Data balance optimization method and device based on multi-satellite collaboration
By selecting the satellite with the largest amount of data in the low-Earth orbit satellite network as the optimization satellite, and selecting neighboring satellites for data offloading based on the data size and class label differences between satellites, and establishing a constrained objective function for optimization, the problems of slow federated learning training and model instability in low-Earth orbit satellite networks are solved, achieving rapid model convergence and improving accuracy.
Patent Information
- Application Number
- CN202511192531.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-25
- Publication Date
- 2025-12-09
AI Technical Summary
In low-Earth orbit satellite networks, federated learning faces challenges such as short and unstable communication windows, large differences in data characteristics leading to unstable and delayed model training results, and slow training processes.
By establishing and optimizing the constraint objective function, the constraint objective function between the optimized satellite and the adjacent satellite is selected for optimization, the target data offloading ratio of the optimized satellite is determined, and data offloading is performed between the optimized satellite and the adjacent satellite.
By offloading data from multiple satellites in a collaborative manner, the training burden on satellites with large amounts of data is reduced, the convergence speed of federated learning models is accelerated, the problem of data heterogeneity is alleviated, and the accuracy and stability of the models are improved.
Smart Images
Figure CN121098356A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] Embodiments of the present disclosure relate to the field of data balancing optimization based on multi-satellite cooperation, and in particular to a data balancing optimization method and device based on multi-satellite cooperation. BACKGROUND
[0002] Low Earth Orbit (LEO) satellites have become one of the key supports for the development of the sixth generation of mobile communication technology (6G) due to their low latency, wide coverage, and high transmission rate, especially in the fields of edge computing and satellite communication. With the rapid development of low-orbit satellite constellations, especially the deployment of SpaceX's Starlink and China's GW, Qianxuan constellation, and other projects, these satellite networks not only make up for the deficiencies of ground networks in remote areas, but also provide strong support for smart cities, disaster monitoring, remote medical care, and other applications. However, due to limited downlink bandwidth and unstable connections, satellites cannot transmit all image data to ground stations for machine learning model training. Therefore, federated learning (FL) as a distributed machine learning method provides an efficient solution in this scenario by training models locally on satellites, avoiding large data transmission, reducing bandwidth consumption, and effectively addressing privacy protection needs.
[0003] Federated learning as a new type of distributed machine learning framework is proposed, allowing each node (such as a low-orbit satellite) to train a model locally and only upload model update parameters rather than raw data. This approach effectively alleviates bandwidth bottlenecks during data transmission and reduces communication latency. However, in low-orbit satellite networks, federated learning faces many challenges. First, the communication window between low-orbit satellites and ground stations is short and unstable. Second, due to the differences in data characteristics (Non-IID) between low-orbit satellites, the local models generated by each satellite may differ significantly, increasing the complexity of global model training and causing large delays in the federated learning training process, which cannot meet real-time requirements and severely affects model accuracy and stability. SUMMARY
[0004] The summary of the present disclosure is intended to introduce the concepts in a simplified form, which will be described in detail in the specific embodiments section. The summary of the present disclosure 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.
[0005] Some embodiments of the present disclosure propose a data balancing optimization method and device based on multi-satellite cooperation, electronic equipment, and computer readable medium to solve the technical problems mentioned in the background section.
[0006] In a first aspect, some embodiments of the present disclosure provide a data balancing optimization method based on multi-satellite cooperation, which comprises: selecting a satellite with the largest local training data set as an optimization satellite; selecting neighboring satellites of the optimization satellite for data offloading according to data size and class label difference between the satellites; establishing a constraint objective function between the optimization satellite and the neighboring satellites, and solving the constraint objective function to determine a target data offloading ratio of the optimization satellite; and performing data offloading between the optimization satellite and the neighboring satellites according to the target data offloading ratio.
