Performance measurement method, device and equipment of satellite-ground collaborative fine tuning system and medium

Abstract

The invention provides a performance measurement method, device and equipment for a satellite-ground coordinated fine tuning system and a medium, and relates to the technical field of satellite computing, and the method comprises the steps: constructing the satellite-ground coordinated fine tuning system which comprises a plurality of satellites and a ground station; the satellite configures a local training module, and the ground station configures a satellite selection module and an aggregation module; the method comprises the following steps: respectively measuring convergence conditions of a satellite-borne model of a satellite-ground coordination fine tuning system under various unstable link conditions based on a TCP protocol and a UDP protocol to obtain a first measurement result used for representing a target protocol adapted to each unstable link condition, and respectively transmitting the first measurement result to a satellite-ground coordination fine tuning system based on various priority transmission strategies to obtain a second measurement result used for representing a satellite-borne model of the satellite-ground coordination fine tuning system; and measuring convergence conditions of a satellite-borne model of the satellite-ground coordination fine tuning system under various unstable link conditions to obtain a second measurement result used for representing a target priority transmission strategy adapted to each unstable link condition under a target protocol adapted to the unstable link condition.

Description

Technical Field

[0001] The present application relates to the field of satellite computing technology, and in particular to a performance measurement method, device, equipment and medium for a satellite-ground collaborative fine-tuning system. Background Art

[0002] With the continuous advancement of satellite technology, an increasing number of satellites are being deployed into low-orbit orbit. These satellites, with their unique architecture and orbital position, specialize in high-precision image acquisition. Furthermore, with the advancement of satellite hardware computing power, an increasing number of intelligent models are being deployed in orbit, significantly enhancing the capabilities of on-orbit services. Although intelligent models based on on-orbit satellites have the potential to enhance on-orbit services, due to limited satellite resources, many satellite-based machine learning tasks still rely on ground-based computing resources for collaboration. Low-orbit satellites transmit data to ground stations via satellite-to-ground links, where model fine-tuning is performed in powerful computing clusters. However, with increasing privacy protection requirements and rising satellite-to-ground link costs, the limitations of this traditional approach are becoming increasingly apparent. Federated learning, due to its advantages in data privacy protection and distributed collaboration, has become a promising solution. Specifically, multiple satellites, acting as users in federated learning, complete local model fine-tuning and then transmit model parameters to a ground station for aggregated updates. This approach alleviates privacy concerns to some extent, but does not completely address the communication overhead associated with satellite-to-ground links.

[0003] More importantly, most current studies assume a stable link and focus on optimizing federated aggregation algorithms or adjusting onboard model fine-tuning algorithms. These studies are helpful in improving the efficiency of satellite-ground collaborative fine-tuning under stable links. However, the instability of satellite-ground links remains a difficult problem that cannot be ignored. Due to the high-speed movement of satellites and the complex space environment in which they are located, satellite-ground links often exhibit high packet loss rates and long delays. Although some studies have explored this issue, most remain at the theoretical analysis stage and have not conducted in-depth quantitative analysis of link conditions and their specific impact on system fine-tuning through experiments, especially the performance disadvantages in high packet loss rates and long delay environments (such as convergence delay multiples, additional communication overhead ratios) and performance differences under different communication protocols.

[0004] Different communication protocols (such as TCP, UDP, and others) can significantly impact system performance in a satellite-ground collaborative fine-tuning system. Specifically, the choice of protocol directly affects data transmission latency, packet loss rate, and network bandwidth utilization, which in turn affects the efficiency of model fine-tuning and the final training results. In environments with high packet loss rates and long delays, the UDP protocol can effectively reduce communication overhead and improve transmission efficiency due to its low latency. While the TCP protocol has advantages in data reliability, its retransmission mechanism and congestion control can significantly reduce efficiency in unstable link environments. Therefore, choosing the right protocol is crucial to the overall performance of the satellite-ground collaborative fine-tuning system.

[0005] Therefore, related technologies generally fail to systematically analyze the performance of different communication protocols (such as TCP and UDP) under varying unstable satellite-ground link environments, particularly in scenarios with high packet loss and latency. The choice of communication protocol directly impacts data transmission efficiency and model fine-tuning results, but existing research lacks quantitative data to support this, hindering the selection of appropriate communication protocols under varying unstable link conditions. Furthermore, many solutions are designed based on idealized link environments and fail to consider the volatility of satellite-ground links, such as dynamic changes in packet loss and communication interruptions. This lack of adaptability to varying link fluctuations makes it difficult to effectively address performance fluctuations in satellite-ground coordinated fine-tuning systems in practical deployments. Furthermore, balancing communication overhead with model accuracy remains a challenge in practical applications. Related technologies fail to address how to dynamically select adaptive parameter transmission strategies under varying unstable link conditions to maximize communication efficiency. Summary of the Invention

[0006] In view of this, embodiments of the present application provide a performance measurement method, apparatus, device, and medium for a satellite-ground coordinated fine-tuning system to overcome the above-mentioned problems or at least partially solve the above-mentioned problems.

[0007] A first aspect of an embodiment of the present application provides a performance measurement method for a satellite-ground coordinated fine-tuning system, the method comprising: A satellite-ground coordinated fine-tuning system is constructed, comprising multiple satellites and a ground station. The satellites are configured with a local training module for performing local training of an onboard model, generating fine-tuning parameters, and transmitting the fine-tuning parameters to the ground station. The ground station is configured with a satellite selection module and an aggregation module. The satellite selection module is configured to select satellites for parameter aggregation, and the aggregation module is configured to aggregate the fine-tuning parameters of the selected satellites for parameter aggregation to generate global model parameters. Measuring, based on the TCP protocol and the UDP protocol, respectively, the convergence of the satellite-onboard model of the satellite-ground coordinated fine-tuning system under multiple unstable link conditions to obtain a first measurement result, where the first measurement result is used to indicate a target protocol adapted for each unstable link condition; According to the first measurement result, the convergence of the satellite-borne model of the satellite-ground coordinated fine-tuning system under multiple unstable link conditions is measured based on multiple priority transmission strategies to obtain a second measurement result; the priority transmission strategy is used to indicate the order of transmitting the fine-tuning parameters, and the second measurement result is used to represent the target priority transmission strategy for adaptation to each unstable link condition under the target protocol adapted to the unstable link condition.

[0008] Optionally, the measuring, based on the TCP protocol and the UDP protocol respectively, the convergence of the satellite-borne model of the satellite-ground coordinated fine-tuning system under multiple unstable link conditions to obtain the first measurement result includes: Using a network simulation tool to simulate various unstable link conditions, including packet loss rate, delay, and network fluctuation; Initialize the onboard model and training rounds; Under different unstable link conditions, TCP and UDP protocols are used to establish communication links between satellites and ground stations. Training the onboard model and recording the convergence of the training, wherein the convergence includes: training time, convergence time, loss value, and accuracy; the convergence time is the time required for the training to reach a preset error threshold; the convergence time is determined based on the loss value of each training round, the current training round, the training round, and the error threshold; According to the convergence of each round of training, the target protocol adapted to each unstable link condition is determined from the TCP protocol and UDP protocol.

[0009] Optionally, the measuring, based on the first measurement result and based on multiple priority transmission strategies respectively, convergence of the satellite-borne model of the satellite-ground coordinated fine-tuning system under multiple unstable link conditions to obtain a second measurement result includes: Using a network simulation tool to simulate various unstable link conditions, including packet loss rate, delay, and network fluctuation; Initialize the onboard model and training rounds; Under each unstable link condition, establishing a communication link between the satellite and the ground station according to a target protocol adapted to the unstable link condition; The onboard model is trained based on multiple priority transmission strategies, and the convergence of the training is recorded, wherein the convergence status includes: training time, convergence time, loss value, and accuracy rate; the convergence time is the time required for the training to reach a preset error threshold; the convergence time is determined based on the loss value of each training round, the current training round, the training round, and the error threshold; According to the convergence of the training, the priority transmission strategy with the highest transmission efficiency among the multiple priority transmission strategies is used as the target priority transmission strategy for adaptation to the unstable link condition.

[0010] Optionally, the multiple priority transmission strategies include: no priority transmission strategy, gradient assignment priority transmission strategy and dynamic priority transmission strategy; The non-priority transmission strategy means that the fine-tuning parameters of each satellite after local training are transmitted in sequence; The gradient assignment priority transmission strategy indicates that the transmission priority of the fine-tuning parameters after local training of each satellite is determined according to the gradient amplitude of the fine-tuning parameters; The dynamic priority transmission strategy means that the transmission priority of the fine-tuning parameters after local training of each satellite is dynamically determined according to the contribution of each fine-tuning parameter to the convergence speed in each round of training.

