A method and system for resolving dynamic constraint conflicts in large-scale satellite missions based on heterogeneous graph learning

Abstract

This invention proposes a method and system for resolving dynamic constraint conflicts in large-scale satellite missions based on heterogeneous graph learning. The method includes: mathematically describing the mission planning problem and designing an optimization objective function; analyzing the characteristics and mathematically expressing the constraints; constructing a static constraint conflict graph model for the satellite missions based on static constraints; constructing a dynamic constraint conflict graph model for the satellite missions based on dynamic constraints; solving the static constraint conflict graph model; solving the dynamic constraint conflict graph model to obtain the satellite mission selection probabilities; and performing decision optimization on the satellite missions to finally obtain the satellite mission planning result. This invention achieves efficient scheduling and allocation decisions for large-scale, dense multi-satellite missions, solving the problem of efficient computation for constraint screening and conflict resolution in multi-satellite mission planning under dense mission conditions, and significantly reducing the number of conflict resolution operations in the multi-satellite mission planning process.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing satellite mission planning, specifically relating to a method and system for resolving dynamic constraint conflicts in large-scale satellite missions based on heterogeneous graph learning. Background Technology

[0002] The remote sensing satellite mission planning problem is widely considered a nondeterministic polynomial (NP) problem with complex constraints. To address this problem, scholars both domestically and internationally have conducted extensive research.

[0003] Exact and Heuristic Algorithms for Solving Satellite Mission Planning Problems: Early solutions to multi-satellite mission planning problems were typically categorized into exact 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, establishing a deterministic model is practically impossible. Heuristic methods can accelerate the solution process by finding satisfactory solutions, but these methods often rely on specific heuristic strategies. They are effective for optimizing small-scale problems but struggle with large-scale ones.

[0004] Metaheuristic algorithms for solving satellite mission planning problems: Metaheuristic algorithms, guided by a set of search strategies and independent of specific problem-specific heuristics, are considered a high-level framework for solving combinatorial optimization problems. Typical metaheuristic algorithms include Genetic Algorithms (GA), Particle Swarm Optimization (PSO), Tabu Search, and Simulated Annealing (SA). They can provide sufficiently high-quality optimization solutions for most NP-hard problems and have been widely used in engineering fields such as satellite mission planning and scheduling, and vehicle routing optimization for decades. Although metaheuristic algorithms can achieve better computational results, they inevitably get trapped in local optima and sometimes fail to provide satisfactory convergence performance. Their computational efficiency is often limited by the handling of complex constraints and conflict resolution processes. Mission-intensive multi-satellite mission planning problems require algorithms capable of handling even larger and more complex constraints and conflict resolution.

[0005] Deep reinforcement learning algorithms for solving satellite mission planning problems: In recent years, algorithms combining 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 early stages, with most research remaining focused on solving single-satellite problems; on the other hand, as the number of missions increases, some deep reinforcement learning algorithms still fall short of baseline algorithms such as genetic algorithms in terms of solution quality; furthermore, whether these methods are suitable for large-scale multi-satellite mission planning problems with tens of thousands of missions has not yet been verified and discussed.

[0006] The above research can effectively solve combinatorial optimization problems such as satellite mission planning to a certain extent. However, when facing the problem of satellite mission planning for large-scale, dense missions, the task constraint checking and conflict resolution under the huge solution space become important factors affecting the efficiency of the algorithm. However, very few studies have focused on this important link of constraint checking and conflict resolution in the satellite mission planning process. Finding an efficient solution method for satellite mission planning for large-scale, dense missions still faces the following challenges:

[0007] Due to the operational characteristics of satellites, the single-satellite mission planning problem is inherently highly complex. The constraints between satellite resource capabilities and mission requirements are difficult to characterize and describe, and simple greedy construction algorithms or heuristic strategies are unlikely to yield high-quality solutions.

[0008] Large-scale multi-satellite mission planning problems often lead to an explosive increase in problem size and mission density. However, the effectiveness of most current algorithms in solving large-scale multi-satellite mission planning problems has not been verified or discussed. Therefore, it is urgent to propose an efficient algorithm for solving large-scale multi-satellite mission planning problems.

[0009] The methods for checking and resolving mission constraints are important factors affecting the computational efficiency of satellite mission planning algorithms. Most current research has not yet conducted theoretical studies and discussions on the essential relationships of constraints in satellite mission planning. Summary of the Invention

[0010] Therefore, this invention proposes a method for resolving dynamic constraint conflicts in large-scale satellite missions based on heterogeneous graph learning. By introducing the powerful representation capability of heterogeneous graphs for relationships between unstructured and multi-class data, a dynamic constraint conflict graph model for satellite missions based on heterogeneous graphs is constructed. The temporal variation characteristics of complex dynamic constraints of satellites are transformed into spatial topological features of multiple types of nodes and edges under heterogeneous graphs. Through learning dynamic constraint conflict resolution strategies for satellite missions and designing multi-satellite collaborative mission planning and solving methods, the number of conflict resolutions in the multi-satellite mission planning and solving process is greatly reduced, enabling efficient scheduling and allocation decisions for large-scale and dense multi-satellite missions.

