Satellite cooperation-based edge computing offloading prediction method and device, and storage medium

Abstract

The application discloses a satellite cooperation-based edge computing offloading prediction method and device and a storage medium. The method comprises the following steps: determining a satellite constellation corresponding to a target satellite and a task computation amount of each satellite in the satellite constellation for a same region in a current period; taking each satellite as a first node and a task offloading relationship between adjacent two satellites as an edge to construct a first graph structure; inputting the first graph structure into a pre-trained graph neural network model, iteratively updating the first graph structure according to a pre-set calculation rule by using the graph neural network model, and generating a second graph structure; and determining a target feature vector corresponding to the target satellite in the second graph structure, and inputting the target feature vector into a pre-trained first task amount prediction model to predict a task offloading amount received by the target satellite.

Description

Technical Field

[0001] This application relates to the field of satellite computing technology, and in particular to an edge computing offloading prediction method, apparatus and storage medium based on satellite collaboration. Background Technology

[0002] In a multi-satellite constellation, when any one of the satellites is overloaded with computing power, it can offload some of the tasks to other satellites with available computing resources. This ensures computational efficiency and improves the utilization of computing resources. This operation is called inter-satellite computing task offloading and is one of the core advantages of modern satellite network design.

[0003] Figure 1 This is a schematic diagram of existing multi-satellite computing task offloading. (Reference) Figure 1 As shown, the system includes multiple satellites 1 to 3. Satellite 1 is connected to Satellite 2 via an inter-satellite link, and Satellite 1 is also connected to Satellite 3 via an inter-satellite link. Therefore, if Satellite 1 has a large number of computational tasks that it cannot complete using only its own computing resources, Satellite 1 can, for example, send 10% of the computational tasks to Satellite 2, and another 10% to Satellite 3. This ensures the successful completion of the computational tasks.

[0004] On the other hand, in addition to its own computational tasks, Satellite 2 (or Satellite 3) also needs to compute some of the tasks offloaded by Satellite 1. However, due to the limited computational resources of Satellite 2 (or Satellite 3), and the potential influence of other objective factors, predicting the optimal amount of offloaded tasks from other satellites for Satellite 2 (or Satellite 3) to ensure computational efficiency is a critical issue. These objective factors could include, for example, the distance between Satellite 1 and Satellite 2 (or Satellite 3), and the attenuation coefficient between them.

[0005] There is currently no effective solution to the technical problem in the existing technology of how to predict how much workload from other satellites is most suitable for a satellite to receive in order to ensure the computational efficiency of the mission, due to the limited computing resources of the satellite. Summary of the Invention

[0006] The embodiments of this disclosure provide an edge computing offload prediction method, apparatus, and storage medium based on satellite collaboration, to at least solve the technical problem in the prior art of how to predict how much offloaded task a satellite is most suitable to receive from other satellites in order to ensure the computational efficiency of the task due to the limited computing resources of satellites.

[0007] According to one aspect of the present disclosure, an edge computing offloading prediction method based on satellite cooperation is provided, comprising: determining a satellite constellation corresponding to a target satellite, and the task computing volume of each satellite in the constellation for the same area during the current time period; constructing a first graph structure with each satellite as a first node and the task offloading relationship between two adjacent satellites as edges, wherein the node attributes of the first node represent the task computing volume corresponding to each satellite, and the edge attributes of the first edge represent the task offloading weight, distance weight, and attenuation coefficient weight between two adjacent satellites; inputting the first graph structure into a pre-trained graph neural network model, iteratively updating the first graph structure using the graph neural network model and according to a pre-set calculation rule to generate a second graph structure; and determining the target feature vector corresponding to the target satellite in the second graph structure, and inputting the target feature vector into a pre-trained first task volume prediction model to predict the task offloading volume received by the target satellite.

[0008] According to another aspect of the present disclosure, a storage medium is also provided, the storage medium including a stored program, wherein, when the program is executed, a processor performs any of the methods described above.

[0009] According to another aspect of the present disclosure, an edge computing offloading prediction device based on satellite cooperation is also provided, comprising: a first task computation volume determination module, configured to determine the satellite constellation corresponding to the target satellite, and the task computation volume of each satellite in the satellite constellation for the same area in the current time period; a first graph structure construction module, configured to construct a first graph structure with each satellite as a first node and the task offloading relationship between two adjacent satellites as edges, wherein the node attributes of the first node represent the task computation volume corresponding to each satellite, and the edge attributes of the first edge represent the task offloading weight, distance weight, and attenuation coefficient weight between two adjacent satellites; a second graph structure generation module, configured to input the first graph structure into a pre-trained graph neural network model, and iteratively update the first graph structure using the graph neural network model and according to a pre-set calculation rule to generate a second graph structure; and a task offloading volume prediction module, configured to determine the target feature vector corresponding to the target satellite in the second graph structure, and input the target feature vector into a pre-trained first task volume prediction model to predict the task offloading volume received by the target satellite.

[0010] According to another aspect of the present disclosure, an edge computing offloading prediction device based on satellite cooperation is also provided, comprising: a processor; and a memory connected to the processor, configured to provide the processor with instructions for processing the following steps: determining a satellite constellation corresponding to a target satellite, and the task computation volume of each satellite in the constellation for the same area during the current time period; constructing a first graph structure with each satellite as a first node and the task offloading relationship between two adjacent satellites as edges, wherein the node attributes of the first nodes represent the task computation volume corresponding to each satellite, and the edge attributes of the first edges represent the task offloading weight, distance weight, and attenuation coefficient weight between two adjacent satellites; inputting the first graph structure into a pre-trained graph neural network model, iteratively updating the first graph structure using the graph neural network model and according to a pre-set calculation rule to generate a second graph structure; and determining the target feature vector corresponding to the target satellite in the second graph structure, and inputting the target feature vector into a pre-trained first task volume prediction model to predict the task offloading volume received by the target satellite.