[0011] Optionally, the method further includes: When the target protocol is UDP, the onboard model is trained based on the UDP protocol, and the convergence of the training is recorded, wherein the convergence status includes: training time, convergence time, loss value, and accuracy rate; the convergence time is the time required for the training to reach a preset error threshold; the convergence time is determined based on the loss value of each training round, the current training round, the training round, and the error threshold; Based on whether the training convergence reaches the expected target convergence, it is determined whether the satellite-ground collaborative fine-tuning system should adopt a retransmission mechanism in actual application; When the convergence of the training reaches the desired first target convergence, determining not to adopt the retransmission mechanism in the satellite-ground coordinated fine-tuning system; When the convergence of the training reaches the desired second target convergence, determining to adopt the retransmission mechanism on the satellite-ground coordinated fine-tuning system; Among them, the first target convergence situation is better than the second target convergence situation; the retransmission mechanism indicates that after the satellite transmits the fine-tuning parameters of the screened satellites participating in parameter aggregation to the ground station, and when the communication resources of the communication link between the satellite and the ground station are not used up, the fine-tuning parameters of the satellites participating in parameter aggregation that have not been successfully transmitted are retransmitted.

[0012] Optionally, the method further includes: Under the current unstable link condition, a satellite selection module of the ground station selects multiple satellites as target satellites from multiple satellites within the connection window according to preset rules; the preset rules are set based on the current status of each satellite and the quality and relevance of local data collected by the satellites; The multiple target satellites establish communication links with the ground station through a communication module according to a target protocol adapted for the current unstable link condition, and transmit fine-tuning parameters corresponding to each of the multiple target satellites to the ground station according to a target priority transmission strategy adapted for the current unstable link condition; the fine-tuning parameters are obtained by the target satellites performing local training on a locally deployed onboard model of the target satellites through a local training module based on local data; The ground station aggregates the received multiple fine-tuning parameters through an aggregation module to obtain the global model parameters of this round; The ground station transmits the global model parameters to the multiple satellites via a communication link established with the ground station using a target protocol adapted according to the current unstable link condition, so that the multiple satellites update current model parameters of locally deployed onboard models according to the global model parameters, and reselects new target satellites from the multiple satellites through the satellite selection module to participate in the next round of local training until the onboard model training is completed; Among them, when the satellite-borne model is used to realize the disaster navigation function, the local data is the disaster information data collected by the satellite; when the satellite-borne model is used to realize the earth observation function, the local data is the earth observation data collected by the satellite; when the satellite-borne model is used to realize the climate monitoring function, the local data is the climate monitoring data collected by the satellite.

[0013] Optionally, the ground station aggregates the received multiple fine-tuning parameters through an aggregation module to obtain global model parameters for this round, including: When the fine-tuning parameters of any satellite among the selected satellites participating in parameter aggregation fail to be transmitted, the fine-tuning parameters of multiple satellites that have successfully transmitted the fine-tuning parameters among the selected satellites participating in parameter aggregation and the fine-tuning parameters of the target satellite that failed to transmit in the previous round are aggregated to obtain the global model parameters of this round; When the fine-tuning parameters of all the selected satellites participating in the parameter aggregation are successfully transmitted, the fine-tuning parameters of all the selected satellites participating in the parameter aggregation are aggregated to obtain the global model parameters of this round.

[0014] A second aspect of an embodiment of the present application provides a performance measurement device for a satellite-ground coordinated fine-tuning system, the device comprising: A construction module is configured to construct a satellite-ground coordinated fine-tuning system, the satellite-ground coordinated fine-tuning system comprising multiple satellites and ground stations; the satellites are configured with local training modules, configured to perform local training of onboard models, generate fine-tuning parameters, and transmit them to the ground stations; the ground stations are configured with satellite selection modules and aggregation modules, the satellite selection module is configured to screen satellites for parameter aggregation, and the aggregation module is configured to aggregate the fine-tuning parameters of the screened satellites for parameter aggregation to generate global model parameters; A first measurement module is configured to measure, based on the TCP protocol and the UDP protocol, the convergence of the satellite-onboard model of the satellite-ground coordinated fine-tuning system under multiple unstable link conditions, to obtain a first measurement result, where the first measurement result is used to indicate a target protocol adapted for each unstable link condition; The second measurement module is used to measure the convergence of the satellite-borne model of the satellite-ground coordinated fine-tuning system under multiple unstable link conditions based on multiple priority transmission strategies according to the first measurement result, to obtain a second measurement result; the priority transmission strategy is used to indicate the order of transmitting the fine-tuning parameters, and the second measurement result is used to represent the target priority transmission strategy for adaptation to each unstable link condition under the target protocol adapted to the unstable link condition.

[0015] According to a third aspect of an embodiment of the present application, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the method described in the first aspect.

[0016] According to a fourth aspect of the embodiments of the present application, a computer-readable storage medium is provided, on which a computer program is stored, wherein when the computer program is executed by a processor, the method described in the first aspect is implemented.

[0017] Beneficial effects of this application: The embodiments of the present application provide a performance measurement method, device, equipment and medium for a satellite-ground collaborative fine-tuning system. Through the technical solution of the present application, by constructing a satellite-ground collaborative fine-tuning system, the collaborative working mode of the satellite and the ground station is clarified, providing an infrastructure for subsequent performance measurement. Secondly, by measuring based on the TCP and UDP protocols respectively, the performance differences of different protocols under unstable link conditions are quantitatively analyzed, providing data support for selecting appropriate transmission protocols. Finally, by introducing multiple priority transmission strategies for measurement, the data transmission process of the fine-tuning parameters is further optimized, the communication overhead is reduced, the communication efficiency is improved, and the convergence speed and accuracy of the model are improved, so that the system can achieve efficient and stable model fine-tuning under complex and changeable satellite-ground link conditions, significantly improving the overall performance of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings that constitute a part of this application are used to provide further understanding of this application. The illustrative embodiments of this application and their descriptions are used to explain this application and do not constitute improper limitations on this application.

[0019] In order to more clearly illustrate the technical solution of the present application, the following is a brief introduction to the drawings required for the description of the present application. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0020] Figure 1 This is a flow chart of a performance measurement method for a satellite-ground coordinated fine-tuning system provided in one embodiment of the present application; Figure 2 This is a system configuration diagram of a satellite-ground coordinated fine-tuning system provided in one embodiment of the present application; Figure 3 This is a schematic diagram of the framework of a performance measurement device for a satellite-ground coordinated fine-tuning system provided in one embodiment of the present application. DETAILED DESCRIPTION

[0021] It should be noted that, unless there is any conflict, the embodiments and features in the embodiments of this application can be combined with each other.

[0022] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0023] TCP: Transmission Control Protocol (TCP) is a connection-oriented, reliable, byte stream-based transport layer communication protocol.

[0024] UDP (User Datagram Protocol) uses IP as its underlying protocol and provides a way for applications to send messages to other programs with minimal protocol mechanisms. Its main features are connectionlessness, no guarantee of reliable transmission, and message-oriented nature.

[0025] FL: Federated Learning is a distributed machine learning framework designed to help multiple institutions share knowledge and conduct machine learning modeling while meeting user privacy protection, data security, and government regulations.

[0026] Figure 1 This is a flow chart of a performance measurement method for a satellite-ground coordinated fine-tuning system provided in one embodiment of the present application. Figure 1 An embodiment of the present application provides a performance measurement method for a satellite-ground coordinated fine-tuning system, the method comprising: Step S11: construct a satellite-ground coordinated fine-tuning system, which includes multiple satellites and ground stations; the satellites are equipped with local training modules for performing local training of the satellite-borne model, generating fine-tuning parameters and sending them to the ground stations; the ground stations are equipped with satellite selection modules and aggregation modules, the satellite selection module is used to screen satellites participating in parameter aggregation, and the aggregation module is used to aggregate the fine-tuning parameters of the screened satellites participating in parameter aggregation to generate global model parameters.

[0027] In this embodiment, the satellite-ground coordinated fine-tuning system consists of two parts: a satellite constellation (consisting of multiple satellites) and a ground station. Through the coordinated operation of multiple modules, the system enables efficient data transmission and model fine-tuning between the satellites and the ground station. The core goal of the system is to optimize module design under unstable satellite-ground link conditions to ensure stability and efficiency during the model fine-tuning process.

[0028] In the process of building a satellite-ground coordinated fine-tuning system, we first consider a group of low-Earth orbit satellites and ground stations g for Earth observation. A continuous clock time t and a discrete time index are introduced. i ∈0,1,2,..., each adjacent time index has a time interval of τ. i τ to ( i +1)τ is expressed as [ i ]. When satellite s∈S at any time t∈[ i ] is captured by the ground station g∈G, then the communication link between the satellite s and the ground station g will be established, and the satellite s is defined as the time interval [ i ], the satellite s can use the communication resources during the communication link establishment process. When the satellite s moves out of the signal coverage of the ground station g, the communication link is interrupted, and the satellite s is no longer a satellite connected to the ground station g. The satellite s cannot use the communication resources during the communication link establishment process. The duration of this communication link connection is defined as the connection window, expressed as τd, and τd≤τ. In addition, it is assumed that all ground stations work as a powerful computing cluster without considering the transmission delay between these geographically distributed ground stations. This assumption is reasonable in theory because the structural connections between the various ground stations are always well configured and much faster than the satellite ground links. Therefore, a satellite connected to any ground station can be considered to be connected to a cluster with all ground stations G.

