Task unloading method for zero-trust satellite network

Abstract

The invention discloses a task unloading method for a zero-trust satellite network, and relates to the technical field of communication networks. Comprising the following steps: sending model parameters to other low-earth-orbit satellites so that other low-earth-orbit satellites vote for the low-earth-orbit satellites according to empirical data of other low-earth-orbit satellites and the model parameters of the low-earth-orbit satellites; according to voting results of other low-orbit satellites on the low-orbit satellites, determining a pre-submission state of the low-orbit satellites; when the ratio of the number of the low earth orbit satellites in the pre-submission state in the communication network to the total number of the low earth orbit satellites is greater than a preset proportion threshold value, correcting the model parameters of the low earth orbit satellites in the pre-submission state, and performing global aggregation according to the corrected model parameters to obtain global model parameters; and taking the global model parameters as model parameters of the models of the low-earth-orbit satellites, and calculating task unloading strategies of the corresponding low-earth-orbit satellites through the models of which the parameters are updated. According to the method, the accuracy of the calculation task unloading result of the low-orbit satellite is improved.

Description

Technical Field

[0001] The present invention relates to the field of communication network technology, and in particular to a task offloading method for a zero-trust satellite network. Background Art

[0002] With the rapid development and widespread application of 6G technology, computing-intensive services, such as metaverse entertainment, intelligent transportation, and immersive communication, will be more widely present in people's daily lives, providing convenience and fun for people's lives. However, the above computing-intensive services usually need to frequently upload computing tasks to cloud servers, which brings huge computing load to cloud servers. At the same time, the transmission delay and transmission cost brought by this interactive mode are high, which makes it difficult to meet the quality of service (QoS) requirements of 6G communication technology. In urban areas with good infrastructure, tasks that are delay-sensitive, computing-intensive, and require high confidentiality are usually offloaded to ground edge servers, which can greatly reduce the delay and cost generated by offloading to cloud servers, improve QoS, and reduce the pressure on cloud servers. However, in some remote or harsh areas, the potential income cannot match the cost of deploying and maintaining ground infrastructure such as base stations (BS), so existing communication technologies cannot meet the 6G requirements for seamless wireless transmission and wide coverage.

[0003] In the existing technology, non-ground networks using low Earth Orbit (LEO) satellites can serve as a supplement to ground infrastructure, and computing tasks that should be transmitted to cloud servers or ground edge servers can be offloaded to LEO satellite networks for processing, thereby achieving seamless coverage and high throughput of communications and reducing transmission delays and costs.

[0004] However, there are still huge challenges in the trust issue between distributed LEO satellites: LEO satellites launched by different providers are usually selected to minimize costs and improve QoS. However, this approach will lead to potential trust issues between different LEO satellites, and there will be malicious behaviors such as calculation result falsification and data tampering, which will reduce the reliability of mission results and lead to inaccurate results of LEO satellite computing task offloading. Summary of the invention

[0005] Based on this, it is necessary to provide a task offloading method for zero-trust satellite networks to address the above-mentioned technical problems. This method improves the accuracy of computing task offloading results of low-orbit satellites.

[0006] The present invention adopts the following technical solutions:

[0007] The present invention provides a task offloading method for a zero-trust satellite network, comprising:

[0008] When the low-orbit satellite receives a computing task, a model parameter of a model used by the low-orbit satellite at the current moment to determine a computing task offloading strategy is obtained; the low-orbit satellite is any low-orbit satellite in the communication network; the communication network is a network formed by the low-orbit satellite and other low-orbit satellites accessible within the communication range of the low-orbit satellite;

[0009] Sending model parameters to other low-orbit satellites in the communication network so that other low-orbit satellites vote on the low-orbit satellite based on the experience data of other low-orbit satellites and the model parameters of the low-orbit satellite; the experience data is the data for interaction between other low-orbit satellites and the low-orbit satellites in the communication network;

[0010] Determine the pre-commitment status of the low-orbit satellite according to the voting results of other low-orbit satellites on the low-orbit satellite;

[0011] When the ratio between the number of low-orbit satellites in the pre-submission state and the total number of low-orbit satellites in the communication network is greater than a preset ratio threshold, the model parameters of the low-orbit satellites in the pre-submission state are corrected, and global aggregation is performed through a voting mechanism and a cold start mechanism according to the corrected model parameters to obtain global model parameters;

[0012] The global model parameters are used as the model parameters of each low-orbit satellite model, and the task offloading strategy of the corresponding low-orbit satellite is calculated through the model with updated parameters.

[0013] Preferably, obtaining the empirical data of other low-orbit satellites at the current moment includes:

[0014] According to the positions of other low-orbit satellites at the current moment, the state space of other low-orbit satellites at the current moment is obtained, and the reward values ​​of other low-orbit satellites at the current moment are calculated;

[0015] Analyze the state space based on the local models of other low-orbit satellites to determine the action space of other low-orbit satellites at the next moment;

[0016] According to the action space, the states of other low-orbit satellites are transferred to obtain the state spaces of other low-orbit satellites at the next moment;

[0017] The state space, action space, reward value of other low-orbit satellites at the current moment and the state space of other low-orbit satellites at the next moment are determined as experience data.

[0018] Preferably, other low-orbit satellites k The expression of the state space at time t is:

[0019]

[0020] Where D is the distance between every two low-orbit satellites The set of L(t,s k' ) are other low-orbit satellites k' The calculated load at time t, For other low-orbit satellites k' To low-orbit satellites k The transmission rate, For other low-orbit satellites k' The computing resources required, is the delay threshold.

[0021] Preferably, calculating the reward values ​​of other low-orbit satellites at the current moment includes:

[0022] Obtain the number of successful tasks and the number of failed tasks processed successfully by other low-orbit satellites within a preset time period;

[0023] Determine the active reward value and penalty value based on the number of successful tasks and the number of failed tasks;

[0024] The difference between the active reward value and the penalty value is determined as the reward value of other low-orbit satellites.

[0025] Preferably, voting for the low-orbit satellites based on the empirical data of other low-orbit satellites and the model parameters of the low-orbit satellites includes:

[0026] Calculate the loss rate between the models of other low-orbit satellites and the model of the low-orbit satellite based on the model parameters and empirical data of other low-orbit satellites and the model parameters of the low-orbit satellite;

[0027] If the loss rate is greater than the loss tolerance threshold, it is determined that the voting result of other low-orbit satellites on the low-orbit satellite is in favor;

[0028] If the loss rate is less than or equal to the loss tolerance threshold, it is determined that the voting result of other low-orbit satellites on the low-orbit satellite is against.

