Satellite-Ground Computing Network Resource Allocation Methods, Devices, Equipment, Media, and Program Products

By acquiring the multi-dimensional intent vectors of intelligent agents, calculating channel gain and task latency, and performing resource configuration parsing and dual-transmission selective reception processing, the transmission instability during cross-domain link cutover is resolved, and dynamic matching and continuity of satellite-ground computing network resources are realized.

CN123123955APending Publication Date: 2026-11-17CHINA UNITED NETWORK COMM GRP CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611484464.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-22
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies can cause sudden changes in transmission delay and batch packet loss during cross-domain link cutover, making it difficult to guarantee the transmission determinism and operational stability of delay-sensitive services.

Method used

By acquiring the multi-dimensional intent vectors of intelligent agents, calculating the channel gain of ground and satellite communication links and the task processing latency of edge computing nodes, performing periodic resource allocation parsing, realizing the coordinated allocation of satellite-ground communication resources and edge computing resources, and performing dual-transmission selection and latency compensation processing during link switching, updating the resource allocation matrix to adapt to the actual transmission status.

Benefits of technology

It achieves continuity and determinism in the transmission process before and after link switching, eliminates the latency abrupt changes and packet loss problems caused by traditional switching methods, and ensures the matching and stability of business requirements.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123123955A_ABST
    Figure CN123123955A_ABST
Patent Text Reader

Abstract

This application provides a method, apparatus, device, medium, and program product for configuring satellite-to-ground computing network resources, relating to the field of data processing technology. It is used for configuring satellite-to-ground computing network resources, and the specific technical solution is as follows: Obtaining multi-dimensional intent vectors corresponding to intelligent entities within a target area; calculating the channel gain of the ground communication link and satellite communication link based on the multi-dimensional intent vectors, and calculating the task processing delay of the edge computing node; performing periodic resource configuration parsing based on communication constraint parameters, control state parameters, channel gain, and task processing delay to obtain a resource configuration matrix; performing dual-transmission selection and delay compensation processing on the ground communication link and satellite communication link to obtain measured transmission delay and measured jitter data; updating the resource configuration matrix based on the measured transmission delay and measured jitter data; and configuring communication resources and edge computing resources within the satellite-to-ground computing network based on the updated resource configuration matrix. This is applied to satellite-to-ground computing network resource configuration scenarios.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method, apparatus, equipment, medium and program product for configuring satellite-to-ground computing network resources. Background Technology

[0002] As smart parks evolve towards high automation and unmanned operation, the diverse intelligent entities within the parks, such as Automated Guided Vehicles (AGVs) and embodied intelligent robots, place stringent demands on the low latency, high reliability, and business continuity of network connections. Integrating 5G terrestrial networks with low Earth Orbit (LEO) satellite networks and leveraging multi-access edge computing (MEC) nodes to handle park services locally has become the mainstream networking approach in this scenario.

[0003] A representative implementation scheme is as follows: The campus deploys satellite terminals on the gateway side as disaster recovery links. Daily services are carried by terrestrial 5G micro base stations, and edge computing nodes provide data processing services to each terminal according to pre-set static policies. The system continuously monitors the signal quality of each terminal. When the measured signal-to-noise ratio (SNR) falls below a preset threshold, link redirection is immediately triggered, switching the terminal's services entirely to the satellite link. After the switchover, the satellite link is used only as a transparent transmission channel; the communication and computing sides operate independently, unaware of each other's resource status and load changes.

[0004] However, the above scheme adopts the switching method of "breaking the old link first and then starting the new link", which inevitably causes sudden changes in transmission delay and batch packet loss during cross-domain link cutover, making it difficult to effectively guarantee the transmission determinism and operational stability of delay-sensitive services. Summary of the Invention

[0005] This application provides a method, apparatus, device, medium, and program product for configuring satellite-to-ground computing network resources, which solves the problem that existing methods inevitably cause sudden changes in transmission delay and batch packet loss during cross-domain link cutover, making it difficult to effectively guarantee the transmission determinism and operational stability of delay-sensitive services.

[0006] In a first aspect, embodiments of this application provide a method for configuring satellite-to-ground computing network resources, the method comprising:

[0007] Obtain the multi-dimensional intent vectors corresponding to intelligent agents within the target area. The multi-dimensional intent vectors include three-dimensional spatial coordinates, communication constraint parameters, computational requirement parameters, and control state parameters.

[0008] Based on the aforementioned multivariate intent vectors, the channel gain of the terrestrial communication link and the satellite communication link are calculated respectively, as well as the task processing latency of the edge computing node is calculated.

[0009] Based on the above communication constraint parameters, the above control state parameters, the above channel gain and the above task processing delay, periodic resource configuration analysis is performed to obtain a resource configuration matrix. The above resource configuration matrix is ​​used to control the coordinated configuration of communication resources and edge computing resources of the satellite-ground computing network in the target area. The above resource configuration matrix includes link offloading decision, bandwidth allocation result, computing power allocation result and transmission power result.

[0010] When the above-mentioned link offloading decision instructs the above-mentioned intelligent subject to switch the communication link between the above-mentioned ground communication link and the above-mentioned satellite communication link, dual transmission and reception and delay compensation processing are performed on the above-mentioned ground communication link and the above-mentioned satellite communication link to obtain the measured transmission delay and measured jitter data.

[0011] Update the resource configuration matrix based on the measured transmission latency and jitter data.

[0012] Configure the above-mentioned satellite-ground computing network communication resources and edge computing resources based on the updated resource configuration matrix.

[0013] The technical solution provided in this application brings at least the following beneficial effects: By acquiring the multi-dimensional intent vectors corresponding to intelligent agents within the target area, which include three-dimensional spatial coordinates, communication constraint parameters, computational requirement parameters, and control state parameters, the business requirements of intelligent agents in the three dimensions of communication, computation, and control are parameterized and structured, providing a unified requirement input for subsequent resource allocation and avoiding coarse-grained perception of business requirements; Based on this, the channel gain of the ground communication link and satellite communication link is calculated based on the above multi-dimensional intent vectors, and the task processing delay of the edge computing node is calculated, that is, the satellite-ground channel state and computing queue state of the next time slot are jointly predicted, so that the resource solution is based on the real-time state of the physical layer and the computing layer, improving the matching accuracy between the solution results and the actual operating environment; Furthermore, based on the above communication constraint parameters, the above control state parameters, the above channel gain, and the above task processing delay, periodic resource allocation analysis is performed to obtain a resource allocation matrix including link offloading decision, bandwidth allocation result, computing power allocation result, and transmission power result, realizing the integration of satellite-ground communication resources and edge computing resources. The coordinated allocation of resources ensures that resource allocation decisions are simultaneously constrained by service latency tolerance and control stability requirements, guaranteeing the match between resource allocation results and the intelligent agent's service needs. When the aforementioned link offloading decision instructs the intelligent agent to switch communication links between terrestrial and satellite links, dual-transmission and selective reception with latency compensation are performed on both terrestrial and satellite communication links. This ensures that during the switchover, both terrestrial and satellite links simultaneously carry data and achieve seamless connection through deduplication and latency compensation. This eliminates the inevitable transmission latency spikes and batch packet losses during cross-domain link cutovers caused by the traditional "break-before-send" method. In this process, measured transmission latency and jitter data are obtained. Finally, based on the measured transmission latency and jitter data, the resource allocation matrix is ​​updated, and the communication resources and edge computing resources within the satellite-terrestrial computing network are configured based on the updated matrix. This allows the resource allocation to be dynamically adjusted according to the actual transmission status of the links, ensuring that the resource allocation continuously matches the actual link status and guaranteeing the continuity and determinism of the intelligent agent's transmission process before and after the communication link switchover.

[0014] One possible implementation, whereby updating the resource configuration matrix based on the measured transmission delay and the measured jitter data includes:

[0015] Based on the measured transmission delay and jitter data, the pruning factor of the inference model in the edge computing node is obtained and the control sampling period of the intelligent agent is adjusted. The resource configuration matrix is ​​updated according to the pruning factor and the adjusted control sampling period to obtain the updated resource configuration matrix. The communication resources and edge computing resources within the satellite-ground computing network are then reconfigured based on the updated resource configuration matrix.

[0016] Another possible implementation, as described above, involves obtaining the multi-dimensional intent vectors corresponding to intelligent agents within the target area, including:

[0017] Collect the three-dimensional spatial coordinates, velocity vector, maximum tolerable latency, minimum guaranteed rate, single task computation amount, upper limit of control instability deviation, and service priority identifier of the intelligent subject;

[0018] The communication constraint parameters are constructed based on the maximum tolerable latency and the minimum guaranteed rate of the above services. The computational requirement parameters are constructed based on the computational amount of a single task. The control state parameters are constructed based on the velocity vector, the state error and the network jitter.

[0019] By combining the aforementioned three-dimensional spatial coordinates, communication constraint parameters, computational requirement parameters, and control state parameters according to preset dimensions, a structured multi-dimensional intent vector is obtained.

[0020] Another possible implementation involves calculating the channel gain of the terrestrial communication link and the satellite communication link based on the aforementioned multi-dimensional intent vectors, and calculating the task processing latency of the edge computing node, including:

[0021] Based on the aforementioned multi-intention vectors and ground base station coordinates, combined with reference path loss, path loss index, and shadow fading based on measured channel sample statistics, the channel gain of the aforementioned ground communication link relative to the aforementioned multi-agent is calculated.

[0022] Based on the aforementioned multi-intention vectors and satellite ephemeris data, combined with carrier frequency, satellite-to-ground tilt distance, and Rice factor modulated by elevation angle, the channel gain of the aforementioned satellite communication link relative to the aforementioned multi-intelligence agent is calculated using the Rice model.

[0023] Based on the number of cores, task arrival rate, and service rate of the aforementioned edge computing nodes, the queuing probability is calculated. Based on the queuing probability, the waiting latency and processing latency of the aforementioned intelligent subject's task on the edge computing nodes are calculated. Based on the waiting latency and processing latency, the task processing latency is determined.

[0024] Another possible implementation involves performing periodic resource allocation analysis based on the aforementioned communication constraint parameters, control state parameters, channel gain, and task processing delay to obtain a resource allocation matrix, including:

[0025] Decision variables are constructed based on link offload ratio, bandwidth allocation value, computing power allocation value, and transmit power value;

[0026] A constant matrix is ​​constructed based on the aforementioned channel gain and task processing delay.