[0011] A method for resolving dynamic constraint conflicts in large-scale satellite missions based on heterogeneous graph learning includes:

[0012] This paper provides a mathematical description of the large-scale satellite mission planning problem and designs an optimization objective function. It also analyzes the characteristics and mathematically expresses the constraints in the large-scale satellite mission planning problem, classifying them into static constraints and dynamic constraints based on their different characteristics.

[0013] For static constraints, a static constraint conflict graph model for the satellite mission is constructed; for dynamic constraints, a dynamic constraint conflict graph model for the satellite mission is constructed.

[0014] Inputting satellite missions and their original information, the static constraint conflict graph model is solved to resolve static constraint conflicts between missions, resulting in a set of satellite missions with reduced static constraint conflicts. For this set of satellite missions with reduced static constraint conflicts, a dynamic constraint conflict graph model is solved to resolve dynamic constraint conflicts between missions, obtaining satellite mission selection probabilities. Based on these reduced dynamic constraint conflict probabilities, satellite mission decision optimization is performed, ultimately yielding the satellite mission planning result.

[0015] Furthermore, the static constraints are inherent attributes of the task and are independent of the decision-making sequence; the constraint relationships between tasks do not change with the combination of tasks. The dynamic constraints are related to the decision-making sequence; the constraint relationships between tasks change dynamically with the combination of tasks. The specific satellite mission static constraint conflict graph model is as follows:

[0016] Treat tasks as nodes, V = {task1, task2, ..., task...} k ,...,task K}, treating the constraint conflicts between tasks as edges, E = {conflict} ij |i,j=1,2,...,K,i≠j}.

[0017] Furthermore, the aforementioned satellite dynamic constraint conflict graph model is specifically as follows:

[0018] Construct a heterogeneous graph of dynamic constraint conflicts for satellite missions, mathematically described as follows: Where V and E represent the sets of nodes and edges, respectively, and F and R represent the sets of node and edge types, respectively, satisfying |F|+|R|>2. Each node v∈V is associated with a node type mapping function. Each edge e∈E is associated with an edge type mapping function φ:E→R.

[0019] In the heterogeneous graph of dynamic constraint conflict of satellite mission, the nodes are satellite mission and dynamic constraint critical conflict cluster nodes, and the edges are the connecting edges between energy constraint critical conflict cluster nodes and satellite mission nodes, and the connecting edges between storage constraint critical conflict cluster nodes and satellite mission nodes. Dynamic constraint critical conflict cluster nodes are divided into energy constraint critical conflict cluster nodes and storage constraint critical conflict cluster nodes.

[0020] Furthermore, the dynamic constraint critical conflict cluster node is defined as follows: the satellite mission node connected to the dynamic constraint critical conflict cluster node just exceeds the satellite's energy and storage limits, that is, a dynamic constraint conflict such as energy and storage has occurred. If one mission is removed from these missions, the satellite's energy and storage constraints will be satisfied.

[0021] Furthermore, the method for obtaining the dynamically constrained critical conflict clique nodes employs a sampling-based algorithm for generating dynamically constrained critical conflict clique nodes, with the following specific steps:

[0022] (1) Import all satellite mission sets and determine the number of dynamic constraint critical conflict clusters to be sampled;

[0023] (2) Record the current sampling quantity as i, and initialize the collision clique C. i ;

[0024] (3) Randomly select a mission from the satellite mission set and add it to conflict group C. i middle;

[0025] (4) Regarding the conflict group C i The task in the process is dynamically constrained. If the constraints are met, proceed to step (3); if the constraints are not met, proceed to step (5).

[0026] (5) Determine if the current conflict clique is a critical conflict clique. If so, save the current critical conflict clique minC. i If yes, proceed to step (7); otherwise, proceed to step (6).

[0027] (6) Randomly remove one mission from the current conflict group and proceed to step (5);

[0028] (7) Determine whether the number of conflict clusters meets the sampling requirement. If yes, output the set of all critical conflict clusters, with each critical conflict cluster corresponding to a dynamic constraint critical conflict cluster node; otherwise, i = i + 1, go to step (2) and continue sampling.

[0029] Furthermore, the solution to the dynamic constraint conflict graph model for satellite missions is as follows:

[0030] Input and output element design: The input elements include a heterogeneous graph of task constraints, task node features, and a set of constraint conflict element paths; the output elements are the set of probabilities that the satellite mission will be selected for execution.

[0031] Heterogeneous graph attention network structure design: The heterogeneous graph attention network structure includes two layers of attention mechanisms: node attention and meta-path attention. Node attention mainly represents the importance of each task node, while meta-path attention mainly represents the contribution of different types of dynamic constraint critical conflict clumps to the conflict correlation between tasks.

[0032] Conflict resolution strategy learning and training: The conflict strategy learning and training process adopts an unsupervised training method, and the loss function adopts a quadratic unconstrained optimization function that is improved from discrete to continuous.