[0029] Figure 2 This is a system diagram of the satellite-ground coordinated fine-tuning system provided in one embodiment of the present application. Figure 2 The satellite-ground coordinated fine-tuning system proposed in this application consists of two parts: a satellite constellation and a ground station. Each satellite is equipped with an onboard model to be trained. The system utilizes multiple modules working together to achieve efficient data transmission and model fine-tuning. The system consists of four modules: a local training module, a communication module (divided into satellite and ground station), a satellite selection module, and an aggregation module.

[0030] Satellite constellation part: consists of two main modules: local training module and communication module (satellite side).

[0031] Local Training Module: Each satellite in the constellation uses its own locally collected data, such as terrain and vegetation data from Earth observations, to perform onboard local model training. Training the onboard model locally on the satellite significantly reduces the need to transmit raw data to the ground station, saving bandwidth and time, and improving fine-tuning efficiency. Fine-tuning parameters (such as gradients and weights) derived from local training are then transmitted to the ground station for aggregation across multiple satellites.

[0032] Communication Module (Satellite): The satellite transmits updated fine-tuning parameters (such as fine-tuned gradients and weights) to the ground station for aggregation through the communication module. The communication module transmits fine-tuning parameters using an adapted communication protocol (such as TCP or UDP), ensuring efficient and stable data transmission between the satellite and the ground station.

[0033] Ground station part: The ground station is mainly composed of a satellite selection module, an aggregation module and a communication module (ground station end).

[0034] Satellite Selection Module: The ground station uses the satellite selection module to select satellites for each round of fine-tuning parameter aggregation based on pre-set rules. These rules can be based on the satellite's current state (such as energy level and computing power) or the quality and relevance of the data collected by the satellite. For example, satellites with richer or higher-quality local data collected in a specific area may be prioritized for fine-tuning parameter aggregation of the onboard model.

[0035] Aggregation Module: After the ground station receives fine-tuning parameters transmitted by multiple satellites, the aggregation module is responsible for aggregating these fine-tuning parameters and generating global model parameters. During the aggregation process, the aggregation module can take into account factors such as data quality and computing power of different satellites, and perform a weighted average of the fine-tuning parameters to generate a more representative global model parameter.

[0036] Communication Module (Ground Station): Because the ground station and satellites require multiple rounds of collaborative fine-tuning, the system also includes a feedback mechanism. After aggregating multiple fine-tuning parameters in each round to obtain global model parameters, these global model parameters are sent to each satellite via the communication module (ground station). Each satellite then updates the parameters of its locally deployed onboard model based on the global model parameters and proceeds to the next round of local training. In this way, the satellite and ground station achieve multiple rounds of federated learning collaboration, ultimately achieving the goal of optimizing the global model (i.e., the onboard model obtained through multiple rounds of collaborative fine-tuning).

[0037] Step S12: Based on the TCP protocol and the UDP protocol, the convergence of the satellite-based model of the satellite-ground coordinated fine-tuning system under multiple unstable link conditions is measured to obtain a first measurement result, where the first measurement result is used to represent the target protocol adapted for each unstable link condition.

[0038] In this embodiment, we consider that using different communication protocols to establish the communication link between the satellite and the ground station will significantly affect the convergence of the onboard model of the satellite-ground coordinated fine-tuning system under various unstable link conditions (such as different packet loss rates, long delays, and network fluctuations). Therefore, in order to determine which protocol is more suitable for system operation under different unstable link conditions, it is necessary to measure the system performance under various unstable link conditions based on the TCP protocol and the UDP protocol.

[0039] Among them, the TCP protocol is a connection-oriented, reliable transport layer protocol, which is characterized by the ability to ensure data integrity and sequence. However, its retransmission mechanism and congestion control may lead to reduced transmission efficiency in environments with high packet loss rates and long delays.

[0040] The UDP protocol is a connectionless protocol that does not guarantee reliability. Its advantages are low latency and high throughput, but in an environment with high packet loss rates, additional mechanisms may be required to ensure data integrity.

[0041] By conducting experiments using these two protocols under different link conditions (such as different combinations of packet loss rates and delays), corresponding measurement results (i.e., first measurement results) are obtained. These first measurement results indicate the target protocol (i.e., UDP or TCP) that is adapted for each unstable link condition. By quantitatively analyzing the system performance of different protocols under unstable link conditions, this provides basic data support for subsequent selection of priority transmission strategies. This allows for subsequent fine-tuning of satellite-ground coordination, allowing the first measurement results to be used to directly select the target protocol adapted for the current unstable link conditions to establish a communication link between the satellite and ground station. This improves system performance, enabling optimal convergence of the onboard model, and enhancing both convergence speed and accuracy.

[0042] Step S13: Based on the first measurement result, the convergence of the satellite-borne model of the satellite-ground coordinated fine-tuning system under multiple unstable link conditions is measured based on multiple priority transmission strategies to obtain a second measurement result; the priority transmission strategy is used to indicate the order of transmitting the fine-tuning parameters, and the second measurement result is used to represent the target priority transmission strategy for adaptation to each unstable link condition under the target protocol adapted to the unstable link condition.

[0043] In this embodiment, after determining the communication protocols adapted for different unstable link conditions, in order to further optimize the system performance, this embodiment takes into account the limited communication resources of the satellite-to-ground link. Therefore, it is necessary to further determine how to efficiently transmit the fine-tuning parameters of the satellites participating in the aggregation. Based on this, we believe that when the communication link between the satellite and the ground is unstable, the system performance can be significantly improved by reasonably arranging the transmission order of the fine-tuning parameters. Higher system performance can also lead to better convergence speed and model accuracy of the onboard model.

[0044] Therefore, in this step, multiple pre-set priority transmission strategies are used to measure the convergence of the satellite-ground coordinated fine-tuning system under multiple unstable link conditions to obtain a second measurement result.

[0045] The priority transmission strategy refers to determining the order in which fine-tuning parameters are transmitted to each satellite via the satellite-to-ground communication link, based on factors such as the importance of the fine-tuning parameters or their contribution to model convergence. For example, the priority transmission strategy can be defined based on criteria such as the gradient amplitude of the parameter and the parameter sensitivity. In this way, the data transmission process can be optimized, unnecessary communication overhead can be reduced, and the convergence speed and model accuracy of the onboard model can be improved. The second measurement result is used to indicate the priority transmission strategy that is adapted to each unstable link condition among multiple priority transmission strategies when establishing a satellite-to-ground communication link using the target protocol adapted for that unstable link condition, i.e., the target priority transmission strategy. By verifying the effectiveness of different priority transmission strategies under different link conditions, data support is provided for the system to select the optimal transmission strategy in practical applications.

[0046] Combining the two measurement processes mentioned above, we can obtain the target protocol and target priority transmission strategy for each unstable link condition. In the subsequent actual application of the system, we can select the target protocol and target priority transmission strategy for the current unstable link condition based on the measurement results to conduct satellite-ground collaborative fine-tuning of the onboard model, so as to maximize the performance of the system and improve the convergence speed and model accuracy of the onboard model.

[0047] Through the technical solutions of the above-mentioned embodiments, the performance optimization of the satellite-ground collaborative fine-tuning system under unstable link conditions is achieved. First, by constructing the satellite-ground collaborative fine-tuning system, the collaborative working mode of the satellite and the ground station is clarified, providing an infrastructure for subsequent performance measurement. Secondly, by measuring based on the TCP and UDP protocols respectively, the performance differences of different protocols under unstable link conditions are quantitatively analyzed, providing data support for the selection of appropriate transmission protocols. Finally, by introducing multiple priority transmission strategies for measurement, the data transmission process of the fine-tuning parameters is further optimized, the communication overhead is reduced, and the convergence speed and accuracy of the model are improved, so that the system can achieve efficient and stable model fine-tuning under complex and changeable satellite-ground link conditions, significantly improving the overall performance of the system.

[0048] In combination with the technical solutions of the above embodiments, an embodiment of the present application further provides another performance measurement method for a satellite-ground coordinated fine-tuning system. In this method, the step S12 of "measuring the convergence of the satellite-borne model of the satellite-ground coordinated fine-tuning system under multiple unstable link conditions based on the TCP protocol and the UDP protocol to obtain a first measurement result" specifically includes steps S12-1 to S12-5: Step S12-1: Use a network simulation tool to simulate various unstable link conditions, including packet loss rate, delay, and network fluctuation.

