Federal learning method and device for low earth orbit satellite network, and electronic equipment

Abstract

The embodiment of the invention discloses a federated learning method and device for a low earth orbit satellite network and electronic equipment. A specific embodiment of the method comprises the following steps: sending an initial global parameter to each participating satellite through a ground station; for each participating satellite, the following steps are executed: generating trained parameters based on local data collected by the participating satellite, initial global parameters and a preset model; performing secret sharing on the trained parameters to obtain a secret share parameter set; sending each secret share parameter to each aggregation satellite based on a preset safe path set; for each aggregation satellite, performing weighted summation on the secret share parameter from each participating satellite to obtain a weighted secret share parameter, and sending the weighted secret share parameter to a ground station; global training parameters are generated through the ground station according to the received weighted secret share parameters. According to the embodiment, the practicability and safety of federated learning in the satellite network are improved.

Description

Technical Field

[0001] Embodiments of this disclosure relate to the field of computer technology, and more specifically to federated learning methods, apparatus, and electronic devices for low-Earth orbit satellite networks. Background Technology

[0002] Low Earth Orbit (LEO) satellite network technology has made significant progress in recent years, with its unique advantage being the ability to provide seamless broadband access services to users worldwide. These networks, with their low latency and wide coverage, have demonstrated enormous application potential not only in the communications field but also in environmental monitoring, scientific research, and many other areas. With advancements in aerospace technology and reductions in satellite launch costs, a large number of small satellites have been deployed. Equipped with high-resolution sensors and cameras, these satellites operate in low Earth orbit, providing new solutions for Earth observation and wireless broadband coverage.

[0003] Federated learning, as a distributed machine learning framework, allows multiple devices to collaboratively train models without centrally storing sensitive data, which is crucial for protecting data privacy and security. In low-Earth orbit (LEO) satellite networks, the application of federated learning can not only improve data processing efficiency but also enhance the security and privacy of model training. However, combining LEO satellite networks with federated learning faces a number of challenges, including the dynamic nature of network topology, limitations in communication bandwidth, and security issues during data transmission and model updates. These challenges require researchers to develop new technologies to ensure the effectiveness and security of federated learning in LEO satellite networks.

[0004] Security and privacy protection are particularly important in low-Earth orbit (LEO) satellite networks. Due to their open nature, satellite networks are vulnerable to various security threats, including node exposure, open communication channels, and heterogeneous network interconnection. To address these threats, researchers have proposed various security protection technologies, including cryptography-based protection systems and zero-trust security architectures, aiming to protect the data transmission and system security of satellite internet. Furthermore, with the application of federated learning in satellite networks, data privacy and security protection technologies have become a research hotspot, such as data privacy and security protection in federated learning-assisted satellite mobile edge computing based on a blockchain framework. The development of these technologies is crucial for ensuring the secure transmission and processing of data in satellite networks, while also driving the advancement and expansion of satellite communication technologies. Summary of the Invention

[0005] The summary portion of this disclosure is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description portion. This summary portion is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0006] Some embodiments of this disclosure propose federated learning methods, apparatuses, and electronic devices for low-Earth orbit satellite networks to address the technical problems mentioned in the background section above.

[0007] In a first aspect, some embodiments of this disclosure provide a federated learning method for low-Earth orbit satellite networks. The method includes: sending initial global parameters to each participating satellite in a preset set of participating satellites via a ground station; for each participating satellite in the set of participating satellites, performing the following steps: generating trained parameters based on local data collected by the participating satellites, the initial global parameters, and a preset model; secretly sharing the trained parameters to obtain a set of secret share parameters; sending each secret share parameter in the set of secret share parameters to each aggregated satellite in a preset set of aggregated satellites based on a preset set of secure paths; weighting and summing the secret share parameters from each participating satellite using each aggregated satellite in the set of aggregated satellites to obtain weighted secret share parameters, and sending the weighted secret share parameters to the ground station; and generating global training parameters using the ground station based on the received weighted secret share parameters.