[0033] Furthermore, the input and output elements are specifically as follows:

[0034] Input elements include: satellite mission dynamic constraint conflict heterogeneity map Satellite mission node characteristics Satellite mission node features include satellite number, observation target number, mission number, observation window start time, observation window end time, duration, satellite attitude angle, imaging energy consumption, storage usage, and mission benefits; the dynamic constraint conflict metapath set metaPath = {Φ1, Φ2, ..., Φ m ,...,Φ M};Conflicting metapath Φ m This refers to various sets of nodes communicating through "satellite mission tasks". i - Dynamically constrained critical conflict clique nodes of category m m - Satellite mission task j "A path formed by relationships;"

[0035] The output element is the set of probabilities that each satellite mission will be selected for execution, i.e., probs = {p1, p2, ..., p...} i ,...,p N}

[0036] Furthermore, the specific steps for learning and training conflict resolution strategies are as follows:

[0037] (1) Based on the satellite mission probability values ​​output by the heterogeneous graph attention network, determine whether each one is greater than the set threshold. If so, add it to the set of execution mission nodes and go to step (3); otherwise, add it to the set of deleted mission nodes and go to step (2).

[0038] (2) Based on the deleted task node, disband the critical conflict group to which it belongs, update the critical conflict group node set, and proceed to step (3).

[0039] (3) Calculate the penalty matrix P: The penalty matrix P is obtained by taking the current set of critical conflict cluster nodes, obtaining the satellite task nodes that are connected to the critical conflict cluster nodes by edges, and counting the number of edges connected to the critical conflict cluster nodes by these satellite task nodes. The penalty mechanism can adopt single-point-single penalty, single-point-cumulative penalty, multi-point-single penalty, multi-point-cumulative penalty, etc., to obtain the task penalty matrix.

[0040] (4) Calculate the loss function loss = X T (R+P)X, where matrix R is the reward matrix of the task, such as the reward or priority of each task; P is the penalty matrix; and X is the selection probability matrix of the task.

[0041] (5) Determine whether the maximum number of iterations has been reached or whether the loss function has converged. If yes, go to (6); otherwise, update the parameters of the heterogeneous graph attention network and continue training.

[0042] (6) Output satellite mission selection probability.

[0043] Furthermore, the optimization of satellite mission decisions specifically includes:

[0044] (1) Initialize the satellite mission planning decision result vector, and initialize the conflict resolution statistics count = 0;

[0045] (2) Sort the satellite mission set according to the probability values ​​obtained from the satellite mission constraint conflict diagram model and conflict resolution.

[0046] (3) Perform constraint checks on the selected set of tasks to be executed. If all constraints are met, proceed to step (6); if the constraints are not met, proceed to step (4).

[0047] (4) In the current set of tasks with constraint conflicts, delete the task with the smallest probability value;

[0048] (5) Update the task planning decision vector and update the conflict resolution statistics count = count + 1, then go to step (3);

[0049] (6) Output the task planning decision result vector and the number of conflict resolutions.

[0050] A system for implementing the above-described large-scale satellite mission planning method based on dynamic constraint conflict graph learning includes: a satellite mission input module, a static constraint conflict graph model construction module, a dynamic constraint conflict graph model construction module, a static constraint conflict resolution module, a dynamic constraint conflict resolution module, a satellite mission decision optimization module, and a satellite mission planning result output module.

[0051] Beneficial effects:

[0052] To address the challenges of large-scale solution space, complex and variable task constraints, time-consuming conflict resolution, and low efficiency in multi-satellite mission planning under intensive mission scenarios, this invention proposes a large-scale satellite mission dynamic constraint conflict resolution method based on heterogeneous graph learning. This method achieves efficient scheduling and allocation decisions for large-scale, intensive multi-satellite missions through the construction of a multi-satellite mission constraint conflict graph model, the learning of dynamic constraint conflict resolution strategies based on heterogeneous graph attention networks, and the design of a multi-satellite collaborative mission planning solution method. It solves the problem of efficient computation for constraint resolution and conflict resolution in multi-satellite mission planning under intensive mission scenarios. By constructing and learning heterogeneous graphs, it extracts common knowledge of complex dynamic constraint conflict relationships under different mission scenarios, significantly reducing the number of conflict resolution operations in the multi-satellite mission planning solution process. This provides important theoretical and technical support for further improving the intelligence and efficiency of future satellite management. Attached Figure Description

[0053] Figure 1 The overall process of a dynamic constraint conflict resolution method for large-scale satellite missions based on heterogeneous graph learning;

[0054] Figure 2 Example of a conflict diagram between static and dynamic constraints;

[0055] Figure 3 Explanation of the time-varying characteristics of dynamic constraints for satellite missions;

[0056] Figure 4 The flowchart of the dynamic constraint critical conflict clique generation algorithm based on sampling strategy is as follows;

[0057] Figure 5 The overall algorithm flow;

[0058] Figure 6 Design and conflict resolution learning training for heterogeneous graph attention network structures;

[0059] Figure 7 The flowchart of a multi-star collaborative task planning algorithm based on a probabilistic greedy strategy;

[0060] Figure 8 This is a simulation training time curve;

[0061] Figure 9 Choose a probability distribution histogram for the simulation task: (a) 1000 task nodes, (b) 2000 task nodes, (c) 3000 task nodes, (d) 4000 task nodes, (e) 5000 task nodes.

[0062] Figure 10 Statistics on average planning indicators under different task scales: (a) average number of tasks executed, (b) average total revenue from executed tasks, (c) average number of task conflict resolutions during the planning process, and (d) average planning calculation time. Detailed Implementation

[0063] The technical solution of the present invention will be further described below with reference to the embodiments, but it is not limited thereto. Any modifications or equivalent substitutions to the technical solution of the present invention without departing from the spirit and scope of the technical solution of the present invention should be covered within the protection scope of the present invention.