[0011] This application provides an edge computing offload prediction method based on satellite cooperation. First, the processor determines the satellite constellation corresponding to the target satellite and the computational load of each satellite in the constellation for the same region during the current time period. Then, the processor constructs a first graph structure using each satellite as the first node and the offload relationship between adjacent satellites as edges. Further, the processor outputs the first graph structure to a pre-trained graph neural network model, iteratively updating the first graph structure using the graph neural network model according to pre-defined computation rules to generate a second graph structure. Finally, the processor determines the target feature vector corresponding to the target satellite in the second graph structure and inputs the target feature vector into a pre-trained first task load prediction model to predict the offload load received by the target satellite.

[0012] As described above, the node attributes of the first node in the first graph structure constructed in this application represent the computational workload of each satellite, and the edge attributes of the first edge represent the task offloading weight, distance weight, and attenuation coefficient weight between two adjacent satellites. Therefore, the first graph structure constructed in this application not only considers the computational capabilities of the target satellite itself, but also the factors influencing task offloading, communication quality, and computational efficiency between the target satellite and other satellites.

[0013] Therefore, by iteratively updating the first graph structure and using the target feature vector corresponding to the target satellite in the second graph structure for prediction, the optimal task unloading amount for the target satellite to receive can be predicted.

[0014] In other words, determining the optimal amount of unloaded tasks that the target satellite can receive not only ensures successful mission completion but also improves the target satellite's computational efficiency. This avoids situations where the target satellite receives too much unloaded task, leading to mission failure, or too little, resulting in low utilization of computational resources. Therefore, it solves the technical problem in existing technologies of how to predict the optimal amount of unloaded tasks from other satellites to ensure mission computational efficiency, given the limited computational resources of the satellite. Attached Figure Description

[0015] The accompanying drawings, which are included to provide a further understanding of this disclosure and form part of this application, illustrate exemplary embodiments of this disclosure and are used to explain this disclosure, but do not constitute an undue limitation of this disclosure. In the drawings:

[0016] Figure 1 This is a schematic diagram illustrating the offloading of multi-satellite computing tasks in existing technologies;

[0017] Figure 2 This is a schematic diagram of the edge computing offloading prediction system based on satellite cooperation as described in Embodiment 1 of this application;

[0018] Figure 3A This is a schematic diagram of the hardware architecture of satellites 1 to m according to Embodiment 1 of this application;

[0019] Figure 3B This is a schematic diagram of the hardware architecture of the ground system 20 according to Embodiment 1 of this application;

[0020] Figure 4 This is a modular schematic diagram of the edge computing offloading prediction system based on satellite cooperation as described in Embodiment 1 of this application;

[0021] Figure 5 This is a flowchart of the edge computing offloading prediction method based on satellite cooperation according to Embodiment 1 of this application;

[0022] Figure 6 This is a schematic diagram of the structure of the first figure according to Embodiment 1 of this application;

[0023] Figure 7 This is a model architecture diagram of the graph neural network model according to Embodiment 1 of this application;

[0024] Figure 8 This is a schematic diagram of the first task volume prediction model according to Embodiment 1 of this application;

[0025] Figure 9 This is a schematic diagram of the satellite-cooperative edge computing offloading prediction device according to Embodiment 2 of this application; and

[0026] Figure 10 This is a schematic diagram of the edge computing offloading prediction device based on satellite cooperation as described in Embodiment 3 of this application. Detailed Implementation

[0027] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this disclosure.

[0028] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0029] Example 1

[0030] According to this embodiment, a method embodiment for edge computing offloading prediction based on satellite cooperation is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Also, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0031] Figure 2 This is a schematic diagram of the satellite-cooperative edge computing offloading prediction system according to Embodiment 1 of this application. (Reference) Figure 2 As shown, the edge computing offloading prediction system includes multiple satellites 1 to m. Each satellite 1 to m in the satellite constellation can establish a communication connection with the ground system 20, thereby ensuring that satellite 1 to m can send prediction requests to the ground system, and that the ground system 20 can send the predicted optimal task offloading amount to satellite 1 to m.

[0032] Figure 3A This is a schematic diagram of the hardware architecture of satellites 1 to m according to Embodiment 1 of this application. (Reference) Figure 3A As shown, satellites 1-m include an integrated electronic system, which includes a processor, a memory, a bus management module, and a communication interface. The memory is connected to the processor, allowing the processor to access the memory, read program instructions stored in the memory, read data from the memory, or write data to the memory. The bus management module is connected to the processor and also to a bus such as a CAN bus. Thus, the processor can communicate with onboard peripherals connected to the bus through the bus managed by the bus management module. Furthermore, the processor also communicates with devices such as cameras, star sensors, telemetry and command transponders, and data transmission equipment via the communication interface. Those skilled in the art will understand that… Figure 3A The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, a satellite system may also include... Figure 3A The more or fewer components shown, or having the same Figure 3A The different configurations shown.

[0033] Figure 3B Further shown Figure 1 A schematic diagram of the hardware architecture of the ground system 20. (Reference) Figure 3B As shown, the ground system 20 may include one or more processors (processors may include, but are not limited to, microprocessors such as MCUs or programmable logic devices such as FPGAs), a memory for storing data, a transmission device for communication functions, and an input / output interface. The memory, transmission device, and input / output interface are connected to the processor via a bus. In addition, it may also include a display, keyboard, and cursor control device connected to the input / output interface. Those skilled in the art will understand that... Figure 3B The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, the ground system may also include... Figure 3B The more or fewer components shown, or having the same Figure 3B The different configurations shown.