[0008] Secondly, some embodiments of this disclosure provide a federated learning apparatus for low-Earth orbit satellite networks. The apparatus includes: a transmitting unit configured to transmit initial global parameters to each participating satellite in a preset set of participating satellites via a ground station; an execution unit configured to perform the following steps for each participating satellite in the set of participating satellites: generating trained parameters based on local data collected by the participating satellites, the initial global parameters, and a preset model; generating a secret share parameter set based on the trained parameters, wherein each secret share parameter in the set of secret share parameters corresponds to a serial number; transmitting each secret share parameter in the set of secret share parameters to each aggregated satellite in a preset set of aggregated satellites based on a preset set of secure paths; a weighted summation unit configured to perform a weighted summation of the secret share parameters from each participating satellite via each aggregated satellite in the set of aggregated satellites to obtain weighted secret share parameters, and transmitting the weighted secret share parameters to the ground station; and a generation unit configured to generate global training parameters via the ground station based on the received weighted secret share parameters.

[0009] Thirdly, some embodiments of this disclosure provide an electronic device, including: one or more processors; and a storage device having one or more programs stored thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation of the first aspect above.

[0010] Fourthly, some embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method described in any of the implementations of the first aspect above.

[0011] The various embodiments disclosed above have the following beneficial effects: Through the federated learning method for low-Earth orbit satellite networks in some embodiments of this disclosure, distributed model training is achieved without centralized storage of sensitive data, effectively protecting the privacy of data collected by satellites; the use of secret share splitting technology enhances the security of model parameters during inter-satellite transmission, ensuring that even if some shares are intercepted, the original data cannot be recovered, thus significantly improving data confidentiality; and by implementing a path protection mechanism, the data transmission path is dynamically adjusted, preventing data security risks caused by path leakage and enhancing the system's resistance to attacks. The technical solution of this invention can adapt to the dynamic and intermittent connectivity characteristics of low-Earth orbit satellite networks, improving the practicality and stability of federated learning in satellite networks. Attached Figure Description

[0012] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0013] Figure 1 This is a flowchart of some embodiments of a federated learning method for low-Earth orbit satellite networks according to the present disclosure;

[0014] Figure 2 This is a schematic diagram of the structure of some embodiments of a federated learning device for low-Earth orbit satellite networks according to the present disclosure;

[0015] Figure 3 This is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure;

[0016] Figure 4 This is a schematic diagram of a federated learning process according to some embodiments of the federated learning method for low-Earth orbit satellite networks disclosed herein;

[0017] Figure 5 This is a schematic diagram of a secret sharing process according to some embodiments of the federated learning method for low-Earth orbit satellite networks disclosed herein;

[0018] Figure 6 This is a flowchart illustrating some embodiments of the federated learning method for low-Earth orbit satellite networks according to the present disclosure. Detailed Implementation

[0019] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0020] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0021] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0022] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0023] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0024] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0025] Figure 1 A flow 100 is shown, illustrating some embodiments of a federated learning method for low-Earth orbit satellite networks according to this disclosure. This federated learning method for low-Earth orbit satellite networks includes the following steps:

[0026] Step 101: Send the initial global parameters to each participating satellite in the preset set of participating satellites via the ground station.

[0027] In some embodiments, the implementer of the federated learning method for low-Earth orbit satellite networks (e.g., a computing device) can send initial global parameters to each participating satellite in a predefined set of participating satellites via a ground station. In practice, this step is the starting point for executing the federated learning task, ensuring that all participating satellites begin training with the same model parameters, laying the foundation for subsequent collaborative learning.

[0028] In some optional implementations of certain embodiments, the execution entity sends the initial global parameters to each participating satellite in a preset set of participating satellites via a ground station, which may include the following steps:

[0029] The first step involves transmitting initial global parameters to each visible low-Earth orbit (LEO) satellite via a ground station. These visible LEO satellites can be any LEO satellite within the ground station's line of sight. In practice, this is first achieved by initializing a set of global parameters at the ground station. These initial global parameters can be parameters corresponding to a pre-defined model, which can be a neural network model. Secondly, the ground station can transmit these initial global parameters to each LEO satellite within its current line of sight via a secure communication channel.

