Multi-satellite distributed cooperative task planning method based on multi-agent game
By employing multi-agent game theory, a multi-star distributed network system without a master star is constructed, enabling autonomous negotiation and local optimization in multi-star collaborative task planning. This solves the problems of computational complexity and low optimization efficiency in multi-star distributed task planning, and improves task response speed and system robustness.
Patent Information
- Application Number
- CN202511845075.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-09
- Publication Date
- 2026-04-24
AI Technical Summary
In existing technologies, multi-satellite distributed collaborative mission planning suffers from problems such as computational complexity, high strip overlap rate, failure to fully utilize the maneuverability of agile satellites, and low efficiency of optimization algorithms.
A multi-satellite distributed collaborative task planning method based on multi-agent game theory is adopted to construct a masterless multi-satellite distributed network system. Task planning is achieved through inter-satellite interaction and game learning. Each satellite agent performs local optimization and autonomous negotiation. Using concepts such as potential game, best response and regret value, distributed learning rules are designed to achieve task planning with optimal or near-optimal global benefits.
It enables rapid calculation of near-optimal multi-satellite collaborative mission planning without relying on a centralized control center, reducing the burden on ground operators and operating costs, and improving the response speed of observation missions and system robustness.
Smart Images

Figure CN121923693A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of satellite observation mission planning technology, specifically to a multi-satellite distributed collaborative mission planning method based on multi-agent game theory. Background Technology
[0002] With the increasing demand for Earth observation and the booming development of commercial spaceflight, large-scale distributed remote sensing systems are gradually becoming a new paradigm for space-based Earth observation. These systems offer higher spatial, temporal, and spectral resolution, enabling comprehensive perception and understanding of target areas. Compared to single large remote sensing satellite platforms, they possess enormous potential and advantages in terms of rapid mission response, low-cost spaceflight, resilience, and on-orbit reconfiguration.
[0003] Traditional satellites operate under a centralized, ground-based control model. Ground control centers pre-program observation tasks and transmit them to the satellite platforms as specific action commands. This system is simple in structure and easy to manage. However, limitations imposed by factors such as the communication window between the satellite platform and the ground control center, and exponential growth, make this operational model ill-suited to the demands of large-scale, highly dynamic, and emergency-response-intensive observation applications. These shortcomings become increasingly apparent, especially with the ever-increasing number of satellites in orbit. Furthermore, the lack of system resilience and robustness due to single-point failures makes this centralized control model ill-suited for multi-satellite collaboration in highly competitive environments.
[0004] The distributed operation mode employs different control mechanisms, effectively utilizing the satellite platform's real-time sensing, neighborhood interaction, and autonomous negotiation capabilities to achieve autonomous system operation and collaborative observation under decentralized conditions. Compared to a centralized framework, it offers advantages such as greater robustness, faster response speed, and lower operating costs.
[0005] Considering the constraints of satellite platform physical characteristics, complex space communication environment, and dynamically uncertain observation mission requirements, the lack of central nodes and global information poses a significant challenge to multi-satellite autonomous collaborative mission planning technology. Therefore, researching and implementing a distributed collaborative mission planning method based on inter-satellite autonomous interaction and negotiation to meet the collaborative management needs of large-scale satellite constellations in orbit has significant practical engineering application value.
[0006] Multi-satellite distributed collaborative mission planning refers to the process by which satellites, without relying on ground or on-orbit centralized control nodes, obtain globally optimal or near-optimal observation mission planning schemes through single-satellite planning and inter-satellite negotiation, without violating any hard constraints. Bonnet studied the multi-objective constellation scheduling problem using adaptive multi-agent theory oriented towards scheme optimization and proposed a decentralized distributed collaborative strategy. Van Der Horst studied a mission allocation method based on a market auction mechanism, where each member satellite uses its relative fitness as a bid value based on its remaining resource level, enabling mission allocation without global state knowledge. Gao Yuan studied a mission planning algorithm framework based on a contract network protocol and designed collaborative algorithms for moving and sensitive targets. Song Nan, based on regional grid segmentation preprocessing, studied a multi-objective regional collaborative observation method based on an improved contract network and proposed an evaluation index for imaging performance. Liu Zilin extended the contract network mission allocation mechanism, proposing a recursive allocation mechanism to reduce communication costs in negotiation. In addition, researchers also studied self-organizing methods inspired by swarm intelligence. Inspired by ant colony pheromone trajectories, Iacopino et al. designed a task planning method based on ant colony optimization meta-heuristic operators. Addressing the inherent limitations of microsatellites and nanosatellites, Tripp et al. proposed a consensus-based initiative-driven task planning method, investigating multi-satellite coordination mechanisms based solely on indirect interaction when direct inter-satellite communication is lacking. This method significantly reduces the burden of inter-satellite communication but cannot guarantee the satisfaction of repetitive constraints or further improve system collaboration performance. Literature review shows that domestic and international research in this field mainly focuses on theoretical methods such as contract networks, market auction mechanisms, and biological swarm intelligence. These methods can achieve task allocation based on interaction and negotiation to a certain extent, but they still do not completely eliminate the dependence on centralized nodes, and the algorithms still have significant room for improvement in robustness and global benefit optimization. Summary of the Invention
[0007] In view of the above-mentioned technical problems, this invention proposes a multi-satellite distributed collaborative mission planning method based on multi-agent game theory to solve the problems of complex calculation of regional target observation planning, high strip overlap rate, failure to fully utilize the maneuverability of agile satellites, and low efficiency of optimization algorithms in the existing technology.