[0034] It should be noted that, Figure 3A and Figure 3B One or more processors and / or other data processing circuits shown herein may generally be referred to as "data processing circuitry". This data processing circuitry may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuitry may be a single, independent processing module, or may be integrated, in whole or in part, into any other element in a computing device. As involved in embodiments of this disclosure, the data processing circuitry serves as processor control (e.g., selection of a variable resistor termination path connected to an interface).

[0035] Figure 3A and Figure 3BThe memory shown can be used to store software programs and modules of application software, such as the program instruction / data storage device corresponding to the satellite-cooperative edge computing offloading prediction method in this embodiment of the present disclosure. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby realizing the satellite-cooperative edge computing offloading prediction method of the above-mentioned application. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory.

[0036] It should be noted here that, in some optional embodiments, the above... Figure 3A and Figure 3B The device shown may include hardware elements (including circuitry), software elements (including computer code stored on a computer-readable medium), or a combination of both hardware and software elements. It should be noted that... Figure 3A and Figure 3B This is only one instance of a specific particular instance, and is intended to illustrate the types of components that may exist in the aforementioned devices.

[0037] Figure 4 This is a modular schematic diagram of a satellite-cooperative edge computing offloading prediction system according to an embodiment of this application. (Reference) Figure 4 As shown, the edge computing offloading prediction system includes a data acquisition module, a first prediction module, a graph structure construction module, an iterative update module, a second prediction module, and a storage module.

[0038] The system comprises the following modules: a data acquisition module and a storage module, used to acquire the computational workload of each satellite in the constellation for the same region during various historical time periods. A first prediction module, also connected to the data acquisition module, calculates the computational workload of each satellite for the same region during the current time period based on the computational workload of each satellite during various historical time periods. A graph structure construction module, connected to both the first prediction module and the storage module, acquires data corresponding to the node attributes of the first node and the edge attributes of the first edge from the storage module, and constructs a first graph structure. An iterative update module, connected to the graph structure construction module, iteratively updates the first graph structure using a pre-trained graph neural network and generates a corresponding second graph structure. A second prediction module, connected to the iterative update module, predicts the unloading workload received by the target satellite based on the second graph structure.

[0039] Under the aforementioned operating environment, according to the first aspect of this embodiment, an edge computing offloading prediction method based on satellite cooperation is provided. This method consists of... Figure 4 The edge computing offloading prediction system implementation is shown in the figure. Figure 4A flowchart illustrating the method is shown below. (Refer to...) Figure 5 As shown, the method includes:

[0040] S502: Determine the satellite constellation corresponding to the target satellite, and the computational workload of each satellite in the constellation for the same area during the current time period;

[0041] S504: The first graph structure is constructed with each satellite as the first node and the task offloading relationship between two adjacent satellites as the edge. The node attribute of the first node represents the task computation amount corresponding to each satellite, and the edge attribute of the first edge represents the task offloading weight, distance weight and attenuation coefficient weight between two adjacent satellites.

[0042] S506: Input the first graph structure into a pre-trained graph neural network model, and use the graph neural network model to iteratively update the first graph structure according to pre-set calculation rules to generate the second graph structure; and

[0043] S508: Determine the target feature vector corresponding to the target satellite in the second graph structure, and input the target feature vector into the pre-trained first task load prediction model to predict the task unloading amount received by the target satellite.

[0044] Specifically, firstly, in response to a request from the target satellite to predict the most suitable amount of mission offload for reception in the next time period, the ground system 20 determines the satellite constellation corresponding to the target satellite. The satellite constellation corresponding to the target satellite may include, for example, satellites 1 to m, and satellite 3 may be the target satellite that sent the prediction request to the ground system 20.

[0045] Then, after the ground system 20 determines the satellite constellation corresponding to the target satellite, it uses the data acquisition module to retrieve the task computation amount for the same area for each satellite in the constellation during each historical time period from the storage module. Here, task computation amount represents the total number of tasks calculated when the satellites cover the area. For example, the data acquisition module retrieves historical time periods from the storage module. Within, the computational workload of satellite 1 for region A. Satellite 2's computational workload for region A ...; The computational workload of satellite m for region A

[0046] The data acquisition module retrieves historical time periods from the storage module. Within, the computational workload of satellite 1 for region A. Satellite 2's computational workload for region A ...; The computational workload of satellite m for region A

[0047] And so on.

[0048] The data acquisition module retrieves historical time periods from the storage module. Within, the computational workload of satellite 1 for region A. Satellite 2's computational workload for region A ...; The computational workload of satellite m for region A

[0049] The data acquisition module then sends the acquired data to the first prediction module, which in turn predicts the data based on the historical data of each satellite (1-m) during various time periods. To reduce the computational load of the task, construct a task computation vector Q corresponding to each satellite from 1 to m. 1,1 ~Q 1,m .in,

[0050] Furthermore, the first prediction module will calculate the mission vector Q corresponding to each satellite 1 to m. 1,1 ~Q 1,m The input is fed into a pre-trained second task prediction model, which outputs the task computation Q of each satellite 1 to m for region A in the current time period T2. 2,1 ~Q 2,m (S502).

[0051] Then, the first prediction module calculates the task amount Q for region A for each of the determined satellites 1 to m in the current time period T2. 2,1 ~Q 2,m The data is sent to the graph structure construction module, which uses each satellite as the first node and the task offloading relationship between adjacent satellites as edges to construct a first graph structure (S504). The node attributes of the first node in the first graph structure represent the task computation amount corresponding to each satellite, and the edge attributes of the first edge represent the task offloading weight, distance weight, and attenuation coefficient weight between adjacent satellites.