[0064] The following is a detailed implementation of a method for resolving dynamic constraint conflicts in large-scale satellite missions based on heterogeneous graph learning. The overall process is as follows: Figure 1 As shown, it includes the following steps:

[0065] This paper provides a mathematical description of the large-scale satellite mission planning problem and designs an optimization objective function. It also analyzes the characteristics and mathematically expresses the constraints in the large-scale satellite mission planning problem, classifying them into static constraints and dynamic constraints based on their different characteristics. Figure 2 This is an example of a diagram showing the conflict between static and dynamic constraints.

[0066] The static constraints are inherent attributes of the task and are independent of the decision-making sequence; the constraint relationships between tasks do not change with the combination of tasks. Examples include time constraints in satellite mission planning. Specifically, time constraints mean that a satellite can only perform one observation task within the same time period (time window). If the time windows of different observation tasks for a satellite overlap, it indicates a time conflict.

[0067] Dynamic constraints are related to the decision-making sequence, and the constraint relationships between tasks change dynamically as the tasks are combined. Examples include satellite storage constraints and energy constraints in the satellite mission planning process. Specifically, "storage constraints" refer to the fact that due to limited satellite storage space, the total data volume of all tasks cannot exceed the satellite's storage capacity limit; "energy constraints" refer to the fact that for each satellite, the total energy consumption of the tasks cannot exceed the satellite's battery energy limit. Figure 3 Explanation of the time-varying characteristics of dynamic constraints for satellite missions.

[0068] For static constraints (such as mission time constraints), considering the time-invariant characteristics of static constraints, a static constraint conflict graph model for satellite missions is constructed.

[0069] The aforementioned satellite mission static constraint conflict diagram model is specifically as follows:

[0070] Treat tasks as nodes, V = {task1, task2, ..., task...} k ,...,task K}, treating the constraint conflicts between tasks as edges, E = {conflict} ij |i,j=1,2,...,K,i≠j}.

[0071] The aforementioned satellite mission static constraint conflict graph model is a homogeneous graph-based model. A homogeneous graph is a graph with only one type of node and edge, mathematically described as G = (V, E). For the satellite mission static constraint conflict graph, all satellite missions are classified as nodes of the same type, i.e., V = {task1, task2, ..., task...} k ,...,task K Static constraint conflicts (such as time conflicts) between satellite missions are classified as one type of edge, i.e., E = {conflict} ij |i,j=1,2,...,K,i≠j}.

[0072] Example of a static constraint conflict graph for satellite missions based on homogeneity graphs. Figure 2 As shown in (a), task nodes connected by edges indicate that tasks conflict and cannot be executed simultaneously. Therefore, resolving the static constraint conflict of satellites is transformed into the problem of finding independent sets of nodes in the graph, that is, finding a set of unconnected nodes in the graph. The representation and learning of static constraint conflict relationships can be achieved using neural network algorithms for processing homogeneous graphs, such as Graph Convolutional Network (GCN), Graph Sample and Aggregate (GraphSAGE), and Graph Attention Network (GAT).

[0073] For dynamic constraints (such as energy constraints, storage constraints, and attitude maneuver constraints), considering the higher complexity and time-varying nature of dynamic constraints with decision sequences, a satellite mission dynamic constraint conflict graph model is constructed. This invention constructs a mission dynamic constraint conflict graph model based on heterogeneous graphs, as follows: Figure 2 As shown in (b), taking dynamic constraints such as satellite energy and storage as an example, a graph contains three types of nodes: satellite mission (red dots), energy constraint critical conflict cluster nodes (green squares), and storage constraint critical conflict cluster nodes (blue triangles), as well as two types of edges (green solid lines and blue dashed lines) connecting the storage and energy critical conflict cluster nodes to the satellite mission nodes, respectively.

[0074] The aforementioned satellite dynamic constraint conflict diagram model is specifically as follows:

[0075] Construct a heterogeneous graph of dynamic constraint conflicts for satellite missions, mathematically described as follows: Where V and E represent the sets of nodes and edges, respectively, and F and R represent the sets of node and edge types, respectively, satisfying |F|+|R|>2. Each node v∈V is associated with a node type mapping function. Each edge e∈E is associated with an edge type mapping function φ:E→R.

[0076] In the heterogeneous graph of dynamic constraint conflict of satellite mission, the nodes are satellite mission and dynamic constraint critical conflict cluster nodes, and the edges are the connecting edges between energy constraint critical conflict cluster nodes and satellite mission nodes, and the connecting edges between storage constraint critical conflict cluster nodes and satellite mission nodes. Dynamic constraint critical conflict cluster nodes are divided into energy constraint critical conflict cluster nodes and storage constraint critical conflict cluster nodes.

[0077] The above-mentioned satellite dynamic constraint conflict graph model is a satellite mission dynamic constraint conflict graph based on heterogeneous graphs.