[0029] Preferably, the loss rate between the model of the other low-orbit satellite and the model of the low-orbit satellite is calculated based on the model parameters and empirical data of the other low-orbit satellites and the model parameters of the low-orbit satellite, including:

[0030] Calculate the arrival target time difference of the model of other low-orbit satellites based on the model parameters and empirical data of other low-orbit satellites, and calculate the arrival target time difference of the model of the low-orbit satellite based on the model parameters of the low-orbit satellite;

[0031] The ratio of the arrival target time difference of the models of other low-orbit satellites to the arrival target time difference of the model of the low-orbit satellite is determined as the loss rate between the models of the other low-orbit satellites and the model of the low-orbit satellite.

[0032] Preferably, the calculation formula for the time difference to the target is:

[0033]

[0034] Where γ is the discount factor, are model parameters, Low Earth Orbit Satellites k In the state space Next execution action space The model parameters are Q value when ; Q value represents the model parameter In the state space Execution Action Space The expected value of the reward obtained.

[0035] Preferably, determining the pre-submission status of the low-orbit satellite according to the voting results of other low-orbit satellites on the low-orbit satellite includes:

[0036] Obtain the number of other low-orbit satellites that vote in favor of the low-orbit satellite;

[0037] The ratio of the number of affirmative votes to the total number of other low-orbit satellites is determined as the verification ratio;

[0038] If the verification ratio is greater than the preset verification threshold, the state of the low-orbit satellite is determined to be a pre-submission state.

[0039] Preferably, the model parameters of the low-orbit satellite in the pre-submission state are modified through a voting mechanism and a cold start mechanism, including:

[0040] For any low-orbit satellite in a pre-submission state, obtain a first objection number of other low-orbit satellites whose voting results for the low-orbit satellite are against, and a second objection number of the low-orbit satellites whose voting results for other low-orbit satellites are against;

[0041] Determine the parameters of the constraint correction voting mechanism according to the first objection number and the second objection number;

[0042] According to the accumulated historical reputation and reputation threshold parameters of the low-orbit satellite, the cold start reputation aggregation parameters are determined; the accumulated historical reputation represents the number of consecutive rounds of global aggregation in which the low-orbit satellite successfully participates;

[0043] According to the constraint correction voting mechanism parameters and the cold start reputation aggregation parameters, the model parameters of the low-orbit satellite are corrected to obtain the corrected model parameters.

[0044] Preferably, global aggregation is performed according to the corrected model parameters to obtain global model parameters, including:

[0045] The weighted average of each modified model parameter is performed to obtain the global model parameter.

[0046] The present invention provides a task offloading device for a zero-trust satellite network, comprising:

[0047] an acquisition module, for acquiring model parameters of a model used by the low-orbit satellite to determine a computing task offloading strategy at a current moment when the low-orbit satellite receives a computing task; the low-orbit satellite is any low-orbit satellite in a communication network; the communication network is a network formed by a low-orbit satellite and other low-orbit satellites accessible within a communication range of the low-orbit satellite;

[0048] A voting module is used to send model parameters to other low-orbit satellites in the communication network, so that other low-orbit satellites can vote for the low-orbit satellite based on the experience data of other low-orbit satellites and the model parameters of the low-orbit satellite; the experience data is the data for interaction between other low-orbit satellites and the low-orbit satellites in the communication network;

[0049] A determination module, used to determine the pre-submission status of the low-orbit satellite according to the voting results of other low-orbit satellites on the low-orbit satellite;

[0050] An obtaining module is used for correcting the model parameters of each low-orbit satellite through a voting mechanism and a cold start mechanism when the ratio between the number of low-orbit satellites in the communication network in a pre-submission state and the total number of low-orbit satellites in the communication network is greater than a preset ratio threshold, and performing global aggregation according to the corrected model parameters to obtain global model parameters;

[0051] The calculation module is used to use the global model parameters as the model parameters of each low-orbit satellite model, and calculate the task offloading strategy for the corresponding low-orbit satellite through the model with updated parameters.

[0052] The present invention provides a computer-readable storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned task offloading method for a zero-trust satellite network is implemented.

[0053] The present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the above-mentioned task offloading method for a zero-trust satellite network is implemented.

[0054] At least one of the above technical solutions adopted by the present invention can achieve the following beneficial effects:

[0055] In the present invention, the model parameters of the low-orbit satellite are voted by other low-orbit satellites, and the model parameters are corrected through the voting mechanism and the cold start mechanism according to the voting results between the low-orbit satellites, and the model parameters of each low-orbit satellite are dynamically adjusted, which can promote model convergence and ensure the consistency of the global model parameters. Voting on the model parameters of the low-orbit satellite by other low-orbit satellites is equivalent to considering the consensus problem of each low-orbit satellite in the communication network, and the model parameters are corrected through the voting mechanism and the cold start mechanism according to the voting results between the low-orbit satellites, which can reduce the impact of malicious attack satellites on the global model. In this way, the calculation task offloading strategy of the low-orbit satellite is calculated with the global model parameters after global aggregation, thereby improving the accuracy of the calculation task offloading. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0057] Figure 1 A flowchart of a task offloading method for a zero-trust satellite network provided by the present invention;

[0058] Figure 2 A schematic diagram of a scenario applicable to a task offloading method for a zero-trust satellite network provided by the present invention;

[0059] Figure 3 A flowchart of a LEO satellite initiating synchronization request verification provided by the present invention;

[0060] Figure 4 A flowchart of a single round of training of a BCSA-FRL model provided by the present invention;

[0061] Figure 5 A schematic diagram of the mechanism of CCVM and CSRA provided by the present invention;

[0062] Figure 6 A comparison chart of reward values ​​using or not using the CCVM mechanism provided by the present invention;

[0063] Figure 7 A comparison chart of rewards for replay buffer poisoning under different federated reinforcement learning methods provided by the present invention;

[0064] Figure 8 A comparison chart of rewards under different federated reinforcement learning methods provided by the present invention that are poisoned by model parameters;

[0065] Fig. 9A performance comparison diagram of packet loss rate versus malicious satellite ratio under different federated reinforcement learning methods provided by the present invention;

[0066] Fig.10 A performance comparison diagram showing the average task processing delay as the proportion of malicious satellites changes under different federated reinforcement learning methods provided by the present invention;

[0067] Fig.11 A performance comparison diagram of packet loss rate changing with task load quantity under different task offloading decisions provided by the present invention;

[0068] Fig.12 A performance comparison diagram of average task processing delay as a function of task load quantity under different task offloading decisions provided by the present invention;

[0069] Fig.13 A schematic diagram of a task offloading device for a zero-trust satellite network provided by the present invention;

[0070] Fig.14 A schematic diagram of a computer device for implementing a task offloading method for a zero-trust satellite network provided by the present invention.