[0052] Figure 6 This is a schematic diagram of the structure described in Embodiment 1 of this application. (Reference) Figure 6 As shown, the graph structure construction module uses each satellite 1 to m as the first node. Therefore, the node attributes of the first node include the task computation amount Q in the current time period T2 corresponding to satellite 1. 2,1 The computational workload Q for the current time period T2 corresponding to satellite 2 2,2 ..., the computational load Q of the task in the current time period T2 corresponding to satellite m. 2,m .

[0053] The graph structure construction module uses the task offloading relationship between two adjacent satellites as the first edge. The edge attributes of this first edge include the task offloading weight, distance weight, and attenuation coefficient weight between the two adjacent satellites. The task offloading weight indicates the weighted relationship between the amount of offloaded tasks received by a satellite and the total amount of offloaded tasks by the satellite constellation over multiple historical periods. For example, in historical periods... Inside, Satellite 2 receives the workload unloaded by Satellite 1. Satellite 1 receives the amount of mission data that was offloaded by Satellite 2. Satellite 2 received the amount of mission data that was offloaded by Satellite 3. Satellite 3 received the amount of mission data that was offloaded by Satellite 2. ..., satellite m-1 receives the amount of task unloaded by satellite m as... Satellite m receives the amount of tasks unloaded by satellite m-1. And in historical periods Within, the total amount of tasks unloaded by multiple satellites is q1.

[0054] In historical periods Within, the weighted relationship between the tasks received by satellite 2 from satellite 1 and the total number of tasks is as follows: (i.e., task offloading weight), the weight relationship between the tasks received by satellite 1 and those offloaded by satellite 2 relative to the total number of tasks is as follows: (i.e., task offloading weight), the weight relationship between the tasks received by satellite 2 from satellite 3 offloading and the total number of tasks is as follows: (i.e., task offloading weight), the weight relationship between the tasks offloaded by satellite 2 and the total number of tasks received by satellite 3 is as follows: (i.e., task offloading weights), ..., the weight relationship between the tasks offloaded by satellite m received by satellite m and the total number of tasks is as follows: (i.e., task offloading weight), the weight relationship between the tasks offloaded by satellite m-1 received by satellite m and the total number of tasks is as follows: (i.e., task unloading weight).

[0055] The distance weight is used to indicate the weighted relationship between the distance between two adjacent satellites, using the two closest adjacent satellites as a reference, and the aforementioned reference distance. The edge attributes of the first edge include the distance weight between satellite 2 and satellite 1. Distance weight between satellite 2 and satellite 3 ...distance weight between satellite m-1 and satellite m

[0056] The edge attributes of the first side also include the attenuation coefficient weight between satellite 2 and satellite 1. Attenuation coefficient weighting between satellite 2 and satellite 3 ..., the attenuation coefficient weight between satellite m-1 and satellite m The attenuation coefficient weight between two adjacent satellites is affected by factors such as rainfall, cloud cover, and fog cover.

[0057] Furthermore, when constructing a first graph structure, the graph structure construction module sends the first graph structure to the iterative update module. The iterative update module then inputs the first graph structure into a pre-trained graph neural network model, uses the graph neural network model and follows pre-set calculation rules to iteratively update the first graph structure, generating a second graph structure (S506). In this embodiment, the graph neural network model includes multiple first MLP models for iteratively updating the edge attributes of the first edge of the first graph structure and multiple second MLP models for iteratively updating the node attributes of the first node of the first graph structure.

[0058] Thus, the graph neural network model can utilize multiple first MLP models and construct edge attribute vectors according to the first calculation rules corresponding to the first edge, and use multiple first MLP models to iteratively update the edge attribute vectors of the first edge, thereby generating edge features corresponding to the second graph structure.

[0059] Similarly, the graph neural network model can utilize multiple second MLP models and construct node attribute vectors based on the second extreme rules corresponding to the first node. Iterative updates of the node attribute vectors of the first node are then performed using multiple second MLP models to generate node features corresponding to the second graph structure. The above will be described in detail later, and therefore will not be repeated here.

[0060] The iterative update module then sends the second graph structure to the prediction module. The prediction module determines the target feature vector corresponding to the target satellite in the second graph structure and inputs this target feature vector into a pre-trained first task load prediction model to predict the task unloading amount received by the target satellite (S508). Specifically, firstly, the prediction module determines the second node corresponding to the target satellite in the second graph structure and the target feature vector corresponding to the second node. The target feature vector includes the node features corresponding to the second node, the edge features of the two adjacent second edges of the second node, and the node features of the second nodes corresponding to the two adjacent second edges. Then, the prediction module inputs the target feature vector into the first task load prediction model and determines the task unloading amount corresponding to the target satellite. That is, the prediction module outputs the task unloading amount from other satellites that the target satellite is most suitable to receive. The above will be described in detail later, so it will not be repeated here.

[0061] As described in the background section, from another perspective, in addition to its own computational tasks, Satellite 2 (Satellite 3) also needs to compute some of the tasks offloaded by Satellite 1. However, due to the limited computational resources of Satellite 2 (Satellite 3) and the potential influence of other objective factors, predicting how much of the offloaded workload from other satellites is most suitable for Satellite 2 (Satellite 3) to ensure computational efficiency is a pressing issue. These objective factors may include, for example, the distance between Satellite 1 and Satellite 2 (Satellite 3) and the attenuation coefficient between them.

[0062] In view of this, this application provides an edge computing offload prediction method based on satellite cooperation. Furthermore, as described above, the node attributes of the first node in the first graph structure constructed in this application represent the computational load of each satellite's corresponding task, and the edge attributes of the first edge represent the task offload weight, distance weight, and attenuation coefficient weight between two adjacent satellites. Therefore, the first graph structure constructed in this application not only considers the computational task capability of the target satellite itself but also the influencing factors of task offload between the target satellite and other satellites.

