Large-scale satellite task planning method and system based on static constraint conflict graph learning

Abstract

The invention provides a large-scale satellite task planning method and system based on static constraint conflict graph learning, and the method comprises the steps: carrying out the mathematical description of a large-scale satellite task planning problem, and designing an optimization objective function; carrying out constraint analysis and mathematical expression in a large-scale satellite task planning problem; dividing into static constraints and dynamic constraints according to different features of the constraints; constructing a static constraint conflict graph model for the static constraint; and inputting satellite tasks and original information thereof, solving the static constraint conflict graph model to obtain an initial satellite task combination sequence without static constraint conflicts, solving dynamic constraints in the initial satellite task combination sequence, and finally obtaining a satellite task planning result. According to the method, the problem of multi-satellite collaborative scheduling in a large-scale intensive task scene is solved, the calculation efficiency of solving a large-scale satellite task planning problem is greatly improved, and a new method reference is provided for further improving the intelligent management and control level of satellites.

Description

Technical Field

[0001] The invention belongs to the field of remote sensing satellite mission planning, and in particular relates to a large-scale satellite mission planning method and system based on static constraint conflict graph learning. Background Art

[0002] Remote sensing satellite mission planning is widely considered to be a non-deterministic polynomial (NP) hard problem with complex constraints. In order to solve this problem, domestic and foreign scholars have conducted a lot of research.

[0003] Exact algorithms and heuristic algorithms for solving satellite mission planning problems: In the early days of solving multi-satellite mission planning problems, the algorithms commonly used were usually divided into exact algorithms and heuristic algorithms. Exact methods, such as branch and bound and mixed integer linear programming (MILP), have deterministic solution steps and can solve problems in polynomial time in most cases; however, for larger-scale problems, it is actually impossible to establish a deterministic model. Heuristic methods can speed up the solution process by finding a satisfactory solution, but these methods usually rely on specific heuristic strategies, which are effective for optimizing small-scale problems, but difficult to solve large-scale problems.

[0004] Metaheuristic algorithms for solving satellite mission planning problems: Metaheuristic algorithms guided by a set of search strategies are considered to be an advanced solution framework for combinatorial optimization problems, independent of the heuristic strategy of a specific problem. According to the different search strategies, metaheuristic algorithms can be divided into four branches: evolutionary algorithms, physics-based algorithms, swarm-based algorithms, and human-based algorithms. Specifically, the typical metaheuristic algorithms that have a greater impact on solving optimization problems include genetic algorithms (GA), particle swarm optimization algorithms (PSO), taboo search, simulated annealing algorithms (SA), etc., which mainly belong to the branches of evolutionary algorithms and swarm algorithms. Because they can provide sufficiently high-quality optimization solutions for most NP problems, they have been widely used in engineering fields such as satellite mission planning and scheduling, vehicle path optimization, etc. for decades. Although metaheuristic algorithms can obtain better calculation results, they still inevitably fall into local optimality and sometimes cannot provide satisfactory convergence performance, especially when dealing with complex constraints in the task-intensive multi-satellite mission planning process.

[0005] Deep reinforcement learning algorithms for solving satellite mission planning problems: In recent years, algorithms that combine deep learning, reinforcement learning, and other neural networks have been proposed to solve multi-satellite mission planning problems. Deep reinforcement learning (DRL) has shown strong capabilities and potential in solving multi-satellite mission planning and scheduling problems. However, on the one hand, the use of deep reinforcement learning algorithms to solve multi-satellite mission planning problems is still in its infancy, and most research is still at the stage of solving single-satellite problems; on the other hand, as the number of missions increases, some deep reinforcement learning algorithms are still inferior to baseline algorithms such as genetic algorithms in terms of solution quality; in addition, whether these methods are applicable to large-scale multi-satellite mission planning problems with tens of thousands of missions has not been verified and discussed.

[0006] The above research can effectively solve combinatorial optimization problems such as satellite mission planning to a certain extent. However, there are still challenges in finding an efficient and high-quality method for satellite mission planning problems with large-scale intensive tasks:

[0007] First, due to the working characteristics of satellites, the single-satellite mission planning problem itself is highly complex, and it is difficult to obtain high-quality solutions using simple greedy algorithms or heuristic strategies.