[0027] Based on the maximum tolerable delay of the above-mentioned service in the above-mentioned communication constraint parameters, a delay constraint is constructed; based on the minimum guaranteed rate in the above-mentioned communication constraint parameters, a rate constraint is set; and based on the upper bound of the control instability deviation in the above-mentioned control state parameters, a control stability constraint is constructed.

[0028] Under the constraints of the aforementioned time delay, rate, and control stability, periodic resource allocation analysis is performed on the aforementioned decision variables and constant matrix to obtain the aforementioned resource allocation matrix.

[0029] Another possible implementation involves performing dual-transmission selective reception and delay compensation processing on the aforementioned ground communication link and satellite communication link to obtain measured transmission delay and measured jitter data, including:

[0030] The data packets carrying the incrementing sequence number are sent synchronously along the aforementioned ground communication link and the aforementioned satellite communication link, resulting in two arriving data packets;

[0031] Based on the time synchronization protocol, clock synchronization is performed on the two incoming data packets to obtain the propagation delay difference between the two links, and a jitter compensation window is set based on the propagation delay difference.

[0032] Within the jitter compensation window mentioned above, deduplication is performed on the data packets arriving from both links, retaining the first valid data packet that arrives.

[0033] The measured transmission delay and jitter data were calculated based on the start and arrival times of the end-to-end transmission of the aforementioned valid data packets.

[0034] Another possible implementation involves obtaining the pruning factor of the inference model in the edge computing node based on the measured transmission delay and jitter data, adjusting the control sampling period of the intelligent agent, and updating the resource configuration matrix according to the pruning factor and the adjusted control sampling period to obtain the updated resource configuration matrix, including:

[0035] When the measured transmission delay exceeds the maximum tolerable delay of the service and there are no adjustable communication resources, the pruning factor is calculated based on the delay excess, the current remaining computing power of the edge computing node, and the current number of channels in the above inference model.

[0036] The channels of the above inference model are pruned according to the above pruning factor to obtain the pruned model, which requires less computation.

[0037] Based on the measured jitter data, the control sampling period predicted by the above-mentioned intelligent subject execution model is increased by a preset step size to obtain the above-mentioned adjusted control sampling period.

[0038] Based on the computational cost of the trimmed model and the adjusted control sampling period, the resource configuration matrix is ​​updated to obtain the updated resource configuration matrix.

[0039] Another possible implementation involves constructing the control state parameters based on the aforementioned velocity vector, state error, and network jitter, including:

[0040] Obtain the state error vector and control input vector of the intelligent agent. The state error vector represents the deviation between the actual motion trajectory and the desired motion trajectory.

[0041] By introducing a positive definite weight matrix, quadratic form calculations are performed on the aforementioned state error vector and the aforementioned control input vector respectively to obtain the basic loss of control instability.

[0042] Based on the above network jitter and the jitter penalty term coefficients corresponding to the above network jitter, a jitter penalty term is constructed. The jitter penalty term is then fused with the above control instability basic loss to obtain the second-order control instability loss function.

[0043] The above-mentioned second-order control instability loss function and the above-mentioned upper bound of control instability deviation are used to construct the above-mentioned control state parameters.

[0044] Secondly, embodiments of this application provide a satellite-to-ground computing network resource configuration device, comprising:

[0045] The acquisition module is used to acquire the multi-dimensional intent vectors corresponding to the intelligent agents within the target area. The multi-dimensional intent vectors include three-dimensional spatial coordinates, communication constraint parameters, computational requirement parameters, and control state parameters.

[0046] The calculation module is used to calculate the channel gain of the terrestrial communication link and the satellite communication link based on the above multi-dimensional intent vectors, and to calculate the task processing latency of the edge computing node.

[0047] The parsing module is used to perform periodic resource configuration parsing based on the above-mentioned communication constraint parameters, control state parameters, channel gain and task processing delay to obtain a resource configuration matrix. The resource configuration matrix is ​​used to control the coordinated configuration of communication resources and edge computing resources of the satellite-ground computing network in the target area. The resource configuration matrix includes link offloading decision, bandwidth allocation result, computing power allocation result and transmission power result.

[0048] The processing module is used to perform dual-transmission and reception selection and delay compensation processing on the ground communication link and the satellite communication link when the above-mentioned link offloading decision instructs the intelligent subject to switch the communication link between the above-mentioned ground communication link and the above-mentioned satellite communication link, so as to obtain the measured transmission delay and measured jitter data.

[0049] The update module is used to update the resource configuration matrix based on the measured transmission delay and the measured jitter data.

[0050] The configuration module is used to configure the communication resources and edge computing resources within the aforementioned satellite-to-ground computing network based on the updated resource configuration matrix.

[0051] Another possible implementation, the aforementioned update module is also used for:

[0052] Based on the measured transmission delay and jitter data, the pruning factor of the inference model in the edge computing node is obtained and the control sampling period of the intelligent agent is adjusted. The resource configuration matrix is ​​updated according to the pruning factor and the adjusted control sampling period to obtain the updated resource configuration matrix. The communication resources and edge computing resources within the satellite-ground computing network are then reconfigured based on the updated resource configuration matrix.

[0053] Another possible implementation is that the above-mentioned acquisition module is specifically used for:

[0054] Collect the three-dimensional spatial coordinates, velocity vector, maximum tolerable latency, minimum guaranteed rate, single task computation amount, upper limit of control instability deviation, and service priority identifier of the intelligent subject;

[0055] The communication constraint parameters are constructed based on the maximum tolerable latency and the minimum guaranteed rate of the above services. The computational requirement parameters are constructed based on the computational amount of a single task. The control state parameters are constructed based on the velocity vector, the state error and the network jitter.

[0056] By combining the aforementioned three-dimensional spatial coordinates, communication constraint parameters, computational requirement parameters, and control state parameters according to preset dimensions, a structured multi-dimensional intent vector is obtained.

[0057] One possible implementation is that the above-mentioned computing module is specifically used for:

[0058] Based on the aforementioned multi-intention vectors and ground base station coordinates, combined with reference path loss, path loss index, and shadow fading based on measured channel sample statistics, the channel gain of the aforementioned ground communication link relative to the aforementioned multi-agent is calculated.

[0059] Based on the aforementioned multi-intention vectors and satellite ephemeris data, combined with carrier frequency, satellite-to-ground tilt distance, and Rice factor modulated by elevation angle, the channel gain of the aforementioned satellite communication link relative to the aforementioned multi-intelligence agent is calculated using the Rice model.

[0060] Based on the number of cores, task arrival rate, and service rate of the aforementioned edge computing nodes, the queuing probability is calculated. Based on the queuing probability, the waiting latency and processing latency of the aforementioned intelligent subject's task on the edge computing nodes are calculated. Based on the waiting latency and processing latency, the task processing latency is determined.

[0061] Another possible implementation, the parsing module described above is specifically used for:

[0062] Decision variables are constructed based on link offload ratio, bandwidth allocation value, computing power allocation value, and transmit power value;

[0063] A constant matrix is ​​constructed based on the aforementioned channel gain and task processing delay.

[0064] Based on the maximum tolerable delay of the above-mentioned service in the above-mentioned communication constraint parameters, a delay constraint is constructed; based on the minimum guaranteed rate in the above-mentioned communication constraint parameters, a rate constraint is set; and based on the upper bound of the control instability deviation in the above-mentioned control state parameters, a control stability constraint is constructed.

[0065] Under the constraints of the aforementioned time delay, rate, and control stability, periodic resource allocation analysis is performed on the aforementioned decision variables and constant matrix to obtain the aforementioned resource allocation matrix.

[0066] Another possible implementation, in which the above processing module is specifically used for:

[0067] The data packets carrying the incrementing sequence number are sent synchronously along the aforementioned ground communication link and the aforementioned satellite communication link, resulting in two arriving data packets;

[0068] Based on the time synchronization protocol, clock synchronization is performed on the two incoming data packets to obtain the propagation delay difference between the two links, and a jitter compensation window is set based on the propagation delay difference.

[0069] Within the jitter compensation window mentioned above, deduplication is performed on the data packets arriving from both links, retaining the first valid data packet that arrives.

[0070] The measured transmission delay and jitter data were calculated based on the start and arrival times of the end-to-end transmission of the aforementioned valid data packets.

[0071] Another possible implementation is that the above-mentioned update module is specifically used for:

[0072] When the measured transmission delay exceeds the maximum tolerable delay of the service and there are no adjustable communication resources, the pruning factor is calculated based on the delay excess, the current remaining computing power of the edge computing node, and the current number of channels in the above inference model.

[0073] The channels of the above inference model are pruned according to the above pruning factor to obtain the pruned model, which requires less computation.

[0074] Based on the measured jitter data, the control sampling period predicted by the above-mentioned intelligent subject execution model is increased by a preset step size to obtain the above-mentioned adjusted control sampling period.

[0075] Based on the computational cost of the trimmed model and the adjusted control sampling period, the resource configuration matrix is ​​updated to obtain the updated resource configuration matrix.

[0076] Another possible implementation is that the above-mentioned acquisition module is specifically used for:

[0077] Obtain the state error vector and control input vector of the intelligent agent. The state error vector represents the deviation between the actual motion trajectory and the desired motion trajectory.

[0078] By introducing a positive definite weight matrix, quadratic form calculations are performed on the aforementioned state error vector and the aforementioned control input vector respectively to obtain the basic loss of control instability.

[0079] Based on the above network jitter and the jitter penalty term coefficients corresponding to the above network jitter, a jitter penalty term is constructed. The jitter penalty term is then fused with the above control instability basic loss to obtain the second-order control instability loss function.

[0080] The above-mentioned second-order control instability loss function and the above-mentioned upper bound of control instability deviation are used to construct the above-mentioned control state parameters.

[0081] Thirdly, this application provides an electronic device comprising: a processor and a memory; the memory stores a program or instructions executable on the processor, wherein the program or instructions, when executed by the processor, implement the method of the first aspect described above.

[0082] Fourthly, this application provides a readable storage medium on which a program or instructions are stored, which, when executed by a computer, implement the method of the first aspect described above.

[0083] Fifthly, this application provides a computer program product stored in a storage medium, which, when executed by a computer, implements the method described in the first aspect.