[0063] Therefore, by iteratively updating the first graph structure and using the target feature vector corresponding to the target satellite in the second graph structure for prediction, the optimal task unloading amount for the target satellite to receive can be predicted.

[0064] In other words, determining the optimal amount of unloaded tasks that the target satellite can receive not only ensures successful mission completion but also improves the target satellite's computational efficiency. This avoids situations where the target satellite receives too much unloaded task, leading to mission failure, or too little, resulting in low utilization of computational resources. Therefore, it solves the technical problem in existing technologies of how to predict the optimal amount of unloaded tasks from other satellites to ensure mission computational efficiency, given the limited computational resources of the satellite.

[0065] Optionally, the operation of determining the task computation volume of each satellite in the satellite constellation for the same region in the current time period includes: determining the task computation volume of each satellite in the satellite constellation for the region in each historical time period; constructing a task computation vector corresponding to each satellite based on the task computation volume of each satellite in each historical time period; and inputting each task computation vector into a pre-trained second task volume prediction model and outputting the task computation volume of each satellite for the region in the current time period.

[0066] Specifically, firstly, the data acquisition module retrieves historical time periods from the storage module. Within, the computational workload of satellite 1 for region A. Satellite 2's computational workload for region A ...; The computational workload of satellite m for region A

[0067] The data acquisition module retrieves historical time periods from the storage module. Within, the computational workload of satellite 1 for region A. Satellite 2's computational workload for region A ...; The computational workload of satellite m for region A

[0068] And so on.

[0069] The data acquisition module retrieves historical time periods from the storage module. Within, the computational workload of satellite 1 for region A. Satellite 2's computational workload for region A ...; The computational workload of satellite m for region A

[0070] The data acquisition module then sends the acquired data to the first prediction module, which in turn predicts the data based on the historical data of each satellite (1-m) during various time periods. To reduce the computational load of the task, construct a task computation vector Q corresponding to each satellite from 1 to m. 1,,1 ~Q 1,m .in,

[0071] Furthermore, the first prediction module will calculate the mission vector Q corresponding to each satellite 1 to m. 1,1 ~Q 1,m The input is fed into a pre-trained second task workload prediction model. This second task workload prediction model is a neural network model, comprising an input layer, a fully connected layer, and an output layer. Thus, the second task workload prediction model can output the task computation Q of each satellite 1 to m for region A in the current time period T2. 2,1 ~Q 2,m .

[0072] Optionally, the graph neural network model includes multiple first MLP models, and the operation of using the graph neural network model and iteratively updating the first graph structure according to a pre-defined calculation rule to generate the second graph structure includes: inputting the first graph structure into the graph neural network model, and constructing an edge attribute vector using a first calculation rule corresponding to the first edge, wherein the edge attribute vector includes the edge attribute of the first edge and the node attributes of the two first nodes adjacent to the first edge; and iteratively updating the edge attribute vector of the first edge using multiple first MLP models to generate edge features corresponding to the second graph structure.

[0073] Specifically, Figure 7 This is a model architecture diagram of the graph neural network model according to Embodiment 1 of this application. (Reference) Figure 7 As shown, the graph neural network model includes multiple first MLP models. These multiple first MLP models are used to iteratively update the edge attribute vector of the first edge in the first graph structure and generate edge features corresponding to the second graph structure.

[0074] Therefore, the iterative update module first constructs an edge attribute vector using the first calculation rule corresponding to the first edge. For example, the iterative update module constructs the first edge in the first graph structure according to the first calculation rule. (That is, the edge attribute vector corresponding to Satellite 1 receiving the offloading task from Satellite 2) in, For example, the iterative update module constructs the first edge in the first graph structure according to the first calculation rule. (That is, the edge attribute vector corresponding to Satellite 2 receiving the offloading task from Satellite 1) in, For example, the iterative update module constructs the first edge in the first graph structure according to the first calculation rule. (That is, the edge attribute vector corresponding to Satellite 2 receiving the offloading task from Satellite 3) in,

[0075] Then, the iterative update module inputs the constructed edge attribute vectors corresponding to each first edge into multiple first MLP models, thereby outputting the iteratively updated edge features corresponding to the second graph structure. For example, the edge features of the second edge corresponding to the unloading task of satellite 1 receiving satellite 2. For example, the edge features of the second edge corresponding to the unloading task from satellite 1 received by satellite 2. For example, the edge features of the second edge corresponding to the unloading task from satellite 3 received by satellite 2.

[0076] And so on.

[0077] The iterative update module can use multiple first MLP models to iteratively update the edge attribute vectors of each first edge in the first graph structure, and generate edge features corresponding to the second graph structure. The iterative update method is the same as described above, so it will not be repeated here.

[0078] Optionally, the graph neural network model includes multiple second MLP models, and the operation of using the graph neural network model to iteratively update the first graph structure according to a pre-defined calculation rule to generate the second graph structure includes: inputting the first graph structure into the graph neural network model, and constructing a node attribute vector using the second calculation rule corresponding to the first node, wherein the node attribute vector includes the node attribute of the first node, the edge attributes of the two adjacent first edges of the first node, and the node attributes of the first nodes corresponding to the two adjacent first edges respectively; and iteratively updating the node attribute vector of the first node using multiple second MLP models to generate node features corresponding to the second graph structure.

[0079] Specifically, refer to Figure 6 As shown, the graph neural network model includes multiple second MLP models. These second MLP models are used to iteratively update the node attribute vector of the first node in the first graph structure and generate node features corresponding to the second graph structure.