[0071] Description of reference numerals:

[0072] 101. Low-orbit satellite; 102. Ground users. DETAILED DESCRIPTION

[0073] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the specific embodiments of the present invention and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0074] For the task offloading problem in LEO satellite networks, existing methods mainly model the problem as a Markov decision problem, and use deep reinforcement learning and other methods to establish an artificial intelligence neural network, a highly robust model, to solve the problems of frequent changes in LEO satellite topology, different deployment altitudes, and diverse ground service QoS requirements. On the other hand, deep reinforcement learning requires unified training of data uploaded by users, ignoring the privacy of user data. In order to protect private data during the training process of artificial intelligence (AI) models, Federated Reinforcement Learning (FRL), which combines federated learning with deep reinforcement learning, protects the privacy of user data by transmitting only model parameters rather than specific data during the training process.

[0075] However, there are several urgent problems to be solved in task offloading in LEO satellite networks. First, the trust issue between distributed LEO satellites is still a huge challenge: service providers may choose LEO satellites launched by different companies to minimize costs and improve QoS, which leads to potential trust issues between different LEO satellites, and then leads to security risks; malicious voting attacks launched by malicious satellites during the verification process also seriously hinder the security consensus of LEO satellites. In this Zero Trust (Zero Trust, ZT) architecture, the traditional network security boundary assumes that there is inherent trust between devices within the preset network boundary. It is therefore crucial to propose a continuous and robust authentication, authorization, and attack detection scheme. Secondly, in this zero-trust architecture, federated reinforcement learning also has drawbacks: since the zero-trust architecture lacks a central reputation evaluation system, it is impossible to guide the federated learning to aggregate model parameters, and it is difficult for each LEO satellite to reach a secure consensus by evaluating reputation in a distributed manner; network attacks such as data poisoning and malicious models against FRL have further deteriorated the credibility between LEO satellites, and existing technologies are difficult to defend against the above attacks; for the zero-trust architecture, blockchain technology can use distributed ledgers to ensure the transparency and traceability of historical behaviors, and use smart contracts for voting consensus. However, in the context of federated learning, smart contracts can only bipolarly distinguish between malicious models and normal models, and the sub-model that has just recovered from an attack has a significantly smaller contribution than other continuously credible models, requiring extra time for regular contributions, and commonly used smart contracts cannot defend against malicious voting.

[0076] Based on this, the present invention provides a task offloading method for zero-trust satellite networks, which weights the local model according to the voting results of blockchain participants, promotes model convergence and reduces the impact of malicious attack satellites on the global model; secondly, the invention can dynamically adjust the model parameters of each low-orbit satellite in persistent and continuous malicious satellite attacks, update the global model parameters, and ensure the consistency of the global model; finally, the invention comprehensively considers the security consensus problem between heterogeneous low-orbit satellites, the impact of malicious attack satellites on the global model and the QoS requirements of services, and formulates a flexible task offloading strategy to solve the zero-trust problem of distributed low-orbit satellites and the problem that federated learning is vulnerable to malicious network attacks, and provide high-quality, high-stability and customized communication transmission services for users in remote or harsh areas.

[0077] The technical solutions provided by various embodiments of the present invention are described in detail below in conjunction with the accompanying drawings.

[0078] In an exemplary embodiment, Figure 1 As shown, Figure 1 The present invention is a flowchart of a task offloading method for a zero-trust satellite network, which specifically includes the following steps:

[0079] S101, when a low-orbit satellite receives a computing task, obtain model parameters of a model used by the low-orbit satellite at the current moment to determine a computing task offloading strategy; the low-orbit satellite is any low-orbit satellite in a communication network.

[0080] The communication network is a network formed by low-orbit satellites and other low-orbit satellites that can be accessed within the communication range of low-orbit satellites. Figure 2 As shown, the ground user 102 offloads the computing task to the low-orbit satellite 101, and the low-orbit satellite 101 makes intelligent offloading decisions on the computing task through the model.

[0081] Each low-orbit satellite corresponds to a model, which is used to calculate the task offloading strategy, and the model parameters are the parameters of the model.

[0082] S102, sending model parameters to other low-orbit satellites in the communication network, so that other low-orbit satellites vote for the low-orbit satellite based on the experience data of other low-orbit satellites and the model parameters of the low-orbit satellite; the experience data is the data for interaction between other low-orbit satellites and the low-orbit satellites in the communication network.

[0083] Among them, voting on low-orbit satellites is carried out through smart contracts on the blockchain.

[0084] In an exemplary embodiment, the experience data of other low-orbit satellites at the current moment is obtained, including: obtaining the state space of other low-orbit satellites at the current moment according to the positions of other low-orbit satellites at the current moment, and calculating the reward values ​​of other low-orbit satellites at the current moment; analyzing the state space according to the local models of other low-orbit satellites to determine the action space of other low-orbit satellites at the next moment; performing state transfer on other low-orbit satellites according to the action space to obtain the state space of other low-orbit satellites at the next moment; determining the state space, action space, reward value of other low-orbit satellites at the current moment and the state space of other low-orbit satellites at the next moment as experience data.

[0085] All low-orbit satellites can be modeled in the satellite orbit simulation software (Satellite Tool Kit, STK) to obtain the low-orbit satellite s at each time. k coordinate And build the network topology map of low-orbit satellites according to the coordinates of low-orbit satellites. After obtaining the network topology map, initialize the low-orbit satellite s k ∈{s1,…,s k ,…,s K} and global model parameters; initialize the position-related parameters of the low-orbit satellite; the system time node is recorded as t∈{1,...,t,...,T}, and the state space at time t is recorded as The action space is denoted as The reward value is recorded as The verification experience buffer is denoted as The verification buffer stores the empirical data; the task set is represented by C = {c1,...,c i ,...,c I}.

[0086] Other low-orbit satellites k' For example, other low-orbit satellites s k' The expression of the state space at time t is:

[0087]

[0088] Where D is the distance between every two low-orbit satellites The set of L(t,s k' ) are other low-orbit satellites k' The calculated load at time t, For other low-orbit satellites k' To low-orbit satellites k The transmission rate, For other low-orbit satellites k' The computing resources required, is the delay threshold.

