A satellite selection method based on multi-satellite resource balancing

Abstract

The application belongs to the technical field of satellite communication and relates to a satellite selection method based on multi-satellite resource balancing, which comprises the following steps: each user i selects a satellite with the highest current signal strength for random access; each user i uploads a data set to the accessed satellite to train a global model; each user i calculates a utility value and sends the utility value to a ground controller; the ground controller determines whether a dynamic balance state is reached according to the utility value; if yes, the controller informs all users to camp on the currently accessed satellite; otherwise, the ground controller calculates an average utility value according to the utility value and broadcasts the average utility value and the utility value of all accessed satellites j to all users; each user i reselects a satellite for access and trains a global model in the next round; the application dynamically selects a suitable satellite for model updating and data transmission according to different requirements of satellite resources and federated learning, thereby realizing balanced utilization of satellite resources.

Description

Technical Field

[0001] This invention belongs to the field of satellite communication technology, and specifically relates to a satellite selection method based on multi-satellite resource balancing. Background Technology

[0002] To better respond to natural disasters, technicians often use data collected by IoT sensors distributed in various isolated areas (such as forests, oceans, and mountains) to train machine learning models. However, many IoT sensors used for natural disaster prediction are located in areas with poor communication infrastructure, lacking both terrestrial base stations and network access points. Therefore, satellite communication, as a seamless global connectivity solution, has shown great potential in these regions.

[0003] Federated learning is a distributed machine learning paradigm that allows multiple devices to collaboratively train a shared model while maintaining data locality. This approach is particularly suitable for data privacy-sensitive and communication-constrained scenarios, such as IoT devices in remote areas. With the aid of satellite communication, Federated Learning (FL) can transcend geographical boundaries, connecting devices distributed globally to jointly train models and thus provide intelligent services.

[0004] However, with low Earth orbit satellites providing coverage, each user is highly likely to be covered by multiple satellites simultaneously. When a user initiates an access request to a satellite, they can select from multiple satellites. But since users are unaware of each other's information, it is highly probable that a large number of users will choose the same satellite for their FL missions.

[0005] Therefore, how to dynamically select appropriate satellites for model updates and data transmission based on satellite resource availability and different user FL mission requirements, thereby making efficient and balanced use of the computing resources of each satellite, is a complex decision-making problem. Summary of the Invention

[0006] To address the aforementioned problems in the prior art, this invention employs a satellite selection method based on multi-satellite resource balancing, comprising:

[0007] S1: Each user Select the satellite with the strongest current signal for random access;

[0008] S2: Each user Upload the local dataset to the connected satellite to train the global model. ;

[0009] S3: All users Based on the signal strength of the accessed satellites, the size of the dataset uploaded to the accessed satellites, and the global model Calculate the utility value of each training time. and their respective utility values Send to the ground controller; among which, This refers to the number of training iterations for the global model.

[0010] S4: Ground controller based on utility value Determine whether a dynamic equilibrium state has been reached. If so, proceed to step S5; otherwise, proceed to step S6.

[0011] S5: The ground controller notifies all users to stay on the currently accessed satellite and completes the satellite selection;

[0012] S6: The ground controller determines the utility value of each user. Calculate the average utility value , will each user utility value As users Access to satellites utility value The average utility value and all connected satellites utility value Broadcast to all users and proceed to step S7;

[0013] S7: All users Calculate the current temperature According to all accessed satellites utility value Average utility value and current temperature Select a satellite for access and execute step S2 to train the next round of global model.

[0014] The local datasets uploaded by each user to the access satellites are non-sensitive datasets; each user Upload the local dataset to the connected satellite to train the global model. include:

[0015] S21: After the connection is established, each user Upload non-sensitive datasets to the accessed satellites;

[0016] S22: After each access satellite successfully receives the non-sensitive data set, it sends a notification to the corresponding user. Send confirmation message;

[0017] S23: Each access satellite uses the non-sensitive datasets received from multiple users to train a local model;

[0018] S24: After each access satellite has trained its local model, the trained local model is sent to one of the satellites for aggregation to obtain the trained global model, and then the trained global model is sent back to each user. .

[0019] Ground controller based on utility value Determining whether a dynamic equilibrium state has been reached includes:

[0020] Set threshold If the utility value of all users in the previous round Utility value in the current round The differences are all less than the threshold. Then a dynamic equilibrium is reached.

[0021] Each user Calculating the current temperature includes: setting the annealing factor and obtaining the temperature from the previous training round. The temperature from the previous training round Multiply by the annealing factor to get the current temperature. ; where the initial temperature The process includes: the ground controller calculating the utility value of each user during the first training round. variance According to variance Calculate the initial temperature and the initial temperature The broadcast was sent to all users.