[0078] A heterogeneous graph is a graph containing multiple different types of nodes and edges, mathematically described as follows: Where V and E represent the sets of nodes and edges, respectively, and F and R represent the sets of node and edge types, respectively, satisfying |F|+|R|>2. Each node v∈V is associated with a node type mapping function. Each edge e∈E is associated with an edge type mapping function φ:E→R.

[0079] Based on the fundamental principles of heterogeneous graphs and the characteristics of satellite dynamic constraints, heterogeneous graphs are used to describe the dynamic constraint conflict relationships of satellite missions, transforming the temporal variation characteristics of complex dynamic constraints of satellites into spatial topological features of multiple types of nodes and edges under heterogeneous graphs.

[0080] A dynamically constrained critical conflict cluster node is defined as follows: the satellite mission nodes connected to the dynamically constrained critical conflict cluster node just exceed the satellite's energy and storage limits, that is, dynamic constraint conflicts such as energy and storage have occurred. If one of these missions is removed, the satellite's energy and storage constraints will be satisfied.

[0081] Obtaining the nodes of dynamically constrained critical conflict clusters is crucial for constructing the dynamic constraint heterogeneous graph of a satellite mission. Since the dynamic constraint relationships between mission nodes are difficult to describe accurately, and it is also difficult to obtain all dynamically constrained conflict clusters in polynomial time, this invention employs a sampling-based algorithm for generating dynamically constrained critical conflict clusters. The basic algorithm flow is as follows: Figure 4 As shown.

[0082] The method for obtaining dynamically constrained critical conflict clique nodes adopts a sampling-based algorithm for generating dynamically constrained critical conflict clique nodes. The specific steps are as follows:

[0083] (1) Import all satellite mission sets and determine the number of dynamic constraint critical conflict clusters to be sampled;

[0084] (2) Record the current sampling quantity as i, and initialize the collision clique C. i ;

[0085] (3) Randomly select a mission from the satellite mission set and add it to conflict group C. i middle;

[0086] (4) Regarding the conflict group C i The task in the process is dynamically constrained. If the constraints are met, proceed to step (3); if the constraints are not met, proceed to step (5).

[0087] (5) Determine if the current conflict clique is a critical conflict clique. If so, save the current critical conflict clique minC. i If yes, proceed to step (7); otherwise, proceed to step (6).

[0088] (6) Randomly remove one mission from the current conflict group and proceed to step (5);

[0089] (7) Determine whether the number of conflict clusters meets the sampling requirement. If so, output the set of all critical conflict clusters, with each critical conflict cluster corresponding to a dynamic constraint critical conflict cluster node; otherwise, i = i + 1, go to step (2), and continue sampling. Input the satellite mission and its original information, solve the static constraint conflict graph model, resolve the static constraint conflicts between missions, and obtain the set of satellite missions with reduced static constraint conflicts; for the set of satellite missions with reduced static constraint conflicts, solve the dynamic constraint conflict graph model of the satellite missions formed by it, resolve the dynamic constraint conflicts between missions, and obtain the satellite mission selection probability; based on the satellite mission selection probability after the reduction of dynamic constraint conflicts, perform decision optimization on the satellite missions, and finally obtain the satellite mission planning result.

[0090] The overall process of the satellite mission dynamic constraint conflict resolution strategy learning algorithm based on heterogeneous graph attention network is as follows: Figure 5 As shown.

[0091] The solution to the dynamic constraint conflict graph model for satellite missions is as follows:

[0092] Input and output element design: The input elements include a heterogeneous graph of task constraints, task node features, and a set of constraint conflict element paths; the output elements are the set of probabilities that the satellite mission will be selected for execution.

[0093] Heterogeneous graph attention network structure design: The heterogeneous graph attention network structure includes two layers of attention mechanisms: node attention and meta-path attention. Node attention mainly represents the importance of each task node, while meta-path attention mainly represents the contribution of different types of dynamic constraint critical conflict clumps to the conflict correlation between tasks; the heterogeneous graph attention network structure is as follows: Figure 6 As shown in (a), there are two attention mechanisms: node attention and meta-path attention. Node attention mainly represents the importance of each task node, while meta-path attention mainly represents the contribution of different types of dynamic constraint critical conflict groups to the conflict correlation between tasks.

[0094] Conflict resolution strategy learning and training: The conflict resolution strategy learning and training process adopts an unsupervised training method, and the loss function is a quadratic unconstrained optimization function (QUBO) that has been improved from discrete to continuous. The basic process of conflict resolution strategy learning and training is as follows: Figure 6 As shown in (b).

[0095] The input and output elements are specifically as follows:

[0096] Input elements include: satellite mission dynamic constraint conflict heterogeneity map Satellite mission node characteristics Satellite mission node features include satellite number, observation target number, mission number, observation window start time, observation window end time, duration, satellite attitude angle, imaging energy consumption, storage usage, and mission benefits; the dynamic constraint conflict metapath set metaPath = {Φ1, Φ2, ..., Φ m ,...,Φ M};Conflicting metapath Φ m This refers to various node sets (satellite mission nodes, critical conflict cluster nodes) communicating through "satellite mission tasks". i - Dynamically constrained critical conflict clique nodes of category m m - Satellite mission task j "A path formed by relationships;"

[0097] The output element is the set of probabilities that each satellite mission will be selected for execution, i.e., probs = {p1, p2, ..., p...} i ,...,p N}

[0098] The specific steps for learning and training conflict resolution strategies are as follows:

[0099] (1) Based on the satellite mission probability values ​​output by the heterogeneous graph attention network, determine whether each one is greater than the set threshold. If so, add it to the set of execution mission nodes and go to step (3); otherwise, add it to the set of deleted mission nodes and go to step (2).