[0089] Optionally, the state space The transmission rate The specific derivation process is as follows:

[0090] The communication link between LEO satellites can be approximated as propagation in free space, and its free space path loss coefficient PL F The calculation expression is as follows:

[0091]

[0092] Among them, G T is the LEO satellite transmitting antenna gain, G R is the LEO satellite receiving antenna gain, λ is the wavelength of the signal, Is launching satellites k' and receive satellites k The distance between them, L is the system loss coefficient.

[0093] In addition, in other practical environments, the average received signal power decreases logarithmically with distance, so the free space path loss coefficient PL LD The calculation expression is as follows:

[0094]

[0095] Where d0 is the reference distance and n is the environmental correlation coefficient.

[0096] At the same time, the path loss also depends on the surrounding environment between the transmitting satellite and the receiving satellite. Therefore, in order to simulate the path loss in the actual communication environment, a Gaussian random variable X with a mean of 0 and a standard deviation of σ is used. σ Further description of the free space path loss coefficient PL NOR , and its calculation expression is as follows:

[0097]

[0098] Therefore, the transmission rate between satellite links The calculation expression is as follows:

[0099]

[0100] in, is the channel bandwidth, LEO satellites k' The signal transmission power, is the spatial noise power.

[0101] Optionally, in state space, the actual delay of the task It consists of three parts: transmission delay, waiting delay and calculation delay. Its calculation expression is as follows:

[0102]

[0103] in, is the computation task c i The size of α is a binary factor with a value range of [0,1], indicating local processing or forwarding to other satellites for processing. is the queue delay time that the task waits in the queue, is the computation task c i The computing resources required, Other low-orbit satellites k' computing power.

[0104] There are other computing tasks in other low-orbit satellites c i' When the queue delay The calculation expression is as follows:

[0105]

[0106] Among them, in the calculation task c i' Timeout case β i' The value is 0. If there is no timeout, the value is 1.

[0107] Optionally, the reward values ​​of other low-orbit satellites at the current moment are calculated, including: obtaining the number of successful tasks successfully processed and the number of failed tasks failed to be processed by other low-orbit satellites within a preset time; determining the active reward value and the penalty value according to the number of successful tasks and the number of failed tasks; and determining the difference between the active reward value and the penalty value as the reward value of other low-orbit satellites.

[0108] The preset duration represents the interval duration ΔT, based on other low-orbit satellites s k' Take time t as an example to illustrate, other low-orbit satellites s k' The reward function at time t is It is the difference between the tasks that are successfully processed and the tasks that fail to be processed within the time interval ΔT. Its calculation expression is as follows:

[0109]

[0110] in, represents the number of computing tasks successfully processed within ΔT, represents the number of computing tasks that failed to be processed within ΔT, r t,+ represents the active reward value for successful processing within ΔT, r t,- It represents the penalty value for processing failure within ΔT.

[0111] The state space is analyzed according to the models of other low-orbit satellites to determine the action space of other low-orbit satellites at the next moment, including: inputting the state space of other low-orbit satellites into the models corresponding to other low-orbit satellites to obtain the action space of other low-orbit satellites at the next moment output by the model; wherein the model can be a federated learning model, with other low-orbit satellites s k' The action space at time t For example, the action space The expression is:

[0112]

[0113] The action space Other low-orbit satellites k The set of offloading destinations that can be reached within two hops; that is, the low-orbit satellites to which the computing task can be offloaded within two hops. For example, if Other low-orbit satellites k' The computing task can be offloaded to any low-orbit satellite among s1, s5, and s6. k' It is reachable to s1, s5 and s6 in two hops.

[0114] The state of the low-orbit satellite is transferred according to the action space to obtain the state space of other low-orbit satellites at the next moment, including: selecting the computing task unloading destination (other low-orbit satellite) of other low-orbit satellites from the action space, then unloading the computing task to the corresponding destination, and then observing the state of the low-orbit satellite to obtain the state space of other low-orbit satellites at the next moment; wherein, selecting the computing task unloading destination of other low-orbit satellites from the action space includes: randomly selecting a low-orbit satellite from the action space as the computing task unloading destination.

[0115] Finally, the state space, action space, reward value of other low-orbit satellites at the current moment and the state space of other low-orbit satellites at the next moment are determined as empirical data

[0116] When a low-orbit satellite receives a computing task, the low-orbit satellite may send the model parameters of the low-orbit satellite to other low-orbit satellites, wherein the other low-orbit satellites here may be all low-orbit satellites in the communication network except the low-orbit satellite.

[0117] After receiving the model parameters sent by the low-orbit satellite, other low-orbit satellites may vote for the low-orbit satellite according to the model parameters. Optionally, other low-orbit satellites vote for the low-orbit satellite according to the experience data of other low-orbit satellites and the model parameters of the low-orbit satellite, including: calculating the loss rate between the models of other low-orbit satellites and the model of the low-orbit satellite according to the model parameters and experience data of other low-orbit satellites and the model parameters of the low-orbit satellite; if the loss rate is greater than the loss tolerance threshold, determining that the voting result of other low-orbit satellites for the low-orbit satellite is in favor; if the loss rate is less than or equal to the loss tolerance threshold, determining that the voting result of other low-orbit satellites for the low-orbit satellite is against.

[0118] Among them, the loss rate between the models of other low-orbit satellites and the model of the low-orbit satellite is calculated according to the model parameters and empirical data of other low-orbit satellites, as well as the model parameters of the low-orbit satellite, including: calculating the target arrival time difference of the models of other low-orbit satellites according to the empirical data of other low-orbit satellites, and calculating the target arrival time difference of the model of the low-orbit satellite according to the model parameters of the low-orbit satellite; the ratio of the target arrival time difference of the models of other low-orbit satellites to the target arrival time difference of the model of the low-orbit satellite is determined as the loss rate between the models of other low-orbit satellites and the model of the low-orbit satellite.

[0119] Optionally, the calculation formula for the time difference to the target is:

[0120]

[0121] Where γ is the discount factor, are model parameters, Low Earth Orbit Satellites k In the state space Next execution action space The model parameters are Q value when ; Q value represents the model parameter In the state space Execution Action Space The expected value of the reward obtained.

[0122] The action space May include a variety of uninstall results, It represents the Q value corresponding to the maximum Q value of a certain offloading result in the execution action space.