[0084] In a sixth aspect, embodiments of this application provide a chip including a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run programs or instructions to implement the method described in the first aspect.

[0085] The beneficial effects of the second to sixth aspects mentioned above are described in the corresponding description of the first aspect and will not be repeated here. Attached Figure Description

[0086] Figure 1 A schematic diagram of the network architecture for an application of a satellite-to-ground computing network resource allocation method provided in this embodiment of the application;

[0087] Figure 2 A flowchart illustrating a satellite-to-ground computing network resource allocation method provided in an embodiment of this application;

[0088] Figure 3 A flowchart illustrating another method for configuring satellite-to-ground computing network resources provided in this application embodiment;

[0089] Figure 4 A flowchart illustrating another method for configuring satellite-to-ground computing network resources provided in this application embodiment;

[0090] Figure 5 A system overall hierarchical architecture diagram provided for embodiments of this application;

[0091] Figure 6 A flowchart of 3C bidirectional closed-loop and collaborative control is provided for embodiments of this application;

[0092] Figure 7 A schematic diagram of a satellite-to-ground dual-transmission selective reception (PRE) mechanism based on DetNet provided for embodiments of this application;

[0093] Figure 8 This is a schematic diagram of the structure of a satellite-to-ground computing network resource allocation device provided in an embodiment of this application;

[0094] Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0095] The following will describe in detail, with reference to the accompanying drawings, the satellite-to-ground computing network resource configuration method, apparatus, equipment, media, and program products provided in this application.

[0096] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0097] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0098] The terms "at least one," "at least one of," etc., used in the specification and claims of this application refer to any one, any two, or a combination of two or more of the included items. For example, at least one of a, b, and c can mean: "a," "b," "c," "a and b," "a and c," "b and c," and "a, b, and c," where a, b, and c can be single or multiple. Similarly, "at least two" refers to two or more items, and its meaning is similar to that of "at least one."

[0099] In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0100] The embodiments of this application provide a method, apparatus, device, medium, and program product for configuring satellite-to-ground computing network resources, which can be applied to satellite-to-ground computing network resource configuration scenarios.

[0101] A representative implementation scheme is as follows: The campus deploys satellite terminals on the gateway side as disaster recovery links. Daily services are carried by terrestrial 5G micro base stations, and edge computing nodes provide data processing services to each terminal according to pre-set static policies. The system continuously monitors the signal quality of each terminal. When the measured signal-to-noise ratio (SNR) falls below a preset threshold, link redirection is immediately triggered, switching the terminal's services entirely to the satellite link. After the switchover, the satellite link is used only as a transparent transmission channel; the communication and computing sides operate independently, unaware of each other's resource status and load changes.

[0102] However, the above scheme adopts the switching method of "breaking the old link first and then starting the new link", which inevitably causes sudden changes in transmission delay and batch packet loss during cross-domain link cutover, making it difficult to effectively guarantee the transmission determinism and operational stability of delay-sensitive services.

[0103] To address the aforementioned technical problems, this application provides a method, apparatus, device, medium, and program product for configuring satellite-ground computing network resources. By acquiring multi-dimensional intent vectors corresponding to intelligent entities within a target area—including three-dimensional spatial coordinates, communication constraint parameters, computing requirement parameters, and control state parameters—the service requirements of intelligent entities in the three dimensions of communication, computing, and control are parameterized and structured, providing a unified requirement input for subsequent resource configuration and avoiding coarse-grained perception of service requirements. Based on this, the channel gain of the ground communication link and satellite communication link is calculated based on the aforementioned multi-dimensional intent vectors, as well as the task processing delay of edge computing nodes. This involves jointly predicting the satellite-ground channel state and computing queue state for the next time slot, ensuring that resource calculation is based on the real-time states of the physical and computing layers, improving the matching accuracy between the calculation results and the actual operating environment. Furthermore, based on the aforementioned communication constraint parameters, control state parameters, channel gain, and task processing delay, periodic resource configuration analysis is performed to obtain a resource configuration matrix including link offloading decisions, bandwidth allocation results, computing power allocation results, and transmit power results, thereby achieving… The coordinated allocation of satellite-to-ground communication resources and edge computing resources ensures that resource allocation decisions are simultaneously constrained by service latency tolerance and control stability requirements, guaranteeing the match between resource allocation results and the intelligent agent's service needs. When the aforementioned link offloading decision instructs the intelligent agent to switch between ground and satellite communication links, dual-transmission and selective reception with latency compensation processing is performed on both ground and satellite communication links. This ensures that during the switchover, both ground and satellite links simultaneously carry data and achieve seamless connection through deduplication and latency compensation, eliminating the traditional "first-disconnect, then-receive" transmission mechanism. During cross-domain link cutover, the switching method inevitably causes sudden changes in transmission delay and batch packet loss, and in this process, measured transmission delay and jitter data are obtained. Finally, based on the measured transmission delay and jitter data, the resource configuration matrix is ​​updated, and the communication resources and edge computing resources within the satellite-ground computing network are configured based on the updated resource configuration matrix. This allows the resource configuration to be dynamically corrected according to the actual transmission status of the link, ensuring that the resource configuration continuously matches the real state of the link and guaranteeing the continuity and determinism of the transmission process of the intelligent subject before and after the communication link switchover.

[0104] The following description, in conjunction with the accompanying drawings, details the satellite-to-ground computing network resource configuration method, apparatus, equipment, medium, and program products provided in the embodiments of this application.

[0105] Figure 1 This illustration shows the network architecture of a satellite-to-ground computing network resource configuration method provided in an embodiment of this application. For example... Figure 1As shown, the network architecture includes a satellite-to-ground computing network resource configuration device 101 and a terminal device 102. The satellite-to-ground computing network resource configuration device 101 and the terminal device 102 are interconnected.

[0106] In some embodiments, the satellite-to-ground computing network resource configuration device 101 may be a server, a computer, or a processor or processing unit within a server or computer. The server may be a single server or a server cluster consisting of multiple servers. It should be noted that this application embodiment does not limit the specific device form of the satellite-to-ground computing network resource configuration device 101. Figure 1 The example shown is a single server using the satellite-to-ground computing network resource configuration device 101.

[0107] In some embodiments, the terminal device 102 may be a mobile phone, tablet computer, laptop computer, handheld computer, in-vehicle electronic device, mobile internet device (MID), augmented reality (AR) / virtual reality (VR) device, robot, wearable device, personal computer (PC), ultra-mobile personal computer (UMPC), netbook, or personal digital assistant (PDA), etc., and the embodiments of this application do not specifically limit it. Figure 1 The example shown is a mobile phone, with terminal device 102 as an example.

[0108] In some embodiments, the satellite-to-ground computing network resource configuration device 101 is used to acquire multi-dimensional intent vectors corresponding to intelligent agents within a target area. These multi-dimensional intent vectors include three-dimensional spatial coordinates, communication constraint parameters, computational requirement parameters, and control state parameters. Based on these multi-dimensional intent vectors, the device calculates the channel gain of the ground communication link and the satellite communication link, and calculates the task processing delay of the edge computing node. Based on the communication constraint parameters, the control state parameters, the channel gain, and the task processing delay, the device performs periodic resource configuration parsing to obtain a resource configuration matrix. This resource configuration matrix is ​​used to control the communication resources of the satellite-to-ground computing network within the target area. The resource configuration matrix, which coordinates the allocation of edge computing resources, includes link offloading decisions, bandwidth allocation results, computing power allocation results, and transmission power results. When the link offloading decision instructs the intelligent agent to switch communication links between the ground communication link and the satellite communication link, dual-transmission selective reception and delay compensation processing is performed on the ground communication link and the satellite communication link to obtain measured transmission delay and measured jitter data. Based on the measured transmission delay and measured jitter data, the resource configuration matrix is ​​updated. Based on the updated resource configuration matrix, communication resources and edge computing power resources within the satellite-ground computing network are configured. Terminal device 102 is used to monitor the operating status of the satellite-ground computing network resource configuration device 101.

[0109] It should be noted that the network architecture described in the embodiments of this application is for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and does not constitute a limitation on the technical solutions provided in the embodiments of this application. As network architectures evolve, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0110] See Figure 2 This is a flowchart illustrating a satellite-to-ground computing network resource configuration method provided in an embodiment of this application. Figure 2 As shown in the figure, the satellite-to-ground computing network resource configuration method provided in this application embodiment specifically includes the following steps 201 to 206.

[0111] Step 201: The satellite-ground computing network resource configuration device acquires the multi-dimensional intent vector corresponding to the intelligent subject within the target area. The multi-dimensional intent vector includes three-dimensional spatial coordinates, communication constraint parameters, computing requirement parameters, and control state parameters.

[0112] In some embodiments, the satellite-to-ground computing network resource configuration device collects the three-dimensional spatial coordinates, velocity vector, maximum tolerable latency for services, minimum guaranteed rate, single task computation amount, upper bound of control instability deviation, and service priority identifier of the intelligent subject; constructs communication constraint parameters based on the maximum tolerable latency for services and the minimum guaranteed rate, constructs computation requirement parameters based on the single task computation amount, and constructs control state parameters based on the velocity vector, state error, and network jitter; and combines the three-dimensional spatial coordinates, communication constraint parameters, computation requirement parameters, and control state parameters according to preset dimensions to obtain a structured multi-dimensional intent vector.

[0113] In some embodiments, three-dimensional spatial coordinates represent the real-time spatial position of the intelligent subject within the target area, which can be obtained through the positioning information reported by the intelligent subject; velocity vector represents the displacement components of the intelligent subject in each coordinate axis direction in three-dimensional space per unit time, used to reflect the movement status of the intelligent subject; maximum tolerable latency of services represents the longest time allowed for the services carried by the intelligent subject from issuance to completion; minimum guaranteed rate represents the lower limit of the rate that the intelligent subject's service transmission needs to reach; single task computation represents the number of floating-point operations required by the intelligent subject to issue a single task, in units of floating-point operations per second (FLOPs); upper bound of control instability deviation represents the maximum deviation that the intelligent subject's control process can withstand; service priority identifier represents the order of occupation of services carried by the intelligent subject during resource contention, for example, it can be classified into levels according to embodied control, high perception and conventional IoT; state error represents the deviation between the actual movement trajectory and the expected movement trajectory of the intelligent subject; network jitter represents the degree of variation in the arrival time of data packets.