[0080] Therefore, the iterative update module first constructs the edge attribute vector using the second calculation rule corresponding to the first node. For example, the iterative update module constructs the edge attribute vector corresponding to the first node in the first graph structure according to the second calculation rule. (That is, the node attribute vector corresponding to the first node corresponding to satellite 1) in, For example, the iterative update module constructs the first node in the first graph structure according to the second calculation rule. (That is, the node attribute vector corresponding to the second node corresponding to satellite 2) in,

[0081] Then, the iterative update module inputs the constructed node attribute vectors corresponding to each first node into multiple second MLP models, thereby outputting the iteratively updated node features corresponding to the second graph structure. For example, the node features of the second node corresponding to satellite 1. For example, the node characteristics of the second node corresponding to satellite 2.

[0082] And so on.

[0083] The iterative update module can use multiple second MLP models to iteratively update the node attribute vectors of each first node in the first graph structure, and generate node features corresponding to the second graph structure. The iterative update method is the same as described above, so it will not be repeated here.

[0084] Optionally, the operation of inputting the target feature vector into a pre-trained first task load prediction model to predict the task unloading amount received by the target satellite includes: determining the second node in the second graph structure corresponding to the target satellite and the target feature vector corresponding to the second node, wherein the target feature vector includes the node features corresponding to the second node, the edge features of the two adjacent second edges of the second node, and the node features of the second nodes corresponding to the two adjacent second edges respectively; and inputting the target feature vector into the first task load prediction model and determining the task unloading amount corresponding to the target satellite, wherein the first task load prediction model includes an input layer, a fully connected layer, an activation function layer, and an output layer.

[0085] Specifically, Figure 8 This is a schematic diagram of the first task volume prediction model according to Embodiment 1 of this application. (Reference) Figure 8 As shown, the first task prediction model includes an input layer, a fully connected layer, an activation function layer, and an output layer.

[0086] First, the prediction module determines the second node in the second graph structure corresponding to the target satellite and the target feature vector corresponding to the second node. For example, if the target satellite is satellite 3, then the prediction module determines the second node in the second graph structure corresponding to satellite 3 and the target feature vector corresponding to the second node. Corresponding target feature vector

[0087] The prediction module then inputs the target feature vector into the first task quantity prediction model to determine the unloading task quantity that the target satellite (i.e., satellite 3) is most suitable to receive from other satellites.

[0088] Thus, according to the first aspect of this embodiment, the technical effect of being able to predict the amount of unloaded tasks from other satellites most suitable for satellite reception is achieved, thereby ensuring the computational efficiency of the task.

[0089] In addition, before applying the graph neural network model, the first task volume prediction model, and the second task volume prediction model, this application also needs to use the training module to pre-train the graph neural network model, the first task volume prediction model, and the second task volume prediction model.

[0090] The training module trains the graph neural network model, the first task prediction model, and the second task prediction model in the same way that the application phase uses the graph neural network model for feature extraction, the first task prediction model for prediction, and the second task prediction model for prediction. Therefore, it will not be elaborated here.

[0091] Furthermore, the samples collected by the training module can correspond to various sampling time periods. The format of the training sample data is the same as that of the data collected in the application phase described above, and will not be repeated here.

[0092] In addition, refer to Figure 1 As shown, according to a second aspect of this embodiment, a storage medium is provided. The storage medium includes a stored program, wherein, when the program is executed, a processor performs any of the methods described above.

[0093] Thus, according to this embodiment, the technical effect of being able to predict the amount of unloaded tasks from other satellites that are most suitable for satellite reception is achieved, thereby ensuring the computational efficiency of the task.

[0094] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0095] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0096] Example 2

[0097] Figure 9 An edge computing offloading prediction apparatus 900 based on satellite cooperation according to this embodiment is shown, which corresponds to the method described according to embodiment 1. Reference Figure 9As shown, the device 900 includes: a first task computation volume determination module 910, used to determine the satellite constellation corresponding to the target satellite, and the task computation volume of each satellite in the constellation for the same area during the current time period; a first graph structure construction module 920, used to construct a first graph structure with each satellite as the first node and the task unloading relationship between two adjacent satellites as the edge, wherein the node attributes of the first node represent the task computation volume corresponding to each satellite, and the edge attributes of the first edge represent the task unloading weight, distance weight, and attenuation coefficient weight between two adjacent satellites; a second graph structure generation module 930, used to input the first graph structure into a pre-trained graph neural network model, use the graph neural network model and iteratively update the first graph structure according to a pre-set calculation rule to generate a second graph structure; and a task unloading volume prediction module 940, used to determine the target feature vector corresponding to the target satellite in the second graph structure, and input the target feature vector into a pre-trained first task volume prediction model to predict the task unloading volume received by the target satellite.

[0098] Optionally, the first task computation volume determination module 910 includes: a second task computation volume determination module, used to determine the task computation volume of each satellite in the satellite constellation for the region under each historical period; a vector construction module, used to construct a task computation vector corresponding to each satellite based on the task computation volume of each satellite under each historical period; and a task computation volume prediction module, used to input each task computation vector into the pre-trained second task volume prediction model and output the task computation volume of each satellite for the region under the current period.

[0099] Optionally, the graph neural network model includes multiple first MLP models, and the second graph structure generation module 930 includes: an edge attribute vector construction module, used to input the first graph structure into the graph neural network model and construct an edge attribute vector using a first calculation rule corresponding to the first edge, wherein the edge attribute vector includes the edge attribute of the first edge and the node attributes of the two first nodes adjacent to the first edge; and a first iterative update module, used to iteratively update the edge attribute vector of the first edge using multiple first MLP models to generate edge features corresponding to the second graph structure.

