Pt-1, At, Om, β) in the genetic algorithm, where Pt-1 is the population at time t-1, Oc is the crossover operator, Om is the mutation operator, α is the crossover probability, and β is the mutation probability.
4. The large-scale satellite measurement and control scheduling method based on the self-organizing learning algorithm according to claim 3 is characterized in that: It also includes fitness evaluation based on the obtained population Pt, and calculation of rewards based on the generated measurement and control scheduling plan benefit value, and the reward Rt = f t+1 –f t .
5. The large-scale satellite measurement and control scheduling method based on the self-organizing learning algorithm according to claim 1 is characterized in that: The policy probabilities are calculated using the softmax Greedy policy.
6. A large-scale satellite measurement, control and dispatching system based on a self-organizing learning algorithm, characterized in that: include: An inter-satellite link satellite measurement and control scheduling model building module is used to establish a large-scale satellite measurement and control scheduling model, wherein the optimization goal of the large-scale satellite measurement and control scheduling model is to maximize the mission benefit; The measurement and control scheduling plan generation module generates a measurement and control scheduling plan using a rule-based algorithm; Iterative optimization module, when the termination condition is not met, calculates the benefits and updates the state at the next moment, calculates the strategy probability, and generates an action selection operator based on the strategy probability; Use the selection operator to generate a new measurement and control scheduling plan, repeat the process until the termination condition is met and output the final plan.
7. The large-scale satellite measurement, control and dispatching system based on the self-organizing learning algorithm according to claim 6 is characterized in that: The optimization goal of the large-scale satellite measurement and control scheduling model is: The constraints are as follows: 1) Target uniqueness constraint: 2) Flow balance constraints, connecting decision variables and 3) The task sequence of each round starts from the virtual starting task s and ends with the virtual ending task e: 4)Conversion time constraints: 5) Arc segment time constraints: 6) Task time constraints: 7) Tasks can only be scheduled on executable arcs: 8) Conflicting arcs cannot be used at the same time, that is, if a task has been executed on an arc, the conflicting arc cannot execute any task: in, Whether tasks i and j are both executed in arc segment k, is whether the virtual starting task s and task j are both executed in arc segment k, is whether task j and virtual termination task e are both executed in arc segment k, is whether task i is executed in arc segment k, T is the task set, T = {i = 1, ..., |T|, O is the arc segment set, O = {k = 1, ..., |o|}, d i is the duration of task i, p i is the benefit of task i, is the visibility of task i in arc k, Indicates that task i is visible in arc segment k, otherwise it is 0, es i ,le i is the earliest start time and the latest end time of task i, st k ,et k is the measurement and control start time and measurement and control end time of arc segment k, k0 is the conflicting arc segment of arc segment k, t i ,t j are the observation start time of tasks i and j respectively, and M represents the coefficient.
8. The large-scale satellite measurement, control and dispatching system based on the self-organizing learning algorithm according to claim 6 is characterized in that: In the iterative optimization module, a deep reinforcement learning algorithm is used to select appropriate actions, and the selected action At is used to generate a new measurement and control scheduling plan. The actions include the crossover operation (Pt-1, At, Oc, α) and mutation operation (Pt-1, At, Om, β) in the genetic algorithm, where Pt-1 is the population at time t-1, Oc is the crossover operator, Om is the mutation operator, α is the crossover probability, and β is the mutation probability.
9. The large-scale satellite measurement and control scheduling method based on the self-organizing learning algorithm according to claim 8 is characterized in that: It also includes fitness evaluation based on the obtained population Pt, and calculation of rewards based on the generated measurement and control scheduling plan benefit value, and the reward Rt = f t+1 –f t .
10. A computer storage medium having a computer program stored thereon, characterized in that: When the computer program is executed, the large-scale satellite measurement, control and scheduling method based on the self-organizing learning algorithm described in any one of claims 1 to 5 is implemented.