[0022] user According to all accessed satellites utility value Average utility value and current temperature Choosing satellite access includes:

[0023] user Calculate its own utility value With average utility value difference , if the difference If the value is greater than 0, then remain on the currently accessed satellite; otherwise, determine if there is a utility value among all accessed satellites. Higher than users The utility value of currently connected satellites If any satellites are available, then filter out the utility values ​​from all accessible satellites. Higher than users The utility value of currently connected satellites The satellites, to obtain the satellite set Weighted random sampling was used to sample the satellite ensemble. If one satellite is selected for access, then a satellite is selected from all the satellites to be accessed using a weighted random sampling method. The acceptance probability mechanism of the simulated annealing algorithm is then used to determine whether to access the selected satellite.

[0024] Weighted random sampling was used from the satellite ensemble. Selecting a satellite includes: calculating the satellite set Each satellite in the user The utility difference of the currently accessing satellites is normalized to obtain the selection probability of each satellite. According to the selection probability from Choose one satellite.

[0025] The acceptance probability mechanism of the simulated annealing algorithm determines whether to connect to a selected satellite, including: based on the utility difference of the selected satellite and the current temperature. Calculate the probability of acceptance Generate a random number ,like Then the user Access the selected satellite; otherwise, remain on the currently accessed satellite.

[0026] Calculate the probability of acceptance include:

[0027] in, For the selected satellite, For the selected satellite Poor utility It is an exponential function with Euler's number e as the base.

[0028] The beneficial effects of this invention are:

[0029] This invention dynamically adjusts the satellite's resource conditions and the different needs of the user's federated learning task, namely the satellite's signal strength, the size of the dataset, and the global model. By allocating sufficient training time and selecting appropriate satellites for model updates and data transmission, the resource overload problem caused by a large number of users selecting the same satellite is avoided, thus achieving balanced utilization of satellite resources. Attached Figure Description

[0030] Figure 1 A schematic diagram of a system model for a satellite selection method based on multi-satellite resource balancing provided in an embodiment of the present invention;

[0031] Figure 2 A flowchart illustrating a satellite selection method based on multi-satellite resource balancing, provided as an embodiment of the present invention;

[0032] Figure 3 A flowchart of the four-step random access process provided in this embodiment of the invention;

[0033] Figure 4 This is a flowchart illustrating a user decision on whether to replace a satellite for the next round of global model training, provided in an embodiment of the present invention. Detailed Implementation

[0034] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0035] The framework of the satellite system model is as follows Figure 1 As shown, the entire system consists of three parts: a low-Earth orbit satellite (S), a user terminal (UE), and a controller. The user terminal communicates bidirectionally with the low-Earth orbit satellite, and the user terminal communicates bidirectionally with the controller via a low-power wide-area network (LPWAN). The controller only communicates with each user terminal. The LPWAN uses LoRa (Long Range) technology.

[0036] A communication scenario consisting of N low-Earth orbit satellites, K ground terminal users, and 1 ground controller is constructed. A wildfire prediction model is trained through collaborative federated learning between the satellites and the ground terminal users. The ground terminal users are distributed in different isolated areas (such as forests and mountains) to collect natural environmental data, including air temperature, temperature, wind speed, wind direction, and soil moisture.

[0037] Based on the above communication scenarios, this embodiment of the invention proposes a satellite selection method based on multi-satellite resource balancing, such as... Figure 2 As shown, it includes:

[0038] S1: At the start of the first round of global model training, each user Select the satellite with the strongest current signal for random access;

[0039] In communication scenarios, ground-based terminal users in various locations will be covered by n low-Earth orbit satellites at a certain moment ( During the initial global model training, the user terminal will select the satellite with the highest current signal strength and use methods such as... Figure 3 The four-step random access method shown establishes a connection with the satellite;

[0040] The process of establishing a connection with a satellite using four-step random access includes:

[0041] S11: The ground terminal user randomly selects a preamble sequence (Msg1) and sends it to the satellite via the Physical Random Access Channel (PRACH) to request the establishment of a communication connection;

[0042] S12: After receiving the request, the satellite sends a response message (Msg2) to the ground terminal user through the Physical Downlink Shared Channel (PDSCH), including uplink resource allocation and timing adjustment information, so that the ground terminal user can perform uplink synchronization;

[0043] S13: Based on the resource allocation in the response message, the ground terminal user sends a connection establishment request (Msg3) to the satellite via the Physical Uplink Shared Channel (PUSCH), which includes the necessary connection information and user data;

[0044] S14: After receiving the connection establishment request, the satellite verifies and processes it, and then sends a connection establishment confirmation message (Msg4) to the ground terminal user through the Physical Downlink Shared Channel (PDSCH), thereby successfully establishing a communication connection.