[0049] In this embodiment, in the satellite-ground coordinated fine-tuning system, the communication link between the satellite and the ground station is often unstable due to factors such as the satellite's high-speed motion, atmospheric interference, and the complexity of the space environment. This instability is primarily manifested in high packet loss rates, long delays, and intermittent link interruptions. To accurately evaluate the performance of different communication protocols under these complex, unstable link conditions, a network simulation tool is used to simulate various unstable link conditions. Different possible instabilities in the satellite-ground communication link are simulated, including packet loss rates (e.g., 0%, 5%, 10%, 20%) and delays (e.g., 100ms, 300ms, 500ms), and varying network volatility. Training experiments are conducted for each condition. Using network simulation tools (such as NS3), the parameters of these conditions can be precisely set and adjusted, providing a controllable and diverse testing environment for subsequent experiments. This step is the foundation of the entire performance measurement method, ensuring the reliability and repeatability of the measurement results.

[0050] Step S12-2: Initialize the onboard model and training rounds.

[0051] In this embodiment, the onboard model needs to be initialized before performance measurements are performed. An onboard model refers to a machine learning model deployed on the satellite, used for local training and fine-tuning of data collected by the satellite. The initialization process includes setting the model's initial parameters, selecting an appropriate training algorithm, and determining the number of training rounds. Training rounds refer to the number of iterative optimizations performed on the model during training, typically determined by the model's complexity and the size of the dataset. Initializing the onboard model and training rounds ensures that experiments are conducted under identical starting conditions, enabling comparable performance comparisons across different protocols and under varying unstable link conditions.

[0052] For example, the onboard model can be a classic deep learning model (such as a convolutional neural network (CNN)), which is suitable for tasks such as image recognition or data analysis. The number of training rounds is set, such as 100 rounds, and the loss value and accuracy of the onboard model in each round are recorded.

[0053] Step S12-3: Under different unstable link conditions, the TCP protocol and the UDP protocol are used to establish a communication link between the satellite and the ground station.

[0054] In this example, TCP and UDP protocols were used to establish communication links between the satellite and the ground station under simulated unstable link conditions. By using both protocols to establish communication links under the same unstable link conditions, the system performance of the two protocols under different unstable link conditions can be compared and analyzed, providing data support for selecting the appropriate target protocol in subsequent practical applications.

[0055] Step S12-4: train the onboard model and record the convergence of the training, where the convergence includes: training time, convergence time, loss value, and accuracy. The convergence time is the time required for the training to reach a preset error threshold. The convergence time is determined based on the loss value of each round of training, the current training round, the training round, and the error threshold.

[0056] In this embodiment, after establishing a communication link, the onboard model is collaboratively fine-tuned using different protocols under various unstable link conditions. This involves local satellite training and parameter aggregation at the ground station. Convergence is recorded during the fine-tuning process. Convergence metrics include training time, convergence time, loss (e.g., mean squared error (MSE)), and accuracy. Training time represents the total training time, estimated by using the time per round (in seconds). Convergence time refers to the time required for training to reach a preset error threshold (e.g., MSE < 0.01). This time is determined based on the loss value per round, the current training round, the total number of training rounds, and the error threshold. Loss measures the model's error during training, while accuracy reflects the model's ability to classify or predict data. By recording these key metrics, the performance of different protocols under different link conditions can be comprehensively evaluated, particularly their impact on model convergence speed and accuracy.

[0057] For example, the formula for calculating the convergence time is: ;in, Represents the loss of each round of training, t is the current training round, N is the total rounds, is the error threshold.

[0058] For example, the training time is calculated as: ;in, represents the training time of the tth round.

[0059] Step S12-5: According to the convergence of each round of training, a target protocol adapted to each unstable link condition is determined from the TCP protocol and the UDP protocol.

[0060] In this embodiment, after completing the training and performance recording of different protocols under different link conditions, based on the convergence of each round of training, a protocol adapted to each unstable link condition is determined from the TCP protocol and the UDP protocol as the target protocol.

[0061] For example, if, under certain link conditions, the UDP protocol converges significantly faster than the TCP protocol, while also exhibiting superior performance in terms of loss and accuracy, then UDP can be considered the target protocol for that link condition. This step provides data support for fine-tuning the communication protocol selection for the satellite-ground collaborative system under different link conditions, thereby optimizing the system's overall performance.

[0062] Through the technical solutions of the above embodiments, a network simulation tool is used to simulate a variety of unstable link conditions, providing a controllable and diverse test environment for the experiment, ensuring the reliability and repeatability of the experimental results. Secondly, by initializing the onboard model and training rounds, a unified benchmark is provided for subsequent training and performance evaluation. Then, under different unstable link conditions, the TCP protocol and the UDP protocol are used to establish communication links, and the onboard model is trained, and the convergence of the training is recorded. This process provides data support for the selection of appropriate protocols by comparing and analyzing the performance of different protocols under different link conditions. Finally, based on the convergence of each round of training, the target protocol adapted to each unstable link condition is determined.

[0063] In combination with the technical solutions of the above embodiments, an embodiment of the present application further provides another performance measurement method for a satellite-ground coordinated fine-tuning system. In this method, the step S13 of "measuring, based on the first measurement result, the convergence of the satellite-borne model of the satellite-ground coordinated fine-tuning system under multiple unstable link conditions based on multiple priority transmission strategies to obtain a second measurement result" specifically includes steps S13-1 to S13-5: Step S13-1: Use a network simulation tool to simulate various unstable link conditions, including packet loss rate, delay, and network fluctuation. In this embodiment, this step refers to the content of step S12-1 above and will not be repeated here.

[0064] Step S13-2: Initialize the onboard model and training rounds. In this embodiment, this step refers to the content of step S12-2 above and will not be repeated here.

[0065] Step S13-3: Under each unstable link condition, a communication link is established between the satellite and the ground station according to the target protocol adapted to the unstable link condition. In this embodiment, this step is similar to the above-mentioned step S12-3 and will not be described in detail here.

[0066] Step S13-4: Train the onboard model based on multiple priority transmission strategies, and record the convergence of the training, which includes: training time, convergence time, loss value, and accuracy; the convergence time is the time required for training to reach a preset error threshold; the convergence time is determined based on the loss value of each round of training, the current training round, the training round, and the error threshold.

[0067] In this embodiment, after establishing a communication link based on the target protocol, the onboard model is trained using various priority transmission strategies, and the training convergence status is recorded. Priority transmission strategies prioritize the transmission of fine-tuning parameters based on their importance or contribution to model convergence. Furthermore, the difference between the error values ​​during training and verification is used to measure the impact of different priority transmission strategies on convergence speed and accuracy.

[0068] The difference between the error value of the training process (obtained using the training data set in the local data) and the error value of the verification process (obtained using the verification data set in the local data) is used to measure the impact of different priority transmission strategies on convergence speed and accuracy. The specific process is as follows: ; Among them, Convergence difference means difference, Represents the error value of the training process, Represents the error value of the verification process.

[0069] When the difference between the training error and the validation error is small and both tend to be stable, the current onboard model has good generalization and convergence performance. When the difference between the training error and the validation error is large, this may indicate that the model performs well on the training data but poorly on new, unseen data (i.e., the validation dataset), indicating that the onboard model is overfitting. For more information on convergence, refer to the above step S12-4 and will not be repeated here.

[0070] Step S13-5: Based on the convergence of the training, the priority transmission strategy with the highest transmission efficiency among the multiple priority transmission strategies is used as the target priority transmission strategy for adaptation to the unstable link condition.

[0071] In this embodiment, after completing model fine-tuning and recording measurement results for multiple priority transmission strategies under different link conditions, under the same unstable link condition, the priority transmission strategy with the highest transmission efficiency is selected from the multiple priority transmission strategies based on the convergence of each training round and serves as the target priority transmission strategy for adaptation to the unstable link condition. For example, if, under a certain unstable link condition, the gradient-assigned priority transmission strategy results in the shortest model convergence time and the highest accuracy, then this strategy can be considered the target priority transmission strategy for that link condition. This step provides data support for the satellite-ground collaborative fine-tuning system's selection of priority transmission strategies under different link conditions, thereby further optimizing the system's overall performance.

[0072] For example, the calculation formula for transmission efficiency is: ; Where M is the number of fine-tuning parameters transmitted, is the size of the i-th fine-tuning parameter, is the total transmission time.

[0073] In combination with the technical solutions of the above embodiments, an embodiment of the present application further provides another performance measurement method for a satellite-ground coordinated fine-tuning system, in which the multiple priority transmission strategies include: a no-priority transmission strategy, a gradient-assigned priority transmission strategy, and a dynamic priority transmission strategy; The non-priority transmission strategy means that the fine-tuning parameters of each satellite after local training are transmitted in sequence.

[0074] Specifically, in the non-prioritized transmission strategy, the fine-tuning parameters of each satellite after local training are transmitted sequentially. In this strategy, all fine-tuning parameters are considered equally important, regardless of their contribution to model convergence.

[0075] The gradient assignment priority transmission strategy indicates that the transmission priority of the fine-tuning parameters after local training of each satellite is determined according to the gradient amplitude of the fine-tuning parameters.