[0007] In a second aspect, some embodiments of the present disclosure provide a data balancing optimization device based on multi-satellite cooperation, which comprises: a first selection unit configured to select a satellite with the largest local training data set as an optimization satellite; a second selection unit configured to select neighboring satellites of the optimization satellite for data offloading according to data size and class label difference between the satellites; an establishment unit configured to establish a constraint objective function between the optimization satellite and the neighboring satellites, and solve the constraint objective function to determine a target data offloading ratio of the optimization satellite; and an offloading unit configured to perform data offloading between the optimization satellite and the neighboring satellites according to the target data offloading ratio.
[0008] In a third aspect, some embodiments of the present disclosure provide an electronic device, which comprises: one or more processors; and a storage device having one or more programs stored thereon, wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the method described in any of the implementations of the first aspect.
[0009] In a fourth aspect, some embodiments of the present 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.
[0010] The above-mentioned various embodiments of the present disclosure have the following beneficial effects: through the data balancing optimization method based on multi-satellite cooperation of some embodiments of the present disclosure, the training burden of a satellite with a large amount of data is reduced through multi-satellite cooperative data offloading, the convergence speed of the federated learning model is accelerated, and the problem of slow training convergence in a low-orbit giant constellation is effectively addressed. By considering the difference in data size and class label distribution between satellites for data offloading, the data heterogeneity problem is alleviated, and the accuracy of the federated learning model is improved. The federated learning framework is used for model training, each satellite uses its local data set for training, data transmission is avoided, the bandwidth pressure in the transmission process is reduced, and the risk of data leakage is reduced. BRIEF DESCRIPTION OF DRAWINGS
[0011] The above and other features, aspects and advantages of the present disclosure will become more apparent after a reading of the following detailed description together with the accompanying drawings. Throughout the drawings, similar reference numerals are used to represent similar elements. It should be understood that the drawings are schematic and elements and features do not necessarily appear in proportion in the drawings.
[0012] Figure 1 is a flowchart of some embodiments of a data balancing optimization method based on multi-satellite cooperation according to the present disclosure;
[0013] Figure 2 is a structural schematic diagram of some embodiments of a data balancing optimization apparatus based on multi-satellite cooperation according to the present disclosure;
[0014] Figure 3 is a structural schematic diagram of an electronic device suitable for implementing some embodiments of the present disclosure. DETAILED DESCRIPTION
[0015] Embodiments of the present disclosure will be described in detail with reference to the drawings, wherein the same or similar components are denoted by the same reference numerals, and thus, their repetitive description is omitted. As those skilled in the art would realize, the present disclosure can be carried out in many different configurations of software, hardware, and / or firmware. The present disclosure, thus, can be implemented in various ways, and should not be limited to the embodiments set forth in the examples below. Also, aspects of the present disclosure can comprise any combination of features from different embodiments.
[0016] It should also be noted that, in the drawings, only parts related to the present disclosure are shown. The embodiments in the present disclosure and the features in the embodiments can be combined with each other in the case of no conflict.
[0017] It should be noted that the terms "first", "second", and the like in the present disclosure are used only to distinguish different devices, modules or units, and do not imply the order or sequence of the functions executed by the devices, modules or units.
[0018] It should be noted that the terms "one", "multiple" in the present disclosure are illustrative and not restrictive, and those skilled in the art should understand that "one" or "multiple" should be understood as "one or more" unless otherwise explicitly indicated in the context.
[0019] The names of the messages or information exchanged between the devices in the embodiments of the present disclosure are only for illustrative purposes, and are not used to limit the scope of the messages or information.
[0020] The present disclosure will be described in detail with reference to the accompanying drawings and embodiments.
[0021] Figure 1is a flow 100 of some embodiments of the multi-satellite coordination based data balancing optimization method of some embodiments of the present disclosure. The multi-satellite coordination based data balancing optimization method includes the following steps:
[0022] Step 101, select the satellite with the largest local training dataset as the optimization satellite.