[0100] Optionally, the graph neural network model includes multiple second MLP models, and the second graph structure generation module 930 includes: a node attribute vector construction module, used to input the first graph structure into the graph neural network model and construct a node attribute vector using a second calculation rule corresponding to the first node, wherein the node attribute vector includes the node attribute of the first node, the edge attribute of the two adjacent first edges of the first node, and the node attribute of the first node corresponding to the two adjacent first edges respectively; and a second iterative update module, used to iteratively update the node attribute vector of the first node using multiple second MLP models to generate node features corresponding to the second graph structure.

[0101] Optionally, the mission unloading prediction module 940 includes: a target feature vector determination module, used to determine the second node corresponding to the target satellite in the second graph structure and the target feature vector corresponding to the second node, wherein the target feature vector includes the node feature corresponding to the second node, the edge feature of the two adjacent second edges of the second node, and the node feature of the second node corresponding to the two adjacent second edges respectively; and a mission unloading determination module, used to input the target feature vector into the first mission load prediction model and determine the mission unloading amount corresponding to the target satellite, wherein the first mission load prediction model includes an input layer, a fully connected layer, an activation function layer, and an output layer.

[0102] Thus, according to this embodiment, the technical effect of being able to predict the amount of unloaded tasks from other satellites that are most suitable for satellite reception is achieved, thereby ensuring the computational efficiency of the task.

[0103] Example 3

[0104] Figure 10 An edge computing offloading prediction apparatus 1000 based on satellite cooperation according to this embodiment is shown, which corresponds to the method described according to Embodiment 1. Reference Figure 10As shown, the device 1000 includes: a processor 1010; and a memory 1020 connected to the processor 1010, used to provide the processor 1010 with instructions to process the following steps: determining the satellite constellation corresponding to the target satellite, and the task computation amount of each satellite in the constellation for the same area during the current time period; constructing a first graph structure with each satellite as a first node and the task offloading relationship between two adjacent satellites as edges, wherein the node attributes of the first node represent the task computation amount corresponding to each satellite, and the edge attributes of the first edge represent the task offloading weight, distance weight, and attenuation coefficient weight between two adjacent satellites; inputting the first graph structure into a pre-trained graph neural network model, using the graph neural network model and according to a pre-set calculation rule to iteratively update the first graph structure to generate a second graph structure; and determining the target feature vector corresponding to the target satellite in the second graph structure, and inputting the target feature vector into a pre-trained first task load prediction model to predict the task offloading amount received by the target satellite.

[0105] Optionally, the operation of determining the task computation volume of each satellite in the satellite constellation for the same region in the current time period includes: determining the task computation volume of each satellite in the satellite constellation for the region in each historical time period; constructing a task computation vector corresponding to each satellite based on the task computation volume of each satellite in each historical time period; and inputting each task computation vector into a pre-trained second task volume prediction model and outputting the task computation volume of each satellite for the region in the current time period.

[0106] Optionally, the graph neural network model includes multiple first MLP models, and the operation of using the graph neural network model and iteratively updating the first graph structure according to a pre-defined calculation rule to generate the second graph structure includes: inputting the first graph structure into the graph neural network model, and constructing an edge attribute vector using a first calculation rule corresponding to the first edge, wherein the edge attribute vector includes the edge attribute of the first edge and the node attributes of the two first nodes adjacent to the first edge; and iteratively updating the edge attribute vector of the first edge using multiple first MLP models to generate edge features corresponding to the second graph structure.

[0107] Optionally, the graph neural network model includes multiple second MLP models, and the operation of using the graph neural network model to iteratively update the first graph structure according to a pre-defined calculation rule to generate the second graph structure includes: inputting the first graph structure into the graph neural network model, and constructing a node attribute vector using the second calculation rule corresponding to the first node, wherein the node attribute vector includes the node attribute of the first node, the edge attributes of the two adjacent first edges of the first node, and the node attributes of the first nodes corresponding to the two adjacent first edges respectively; and iteratively updating the node attribute vector of the first node using multiple second MLP models to generate node features corresponding to the second graph structure.

[0108] Optionally, the operation of inputting the target feature vector into a pre-trained first task load prediction model to predict the task unloading amount received by the target satellite includes: determining the second node in the second graph structure corresponding to the target satellite and the target feature vector corresponding to the second node, wherein the target feature vector includes the node features corresponding to the second node, the edge features of the two adjacent second edges of the second node, and the node features of the second nodes corresponding to the two adjacent second edges respectively; and inputting the target feature vector into the first task load prediction model and determining the task unloading amount corresponding to the target satellite, wherein the first task load prediction model includes an input layer, a fully connected layer, an activation function layer, and an output layer.

[0109] Thus, according to this embodiment, the technical effect of being able to predict the amount of unloaded tasks from other satellites that are most suitable for satellite reception is achieved, thereby ensuring the computational efficiency of the task.

[0110] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0111] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0112] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0113] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0114] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0115] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0116] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for edge computing offloading prediction based on satellite cooperation, characterized in that, include: Determine the satellite constellation corresponding to the target satellite, and the computational workload of each satellite in the constellation for the same area during the current time period; A first graph structure is constructed with each satellite as the first node and the task offloading relationship between two adjacent satellites as the edge. The node attributes of the first node represent the task computation amount corresponding to each satellite, and the edge attributes of the first edge represent the task offloading weight, distance weight, and attenuation coefficient weight between two adjacent satellites. The first graph structure is input into a pre-trained graph neural network model, and the first graph structure is iteratively updated using the graph neural network model and according to a pre-set calculation rule to generate a second graph structure. as well as The target feature vector corresponding to the target satellite in the second graph structure is determined, and the target feature vector is input into the pre-trained first task load prediction model to predict the task unloading amount received by the target satellite. The target feature vector includes the node feature corresponding to the second node, the edge feature of the two adjacent second edges of the second node, and the node feature of the second node corresponding to the two adjacent second edges.