[0100] (2) Based on the deleted task node, disband the dynamic constraint critical conflict group to which it belongs, update the set of nodes of the dynamic constraint critical conflict group, and go to step (3).

[0101] (3) Calculate the penalty matrix P: The penalty matrix P is obtained by obtaining the satellite mission nodes that are connected to the dynamic constraint critical conflict cluster nodes based on the current set of dynamic constraint critical conflict cluster nodes, and counting the number of edges connected to these satellite mission nodes and the dynamic constraint critical conflict cluster nodes. The penalty mechanism can adopt strategies such as single-point-single penalty, single-point-cumulative penalty, multi-point-single penalty, and multi-point-cumulative penalty to obtain the task penalty matrix.

[0102] (4) Calculate the loss function loss = X T (R+P)X, where matrix R is the reward matrix of the task, such as the reward or priority of each task; P is the penalty matrix; and X is the selection probability matrix of the task.

[0103] (5) Determine whether the maximum number of iterations has been reached or whether the loss function has converged. If yes, go to (6); otherwise, update the parameters of the heterogeneous graph attention network and continue training.

[0104] (6) Output satellite mission selection probability.

[0105] Based on the satellite mission selection probability after dynamic constraint conflict reduction, a multi-satellite collaborative mission planning solution method based on a probabilistic greedy strategy is designed to optimize satellite mission decisions. By utilizing the constraint conflict graph model of satellite missions and the constraint conflict resolution network module provided by the constraint conflict resolution experience knowledge, the difficulty of conflict resolution and the time consumption of constraint checking in the large-scale dense mission planning decision process can be effectively reduced, and the satellite mission planning results can be obtained quickly, further improving the efficiency of satellite mission planning solution.

[0106] The overall process of the multi-star cooperative task planning and solving algorithm based on the probabilistic greedy strategy is as follows: Figure 7 As shown.

[0107] The specific optimization of satellite mission decisions includes:

[0108] (1) Initialize the satellite mission planning decision result vector, and initialize the conflict resolution statistics count = 0;

[0109] (2) Sort the satellite mission set according to the probability values ​​obtained from the satellite mission constraint conflict diagram model and conflict resolution.

[0110] (3) Perform constraint checks on the selected set of tasks to be executed. If all constraints are met, proceed to step (6); if the constraints are not met, proceed to step (4).

[0111] (4) In the current set of tasks with constraint conflicts, delete the task with the smallest probability value;

[0112] (5) Update the task planning decision vector and update the conflict resolution statistics count = count + 1, then go to step (3);

[0113] (6) Output the task planning decision result vector and the number of conflict resolutions.

[0114] This invention also proposes a system for implementing the above-mentioned large-scale satellite mission planning method based on dynamic constraint conflict graph learning. The system includes: a satellite mission input module, a static constraint conflict graph model construction module, a dynamic constraint conflict graph model construction module, a static constraint conflict resolution module, a dynamic constraint conflict resolution module, a satellite mission decision optimization module, and a satellite mission planning result output module.

[0115] Simulation verification:

[0116] Using actual satellite mission data as a reference, the effectiveness of the large-scale satellite mission dynamic constraint conflict resolution method based on heterogeneous graph learning was verified under simulation mission scenarios of different scales, including 10,000, 20,000, 30,000, 40,000, and 50,000 satellite missions respectively.

[0117] The simulation results are as follows:

[0118] 1. Analysis of Training Results of Dynamic Constraint Conflict Resolution Network Model

[0119] The dynamic constraint conflict resolution graph network model was simulated and trained using task datasets of different sizes ranging from 10,000 to 50,000. The effectiveness of the dynamic constraint conflict resolution method for satellite missions was verified. The core network modules were respectively adopted by graph convolutional network (GCN), sampled aggregate graph network (SAGE), and heterogeneous graph attention network (HAN), and comparative analysis was performed.

[0120] Regarding training time, the training time for different networks is as follows: Figure 8 As shown, the training time of the model gradually increases with the increase in the number of task nodes. Under the same task scale, the HAN model requires the shortest training time, while the GCN model requires the longest training time, and the SAGE model is in between.

[0121] Figure 9The histograms show the probability distribution of selected tasks under different task sizes, with the highly sparse arrangement visually reflecting the model's decision preference for task selection. The GCN model produces a dense task selection probability distribution, with almost all task selection probabilities close to 1, exhibiting aggressive and greedy behavior. In contrast, the SAGE model is more conservative, with a sparse task selection probability distribution, and most task selection probabilities close to 0. The HAN model's probability distribution sparsity falls between that of the GCN and SAGE models.

[0122] In summary, through analysis of the training process and training results, compared with the GCN and SAGE models, the HAN model proposed in this patent shows significant performance advantages in several aspects, such as the convergence of the training loss function, training time, and task selection probability distribution.