[0114] In some embodiments, the process of constructing control state parameters is as follows: The state error vector and control input vector of the intelligent agent are obtained, whereby the state error vector represents the deviation between the actual trajectory and the desired trajectory. A positive definite weight matrix is ​​introduced to perform quadratic form calculations on the state error vector and control input vector respectively, yielding the basic loss of control instability. A jitter penalty term is constructed based on network jitter and the corresponding jitter penalty term coefficients. This jitter penalty term is then fused with the basic loss of control instability to obtain a second-order control instability loss function. The second-order control instability loss function and the upper bound of the control instability deviation are used to construct the control state parameters. Here, the state error vector is a vector composed of deviations in each state dimension, the control input vector is a vector composed of control quantities applied to the intelligent agent, and the positive definite weight matrix is ​​a weighted matrix where all eigenvalues ​​are greater than zero, used to measure the degree of influence between the state error and the control input.

[0115] Step 202: The satellite-ground computing network resource configuration device calculates the channel gain of the ground communication link and the satellite communication link based on the above-mentioned multi-dimensional intent vectors, and calculates the task processing delay of the edge computing node.

[0116] In some embodiments, the satellite-to-ground computing network resource allocation device calculates the channel gain of the ground communication link relative to the intelligent agent based on the three-dimensional spatial coordinates in the multivariate intent vector and the coordinates of the ground base station, combined with reference path loss, path loss index, and shadow fading based on measured channel sample statistics. Here, reference path loss is the basic path loss value at a reference distance from the ground base station; path loss index is the rate at which signal strength decreases with increasing propagation distance; shadow fading is a statistical measure of signal fading caused by environmental factors such as building obstruction; and channel gain characterizes the ratio of received signal power to transmitted signal power.

[0117] In some embodiments, the satellite-to-ground computing network resource allocation device calculates the channel gain of the satellite communication link relative to the intelligent agent using the Ricean model in a non-terrestrial network (NTN), based on the three-dimensional spatial coordinates in the multivariate intent vector and satellite ephemeris data, combined with the carrier frequency, satellite-to-ground tilt distance, and Ricean factor modulated by the elevation angle. Here, ephemeris data describes the change of the satellite's orbital position over time; the carrier frequency is the frequency of the electromagnetic waves used for satellite communication; the satellite-to-ground tilt distance is the straight-line propagation distance between the intelligent agent and the satellite; and the Ricean factor is the ratio of direct path power to scattered path power, and its value increases with the elevation angle, where the elevation angle is the altitude angle of the satellite relative to the horizontal plane where the intelligent agent is located.

[0118] In some embodiments, the satellite-to-ground computing network resource allocation device calculates the queuing probability based on the number of cores of the edge computing nodes, the task arrival rate, and the service rate. Based on the queuing probability, it calculates the waiting latency and processing latency of the intelligent subject's tasks at the edge computing nodes, and determines the task processing latency based on the waiting latency and processing latency. Here, the edge computing node is a node deployed within the target area, close to the intelligent subject, providing local data processing capabilities; the queuing probability is the probability that all cores within the node are occupied and a task needs to wait in a queue when it arrives, which can be solved based on a multi-server queuing model; the task arrival rate is the number of tasks arriving at the edge computing node per unit time; the service rate is the number of tasks that can be completed by a single core per unit time; the waiting latency is the time spent waiting in the queue; and the processing latency is the time spent by the task occupying a core to perform computation.

[0119] Step 203: The satellite-to-ground computing network resource configuration device performs periodic resource configuration analysis based on the above-mentioned communication constraint parameters, control state parameters, channel gain and task processing delay to obtain a resource configuration matrix. The resource configuration matrix is ​​used to control the coordinated configuration of communication resources and edge computing resources of the satellite-to-ground computing network within the target area. The resource configuration matrix includes link offloading decision, bandwidth allocation result, computing power allocation result and transmission power result.

[0120] In some embodiments, the satellite-to-ground computing network resource allocation device constructs decision variables based on link offloading ratio, bandwidth allocation value, computing power allocation value, and transmit power value; constructs a constant matrix based on channel gain and task processing delay calculated in step 202; constructs delay constraints based on the maximum tolerable delay of services in the communication constraint parameters; sets rate constraints based on the minimum guaranteed rate in the communication constraint parameters; and constructs control stability constraints based on the upper bound of control instability deviation in the control state parameters. Under the constraints of delay constraints, rate constraints, and control stability constraints, a periodic resource allocation analysis is performed on the decision variables and constant matrix using Mixed Integer Nonlinear Programming (MINLP). Alternatively, a method based on Lagrange dual decomposition combined with Multi-Agent Deep Deterministic Policy Gradient (MADDPG) can be used to solve the resource allocation matrix. Among them, the link offloading ratio is the proportion of the intelligent subject's task transmitted via the terrestrial communication link and the satellite communication link respectively; the bandwidth allocation value is the spectrum width allocated to the intelligent subject; the computing power allocation value is the core computing power share allocated to the intelligent subject; and the transmit power value is the power used by the intelligent subject or access node to transmit signals. The constant matrix is ​​a matrix whose elements are determined by the channel gain and the task processing delay and whose values ​​remain unchanged during a single analysis process. The delay constraint limits the total end-to-end delay of the task to not exceed the maximum tolerable delay of the service; the rate constraint limits the transmission rate to not be lower than the minimum guaranteed rate; and the control stability constraint limits the control instability loss to not exceed the upper limit of the control instability deviation.

[0121] In some embodiments, periodic resource configuration parsing is repeatedly executed according to a fixed parsing period. Each parsing period takes the latest multivariate intent vector, channel gain and task processing delay as input and outputs the resource configuration matrix of the current time slot, so that the resource configuration is updated in a rolling manner as the location of the intelligent agent and the service requirements change.

[0122] Step 204: When the above-mentioned link offloading decision instructs the above-mentioned intelligent subject to switch the communication link between the above-mentioned ground communication link and the above-mentioned satellite communication link, the satellite-ground computing network resource configuration device performs dual-transmission and reception selection and delay compensation processing on the above-mentioned ground communication link and the above-mentioned satellite communication link to obtain measured transmission delay and measured jitter data.

[0123] In some embodiments, when the link offloading decision instructs the intelligent agent to switch communication links between the ground communication link and the satellite communication link, the satellite-ground computing network resource configuration device synchronously sends data packets carrying incremental sequence numbers along the ground communication link and the satellite communication link to obtain two arriving data packets; performs clock synchronization on the two arriving data packets based on the time synchronization protocol to obtain the propagation delay difference between the two links, and sets a jitter compensation window based on the propagation delay difference; performs deduplication processing on the data packets arriving from the two links within the jitter compensation window, retaining the first arriving valid data packet; and calculates the measured transmission delay and measured jitter data based on the start time and arrival time of the end-to-end transmission of the valid data packets. The dual-transmission and selective reception mechanism represents the processing method where the same data packet is simultaneously transmitted through two links, and the receiving side selects one of them for reception. This can be implemented using the Packet Replication and Elimination (PRE) mechanism in Deterministic Networking (DetNet). The incrementing sequence number is the data packet number that increments by one in the order of transmission, used to identify duplicate packets belonging to the same data. The time synchronization protocol can use the Precision Time Protocol (PTP) to achieve clock alignment between the transmitting and receiving ends of the two links. The propagation delay difference is the difference between the time taken for the data packet to propagate through the terrestrial communication link and the time taken to propagate through the satellite communication link. Since the satellite propagation delay is usually greater than the terrestrial propagation delay, this difference serves as the basis for setting the jitter compensation window. The jitter compensation window is the waiting time set at the receiving end. Duplicate data packets that arrive late on the link within this waiting time can be identified and discarded. The valid data packet is the first data packet that arrives within the jitter compensation window and has not been discarded. The measured transmission delay is the actual time taken for the valid data packet from the start time to the arrival time, and the measured jitter data is the variation amplitude between multiple measured transmission delays.

[0124] For example, if a data packet with sequence number 100 is sent simultaneously via a terrestrial communication link and a satellite communication link, and the terrestrial link copy arrives first, followed by the satellite link copy within the jitter compensation window, then the satellite link copy is discarded, and the terrestrial link copy is retained as a valid data packet. Thus, data is neither interrupted nor duplicated during the handover.

[0125] Step 205: The satellite-to-ground computing network resource configuration device updates the resource configuration matrix based on the measured transmission delay and the measured jitter data.

[0126] In some embodiments, the satellite-to-ground computing network resource configuration device obtains the pruning factor of the inference model in the edge computing node and adjusts the control sampling period of the intelligent agent based on the measured transmission delay and measured jitter data, and updates the resource configuration matrix according to the pruning factor and the adjusted control sampling period to obtain the updated resource configuration matrix.

[0127] In some embodiments, when the measured transmission delay exceeds the maximum tolerable delay of the service and there are no adjustable communication resources, the satellite-ground computing network resource configuration device calculates a pruning factor based on the delay excess, the current remaining computing power of the edge computing node, and the current number of channels of the inference model; prunes the channels of the inference model according to the pruning factor to obtain a pruned model, which reduces the computational load required by the pruned model; expands the control sampling period for the intelligent agent to perform model prediction by a preset step size according to the measured jitter data to obtain an adjusted control sampling period; and updates the resource configuration matrix based on the computational load corresponding to the pruned model and the adjusted control sampling period to obtain an updated resource configuration matrix. The inference model is a neural network model deployed within edge computing nodes to assist intelligent agents in performing model predictions. Channels are parallel convolutional units with similar functions within the inference model; the number of channels represents the number of parallel convolutional units. The latency excess is the difference between the measured transmission latency and the maximum tolerable latency of the service. No adjustable communication resources indicate that bandwidth and transmission power have been fully utilized according to the existing allocation, with no remaining adjustable margin. The pruning factor represents the proportion of pruned channels to the current number of channels. Remaining computing power is the core computing power within the edge computing node that is not yet occupied. After pruning, the computing power required by the model decreases as the number of pruned channels increases, compensating for the increase in air interface transmission time by reducing computation time. The control sampling period is the time interval between two consecutive model predictions performed by the intelligent agent. The preset step size is the fixed increment of the control sampling period expansion. After the control sampling period is expanded, the intelligent agent maintains stable operation by actively reducing speed or increasing the safety distance.