[0008] In addition, large-scale multi-satellite mission planning problems often lead to an explosive increase in the problem size and task density, which places high demands on the efficiency of scheduling algorithms. However, the effectiveness of most algorithms in solving multi-satellite mission planning problems under large-scale tasks has not yet been verified or discussed. Summary of the invention

[0009] In order to solve the problems existing in the above-mentioned prior art, the present invention proposes a large-scale satellite mission planning method and system based on static constraint conflict graph learning. By creatively constructing a static constraint conflict graph model for satellite missions and taking advantage of the graph neural network's ability to identify the correlation between "unstructured" data, combined with the optimization search capability of traditional metaheuristic algorithms, a generalized solution framework for multi-satellite collaborative planning based on graph neural networks and metaheuristic algorithms is designed. The large-scale satellite mission planning method based on static constraint conflict graph learning includes:

[0010] Mathematically describe the large-scale satellite mission planning problem and design the optimization objective function; perform constraint analysis and mathematical expression on the large-scale satellite mission planning problem; classify the constraints into static constraints and dynamic constraints according to their different characteristics;

[0011] For static constraints, a static constraint conflict graph model is constructed;

[0012] Input the satellite mission and its original information, solve the static constraint conflict graph model, and obtain the initial satellite mission combination sequence without static constraint conflicts. Then solve the dynamic constraints in the initial satellite mission combination sequence, and finally obtain the satellite mission planning result.

[0013] Furthermore, for the optimization objective function, the optimization goal is set to maximize the sum of observed benefits of executing tasks under the constraints of time, energy, storage, etc. Its mathematical expression is:

[0014]

[0015] in, represents the observation benefit of satellite m performing mission k; Is a decision variable, indicating whether to choose to execute task k. If Then execute task k, otherwise, do not execute task k; and are the start and end times of the observation window of satellite m for mission k, respectively; The attitude adjustment stabilization time required for satellite m to perform mission k and mission k+1; and are the storage and energy consumed by satellite m to perform mission k; STORE m and ENERGY m are the total storage and total energy limits of satellite m respectively.

[0016] The static constraints are inherent attributes of the tasks and are independent of the decision sequence. The constraint associations between tasks do not change with the combination of tasks. The dynamic constraints are related to the decision sequence, and the constraint associations between tasks change dynamically with the combination of tasks.

[0017] Furthermore, the constructed static constraint conflict graph model is specifically as follows:

[0018] Consider tasks as nodes, V = {task1,task2,...,task k ,...,task K}, taking the constraint conflicts between tasks as edges, E = {conflict ij |i,j=1,2,...,K,i≠j}.

[0019] Furthermore, the steps for solving the static constraint conflict graph model are as follows:

[0020] The sampling aggregation graph network is used to express the characteristics of the task nodes in the satellite static constraint conflict graph and obtain the selection probability of the task.

[0021] Calculate the loss function based on the task selection probability;

[0022] The gradient of the loss function is calculated to optimize the sampling aggregation graph network and better realize the feature expression of the task nodes.

[0023] Furthermore, the task nodes in the satellite static constraint conflict graph are characterized and the specific steps to obtain the selection probability of the task are as follows:

[0024] Sampling subgraph: For each task node, the subgraph cutting method is used to randomly sample some neighbor nodes as aggregated feature nodes, thereby obtaining a subgraph containing the central node and neighbor nodes;

[0025] Aggregation: Based on the subgraph containing the central node and neighboring nodes, the features of the neighboring nodes of the central node are aggregated, and aggregation calculations are performed from the outermost nodes to the central node based on different aggregation functions to obtain the feature expression of the central node;

[0026] Task prediction: Based on the feature expression of the central node, nonlinear transformation is performed to obtain the probability of task node selection.

[0027] The loss function is mean square error, cross entropy or QUBO function, and the aggregation function used in the aggregation step of the sampling aggregation graph network is one or more combinations of average aggregation (mean), inductive aggregation, pooling aggregation (pool), and LSTM aggregation function.

[0028] When the loss function is designed based on the QUBO function, it is specifically:

[0029] Loss = -X T (α·R+β·P)X

[0030] Wherein, matrix R is the reward matrix of the task; X is the decision variable of the task, and X is expressed as the probability value of the task to be executed, that is, X = {(p1, p2, ..., p k ,...,p K )|0<p k <1}.

[0031] Furthermore, a meta-heuristic algorithm is used to solve the dynamic constraints, such as tabu search, simulated annealing, ant colony algorithm, genetic algorithm or particle swarm algorithm.

