Giant constellation operation and maintenance and resource management and control system and method for computing network integration

Abstract

The invention discloses a giant constellation operation and maintenance and resource management and control system and method oriented to computing network fusion, the system comprises # imgabs0 # MEO satellites and # imgabs1 # LEO satellites, and each MEO satellite manages a plurality of LEO satellites in each time slot of each scheduling period; each LEO satellite is used for acquiring data of a calculation task in a preset time period, and executing data calculation in each time slot # imgabs2 # of each scheduling period in the preset time period according to resource scheduling information which is sent by the MEO satellite and aims at the time slot # imgabs3 #; and each MEO satellite is used for generating a network by adopting the trained resource scheduling information, generating resource scheduling information of a time slot # imgabs5 # according to a resource task state of the MEO satellite in the time slot # imgabs4 #, sending the resource scheduling information to each LEO satellite managed in the time slot # imgabs6 #, and performing collaborative data calculation of the managed LEO satellites according to the resource scheduling information of the time slot # imgabs7 #. The method has higher real-time performance, better resource management and control scheduling capability and higher task completion rate.

Description

Technical Field

[0001] The present invention belongs to the field of satellite communication technology, and specifically relates to a giant constellation operation and maintenance and resource management system and method for computing-network integration. Background Art

[0002] A giant constellation is a satellite network system consisting of hundreds to tens of thousands of satellites, which can provide all-time and continuous communication services. Hybrid multi-layer giant constellations, including low earth orbit (LEO) and medium earth orbit (MEO) satellites, have gradually become the first choice for coping with complex mission requirements due to their wide coverage, fast response speed and low communication delay. With the explosive growth of mission requirements, the drawbacks of the model of satellite backhaul to ground data centers for processing have gradually become prominent. On-orbit computing can process data directly on the satellite and calculate data from sensors or other devices without transmitting the data back to the ground, reducing communication delays, improving response speed, and effectively saving bandwidth and ground processing resources. At present, it has become a trend to actively launch and deploy high-computing satellites to provide powerful data processing and real-time decision-making capabilities for multiple fields.

[0003] Therefore, it is urgent to propose a new satellite operation and maintenance control architecture, which can coordinate the computing resources of multiple satellites through intersatellite links to form an intelligent operation and maintenance control mechanism with deep integration of computing power and network. Give full play to the synergy of giant constellation computing and communication resources, realize more flexible and efficient task scheduling and resource allocation, support more complex space applications, and promote the development of satellite networks in a more autonomous and intelligent direction. However, this multi-level giant constellation structure also brings huge challenges to operation and maintenance and resource management. The current satellite network operation and maintenance and resource management methods mainly rely on independent satellite computing or only schedule planning for static time slots. However, in giant constellation networks, efficient computing, communication and energy resource allocation and scheduling between different orbital layers bring huge challenges. First, how to coordinate the allocation of computing and communication resources between satellites at different levels to improve task processing efficiency is an important issue facing operation and maintenance and resource management; second, the tasks and resources in satellite networks are highly dynamic and uncertain, and the task load, resource availability and communication link status change over time. Traditional static control strategies cannot effectively cope with these changes. Therefore, how to introduce the idea of ​​computing-network collaboration into giant constellation networks and redesign the operation and maintenance and network management architecture and methods has become an important development direction for future giant constellations.

[0004] At present, there are some studies on the architecture and methods of giant constellation operation and maintenance and resource management. However, these studies have the following problems: The current architecture and scheduling methods are often limited to fixed or predictable tasks and cannot cope with dynamic and complex task requirements; the current methods are limited to collaborative computing through the inter-satellite transmission of a large amount of actual task data, which will occupy a large amount of communication bandwidth and cause high latency and task response time; at the same time, the existing satellite computing collaboration methods also rely on the management and operation of ground control stations; therefore, it is impossible to obtain higher real-time performance and better resource management and scheduling capabilities. Summary of the invention

[0005] In order to solve the above problems existing in the prior art, the present invention provides a giant constellation operation and maintenance and resource management system and method for computing-network integration.

[0006] The technical problem to be solved by the present invention is achieved through the following technical solutions: The present invention provides a giant constellation operation and maintenance and resource management system for computing-network integration, comprising: MEO satellite collaborative management and control architecture with The LEO satellite edge-level management and control architecture of multiple LEO satellites, MEO satellites and LEO satellites are connected by communication links, and each MEO satellite manages multiple LEO satellites in each time slot of each scheduling cycle; Each LEO satellite is used to obtain data of the computing task in a preset time period and in each time slot of each scheduling cycle in the preset time period. According to the time slot sent by the MEO satellite resource scheduling information to perform data calculation of the computing task; Each MEO satellite is used to acquire itself in the time slot The resource task status is obtained by using the trained resource scheduling information to generate the network according to its own The resource task status is generated for the time slot The resource scheduling information is sent to the Each LEO satellite managed within the The resource scheduling information is used to perform collaborative data calculations of the computing tasks of each LEO satellite managed; the time slot The resource scheduling information indicates the time slot In the computing tasks of each LEO satellite managed by the MEO satellite, the ratio of the amount of data calculated by the MEO satellite to the total amount of data in the computing tasks of the LEO satellite; the value range of the resource scheduling information is 0 to 1.

[0007] The present invention also provides a giant constellation operation and maintenance and resource management method for computing-network integration, which is applied to each MEO satellite in the above giant constellation operation and maintenance and resource management system for computing-network integration. The method includes: Get the resource task status of the current time slot in the current scheduling cycle; The resource scheduling information generation network is trained to generate resource scheduling information for the current time slot according to its own resource task status in the current time slot; the trained resource scheduling information generation network is trained by using local reinforcement learning and federated learning methods; the resource scheduling information of the current time slot represents the proportion of the amount of data calculated by the MEO satellite to the total amount of data of the LEO satellite's computing tasks in the computing tasks of each LEO satellite managed in the current time slot; Sending resource scheduling information for the current time slot to each LEO satellite managed in the current time slot; Determine, according to resource scheduling information for the current time slot, whether to perform data calculation of the computing task of each managed LEO satellite in the current time slot, and the amount of data that needs to be calculated when performing the data calculation; When data calculation of the computing task of each managed LEO satellite needs to be performed in the current time slot, the corresponding data amount of the computing task of each managed LEO satellite is calculated in the current time slot according to the data amount that needs to be calculated.