[0076] Specifically, in the gradient assignment priority transmission strategy, the transmission priority of fine-tuning parameters after local training on each satellite is determined by the magnitude of the fine-tuning parameter's gradient. Parameters with larger gradient magnitudes typically have a greater impact on the model's convergence speed. Therefore, prioritizing the transmission of these parameters can accelerate model convergence, effectively utilizing limited communication resources. Prioritizing the transmission of parameters that are more important to model convergence improves training efficiency.

[0077] The dynamic priority transmission strategy means that the transmission priority of the fine-tuning parameters after local training of each satellite is dynamically determined according to the contribution of each fine-tuning parameter to the convergence speed in each round of training.

[0078] Specifically, in the dynamic priority transmission strategy, the transmission priority of each satellite's locally trained fine-tuned parameters is dynamically determined based on their contribution to convergence speed during each round of training. Unlike the gradient-based priority transmission strategy, the dynamic priority transmission strategy dynamically adjusts the parameter transmission order based on the current training status and link conditions.

[0079] For example, in some training rounds, certain fine-tuning parameters may contribute more to the convergence speed, so these fine-tuning parameters are transmitted first. In other training rounds, other fine-tuning parameters are more important, so they are transmitted first. By dynamically adjusting the transmission priority, high communication efficiency and model convergence speed can be maintained under different link conditions.

[0080] Through the technical solutions of the above embodiments, a no-priority transmission strategy, a gradient-assigned priority transmission strategy, and a dynamic priority transmission strategy are introduced, which can flexibly select the most appropriate priority transmission strategy according to different link conditions and training requirements.

[0081] In combination with the technical solutions of the above embodiments, an embodiment of the present application further provides another performance measurement method for a satellite-ground coordinated fine-tuning system, wherein the method further includes steps S21 to S24: Step S21: When the target protocol is the UDP protocol, the onboard model is trained based on the UDP protocol, and the convergence of the training is recorded. The convergence status includes: training time, convergence time, loss value and accuracy rate; the convergence time is the time required for the training to reach a preset error threshold; the convergence time is determined based on the loss value of each round of training, the current training round, the training round, and the error threshold.

[0082] In this embodiment, in the satellite-ground coordinated fine-tuning system, when the target protocol is UDP, it is necessary to use UDP to train the satellite model and record the convergence status during the training process in detail. The convergence status is referred to the above content and will not be repeated here. Step S22 , judging whether the satellite-ground coordinated fine-tuning system adopts a retransmission mechanism in actual application according to whether the convergence of the training reaches the expected target convergence.

[0083] In this embodiment, after completing UDP-based training and recording convergence, it is necessary to determine whether the training convergence reaches the desired target convergence level and whether the satellite-ground coordinated fine-tuning system should adopt a retransmission mechanism in actual application. The introduction of a retransmission mechanism can improve the data transmission reliability of the fine-tuning parameters, ensuring that the fine-tuning parameters of the onboard model reach the ground station completely and accurately, thereby improving the convergence speed and model accuracy of the onboard model.

[0084] Step S23: When the convergence of the training reaches the desired first target convergence, it is determined that the retransmission mechanism is not adopted in the satellite-ground coordinated fine-tuning system.

[0085] In this embodiment, when the convergence condition is good, that is, the first target convergence condition is reached, it means that the current performance of the satellite-borne model is good. At this time, the fine-tuning parameters of the failed transmission have a relatively small impact on the satellite-borne model. Therefore, it is possible to choose not to retransmit these fine-tuning parameters of the failed transmission, but to prioritize the subsequent satellite-ground coordinated fine-tuning process of the satellite-borne model for the purpose of reducing communication resources.

[0086] Step S24: When the convergence of the training reaches the desired second target convergence, it is determined to adopt the retransmission mechanism in the satellite-ground coordinated fine-tuning system.

[0087] In this embodiment, when the convergence is relatively insufficient, that is, the second target convergence is reached, it indicates that the current performance of the satellite-borne model is poor. At this time, the fine-tuning parameters that failed to be transmitted have a relatively large impact on the satellite-borne model. Therefore, it is necessary to retransmit these fine-tuning parameters that failed to be transmitted to ensure the quality of the subsequent satellite-ground collaborative fine-tuning process of the satellite-borne model.

[0088] Among them, the first target convergence situation is better than the second target convergence situation; the retransmission mechanism indicates that after the satellite transmits the fine-tuning parameters of the screened satellites participating in parameter aggregation to the ground station, and when the communication resources of the communication link between the satellite and the ground station are not used up, the fine-tuning parameters of the satellites participating in parameter aggregation that have not been successfully transmitted are retransmitted.

[0089] Suppose that during a data transmission, the satellite transmits 100 fine-tuning parameters to the ground station, but 10 parameters fail to arrive. If communication resources are not exhausted, the satellite retransmits these 10 unsuccessful parameters to the ground station. This ensures that all parameters reach the ground station intact, thereby improving the model's convergence speed and accuracy.

[0090] For example, assuming the error threshold is 0.01, during training using the UDP protocol, we record the following data: In the 10th round of training, the loss value is 0.02 and the accuracy is 95%; In the 15th round of training, the loss value is 0.009 and the accuracy is 96%; At the 20th round of training, the loss value was 0.005 and the accuracy was 97%; At the 30th round of training, the loss value was 0.003 and the accuracy was 98%.

[0091] Based on this data, we can calculate that the loss value is below 0.01 for the first time during the 15th round of training, so the convergence time is the time it takes to complete the 15th round of training. In addition, we define two target convergence conditions. During each round of training, we can decide whether to use the retransmission mechanism based on the above recorded convergence conditions. First objective convergence: loss value is less than 0.004, accuracy is higher than 97%; Second objective convergence: loss value is less than 0.01 and accuracy is higher than 95%.

[0092] In the 10th round of training, the loss value is 0.02 and the accuracy is 95%, which meets the second target convergence. In this case, higher performance requirements are required, and a retransmission mechanism is needed to further improve the reliability of the system and the convergence quality of the model.

[0093] In the 20th round of training, the loss value was 0.005 and the accuracy was 97%, meeting the first target convergence condition. At this time, there was no need to adopt a retransmission mechanism. In other words, under the current link conditions and UDP protocol, the performance of the satellite-borne model has met the basic requirements, and no additional retransmission mechanism is needed to improve reliability.

[0094] The technical solutions of the above-described embodiments enable the dynamic decision to adopt a retransmission mechanism based on the convergence of training, even when using the UDP protocol, thereby achieving a balance between communication efficiency and reliability. This dynamic introduction of a retransmission mechanism improves data transmission reliability without significantly increasing communication overhead, ensuring that onboard model parameters reach the ground station completely and accurately, thereby improving the model's convergence speed and accuracy. Furthermore, by dynamically selecting whether to adopt a retransmission mechanism based on different convergence conditions, the system can flexibly adjust transmission strategies under varying link conditions and performance requirements, optimizing overall system performance.

[0095] In combination with the technical solutions of the above embodiments, an embodiment of the present application further provides another performance measurement method for a satellite-ground coordinated fine-tuning system, wherein the method further includes steps S31 to S34: Step S31: Under the current unstable link condition, the satellite selection module of the ground station selects multiple satellites as target satellites from multiple satellites within the connection window according to preset rules; the preset rules are set based on the current status of each satellite and the quality and relevance of the local data collected by the satellite.

[0096] In this embodiment, after the aforementioned measurement process is completed, the actual onboard model fine-tuning process can be performed via the satellite-ground coordinated fine-tuning system. To ensure that the satellites participating in model fine-tuning have high-quality data and sufficient resources, the ground station's satellite selection module selects multiple satellites as target satellites from the multiple satellites within the connection window. This selection process is based on preset rules that comprehensively consider each satellite's current state (such as energy level and computing power) as well as the quality and relevance of the local data collected by the satellite. Specifically, the satellite's current state reflects its ability to participate in fine-tuning. For example, satellites with higher energy levels can more stably perform computing tasks. The quality and relevance of local data directly influence the effectiveness of onboard model fine-tuning. High-quality, mission-relevant data can improve onboard model performance. In this way, the satellite selection module can select the most suitable satellites for the current fine-tuning task, laying the foundation for subsequent efficient data transmission and model optimization.

[0097] In step S32, the multiple target satellites establish communication links with the ground station through the communication module according to the target protocol adapted to the current unstable link conditions, and transmit the fine-tuning parameters corresponding to the multiple target satellites to the ground station according to the target priority transmission strategy adapted to the current unstable link conditions; the fine-tuning parameters are obtained by the target satellite through local training of the locally deployed onboard model of the target satellite based on local data through the local training module.