[0023] In some embodiments, the execution subject (e.g., a computing device) of the multi-satellite coordination based data balancing optimization method can select the satellite with the largest local training dataset as the optimization satellite. The satellite with the largest local training dataset is selected as the optimization satellite v a as the current optimization object (optimization satellite).
[0024] Step 102, select the neighboring satellites of the optimization satellite for data offloading according to the data size and class label difference between satellites.
[0025] In some embodiments, the execution subject can select the neighboring satellites of the optimization satellite for data offloading according to the data size and class label difference between satellites.
[0026] In practice, the execution subject can select the neighboring satellites of the optimization satellite for data offloading according to the data size and class label difference between satellites by the following steps:
[0027] First, for each satellite that establishes a direct communication connection with the optimization satellite, the following processing steps are performed:
[0028] 1. Determine the class label difference between the satellite and the optimization satellite. Based on the performance bottleneck of the current satellite v a , optimization is performed. Through optimization, the satellite v a unloads part of the computing task to the neighboring satellite, thereby reducing its own training burden and improving the overall training speed. The neighboring satellite refers to a limited set that establishes a direct communication connection with the current optimization satellite through an inter-satellite link, not all satellites in the constellation. In v a selecting the neighboring satellite for offloading, not only the size of the data amount difference between the two satellites is considered, but also the difference in class labels in the data, which can be represented as:
[0029] .
[0030] where D a is the dataset of satellite v a , specifically the observation data samples (such as images or sensor records) and their corresponding class labels collected in orbit by the satellite v a represents the dataset D a the size of the i-th class label of the data (the class to which the data belongs, in the simulation, it is usually the standard class of MNIST or CIFAR-10), C represents the sum of the categories of the class labels of the data set, k is a weighting parameter for balancing the weight between the data amount difference and the data class label distribution.
[0031] 2. Determine the communication delay between the satellite and the above-mentioned optimized satellite. The delay of each satellite consists of two parts: data offloading delay and model training delay. The data offloading delay refers to the delay generated when the satellite v a transmits data to its adjacent satellite v b :
[0032] .
[0033] wherein, represents the data offloading delay of the satellite v a , wherein, represents the offloading ratio of satellite data, and for convenience, the offloading ratio of satellite data is represented as I represents I ab , I ab represents the amount of data offloaded from the satellite v a , B a is the bandwidth of the satellite v a , represents the communication power allocated to the satellite v a , , is the channel fading coefficient, and N0 is the variance of the complex Gaussian white channel noise. The offloading method finally adopted is a multi-hop flow-based offloading method, which will have multiple offloading flows, different flows do not intersect, a satellite can only be in one offloading flow, and the satellite in each flow set can be represented as , f={1, 2, 3…F}, the time of each offloading flow is the sum of the offloading delays of the satellites in the flow, which can be represented as:
[0034] .
[0035] wherein, represents the total offloading delay of the offloading flow f. For all satellites participating in the offloading flow f, the offloading delay of each satellite is calculated respectively . Then the maximum one is taken as the delay of the entire offloading flow.
[0036] In the process of offloading data between two satellites, after the satellite completes the data offloading, machine learning training is performed, and the delay of model training calculation of the satellite v a can be represented as:
[0037] .
[0038] where, denotes the total number of epochs, is the CPU frequency of satellite v a , in cycles per second or Hz, m a is the number of CPU cycles that satellite v a takes to process one data sample. The relationship between computational power and frequency can be calculated as:
[0039] .
[0040] where, is the effective capacitance coefficient of satellite v a , is the computational power of satellite v a .
[0041] In a low earth orbit satellite constellation, the training latency of federated learning is determined by the slowest satellite. The total latency in a round can be expressed as:
[0042] .
[0043] where, is the time taken by the satellite to aggregate at the ground station after training, which is a fixed value. N is the number of satellites in the constellation, and R represents the round of federated learning. The total system latency required for federated learning can be expressed as:
[0044] .