[0032] The present invention further proposes a system of a large-scale satellite mission planning method based on static constraint conflict graph learning, the system comprising: a satellite mission input module, a static constraint conflict graph model construction module, a graph network solution module for static constraint conflict resolution, a meta-heuristic solution module for dynamic constraint conflict resolution, and a satellite mission planning result output module.

[0033] Beneficial effects:

[0034] The present invention proposes a complete large-scale satellite mission planning method based on static constraint conflict graph learning. By creatively constructing a static constraint conflict graph model for satellite missions and utilizing the advantages of graph neural networks in identifying the correlation between "unstructured" data, combined with the optimization search capabilities of typical meta-heuristic algorithms such as particle swarm optimization and genetic algorithms, a general solution framework for multi-satellite collaborative planning based on graph neural networks and meta-heuristic algorithms is designed. This solves the problem of multi-satellite collaborative scheduling in large-scale intensive mission scenarios, greatly improves the computational efficiency of meta-heuristic algorithms such as genetic algorithms and particle swarm optimization in solving large-scale satellite mission planning problems, and provides a new method reference for further improving the level of satellite intelligent management and control. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 The overall process of large-scale satellite mission planning method based on static constraint conflict graph learning;

[0036] Figure 2 Build examples for static constraint conflict graphs;

[0037] Figure 3 Solving process for large-scale satellite mission planning based on static constraint conflict graph learning;

[0038] Figure 4 Design of network structure for sampling aggregation graph;

[0039] Figure 5 Coding methods for multi-satellite mission planning based on particle swarm optimization;

[0040] Figure 6 The basic process of large-scale multi-satellite mission planning solution based on particle swarm optimization;

[0041] Figure 7 Coding methods for multi-satellite mission planning based on genetic optimization;

[0042] Figure 8 The basic process of solving the large-scale multi-satellite mission planning method based on genetic optimization;

[0043] Fig. 9 Comparison of the training performance of the simulation model for tasks of different scales. (a) is the maximum number of independent set tasks that meet the static constraints, and (b) is the training time statistics.

[0044] Fig.10 Comparison of the model's test performance for tasks of different scales. (a) is the maximum number of independent set tasks that meet static constraints, and (b) is the test duration statistics.

[0045] Fig.11 Comparison curve for the average number of tasks executed;

[0046] Fig.12 Compare the average task planning solution time curve;

[0047] Fig.13 Comparison curve for the average number of tasks executed;

[0048] Fig.14 Comparison curve for average task planning solution time. DETAILED DESCRIPTION

[0049] The technical solution of the present invention is further described below in conjunction with the embodiments, but is not limited thereto. Any modification or equivalent replacement of the technical solution of the present invention without departing from the spirit and scope of the technical solution of the present invention should be included in the protection scope of the present invention.

[0050] A large-scale satellite mission planning method based on static constraint conflict graph learning. The overall process of the planning method is shown in Figure 1 , including:

[0051] Mathematically describe the large-scale satellite mission planning problem and design the optimization objective function; perform constraint analysis and mathematical expression on the large-scale satellite mission planning problem; and classify the constraints into static constraints and dynamic constraints according to their different characteristics.

[0052] For the optimization objective function, the optimization goal is set to maximize the sum of observed benefits of executing tasks under the constraints of time, energy, storage, etc. Its mathematical expression is:

[0053]

[0054] in, represents the observation benefit of satellite m performing mission k; Is a decision variable, indicating whether to choose to execute task k. If Then execute task k, otherwise, do not execute task k; and are the start and end times of the observation window of satellite m for mission k, respectively; The attitude adjustment stabilization time required for satellite m to perform mission k and mission k+1; and are the storage and energy consumed by satellite m to perform mission k; STORE m and ENERGY m are the total storage and total energy limits of satellite m respectively.

[0055] The static constraints are inherent attributes of the tasks and are independent of the decision sequence. The constraint associations between tasks do not change with the combination of tasks, such as the time constraints of tasks in the satellite mission planning process.

[0056] Specifically, time constraint means that a satellite can only perform one observation task within the same time period (time window). If the time windows of different observation tasks of a satellite overlap, it means a time conflict occurs.

[0057] The dynamic constraints are related to the decision sequence, and the constraint association relationship between tasks changes dynamically with the combination of tasks, such as satellite storage constraints and energy constraints in the satellite mission planning process.