[0008] The technical solution to the technical problem of this invention is: a multi-star distributed cooperative task planning method based on multi-agent game theory, comprising the following steps:
[0009] Step S1: Construct a multi-satellite distributed network system without a master star, in which each satellite agent has the same equal status;
[0010] Step S2: The observation task is either injected from the ground to any satellite or generated autonomously by any satellite, and broadcast throughout the network via the multi-satellite distributed network system to all satellite agents;
[0011] Step S3: Each satellite agent performs converged analysis on the received tasks, and plans and optimizes local observation schemes based on task type, its own status and capabilities to generate an initial task planning scheme.
[0012] Step S4: Each satellite agent broadcasts its current mission planning scheme to the entire network and receives mission planning schemes from other satellite agents.
[0013] Step S5: Each satellite agent optimizes its own task planning scheme locally based on the task planning schemes received from neighboring satellite agents, and obtains the optimized task planning scheme.
[0014] Step S6: Each satellite agent calculates its own regret value, which is the difference between the local benefits of the optimized mission planning scheme and the current mission planning scheme. The regret value is then broadcast to the entire network, while the regret values of other satellite agents are received.
[0015] Step S7: Determine whether the regret values of all satellite agents meet the preset convergence conditions;
[0016] If the conditions are met, the coordination and optimization process is stopped, and the current task planning scheme of each satellite agent is used as the final task allocation result.
[0017] If not satisfied, proceed to step S8;
[0018] Step S8: Each satellite agent iteratively updates its current task planning scheme using the optimized task planning scheme according to the preset game learning rules, and then jumps to step S6.
[0019] According to one technical solution of the present invention, in step S1, the multi-star distributed network system without a master star is composed of... Composed of remote sensing satellites, represented as , Each satellite Defined as:
[0020]
[0021] in, and These represent the satellite's unique identifier and network address, respectively. Indicates satellite orbit information, Indicates the load field of view angle, Indicates load type, Indicates the maximum lateral sway angle, Indicates the shortest boot time, Indicates attitude yaw rate, Indicates the longest startup time per lap. Indicates the maximum observation resolution, This indicates the local minimum solar altitude angle required for the payload to conduct effective observations.
[0022] Each satellite intelligent agent uses inter-satellite links to form a real-time interconnected, masterless peer-to-peer network.
[0023] According to one technical solution of the present invention, in step S2, the set of observation tasks is denoted as... , Each task Defined as:
[0024]
[0025] in, Indicates the observational gains of the task, Indicates the type of task, Indicates the latitude and longitude coordinates of the mission. Indicate the required observation type, [ Indicates the desired observation time interval, This indicates the minimum resolution required to meet the requirements. This indicates the frequency of observations within a certain period of time. This represents the minimum time difference that must be allowed between two effective imaging missions.
[0026] According to a technical solution of the present invention, step S3, the local observation scheme planning and local optimization, specifically includes:
[0027] Step S301: Each satellite agent calculates the visible observation window for each mission based on its own orbital parameters and mission geographical location;
[0028] Step S302: Construct a set of feasible observation windows that satisfy the filtering constraints, wherein the filtering constraints include at least: observation type matching, side swing angle range, minimum power-on time, solar altitude angle requirement and resolution requirement;
[0029] Step S303: Under the premise of satisfying the attitude transition time constraint and the cumulative power-on time constraint, with the goal of maximizing local observation benefits, select a series of observation windows from the set of feasible observation windows to form the initial task planning scheme. .