[0045] S2: Each user Upload the local dataset to the connected satellite to train the global model. ;

[0046] Upload the local dataset to the satellite to train the global model. include:

[0047] S21: After the connection is established, the user uploads non-sensitive datasets to the access satellite;

[0048] user The total number of data samples to be processed is Among them, the non-sensitive datasets that need to be offloaded to the satellite are ,Right now The remaining data is provided by the user. Perform local model updates; each user needs to identify the non-sensitive datasets to be offloaded to the satellite before training begins; after the connection is established, the user terminal will offload the non-sensitive datasets to the satellite. The above is used to perform model training;

[0049] S22: After each access satellite successfully receives the non-sensitive data set, it sends a notification to the corresponding user. Send confirmation message;

[0050] S23: Each access satellite uses the non-sensitive datasets received from multiple users to train a local model;

[0051] S24: After each access satellite has trained its local model, the trained local model is sent to one of the satellites for aggregation to obtain the trained global model, and then the trained global model is sent back to each user. .

[0052] S3: All users Based on the signal strength of the accessed satellites, the size of the dataset uploaded to the accessed satellites, and the global model Calculate the utility value of each training time. and their respective utility values Send to the ground controller; among which, This refers to the number of training iterations for the global model.

[0053] S4: Ground controller based on utility value Determine whether a dynamic equilibrium state has been reached. If so, proceed to step S5; otherwise, proceed to step S6.

[0054] The ground controller determines whether a dynamic equilibrium state has been reached based on utility values, including:

[0055] Set threshold Preferably, The value is 0.05; if the utility value of all users in the previous round is... Utility value in the current round The differences are all less than the threshold. :

[0056]

[0057] This indicates that a dynamic equilibrium has been reached.

[0058] S5: The ground controller notifies all users to stay on the currently accessed satellite and completes the satellite selection;

[0059] S6: The ground controller determines the utility value of each user. Calculate the average utility value , will each user utility value As users Access to satellites utility value The average utility value and all connected satellites utility value Broadcast to all users and proceed to step S7;

[0060] Ground controller calculates average utility value include:

[0061]

[0062] Where N is the total number of users, and t is the current training round;

[0063] S7: All users Calculate the current temperature And based on all accessed satellites utility value Average utility value and current temperature Select a satellite for access and execute step S2 to train the next round of global model.

[0064] like Figure 4 As shown, the user According to all accessed satellites utility value Average utility value and current temperature Choosing a satellite for access includes:

[0065] user Calculate its own utility value With average utility value difference , if the difference If the value is greater than 0, then remain on the currently accessed satellite; otherwise, determine if there is a utility value among all accessed satellites. Higher than users The utility value of currently connected satellites If any satellites are available, then filter out the utility values ​​from all accessible satellites. Higher than users The utility value of currently connected satellites The satellites, to obtain the satellite set Weighted random sampling was used to sample the satellite ensemble. If one satellite is selected for access, then a satellite is selected from all the satellites to be accessed using a weighted random sampling method. The acceptance probability mechanism of the simulated annealing algorithm is then used to determine whether to access the selected satellite.

[0066] Using weighted random sampling method from Selecting a satellite includes:

[0067] calculate Each candidate satellite in the set poor utility :

[0068]

[0069] Normalize the utility difference of each candidate satellite to obtain the selection probability of each candidate satellite. :

[0070]

[0071] Where s is The index of the satellite in the set.

[0072] Based on the probability of selection from Choose one satellite to train the global model for the next round.

[0073] The use of the simulated annealing algorithm's acceptance probability mechanism to determine whether to connect to selected satellites for global model training includes:

[0074] Based on the selected satellite poor utility and current temperature Calculate the probability of acceptance :

[0075]

[0076] The user generates a random number. ,like If the selected satellite is selected, it will be used in the next round of training; otherwise, the currently accessed satellite will remain in service.

[0077] Each user calculates the temperature for the current training round:

[0078]

[0079] initial temperature :

[0080]

[0081] Where c is a proportionality constant; the larger c is, the higher the initial temperature and the more possibilities are available for exploration; the smaller c is, the lower the initial temperature, which is beneficial for faster convergence to a local optimum. In this embodiment, c is set to 1. It is the utility value of each user in the first training round. Variance of:

[0082]

[0083] in, This is the annealing coefficient. This affects the rate of temperature decrease; a slower rate of temperature decrease is beneficial for fully searching the solution space. In this embodiment, it is set... It is 0.85.