[0030] The second step involves sending the initial global parameters to each participating satellite in a pre-defined set of participating satellites via the aforementioned visible low-Earth orbit (LEO) satellites, based on a pre-defined global routing table. The global routing table can be a data table corresponding to routing information in the satellite network, used to determine the forwarding path of data packets. Each participating satellite in the set of participating satellites can be a satellite participating in collaborative learning. In practice, the executing entity can control each visible LEO satellite to broadcast the initial global parameters to all participating satellites via inter-satellite links based on the global routing table. Optionally, the executing entity can control each visible LEO satellite to broadcast the initial global parameters to all participating satellites via inter-satellite links based on a flooding routing algorithm.

[0031] Step 102, for each participating satellite in the participating satellite set, perform the following steps:

[0032] Step 1021: Generate post-training parameters based on the local data collected by the participating satellites, the initial global parameters, and the preset model.

[0033] In some embodiments, the executing entity can generate post-trained parameters based on the local data collected by the participating satellites, the initial global parameters, and a preset model. The preset model can be the same as the model corresponding to the initial global parameters. The local data can be the dataset collected by the participating satellites. This dataset can be a dataset used for federated learning. For example, when the federated learning task is a land cover category detection task, the dataset can be an image dataset, without specific limitations. In practice, firstly, each participating satellite can collect local data using its onboard sensors and cameras. Secondly, the executing entity can use the participating satellites to train the model based on the local data and the initial global parameters to obtain post-trained parameters. During the training process, the participating satellites can use a gradient descent algorithm to update the initial global parameters, obtaining updated parameters as post-trained parameters.

[0034] In practice, the above steps are performed entirely locally without involving data transmission, thus effectively protecting data privacy.

[0035] Step 1022: Secretly share the trained parameters to obtain the secret share parameter set.

[0036] In some embodiments, the executing entity may secretly share the trained parameters to obtain a secret share parameter set. Each secret share parameter in the secret share parameter set may have the same structure and size as the trained parameters, but different element values. The sum of the element values ​​at the same position of each secret share parameter in the secret share parameter set is equal to the element value at the corresponding position of the trained parameters.

[0037] As an example, assuming the trained parameters are a 3×5 matrix, each secret share parameter in the secret share parameter set is also a 3×5 matrix, but the element values ​​are different. Adding the elements in the first row and first column of each secret share parameter in the secret share parameter set yields a sum equal to the element value in the first row and first column of the trained parameters.

[0038] In some optional implementations of certain embodiments, the execution entity secretly shares the trained parameters to obtain a secret share parameter set, which may include the following steps:

[0039] The first step, for each element included in the above post-training parameters, is to perform the following steps:

[0040] The first sub-step involves generating a random number set based on a preset number of shares. The number of random numbers in this set is equal to the preset number of shares minus one. The preset number of shares can be a positive integer. The preset number of shares can also be equal to the number of secret share parameters in the preset set of secret share parameters.

[0041] The second sub-step is to determine the sum of each random number in the above random number set, thus obtaining the random number sum.

[0042] The third sub-step is to determine the difference between the above elements and the above random number sum as a conventional number.

[0043] Here, by adding the above-mentioned conventional number to the above-mentioned random number, the numerical value corresponding to the above-mentioned element can be obtained.

[0044] The fourth sub-step involves generating a set of secret share elements corresponding to the aforementioned elements, based on the aforementioned set of random numbers and the aforementioned agreed-upon numbers. In practice, each random number in the aforementioned set of random numbers and the aforementioned agreed-upon numbers can be jointly determined as a secret share element to obtain the set of secret share elements. Here, each secret share element in the aforementioned set of secret share elements can correspond to a serial number.

[0045] As an example, assuming the above share number is 3, then the above secret share element set contains three secret share elements, with corresponding serial numbers 1, 2, and 3 respectively.

[0046] The second step is to generate a set of secret share parameters based on the generated sets of secret share elements. In practice, secret share elements with the same index in the generated sets of secret share elements can be combined to form a set of secret share parameters. Here, each secret share parameter in the above set of secret share parameters can correspond to an index.