[0123] It should be noted that after receiving the model parameters sent by the low-orbit satellite, other low-orbit satellites can calculate the action space and reward value in the state space through the model parameters of the low-orbit satellite, and determine the arrival target time difference of the low-orbit satellite model based on the action space and reward value calculated by the model parameters of the low-orbit satellite.

[0124] The low-orbit satellite s can be calculated according to the above formulas (12)-(14) k The arrival time difference of the model is calculated by the same principle for other low-orbit satellites s k' The arrival time difference of the model.

[0125] In addition, other low-orbit satellites Model and low-orbit satellite The loss rate between the models The calculation expression is as follows:

[0126]

[0127] S103, determining the pre-submission status of the low-orbit satellite according to the voting results of other low-orbit satellites on the low-orbit satellite.

[0128] Preferably, the pre-submission status of the low-orbit satellite is determined based on the voting results of other low-orbit satellites on the low-orbit satellite, including: obtaining the number of votes in favor of the low-orbit satellite from other low-orbit satellites that voted in favor of the low-orbit satellite; determining the ratio of the number of votes in favor to the total number of other low-orbit satellites as the verification ratio; if the verification ratio is greater than a preset verification threshold, determining that the status of the low-orbit satellite is in the pre-submission status.

[0129] Low Earth Orbit Satellites k Verification ratio The calculation formula can be:

[0130]

[0131] Among them, Number(V T ) is the number of votes received that are in favor, Number(V F ) is the number of votes received with a negative result, and the sum of the number of votes in favor and against is the total number of other low-orbit satellites.

[0132] If the verification ratio is greater than the preset verification threshold, the state of the low-orbit satellite is determined to be set to the pre-submission state, otherwise, it does not participate in the global aggregation process. The preset verification threshold may be 50%.

[0133] S104, when the ratio between the number of low-orbit satellites in the pre-submission state in the communication network and the total number of low-orbit satellites in the communication network is greater than a preset ratio threshold, the model parameters of the low-orbit satellites in the pre-submission state are corrected through a voting mechanism and a cold start mechanism, and global aggregation is performed based on the corrected model parameters to obtain global model parameters.

[0134] Obtain the voting results of the LEO satellites stored in the blockchain, calculate the ratio of the LEO satellites in the current pre-submission state to all LEO satellites, and determine whether to enter the global parameter aggregation stage. According to the voting results of the LEO satellites for the remaining satellites, use the CCVM mechanism to correct the model parameters of the local model. Establish a CSRA scheme, consider the historical accumulated reputation of LEO satellites, and correct the model parameters of their local models.

[0135] When the number of low-orbit satellites in the pre-submission state in the communication network is less than or equal to the preset ratio threshold, the model parameters of the local models of all low-orbit satellites are traced back to the global model parameters of the previous round of training. When the number of low-orbit satellites in the pre-submission state in the communication network is greater than the preset ratio threshold, the model parameters of the low-orbit satellites in the pre-submission state are corrected, and global aggregation is performed based on the corrected model parameters to obtain the global model parameters.

[0136] Optionally, the model parameters of the low-orbit satellite in the pre-submission state are corrected through a voting mechanism and a cold start mechanism, including the following steps: for any low-orbit satellite in the pre-submission state, obtain a first number of objections to other low-orbit satellites that have voted against the low-orbit satellite, and a second number of objections to low-orbit satellites that have voted against other low-orbit satellites; determine constraint correction voting mechanism parameters according to the first number of objections and the second number of objections; determine cold start reputation aggregation parameters according to the accumulated historical reputation and reputation threshold parameters of the low-orbit satellite; the accumulated historical reputation represents the number of consecutive rounds of global aggregation in which the low-orbit satellite has successfully participated; and correct the model parameters of the low-orbit satellite according to the constraint correction voting mechanism parameters and the cold start reputation aggregation parameters to obtain corrected model parameters.

[0137] The local model parameters can be corrected according to the Constrained Correction Voting Mechanism (CCVM) and the Cold Start Reputation Aggregation (CSRA) scheme, and the federated learning model can be globally aggregated to adaptively make intelligent offloading decisions for computing tasks in the low-orbit satellite network.

[0138] Specifically, malicious voting attack satellites can continuously vote against other satellites, hindering model convergence and interfering with consensus. In order to limit the impact of malicious voting attack satellites on the global model, by calculating any low-orbit satellite s k The voting results of other low-orbit satellites are used to obtain the constraint correction voting mechanism parameters of the low-orbit satellite. The calculation expression is as follows:

[0139]

[0140] Where K is the total number of low-orbit satellites, is the first objection number of other low-orbit satellites whose voting results are against the low-orbit satellite, is the second number of objections to other LEO satellites whose voting results are against, Indicates that except for low-orbit satellites k For other low-orbit satellites other than , sigmoid is a function that maps values ​​to [0,1], and its calculation expression is:

[0141]

[0142] In the constraint correction voting mechanism, the penalty for satellites that cannot make pre-submissions is to prevent them from participating in the current round of global aggregation. However, the low-orbit satellite may be restored in the future. In order to avoid long-term impact in the future, the CSRA mechanism is established to use low-orbit satellites to k Historical voting records, to obtain the cold start reputation aggregation parameters of the low-orbit satellite The calculation expression is as follows:

[0143]

[0144] in, Low Earth Orbit Satellites k The cumulative historical reputation of the satellite represents the number of consecutive rounds of global aggregation that the satellite successfully participated in, H Rep is the reputation threshold parameter.

[0145] Get CCVM correction parameters and CSRA correction parameters After that, for low-orbit satellites k Model parameters of the model The calculation expression is as follows:

[0146]

[0147] in, Indicates low-orbit satellites k The model parameters of the model are represents the modified model parameters.

[0148] Performing global aggregation based on the modified model parameters to obtain global model parameters includes: performing weighted average on each modified model parameter to obtain the global model parameter. Among them, the global model parameter θ K The calculation expression is as follows:

[0149]

[0150] Where M is the number of low-orbit satellites in the pre-submission state.

[0151] After obtaining the global model parameters, the global model parameters are recorded in the blockchain.

[0152] S105, using the global model parameters as model parameters of each low-orbit satellite model, and calculating the task offloading strategy for the corresponding low-orbit satellite through the model with updated parameters.

[0153] In an exemplary embodiment, the present invention further provides a task offloading method for a zero-trust satellite network, the method comprising the following steps:

[0154] S301, using satellite orbit simulation software to simulate the spatial position of the LEO satellite, and constructing a network topology map of the LEO satellite according to the coordinate parameters obtained by the simulation. Using the blockchain-based smart contract mechanism proposed by the present invention, the synchronization consensus step of the LEO satellite is completed.