[0008] Compared with the prior art, the present invention has the following beneficial effects: The system proposed by the present invention includes two-level architectures, namely, a MEO satellite collaborative level and a LEO satellite edge level. Each MEO satellite in the MEO satellite collaborative level architecture can obtain its own real-time resource task status, generate real-time resource scheduling information according to its own real-time resource task status, and perform collaborative data calculation of computing tasks of each managed LEO satellite according to the real-time resource scheduling information, and enable each managed LEO satellite to perform data calculation of computing tasks according to the real-time resource scheduling information. Therefore, the real-time allocation of computing tasks of the entire satellite network and the comprehensive utilization of inter-layer resources can be performed through the MEO satellite collaborative level architecture. The edge-level architecture can perform real-time collaborative execution of real-time computing and communication tasks of edge nodes, so that resources can be dynamically scheduled and efficiently utilized according to real-time mission requirements without relying on the control of the ground controller. It is not limited to fixed or predictable tasks. In addition, there is no problem of high latency and mission response time caused by transmitting large amounts of mission data between different MEOs to enable collaborative computing between MEO satellites. It can achieve efficient computing scheduling for dynamic and complex mission requirements, improve the mission completion volume of the overall satellite network, and not only has higher real-time performance and better resource control and scheduling capabilities, but also has a higher mission completion rate.

[0009] The present invention will be further described in detail below with reference to the accompanying drawings and specific implementation methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 It is a schematic diagram of the architecture of a giant constellation operation and maintenance and resource management and control system for computing-network integration provided by an embodiment of the present invention; Figure 2 It is another schematic diagram of the architecture of a giant constellation operation and maintenance and resource management system for computing-network integration provided by an embodiment of the present invention; Figure 3 It is a flow chart of a method for giant constellation operation and maintenance and resource management and control oriented to computing-network integration provided by an embodiment of the present invention; Figure 4 is a schematic diagram of task completion obtained through a simulation experiment provided by an embodiment of the present invention; Figure 5 This is a diagram of energy consumption results obtained through simulation experiments provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0011] The present invention is further described in detail below with reference to specific embodiments, but the embodiments of the present invention are not limited thereto.

[0012] The present invention provides a giant constellation operation and maintenance and resource management system for computing-network integration, the system comprising: MEO satellite collaborative management and control architecture with LEO satellite edge-level control architecture for LEO satellites, and are all positive integers greater than 1. MEO satellites and LEO satellites are connected through communication links, and each MEO satellite manages multiple LEO satellites in each time slot of each scheduling cycle. The number of LEO satellites managed by each MEO satellite in different time slots is different or the same. The number of LEO satellites managed by different MEO satellites in different time slots is the same or different.

[0013] Each LEO satellite is used to obtain the data of the computing task in the preset time period and in each time slot of each scheduling cycle in the preset time period. According to the time slots sent by the MEO satellite Each MEO satellite is used to obtain its own resource scheduling information in the time slot. The resource task status is determined by using the trained resource scheduling information to generate a network based on its own The resource task status is generated for the time slot The resource scheduling information is sent to the Each LEO satellite managed within the Resource scheduling information, perform collaborative data calculations for computing tasks of each LEO satellite managed; time slot The resource scheduling information is represented in the time slot The ratio of the amount of data calculated by the MEO satellite to the total amount of data in the computing tasks of the LEO satellites managed within the MEO satellite. Here, each LEO satellite collects data from the outside world according to the computing task, so as to obtain the data of the corresponding computing task. For example, the data of the ecological monitoring task is obtained by ecological monitoring of the environment in which it is located, the data of the meteorological monitoring task is obtained by meteorological observation of the environment in which it is located, and the data of the computing task related to the remote sensing image is obtained by acquiring the remote sensing image of the environment in which it is located, and so on. Each computing task corresponds to a preset time period to indicate that the calculation of the computing task needs to be performed within the time period and the calculation of the computing task needs to be completed. Exemplarily, Figure 1 is a schematic diagram of the architecture of a giant constellation operation and maintenance and resource management system for computing-network integration provided by an embodiment of the present invention. Figure 3 As shown, the system includes three MEO satellites. There are communication links between the MEO satellites, and there are communication links between the MEO satellites and the LEO satellites. The area formed by the dotted line under each MEO satellite represents the signal radiation range of the MEO satellite. Figure 3The semicircle in the figure represents the earth, and the arrow from the earth to LEO represents the acquisition of ground data from LEO. Each MEO satellite manages multiple LEO satellites in each time slot, and each LEO satellite can obtain data for computing tasks from the outside world. For example, it can obtain task data by receiving data from ships, cars and other equipment on the ground.

[0014] Here, the value range of the resource scheduling information is 0 to 1, so that according to the resource scheduling information of each time slot, the MEO satellite and the LEO satellite can determine whether to perform data calculation of the computing task in the time slot, and the amount of data that needs to be calculated when the data calculation of the computing task is required; wherein, when the value range of the resource scheduling information of a time slot is 0, it means that in the time slot, the MEO satellite does not need to undertake any data calculation of the total amount of data required to be calculated in the computing task of each LEO satellite managed in the time slot, and each LEO satellite managed by the MEO satellite needs to undertake the data calculation of the total amount of data required to be calculated in the computing task of its own in the time slot; when the value range of the resource scheduling information is 0, it means that in the time slot, the MEO satellite does not need to undertake any data calculation of the total amount of data required to be calculated in the computing task of each LEO satellite managed in the time slot, and each LEO satellite managed by the MEO satellite needs to undertake the data calculation of the total amount of data required to be calculated in the computing task of its own in the time slot; When it is 1, it means that the MEO satellite needs to perform data calculations for all the data required to be calculated in the time slot in the calculation tasks of each LEO satellite it manages, while each LEO satellite managed by the MEO satellite does not need to undertake any data calculations for all the data required to be calculated in its own calculation tasks in the time slot; when the value of the resource scheduling information of a time slot is 0.4, it means that in the time slot, the MEO satellite needs to undertake 40% of the data calculations for all the data required to be calculated in the time slot in the calculation tasks of each LEO satellite it manages, while each LEO satellite managed by the MEO satellite needs to undertake 60% of the data calculations for all the data required to be calculated in its own calculation tasks in the time slot. Exemplarily, the value of the resource scheduling information can be one of {0, 0.2, 0.4, 0.6, 0.8, 1}. Based on this, each LEO satellite is specifically used to calculate the total amount of data required to be calculated in the time slot according to the resource scheduling information sent by the MEO satellite. Resource scheduling information to determine the time slot Whether to perform data calculation of the computing task within the time slot, and the amount of data required to be calculated when performing data calculation, and the amount of data required to be calculated in the time slot Each MEO satellite is specifically used to calculate the corresponding data volume of the computing task according to the time slot. Resource scheduling information to determine the time slot Whether to perform data calculation for each LEO satellite computing task managed within the time slot, and the amount of data required to be calculated during the data calculation, and the amount of data required to be calculated in the time slot The corresponding data volume of the computing task of each LEO satellite managed is calculated within.