[0030] According to one technical solution of the present invention, the local benefit is calculated as follows: satellite intelligent agent Local benefits The incremental benefit it brings to the overall system observation is as follows:
[0031]
[0032] in, and These represent the number of point target tasks and the number of area tasks, respectively. Indicates the first A collection area of tasks in each region. Indicates all valid bands and The intersection area For the area, For satellite Point-to-point tasks under the current task planning scheme The number of valid score observations For satellite The resulting regional tasks The effective observation area.
[0033] According to a technical solution of the present invention, step S5, which involves performing local optimization of the self-planning scheme, specifically includes:
[0034] Knowing the current mission planning schemes of other satellites, the satellite agent replans its own observation window sequence to maximize the local gains, thus obtaining an optimized mission planning scheme. And called the optimized task planning scheme This is the best response for the satellite intelligent agent.
[0035] According to a technical solution of the present invention, in step S6, the satellite intelligent agent In the decision-making round Regret value Represented as:
[0036]
[0037] in, This is the optimized task planning scheme obtained in step S5. This indicates the current task planning scheme. This is a function for calculating local benefits.
[0038] According to one technical solution of the present invention, in step S7, the preset convergence condition is: the regret value of all satellite agents is 0.
[0039] According to one technical solution of the present invention, in step S8, the preset game learning rule is an asynchronous coordination update rule, specifically:
[0040] In each iteration, the satellite agent with the strictest regret value among all satellite agents, or the satellite agent with the smallest ID among all satellite agents with the largest regret value, is selected.
[0041] The selected satellite agent updates its current mission planning scheme to the optimized mission planning scheme obtained in step S5. ;
[0042] Satellite agents that were not selected will retain their current planning schemes.
[0043] According to a technical solution of the present invention, in step S8, the preset game learning rule is a synchronous coordination update rule, specifically:
[0044] In each iteration, each satellite agent compares its own regret value with the regret values of all its neighboring satellite agents;
[0045] If a satellite agent's regret value is significantly greater than the regret values of all its neighbors, or if its regret value is equal to the maximum regret value among its neighbors but its ID is less than the IDs of all neighbors with that maximum regret value, then the satellite agent updates its current planning scheme to the optimized task planning scheme obtained in step S5. ;
[0046] Otherwise, the satellite agent will maintain its current mission planning scheme.
[0047] Compared with the prior art, the present invention has the following advantages:
[0048] This invention presents a multi-satellite distributed collaborative task planning method based on multi-agent game theory. Addressing the mixed observation needs of point targets and regional targets, it considers filtering constraints, pairwise constraints, and cumulative constraints in the execution of observation tasks. Utilizing concepts such as potential game theory, optimal response, and regret value, it designs and implements a distributed task collaborative allocation and planning method that does not rely on centralized decision-making by a central node through a combination of "single-satellite local optimization" and "multi-satellite autonomous negotiation."
[0049] In terms of single-satellite planning, the focus is on optimizing the local observation window sequence using heuristic methods based on mission requirements, single-satellite capabilities, constraint models, and planning results of other satellites, in order to obtain improved strategies or optimal responses under the current system planning scheme. In terms of multi-satellite coordination, the planning schemes and regret values of neighboring satellites are obtained through inter-satellite communication, and synchronous or asynchronous coordination updates of strategies are achieved based on specific game learning rules, which can ensure the autonomous emergence and rapid convergence of Nash equilibrium cooperative schemes.
[0050] This invention overcomes the excessive reliance of traditional multi-satellite mission allocation algorithms on centralized control centers, distributing the massive computational and optimization burden across various satellite platforms. At the same time, by utilizing inter-satellite interaction and game-theoretic learning frameworks, it ensures that the system mission planning results obtained during the iteration process exhibit a monotonically increasing trend in global benefits. Under the premise of reducing the burden on ground operators and operating costs, improving the response speed of observation missions and the robustness of the system, it can achieve rapid calculation of near-optimal planning schemes.
[0051] Furthermore, this invention can not only be used to solve the distributed satellite task allocation problem, but its modeling method and solution mechanism can also be applied to distributed task allocation and decision-making problems in other fields, such as unmanned vehicle path planning and multi-robot task allocation. Attached Figure Description
[0052] Figure 1 This schematic diagram illustrates a multi-star distributed cooperative task planning method based on multi-agent game theory provided in one embodiment of the present invention.
[0053] Figure 2 This diagram illustrates an example of an iterative multi-agent game learning method provided in one embodiment of the present invention.