[0058] Specifically, the "storage constraint" means that due to the limited satellite storage space, the data volume of all tasks cannot exceed the satellite storage capacity limit; the "energy constraint" means that for each satellite, the total energy consumption of the tasks cannot exceed the satellite's battery energy limit.

[0059] For static constraints, a static constraint conflict graph model is constructed.

[0060] Graph theory is introduced to represent the constraint conflicts between tasks. In mathematics, graphs are usually represented by G = (V, E), where V is the set of vertices and E is the set of edges. The static constraint conflict graph model constructed is as follows:

[0061] Consider tasks as nodes, V = {task1,task2,...,task k ,...,task K}, taking the constraint conflicts between tasks as edges, E = {conflict ij |i,j=1,2,...,K,i≠j}.

[0062] Taking the task time constraint as an example, a typical static constraint is constructed. Figure 2 As shown. The left side of the figure is a Gantt chart of the time window of a satellite observation task. It can be clearly seen that there are overlapping windows between task2 and task1, task3, and task4, that is, there is a conflict in time constraints. According to their conflict relationship, a graph structure as shown on the right side of the figure can be constructed, with nodes representing tasks and edges representing conflict relationships between tasks. In this way, the satellite tasks and the complex relationships between them can be clearly expressed. This representation method is also suitable for processing large-scale satellite tasks of the order of tens of thousands. For the resolution and optimization of static constraints in the satellite mission planning process, it is transformed into the problem of finding independent sets of the graph. Independent sets refer to the set of non-adjacent vertices in the graph G.

[0063] Input the satellite mission and its original information, solve the static constraint conflict graph model, obtain the initial satellite mission combination sequence without static constraint conflict, and then solve the dynamic constraints in the initial satellite mission combination sequence to finally obtain the satellite mission planning result. The mission planning solution process is shown in Figure 3 .

[0064] The specific steps for solving the static constraint conflict graph model are:

[0065] The sampling aggregation graph network is used to express the characteristics of the task nodes in the satellite static constraint conflict graph and obtain the selection probability of the task. The sampling aggregation graph network structure design is shown in Figure 4 ;

[0066] Calculate the loss function based on the task selection probability;

[0067] The gradient of the loss function is calculated to optimize the sampling aggregation graph network and better realize the feature expression of the task nodes.

[0068] The specific steps for expressing the characteristics of the task nodes in the satellite static constraint conflict graph and obtaining the selection probability of the task are as follows:

[0069] Sampling subgraph: For each task node, the subgraph cutting method is used to randomly sample some neighbor nodes as aggregated feature nodes, thereby obtaining a subgraph containing the central node and neighbor nodes;

[0070] Aggregation: Based on the subgraph containing the central node and neighboring nodes, the features of the neighboring nodes of the central node are aggregated, and aggregation calculations are performed from the outermost nodes to the central node based on different aggregation functions to obtain the feature expression of the central node;

[0071] Task prediction: Based on the feature expression of the central node, nonlinear transformation is performed to obtain the probability of task node selection.

[0072] The loss function is mean square error, cross entropy or QUBO function, and the aggregation function used in the aggregation step of the sampling aggregation graph network is one or more combinations of average aggregation (mean), inductive aggregation, pooling aggregation (pool), and LSTM aggregation function.

[0073] When the loss function is designed based on the QUBO function, it is specifically:

[0074] Loss = -X T (α·R+β·P)X

[0075] Where, matrix R is the reward matrix of the task, such as the benefit or priority of each task; matrix P is usually symmetrical or upper triangular, and is the penalty matrix when there is a conflict between tasks (edges in the graph); X is the decision variable of the task. In large-scale satellite mission planning problems, the decision variables of the task are defined as binary, which leads to the discrete optimization objective function. In order to better play the expression and extraction capabilities of the sampling aggregation graph network, X is expressed as the probability value of the task to be decided, that is, X = {(p1, p2, ..., p k ,...,p K )|0<p k <1}, thereby converting the loss function from a discrete function to a continuous optimization function.

[0076] The solution method for dynamic constraints adopts meta-heuristic algorithm, which is taboo search, simulated annealing, ant colony algorithm, genetic algorithm or particle swarm algorithm.