[0015] Figure 2 FIG. 1 is another schematic diagram of the architecture of the giant constellation operation and maintenance and resource management system for computing-network integration provided by the present invention. Figure 2 As shown, each MEO satellite includes: a resource task status monitoring module, a data calculation processing module and a computing network fusion scheduling optimization module. The resource task status monitoring module is used to obtain each time slot of each scheduling cycle. The resource usage and computing task execution status of the MEO satellite itself and each LEO satellite it manages are obtained in the time slot. It should be noted that resource usage includes remaining power and communication resources; computing task execution status includes: whether computing tasks are being executed, and the amount of data remaining to be calculated, etc. In the present invention, each MEO satellite is in a time slot Resource task status Includes: MEO satellites in time slots Energy state , Link Status and onboard mission data status ; Among them, MEO satellites are in time slots Energy state characterization of MEO satellites in time slots The remaining power in the MEO satellite Link state representation: In the time slot The sum of the link capacities of the communication links between the MEO satellites and the managed LEO satellites; the MEO satellites in the time slot On-board mission data status representation: In the time slot The sum of the data volumes of the LEO satellite computing tasks managed by the MEO satellite.

[0016] The data calculation and processing module is used to When the data calculation of the calculation task of each LEO satellite managed is needed within the time slot, the data of the calculation task of each LEO satellite managed is received, and the amount of data to be calculated is calculated in the time slot The corresponding data volume of the computing task of each LEO satellite managed is calculated within the time slot. It should be noted that when the MEO satellite needs to perform data calculations for the computing tasks of each LEO satellite managed in a certain time slot, the corresponding data volume is obtained from each LEO satellite managed and calculated. The computing network fusion scheduling optimization module is used to train the resource scheduling information generation network using local reinforcement learning and federated learning methods to obtain the trained resource scheduling information generation network, and in the process of training the resource scheduling information generation network, it is combined with the remaining The resource scheduling information is shared and aggregated among the MEO satellites to generate the network parameters of the network, and the trained resource scheduling information is used to generate the network according to the MEO satellite's own time slot The resource task status is generated for the time slot The resource scheduling information will be targeted at the time slot The resource scheduling information is sent to In some embodiments, continue to refer to the above Figure 2 The computing network fusion scheduling optimization module can include an intelligent learning submodule and a cooperative sharing submodule. The intelligent learning submodule is used to train the resource scheduling information generation network using local reinforcement learning and federated learning methods to obtain the trained resource scheduling information generation network, and use the trained resource scheduling information generation network to generate the resource scheduling information according to the MEO satellite in the time slot. The resource task status is generated for the time slot The cooperative sharing submodule is used to share the resource scheduling information with the remaining MEO satellites share and aggregate resource scheduling information to generate network parameters, and The resource scheduling information is sent to Every LEO satellite managed within.

[0017] In the present invention, Figure 2 As shown, each LEO satellite includes: a data acquisition module, a resource task status monitoring module, and a data calculation and processing module. The data acquisition module is used to collect task data in the environment in which the LEO satellite itself is located. The resource task status monitoring module is used to obtain the computing resources, communication resources and computing task execution status of the LEO satellite itself in real time. The data calculation and processing module is used to process the collected data and perform edge computing tasks. On this basis, in some embodiments, when each LEO satellite among the multiple LEO satellites managed by the same MEO satellite performs corresponding data calculations according to real-time resource scheduling information, it can use existing algorithms to perform collaborative calculations with the remaining LEO satellites among the multiple LEO satellites managed by the same MEO satellite. Based on this, as Figure 2As shown, each LEO satellite also includes a performance optimization module, which is used to obtain the LEO satellite's own operating parameters and the resource scheduling strategy for the remaining LEO satellites based on the status of the LEO satellite's own computing resources and communication resources and the computing task execution status, using the existing LEO satellite collaborative scheduling method to improve system performance. It should be noted that the LEO satellite's own operating parameters and the resource scheduling strategy for the remaining LEO satellites need to be determined based on the specific existing algorithms used by the LEO satellite, and the present invention does not limit this. Based on this, the present invention, on the basis of introducing the computing and communication coordination mechanism and the strategic parameter coordination mechanism between MEO satellites, can further improve the task scheduling capability of the entire satellite network by introducing the collaborative scheduling and control mechanism of inter-layer satellites.

[0018] In the present invention, the trained resource scheduling information generation network is trained by local reinforcement learning and federated learning methods with the minimum energy consumption of the system in a scheduling cycle as the optimization goal, and with the energy constraints of each LEO satellite, the energy constraints of each MEO satellite, the computing constraints of each LEO satellite, the computing constraints of each MEO satellite, and the communication constraints of each LEO satellite as the constraints of the optimization goal. Exemplarily, the resource scheduling information generation network is an Actor network.