[0047] In practice, by following the steps above, the trained parameters can be divided into multiple secret shares to ensure that even if some shares are intercepted, the original data cannot be recovered, thereby enhancing the security of data transmission.

[0048] Step 1023: Based on the preset secure path set, send each secret share parameter in the secret share parameter set to each aggregated satellite in the preset aggregated satellite set.

[0049] In some embodiments, the executing entity can send each secret share parameter in the secret share parameter set to each aggregated satellite in the preset aggregated satellite set, based on a preset set of secure paths. The number of secure paths in the secure path set can be equal to the number of shares. Each secure path in the secure path set can be a link composed of multiple satellites. Each secure path in the secure path set can correspond one-to-one with an aggregated satellite in the aggregated satellite set. The last satellite node of each secure path in the secure path set can be an aggregated satellite in the aggregated satellite set. The number of aggregated satellites in the aggregated satellite set can be equal to the number of shares. The secret share parameters in the secret share parameter set can correspond one-to-one with the secure paths in the secure path set, and simultaneously, the secret share parameters can correspond one-to-one with the aggregated satellites in the aggregated satellite set. Each aggregated satellite in the aggregated satellite set can also correspond to a serial number. The secret share parameter in the secret share parameter set can be sent to the aggregated satellite with the same serial number. Here, each aggregated satellite can be selected from the various visible low-Earth orbit satellites based on link stability, communication quality, topology location, and resource status. A schematic diagram of the federated learning process based on preset paths is shown below. Figure 4 As shown. Figure 4 In this context, "source satellite" refers to the participating satellites, "target satellite" refers to the target satellites, "local model" describes the trained parameters, and "partial aggregation satellites" refers to the aggregated satellites.

[0050] As an example, when ground station A needs to select 3 satellites from overhead satellites to receive federated learning parameters, the decision-making process is as follows: First, satellites with insufficient resources are directly eliminated (e.g., satellite B is excluded because its CPU load exceeds 75%). Then, the remaining satellites are sorted by visible duration to form a preliminary list (satellite C's 10 minutes > satellite D's 9 minutes > satellite A's 8 minutes). Next, dynamic fine-tuning is performed: priority is given to ensuring communication quality. If the satellite signal strength is below the threshold (e.g., satellite C's 22dB is close to the critical value of 20dB), it is postponed to the next one. Finally, network topology optimization is introduced. When the satellite performance is similar (e.g., satellites D and A both meet the requirements), network hub nodes are selected first (satellite D is a secondary hub, superior to edge node satellite A). Finally, the receiving satellites are determined to be C (high stability), D (balanced hub), and A (safety net node).

[0051] Optionally, the above set of secure paths can be generated through the following steps:

[0052] The first step is to generate an initial path sequence corresponding to the participating satellites based on a pre-defined global routing table. Each initial path in the initial path sequence can contain multiple participating satellites. The number of initial paths in the initial path sequence can be equal to the aforementioned share. Each initial path in the initial path sequence can be a link composed of multiple satellites. Each initial path in the initial path sequence can correspond one-to-one with an aggregated satellite in the aggregated satellite set. Each initial path in the initial path sequence is arranged in descending order according to the number of participating satellites it contains. The length of each initial path in the initial path sequence can be represented by the number of participating satellites it contains. In practice, for each aggregated satellite in the aggregated satellite set, the shortest path between the participating satellite and the aggregated satellite can be selected from the global routing table as the initial path.

[0053] The second step involves determining the number of initial paths traversed by each participating satellite in the aforementioned participating satellite set as the number of repeated transmissions. This repeated transmission count characterizes the number of initial paths passing through the participating satellite. Here, when a participating satellite is currently performing federated learning training, its repeated transmission count is not counted.

[0054] As an example, suppose there are three initial paths in the initial path sequence: "Participating Satellite 1, Participating Satellite 2, Participating Satellite 3, Aggregating Satellite 1"; "Participating Satellite 2, Participating Satellite 5, Aggregating Satellite 2"; and "Participating Satellite 2, Participating Satellite 4, Aggregating Satellite 3". Then, the number of repeated transmissions corresponding to Satellite 1 is 1. The number of repeated transmissions corresponding to Satellite 2 is 3.