[0084] The above-described embodiments further illustrate the purpose, technical solution, and advantages of the present invention. It should be understood that the above-described embodiments are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made to the present invention within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A satellite selection method based on multi-satellite resource balancing, characterized in that, include: S1: All users Select the satellite with the strongest current signal for random access; S2: Each user Upload the local dataset to the connected satellite to train the global model. ; S3: All users Based on the signal strength of the accessed satellites, the size of the dataset uploaded to the accessed satellites, and the global model Calculate the utility value of each training time. and their respective utility values Send to the ground controller; among which, This refers to the number of training iterations for the global model. S4: Ground controller based on utility value Determine whether a dynamic equilibrium state has been reached. If so, proceed to step S5; otherwise, proceed to step S6. S5: The ground controller notifies all users to stay on the currently accessed satellite and completes the satellite selection; S6: The ground controller determines the utility value of each user. Calculate the average utility value , will each user utility value As users Access to satellites utility value Average utility value and all connected satellites utility value Broadcast to all users and proceed to step S7; S7: All users Calculate the current temperature According to all accessed satellites utility value Average utility value and current temperature Select a satellite for access and execute step S2 to train the next round of global model.

2. The satellite selection method based on multi-satellite resource balancing according to claim 1, characterized in that, The local datasets uploaded by each user to the access satellites are non-sensitive datasets; each user Upload the local dataset to the connected satellite to train the global model. include: S21: After the connection is established, each user Upload non-sensitive datasets to the accessed satellites; S22: After each access satellite successfully receives the non-sensitive data set, it sends a notification to the corresponding user. Send confirmation message; S23: Each access satellite uses the non-sensitive datasets received from multiple users to train a local model; S24: After each access satellite has trained its local model, the trained local model is sent to one of the satellites for aggregation to obtain the trained global model, and then the trained global model is sent back to each user. .

3. The satellite selection method based on multi-satellite resource balancing according to claim 1, characterized in that, Ground controller based on utility value Determining whether a dynamic equilibrium state has been reached includes: Set threshold If the utility value of all users in the previous round Utility value in the current round The differences are all less than the threshold. Then a dynamic equilibrium is reached.

4. The satellite selection method based on multi-satellite resource balancing according to claim 1, characterized in that, Each user Calculating the current temperature includes: setting the annealing factor and obtaining the temperature from the previous training round. The temperature from the previous training round Multiply by the annealing factor to get the current temperature. ; where the initial temperature The process includes: the ground controller calculating the utility value of each user during the first training round. variance According to variance Calculate the initial temperature and the initial temperature The broadcast was sent to all users.

5. The satellite selection method based on multi-satellite resource balancing according to claim 1, characterized in that, user According to all accessed satellites utility value Average utility value and current temperature Choosing satellite access includes: user Calculate its own utility value With average utility value difference , if the difference If the value is greater than 0, then remain on the currently accessed satellite; otherwise, determine if there is a utility value among all accessed satellites. Higher than users The utility value of currently connected satellites If any satellites are available, then filter out the utility values ​​from all accessible satellites. Higher than users The utility value of currently connected satellites The satellites, to obtain the satellite set Weighted random sampling was used to sample the satellite ensemble. If one satellite is selected for access, then a satellite is selected from all the satellites to be accessed using a weighted random sampling method. The acceptance probability mechanism of the simulated annealing algorithm is then used to determine whether to access the selected satellite.

6. The satellite selection method based on multi-satellite resource balancing according to claim 5, characterized in that, Weighted random sampling was used from the satellite set. Selecting a satellite includes: calculating the satellite set Each satellite in the system is connected to the user. The utility difference of the currently accessing satellites is normalized to obtain the selection probability of each satellite. According to the selection probability from Choose one satellite.

7. A satellite selection method based on multi-satellite resource balancing according to claim 6, characterized in that, The acceptance probability mechanism of the simulated annealing algorithm determines whether to connect to a selected satellite, including: based on the utility difference of the selected satellite and the current temperature. Calculate the probability of acceptance Generate a random number ,like Then the user Access the selected satellite; otherwise, remain on the currently accessed satellite.

8. A satellite selection method based on multi-satellite resource balancing according to claim 7, characterized in that, Calculate the probability of acceptance include: in, For the selected satellite, For the selected satellite Poor utility It is an exponential function with Euler's number e as the base.

Images