2. The method according to claim 1, characterized in that, The operations for determining the computational workload of each satellite in the satellite constellation for the same region during the current time period include: Determine the computational workload of each satellite in the satellite constellation for the region during each historical period; Based on the computational workload of each satellite in each historical time period, a corresponding computational vector is constructed for each satellite; and Each task calculation vector is input into a pre-trained second task quantity prediction model, and the task calculation quantity of each satellite for the region during the current time period is output.

3. The method according to claim 1, characterized in that, The graph neural network model includes multiple first MLP models, and the operation of generating a second graph structure by iteratively updating the first graph structure using the graph neural network model according to a pre-set calculation rule includes: The first graph structure is input into the graph neural network model, and an edge attribute vector is constructed using the first calculation rule corresponding to the first edge. The edge attribute vector includes the edge attribute of the first edge and the node attributes of the two first nodes adjacent to the first edge. The edge attribute vector of the first edge is iteratively updated using the multiple first MLP models to generate edge features corresponding to the second graph structure.

4. The method according to claim 1, characterized in that, The graph neural network model includes multiple second MLP models, and the operation of generating a second graph structure by iteratively updating the first graph structure using the graph neural network model according to a pre-defined calculation rule includes: The first graph structure is input into the graph neural network model, and a node attribute vector is constructed using the second calculation rule corresponding to the first node. The node attribute vector includes the node attribute of the first node, the edge attributes of the two adjacent first edges of the first node, and the node attributes of the first nodes corresponding to the two adjacent first edges. The node attribute vector of the first node is iteratively updated using the multiple second MLP models to generate node features corresponding to the second graph structure.

5. The method according to claim 1, characterized in that, The operation of inputting the target feature vector into a pre-trained first task load prediction model to predict the task unloading amount received by the target satellite includes: Determine the second node in the second graph structure corresponding to the target satellite and the target feature vector corresponding to the second node; and The target feature vector is input into the first task load prediction model to determine the task unloading amount corresponding to the target satellite. The first task load prediction model includes an input layer, a fully connected layer, an activation function layer, and an output layer.

6. A storage medium, characterized in that, The storage medium includes a stored program, wherein, when the program is executed, the method described in any one of claims 1 to 5 is performed by a processor.

7. An edge computing offloading prediction device based on satellite collaboration, characterized in that, include: The first task computation volume determination module is used to determine the satellite constellation corresponding to the target satellite, and the task computation volume of each satellite in the satellite constellation for the same area in the current time period; The first graph structure construction module is used to construct a first graph structure with each satellite as the first node and the task unloading relationship between two adjacent satellites as the edge. The node attributes of the first node represent the task computation amount corresponding to each satellite, and the edge attributes of the first edge represent the task unloading weight, distance weight, and attenuation coefficient weight between two adjacent satellites. The second graph structure generation module is used to input the first graph structure into a pre-trained graph neural network model, and use the graph neural network model to iteratively update the first graph structure according to a pre-set calculation rule to generate the second graph structure. as well as The task unloading volume prediction module is used to determine the target feature vector corresponding to the target satellite in the second graph structure, and input the target feature vector into the pre-trained first task volume prediction model to predict the task unloading volume received by the target satellite. The target feature vector includes the node feature corresponding to the second node, the edge feature of the two adjacent second edges of the second node, and the node feature of the second node corresponding to the two adjacent second edges.

8. The apparatus according to claim 7, characterized in that, The first task computation amount determination module includes: The second task computation volume determination module is used to determine the task computation volume of each satellite in the satellite constellation for the region under each historical period. The vector construction module is used to construct a task computation vector corresponding to each satellite based on the task computation volume of each satellite in each historical time period; and The task computation volume prediction module is used to input each task computation vector into the pre-trained second task volume prediction model and output the task computation volume of each satellite for the region in the current time period.

9. The apparatus according to claim 7, characterized in that, The graph neural network model includes multiple first MLP models, and the second graph structure generation module includes: An edge attribute vector construction module is used to input the first graph structure into the graph neural network model and construct an edge attribute vector using a first calculation rule corresponding to the first edge, wherein the edge attribute vector includes the edge attribute of the first edge and the node attributes of the two first nodes adjacent to the first edge; and The first iterative update module is used to iteratively update the edge attribute vector of the first edge using the multiple first MLP models to generate edge features corresponding to the second graph structure.

10. An edge computing offloading prediction device based on satellite collaboration, characterized in that, include: processor; as well as A memory, connected to the processor, for providing the processor with instructions to perform the following processing steps: Determine the satellite constellation corresponding to the target satellite, and the computational workload of each satellite in the constellation for the same area during the current time period; A first graph structure is constructed with each satellite as the first node and the task offloading relationship between two adjacent satellites as the edge. The node attributes of the first node represent the task computation amount corresponding to each satellite, and the edge attributes of the first edge represent the task offloading weight, distance weight, and attenuation coefficient weight between two adjacent satellites. The first graph structure is input into a pre-trained graph neural network model, and the first graph structure is iteratively updated using the graph neural network model and according to a pre-set calculation rule to generate a second graph structure. And determine the target feature vector corresponding to the target satellite in the second graph structure, and input the target feature vector into the pre-trained first task volume prediction model to predict the task unloading amount received by the target satellite, wherein the target feature vector includes the node feature corresponding to the second node, the edge feature of the two adjacent second edges of the second node, and the node feature of the second node corresponding to the two adjacent second edges respectively.

Images