[0055] The third step involves identifying the participating satellites corresponding to each of the determined repeated transmission counts that is greater than or equal to a preset safety count as satellites to be adjusted, thus obtaining the set of satellites to be adjusted. The aforementioned safety count can be a positive integer. The value of the aforementioned safety count can be less than or equal to the aforementioned share number. The specific value of the aforementioned safety count is not specifically limited here.

[0056] Fourth, for each initial path in the initial path sequence, perform the following steps:

[0057] The first sub-step involves adjusting the nodes of the initial path in response to the determination that at least one satellite from the set of satellites to be adjusted exists in the initial path, thereby obtaining a safe path. In practice, after adjusting the nodes of the initial path, the steps in the second and third steps can be re-executed to update the number of repeated transmissions corresponding to each participating satellite in the participating satellite set, as well as to update the set of satellites to be adjusted.

[0058] In practice, since the initial paths in the above initial path sequence are arranged from longest to shortest, the longest initial path in the sequence is processed first. Because the longest path contains more nodes and has a longer transmission time, its satellite nodes are more likely to be reused, and its exposure window for attacks is significantly higher. Starting with the longest path can quickly eliminate the main risks and reduce computational load. At the same time, it avoids modifying the short, low-latency paths (replanning could lead to a surge in latency), minimizing the impact on system efficiency while improving security.

[0059] The second sub-step is to determine the initial path as a safe path in response to the determination that there is no satellite to be adjusted in the set of satellites to be adjusted in the initial path.

[0060] Optionally, the execution entity may adjust the nodes of the initial path to obtain a safe path, which may include the following steps:

[0061] First, based on the initial path described above, perform the following adjustments:

[0062] The first sub-step involves determining the participating satellite corresponding to the previous node of the first satellite to be adjusted in the initial path as the previous hop satellite. In practice, firstly, when the initial path contains two or more satellites to be adjusted, the satellite that appears first in the initial path can be determined as the first satellite to be adjusted. Secondly, the participating satellite corresponding to the previous node of the first satellite to be adjusted can be determined as the previous hop satellite.

[0063] The second sub-step involves identifying the last participating satellite in the initial path as the target satellite. This target satellite can be an aggregated satellite from the aforementioned aggregated satellite set.

[0064] The third sub-step involves generating the shortest path based on the previous-hop satellite and the target satellite. In practice, this is done by first identifying the nearest neighbor satellite to the target satellite among the neighboring nodes of the previous-hop satellite (excluding the satellite to be adjusted). Then, the shortest path from that neighbor satellite to the target satellite is found using the global routing table.

[0065] The fourth sub-step involves generating an updated path based on the initial path, the previous hop satellite, and the shortest path described above. In practice, the shortest path can be concatenated with a portion of the initial path preceding the previous hop satellite to form a new path as the updated path. After generating the updated path, steps two and three can be re-executed to update the number of repeated transmissions for each participating satellite in the participating satellite set, as well as to update the set of satellites to be adjusted.

[0066] The fifth sub-step is to determine that, in response to the determination that there is no satellite to be adjusted in the above-mentioned set of satellites to be adjusted in the above-mentioned updated path, the above-mentioned updated path is determined as a safe path.

[0067] The second step is to determine that at least one satellite from the set of satellites to be adjusted exists in the updated path, and then use the updated path as the initial path to perform the adjustment steps again.

[0068] In practice, since the encrypted parameters need to be transmitted between satellites, if multiple paths pass through the same satellite node, the original parameters can be obtained by adding the values ​​of the traversed satellites. For satellite nodes whose repeated transmissions reach a security threshold, the leakage of secret shares can be avoided by dynamically selecting neighboring nodes and finding the shortest path in the above steps.

[0069] Step 103: For each aggregated satellite in the aggregated satellite set, perform a weighted summation of the secret share parameters from each participating satellite to obtain a weighted secret share parameter, and send the weighted secret share parameter to the ground station.