[0019] Exemplarily, the expressions of the energy constraint conditions of each LEO satellite and the energy constraint conditions of each MEO satellite are as follows: ; ; in, , , , For the system LEO satellites constitute a LEO satellite collection, is the first satellite in the LEO satellite collection LEO satellites, The value of , For a scheduling cycle In each time slot, For the system The MEO satellite cluster consists of MEO satellites. It is the first in the MEO satellite collection MEO satellites, The value of , For the Maximum battery capacity for LEO satellites, The percentage of the maximum discharge depth of the LEO satellite describes the lowest voltage level that the battery can be safely discharged to during use. In other words, exceeding the maximum discharge depth may cause a decrease in battery capacity or performance degradation. For the The amount of power available for each LEO satellite in each time slot, For the The amount of electricity charged by a LEO satellite in each time slot, For the The transmission energy consumption of a LEO satellite in each time slot is For the The computational energy consumption of each LEO satellite in each time slot; For the Maximum battery capacity for a MEO satellite, is the percentage of the maximum discharge depth of the MEO satellite, For the The amount of power available for each MEO satellite in each time slot, For the The amount of electricity charged by a MEO satellite in each time slot, For the The receiving energy consumption of a MEO satellite in each time slot is: For the The calculated energy consumption of a MEO satellite in each time slot. It should be noted that the amount of electricity charged in each time slot of each satellite in the system and the link capacity between satellites in the system need to be acquired in advance. Specifically, it can be obtained by pre-acquiring the satellite link capacity matrix and satellite energy acquisition matrix of the system in the required time period. The satellite link capacity matrix includes the link capacity matrix of each time slot between all LEO satellites and MEO satellites in the system in each scheduling cycle, and each element in the link capacity matrix represents the link capacity between a certain LEO and a certain MEO in the time slot, or the link capacity between two MEOs. 0 in the link capacity matrix means that the corresponding satellites are not visible, and other values ​​greater than 0 in the link capacity matrix represent the link capacity size between the corresponding satellites. The satellite energy acquisition matrix includes: LEO satellite energy acquisition matrix and MEO satellite energy acquisition matrix. The LEO satellite energy acquisition matrix includes the dynamic energy acquisition matrix of the LEO satellite in the system in each time slot of each scheduling cycle, and the MEO satellite energy acquisition matrix includes the dynamic energy acquisition matrix of the MEO satellite in the system in each time slot of each scheduling cycle. Each element in the dynamic energy acquisition matrix of the LEO satellite represents the amount of charge charged by a corresponding LEO satellite in the time slot. When an element is 0, it means that the corresponding LEO satellite is covered by the shadow of the earth in the time slot and cannot obtain solar energy. The dynamic energy acquisition matrix of the MEO satellite is similar.

[0020] Exemplarily, the expression of the computational constraints of each MEO satellite is as follows: ; in, For the MEO satellites in each time slot The amount of data processed, Indicates the maximum computational upper limit of the MEO satellite.

[0021] Exemplarily, the expression of the computational constraints for each LEO satellite is as follows: ; in, For the LEO satellites in each time slot The amount of data processed, Indicates the maximum computational upper limit of LEO satellites.

[0022] Exemplarily, the expression of the communication constraint condition of each LEO satellite is as follows: ; in, For the LEO satellites In time slot A MEO satellite managed by It is a time slot from arrive The link capacity on the link of It is a time slot from arrive The amount of data transferred.

[0023] Exemplarily, the expression of the energy consumption of the system is as follows: ; ; ; ; ; in, The scheduling strategy set obtained by the resource scheduling information generation network for all MEO satellites, where the scheduling strategy is the network parameter of the resource scheduling information generation network, is the total energy consumption of the system in a scheduling cycle, represents the minimum total energy consumption of the system in a scheduling cycle, For each time slot The resource mission status of the satellite, Indicated in The set of scheduling strategies for all MEO satellites in state, For the The transmit power of each LEO satellite is For time slot The time slot length, For the Hardware calculation coefficients for LEO satellites, For the LEO satellites in each time slot The calculation rate of For the The received power of a MEO satellite, for Middle MEO satellites in the time slot Managed and MEO satellites A LEO satellite transmitting data, It is a time slot from arrive The link capacity on the link of It is in the time slot from arrive The amount of data transferred, For the MEO satellites The hardware calculation coefficient of For the MEO satellites In each time slot The calculation rate of .

[0024] The present invention takes into account the dynamic fluctuations in energy acquisition caused by the satellite's periodic switching between the earth's shadow area and the direct sunlight area, and based on these actual fluctuation characteristics, models the dynamic consumption and acquisition of energy corresponding to the satellite in the computing network collaboration process. Based on the modeling of the dynamic consumption and acquisition of energy corresponding to the satellite in the computing network collaboration process, an optimization target for training resource scheduling information to generate a network and constraints on the optimization target are constructed, making the present invention more efficient and more in line with the actual application environment, so that high-energy consumption computing tasks in actual satellite network scenarios can be efficiently completed, avoiding performance degradation or service interruption when performing high-energy consumption services such as on-board computing.

[0025] In some embodiments, the trained resource scheduling information generation network deployed on each MEO satellite is trained using the method described in S101 to S106 below. Take this as an example, the steps of this training method are as follows: S101, in The first round of training Get MEO satellites at time slots In the Resource task status of time slots ; Each round of training corresponds to a scheduling cycle, and each scheduling cycle has time slots, every preset update step A training is performed once per time slot; MEO satellite In the Resource task status of time slots This is the default state.

[0026] It should be noted that in each scheduling cycle, every preset update step A training is performed once per time slot. is a positive integer, and The value starts from 1, for example, , is a preset integer. In the first time slot of the first round of training, the MEO satellites acquired The resource task state in the first time slot is a preset state, and the preset state can be set according to actual needs, which is not limited in the present invention.

[0027] S102. According to The resource scheduling information obtained in the time slot generates the network and the Resource task status of time slots , determine the Resource scheduling information for time slots Among them, when Time The resource scheduling information obtained in the time slot generates the network The resource scheduling information generation network obtained from the last training in the round of training.

[0028] Specifically, the Resource task status of time slots Enter the In the resource scheduling information generation network for time slots, The resource scheduling information of time slots generates the network output Resource scheduling information for time slots .

[0029] S103, based on Resource scheduling information for time slots , determine the MEO satellite No. Rewards for time slots and Resource task status of time slots .