[0155] After obtaining the LEO satellite network topology, the LEO satellite s k ∈{s1,…,s k ,…,s K} and global model parameters; initialize the position-related parameters of the LEO satellite; the system time node is recorded as t∈{1,...,t,...,T}, and the state space at time t is recorded as The action space is denoted as The reward function is recorded as The verification experience buffer is denoted as The task set is represented by C = {c1,...,c i ,...,c I Before each round of algorithm operation, the voting result graph V is initialized, which marks whether the LEO satellite has voted against or in favor of any LEO satellite. If no vote has been cast for a LEO satellite, the voting result is recorded as empty. For each LEO satellite, a verification request is initiated to all remaining LEO satellites, and then the verification experience buffer is used to verify the LEO satellite. Get verification data Calculate the time difference (TD) of the verification data in the local model of the current LEO satellite and use it as the evaluation index of the verification; the current satellite uses the verification data Calculate the arrival time difference TD of the local model of the LEO satellite to be voted; calculate the loss rate of the two local models, and vote in favor or against according to the loss tolerance threshold E; record the voting results of each LEO satellite and upload them to the blockchain for storage.

[0156] Specifically, the operation steps of the LEO satellite network topology simulation and blockchain-based synchronization consensus in S301 are:

[0157] S1, initialize the global time t = 0, model all LEO satellites in the satellite orbit simulation software STK, and obtain the LEO satellite s at each time k coordinate And calculate the distance between each LEO satellite based on the coordinates

[0158] S2, initialization voting result graph Initialize LEO satellite state space Action Space Reward Function and verify experience buffer

[0159] S3, select a LEO satellite that has not voted yet k , and vote for all remaining satellites in turn, and record the voting results in the set V of the blockchain, such as Figure 3 As shown, Figure 3 Flowchart for initiating synchronization request verification for a LEO satellite.

[0160] Specifically, LEO satellites k Select a satellite that has not yet been verified k' , from the verification experience buffer Get a batch of verification data in In the state Select Action to enter the state; LEO satellites k Calculate the local model's time difference of arrival TD, LEO satellites k' The arrival time difference TD of the local model is calculated in the same way to obtain the loss rate like If the threshold E is exceeded, the LEO satellite s k For LEO satellites k' Vote for T , otherwise vote against V F ; Determine the current LEO satellites k Have all other satellites been voted for? If yes, go to S4; if not, continue to execute S3.

[0161] S4, judging whether the current voting result map records the voting results of all LEO satellites, if so, proceeding to S5, if not, proceeding to S3.

[0162] S5, upload the voting result graph V to the blockchain for storage. Each LEO satellite obtains the voting result record graph V. If other LEO satellites k The verification success rate exceeds 50%, and LEO satellites k The status is set to pre-submission status, otherwise it does not participate in the global parameter aggregation process.

[0163] S301, using CCVM and CSRA to correct local model parameters, globally aggregate the federated reinforcement learning model, and adaptively make intelligent offloading decisions for computing tasks in the LEO satellite network through global model parameters.

[0164] Specifically, the satellite network voting results stored in the blockchain are obtained, and the ratio of the current pre-submitted LEO satellites to all LEO satellites is calculated to determine whether to enter the global parameter aggregation stage. k Based on the voting results of the remaining satellites, the CCVM mechanism is used to correct the model parameters of the local model. k The task offloading method for zero-trust satellite network of the present invention includes a cold start model aggregation task offloading scheme based on federated reinforcement learning (Blockchain-enabled ColdStart Aggregation Federated Reinforcement Learning, BCSA-FRL), such as Figure 4 As shown, Figure 4 This is a flowchart of a single round of training of the BCSA-FRL model. The BCSA-FRL framework is used to perform global aggregation of model parameters and record the global model parameters in the blockchain. Specifically, the following steps are included:

[0165] S401, determine whether all LEO satellites have completed verification. If all LEO satellites have completed verification, execute step S402, otherwise execute step S405.

[0166] S402, determine whether the proportion of LEO satellites in the pre-submission state is greater than 50%. If so, correct the model parameters of the local model of each LEO satellite in the pre-submission state, perform global parameter distance, and calculate the global model parameters. Otherwise, trace back the parameters of all local models to the global model parameters of the previous round of training.

[0167] S403, the LEO satellite initiates a model synchronization request and broadcasts the local model.

[0168] S404, the remaining LEO satellites perform local verification and broadcast their verification results.

[0169] S405, determining whether the LEO satellite has received more than half of the successful verification results, if so, the LEO satellite enters a pre-submission state, otherwise, the LEO satellite is marked as an abnormal node.

[0170] Specifically, the specific steps of determining whether to perform global parameter aggregation in S302 are:

[0171] S6, calculate the ratio of the LEO satellites in the current pre-submission state to all LEO satellites. If the ratio exceeds 50%, go to S7, otherwise, trace back the parameters of all local models to the global model parameters of the previous round of training and go to S3.

[0172] Specifically, the specific process of global parameter aggregation in S302 is as follows:

[0173] S7, calculate LEO satellites k Constraints modify voting mechanism parameters and cold start reputation aggregation parameters and correct LEO satellites k Model parameters like Figure 5 As shown, Figure 5 Schematic diagram of the mechanism of CCVM and CSRA.

[0174] S8, based on the corrected model parameters of all LEO satellites, global aggregation is performed to obtain the global model parameters And submit the global model parameters to the blockchain.

[0175] S9, determine whether the training is finished, if it is finished, end the process, if not, go to S3.

[0176] In an exemplary embodiment, without loss of generality, the simulation system of the task offloading method for zero-trust satellite network is assigned in this embodiment. This embodiment considers that there are multiple areas on the ground, each area contains multiple terminals as task creators. After the ground terminal creates a task, it will send a task calculation request to any LEO satellite according to its position. At any time node t, the number of calculation tasks is [200, 400], the step size is 100, and the probability obeys the Poisson distribution with a parameter of 100. The size of the calculation task file obeys the standard normal distribution with a mean of 15MB. In the space layer, there is a total of 1 LEO satellite orbit, each orbit has 15 LEO satellites, its altitude is 800km, its computing power is [1, 4]GB, the channel bandwidth is 2MHz, the transmission power is 150W, the environmental correlation coefficient is 2, the transmitting antenna gain and the receiving antenna gain are both 1, the system loss coefficient is 1, and the spatial noise power is 1w. The learning rate in the neural network is 0.001, the exploration decay rate is 0.95, the batch size of the experience replay pool is 32, and the batch size of the verification experience pool is 512.