[0128] Step 206: The satellite-to-ground computing network resource configuration device configures the communication resources and edge computing resources within the satellite-to-ground computing network based on the updated resource configuration matrix.

[0129] In some embodiments, the satellite-to-ground computing network resource configuration device reconfigures the communication resources and edge computing resources within the satellite-to-ground computing network based on the updated resource configuration matrix. This involves sending bandwidth allocation results and transmission power results to the ground base station and satellite access side, and sending link offloading decisions and computing power allocation results to the edge computing nodes. This ensures that the tasks of the intelligent agent are transmitted via the ground communication link or satellite communication link according to the share indicated by the link offloading decision, and that the edge computing nodes handle the tasks according to the computing power allocation results. The communication resources and edge computing resources are synchronously adjusted to their positions along with the updated resource configuration matrix. Subsequently, the device substitutes the computational load corresponding to the pruned model and the adjusted control sampling period into the multivariate intent vector of the next time slot, returns to step 201 to re-acquire the multivariate intent vector, and performs periodic resource configuration parsing, ensuring that the resource configuration of the satellite-to-ground computing network is continuously updated in accordance with changes in the intelligent agent's state and link state.

[0130] In some embodiments, combined with Figure 2 ,like Figure 3 As shown, step 205 can also be achieved through step 205a.

[0131] Step 205a: The satellite-to-ground computing network resource configuration device obtains the pruning factor of the inference model in the edge computing node and adjusts the control sampling period of the intelligent subject based on the measured transmission delay and the measured jitter data. It updates the resource configuration matrix based on the pruning factor and the adjusted control sampling period to obtain the updated resource configuration matrix. Based on the updated resource configuration matrix, it reconfigures the communication resources and edge computing resources in the satellite-to-ground computing network.

[0132] In some embodiments, the satellite-to-ground computing network resource configuration device receives the measured transmission delay and measured jitter data fed back in step 204, determines whether the measured transmission delay exceeds the maximum tolerable delay of the service and whether there are adjustable communication resources; when the measured transmission delay exceeds the maximum tolerable delay of the service and there are no adjustable communication resources, it calculates the pruning factor of the inference model based on the delay excess, the current remaining computing power of the edge computing node and the current number of channels of the inference model, and performs pruning on the channels of the inference model according to the pruning factor to obtain the pruned model, which reduces the amount of computation required by the pruned model; at the same time, it expands the control sampling period for the intelligent subject to perform model prediction according to the measured jitter data by a preset step size to obtain the adjusted control sampling period; then, it updates the resource configuration matrix according to the pruning factor and the adjusted control sampling period to obtain the updated resource configuration matrix, and reconfigures the communication resources and edge computing power resources in the satellite-to-ground computing network based on the updated resource configuration matrix.

[0133] In some embodiments, "no adjustable communication resources" indicates that both bandwidth and transmit power have been exhausted according to the current allocation results, with no remaining adjustable margin. That is, the pure communication side can no longer compress transmission latency by reallocating the bandwidth allocation results and transmit power results. The latency excess is the difference between the measured transmission latency and the maximum tolerable latency of the service. The current remaining computing power is the core computing power that has not yet been occupied within the edge computing node. The current number of channels is the existing number of parallel convolutional units with similar functions in the inference model. The pruning factor represents the proportion of the number of pruned channels to the current number of channels. The larger the latency excess and the less the current remaining computing power, the larger the value of the pruning factor and the more channels are pruned. After pruning, the computing power required by the model decreases as the number of pruned channels increases, and the computation time of the edge computing node is compressed accordingly. This compensates for the increase in air interface transmission time and brings the total task latency back to within the maximum tolerable latency of the service.

[0134] In some embodiments, the preset step size is a fixed increment that controls the expansion of the sampling period each time. The greater the change in the measured jitter data, the greater the expansion of the sampling period needs to be. The control sampling period is the time interval between two adjacent model predictions performed by the intelligent subject. After the control sampling period is expanded, the intelligent subject maintains stable operation by actively reducing speed or increasing the safety distance, so as to perform defensive degradation on the physical security dimension.

[0135] In some embodiments, the process of updating the resource configuration matrix based on the pruning factor and the adjusted control sampling period is as follows: The computational load corresponding to the pruned model is substituted into the computational requirement parameters, and the adjusted control sampling period is substituted into the control state parameters. Under the constraints of latency constraints, rate constraints, and control stability constraints, resource configuration parsing is re-executed to obtain the updated resource configuration matrix. The computing power allocation result in the updated resource configuration matrix matches the reduced computational load required by the pruned model. The link offloading decision, bandwidth allocation result, and transmit power result are also adjusted synchronously with the service load changes reflected by the adjusted control sampling period. Subsequently, the space-to-ground computing network resource configuration device reconfigures the communication resources and edge computing resources within the space-to-ground computing network based on the updated resource configuration matrix. Specifically, the adjusted computing power allocation result is sent to the edge computing nodes, the adjusted bandwidth allocation result, transmit power result, and link offloading decision are sent to the access side, and the computational load corresponding to the pruned model and the adjusted control sampling period are substituted into the multivariate intent vector of the next time slot for use in the next parsing cycle.

[0136] For example, the maximum tolerable latency for the intelligent subject's services is 20 milliseconds. After dual-transmission and selective reception, the actual transmission latency is 35 milliseconds, exceeding the latency limit by 15 milliseconds. At this point, both bandwidth and transmission power have no adjustable margin. The satellite-ground computing network resource configuration device calculates a pruning factor based on the 15-millisecond latency limit, the current remaining computing power of the edge computing nodes, and the current number of channels in the inference model. It prunes some channels according to the pruning factor to reduce the computational load required by the inference model. At the same time, it expands the control sampling period from 10 milliseconds to 20 milliseconds according to a preset step size to obtain the adjusted control sampling period. Based on this, it updates the resource configuration matrix and reconfigures communication resources and edge computing resources to bring the total task latency back to within the maximum tolerable latency for the services.

[0137] Thus, when the measured transmission latency exceeds the maximum tolerable latency of the service and there are no adjustable communication resources, a pruning factor is obtained based on the latency excess, the current remaining computing power, and the current number of channels to reduce the computational load required by the inference model. At the same time, the control sampling period is expanded by a preset step size based on the measured jitter data to degrade the defensive capabilities of the intelligent subject. Then, the resource configuration matrix is ​​updated and the communication resources and edge computing power resources within the satellite-ground computing network are reconfigured based on the pruning factor and the adjusted control sampling period. In this way, when communication resources cannot be adjusted, the incremental transmission latency is compensated by compressing the computation time, ensuring the determinism and operational stability of the transmission process of latency-sensitive services.

[0138] The satellite-to-ground computing network resource allocation method provided in this application obtains multi-dimensional intent vectors corresponding to intelligent entities within the target area. These multi-dimensional intent vectors include three-dimensional spatial coordinates, communication constraint parameters, computing requirement parameters, and control state parameters. This parameterizes and structures the business requirements of intelligent entities in the three dimensions of communication, computing, and control, providing a unified requirement input for subsequent resource allocation and avoiding coarse-grained perception of business requirements. Based on this, the channel gain of the ground communication link and satellite communication link is calculated based on the multi-dimensional intent vectors, as well as the task processing delay of the edge computing node. This involves jointly predicting the satellite-to-ground channel state and computing queue state for the next time slot, ensuring that resource calculation is based on the real-time state of the physical and computing layers, improving the matching accuracy between the calculation results and the actual operating environment. Furthermore, based on the communication constraint parameters, control state parameters, channel gain, and task processing delay, periodic resource allocation analysis is performed to obtain a resource allocation matrix including link offloading decisions, bandwidth allocation results, computing power allocation results, and transmission power results, thereby realizing the integration of satellite-to-ground communication resources and edge computing resources. The coordinated allocation ensures that resource allocation decisions are simultaneously constrained by service latency tolerance and control stability requirements, guaranteeing the match between resource allocation results and the intelligent agent's service needs. When the aforementioned link offloading decision instructs the intelligent agent to switch communication links between ground and satellite links, dual-transmission and selective reception with latency compensation are performed on both ground and satellite communication links. This ensures that both ground and satellite links synchronously carry data during the switchover and achieve seamless connection through deduplication and latency compensation. This eliminates the inevitable transmission latency spikes and batch packet losses during cross-domain link cutovers caused by the traditional "break-then-switch" method. In this process, measured transmission latency and jitter data are obtained. Finally, based on the measured transmission latency and jitter data, the resource allocation matrix is ​​updated, and the communication resources and edge computing resources within the satellite-ground computing network are configured based on the updated matrix. This allows the resource allocation to be dynamically adjusted according to the actual transmission status of the links, ensuring that the resource allocation continuously matches the actual link status and guaranteeing the continuity and determinism of the intelligent agent's transmission process before and after the communication link switchover.

[0139] In some embodiments, combined with Figure 2 ,like Figure 4 As shown, step 201 above can be implemented through steps 201a to 201c.

[0140] Step 201a: The satellite-ground computing network resource configuration device collects the three-dimensional spatial coordinates, velocity vector, maximum tolerable latency of services, minimum guaranteed rate, single task computation amount, upper limit of control instability deviation, and service priority identifier of the intelligent subject.

[0141] In some embodiments, the satellite-to-ground computing network resource allocation device periodically collects the three-dimensional spatial coordinates, velocity vector, maximum tolerable latency for services, minimum guaranteed rate, single task computation load, upper limit of control instability deviation, and service priority identifier of the intelligent subject within the target area. Specifically, the three-dimensional spatial coordinates represent the real-time spatial position of the intelligent subject within the target area, which can be read from the positioning information reported by the intelligent subject; the velocity vector represents the displacement components of the intelligent subject in each coordinate axis direction per unit time, reflecting the movement status of the intelligent subject; the maximum tolerable latency for services represents the longest allowed time from dispatch to completion of the services carried by the intelligent subject; the minimum guaranteed rate represents the lower limit of the rate required for service transmission by the intelligent subject; the single task computation load represents the number of floating-point operations required for a single task dispatch by the intelligent subject, measured in floating-point operations per second (FLOPs); the upper limit of control instability deviation represents the maximum deviation that the intelligent subject's control process can withstand; and the service priority identifier represents the order in which the services carried by the intelligent subject occupy resources during resource contention, for example, priority levels can be divided according to embodied control, high-sensitivity, and conventional IoT.