[0030] Specifically, MEO satellites According to Resource scheduling information for time slots After performing the corresponding data calculation, we get the MEO satellite No. Rewards for time slots , and MEO satellites No. Resource task status of time slots ,in, , For MEO satellites In the The time slot and the The total energy consumption of the local system consisting of LEO satellites managed by time slots is For MEO satellites In the The receiving energy consumption of each time slot is For MEO satellites In the The computational energy consumption of each time slot is For MEO satellites In the The set of LEO satellites managed by time slots, For MEO satellites In the Each LEO satellite managed by the time slot is The transmission energy consumption of each time slot is For MEO satellites In the Each LEO satellite managed by the time slot is The computational energy consumption of each time slot. The resource scheduling information of each time slot determines the MEO satellite No. The rewards for the time slot and the After the resource task status of the time slot, it also includes: Resource task status of time slots , No. Resource task status of time slots , No. Resource scheduling information for time slots and Rewards for time slots As an empirical sample , , , >Stored on MEO satellite experience buffer.

[0031] S104, based on , preset update step and The resource scheduling information generation network obtained in the first time slot determines whether to update the network parameters of the current resource scheduling information generation network, and obtains the first The resource scheduling information of time slots generates a network.

[0032] Specifically, yes and preset update step size Perform a remainder operation to obtain the second operation result; when the second operation result is 0, use the loss function and the experience samples in the experience buffer to update the network parameters of the current Actor network to obtain the first The Actor network obtained in the time slot is then updated using the loss function and the experience samples in the experience buffer to obtain the network parameters of the current Critic network. The Critic network obtained in the time slot; when the second operation result is not 0, the network parameters are not updated, and the first The Actor network obtained in the time slot is used as the An Actor network with time slots is sufficient.

[0033] Specifically, when updating the network parameters of the Actor network, the MEO satellite Get some experience samples from the experience buffer, and use the obtained experience samples and loss function Calculate the loss and use the gradient descent method to get the Actor network Network parameters for time slots , using the Actor network The network parameters of the time slot can be obtained Actor network obtained in time slots.

[0034] For example, The expression is as follows: ; ; in, is and The advantage function measures the degree of goodness of choosing a specific action (i.e., specific resource scheduling information) compared to the average choice of all possible actions under a given state. That is, the advantage function measures the Take Whether the reward is high or low relative to the average. is Take The action value function is expressed as Take After that, the expected value of all possible cumulative rewards in the future. is The state value function under The expected value of all possible cumulative rewards in the future. is the probability ratio of the strategy, where strategy represents the network parameters of the Actor network and the probability ratio of the strategy refers to the new strategy With the old strategy The probability ratio under the same state and action is currently The new strategy refers to the Actor network's The old strategy refers to the network parameters of the Actor network. The network parameters for each time slot, It is to trim the probability ratio of the strategy, and the trimming threshold is a hyperparameter used to control the step size of the policy update. By clipping, the probability ratio of the policy is limited to To prevent the policy from updating too much, represents mathematical expectation.

[0035] For example, The expression is as follows: ; in, Represents the Actor network The network parameters for each time slot, is the learning rate, which is used to control the step size of parameter update. Express Derivative.

[0036] Specifically, when updating the network parameters of the Critic network, some experience samples are obtained from the experience buffer of the MEO satellite, and the obtained experience samples and the loss function are used to update the network parameters of the Critic network. Calculate the loss and use the gradient descent method to get the Critic network Network parameters for time slots , using the Critic network Network parameters for time slots , we can get the Critic network with time slots.

[0037] For example, The expression is as follows: ; ; in, It is the estimated value of the expected cumulative reward after taking an action in the current state, which is a target value. R represents the current reward, is the Critic network's response to the next state The estimated value of the reward, representing the reward from the next state At the beginning, the expected value of all possible cumulative rewards in the future is the state output by the Critic network The estimate, is the discount factor, Used to measure The importance of the discount factor is closer to 1, indicating The larger the proportion in the target value, Represents the estimated value of the Critic network output With target value During the training process, the Critic network minimizes the target value and estimated values The mean square error between them is used to adjust its parameters so that its output is an estimate of Closer to the target value .

[0038] For example, The expression is as follows: ; in, is the first The network parameters for each time slot, is the learning rate, which is used to control the step size of parameter update. Yes Derivative.

[0039] S105, based on , preset aggregation step and The resource scheduling information obtained in each time slot generates a network to determine whether to use the remaining MEO satellites aggregate network parameters and obtain the latest The resource scheduling information obtained in each time slot generates a network.

[0040] Specifically, yes and preset aggregation step size Perform a remainder operation to obtain the first operation result; when the first operation result is 0, obtain the remainder The first MEO satellite The Actor network obtained in time slots and the preset aggregation update matrix MEO satellite With the remaining The first MEO satellite The network parameters of the Actor network obtained in each time slot are aggregated to obtain the MEO satellite No. The network parameters after the aggregation of time slots are obtained by using MEO satellite No. The network parameters after the aggregation of time slots are updated for MEO satellites No. Actor network obtained in time slots, MEO satellite The latest When the result of the first operation is not 0, the MEO satellite No. Actor network obtained in time slots as MEO satellite The latest The Actor network obtained in time slots. The expressions of network parameters after time slot aggregation are as follows: ; ; in, For the The network parameters after the time slot aggregation, for The first MEO satellite The set of network parameters of the Actor network obtained in time slots, Update the matrix for the preset aggregation, is the federation aggregation factor. It represents the weight of the elements on the diagonal, which usually represents the degree to which each MEO satellite retains its own network parameters. are off-diagonal elements, Indicates other The impact of a MEO satellite on the network parameters of the MEO satellites that share the network parameters.

[0041] S106, according to MEO satellite No. Resource task status and MEO satellites in each time slot The latest The resource scheduling information obtained in the time slot generates a network, and then determines the The first round of training The latest resource scheduling information of time slots generates the network until the Round 1 of training Time slot, complete the Round of training, and continue with the Rounds of training are performed until the iteration termination condition is reached, and the MEO satellite is obtained. The trained resource scheduling information generation network.

[0042] For example, when the number of training rounds reaches a preset round number threshold, it can be considered that the iteration termination condition has been reached, so that the Actor network obtained in the last training in the last round can be used as the MEO satellite The trained Actor network.