[0070] In some embodiments, the aforementioned executing entity may use each of the aggregated satellites in the aforementioned aggregated satellite set to perform a weighted summation of the secret share parameters from each participating satellite to obtain a weighted secret share parameter, and then send the weighted secret share parameter to the aforementioned ground station.

[0071] In some optional implementations of certain embodiments, the execution entity performs a weighted summation of the secret share parameters from each participating satellite in the aggregated satellite set to obtain a weighted secret share parameter, and then sends the weighted secret share parameter to the ground station. This may include the following steps:

[0072] The first step is to determine the participating satellite weight for each participating satellite in the aforementioned participating satellite set, based on the local data of each participating satellite. In practice, this can be achieved by first summing the memory usage of the local data for each participating satellite in the set to obtain the total dataset memory value. Then, the ratio between the memory usage of the local data for each participating satellite and the total dataset memory value can be used to determine the participating satellite weight for that satellite.

[0073] The second step involves generating a weighted secret share parameter for each aggregated satellite in the aforementioned aggregated satellite set, based on the secret share parameters received from each parameter satellite and the corresponding parameter satellite weights. This weighted secret share parameter is then transmitted to the aforementioned ground station. Each aggregated satellite in the aforementioned aggregated satellite set can receive secret share parameters with the same sequence number from each parameter satellite. In practice, the weighted secret share parameter for each aggregated satellite in the aforementioned aggregated satellite set can be generated using the following formula.

[0074]

[0075] Where j represents the sequence number. This represents the weighted secret share parameter for the j-th unit. m represents the number of participating satellites in the satellite pool. This represents the memory size of the local data for the i-th participating satellite. L represents the total memory size of the dataset. This represents the weight of the participating satellite corresponding to the i-th participating satellite. This represents the j-th secret share parameter of the i-th participating satellite.

[0076] In practice, the secret sharing process corresponding to steps 102 to 103 above is as follows: Figure 5 As shown. Figure 5 In the text, "training satellites" refers to participating satellites, "partially aggregated satellites" refers to aggregated satellites, and "curious satellites" refers to satellites awaiting adjustment.

[0077] Step 104: Generate global training parameters through the ground station based on the received weighted secret share parameters.

[0078] In some embodiments, the aforementioned executing entity can generate global training parameters based on the received weighted secret share parameters via the aforementioned ground station. In practice, the global training parameters can be generated based on the received weighted secret share parameters using the following formula.

[0079]

[0080] in, This represents the global training parameters. n represents the number of shares.

[0081] Optionally, the execution entity may, in response to determining that the global training model corresponding to the global training parameters does not meet the preset convergence condition, use the global training parameters as new initial global parameters and regenerate the global training parameters. In practice, when the global training model corresponding to the global training parameters does not meet the preset convergence condition, the global training parameters can be used as new initial global parameters, and the operations in steps 101 to 104 can be performed again to generate global training parameters. The process is as follows: Figure 6 As shown.

[0082] Further reference Figure 2 As an implementation of the methods shown in the above figures, this disclosure provides some embodiments of a federated learning device for low-Earth orbit satellite networks, these device embodiments being similar to... Figure 1 Corresponding to the method embodiments shown, this federated learning device for low-Earth orbit satellite networks can be specifically applied to a variety of electronic devices.

[0083] like Figure 2As shown, a federated learning device 200 for low-Earth orbit satellite networks in some embodiments includes: a sending unit 201, an execution unit 202, a weighted summation unit 203, and a generation unit 204. The sending unit 201 is configured to send initial global parameters to each participating satellite in a preset set of participating satellites via a ground station; the execution unit 202 is configured to perform the following steps for each participating satellite in the set of participating satellites: generating trained parameters based on local data collected by the participating satellites, the initial global parameters, and a preset model; generating a secret share parameter set based on the trained parameters, wherein each secret share parameter in the set of secret share parameters corresponds to a serial number; sending each secret share parameter in the set of secret share parameters to each aggregated satellite in a preset set of aggregated satellites based on a preset set of secure paths; the weighted summation unit 203 is configured to perform a weighted summation of the secret share parameters from each participating satellite via each aggregated satellite in the set of aggregated satellites to obtain a weighted secret share parameter, and sending the weighted secret share parameter to the ground station; and the generation unit 204 is configured to generate global training parameters via the ground station based on the received weighted secret share parameters.