[0045] According to the above analysis, the optimization problem of the system delay of federated learning can be expressed as:
[0046] .
[0047] Specifically, constraint C1 ensures that the proportion of data offloaded by the satellite does not exceed the upper limit of the maximum sensitive data proportion. Since the physical resources of the satellite, such as computing power and communication bandwidth, are limited, its maximum power is also strictly limited. In order to avoid excessive consumption of satellite resources, it is necessary to control the amount of offloaded data to ensure that it does not exceed a reasonable threshold. Constraints C2 and C3 respectively constrain the communication power and computing power of the satellite, requiring that both power consumptions must be within the maximum power limit of each satellite. Specifically, the communication and computing power of each satellite needs to be reasonably set according to the hardware capabilities of the satellite to ensure that it does not exceed its maximum available power, otherwise it will cause the satellite to be overloaded or unable to complete the task. The purpose of constraint C4 is to ensure that the optimized delay time does not exceed the original delay without data offloading. In practical applications, although the data transmission rate between satellites is already quite fast, the computing power of the satellite is usually limited, so the offloading delay of the satellite in federated learning is usually lower than the model training delay. However, in some cases, if the amount of data transmitted is large and the communication capability of the satellite is insufficient to support high-speed transmission, the offloading delay may become very high. For example, when transmitting high-resolution images, the satellite communication is weak, the number of rounds of federated learning is small, and the data offloading delay may be high. In this case, data offloading will not result in any delay improvement. In this case, continuing data offloading may bring a lot of additional delay, not only wasting resources, but also affecting the performance of the overall system. Therefore, the introduction of constraint C4 is to prevent the delay of offloaded data from being too high, so that the offloading operation becomes invalid or no longer meaningful.
[0048] In a second step, the neighboring satellite of the optimized satellite is determined as the satellite for data offloading according to the class label difference and the communication delay. For example, the satellite with the smallest difference in class label difference and satisfying the optimization problem of the system delay of the federated learning can be determined as the neighboring satellite, as the satellite for data offloading.
[0049] In step 103, a constraint objective function between the optimized satellite and the neighboring satellite is established, and the constraint objective function is solved to determine the target data offloading proportion of the optimized satellite.
[0050] In some embodiments, the execution subject can establish a constraint objective function between the optimized satellite and the neighboring satellite, and solve the constraint objective function to determine the target data offloading proportion of the optimized satellite.
[0051] For example, the delay optimization problem described above is a non-convex optimization problem, and the system training delay is converted into a constraint objective function between two satellites, which can be expressed as:
[0052] .
[0053] The above function can be converted and split into two sub-problems, one main problem, and the optimization of unloading delay for sub-problem 1:
[0054] .
[0055] The derivative of the sub-problem is a monotonically decreasing function, and the derivative of the sub-problem The optimization of the training delay of the two satellites is also a monotonous function, and the derivative can get the maximum value. The optimization takes the maximum value of the calculation power And the transmission power , and the main problem is obtained after being brought in:
[0056] .
[0057] .
[0058] The main problem is substituted into the solver to obtain the final target data offloading ratio: .
[0059] In step 104, according to the above target data offloading ratio, data offloading is performed between the above optimization satellite and the above adjacent satellite.
[0060] In some embodiments, the execution subject can perform data offloading between the above optimization satellite and the above adjacent satellite according to the above target data offloading ratio.
[0061] After determining the optimal offloading amount (target data offloading ratio), the class labels between the data of the v a and v b satellites are offloaded, so that the class labels between the data are offloaded on the basis of ensuring the minimum delay, and the class label distribution between the two satellites is uniform. The method adopted by the present application is to first calculate the sum of the offloaded data amount of each class label to achieve balance, and then calculate the difference between the optimal offloading data amount and the increase or decrease of the optimal class label offloading amount in the same proportion to make their sum equal to the optimal data offloading amount. The specific method is to first calculate the optimal offloading amount of each class label , i represents the i-th class label, so that the class label distribution between the two satellites is balanced , then calculate the sum of the of each class label , and compare it with the target data offloading ratio calculated before:
[0062] In practice, the execution subject can offload data between the above optimization satellite and the above adjacent satellite by the following steps:
[0063] A first step is to determine the optimal offloading amount of each class label between the optimization satellite and the adjacent satellite .