[0177] The present invention provides a task offloading method for zero-trust satellite networks. First, a consensus is reached in distributed LEO satellites through smart contracts of blockchain, and the network is transitioned from one consistency state to another consistency state under the premise of ensuring that all files are globally accessible in the LEO satellite network. Secondly, after reaching a consensus between satellites, the present invention proposes CCVM and CSRA schemes, which not only take into account the existence of malicious attack satellites in the satellite network, and correct the influence of the local model of the satellite on the global model through the success or failure of the verification of a certain satellite by other satellites, but also take into account the process of the malicious attack satellite gradually changing from the attack state to the normal state, and dynamically adjust the influence of any satellite on the global model by evaluating its historical reputation, thereby improving the robustness of the global model and accelerating the speed of model convergence. Finally, the present invention not only takes into account the collaborative work between heterogeneous LEO satellites, but also flexibly offloads tasks to LEO satellites with different computing powers according to the needs of ground users, thereby reducing the network packet loss rate and network delay, and providing customized QoS for ground users. Figure 6 As shown, Figure 6 is a comparison chart of rewards with or without CCVM mechanism, Figure 6 It can be seen that the reward value of the BCSA-FRL algorithm of the present invention can converge to the optimal value of 25 with the CCVM mechanism, and its defense performance is significantly better than the reward value of 10 without the CCVM mechanism. Figure 7 and Figure 8 As shown, Figure 7 Comparison of rewards for different federated reinforcement learning methods subject to replay buffer poisoning. Figure 8Comparison of rewards under different federated reinforcement learning methods with model parameter poisoning. Figure 7 and Figure 8 It can be concluded that the task offloading method for zero-trust satellite networks of the present invention can increase the reward value by 100% to the optimal value of 26 in the attack of replay buffer poisoning compared with the conventional federated reinforcement learning algorithm (FedAvg-FRL) algorithm; in the attack of model poisoning, the reward value can be increased by 50.29% to the optimal value of 26, and converge faster; Fig. 9 and Fig.10 As shown, Fig. 9 This is a performance comparison chart of packet loss rate changing with the proportion of malicious satellites under different federated reinforcement learning methods. Fig.10 This is a performance comparison chart of average task processing delay under different federated reinforcement learning methods as the proportion of malicious satellites changes. Fig. 9 and Fig.10 It can be concluded that when the proportion of malicious satellites is less than 50%, the packet loss rate can be reduced by 8.75% to an average level of 5%, and the average processing delay can be reduced by 2.8ms to an average level of 6ms. Fig.11 and Fig.12 As shown, Fig.11 This is a performance comparison chart of packet loss rate changing with the number of task loads under different task offloading decisions. Fig.12 The performance comparison chart of average task processing delay as the number of task loads changes under different task offloading decisions is shown in Figure 1. Fig.11 and Fig.12 It can be concluded that the task offloading method for zero-trust satellite network of the present invention can reduce the packet loss rate by an average of 13.89% and 34.18% when the task load is low, reaching an average level below 6.16%, and the average processing delay can be reduced by 1.45ms and 3.35ms, reaching an average level below 5.95ms; when the task load is high, the packet loss rate and the average task processing delay do not increase significantly, and the packet loss rate can be reduced by an average of 13.29% and 38.81%, reaching an average level below 8.29%, and the average processing delay can be reduced by 2.27ms and 3.88ms, reaching an average level below 6.08ms.

[0178] The present invention provides a task offloading method for zero-trust satellite networks. First, voting consensus is performed through the smart contract of blockchain to ensure the transparency and traceability of historical behaviors. Secondly, the present invention further proposes a CCVM scheme, which can correct the influence of malicious voting satellites, evaluate the credibility of uploaded sub-models, and ensure the trust between distributed satellites. In addition, the present invention proposes a CSRA scheme, which can seriously reduce the reputation of attacking participants when an attack is detected, and gradually restore the malicious attacking satellite to a normal state in consideration of the clearing time of the memory replay buffer, thereby ensuring the normal convergence of the model. Finally, the BCSA-FRL scheme of the present invention takes into account the synergy between distributed satellites and flexibly offloads tasks to different LEO satellites according to the needs of ground users, thereby ensuring the efficiency and security of the task offloading strategy.

[0179] When applying the task offloading method for zero-trust satellite network provided by the present invention, it is not necessary to Figure 1 The steps are executed in the order shown. The specific execution order of the steps can be determined according to needs, and the present invention does not limit this.

[0180] The above is a task offloading method for a zero-trust satellite network provided by one or more embodiments of the present invention. Based on the same idea, the present invention also provides a corresponding task offloading device for a zero-trust satellite network, such as Fig.13 shown.

[0181] Fig.13 A schematic diagram of a task offloading device for a zero-trust satellite network provided by the present invention, the device 1300 includes:

[0182] The acquisition module 1301 is used to acquire the model parameters of the model used by the low-orbit satellite to determine the computing task offloading strategy at the current moment when the low-orbit satellite receives the computing task; the low-orbit satellite is any low-orbit satellite in the communication network; the communication network is a network formed by the low-orbit satellite and other low-orbit satellites that can be accessed within the communication range of the low-orbit satellite;

[0183] The voting module 1302 is used to send model parameters to other low-orbit satellites in the communication network, so that other low-orbit satellites vote for the low-orbit satellite according to the experience data of other low-orbit satellites and the model parameters of the low-orbit satellite; the experience data is the data exchanged between other low-orbit satellites and the low-orbit satellites in the communication network;

[0184] A determination module 1303 is used to determine the pre-submission status of the low-orbit satellite according to the voting results of other low-orbit satellites on the low-orbit satellite;

[0185] The obtaining module 1304 is used for correcting the model parameters of each low-orbit satellite through a voting mechanism and a cold start mechanism when the ratio between the number of low-orbit satellites in the communication network in a pre-submission state and the total number of low-orbit satellites in the communication network is greater than a preset ratio threshold, and performing global aggregation according to the corrected model parameters to obtain global model parameters;

[0186] The calculation module 1305 is used to use the global model parameters as the model parameters of each low-orbit satellite model, and calculate the task offloading strategy for the corresponding low-orbit satellite through the model with updated parameters.