[0084] It is understandable that the units and references described in the Federated Learning Device 200 for Low Earth Orbit Satellite Networks Figure 1 The steps in the described method correspond accordingly. Therefore, the operations, features, and beneficial effects described above for the method are also applicable to the Federated Learning Device 200 for low-Earth orbit satellite networks and the units contained therein, and will not be repeated here.

[0085] The following is for reference. Figure 3 It illustrates a schematic diagram of the structure of an electronic device (such as a computing device) suitable for implementing some embodiments of the present disclosure. Figure 3 The electronic device shown is merely an example and should not be construed as limiting the functionality or scope of the embodiments of this disclosure. Figure 3 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The memory may include a non-volatile storage medium and internal memory. The non-volatile storage medium may store an operating system and a computer program. The computer program includes program instructions that, when executed, cause the processor to perform any of the methods described above. The processor provides computational and control capabilities to support the operation of the entire computer device. The internal memory provides an environment for the execution of the computer program in the non-volatile storage medium; when executed by the processor, the computer program causes the processor to perform any of the methods described above. The network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that... Figure 3The structure shown is merely a block diagram of a portion of the structure related to the present disclosure and does not constitute a limitation on the computer device to which the present disclosure is applied. A specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0086] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0087] In one embodiment, the processor is configured to run a computer program stored in a memory to perform the following steps: sending initial global parameters to each participating satellite in a preset set of participating satellites via a ground station; for each participating satellite in the set of participating satellites, performing the following steps: generating trained parameters based on local data collected by the participating satellites, the initial global parameters, and a preset model; secretly sharing the trained parameters to obtain a set of secret share parameters; sending each secret share parameter in the set of secret share parameters to each aggregated satellite in a preset set of aggregated satellites based on a preset set of secure paths; performing a weighted summation of the secret share parameters from each participating satellite via each aggregated satellite in the set of aggregated satellites to obtain a weighted secret share parameter, and sending the weighted secret share parameter to the ground station; and generating global training parameters via the ground station based on the received weighted secret share parameters.

[0088] This disclosure also provides a computer-readable storage medium storing a computer program, the computer program including program instructions, and the method implemented when the program instructions are executed can be referred to the various embodiments of the methods described above.

[0089] The aforementioned computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiments, such as the hard disk or memory of the computer device. Alternatively, the aforementioned computer-readable storage medium may be an external storage device of the computer device, such as a plug-in hard disk, SmartMedia Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the computer device.

[0090] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0091] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A federated learning method for low-Earth orbit satellite networks, characterized in that, include: The initial global parameters are sent from the ground station to each participating satellite in the preset set of participating satellites; For each participating satellite in the participating satellite set, perform the following steps: Based on the local data collected by the satellite, the initial global parameters, and the preset model, the trained parameters are generated. The trained parameters are secretly shared to obtain a secret share parameter set; Based on a preset set of secure paths, each secret share parameter in the set of secret share parameters is sent to each aggregated satellite in the preset set of aggregated satellites; For each of the aggregated satellites in the aggregated satellite set, the secret share parameters from each participating satellite are weighted and summed to obtain the weighted secret share parameters, and the weighted secret share parameters are sent to the ground station. The ground station generates global training parameters based on the received weighted secret share parameters.

2. The method according to claim 1, characterized in that, The method further includes: In response to the determination that the global training model corresponding to the global training parameters does not meet the preset convergence condition, the global training parameters are used as new initial global parameters to generate global training parameters again.

3. The method according to claim 1, characterized in that, The step of sending the initial global parameters to each participating satellite in a preset set of participating satellites via a ground station includes: Initial global parameters are sent to each visible low-Earth orbit satellite via ground station; Based on a preset global routing table, the initial global parameters are sent to each participating satellite in a preset set of participating satellites via each visible low-Earth orbit satellite.