[0098] In this embodiment, after selecting target satellites, these satellites establish a communication link with the ground station through the communication module, using the first measurement result to determine a target protocol suitable for the current unstable link conditions. Furthermore, based on the second measurement result, the target priority transmission strategy suitable for the current unstable link conditions is determined, and the corresponding fine-tuning parameters are transmitted to the ground station. The fine-tuning parameters are obtained by the target satellite using a local training module to locally train a locally deployed onboard model based on local data. This maximizes communication efficiency even under unstable link conditions, reduces the impact of data transmission on model convergence, and ensures the efficiency and stability of the model fine-tuning process.

[0099] In step S33, the ground station aggregates the received multiple fine-tuning parameters through an aggregation module to obtain the global model parameters of this round.

[0100] In this embodiment, after the ground station receives fine-tuning parameters transmitted from multiple target satellites, it aggregates the received parameters through an aggregation module to generate the global model parameters for this round. The aggregation process can integrate local model updates from different satellites into a global model, thereby achieving collaborative optimization of model parameters. The aggregation module can perform a weighted average of the fine-tuning parameters of different satellites based on factors such as the data quality and computing power of each satellite, thereby improving the generalization capability of the onboard model and ensuring the robustness of the onboard model on data from different satellites. In this way, the ground station can effectively integrate dispersed satellite resources, enhance the collaborative working ability of the entire system, and provide optimized global model parameters for subsequent model updates and the next round of fine-tuning.

[0101] In step S34, the ground station transmits the global model parameters to the multiple satellites through the communication link established with the ground station using the target protocol adapted according to the current unstable link conditions, so that the multiple satellites update the current model parameters of the locally deployed onboard model according to the global model parameters, and reselect new target satellites from the multiple satellites through the satellite selection module to participate in the next round of local training until the onboard model training is completed.

[0102] Among them, when the satellite-borne model is used to realize the disaster navigation function, the local data is the disaster information data collected by the satellite; when the satellite-borne model is used to realize the earth observation function, the local data is the earth observation data collected by the satellite; when the satellite-borne model is used to realize the climate monitoring function, the local data is the climate monitoring data collected by the satellite.

[0103] In this embodiment, the ground station transmits the aggregated global model parameters to multiple satellites via a communication link. The satellites then update the current model parameters of their locally deployed onboard models based on the global model parameters and, through the satellite selection module, reselect new target satellites for the next round of local training until onboard model training is complete. By transmitting the global model parameters back to the satellites, they can perform local updates based on the latest global model, thereby better adapting to the global optimization objectives in the next round of fine-tuning. Simultaneously, the satellite selection module reselects target satellites based on the updated model performance and satellite status, ensuring that each round of fine-tuning is performed with the optimal satellite combination. This iterative process not only improves the convergence speed of the onboard model but also enhances the system's adaptability, enabling efficient and stable model training under complex, changing, and unstable link conditions.

[0104] In addition, depending on the specific application function of the spaceborne model (such as disaster navigation, Earth observation or climate monitoring), the type of local data will also vary, thereby affecting the training direction and performance optimization goals of the model, resulting in a personalized spaceborne model.

[0105] For example, for disaster navigation, local data includes disaster information collected by satellites, such as images of fire areas and flood monitoring data. For Earth observation, local data includes Earth observation data collected by satellites, such as topographic maps and vegetation coverage. For climate monitoring, local data includes climate monitoring data collected by satellites, such as atmospheric temperature and humidity. Different local data directly influences the satellite's local training process and the generation of fine-tuning parameters, thus affecting the performance of the final onboard model.

[0106] Through the technical solutions of the above-mentioned embodiments, efficient operation of the satellite-ground coordinated fine-tuning system under unstable link conditions is achieved. First, by optimizing the satellite selection process, it is ensured that the satellites participating in the fine-tuning have high-quality data and sufficient resources, thereby improving training efficiency. Second, through a flexible communication strategy, the target protocol and target priority transmission strategy are dynamically selected based on the link conditions and the first and second measurement results to ensure efficient and reliable data transmission. Next, the global model parameters are generated through the aggregation module and transmitted back to the satellite, achieving multiple rounds of iterative optimization and further improving model performance. Finally, through cyclic iteration and dynamic satellite selection, the system can adaptively optimize the model training process to ensure efficient and stable operation in complex and changing environments.

[0107] In combination with the technical solutions of the above embodiments, an embodiment of the present application further provides another performance measurement method for a satellite-ground coordinated fine-tuning system. In this method, the step S33 of "the ground station aggregates the received multiple fine-tuning parameters through the aggregation module to obtain the global model parameters of this round" specifically includes steps S33-1 to S33-2: Step S33-1: When the fine-tuning parameter transmission of any satellite among the selected satellites participating in parameter aggregation fails, the fine-tuning parameters of multiple satellites that have successfully transmitted the fine-tuning parameters among the selected satellites participating in parameter aggregation and the fine-tuning parameters of the target satellite that failed to transmit the fine-tuning parameters in the previous round are aggregated to obtain the global model parameters of this round.

[0108] In this embodiment, due to link instability, the transmission of fine-tuning parameters for some satellites may fail in the satellite-ground coordinated fine-tuning system. To ensure the accuracy and integrity of the global model parameters, the ground station's aggregation module needs to process the received fine-tuning parameters. Specifically, if the transmission of fine-tuning parameters for any satellite selected for parameter aggregation fails, the aggregation module aggregates the fine-tuning parameters of multiple satellites that successfully transmitted fine-tuning parameters from the selected satellites participating in parameter aggregation with the fine-tuning parameters of the target satellite that failed to transmit fine-tuning parameters in the previous round to obtain the global model parameters for the current round.

[0109] Specifically, in actual satellite-to-ground communications, fine-tuning parameters for some satellites may not be successfully transmitted to the ground station due to issues such as high packet loss, long latency, or intermittent connectivity. To address this, the aggregation module retains the fine-tuning parameters successfully transmitted from the previous round and aggregates them with the fine-tuning parameters successfully transmitted in the current round. The aggregation module then applies a weighted average or other aggregation algorithm to the received fine-tuning parameters to generate the global model parameters for this round. This approach ensures that even in the event of partial data loss, the global model parameters still reflect the updates from all participating satellites, thereby maintaining model stability and convergence.

[0110] For example, suppose that in a training round, three satellites (satellites A, B, and C) participate in parameter aggregation. The fine-tuning parameters of satellites A and B are successfully transmitted to the ground station, but the fine-tuning parameters of satellite C fail to be transmitted. In this case, the aggregation module aggregates the fine-tuning parameters of satellites A and B with the fine-tuning parameters of satellite C in the previous round to generate the global model parameters for this round.

[0111] Step S33-2: When the fine-tuning parameters of all the selected satellites participating in parameter aggregation are successfully transmitted, the fine-tuning parameters of all the selected satellites participating in parameter aggregation are aggregated to obtain the global model parameters of this round.

[0112] In this embodiment, when the fine-tuning parameters of all the selected satellites participating in parameter aggregation are successfully transmitted, the aggregation module aggregates the fine-tuning parameters of all the selected satellites participating in parameter aggregation to obtain the global model parameters of this round.

[0113] Ideally, the fine-tuning parameters of all participating satellites are successfully transmitted to the ground station. In this case, the aggregation module directly aggregates these fine-tuning parameters to generate the global model parameters for this round. The aggregation module can use weighted averaging or other aggregation algorithms to weight the fine-tuning parameters based on factors such as the data quality and computing power of each satellite to generate the global model parameters for this round. This approach ensures that the global model parameters accurately reflect the updates of all participating satellites, thereby improving the speed and accuracy of model convergence.

[0114] For example, suppose that in a training round, three satellites (satellites A, B, and C) participate in parameter aggregation, and the fine-tuning parameters of all satellites are successfully transmitted to the ground station. At this time, the aggregation module aggregates the fine-tuning parameters of satellites A, B, and C to generate the global model parameters for this round.

[0115] The technical solutions of the above-described embodiments ensure the continuity and stability of global model parameters by retaining the fine-tuning parameters from the previous round and aggregating them with the successfully transmitted parameters from the current round in the event of a transmission failure. This mechanism enables the system to maintain normal operation under unstable link conditions, reducing the impact of data loss on model training. If all parameters are successfully transmitted, all fine-tuning parameters are directly aggregated to generate the global model parameters for this round. This ensures that the global model parameters accurately reflect the updates of all participating satellites, thereby improving the model's convergence speed and accuracy. Whether partial transmission failures or complete transmission successes can be effectively handled, this flexibility enables the system to adapt to different link conditions and application scenarios, providing high adaptability.

[0116] As a preferred technical solution, to address fine-tuning parameter transmission failures, network coding can be used to encode multiple fine-tuning parameters. This allows them to be recovered via alternative paths even if they are lost during transmission. This technology can be applied in satellite-to-ground links to mitigate packet loss and reduce transmission latency. By dynamically selecting the optimal coding strategy, the robustness and efficiency of data transmission can be improved to a certain extent.