[0043] In the present invention, Is present in the system When there are 1 LEO satellites, the number of MEO satellites deployed to minimize the overall scheduling complexity of the system; the expression of the overall scheduling complexity of the system is as follows: ; ; in, is the overall scheduling complexity of the system, represents the scheduling complexity when each MEO satellite manages multiple LEO satellites, The dimension of the input data for the trained resource scheduling information generation network, The number of neurons in the hidden layer of the trained resource scheduling information generation network, The dimension of the output data of the trained resource scheduling information generation network, for example, is 6. In the expression of The other four items constitute the scheduling complexity of collaborative computing between multiple LEO satellites managed by each MEO satellite. "and" ” are all multiplication symbols. For example, the following Table 1 shows the optimal deployment of MEOs determined by using the above expression of the overall scheduling complexity of the system under different LEO numbers, where: Indicates the optimal number of MEO deployments.

[0044] Table 1 Optimal number of MEO deployments corresponding to different numbers of LEOs

[0045] Existing methods ignore the relationship between the scheduling complexity of giant constellation operation and maintenance and resource management architecture and methods in actual operation and the scale of the number of satellites in the network. The present invention proposes a calculation model for the overall scheduling complexity of the system based on the proposed giant constellation operation and maintenance and resource management system for computing-network integration, so that the giant constellation operation and maintenance and resource management system for computing-network integration proposed in the present invention can obtain the optimal number of MEO satellite deployments with low scheduling complexity under different numbers of LEO satellite deployments, so that the present invention can provide guidance for the construction and deployment of actual satellite constellations.

[0046] The present invention also provides a giant constellation operation and maintenance and resource management method for computing-network integration, which is applied to each MEO satellite in the above system, such as Figure 3 As shown, the method includes: S201. Obtain the resource task status of the current time slot in the current scheduling cycle.

[0047] S202. Use the trained resource scheduling information generation network to generate resource scheduling information for the current time slot according to its own resource task status in the current time slot; the trained resource scheduling information generation network is trained using local reinforcement learning and federated learning methods; the resource scheduling information of the current time slot represents the ratio of the amount of data calculated by the MEO satellite to the total amount of data of the LEO satellite's computing tasks in the computing tasks of each LEO satellite managed in the current time slot.

[0048] S203: Send resource scheduling information for the current time slot to each LEO satellite managed in the current time slot.

[0049] S204: Determine, based on the resource scheduling information for the current time slot, whether to perform data calculation for the computing task of each managed LEO satellite in the current time slot, and the amount of data that needs to be calculated when performing the data calculation.

[0050] S205. When data calculation of the computing task of each managed LEO satellite needs to be performed in the current time slot, the corresponding data amount of the computing task of each managed LEO satellite is calculated in the current time slot according to the data amount that needs to be calculated.

[0051] It should be noted that the specific principles of each step in the giant constellation operation and maintenance and resource management method for computing-network integration have been described in detail in the above introduction to the giant constellation operation and maintenance and resource management system for computing-network integration, and will not be repeated here.

[0052] Compared with the prior art, the present invention has the following advantages: (1) Aiming at the complex multi-layer structure and computing control of giant constellations, the present invention proposes a giant constellation operation and maintenance and resource control system and method for computing-network integration. The system includes two levels of control units, namely the MEO satellite coordination level and the LEO satellite edge level. By introducing computing and communication coordination mechanisms, collaborative scheduling and control mechanisms for inter-layer satellites, and strategic parameter coordination mechanisms between MEO satellites, dynamic scheduling and efficient utilization of resources are achieved. At the MEO coordination level, it is mainly responsible for the global operation and maintenance and resource control of the entire satellite network, including the allocation of computing tasks and the comprehensive utilization of inter-layer resources; at the LEO edge level, it focuses on the collaborative execution of real-time computing and communication tasks of edge nodes. This feature enables the present invention to perform efficient computing scheduling in practical applications, without relying on the control of ground controllers, and has higher real-time performance and better resource control and scheduling capabilities.

[0053] (2) The present invention is aimed at the actual giant constellation multi-layer satellite network scenario, taking into account the dynamic and uncertain computing tasks, the dynamically fluctuating satellite energy acquisition and consumption process, and resource limitations in practice. In order to ensure that high-energy consumption computing tasks can be completed efficiently in the actual satellite network scenario, the present invention focuses on the dynamic fluctuations in energy acquisition caused by the satellite periodically switching between the earth's shadow area and the direct sunlight area, as well as the arrival of dynamically uncertain computing tasks, and constructs the computing and network coordination process of computing and communication resources in the multi-layer giant constellation and its corresponding dynamic energy consumption and acquisition process based on these actual fluctuation characteristics. This feature makes the present invention more efficient and more in line with the real application environment.

[0054] (3) Different from the existing planning technology, the present invention uses continuous learning to decide the proportion of task data on the LEO satellite that is assisted by MEO satellites in each time slot through collaborative calculation. At the same time, it considers the long-term optimal resource management decision under the dynamic task arrival to achieve the collaboration between LEO satellites and MEO satellites. The present invention performs computing network collaboration between MEO satellites through sharing, interaction and aggregation of strategy parameters between MEO satellites (rather than relay transmission of task data), which can reduce the occupancy of communication bandwidth and latency, and achieve computing network collaboration of the entire satellite network. The present invention achieves efficient computing network integration management and control of giant constellations through efficient LEO-MEO inter-layer computing network collaboration and strategic parameter aggregation collaboration between MEO satellites, further improving the task completion of the entire satellite network. This feature enables the present invention to have better dynamic decision-making capabilities for the entire network and a higher task completion rate.

[0055] (4) Existing methods ignore the scheduling complexity of giant constellation operation and maintenance and resource management architecture and methods in actual operation and the relationship between the number of satellites in the network. The present invention proposes a giant constellation operation and maintenance and resource management system and method for computing-network integration in response to the multi-layered complex structure and computing management problems of giant constellations. According to this system and method, the optimal number of MEO satellite deployments under low scheduling complexity can be obtained under different numbers of LEO satellite deployments. This feature enables the present invention to provide guidance for the construction and deployment of actual satellite constellations.