[0142] In some embodiments, the three-dimensional spatial coordinates and velocity vectors change in real time with the movement of the intelligent agent. The maximum tolerable latency, minimum guaranteed rate, single task computation amount, upper limit of control instability deviation, and service priority identifier are determined according to the service type. Together, they constitute the original inputs for constructing communication constraint parameters, calculation requirement parameters, and control state parameters in step 201b.

[0143] Step 201b: The satellite-to-ground computing network resource configuration device constructs the communication constraint parameters based on the maximum tolerable latency of the above services and the minimum guaranteed rate, constructs the computing requirement parameters based on the computing volume of a single task, and constructs the control state parameters based on the velocity vector, the state error, and the network jitter.

[0144] In some embodiments, the satellite-to-ground computing network resource allocation device constructs communication constraint parameters based on the maximum tolerable latency and the minimum guaranteed rate of the service. That is, the maximum tolerable latency of the service limits the upper limit of the latency of the service, and the minimum guaranteed rate limits the lower limit of the transmission rate of the service. The two constitute the hard requirements for the configuration of satellite-to-ground computing network communication resources. The device also constructs computing requirement parameters based on the computing volume of a single task. That is, the computing volume of a single task represents the computing scale required for the task to be executed on the edge computing node, and serves as the quantitative basis for the allocation of edge computing power resources.

[0145] In some embodiments, the process by which the satellite-to-ground computing network resource allocation device constructs control state parameters based on velocity vector, state error, and network jitter is as follows: The state error vector and control input vector of the intelligent agent are obtained. The state error vector represents the deviation between the actual motion trajectory and the desired motion trajectory, and the control input vector is a vector composed of control quantities applied to the intelligent agent. A positive definite weight matrix is ​​introduced to perform quadratic form calculations on the state error vector and the control input vector respectively to obtain the basic loss of control instability. A jitter penalty term is constructed based on the network jitter and the corresponding jitter penalty term coefficient. The jitter penalty term is fused with the basic loss of control instability to obtain a second-order control instability loss function. The second-order control instability loss function and the upper bound of the control instability deviation are used to construct the control state parameters.

[0146] In some embodiments, the second-order control instability loss function can be expressed as:

[0147]

[0148] Wherein, x represents the state error vector of the intelligent agent, characterizing the deviation between the actual trajectory and the desired trajectory; u represents the control input vector of the intelligent agent, consisting of the control quantities applied to the agent; Q represents the positive definite weight matrix weighted by the state error vector, with all eigenvalues ​​greater than zero, used to measure the influence of each component of the state error on control stability; R represents the positive definite weight matrix weighted by the control input vector, with all eigenvalues ​​greater than zero, used to measure the influence of each component of the control quantity on control stability; T represents the matrix transpose operation, where the vector, after being weighted by the positive definite weight matrix, is multiplied by itself to obtain the quadratic scalar loss; λ represents the jitter penalty coefficient, a non-negative weight used to measure the influence of network jitter on control stability; and δ represents network jitter, characterizing the degree of variation in data packet arrival time, with the jitter penalty λδ² increasing as network jitter increases.

[0149] In some embodiments, the larger the value of the second-order control instability loss function, the higher the degree of deviation of the intelligent agent's control process from stability; the upper bound of the control instability deviation is the upper limit that the value of the second-order control instability loss function is not allowed to exceed. This upper bound and the second-order control instability loss function together constitute the control state parameters, enabling subsequent resource allocation analysis to perceive the intelligent agent's tolerance margin for transmission jitter.

[0150] For example, when an intelligent robot performs a transport task in a park, its state error vector x increases due to path bends, and at the same time, the link network jitter δ increases. The value of the second-order control instability loss function rises accordingly and approaches the upper bound of the control instability deviation. Based on this, the satellite-ground computing network resource allocation device reserves a lower latency and jitter level for the robot in the subsequent resource allocation analysis to avoid control instability.

[0151] Step 201c: The satellite-ground computing network resource configuration device combines the above-mentioned three-dimensional spatial coordinates, the above-mentioned communication constraint parameters, the above-mentioned calculation requirement parameters and the above-mentioned control state parameters according to preset dimensions to obtain the above-mentioned structured multi-dimensional intent vector.

[0152] In some embodiments, the satellite-to-ground computing network resource configuration device combines three-dimensional spatial coordinates, communication constraint parameters, computing requirement parameters, and control state parameters according to preset dimensions to obtain a structured multi-dimensional intent vector. The preset dimensions represent the position and order of each component in the multi-dimensional intent vector, which can be arranged sequentially according to spatial attributes, communication attributes, computing attributes, and control attributes. The combination involves filling the corresponding components of the vector with the aforementioned parameters according to the preset dimensions, transforming the originally scattered and inconsistent business state information into a vector form with unified dimensions and definite meaning. The multi-dimensional intent vector represents the comprehensive business intent of the intelligent agent in the spatial, communication, computing, and control dimensions.

[0153] In some embodiments, the multivariate intent vector of the i-th intelligent agent can be represented as:

[0154]

[0155] Wherein, x_i(t) is the multivariate intent vector of the i-th intelligent agent at time t; i is the intelligent agent's number; p_i is the three-dimensional spatial coordinates of the i-th intelligent agent; v_i is the velocity vector of the i-th intelligent agent; D_max,i is the maximum tolerable latency for the i-th intelligent agent; R_min,i is the minimum guaranteed rate for the i-th intelligent agent; F_i is the computational cost per task for the i-th intelligent agent; B_max,i is the upper bound of the control instability deviation for the i-th intelligent agent; and L_C3,i is the current value of the second-order control instability loss function for the i-th intelligent agent. The service priority identifier can be added as an additional component to the multivariate intent vector according to the service priority level.

[0156] In some embodiments, the first two components p_i and v_i in the multivariate intent vector are the position and motion inputs for calculating the channel gain in step 202, the middle components D_max,i and R_min,i are the source of the delay constraints and rate constraints in step 203, the component F_i is the quantification basis for edge computing power allocation, and the last components B_max,i and L_C3,i are the source of the control stability constraints in step 203. Thus, the multivariate intent vector deconstructs the communication constraint parameters, computing requirement parameters, and control state parameters into mutually independent yet interconnected components, enabling subsequent channel prediction and resource allocation analysis to simultaneously perceive the service requirements of the three dimensions of communication, computing, and control.

[0157] In this way, by periodically collecting three-dimensional spatial coordinates, velocity vectors, maximum tolerable latency of services, minimum guaranteed rate, single task computation amount, upper bound of control instability deviation, and service priority identifiers, a second-order control instability loss function is constructed through quadratic calculation and jitter penalty term fusion. This forms communication constraint parameters, computational requirement parameters, and control state parameters, which are then combined according to preset dimensions into a structured multi-dimensional intent vector. This transforms the service requirements of intelligent agents in the dimensions of communication, computation, and control into a parameterized representation that is consistent in dimensions and can directly participate in channel prediction and resource allocation analysis, providing a structurally consistent input basis for the collaborative configuration of communication resources and edge computing resources within the space-ground computing network.

[0158] In some embodiments, when performing step 202 to calculate the channel gain of the ground communication link and the satellite communication link based on the aforementioned multi-dimensional intent vectors, and to calculate the task processing delay of the edge computing node, the following implementation methods may also be adopted, but are not limited to: calculating the channel gain of the ground communication link relative to the aforementioned multi-dimensional agent based on the aforementioned multi-dimensional intent vectors and the coordinates of the ground base station, combined with the reference path loss, path loss index, and shadow fading amount based on the statistics of measured channel samples; calculating the channel gain of the satellite communication link relative to the aforementioned multi-dimensional agent based on the aforementioned multi-dimensional intent vectors and satellite ephemeris data, combined with the carrier frequency, satellite-to-ground tilt distance, and Rice factor modulated by elevation angle, using the Rice model; calculating the queuing probability based on the number of cores, task arrival rate, and service rate of the aforementioned edge computing node, calculating the waiting delay and processing delay of the aforementioned intelligent agent's task at the edge computing node based on the aforementioned queuing probability, and determining the aforementioned task processing delay based on the aforementioned waiting delay and the aforementioned processing delay.

[0159] In some embodiments, the satellite-to-ground computing network resource configuration device reads the three-dimensional spatial coordinates of the intelligent subject from the multi-dimensional intent vector, obtains the ground base station coordinates of the ground base station in the target area, and calculates the channel gain of the ground communication link relative to the intelligent subject by combining the reference path loss, path loss index and shadow fading based on measured channel sample statistics.

[0160] In some embodiments, the path loss of a terrestrial communication link can be calculated using the following formula:

[0161]

[0162] In this diagram, PL_g represents the path loss of the terrestrial communication link; PL0 represents the reference path loss, which is the basic path loss value at a reference distance from the ground base station, and its value is determined by the operating frequency band and deployment environment of the ground base station; n represents the path loss exponent, which characterizes the rate attenuation of signal strength as the propagation distance increases, and its value increases with the complexity of the propagation environment; d represents the Euclidean distance between the intelligent subject and the ground base station, calculated from the three-dimensional spatial coordinates in the multivariate intent vector and the coordinates of the ground base station; d0 represents the reference distance; and ξ represents the shadow fading amount, which is a statistical measure of signal fading caused by environmental factors such as building obstruction, obtained based on measured channel samples. After the path loss PL_g is obtained, the channel gain of the terrestrial communication link relative to the intelligent subject is determined accordingly; the higher the path loss, the lower the channel gain.

[0163] In some embodiments, channel gain characterizes the ratio of received signal power to transmitted signal power. This channel gain determines the upper limit of the achievable transmission rate of a terrestrial communication link under the same transmit power and bandwidth allocation values, and serves as the basis for rate constraint calculation and link offloading decisions in subsequent periodic resource allocation analysis.

[0164] In some embodiments, the satellite-to-ground computing network resource configuration device reads the three-dimensional spatial coordinates of the intelligent agent from the multi-dimensional intent vector, determines the position of the currently visible satellite by combining satellite ephemeris data, and calculates the channel gain of the satellite communication link relative to the intelligent agent by combining the carrier frequency, satellite-to-ground tilt distance and Rice factor modulated by elevation angle using the Rice model in Non-Terrestrial Network (NTN).