[0054] Figure 3 The illustration shows a dynamic iterative comparison curve between the multi-agent game distributed collaborative task planning method provided in an embodiment of the present invention and the traditional centralized method. Detailed Implementation
[0055] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.
[0056] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. The embodiments cannot be described in detail here, but the embodiments of the present invention are not limited to the following embodiments.
[0057] This invention presents a multi-satellite distributed collaborative mission planning method based on multi-agent game theory. It models a fully equivalent multi-satellite system as a cooperative potential game. By defining a local payoff function consistent with the global objective and designing distributed learning rules based on optimal response and regret value, each satellite can autonomously and rapidly converge to Nash equilibrium through iteration by relying only on limited information interaction with its neighbors, thereby obtaining a high-quality collaborative mission planning scheme.
[0058] like Figure 1 As shown, the present invention provides a multi-agent game-based multi-star distributed cooperative task planning method, comprising the following steps:
[0059] Step S1: Construct a multi-satellite distributed network system without a master star, in which each satellite agent has the same equal status;
[0060] The multi-star distributed network system without a master star consists of Composed of remote sensing satellites, represented as , Each satellite Defined as:
[0061]
[0062] in, and These represent the satellite's unique identifier and network address, respectively. Indicates satellite orbit information, Indicates the load field of view angle, Indicates the payload type (visible light, SAR, etc.). Indicates the maximum lateral sway angle, Indicates the shortest boot time, Indicates attitude yaw rate, Indicates the longest startup time per lap. Indicates the maximum observation resolution, This indicates the local minimum solar altitude angle required for the payload to conduct effective observations.
[0063] Each satellite uses a unique ID and IP address to form a real-time interconnected, masterless multi-satellite distributed network system via inter-satellite links or access to backbone networks such as StarNet, and utilizes distributed data distribution services such as DDS. This eliminates the need for ground-based master-slave node configuration and allows satellite nodes to join and leave at any time.
[0064] Step S2: The observation task is either injected from the ground to any satellite or generated autonomously by any satellite, and broadcast throughout the network via the multi-satellite distributed network system to all satellite agents;
[0065] In step S2, the ground can transmit the task set via a satellite-to-ground link. Upload to any satellite. Each mission Defined as:
[0066]
[0067] in, Indicates the observational gains of the task, Indicates the type of task (point objective or area objective). Indicates the latitude and longitude coordinates of the mission. Indicate the required observation type, [ Indicates the desired observation time interval, This indicates the minimum resolution required to meet the requirements. This indicates the frequency of observations within a certain period of time. This represents the minimum time difference that must be allowed between two effective imaging missions. Furthermore, observation missions can also be autonomously generated by any satellite based on its own maintained situational information.
[0068] Upon receiving a ground mission package or generating a mission autonomously, the node satellite broadcasts the relevant information to the entire network via a distributed inter-satellite network, enabling all satellites to maintain the same global mission requirements in real time. .
[0069] Step S3: Each satellite agent performs converged analysis on the received tasks, and plans and optimizes local observation schemes based on task type, its own status, and capabilities to generate an initial task planning scheme. The local observation scheme planning and optimization specifically includes:
[0070] Step S301: Each satellite agent calculates the visible observation window for each mission based on its own orbital parameters and mission geographical location;
[0071] Step S302: Construct a set of feasible observation windows that satisfy the filtering constraints, wherein the filtering constraints include at least: observation type matching, side swing angle range, minimum power-on time, solar altitude angle requirement and resolution requirement;
[0072] Step S303: Under the premise of satisfying the attitude transition time constraint and the cumulative power-on time constraint, with the goal of maximizing local observation benefits, select a series of observation windows from the set of feasible observation windows to form the initial task planning scheme. .
[0073] Each satellite first undergoes orbital extrapolation based on its orbital parameters. Combining this with the mission's geographical location and payload characteristics, the visible window information for each mission is obtained through discretized calculations of the yaw and pitch degrees of freedom. Specifically, for regional targets, the activation time for each visible window should be no later than the time of entering the target area, and the deactivation time should be no earlier than the time of leaving the target area.
[0074] Each visible window is defined by the following tuple: , Representing satellite intelligent agents respectively With observation mission , Indicates the start and end times of the observation window. Indicates the lateral swing angle during observation. It is the local solar altitude angle.
[0075] When each satellite uses the obtained set of observation windows to plan and optimize local observation schemes, it essentially selects a set of observation windows while satisfying all hard constraints. This is done to maximize local observational gains. To obtain sufficient observational information, any observation window... It must meet filtering constraints including observation type matching, yaw range, minimum startup time, solar altitude angle, and resolution.