[0064] A second step is to determine the sum of the optimal offloading amount of each class label as the total offloading amount .
[0065] A third step is to, in response to determining that the target data offloading ratio is the same as the total offloading amount, offload the data of the optimization satellite to the adjacent satellite according to the optimal offloading amount of each class label. Case 1, Then, the offloading of each class label is performed according to the optimal offloading amount.
[0066] A fourth step is to, in response to determining that the target data offloading ratio is less than the total offloading amount, subtract an equal amount of offloading amount from the optimal offloading amount of each class label to obtain the balanced offloading amount of the class label, and offload the data of the optimization satellite corresponding to the class label to the adjacent satellite according to the balanced offloading amount. Case 2, At this time, the sum of the optimal offloading amount of each class label exceeds the offloading amount of the minimum time delay, and the method adopted in this case is to subtract an equal amount of offloading amount from the optimal offloading amount of each class label . The updated value of each class label is:
[0067] .
[0068] A fifth step is to, in response to determining that the target data offloading ratio is greater than the total offloading amount, add an equal amount of offloading amount to the optimal offloading amount of each class label to obtain the balanced offloading amount of the class label, and offload the data of the optimization satellite corresponding to the class label to the adjacent satellite according to the balanced offloading amount. Case 2, At this time, the sum of the optimal offloading amount of each class label is less than the offloading amount of the minimum time delay, and the method adopted in this case is to add an equal amount of offloading amount to the optimal offloading amount of each class label . The updated value of each class label is:
[0069] .
[0070] Optionally, the present disclosure ensures that the optimization cost of each round of iteration is kept decreasing in the design, and if the cost cannot be further reduced, the optimization process is terminated, that is, at the beginning of the next round, a new offloading satellite is selected and it is determined whether the constraints are met, if the constraints are met, the offloading amount is calculated and the offloading of the class label between the two satellites is completed, and if the constraints are not met, the loop is ended.
[0071] Optionally, the disclosure designs an unloading mechanism, that is, after the satellite unloads data to the adjacent satellite, the satellite will no longer participate in the subsequent optimization round. The core of this design is to effectively avoid conflicts in the unloading process by limiting the number of satellites participating in optimization, thereby ensuring the formation of multiple independent transmission flows in the low-orbit satellite network. Data transmission within each transmission flow can be carried out in a predetermined order in series, that is, after receiving the data, the satellite can unload the data to the next hop in sequence. This mechanism greatly improves the reliability of data transmission and the utilization efficiency of network resources.
[0072] To achieve efficient decision making, a coordinating satellite or ground station can be designated as the core node for information collection and task coordination, responsible for managing the unloading decisions of the entire low-orbit satellite network, that is, running the optimization algorithm. Through this centralized coordination mechanism, each satellite can execute local optimization decisions in a distributed manner, thereby improving the overall performance of the system while maintaining flexibility. Specifically, all satellites in the low-orbit satellite network will transmit their initial data distribution and physical parameter information to the coordinator. After receiving this information, the coordinator initiates a multi-round sequential optimization process. The core of the optimization is that the coordinator will calculate the unloading strategy centered on each satellite based on the current network state and determine the size of data unloading. In this process, the coordinator maintains a dynamically updated data distribution record, which is updated according to the unloading data size calculated in each round of optimization. This dynamic updating mechanism ensures that subsequent optimization can fully utilize the results of the previous round, achieving iterative improvement. After the optimization process is completed, the coordinator broadcasts the optimized unloading instructions to all satellites. These instructions include two key contents: first, the data unloading size I ab , that is, the amount of data that needs to be transmitted between each pair of satellites; second, the unloading sequence, indicating the priority and order of data transmission in each separate flow. Only after these instructions are published, the actual data transmission in the satellite network can begin. After receiving the instructions from the coordinator, each satellite can execute the corresponding unloading strategy in a distributed manner. This mode can fully utilize the distributed computing power of the satellite network, while ensuring the global optimality and efficiency of the unloading strategy through centralized coordination. In addition, this method also reduces the computational burden of individual nodes, improving the adaptability and flexibility of the network in complex task scenarios. Through this optimization process, the entire satellite network not only effectively reduces system latency, but also significantly improves training accuracy under resource constraints, ultimately achieving overall improvement in system performance and operational efficiency.