[0117] As a preferred technical solution, model compression technology can be used to transmit fine-tuning parameters. By compressing the fine-tuning parameters of the onboard model (e.g., through low-rank decomposition, pruning, and quantization), the amount of data transmitted each time is reduced, thereby alleviating the network transmission burden, especially when satellite-to-ground link bandwidth is limited. This solution can be used in conjunction with the UDP or TCP protocols to further improve transmission efficiency and reduce the impact of packet loss on model training.

[0118] As a preferred technical solution, when the target protocol is TCP, high packet loss rates and long delays can be addressed by optimizing TCP's congestion control algorithms. For example, combining TCP's congestion control algorithms (such as BBR and CUBIC) with fast recovery mechanisms can improve TCP's performance when the satellite-to-ground link is unstable, thus enhancing TCP's stability.

[0119] Figure 3 This is a schematic diagram of a performance measurement device for a satellite-ground coordinated fine-tuning system provided in an embodiment of the present application. Figure 3 Based on the same inventive concept, another embodiment of the present application further provides a performance measurement device for a satellite-ground coordinated fine-tuning system, the device comprising: A construction module 11 is configured to construct a satellite-ground coordinated fine-tuning system, the satellite-ground coordinated fine-tuning system comprising multiple satellites and ground stations; the satellites are configured with local training modules for performing local training of onboard models, generating fine-tuning parameters, and transmitting the fine-tuning parameters to the ground stations; the ground stations are configured with satellite selection modules and aggregation modules, the satellite selection module being configured to screen satellites for parameter aggregation, and the aggregation module being configured to aggregate the fine-tuning parameters of the screened satellites for parameter aggregation to generate global model parameters; A first measurement module 12 is configured to measure, based on the TCP protocol and the UDP protocol, the convergence of the satellite-onboard model of the satellite-ground coordinated fine-tuning system under multiple unstable link conditions, to obtain a first measurement result, where the first measurement result is used to indicate a target protocol adapted for each unstable link condition; The second measurement module 13 is used to measure the convergence of the satellite-borne model of the satellite-ground coordinated fine-tuning system under multiple unstable link conditions based on multiple priority transmission strategies according to the first measurement result, to obtain a second measurement result; the priority transmission strategy is used to indicate the order of transmitting the fine-tuning parameters, and the second measurement result is used to represent the target priority transmission strategy for adaptation to each unstable link condition under the target protocol adapted to the unstable link condition.

[0120] Optionally, the first measurement module 12 includes: A first simulation unit is configured to simulate a plurality of unstable link conditions using a network simulation tool, wherein the link conditions include packet loss rate, delay, and network fluctuation; The first initialization unit is used to initialize the onboard model and training rounds; A first communication link establishing unit is used to establish a communication link between the satellite and the ground station using the TCP protocol and the UDP protocol under different unstable link conditions; a first measurement unit, configured to train the onboard model and record a convergence status of the training, wherein the convergence status includes: training time, convergence time, loss value, and accuracy; the convergence time is the time required for the training to reach a preset error threshold; the convergence time is determined based on the loss value of each training round, the current training round, the training round, and the error threshold; The first recording module is used to determine the target protocol adapted to each unstable link condition from the TCP protocol and the UDP protocol according to the convergence status of each round of training.

[0121] Optionally, the second measurement module 13 includes: A second simulation unit is configured to simulate a plurality of unstable link conditions using a network simulation tool, wherein the link conditions include packet loss rate, delay, and network fluctuation; The second initialization unit is used to initialize the onboard model and training rounds; a second communication link establishing unit, configured to establish, under each unstable link condition, a communication link between the satellite and the ground station according to a target protocol adapted to the unstable link condition; a second measurement unit, configured to train the onboard model based on multiple priority transmission strategies, and record convergence status of the training, wherein the convergence status includes: training time, convergence time, loss value, and accuracy; the convergence time is the time required for the training to reach a preset error threshold; the convergence time is determined based on the loss value of each training round, the current training round, the training round, and the error threshold; The second recording module is used to use the priority transmission strategy with the highest transmission efficiency among the multiple priority transmission strategies as the target priority transmission strategy for adapting to the unstable link condition according to the convergence of the training.

[0122] Optionally, the device further comprises: A third measurement module is configured to, when the target protocol is UDP, train the onboard model based on the UDP protocol and record the convergence of the training, wherein the convergence includes: training time, convergence time, loss value, and accuracy; the convergence time is the time required for the training to reach a preset error threshold; the convergence time is determined based on the loss value of each training round, the current training round, the training round, and the error threshold; The judgment module is used to determine whether the satellite-ground collaborative fine-tuning system adopts a retransmission mechanism in actual application based on whether the training convergence reaches the expected target convergence; A first determining module is configured to determine not to adopt the retransmission mechanism in the satellite-ground coordinated fine-tuning system when the convergence of the training reaches a desired first target convergence; A second determining module is configured to determine whether to adopt the retransmission mechanism in the satellite-ground coordinated fine-tuning system when the convergence of the training reaches a desired second target convergence; Among them, the first target convergence situation is better than the second target convergence situation; the retransmission mechanism indicates that after the satellite transmits the fine-tuning parameters of the screened satellites participating in parameter aggregation to the ground station, and when the communication resources of the communication link between the satellite and the ground station are not used up, the fine-tuning parameters of the satellites participating in parameter aggregation that have not been successfully transmitted are retransmitted.

[0123] Optionally, the device further comprises: a satellite selection module configured to select, under current unstable link conditions, a plurality of satellites from a plurality of satellites within a connection window as target satellites through a satellite selection module of a ground station according to preset rules; the preset rules being set based on the current status of each satellite and the quality and relevance of local data collected by the satellites; a communication module for the multiple target satellites, establishing a communication link with the ground station via the communication module according to a target protocol adapted for the current unstable link condition, and transmitting fine-tuning parameters corresponding to each of the multiple target satellites to the ground station according to a target priority transmission strategy adapted for the current unstable link condition; the fine-tuning parameters are obtained by the target satellites through local training of a locally deployed onboard model of the target satellites based on local data using a local training module; An aggregation module, used for the ground station, aggregating the received multiple fine-tuning parameters through the aggregation module to obtain the global model parameters of this round; a backhaul module, configured for the ground station to transmit the global model parameters to the multiple satellites via a communication link established with the ground station using a target protocol adapted according to current unstable link conditions, so that the multiple satellites update current model parameters of locally deployed onboard models based on the global model parameters, and reselect new target satellites from the multiple satellites through the satellite selection module to participate in the next round of local training until the onboard model training is completed; Among them, when the satellite-borne model is used to realize the disaster navigation function, the local data is the disaster information data collected by the satellite; when the satellite-borne model is used to realize the earth observation function, the local data is the earth observation data collected by the satellite; when the satellite-borne model is used to realize the climate monitoring function, the local data is the climate monitoring data collected by the satellite.

[0124] Optionally, the aggregation module includes: A first aggregation unit is configured to aggregate the fine-tuning parameters of multiple satellites that have successfully transmitted fine-tuning parameters among the selected satellites participating in parameter aggregation and the fine-tuning parameters of the target satellite that failed to transmit fine-tuning parameters in the previous round to obtain global model parameters for this round when fine-tuning parameter transmission of any satellite among the selected satellites participating in parameter aggregation fails; The second aggregation unit is configured to aggregate the fine-tuning parameters of all the selected satellites participating in the parameter aggregation when the fine-tuning parameters of all the selected satellites participating in the parameter aggregation are successfully transmitted, so as to obtain the global model parameters of this round.

[0125] Based on the same inventive concept, another embodiment of the present application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the performance measurement method of the satellite-ground collaborative fine-tuning system as described in any of the above embodiments.

[0126] Based on the same inventive concept, another embodiment of the present application further provides a computer-readable storage medium on which a computer program is stored, wherein when the program is executed by a processor, the performance measurement method of the satellite-ground coordinated fine-tuning system as described in any of the above embodiments is implemented.

[0127] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same and similar parts between the embodiments can be referred to each other. It should be understood by those skilled in the art. The embodiments of the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. The embodiments of the present application may take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The present application is described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the functions in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing terminal device to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including an instruction device, which implements the functions specified in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device, so that a series of operation steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for implementing the process in the process. Figure 1 a process or multiple processes and / or boxes Figure 1The term "comprising" or "including" refers to a step of a function specified in a box or multiple boxes. In this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. The term "comprise", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or terminal equipment including a series of elements includes not only those elements, but also includes other elements not clearly listed, or also includes elements inherent to such process, method, article or terminal equipment. In the absence of more restrictions, the elements limited by the statement "comprising a ..." do not exclude the presence of other identical elements in the process, method, article or terminal equipment including the elements. The above is a detailed introduction to the performance measurement method, device, equipment and medium of a satellite-ground collaborative fine-tuning system provided by the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea; for general technical personnel in this field, based on the ideas of the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A performance measurement method for a satellite-ground coordinated fine-tuning system, characterized in that: The method comprises: A satellite-ground coordinated fine-tuning system is constructed, comprising multiple satellites and a ground station. The satellites are configured with a local training module for performing local training of an onboard model, generating fine-tuning parameters, and transmitting the fine-tuning parameters to the ground station. The ground station is configured with a satellite selection module and an aggregation module. The satellite selection module is configured to select satellites for parameter aggregation, and the aggregation module is configured to aggregate the fine-tuning parameters of the selected satellites for parameter aggregation to generate global model parameters. Measuring, based on the TCP protocol and the UDP protocol, respectively, the convergence of the satellite-onboard model of the satellite-ground coordinated fine-tuning system under multiple unstable link conditions to obtain a first measurement result, where the first measurement result is used to indicate a target protocol adapted for each unstable link condition; According to the first measurement result, the convergence of the satellite-borne model of the satellite-ground coordinated fine-tuning system under multiple unstable link conditions is measured based on multiple priority transmission strategies to obtain a second measurement result; the priority transmission strategy is used to indicate the order of transmitting the fine-tuning parameters, and the second measurement result is used to represent the target priority transmission strategy for adaptation to each unstable link condition under the target protocol adapted to the unstable link condition.