[0056] The following simulation experiments are used to verify some of the technical effects of the present invention.

[0057] Simulation scenario The simulation scenario includes 144 LEO satellites with an orbital altitude of 780km and 6 MEO satellites with an orbital altitude of 8000km. Each satellite is equipped with edge computing and communication capabilities. , 7×10 -26J / cycle and 7×10 -24 J / cycle, J is joule, cycle is rotation, the satellite hardware calculation coefficient indicates how much energy is consumed in one computer clock cycle. In this scenario, each LEO satellite can collect, transmit and process ecological monitoring data. The link capacity between satellites is 300 Mbps. The receiving and transmitting powers of LEO satellites are 8W and 10W respectively. The learning rate , the number of time slots (i.e. scheduling period) , discount factor , total number of training iterations , federation aggregation factor is 0.5, is 2, is 4, the clipping threshold .

[0058] Simulation content and results The present invention is compared with the independent control scheduling algorithm and the non-inter-layer collaborative algorithm. The increase in the task completion of the present invention and the independent control scheduling algorithm and the non-inter-layer collaborative algorithm is as follows Figure 4 The energy consumption results are shown in Figure 5 As shown, Figure 4 and Figure 5 The "proposed collaborative algorithm" in the above represents the present invention, and the task data calculation requirements The unit of is cycles / bit, the unit of task data volume is Gbit, and the unit of energy consumption is KJ (kilojoule). Figure 4 Shows the computational requirements of task data The improvement of the task data completion of the present invention and the independent control scheduling algorithm and the non-inter-layer collaborative algorithm. Figure 4 It can be clearly seen that the present invention can complete the most tasks under different burst task loads due to its ability to adapt to dynamic environments and collaborate through MEO intersatellite federations. Compared with other methods, it can significantly increase the amount of completed task data. In order to further explore and explain the source of the advantages of the present invention, the present invention further studies the energy consumption, such as Figure 5 As shown. Combined Figure 4 and Figure 5 It can be seen that in the task data calculation requirements When the task data computation requirement is small, different methods have similar task completion performance under limited resources, because the resources are sufficient at this time. When the number of tasks is large, the present invention can complete more task data with lower resource consumption by coordinating the available communication and computing resources on different satellites through intelligent federation. Therefore, the present invention completes more task data through intelligent coordination of communication and computing resources under realistic resource-constrained conditions.

[0059] It should be noted that the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0060] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification.

[0061] In the specification, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality of components. Certain measures are recorded in different embodiments, but this does not mean that these measures cannot be combined to produce good effects.

[0062] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.

Claims

1. A giant constellation operation and maintenance and resource management system for computing and network integration, characterized in that: include: have MEO satellite collaborative management and control architecture with The LEO satellite edge-level management and control architecture of multiple LEO satellites, MEO satellites and LEO satellites are connected by communication links, and each MEO satellite manages multiple LEO satellites in each time slot of each scheduling cycle; Each LEO satellite is used to obtain data of the computing task in a preset time period and in each time slot of each scheduling cycle in the preset time period. According to the time slot sent by the MEO satellite resource scheduling information to perform data calculation of the computing task; Each MEO satellite is used to acquire itself in the time slot The resource task status is obtained by using the trained resource scheduling information to generate the network according to its own The resource task status is generated for the time slot The resource scheduling information is sent to the Each LEO satellite managed within the The resource scheduling information is used to perform collaborative data calculations of the computing tasks of each LEO satellite managed; the time slot The resource scheduling information indicates the time slot In the computing tasks of each LEO satellite managed by the MEO satellite, the ratio of the amount of data calculated by the MEO satellite to the total amount of data in the computing tasks of the LEO satellite; the value range of the resource scheduling information is 0 to 1.

2. The giant constellation operation and maintenance and resource management system for computing-network integration according to claim 1 is characterized in that: Each LEO satellite is specifically configured to receive the time slot according to the time slot transmitted by the MEO satellite. The resource scheduling information determines the time slot Whether to perform data calculation of the computing task within the time slot, and the amount of data required to be calculated when performing data calculation, and the amount of data required to be calculated in the time slot Calculate the corresponding data volume of the computing task; Each MEO satellite is specifically configured to provide a time slot for The resource scheduling information determines the time slot Whether to perform data calculation for the computing task of each LEO satellite managed within the time slot, and the amount of data required to be calculated when performing data calculation, and the amount of data required to be calculated according to the time slot The corresponding data volume of the computing task of each LEO satellite managed is calculated.

3. The giant constellation operation and maintenance and resource management system for computing-network integration according to claim 1 is characterized in that: Each MEO satellite in the time slot The resource task status includes: the MEO satellite in the time slot energy state, link state and on-board mission data state of the MEO satellite in the time slot Energy status characterization of the MEO satellite in the time slot The remaining power within the time slot; the MEO satellite Link state characterization: In the time slot The sum of the link capacities of the communication links between the MEO satellite and the managed LEO satellite in the time slot; the MEO satellite in the time slot On-board mission data status representation: in the time slot The sum of the data volumes of the computing tasks of the LEO satellites managed by the MEO satellites.

4. The giant constellation operation and maintenance and resource management system for computing-network integration according to claim 1 is characterized in that: Is there a system When there are 1 LEO satellites, the number of MEO satellites deployed to minimize the overall scheduling complexity of the system; the expression of the overall scheduling complexity of the system is as follows: ； ； in, is the overall scheduling complexity of the system, represents the scheduling complexity when each MEO satellite manages multiple LEO satellites, the dimension of the input data of the trained resource scheduling information generating network, The number of neurons in the hidden layer of the trained resource scheduling information generation network, The dimension of the output data of the trained resource scheduling information generating network.

5. The giant constellation operation and maintenance and resource management and control system for computing-network integration according to claim 1 is characterized in that: The trained resource scheduling information generation network is trained by local reinforcement learning and federated learning methods with the optimization goal of minimizing the energy consumption of the system in a scheduling cycle, and with the energy constraints of each LEO satellite, the energy constraints of each MEO satellite, the computing constraints of each LEO satellite, the computing constraints of each MEO satellite, and the communication constraints of each LEO satellite as constraints of the optimization goal.