[0073] Further reference Figure 2 , as an implementation of the method shown in the above figures, the disclosure provides some embodiments of a data balancing optimization device based on multi-satellite cooperation, which device embodiments are implemented in the same way as Figure 1The method embodiments shown correspond to the multi-satellite cooperative data balancing optimization apparatus, which can be applied to various electronic devices.
[0074] As shown in Figure 2 The multi-satellite cooperative data balancing optimization apparatus 200 of some embodiments includes a first selection unit 201, a second selection unit 202, a building unit 203, and an unloading unit 204. The first selection unit 201 is configured to select a satellite with the largest local training data set as an optimization satellite. The second selection unit 202 is configured to select a neighboring satellite of the optimization satellite for data unloading according to the data size and the class label difference between the satellites. The building unit 203 is configured to build a constraint objective function between the optimization satellite and the neighboring satellite, and solve the constraint objective function to determine a target data unloading ratio of the optimization satellite. The unloading unit 204 is configured to perform data unloading between the optimization satellite and the neighboring satellite according to the target data unloading ratio.
[0075] It can be understood that the units described in the multi-satellite cooperative data balancing optimization apparatus 200 correspond to the steps in the methods described with reference to Figure 1 Thus, the operations, features, and advantages described above for the methods also apply to the multi-satellite cooperative data balancing optimization apparatus 200 and the units included therein, and are not repeated here.
[0076] Reference is made to Figure 3 which shows a structural schematic diagram 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 impose any limitation on the functions and use range of the embodiments of the present disclosure. As Figure 3 shown, the computer device includes a processor, a memory, and a network interface connected through a system bus, wherein the memory can include a non-volatile storage medium and an internal memory. The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions that, when executed, can cause the processor to perform any one of the multi-satellite cooperative data balancing optimization methods. The processor is used to provide computing and control capabilities to support the operation of the entire computer device. The internal memory provides an environment for the computer program in the non-volatile storage medium, which, when executed by the processor, can cause the processor to perform any one of the multi-satellite cooperative data balancing optimization methods. The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art can understand that Figure 3It should be understood that the structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present disclosure, and does not constitute a limitation on the computer device to which the scheme of the present disclosure is applied. A specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0077] It should be understood that the processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.
[0078] In one embodiment, the processor is configured to run a computer program stored in the memory to perform the following steps: selecting a satellite with the largest local training data set as an optimization satellite; selecting neighboring satellites of the optimization satellite for data offloading according to the data size and label difference between the satellites; establishing a constraint objective function between the optimization satellite and the neighboring satellites, and solving the constraint objective function to determine a target data offloading ratio of the optimization satellite; and performing data offloading between the optimization satellite and the neighboring satellites according to the target data offloading ratio.
[0079] The embodiments of the present disclosure further provide a computer readable storage medium, and the computer readable storage medium stores a computer program. The computer program includes program instructions. When the program instructions are executed, the method can refer to each embodiment of the method for data balancing optimization based on multi-satellite cooperation of the present disclosure.