[0077] When the meta-heuristic algorithm is a particle swarm optimization algorithm, the multi-satellite mission planning solution method based on particle swarm optimization is as follows:

[0078] Step 1: Based on the prior knowledge provided by the graph network pre-solving module for static constraints, a group of particle populations are initialized. Each particle represents a planning solution, represented by binary code, and an initial position and velocity are assigned to each particle. Figure 5 Coding methods for multi-satellite mission planning based on particle swarm optimization;

[0079] Step 2: Check satellite mission constraints for each particle and resolve conflicts;

[0080] Step 3: For each particle, the fitness function value of each particle is calculated according to the optimization objective function.

[0081] Step 4: Mark the current individual historical optimal solution for each particle, and the particles collaborate and share information to obtain the global historical optimal solution;

[0082] Step 5: Update the speed and position of each particle according to the speed and position update formula. The update formula of particle speed is:

[0083] V ik (t+1)=w·V ik (t)+c1·r(t)·(lbest(t)-x ik (t))+c2·r(t)·(gbest(t)-x ik (t))

[0084] Where V ik(t) represents the velocity of particle i in the k-th dimension (satellite mission k) at time t (t-th iteration); r(t) is a random number between (0,1); c1, c2 are learning factors, and in the present invention, c1=c2=1.5; lbest(t), gbest(t) are the optimal solution of the current particle and the global optimal solution of the particle swarm, respectively; w is the inertia weight, w max =1.0,w min =0.4, and the inertia weight is dynamically adjusted during the optimization process to balance the globality and convergence speed of convergence.

[0085] The update formula of particle position is:

[0086] x ik (t+1)=x ik (t)+V ik (t)

[0087] In the formula, x ik (t) represents the position of particle i in the k-th dimensional space (satellite mission d) at time t (t-th iteration).

[0088] Step 6: Determine whether the maximum number of iterations has been reached. If so, go to Step 7, otherwise, go to Step 2;

[0089] Step 7: Output the global optimal individual, which is the satellite mission planning result. Figure 6 This paper provides the basic process of large-scale multi-satellite mission planning solution based on particle swarm optimization.

[0090] When the meta-heuristic algorithm is a genetic algorithm, the multi-satellite mission planning solution method based on genetic optimization is as follows:

[0091] Step 1: Initialize the population based on the prior knowledge provided by the graph network pre-solving module for static constraints. The individuals in the population are binary-encoded according to the task decision variables. Each bit of the chromosome represents an observation task, and its value is 0 or 1, indicating whether the task is selected for execution. The length of the chromosome is the number of observation tasks for all satellites for all targets; Figure 7 Coding methods for multi-satellite mission planning based on genetic optimization;

[0092] Step 2: Check the satellite mission constraints of each individual in the population and resolve conflicts;

[0093] Step 3: Calculate the fitness function value of each individual according to the optimization objective function, and mark the best individual in the population;

[0094] Step 4: Determine whether the maximum number of iterations has been reached. If so, go to Step 7; otherwise, go to Step 5.

[0095] Step 5: Based on the elite retention strategy, select individuals to enter the mating pool and perform operations such as crossover and mutation on them. The crossover operation adopts a single-point crossover method, and the mutation operation adopts a random multi-point mutation strategy.

[0096] Step 6: After crossover and mutation, select and retain the better offspring or parent generation, update the population, and go to Step 2;

[0097] Step 7: Output the global optimal individual, which is the satellite mission planning result.

[0098] A system for a large-scale satellite mission planning method based on static constraint conflict graph learning includes: a satellite mission input module, a static constraint conflict graph model building module, a graph network solving module for static constraint conflict resolution, a meta-heuristic solving module for dynamic constraint conflict resolution, and a satellite mission planning result output module. Figure 8 The basic process of large-scale multi-satellite mission planning solution based on genetic optimization is given.

[0099] Simulation verification:

[0100] Taking actual satellite mission data as a reference, the effectiveness of the large-scale satellite mission planning method based on static constraint conflict graph learning was verified in simulation mission scenarios of different scales, including 2000, 4000, 6000, 8000, 10000, 20000, 30000, 40000 and 50000 satellite missions.

[0101] The simulation verification results are as follows:

[0102] 1. Analysis of training results of static constraint conflict resolution network model

[0103] The static constraint conflict resolution graph network model is trained through task datasets of different scales ranging from 2000 to 50000, focusing on the effectiveness verification of the static constraint processing and resolution module. The module uses graph convolutional network (GCN) and sampling aggregation graph network (SAGE).