6. The giant constellation operation and maintenance and resource management system for computing-network integration according to claim 5 is characterized in that: The expressions of the energy constraints of each LEO satellite and each MEO satellite are as follows: ； ； in, , , , For the system LEO satellites constitute a LEO satellite collection, is the number of satellites in the LEO satellite set LEO satellites, The value of , For a scheduling cycle In each time slot, For the system The MEO satellite cluster consists of MEO satellites. is the first in the MEO satellite set MEO satellites, The value of , For the said Maximum battery capacity for LEO satellites, is the percentage of the maximum discharge depth of the LEO satellite, For the said The amount of power available for each LEO satellite in each time slot, For the said The amount of electricity charged by a LEO satellite in each time slot, For the said The transmission energy consumption of a LEO satellite in each time slot is For the said The computational energy consumption of each LEO satellite in each time slot; For the said Maximum battery capacity for a MEO satellite, is the percentage of the maximum discharge depth of the MEO satellite, For the said The amount of power available for each MEO satellite in each time slot, For the said The amount of electricity charged by a MEO satellite in each time slot, For the said The receiving energy consumption of a MEO satellite in each time slot is: For the said The computational energy consumption of a MEO satellite in each time slot.

7. The giant constellation operation and maintenance and resource management and control system for computing-network integration according to claim 2 is characterized in that: Each MEO satellite includes: Resource task status monitoring module, used to obtain each time slot of each scheduling cycle The resource usage and computing task execution status of the MEO satellite itself and each LEO satellite it manages are obtained in the time slot. The resource task status; The data calculation processing module is used in the time slot When data calculation of the computing task of each LEO satellite managed is required, the computing task data of each LEO satellite managed is received, and the amount of data calculated is calculated according to the need in the time slot Calculate the corresponding data volume of the computing task of each LEO satellite managed; The computing network fusion scheduling optimization module is used to train the resource scheduling information generation network using local reinforcement learning and federated learning methods, obtain the trained resource scheduling information generation network, and combine it with the remaining The network parameters of the resource scheduling information generation network are shared and aggregated among the MEO satellites, and the trained resource scheduling information generation network is used to generate the network parameters according to the MEO satellite itself in the time slot. The resource task status is generated for the time slot resource scheduling information, and will target the time slot The resource scheduling information is sent to the Every LEO satellite managed within.

8. A method for giant constellation operation and maintenance and resource management for computing-network integration, characterized in that: Applied to each MEO satellite in the system of any one of claims 1 to 7 above, the method comprises: Get the resource task status of the current time slot in the current scheduling cycle; The resource scheduling information generation network is trained to generate resource scheduling information for the current time slot according to its own resource task status in the current time slot; the trained resource scheduling information generation network is trained by using local reinforcement learning and federated learning methods; the resource scheduling information of the current time slot represents the proportion of the amount of data calculated by the MEO satellite to the total amount of data of the LEO satellite's computing tasks in the computing tasks of each LEO satellite managed in the current time slot; Sending resource scheduling information for the current time slot to each LEO satellite managed in the current time slot; Determine, according to resource scheduling information for the current time slot, whether to perform data calculation of the computing task of each managed LEO satellite in the current time slot, and the amount of data that needs to be calculated when performing the data calculation; When data calculation of the computing task of each managed LEO satellite needs to be performed in the current time slot, the corresponding data amount of the computing task of each managed LEO satellite is calculated in the current time slot according to the data amount that needs to be calculated.

9. The method for giant constellation operation and maintenance and resource management for computing-network integration according to claim 8, characterized in that: The training method of the trained resource scheduling information generation network of each MEO satellite includes: In the The first round of training The MEO satellite is acquired at the time slot The resource task status of time slots; each round of training corresponds to a scheduling cycle, and each scheduling cycle has time slots, and training is performed every preset update step time slots; when The MEO satellite is at The resource task status of each time slot is the preset status; According to The resource scheduling information obtained in the time slot generates a network and the The resource task status of the time slot determines the The resource scheduling information of time slots; When the The resource scheduling information obtained in the time slot generates the network The resource scheduling information generation network obtained from the last training in the round of training; Based on the resource scheduling information of a time slot of the MEO satellite, and determining the The rewards for the time slot and the Resource task status of each time slot; based on , the preset update step size and the The resource scheduling information generation network obtained in the first time slot determines whether to update the network parameters of the current resource scheduling information generation network, and obtains the first The resource scheduling information obtained in each time slot generates a network; based on , preset aggregation step size and the first The resource scheduling information obtained in each time slot generates a network to determine whether to use the remaining MEO satellites aggregate network parameters and obtain the latest The resource scheduling information obtained in each time slot generates a network; According to the MEO satellite The resource task status of the time slot and the latest The resource scheduling information obtained in the time slot generates a network, and then determines the The first round of training The latest resource scheduling information of time slots generates a network until the first Round 1 of training time slot, complete the Round of training, and continue with the The training is repeated for a round until an iteration termination condition is reached, thereby obtaining the trained resource scheduling information generation network of the MEO satellite.

10. The method for giant constellation operation and maintenance and resource management for computing-network integration according to claim 9, characterized in that: The basis , preset aggregation step size and the first The resource scheduling information obtained in each time slot generates a network to determine whether to use the remaining MEO satellites aggregate network parameters and obtain the latest The resource scheduling information obtained in each time slot generates a network, including: right Performing a modulo operation with a preset aggregation step length to obtain a first operation result; When the first operation result is 0, obtain the remaining The first MEO satellite The resource scheduling information obtained in the time slots generates a network, and uses a preset aggregation update matrix to connect the MEO satellite with the remaining The first MEO satellite The resource scheduling information obtained in the time slots generates the network parameters of the network and aggregates them to obtain the first The network parameters after the time slot aggregation are used The network parameters after the aggregation of the time slots update the first The resource scheduling information obtained in the time slot generates a network, and obtains the latest The resource scheduling information obtained in each time slot generates a network; When the first operation result is not 0, the first The resource scheduling information obtained in the time slot generates a network as the latest The resource scheduling information obtained in each time slot generates a network.

Images