[0080] The computer readable storage medium can be an internal storage unit of the computer device, such as a hard disk or a memory of the computer device. The computer readable storage medium can also be an external storage device of the computer device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc.
[0081] It has to be understood that, in the present document, the terms "comprising", "including", "containing", or any other any other variant thereof are intended to cover a non-exclusive inclusion, such that processes, methods, articles, or systems that comprise, include or contain a list of elements, do not only include those elements, but can also include other elements not expressly listed or inherent to such processes, methods, articles, or systems. Without further limitation, an element defined by the phrase "comprising a" does not exclude the presence of additional identical elements in the process, method, article, or system that includes the element.
[0082] The above description is merely exemplary of some embodiments of the present disclosure and of the application principles so as to appreciate the inventive aspects all throughout the description. It is to be understood that the scope of the present disclosure is not limited to particular details of the embodiments described herein and that work incorporating only some of the described embodiments falls within the scope of the present disclosure. Furthermore, other possible embodiments of the present disclosure can be derived from the teaching of the present disclosure, which is to be construed in the broadest sense and to encompass all such embodiments falling within the scope of the inventive concept as set out in the appended claims.
Claims
1. A data balance optimization method based on multi-star collaborative operation, characterized in that, include: Select the satellite with the largest local training dataset as the optimized satellite; Based on the data size between satellites and the differences in class labels, neighboring satellites of the optimized satellite are selected for data offloading; Establish a constraint objective function between the optimized satellite and the neighboring satellites, and solve the constraint objective function to determine the target data offloading ratio of the optimized satellite; Data offloading is performed between the optimized satellite and the adjacent satellite according to the target data offloading ratio.
2. The method according to claim 1, characterized in that, The step of selecting neighboring satellites of the optimized satellite for data offloading based on the data size and class label differences between satellites includes: For each satellite that establishes a direct communication connection with the optimized satellite, the following processing steps are performed: Determine the class label differences between this satellite and the optimized satellite; Determine the communication delay between the satellite and the optimized satellite; Based on the differences in each class label and communication latency, the neighboring satellites of the optimized satellite are determined as satellites for data offloading.
3. The method according to claim 2, characterized in that, The step of offloading data between the optimized satellite and the neighboring satellite according to the target data offloading ratio includes: Determine the optimal offload amount for each class label between the optimized satellite and its neighboring satellites; The sum of the optimal uninstallation amounts for each category tag is determined as the total uninstallation amount; In response to determining that the target data offloading ratio is the same as the total offloading amount, the data of the optimized satellite is offloaded to the adjacent satellite according to the optimal offloading amount for each class label.
4. The method according to claim 3, characterized in that, The step of offloading data between the optimized satellite and the neighboring satellite according to the target data offloading ratio includes: In response to determining that the target data unloading ratio is less than the total unloading amount, the optimal unloading amount for each class label is subtracted by an equal number of unloading amounts to obtain the balanced unloading amount for that class label, and based on the balanced unloading amount, the data corresponding to that class label in the optimized satellite is unloaded to the adjacent satellite. In response to determining that the target data offloading ratio is greater than the total offloading amount, the optimal offloading amount for each class label is added to an equal number of offloading amounts to obtain the balanced offloading amount for that class label, and based on the balanced offloading amount, the data corresponding to that class label in the optimized satellite is offloaded to the adjacent satellite.
5. A data equalization optimization device based on multi-satellite collaboration, characterized in that, include: The first selection unit is configured to select the satellite with the largest local training dataset as the optimized satellite; The second selection unit is configured to select neighboring satellites of the optimized satellite for data offloading based on the data size between satellites and the differences in class labels; The establishment unit is configured to establish a constraint objective function between the optimized satellite and the adjacent satellites, and to solve the constraint objective function to determine the target data offloading ratio of the optimized satellite; The offloading unit is configured to offload data between the optimized satellite and the adjacent satellite according to the target data offloading ratio.
6. 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 4.
7. 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 4.