[0123] 2. Effectiveness Analysis of Dynamic Constraint Conflict Resolution Network Model for Solving Large-Scale Satellite Mission Planning

[0124] To further verify the effectiveness of the proposed task conflict resolution method (module B) based on dynamic constraint heterogeneous graphs for solving large-scale satellite mission planning problems, test datasets of different mission sizes (with some distributional differences from the training dataset) were used. First, a task conflict resolution model based on static constraint homogeneous graphs (module A) was used to handle static constraints, resulting in a set of satellite missions with reduced static constraint conflicts. Then, a task conflict resolution model based on dynamic constraint heterogeneous graphs (module B) was used to handle dynamic constraints. Finally, a task decision optimization method based on a greedy strategy (module C) was used for planning and solving to obtain the final mission planning result.

[0125] Using task datasets of varying sizes, ranging from 10,000 to 50,000, ablation analysis simulation experiments were conducted on each module. Based on the simulation results, the effectiveness of the dynamic constraint conflict resolution model in the task planning and decision-making process was analyzed.

[0126] Test datasets of different task sizes were processed sequentially through modules A, B, and C to obtain task planning results. The average number of tasks scheduled for execution is as follows: Figure 10 As shown in (a), the average reward for performing the task is as follows: Figure 10 As shown in (b), the trends are similar. The average number of task conflict resolutions and the average computation time for scheduling are respectively as follows: Figure 10 (c) and Figure 10 As shown in (d), compared to modules A and C, which only have the ability to handle static constraints, adding module B, which has the ability to handle dynamic constraints, can significantly reduce the pressure of resolving task conflicts during the constraint checking process. Figure 10As shown in (c). Specifically, different models in Module B exhibit different performance characteristics. For different task sizes, the HAN model reduced conflict resolution by an average of 73.48%, SAGE by approximately 54.25%, and GCN by approximately 8.92%. Therefore, scheduling efficiency is improved to some extent while maintaining the quality of the solution, such as... Figure 3 As shown in (a), (b), and (d), the average number of tasks planned for execution and the average revenue from executing tasks remain almost at the same level under different task sizes. Figure 10 As shown in (d), module B becomes more effective in reducing scheduling time as the task size increases. Compared with the GCN and SAGE models, the HAN model clearly has a significant advantage in task conflict resolution and computation time.

[0127] The above description is only a preferred embodiment of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for resolving dynamic constraint conflicts in large-scale satellite missions based on heterogeneous graph learning, comprising: A mathematical description is provided for the large-scale satellite mission planning problem, and an optimization objective function is designed. This paper analyzes the characteristics and mathematical expressions of constraints in large-scale satellite mission planning problems, and classifies them into static constraints and dynamic constraints according to their different characteristics. For static constraints, a static constraint conflict graph model for the satellite mission is constructed; for dynamic constraints, a dynamic constraint conflict graph model for the satellite mission is constructed. Inputting satellite missions and their original information, the static constraint conflict graph model is solved to resolve static constraint conflicts between missions, resulting in a set of satellite missions with reduced static constraint conflicts. For this set of satellite missions with reduced static constraint conflicts, a dynamic constraint conflict graph model is solved to resolve dynamic constraint conflicts between missions, obtaining satellite mission selection probabilities. Based on these reduced dynamic constraint conflict probabilities, satellite mission decision optimization is performed, ultimately yielding the satellite mission planning result. The aforementioned satellite mission dynamic constraint conflict graph model is specifically as follows: Construct a heterogeneous graph of dynamic constraint conflicts for satellite missions, mathematically described as follows: Where V and E represent the sets of nodes and edges, respectively, and F and R represent the sets of node and edge types, respectively, satisfying |F|+|R|>

2. Each node v∈V is associated with a node type mapping function. Each edge e∈E is associated with an edge type mapping function φ:E→R; In the heterogeneous graph of dynamic constraint conflict of satellite mission, the nodes are satellite mission and dynamic constraint critical conflict cluster nodes, and the edges are the connecting edges between energy constraint critical conflict cluster nodes and satellite mission nodes, and the connecting edges between storage constraint critical conflict cluster nodes and satellite mission nodes. Dynamic constraint critical conflict cluster nodes are divided into energy constraint critical conflict cluster nodes and storage constraint critical conflict cluster nodes. The dynamically constrained critical conflict cluster node is: the satellite mission node connected to the dynamically constrained critical conflict cluster node just exceeds the satellite's energy and storage limits. The solution to the dynamic constraint conflict graph model for satellite missions is as follows: Input and output element design: The input elements include a heterogeneous graph of task constraints, task node features, and a set of constraint conflict element paths; the output elements are the set of probabilities that the satellite mission will be selected for execution. Heterogeneous graph attention network structure design: The heterogeneous graph attention network structure includes two layers of attention mechanisms: node attention and meta-path attention. Node attention represents the importance of each task node, while meta-path attention represents the contribution of different types of dynamic constraint critical conflict clumps to the conflict correlation between tasks. Conflict resolution strategy learning and training: The conflict strategy learning and training process adopts an unsupervised training method, and the loss function adopts a quadratic unconstrained optimization function that is improved from discrete to continuous. The set of dynamic constraint conflict metapaths is named metaPath = {Φ1, Φ2, ..., Φ...} m ,...,Φ M };Conflicting metapath Φ m This refers to various sets of nodes communicating through "satellite mission tasks". i - Dynamically constrained critical conflict clique nodes of category m m - Satellite mission task j "A path formed by relationships;" The specific steps for learning and training conflict resolution strategies are as follows: (1) Based on the satellite mission probability values ​​output by the heterogeneous graph attention network, determine whether each one is greater than the set threshold. If so, add it to the set of execution mission nodes and go to step (3); otherwise, add it to the set of deleted mission nodes and go to step (2). (2) Based on the deleted task node, disband the critical conflict group to which it belongs, update the critical conflict group node set, and proceed to step (3). (3) Calculate the penalty matrix P: The penalty matrix P is obtained by taking the current set of critical conflict cluster nodes, obtaining the satellite task nodes that are connected to the critical conflict cluster nodes by edges, and counting the number of edges connected to the critical conflict cluster nodes by these satellite task nodes. The penalty mechanism can adopt single-point-single-time penalty, single-point-cumulative penalty, multi-point-single-time penalty, and multi-point-cumulative penalty strategies to obtain the task penalty matrix. (4) Calculate the loss function loss = X T (R+P)X, where matrix R is the reward matrix of the task, representing the reward or priority of each task; P is the penalty matrix; and X is the selection probability matrix of the task. (5) Determine whether the maximum number of iterations has been reached or whether the loss function has converged. If yes, go to (6); otherwise, update the parameters of the heterogeneous graph attention network and continue training. (6) Output satellite mission selection probability.