[0165] In some embodiments, the satellite-to-ground tilt distance can be calculated using the following formula:

[0166]

[0167] Wherein, d_s represents the satellite-to-ground tilt distance, which is the straight-line propagation distance between the intelligent agent and the satellite; p_sat represents the three-dimensional position of the satellite calculated from satellite ephemeris data; p_i represents the three-dimensional spatial coordinates of the i-th intelligent agent, taken from the multi-dimensional intent vector; and the symbol ‖·‖ represents the vector magnitude operation.

[0168] In some embodiments, the free-space path loss of a satellite communication link can be calculated using the following formula:

[0169]

[0170] Wherein, PL_s represents the free space path loss of the satellite communication link; d_s represents the satellite-to-ground tilt distance; f represents the carrier frequency, which is the frequency of the electromagnetic waves used in satellite communication. The free space path loss increases with the increase of the satellite-to-ground tilt distance d_s and the carrier frequency f; and c represents the speed of light.

[0171] In some embodiments, the Rice factor is modulated by the elevation angle and can be calculated using the following formula:

[0172]

[0173] In this model, K(θ) represents the Rice factor at elevation angle θ, which is the ratio of direct path power to scattered path power; θ represents the elevation angle, the altitude angle of the satellite relative to the horizontal plane where the intelligent agent is located; K_min is the lower bound of the Rice factor, corresponding to the case where the direct path is significantly affected by obstruction and scattering at low elevation angles; K_max is the upper bound of the Rice factor, corresponding to the case where the direct path dominates at high elevation angles; sinθ increases with elevation angle θ, therefore, the higher the elevation angle and the larger the Rice factor, the higher the proportion of the direct path in the received signal. The satellite-to-ground computing network resource allocation device substitutes the free space path loss PL_s and the Rice factor K(θ) into the Rice model to obtain the channel gain of the satellite communication link relative to the intelligent agent, which is used for the generation of subsequent link offloading decisions.

[0174] In some embodiments, the satellite-to-ground computing network resource allocation device calculates the queuing probability based on the number of cores of edge computing nodes, task arrival rate, and service rate, using a multi-server queuing model, which can be solved by the following formula:

[0175]

[0176] In this context, P_queue represents the queuing probability, which is the probability that all cores are occupied and the task needs to wait in a queue when it arrives at the edge computing node; c represents the number of cores on the edge computing node, which is the number of cores within the node that can execute tasks in parallel; λ represents the task arrival rate, which is the number of tasks arriving at the edge computing node per unit time; μ represents the service rate, which is the number of tasks that a single core can complete per unit time; ρ represents the node utilization rate, which is the ratio of the task arrival rate to the total service capacity of all cores. The closer the node utilization rate ρ is to 1, the heavier the node load; and k is the intermediate index in the summation operation, which ranges from 0 to c minus 1. The queuing probability P_queue increases with the increase of the task arrival rate λ and the increase of the node utilization rate ρ.

[0177] In some embodiments, the satellite-to-ground computing network resource allocation device calculates the waiting latency and processing latency of the intelligent agent's task at the edge computing node based on the queuing probability. The waiting latency can be calculated using the following formula:

[0178]

[0179] Where T_wait represents the waiting latency, which is the time spent waiting in the queue for a task; P_queue represents the queuing probability; c represents the number of cores; μ represents the service rate; λ represents the task arrival rate; and cμ−λ represents the remaining service capacity of the edge computing node. The lower the remaining service capacity, the higher the queuing probability P_queue, and the longer the waiting latency T_wait.

[0180] In some embodiments, the task processing latency is determined based on the waiting latency and the processing latency, and can be calculated using the following formula:

[0181]

[0182] Wherein, T_proc represents the task processing latency, which is the total time spent by the intelligent agent in queuing and executing its tasks at the edge computing node; T_wait represents the waiting latency; F_i represents the computational load of the i-th intelligent agent in a single task, taken from the computational demand parameter in the multivariate intent vector; and f_i represents the computing power allocation value assigned to the i-th intelligent agent by the edge computing node. The larger the computational load F_i in a single task or the smaller the computing power allocation value f_i, the longer the processing latency. Therefore, the satellite-ground computing network resource allocation device determines the sum of the waiting latency and the processing latency as the task processing latency. This task processing latency, along with the channel gain of the ground communication link and the satellite communication link, serves as the source for constructing the constant matrix and is input into the periodic resource allocation analysis in step 203.

[0183] For example, an edge computing node is configured with 8 cores, with a task arrival rate of 40 tasks per second and a service rate of 10 tasks per core per second. Based on this, the star-ground computing network resource configuration device solves the queuing probability and waiting latency under the multi-service queuing model. Then, it divides the single task computation of a certain intelligent subject by the allocated computing power value to obtain the processing latency. The two are added together to obtain the task processing latency, which is used for subsequent latency constraint calculation.

[0184] Thus, based on the three-dimensional spatial coordinates in the multivariate intent vector, the channel gain of the ground communication link is predicted by the reference path loss, path loss index, and shadow fading amount according to the logarithmic distance path loss model. The channel gain of the satellite communication link is predicted by the carrier frequency, satellite-to-ground tilt distance, and Rice factor modulated by elevation angle according to the Rice model. The waiting delay is derived from the number of cores, task arrival rate, and service rate through queuing probability. The task processing delay is determined by combining the single task computation amount and computing power allocation value. Thus, the joint prediction results of the satellite-to-ground channel status and edge computing power queue status of the next time slot are obtained before the resource configuration analysis is executed. This provides a state input that matches the actual operating environment for the coordinated configuration of communication resources and edge computing power resources within the satellite-to-ground computing network.

[0185] In some embodiments, when performing step 203 to obtain a resource allocation matrix by performing periodic resource allocation analysis based on the aforementioned communication constraint parameters, control state parameters, channel gain, and task processing delay, the following implementation methods may also be adopted, for example: constructing decision variables based on link offloading ratio, bandwidth allocation value, computing power allocation value, and transmit power value; constructing a constant matrix based on the aforementioned channel gain and task processing delay; constructing a delay constraint based on the aforementioned maximum tolerable delay of services in the aforementioned communication constraint parameters; setting a rate constraint based on the aforementioned minimum guaranteed rate in the aforementioned communication constraint parameters; constructing a control stability constraint based on the upper bound of control instability deviation in the aforementioned control state parameters; and performing periodic resource allocation analysis on the aforementioned decision variables and the aforementioned constant matrix under the constraints of the aforementioned delay constraint, the aforementioned rate constraint, and the aforementioned control stability constraint to obtain the aforementioned resource allocation matrix.

[0186] In some embodiments, the satellite-to-ground computing network resource allocation device constructs decision variables based on link offloading ratio, bandwidth allocation value, computing power allocation value, and transmission power value. The link offloading ratio, denoted as 'a', represents the proportion of tasks transmitted by the intelligent subject via terrestrial and satellite communication links, with a value between 0 and 1. A value closer to 1 indicates a larger share of tasks transmitted via satellite communication links. The two endpoints of the value (0 or 1) correspond to link offloading decisions based on pure terrestrial or pure satellite transmission. The bandwidth allocation value, denoted as 'b', represents the spectrum width allocated to the intelligent subject. The computing power allocation value, denoted as 'f', represents the share of core computing power allocated to the intelligent subject by edge computing nodes. The transmission power value, denoted as 'p', represents the power used by the intelligent subject or access node to transmit signals. Arranging these four types of values ​​for each intelligent subject within the target area according to their numbers constitutes the set of decision variables to be solved. These four types of values ​​are interdependent; for example, increasing the bandwidth allocation value can shorten air interface transmission time, and increasing the computing power allocation value can compress task processing latency. However, the total spectrum and total computing power are limited and compete for resources, requiring joint analysis to obtain a matching combination of values.

[0187] In some embodiments, the satellite-to-ground computing network resource allocation device constructs a constant matrix based on the channel gain and task processing delay calculated in step 202. The constant matrix is ​​a matrix whose element values ​​remain unchanged during a single analysis process. Its elements include the channel gain of the ground communication link and satellite communication link relative to each intelligent agent, and the task processing delay of each intelligent agent's task at the edge computing node. The channel gain determines the upper limit of the achievable transmission rate of the link under the same transmit power and bandwidth allocation values, and the task processing delay provides the baseline of task latency at the edge computing node. After these two elements form the constant matrix, the solution of the decision variables is based on the prediction results of the channel state and computing queue state in the next time slot. Since the analysis is performed in a rolling cycle, the constant matrix is ​​reconstructed at the beginning of each analysis cycle using the latest channel gain and task processing delay at that time.

[0188] In some embodiments, the satellite-to-ground computing network resource allocation device constructs a delay constraint based on the maximum tolerable delay of services in the communication constraint parameters. The delay constraint can be expressed as:

[0189]

[0190] Wherein, T_total represents the total end-to-end latency of the i-th intelligent agent's task; a_i represents the link offloading ratio of the i-th intelligent agent; T_g,i represents the air interface transmission time of the task share via the terrestrial communication link, which is related to the channel gain of the terrestrial communication link; T_s,i represents the air interface transmission time of the task share via the satellite communication link, which is related to the channel gain of the satellite communication link; T_proc,i represents the task processing latency of the i-th intelligent agent; and D_max,i represents the maximum service-tolerable latency of the i-th intelligent agent. The latency constraint ensures that the total end-to-end latency obtained by adding the transmission latency and the task processing latency does not exceed the maximum service-tolerable latency.

[0191] In some embodiments, the satellite-to-ground computing network resource allocation device sets rate constraints based on the minimum guaranteed rate in the communication constraint parameters. The achievable transmission rate of the link is calculated according to the Shannon capacity formula, and can be expressed as:

[0192]

[0193] Wherein, r_i represents the actual achievable transmission rate of the i-th intelligent agent; b_i represents the bandwidth allocation value assigned to the i-th intelligent agent; p_i represents the transmit power value of the i-th intelligent agent; h_i represents the channel gain of the selected communication link, taken from a constant matrix; N0 represents the noise power spectral density; and R_min,i represents the minimum guaranteed rate of the i-th intelligent agent. The rate constraint stipulates that the actual achievable transmission rate must not be lower than the minimum guaranteed rate. It can be seen from this formula that the lower the channel gain, the higher the bandwidth allocation value or transmit power value required to achieve the rate constraint.