[0076]
[0077]
[0078]
[0079]
[0080]
[0081] Meanwhile, to ensure the effectiveness of the entire observation sequence, the time interval constraint between adjacent windows needs to be considered. If yes The immediately adjacent window must meet the following requirements.
[0082]
[0083] To ensure an effective supply of energy, such as onboard storage space and satellite power, each satellite... observation window sequence The following cumulative constraints must be satisfied:
[0084]
[0085] Under the aforementioned hard constraints, each satellite uses a heuristic-based construction method to generate an initial sequence of observation windows to be executed:
[0086] First, the planned observation tasks are sorted in descending order of the single observation benefit of the task;
[0087] Secondly, for each observation task, based on filtering constraints to Filter individual observation windows that meet the constraints, and select unselected observation windows that do not conflict with the selected windows;
[0088] Next, the incremental observation benefit brought by each observation window is calculated sequentially. For point targets, the increment is the sum of the single observation benefits of tasks that meet all observation requirements within that window; for region targets, the observation benefit corresponding to the observation window is calculated using a discretized grid partitioning method. Specifically, all region targets are divided into discrete grids of a certain size and shape, each grid... From its center point This indicates the observation strip corresponding to the observation window. In other words, its effective coverage area is the area where all center points fall after considering existing coverage. The sum of the newly added grid areas in the data, its observation benefit is the percentage of effective coverage area and the total area observation task. income The product of, i.e.:
[0089]
[0090] in, The center point belongs only to the strip The number of grid cells.
[0091] From there, the windows to be selected are sorted in order of increasing revenue increment, increasing ratio of the number of paired conflicts with other windows to the observation revenue, and increasing window start time. The first observation window is then selected and added to the current sequence of observation windows to be executed.
[0092] Finally, the observation windows corresponding to the required observation frequencies for all tasks are selected to generate the initial observation task planning scheme for each satellite. .
[0093] Step S4: Each satellite agent broadcasts its current mission planning scheme to the entire network and receives mission planning schemes from other satellite agents.
[0094] Each satellite sends its initial mission planning scheme to the entire network in the form of a topic through data distribution services such as DDS. At the same time, each satellite obtains the initial mission planning schemes of other satellites as a subscriber to the relevant topic.
[0095] Step S5: Each satellite agent optimizes its own task planning scheme locally based on the task planning schemes received from neighboring satellite agents, and obtains the optimized task planning scheme.
[0096] Each satellite recalculates its observational gains after executing the current window sequence based on the received satellite agent planning scheme. For all given satellite mission planning schemes, the global gain f of the space-based remote sensing system is calculated as follows:
[0097]
[0098] in, and These represent the number of point target tasks and the number of area tasks, respectively. The number of valid score observations for the point target task. and These are the required number of observations and the effective number of observations, respectively. To find the minimum of the two, Indicates the first A collection area of tasks in each region. Indicates all valid bands and The intersection area The area is the region.
[0099] Each satellite recalculates the observation gains after executing the current window sequence based on the planning schemes received from other satellite agents. According to the boundary benefit principle, in the distributed mission planning involved in this invention, each satellite agent... The local benefit is defined as the increment it brings to the overall observation benefits of the space-based remote sensing system, specifically:
[0100]
[0101] in, For satellite Point-to-point tasks under the current allocation strategy The number of valid score observations For satellite The resulting regional tasks The effective observation area.
[0102] Each satellite, based on the mission planning schemes and local benefit functions of other satellites, uses the method in step 3 to readjust and optimize its own planning scheme, thus obtaining an improved planning scheme. Here, the planning scheme that maximizes local benefits is called the satellite's optimal response, and the difference in local benefits between the optimal response and the old scheme is called the satellite's regret value.
[0103] Step S6: Each satellite agent calculates its own regret value, which is the difference between the local benefits of the optimized mission planning scheme and the current mission planning scheme. The regret value is then broadcast to the entire network, while the regret values of other satellite agents are received.
[0104] The satellite intelligent agent In the decision-making round Regret value Represented as:
[0105]
[0106] in, This is the optimized task planning scheme obtained in step S5. This indicates the current task planning scheme. This is a function for calculating local benefits.
[0107] Similarly, each satellite publishes current regret value information in its regret value topic via DDS communication service. Each satellite subscribes to and obtains current regret value information for all satellites through the DDS communication service.
[0108] Step S7: Determine whether the regret values of all satellite agents meet the preset convergence conditions;
[0109] If the conditions are met, the coordination and optimization process is stopped, and the current task planning scheme of each satellite agent is used as the final task allocation result.