4. The method according to claim 1, characterized in that, The step of secretly sharing the trained parameters to obtain a secret share parameter set includes: For each element included in the post-training parameters, perform the following steps: Based on a preset number of shares, a set of random numbers is generated, wherein the number of random numbers in the set is equal to the number of shares minus one; The sum of all random numbers in the set is determined to obtain the random number sum; The difference between the element and the sum of the random numbers is determined as a conventional number; Generate a set of secret share elements corresponding to the element based on the set of random numbers and the agreed number; Generate a set of secret share parameters based on the generated sets of secret share elements.

5. The method according to claim 1, characterized in that, The set of secure paths is generated through the following steps: Based on a preset global routing table, an initial path sequence corresponding to the participating satellites is generated, wherein each initial path in the initial path sequence contains multiple participating satellites, and each initial path in the initial path sequence is arranged in descending order according to the number of participating satellites it contains; For each participating satellite in the participating satellite set, the initial number of paths existing for the participating satellite is determined as the number of repeated transmissions; The participating satellites corresponding to each of the determined repeated transmission counts that are greater than or equal to the preset safety count are identified as satellites to be adjusted, thus obtaining the set of satellites to be adjusted; For each initial path in the initial path sequence, perform the following steps: In response to determining that at least one satellite in the set of satellites to be adjusted exists in the initial path, the initial path is adjusted by nodes to obtain a safe path; In response to determining that none of the satellites to be adjusted from the set of satellites to be adjusted are present in the initial path, the initial path is determined to be a safe path.

6. The method according to claim 5, characterized in that, The step of adjusting the nodes of the initial path to obtain a safe path includes: Based on the initial path, perform the following adjustment steps: The participating satellite corresponding to the previous node of the first satellite to be adjusted in the initial path is determined as the previous hop satellite; The last participating satellite in the initial path is identified as the target satellite; Based on the previous hop satellite and the target satellite, generate the shortest path; The updated path is generated based on the initial path, the previous hop satellite, and the shortest path; In response to determining that none of the satellites to be adjusted in the set of satellites to be adjusted exist in the updated path, the updated path is determined as a safe path; In response to determining that at least one satellite from the set of satellites to be adjusted exists in the updated path, the updated path is used as the initial path, and the adjustment step is executed again.

7. The method according to claim 1, characterized in that, The step of performing a weighted summation of the secret share parameters from each participating satellite through each of the aggregated satellites in the aggregated satellite set to obtain a weighted secret share parameter, and then sending the weighted secret share parameter to the ground station, includes: Based on the local data of each participating satellite in the participating satellite set, determine the participating satellite weight corresponding to each participating satellite in the participating satellite set; Each aggregated satellite in the aggregated satellite set generates a weighted secret share parameter based on the secret share parameters received from each parameter satellite and the parameter satellite weights corresponding to each parameter satellite, and then sends the weighted secret share parameter to the ground station.

8. A federated learning device for low-Earth orbit satellite networks, characterized in that, include: The transmitting unit is configured to transmit initial global parameters to each participating satellite in a preset set of participating satellites via a ground station; The execution unit is configured to perform the following steps for each participating satellite in the participating satellite set: Based on the local data collected by the satellite, the initial global parameters, and the preset model, the trained parameters are generated. Based on the trained parameters, a secret share parameter set is generated, wherein each secret share parameter in the secret share parameter set corresponds to a serial number; Based on a preset set of secure paths, each secret share parameter in the set of secret share parameters is sent to each aggregated satellite in the preset set of aggregated satellites; The weighted summation unit is configured to perform a weighted summation of the secret share parameters from each participating satellite through each of the aggregated satellites in the aggregated satellite set, to obtain a weighted secret share parameter, and to send the weighted secret share parameter to the ground station; The generation unit is configured to generate global training parameters via the ground station based on the received weighted secret share parameters.

9. An electronic device, characterized in that, include: One or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1 to 7.

10. A computer-readable medium, characterized in that, It stores a computer program thereon, wherein the computer program, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images