2. The performance measurement method of the satellite-ground coordinated fine-tuning system according to claim 1, characterized in that: The method measures the convergence of the satellite-borne model of the satellite-ground coordinated fine-tuning system under multiple unstable link conditions based on the TCP protocol and the UDP protocol, and obtains a first measurement result, including: Using a network simulation tool to simulate various unstable link conditions, including packet loss rate, delay, and network fluctuation; Initialize the onboard model and training rounds; Under different unstable link conditions, TCP and UDP protocols are used to establish communication links between satellites and ground stations. Training the onboard model and recording the convergence of the training, wherein the convergence includes: training time, convergence time, loss value, and accuracy; the convergence time is the time required for the training to reach a preset error threshold; the convergence time is determined based on the loss value of each training round, the current training round, the training round, and the error threshold; According to the convergence of each round of training, the target protocol adapted to each unstable link condition is determined from the TCP protocol and UDP protocol.

3. The performance measurement method of the satellite-ground coordinated fine-tuning system according to claim 2, characterized in that: The second measurement result is obtained by measuring, based on the first measurement result and based on multiple priority transmission strategies, the convergence of the satellite-borne model of the satellite-ground coordinated fine-tuning system under multiple unstable link conditions, including: Using a network simulation tool to simulate various unstable link conditions, including packet loss rate, delay, and network fluctuation; Initialize the onboard model and training rounds; Under each unstable link condition, establishing a communication link between the satellite and the ground station according to a target protocol adapted to the unstable link condition; The onboard model is trained based on multiple priority transmission strategies, and the convergence of the training is recorded, wherein the convergence status includes: training time, convergence time, loss value, and accuracy rate; the convergence time is the time required for the training to reach a preset error threshold; the convergence time is determined based on the loss value of each training round, the current training round, the training round, and the error threshold; According to the convergence of the training, the priority transmission strategy with the highest transmission efficiency among the multiple priority transmission strategies is used as the target priority transmission strategy for adaptation to the unstable link condition.

4. The performance measurement method of the satellite-ground coordinated fine-tuning system according to claim 3, characterized in that: The plurality of priority transmission strategies include: no priority transmission strategy, gradient assignment priority transmission strategy and dynamic priority transmission strategy; The non-priority transmission strategy means that the fine-tuning parameters of each satellite after local training are transmitted in sequence; The gradient assignment priority transmission strategy indicates that the transmission priority of the fine-tuning parameters after local training of each satellite is determined according to the gradient amplitude of the fine-tuning parameters; The dynamic priority transmission strategy means that the transmission priority of the fine-tuning parameters after local training of each satellite is dynamically determined according to the contribution of each fine-tuning parameter to the convergence speed in each round of training.

5. The performance measurement method of the satellite-ground coordinated fine-tuning system according to claim 1, characterized in that: The method further comprises: When the target protocol is UDP, the onboard model is trained based on the UDP protocol, and the convergence of the training is recorded, wherein the convergence status includes: training time, convergence time, loss value, and accuracy rate; the convergence time is the time required for the training to reach a preset error threshold; the convergence time is determined based on the loss value of each training round, the current training round, the training round, and the error threshold; Based on whether the training convergence reaches the expected target convergence, it is determined whether the satellite-ground collaborative fine-tuning system should adopt a retransmission mechanism in actual application; When the convergence of the training reaches the desired first target convergence, determining not to adopt the retransmission mechanism in the satellite-ground coordinated fine-tuning system; When the convergence of the training reaches the desired second target convergence, determining to adopt the retransmission mechanism on the satellite-ground coordinated fine-tuning system; Among them, the first target convergence situation is better than the second target convergence situation; the retransmission mechanism indicates that after the satellite transmits the fine-tuning parameters of the screened satellites participating in parameter aggregation to the ground station, and when the communication resources of the communication link between the satellite and the ground station are not used up, the fine-tuning parameters of the satellites participating in parameter aggregation that have not been successfully transmitted are retransmitted.

6. The performance measurement method of the satellite-ground coordinated fine-tuning system according to any one of claims 1 to 5, characterized in that: The method further comprises: Under the current unstable link condition, a satellite selection module of the ground station selects multiple satellites as target satellites from multiple satellites within the connection window according to preset rules; the preset rules are set based on the current status of each satellite and the quality and relevance of local data collected by the satellites; The multiple target satellites establish communication links with the ground station through a communication module according to a target protocol adapted for the current unstable link condition, and transmit fine-tuning parameters corresponding to each of the multiple target satellites to the ground station according to a target priority transmission strategy adapted for the current unstable link condition; the fine-tuning parameters are obtained by the target satellites performing local training on a locally deployed onboard model of the target satellites through a local training module based on local data; The ground station aggregates the received multiple fine-tuning parameters through an aggregation module to obtain the global model parameters of this round; The ground station transmits the global model parameters to the multiple satellites via a communication link established with the ground station using a target protocol adapted according to the current unstable link condition, so that the multiple satellites update current model parameters of locally deployed onboard models according to the global model parameters, and reselects new target satellites from the multiple satellites through the satellite selection module to participate in the next round of local training until the onboard model training is completed; Among them, when the satellite-borne model is used to realize the disaster navigation function, the local data is the disaster information data collected by the satellite; when the satellite-borne model is used to realize the earth observation function, the local data is the earth observation data collected by the satellite; when the satellite-borne model is used to realize the climate monitoring function, the local data is the climate monitoring data collected by the satellite.

7. The performance measurement method of the satellite-ground coordinated fine-tuning system according to claim 6, characterized in that: The ground station aggregates the received multiple fine-tuning parameters through an aggregation module to obtain the global model parameters of this round, including: When the fine-tuning parameters of any satellite among the selected satellites participating in parameter aggregation fail to be transmitted, the fine-tuning parameters of multiple satellites that have successfully transmitted the fine-tuning parameters among the selected satellites participating in parameter aggregation and the fine-tuning parameters of the target satellite that failed to transmit in the previous round are aggregated to obtain the global model parameters of this round; When the fine-tuning parameters of all the selected satellites participating in the parameter aggregation are successfully transmitted, the fine-tuning parameters of all the selected satellites participating in the parameter aggregation are aggregated to obtain the global model parameters of this round.

8. A performance measurement device for a satellite-ground coordinated fine-tuning system, characterized in that: The device comprises: A construction module is configured to construct a satellite-ground coordinated fine-tuning system, the satellite-ground coordinated fine-tuning system comprising multiple satellites and ground stations; the satellites are configured with local training modules, configured to perform local training of onboard models, generate fine-tuning parameters, and transmit them to the ground stations; the ground stations are configured with satellite selection modules and aggregation modules, the satellite selection module is configured to screen satellites for parameter aggregation, and the aggregation module is configured to aggregate the fine-tuning parameters of the screened satellites for parameter aggregation to generate global model parameters; A first measurement module is configured to measure, based on the TCP protocol and the UDP protocol, the convergence of the satellite-onboard model of the satellite-ground coordinated fine-tuning system under multiple unstable link conditions, to obtain a first measurement result, where the first measurement result is used to indicate a target protocol adapted for each unstable link condition; The second measurement module is used to measure the convergence of the satellite-borne model of the satellite-ground coordinated fine-tuning system under multiple unstable link conditions based on multiple priority transmission strategies according to the first measurement result, to obtain a second measurement result; the priority transmission strategy is used to indicate the order of transmitting the fine-tuning parameters, and the second measurement result is used to represent the target priority transmission strategy for adaptation to each unstable link condition under the target protocol adapted to the unstable link condition.

9. An electronic device, characterized in that: The system comprises a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the performance measurement method of the satellite-ground collaborative fine-tuning system according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that A computer program is stored thereon, wherein when the computer program is executed by a processor, the performance measurement method of the satellite-ground coordinated fine-tuning system according to any one of claims 1 to 7 is implemented.

Images