[0110] If not satisfied, proceed to step S8;
[0111] In step S7, each satellite uses the acquired regret value information to determine whether the current iteration process has converged. The preset convergence condition is that the regret value of all satellite agents is 0. That is, if the regret value of all individuals is 0, it indicates that the current system planning scheme has formed a Nash equilibrium, the coordination process ends, and the satellite uses the current task allocation and planning results as the final scheme for observation operations; otherwise, it proceeds to step 8 for distributed negotiation.
[0112] Step S8: Each satellite agent iteratively updates its current task planning scheme using the optimized task planning scheme according to the preset game learning rules, and then jumps to step S6.
[0113] Asynchronous Coordinated Update: In this coordinated update mode, only one satellite agent is selected in each round to adjust its mission planning scheme, which is an asynchronous serial coordination. The selection of satellite agents in each round can be achieved using regret values, randomly generated numbers, or pre-specified methods. During each loop, only that satellite calculates the locally optimal response based on the planning schemes of its neighbors, and updates its current planning scheme until the regret values of all individuals are 0.
[0114] Synchronous Coordinated Update: In each cycle of this coordinated update mode, each satellite updates its local mission planning scheme synchronously and in parallel based on the relative relationship between its own regret value and the regret values of its neighbors.
[0115] In some embodiments of the present invention, in step S8, the preset game learning rule is an asynchronous coordination update rule, specifically:
[0116] In each iteration, the satellite agent with the strictest regret value among all satellite agents, or the satellite agent with the smallest ID among all satellite agents with the largest regret value, is selected.
[0117] The selected satellite agent updates its current mission planning scheme to the optimized mission planning scheme obtained in step S5. ;
[0118] Satellite agents that were not selected will retain their current planning schemes.
[0119] In some embodiments of the present invention, in step S8, the preset game learning rule is a synchronous coordination update rule, specifically:
[0120] In each iteration, each satellite agent compares its own regret value with the regret values of all its neighboring satellite agents;
[0121] If a satellite agent's regret value is significantly greater than the regret values of all its neighbors, or if its regret value is equal to the maximum regret value among its neighbors but its ID is less than the IDs of all neighbors with that maximum regret value, then the satellite agent updates its current planning scheme to the optimized task planning scheme obtained in step S5. ;
[0122] Otherwise, the satellite agent will maintain its current mission planning scheme.
[0123] Figure 2 This example illustrates the steps in an iteration of multi-agent synchronous game learning. Each distributed synchronous collaborative decision-making process includes two communication rounds. In the first communication round, each satellite agent first sends its current planning decision and receives neighborhood information, then uses a single-satellite planning algorithm to calculate an improved planning scheme and a regret value. In the second communication round, the satellite agent broadcasts the regret value to its neighborhood and receives it synchronously, then uses designed neighborhood rules to choose whether to update the current planning scheme or leave it unchanged.
[0124] Figure 3 The graph shows the convergence curve of the global evaluation index generated by the algorithm and the comparison results between it and typical centralized methods. The horizontal axis represents the number of iterations, and the vertical axis represents the global evaluation index or reward. Experimental results show that the multi-satellite distributed collaborative task planning method based on multi-agent game theory proposed in this invention can achieve a fast solution for efficient task planning schemes that are close to those of typical centralized task planning methods in terms of global evaluation index, without relying on centralized control nodes, through inter-satellite information interaction and autonomous negotiation.
[0125] According to one aspect of the present invention, an electronic device is provided, comprising: one or more processors, one or more memories, and one or more computer programs; wherein the processor is connected to the memory, and the one or more computer programs are stored in the memory; when the electronic device is running, the processor executes the one or more computer programs stored in the memory to cause the electronic device to perform a multi-agent game-based multi-star distributed cooperative task planning method as described in any of the above technical solutions.
[0126] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.
[0127] The memory can be an internal storage unit of the terminal device, such as a hard drive or RAM. Alternatively, it can be an external storage device, such as a plug-in hard drive, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card. Furthermore, the memory can include both internal and external storage units. The memory is used to store the computer program and other programs and data required by the terminal device. It can also be used to temporarily store data that has been output or will be output.
[0128] According to one aspect of the present invention, a computer-readable storage medium is provided for storing computer instructions, which, when executed by a processor, implement a multi-agent game-based multi-star distributed cooperative task planning method as described in any of the above technical solutions.