[0187] For the specific definition of the task offloading device for zero-trust satellite networks, please refer to the definition of the task offloading method for zero-trust satellite networks mentioned above, which will not be repeated here. The various modules in the above-mentioned task offloading device for zero-trust satellite networks can be implemented in whole or in part through software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0188] The present invention also provides a computer-readable storage medium, which stores a computer program, which can be used to execute the above Figure 1 A task offloading method for zero-trust satellite networks is provided.

[0189] The present invention also provides Fig.14 The structural diagram of the computer device shown in FIG. Fig.14 As shown in the figure, at the hardware level, the computer device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and may also include other hardware required for the business. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to achieve the above Figure 1 A task offloading method for zero-trust satellite networks is provided.

[0190] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0191] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of the present invention.

Claims

1. A task offloading method for zero-trust satellite network, characterized in that: include: When a low-orbit satellite receives a computing task, a model parameter of a model used by the low-orbit satellite at the current moment to determine a computing task offloading strategy is obtained; the low-orbit satellite is any low-orbit satellite in a communication network; the communication network is a network formed by a low-orbit satellite and other low-orbit satellites accessible within a communication range of the low-orbit satellite; Sending the model parameters to other low-orbit satellites in the communication network so that the other low-orbit satellites vote for the low-orbit satellite based on the experience data of the other low-orbit satellites and the model parameters of the orbiting satellite; the experience data is the data for interaction between the other low-orbit satellites and the low-orbit satellites in the communication network; Determining a pre-submission status of the low-orbit satellite according to voting results of the other low-orbit satellites on the low-orbit satellite; When the ratio between the number of low-orbit satellites in the communication network in a pre-submission state and the total number of low-orbit satellites in the communication network is greater than a preset ratio threshold, the model parameters of the low-orbit satellites in the pre-submission state are corrected through a voting mechanism and a cold start mechanism, and global aggregation is performed according to the corrected model parameters to obtain global model parameters; The global model parameters are used as model parameters of the models of the low-orbit satellites respectively, and the task offloading strategy of the corresponding low-orbit satellite is calculated through the model with updated parameters.

2. The method according to claim 1, characterized in that Acquiring the empirical data of the other low-orbit satellites at the current moment, including: According to the positions of the other low-orbit satellites at the current moment, the state space of the other low-orbit satellites at the current moment is obtained, and the reward values ​​of the other low-orbit satellites at the current moment are calculated; Analyzing the state space according to the local model of the other low-orbit satellites to determine the action space of the other low-orbit satellites at the next moment; Performing state transfer on the other low-orbit satellites according to the action space to obtain the state space of the other low-orbit satellites at the next moment; The state space of the other low-orbit satellites at the current moment, the action space, the reward value and the state space of the other low-orbit satellites at the next moment are determined as the experience data.

3. The method according to claim 2, characterized in that The other low-orbit satellites k' The expression of the state space at time t is: Where D is the distance between every two low-orbit satellites The set of L(t,s k' ) are other low-orbit satellites k' The calculated load at time t, For other low-orbit satellites k' To low-orbit satellites k The transmission rate, For other low-orbit satellites k' The computing resources required, is the delay threshold.

4. The method according to claim 2, characterized in that: Calculating the reward value of the other low-orbit satellites at the current moment includes: Obtaining the number of successful tasks successfully processed by the other low-orbit satellites within a preset time period and the number of failed tasks failed to be processed; Determining an active reward value and a penalty value according to the number of successful tasks and the number of failed tasks; The difference between the active reward value and the penalty value is determined as the reward value of the other low-orbit satellite.

5. The method according to claim 2, characterized in that: The step of voting for the low-orbit satellite according to the empirical data of the other low-orbit satellites and the model parameters of the low-orbit satellite comprises: Calculating a loss rate between the model of the other low-orbit satellite and the model of the low-orbit satellite based on the model parameters and empirical data of the other low-orbit satellites and the model parameters of the low-orbit satellite; If the loss rate is greater than the loss tolerance threshold, determining that the voting result of the other low-orbit satellites on the low-orbit satellite is in favor; If the loss rate is less than or equal to the loss tolerance threshold, it is determined that the voting result of the other low-orbit satellites on the low-orbit satellite is against.

6. The method according to claim 5, characterized in that Calculating the loss rate between the model of the other low-orbit satellite and the model of the low-orbit satellite according to the model parameters and empirical data of the other low-orbit satellites and the model parameters of the low-orbit satellite comprises: Calculating the arrival target time difference of the local model of the other low-orbit satellites according to the model parameters and empirical data of the other low-orbit satellites, and calculating the arrival target time difference of the model of the low-orbit satellite according to the model parameters of the low-orbit satellite; The ratio of the arrival target time difference of the model of the other low-orbit satellite to the arrival target time difference of the model of the low-orbit satellite is determined as the loss rate between the model of the other low-orbit satellite and the model of the low-orbit satellite.

7. The method according to claim 6, characterized in that The calculation formula for the time difference to the target is: Where γ is the discount factor, are model parameters, Low Earth Orbit Satellites k In the state space Next execution action space The model parameters are Q value when ; Q value represents the model parameter In the state space Execution Action Space The expected value of the reward obtained.

8. The method according to claim 5, characterized in that Determining the pre-submission state of the low-orbit satellite according to the voting results of the other low-orbit satellites on the low-orbit satellite includes: Obtaining the number of other low-orbit satellites that vote in favor of the low-orbit satellite; Determine the ratio of the number of approvals to the total number of other low-orbit satellites as a verification ratio; If the verification ratio is greater than a preset verification threshold, it is determined that the state of the low-orbit satellite is a pre-submission state.

9. The method according to claim 5, characterized in that The model parameters of the low-orbit satellites in the pre-submission state are modified through the voting mechanism and the cold start mechanism, including: For any low-orbit satellite in a pre-submission state, obtain a first number of objections from other low-orbit satellites that have voted against the low-orbit satellite and a second number of objections from low-orbit satellites that have voted against other low-orbit satellites; Determine the constraint modified voting mechanism parameters according to the first objection number and the second objection number; Determining a cold start reputation aggregation parameter according to the accumulated historical reputation and reputation threshold parameter of the low-orbit satellite; the accumulated historical reputation represents the number of consecutive rounds of global aggregation in which the low-orbit satellite successfully participates; The model parameters of the low-orbit satellite are corrected according to the constraint correction voting mechanism parameters and the cold start reputation aggregation parameters to obtain corrected model parameters.

10. The method according to claim 1, characterized in that The step of performing global aggregation according to the corrected model parameters to obtain global model parameters includes: The modified model parameters are weighted averaged to obtain the global model parameters.

Images