[0104] The static constraint resolution and optimization in the satellite mission planning process is transformed into the problem of finding the independent set of the mission constraint conflict graph. The more task nodes are included in the independent set and the shorter the training time is, the better the performance of the network model is. Fig. 9 shown.

[0105] From the above analysis, we can see that as the number of tasks increases, the training time required for the model increases. Compared with the traditional graph convolutional neural network (GCN), the sampled aggregate graph network (SAGE) used in this patent can achieve the same training time cost (such as Fig. 9 (b)), find a combination solution that contains more tasks and satisfies the static constraints, that is, it shows a significant advantage under the indicator of the maximum number of independent set tasks that satisfy the static constraints (such as Fig. 9 (as shown in (a)).

[0106] 2. Analysis of test results of static constraint conflict resolution network model

[0107] After the network training is completed, the actual solving ability of the static constraint conflict resolution graph network model is tested through task test data sets of different scales from 2000 to 50000 (with certain distribution differences from the training data). The simulation results are as follows: Fig.10 shown.

[0108] Through analysis, we can know that Fig.10 As shown in (b), once the network model is trained, the model calculation time is stable within 2 seconds for tasks of different sizes, indicating that the construction of the static constraint conflict graph model can effectively improve the efficiency of solving large-scale satellite mission planning. At the same time, in terms of solution quality, compared with the graph convolutional network GCN, the sampling aggregation graph network SAGE proposed in this patent can provide a better maximum independent set of tasks that meet static constraints, and can ensure the elimination of static constraints on tasks while taking into account the requirements of completing more tasks.

[0109] 3. Analysis of simulation results of multi-satellite mission planning solution method based on particle swarm optimization

[0110] The multi-satellite mission planning solution method based on particle swarm optimization was verified. For satellite missions of different sizes from 2,000 to 50,000, the satellite mission planning method based on traditional particle swarm optimization (random-PSO) and the satellite mission planning method based on static constraint conflict graph network and particle swarm optimization (SAGE-PSO) were used for simulation calculations to verify the effectiveness of the solution method and the performance superiority of the proposed method. In each mission scale scenario, 10 plannings were performed, and the average statistical results are shown in the figure below. Fig.11 , Fig.12 shown.

[0111] like Fig.11 The figure shows the average number of tasks executed in the task planning results. Fig.12The figure shows the average planning solution time. As the number of tasks increases, the required planning solution time increases. However, under the same number of tasks, the solution time of each algorithm is slightly different. However, in terms of the average number of executed tasks in the planning results, the SAGE-PSO algorithm proposed in this patent has significant advantages over the traditional particle swarm algorithm random-PSO, and can obtain higher quality task planning solutions at the same computing cost.

[0112] 4. Analysis of simulation results of multi-satellite mission planning solution method based on genetic optimization

[0113] The multi-satellite mission planning solution method based on genetic optimization was verified. For satellite missions of different numbers ranging from 2,000 to 50,000, satellite mission planning based on traditional genetic optimization (random-GA), satellite mission planning based on greedy genetic optimization (greedy-GA), and satellite mission planning method based on static constraint conflict graph network and genetic optimization (SAGE-GA) were used for simulation calculations to verify the effectiveness of the solution method and the performance superiority of the proposed method. In each mission scale scenario, 10 plans were made, and the simulation results are shown in the figure below. Fig.13 , Fig.14 shown.

[0114] like Fig.13 The figure shows the average number of tasks executed in the task planning results. Fig.14 The figure shows the average planning solution time. As the number of tasks increases, the required planning solution time increases. Under the same number of tasks, when the task scale is small, the difference in solution time of each algorithm is small. When the number of tasks is large, such as when the number of tasks reaches more than 30,000, the SAGE-GA proposed in this patent has higher computational efficiency than the traditional genetic algorithms random-GA and greedy-GA without static constraint conflict graph models. In terms of the average number of executed tasks in the planning results, SAGE-GA has a significant advantage, and even exceeds the calculation results of the final iteration of the traditional genetic algorithm at the beginning of the iteration.