[0129] For example, computer-readable storage media can be read-only memory (ROM), random access memory (RAM), read-only optical disc (CD-ROM), magnetic tape, floppy disk, and optical data storage devices. They can be implemented using computer-executable program code, thus allowing them to be stored in a storage device for execution by a computing device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Therefore, this invention is not limited to any particular hardware and software combination.
[0130] In summary, the present invention provides a multi-star distributed cooperative task planning method based on multi-agent game theory, which utilizes agent game theory learning to solve the multi-star cooperative task planning problem in a distributed environment.
[0131] Specifically, in the distributed satellite mission planning process, the overall goal of the system is to obtain the optimal or near-optimal planning result in terms of global evaluation indicators or global benefits. However, due to the lack of centralized control nodes and global information, conflicts of interest inevitably arise among the satellite agents when making autonomous decisions, making distributed conflict resolution, decision convergence, and efficient collaboration challenging. As a modern branch of mathematics that studies the equilibrium behavior among rationally interacting individuals, game theory provides a new research framework and theoretical tool for distributed collaborative decision-making. By constructing a distributed decision-making model based on game theory, concepts such as potential games, optimal responses, and Nash equilibrium can be used to design game learning strategies based on neighborhood interactions. This ensures global performance of system behavior while achieving algorithm convergence, realizing autonomous collaboration and intelligent emergence without centralized control.
[0132] Therefore, multi-agent game theory provides an effective model representation for multi-star distributed collaborative task planning;
[0133] Secondly, as an NP-hard combinatorial optimization problem with complex aerospace constraints, it can be transformed into a Nash equilibrium solution process through game theory modeling. Various game learning algorithms can then be used to solve and refine the Nash equilibrium in a distributed environment. Therefore, through the design of game learning algorithms, efficient solutions for efficient collaborative observation schemes can be achieved under the constraint of local information interaction.
[0134] Based on the above reasons, this invention uses multi-intelligent potential game as a modeling tool to transform the multi-satellite distributed collaborative task planning problem into a process of solving and refining Nash equilibrium solutions. It constructs a global evaluation index for mixed task conditions of point objectives and regional objectives, designs local reward functions for each satellite agent, and gives distributed autonomous decision-making rules based on game learning, providing a more efficient solution for multi-satellite autonomous collaborative task planning in a distributed environment.
[0135] This invention enables various constraints to be satisfied by each satellite agent during mission planning. Because each satellite agent considers all filtering constraints, pairwise constraints, and cumulative constraints when calculating planning schemes, the resulting mission planning satisfies all aerospace physical constraints.
[0136] This invention enables efficient solutions for near-optimal task planning schemes without relying on a centralized decision-making center. Since the optimal decision outcome in a potential game model is necessarily a Nash equilibrium, efficient decision-making can be achieved through distributed solutions for Nash equilibrium schemes. Furthermore, because the distributed collaborative strategy guarantees the existence of a finite number of improvement paths, the system will inevitably converge to the Nash equilibrium planning scheme after a finite number of communication rounds and iterations. By distributing the computational burden of centralized decision nodes to each satellite agent platform, this method is particularly suitable for autonomous collaborative decision-making in large-scale satellite constellations under distributed communication environments.
[0137] This invention can also be applied to distributed task allocation or planning problems in other fields, including but not limited to logistics order delivery, drones, and unmanned vehicles.
[0138] Furthermore, it should be noted that the embodiments of the present invention may take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code.
[0139] Embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0140] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing terminal equipment to cause a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0141] It should also be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.
[0142] Finally, it should be noted that the above description represents a preferred embodiment of the present invention. It should be pointed out that although preferred embodiments have been described, those skilled in the art, once they understand the basic inventive concept of the present invention, can make various improvements and modifications without departing from the principles described herein. These improvements and modifications should also be considered within the scope of protection of the present invention. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present invention.
Claims
1. A multi-agent game-based distributed collaborative task planning method, characterized in that, Includes the following steps: Step S1: Construct a multi-satellite distributed network system without a master star, in which each satellite agent has the same equal status; Step S2: The observation task is either injected from the ground to any satellite or generated autonomously by any satellite, and broadcast throughout the network via the multi-satellite distributed network system to all satellite agents; Step S3: Each satellite agent performs converged analysis on the received tasks, and plans and optimizes local observation schemes based on task type, its own status and capabilities to generate an initial task planning scheme. Step S4: Each satellite agent broadcasts its current mission planning scheme to the entire network and receives mission planning schemes from other satellite agents. Step S5: Each satellite agent optimizes its own task planning scheme locally based on the task planning schemes received from neighboring satellite agents, and obtains the optimized task planning scheme. Step S6: Each satellite agent calculates its own regret value, which is the difference between the local benefits of the optimized mission planning scheme and the current mission planning scheme. The regret value is then broadcast to the entire network, while the regret values of other satellite agents are received. Step S7: Determine whether the regret values of all satellite agents meet the preset convergence conditions; If the conditions are met, the coordination and optimization process is stopped, and the current task planning scheme of each satellite agent is used as the final task allocation result. If not satisfied, proceed to step S8; Step S8: Each satellite agent iteratively updates its current task planning scheme using the optimized task planning scheme according to the preset game learning rules, and then jumps to step S6.