2. The method for resolving dynamic constraint conflicts in large-scale satellite missions based on heterogeneous graph learning according to claim 1, characterized in that, The static constraints are inherent attributes of the task and are independent of the decision-making sequence; the constraint relationships between tasks do not change with the combination of tasks. The dynamic constraints are related to the decision-making sequence; the constraint relationships between tasks change dynamically with the combination of tasks. The satellite mission static constraint conflict diagram model is specifically as follows: Treat tasks as nodes, V = {task1, task2, ..., task...} k ,...,task K }, treating the constraint conflicts between tasks as edges, E = {conflict} ij |i,j=1,2,...,K,i≠j}.

3. The method for resolving dynamic constraint conflicts in large-scale satellite missions based on heterogeneous graph learning according to claim 1, characterized in that, The method for obtaining the dynamically constrained critical conflict clique nodes adopts a dynamic constrained critical conflict clique node generation algorithm based on a sampling strategy, and the specific steps are as follows: (1) Import all satellite mission sets and determine the number of dynamic constraint critical conflict clusters to be sampled; (2) Record the current sampling quantity as i, and initialize the collision clique C. i ; (3) Randomly select a mission from the satellite mission set and add it to conflict group C. i middle; (4) Regarding the conflict group C i The task in the process is dynamically constrained. If the constraints are met, proceed to step (3); if the constraints are not met, proceed to step (5). (5) Determine if the current conflict clique is a critical conflict clique. If so, save the current critical conflict clique minC. i If yes, proceed to step (7); otherwise, proceed to step (6). (6) Randomly remove one mission from the current conflict group and proceed to step (5); (7) Determine whether the number of conflict clusters meets the sampling requirement. If yes, output the set of all critical conflict clusters, with each critical conflict cluster corresponding to a dynamic constraint critical conflict cluster node; otherwise, i = i + 1, go to step (2) and continue sampling.

4. The method for resolving dynamic constraint conflicts in large-scale satellite missions based on heterogeneous graph learning according to claim 1, characterized in that, The input and output elements are specifically as follows: Input elements include: satellite mission dynamic constraint conflict heterogeneity map Satellite mission node characteristics The satellite mission node characteristics include satellite number, observation target number, mission number, observation window start time, observation window end time, duration, satellite attitude angle, imaging energy consumption, storage usage, and mission benefits. The output element is the set of probabilities that each satellite mission will be selected for execution, i.e., probs = {p1, p2, ..., p...} i ,...,p N } 5. The method for resolving dynamic constraint conflicts in large-scale satellite missions based on heterogeneous graph learning according to claim 1, characterized in that, The specific optimization of satellite mission decisions includes: (1) Initialize the satellite mission planning decision result vector, and initialize the conflict resolution statistics count = 0; (2) Sort the satellite mission set according to the probability values ​​obtained from the satellite mission constraint conflict diagram model and conflict resolution. (3) Perform constraint checks on the selected set of tasks to be executed. If all constraints are met, proceed to step (6); if the constraints are not met, proceed to step (4). (4) In the current set of tasks with constraint conflicts, delete the task with the smallest probability value; (5) Update the task planning decision vector and update the conflict resolution statistics count = count + 1, then go to step (3); (6) Output the task planning decision result vector and the number of conflict resolutions.

6. A system for implementing the method for resolving dynamic constraint conflicts in large-scale satellite missions based on heterogeneous graph learning as described in any one of claims 1 to 5, characterized in that, The system includes: a satellite mission input module, a static constraint conflict graph model construction module, a dynamic constraint conflict graph model construction module, a static constraint conflict resolution module, a dynamic constraint conflict resolution module, a satellite mission decision optimization module, and a satellite mission planning result output module.

Images