[0115] The above is only a preferred embodiment of the present invention. It should be pointed out that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A large-scale satellite mission planning method based on static constraint conflict graph learning, comprising: Provide mathematical description for large-scale satellite mission planning problems and design optimization objective functions; Perform constraint analysis and mathematical expression for large-scale satellite mission planning problems; classify constraints into static constraints and dynamic constraints according to their different characteristics; For static constraints, a static constraint conflict graph model is constructed; Input the satellite mission and its original information, solve the static constraint conflict graph model, and obtain the initial satellite mission combination sequence without static constraint conflicts. Then solve the dynamic constraints in the initial satellite mission combination sequence, and finally obtain the satellite mission planning result.

2. The large-scale satellite mission planning method based on static constraint conflict graph learning according to claim 1 is characterized in that: For the optimization objective function, the optimization goal is set to maximize the sum of observed benefits of executing tasks under the constraints of time, energy, storage, etc. Its mathematical expression is: in, represents the observation benefit of satellite m performing mission k; Is a decision variable, indicating whether to choose to execute task k. If Then execute task k, otherwise, do not execute task k; and are the start and end times of the observation window of satellite m for mission k, respectively; The attitude adjustment stabilization time required for satellite m to perform mission k and mission k+1; and are the storage and energy consumed by satellite m to perform mission k; STORE m and ENERGY m are the total storage and total energy limits of satellite m respectively.

3. The large-scale satellite mission planning method based on static constraint conflict graph learning according to claim 1 is characterized in that: The static constraints are inherent attributes of the tasks and are independent of the decision sequence. The constraint associations between tasks do not change with the combination of tasks. The dynamic constraints are related to the decision sequence, and the constraint associations between tasks change dynamically with the combination of tasks.

4. The large-scale satellite mission planning method based on static constraint conflict graph learning according to claim 1 is characterized in that: The constructed static constraint conflict graph model is as follows: Consider tasks as nodes, V = {task1,task2,...,task k ,...,task K }, taking the constraint conflicts between tasks as edges, E = {conflict ij |i,j=1,2,...,K,i≠j}.

5. The large-scale satellite mission planning method based on static constraint conflict graph learning according to claim 1 is characterized in that: The specific steps for solving the static constraint conflict graph model are: The sampling aggregation graph network is used to express the characteristics of the task nodes in the satellite static constraint conflict graph and obtain the selection probability of the task. Calculate the loss function based on the task selection probability; The gradient of the loss function is calculated to optimize the sampling aggregation graph network and better realize the feature expression of the task nodes.

6. The large-scale satellite mission planning method based on static constraint conflict graph learning according to claim 5 is characterized in that: The specific steps for expressing the characteristics of the task nodes in the satellite static constraint conflict graph and obtaining the selection probability of the task are as follows: Sampling subgraph: For each task node, the subgraph cutting method is used to randomly sample some neighbor nodes as aggregated feature nodes, thereby obtaining a subgraph containing the central node and neighbor nodes; Aggregation: Based on the subgraph containing the central node and neighboring nodes, the features of the neighboring nodes of the central node are aggregated, and aggregation calculations are performed from the outermost nodes to the central node based on different aggregation functions to obtain the feature expression of the central node; Task prediction: Based on the feature expression of the central node, nonlinear transformation is performed to obtain the probability of task node selection.

7. The large-scale satellite mission planning method based on static constraint conflict graph learning according to claim 5 is characterized in that: The loss function is mean square error, cross entropy or QUBO function, and the aggregation function used in the aggregation step of the sampling aggregation graph network is one or more combinations of average aggregation, inductive aggregation, pooling aggregation and LSTM aggregation function.

8. The large-scale satellite mission planning method based on static constraint conflict graph learning according to claim 5 is characterized in that: The loss function is designed based on the QUBO function, specifically: Loss=-X T (α·R+β·P)X Wherein, matrix R is the reward matrix of the task; X is the decision variable of the task, and X is expressed as the probability value of the task to be executed, that is, X = {(p1, p2, ..., p k ,...,p K )|0<p k <1}.

9. The large-scale satellite mission planning method based on static constraint conflict graph learning according to claim 1 is characterized in that: The solution method for dynamic constraints adopts meta-heuristic algorithm, which is taboo search, simulated annealing, ant colony algorithm, genetic algorithm or particle swarm algorithm.

10. A system for implementing the large-scale satellite mission planning method based on static constraint conflict graph learning according to any one of claims 1 to 9, characterized in that: The system includes: a satellite mission input module, a static constraint conflict graph model building module, a graph network solving module for static constraint conflict resolution, a meta-heuristic solving module for dynamic constraint conflict resolution, and a satellite mission planning result output module.

Images