2. The method according to claim 1, characterized in that, In step S1, the multi-star distributed network system without a master star is composed of... Composed of remote sensing satellites, represented as , Each satellite Defined as: in, and These represent the satellite's unique identifier and network address, respectively. Indicates satellite orbit information, Indicates the load field of view angle, Indicates load type, Indicates the maximum lateral sway angle, Indicates the shortest boot time, Indicates attitude yaw rate, Indicates the longest startup time per lap. Indicates the maximum observation resolution, This indicates the local minimum solar altitude angle required for the payload to conduct effective observations. Each satellite intelligent agent uses inter-satellite links to form a real-time interconnected, masterless peer network.
3. The method according to claim 1, characterized in that, In step S2, the set of observation tasks is denoted as , Each task Defined as: in, Indicates the observational gains of the task, Indicates the type of task, Indicates the latitude and longitude coordinates of the mission. Indicate the required observation type, [ Indicates the desired observation time interval, This indicates the minimum resolution required to meet the requirements. This indicates the frequency of observations within a certain period of time. This represents the minimum time difference that must be allowed between two effective imaging missions.
4. The method according to claim 3, characterized in that, In step S3, the local observation scheme planning and local optimization specifically includes: Step S301: Each satellite agent calculates the visible observation window for each mission based on its own orbital parameters and mission geographical location; Step S302: Construct a set of feasible observation windows that satisfy the filtering constraints, wherein the filtering constraints include at least: observation type matching, side swing angle range, minimum power-on time, solar altitude angle requirement and resolution requirement; Step S303: Under the premise of satisfying the attitude transition time constraint and the cumulative power-on time constraint, with the goal of maximizing local observation benefits, select a series of observation windows from the set of feasible observation windows to form the initial task planning scheme. .
5. The method according to claim 1 or 4, characterized in that, The calculation method for the local benefits is as follows: satellite intelligent agent Local benefits The incremental benefit it brings to the overall system observation is as follows: in, and These represent the number of point target tasks and the number of area tasks, respectively. Indicates the first A collection area of tasks in each region. Indicates all valid bands and The intersection area For the area, For satellite Point-to-point tasks under the current task planning scheme The number of valid score observations For satellite The resulting regional tasks The effective observation area.
6. The method according to claim 1, characterized in that, In step S5, the local optimization of the planning scheme specifically includes: Knowing the current mission planning schemes of other satellites, the satellite agent replans its own observation window sequence to maximize the local gains, thus obtaining an optimized mission planning scheme. And called the optimized task planning scheme This is the best response for the satellite intelligent agent.
7. The method according to claim 1, characterized in that, In step S6, the satellite intelligent agent In the decision-making round Regret value Represented as: in, This is the optimized task planning scheme obtained in step S5. This indicates the current task planning scheme. This is a function for calculating local benefits.
8. The method according to claim 1, characterized in that, In step S7, the preset convergence condition is: the regret value of all satellite agents is 0.
9. The method according to claim 1, characterized in that, In step S8, the preset game learning rule is an asynchronous coordination update rule, specifically: In each iteration, the satellite agent with the strictest regret value among all satellite agents, or the satellite agent with the smallest ID among all satellite agents with the largest regret value, is selected. The selected satellite agent updates its current mission planning scheme to the optimized mission planning scheme obtained in step S5. ; Satellite agents that were not selected will retain their current planning schemes.
10. The method according to claim 1, characterized in that, In step S8, the preset game learning rule is a synchronous coordination update rule, specifically: In each iteration, each satellite agent compares its own regret value with the regret values of all its neighboring satellite agents; If a satellite agent's regret value is significantly greater than the regret values of all its neighbors, or if its regret value is equal to the maximum regret value among its neighbors but its ID is less than the IDs of all neighbors with that maximum regret value, then the satellite agent updates its current planning scheme to the optimized task planning scheme obtained in step S5. ; Otherwise, the satellite agent will maintain its current mission planning scheme.