[0194] In some embodiments, the satellite-to-ground computing network resource allocation device constructs control stability constraints based on the upper bound of the control instability deviation in the control state parameters, which can be expressed as:

[0195]

[0196] Wherein, L_C3,i represents the current value of the second-order control instability loss function of the i-th intelligent agent, which varies with the network jitter level of the allocated transmission link; B_max,i represents the upper bound of the control instability deviation of the i-th intelligent agent. The control stability constraint limits the value of the second-order control instability loss function to not exceed the upper bound of the control instability deviation, so that the allocation results of communication resources and edge computing power resources are subject to the reverse constraint of the control stability requirements of the intelligent agent's end side.

[0197] In some embodiments, the satellite-to-ground computing network resource allocation device performs periodic resource allocation analysis on decision variables and constant matrices under constraints of time delay, rate, and control stability. The optimization objective of the analytical problem can be expressed as:

[0198]

[0199] Among them, the letter U(a,b,f,p) represents the overall system performance of the campus as it changes with the link offloading ratio, bandwidth allocation value, computing power allocation value and transmission power value. It increases as the task latency of each intelligent subject decreases and the achievement rate increases; the letter ω is the energy consumption weighting coefficient, which is used to balance system performance and transmission power consumption; the letter i is the number of the intelligent subject. This analytical problem can be constructed as a mixed-integer nonlinear programming (MINLP) problem, where the link offloading ratio is represented by integers to indicate the link offloading decision, and the bandwidth allocation, computing power allocation, and transmission power values ​​are in continuous form. Solving this type of problem has high computational complexity and can be solved using a method based on Lagrange dual decomposition combined with multi-agent deep deterministic policy gradient (MADDPG). Specifically, Lagrange dual decomposition is used to handle constraints, and the multi-agent deep deterministic policy gradient method is used to generate decisions for each intelligent agent. The resource allocation matrix of the current time slot is output within a millisecond time interval. The parsing cycle is set in milliseconds, and each parsing cycle is executed with the latest multivariate intent vector and constant matrix, so that the resource allocation is continuously updated according to the changes in the location of the intelligent agents and business needs.

[0200] In some embodiments, the parsed resource allocation matrix includes link offloading decision, bandwidth allocation result, computing power allocation result, and transmission power result. The link offloading decision is obtained by merging the link offloading ratio and is used to indicate whether the intelligent subject transmits via a terrestrial communication link, transmits via a satellite communication link, or switches between the two communication links. The bandwidth allocation result, computing power allocation result, and transmission power result respectively give the values ​​of the bandwidth allocation value, computing power allocation value, and transmission power value corresponding to each intelligent subject.

[0201] For example, the maximum tolerable latency for a certain intelligent entity with embodied control is 20 milliseconds. The satellite-ground computing network resource allocation device analyzes decision variables under the joint constraints of latency, rate, and control stability. If the analysis finds that the channel gain of the ground communication link decreases due to building obstruction and that simply increasing the transmission power value is no longer sufficient to complete the transmission within the latency constraint, the output link offloading decision instructs the intelligent entity to switch to the satellite communication link and simultaneously provides the matching bandwidth allocation result, computing power allocation result, and transmission power result.

[0202] Thus, by constructing decision variables with link offloading ratio, bandwidth allocation value, computing power allocation value and transmit power value, constructing a constant matrix with channel gain and task processing delay, and constructing delay constraints, rate constraints and control stability constraints with the maximum tolerable delay of services, the minimum guaranteed rate and the upper bound of control instability deviation respectively, the decision variables are jointly solved by mixed integer nonlinear programming under the common constraints of the three types of constraints. This makes the allocation of communication resources and edge computing power resources simultaneously constrained by communication time limits, rate bottom lines and end-side control stability requirements, and obtains a resource allocation matrix that matches the next time slot satellite-ground channel state and computing power queue state.

[0203] In some embodiments, when performing step 204 to perform dual-transmission selection and delay compensation processing on the ground communication link and the satellite communication link to obtain measured transmission delay and measured jitter data, the following implementation methods may also be adopted, for example: synchronously transmitting data packets carrying incremental sequence numbers along the ground communication link and the satellite communication link to obtain two arriving data packets; performing clock synchronization on the two arriving data packets based on a time synchronization protocol to obtain the propagation delay difference between the two links; setting a jitter compensation window based on the propagation delay difference; performing deduplication processing on the data packets arriving from the two links within the jitter compensation window, retaining the first arriving valid data packet; calculating the measured transmission delay and the measured jitter data based on the start time and arrival time of the end-to-end transmission of the valid data packet.

[0204] In some embodiments, when the link offloading decision instructs the intelligent agent to switch communication links between the ground communication link and the satellite communication link, the satellite-ground computing network resource allocation device encapsulates the data to be transmitted by the intelligent agent into data packets carrying incremental sequence numbers at the transmitting end, and transmits them synchronously along the ground communication link and the satellite communication link, resulting in two arriving data packets. The dual-transmission and selective reception represents the processing method where the same data packet is transmitted simultaneously through two links, and the receiving side selects one to receive. This can be implemented using the Packet Replication and Elimination (PRE) mechanism in Deterministic Networking (DetNet). The incremental sequence number is a data packet number incremented by one according to the order of transmission, denoted by the letter S. The value increments by one for each encapsulated data packet, and the receiving side identifies duplicate packets belonging to the same data based on the incremental sequence number. Synchronous transmission means that two copies of the same data packet are injected into the ground communication link and the satellite communication link respectively at the same trigger time, so that the order of the two copies at the receiving end is determined only by the propagation and queuing state of the link itself.

[0205] In some embodiments, the incrementing sequence number and the data packet arrival time can be recorded using the following formula:

[0206]

[0207] In this set of data packets, S represents the incrementing sequence number; t_g(S) represents the arrival time of a copy of the data packet with sequence number S at the receiving end via the terrestrial communication link; and t_s(S) represents the arrival time of a copy of the same data packet at the receiving end via the satellite communication link. This triplet is recorded with each data packet and serves as the original basis for subsequent deduplication processing and actual transmission delay calculation.

[0208] In some embodiments, the satellite-to-ground computing network resource configuration device performs clock synchronization on the two arriving data packets based on a time synchronization protocol to obtain the propagation delay difference between the two links, and sets a jitter compensation window based on the propagation delay difference. The time synchronization protocol can employ a Precision Time Protocol (PTP) to achieve clock alignment between the transmitting and receiving ends of the two links. This involves calculating the offset and correcting the receiving end clock through message exchange between the master and slave clocks, ensuring that the comparison of the arrival times of the two links is based on the same time reference. The propagation delay difference between the two links can be calculated using the following formula:

[0209]

[0210] Wherein, Δt represents the difference in propagation delay between the two links; t̄_g represents the statistical average arrival time of multiple data packet replicas via the ground communication link; and t̄_s represents the statistical average arrival time of the same batch of data packet replicas via the satellite communication link. The absolute value is used to ensure the difference is non-negative, as the propagation delay of the satellite communication link is typically greater than that of the ground communication link, and this difference mainly reflects the delay difference caused by the satellite-to-ground propagation distance. The jitter compensation window can be set using the following formula:

[0211]

[0212] In this diagram, W represents the jitter compensation window, which is the waiting time set at the receiving end; β represents the window margin coefficient, an adjustable weight greater than zero, used to absorb fluctuations in link latency; and σ represents the statistical standard deviation of the arrival times of the data packets in this batch. The larger σ is, the stronger the link latency fluctuation, and the longer the jitter compensation window should be. If the jitter compensation window is set too short, the replicas may be misjudged as lost; if it is set too long, the waiting latency at the receiving end will increase. Therefore, a compromise is made, using the propagation delay difference Δt as the baseline and the statistical standard deviation σ as the margin.

[0213] In some embodiments, the satellite-to-ground network resource allocation device performs deduplication processing on data packets arriving from both links within a jitter compensation window, retaining the first arriving valid data packet. Specifically, the deduplication process involves the receiving end registering arriving data packets according to their incrementing sequence numbers within the timing segment of the jitter compensation window W. If a copy with the same incrementing sequence number arrives again, it is identified as a duplicate packet and discarded. The valid data packet is the first arriving copy of the same sequence number within the jitter compensation window that has not been discarded. Therefore, even if one link experiences packet loss or a sudden increase in latency during communication link switching, a copy from the other link can still fill the gap within the jitter compensation window, ensuring that data is neither interrupted nor duplicated.

[0214] For example, a data packet with sequence number 100 is sent simultaneously via a terrestrial communication link and a satellite communication link. The terrestrial link copy arrives 10 milliseconds after the start time, and the satellite link copy arrives 260 milliseconds after the start time. If the jitter compensation window W, jointly set by the propagation delay difference Δt and the margin term, is 300 milliseconds, then the satellite link copy arrives within the window. The receiving end recognizes that its incrementing sequence number is the same as the already retained copy, discards it as a duplicate packet, and retains the terrestrial link copy as a valid data packet. Conversely, if the terrestrial link copy is lost, the satellite link copy is retained as a valid data packet upon arrival, and the service data is not interrupted due to the failure of a single link.

[0215] In some embodiments, the satellite-to-ground network resource configuration device calculates the measured transmission delay based on the start and arrival times of the effective data packet end-to-end transmission, which can be calculated using the following formula:

[0216]

[0217] Wherein, T_m(S) represents the measured transmission delay of the valid data packet with incrementing sequence number S; t_arr(S) represents the arrival time of the valid data packet; and t_start(S) represents the start time of the data packet at the sending end, i.e., the time when two copies are injected into the link during synchronous transmission. The measured transmission delay is the actual time elapsed from the start time to the arrival time of the valid data packet, covering both terrestrial communication link and satellite communication link transmission scenarios, thus reflecting the true transmission status of the link before and after the handover.

[0218] In some embodiments, the satellite-to-ground network resource configuration device calculates the measured jitter data based on the measured transmission delay of multiple consecutive valid data packets, which can be calculated using the following formula:

[0219]

[0220] Wherein, J_m represents the measured jitter data, which is the statistical average of the variation in the measured transmission delay between adjacent valid data packets; N represents the number of valid data packets included in the statistics; S_k represents the incrementing sequence number of the k-th valid data packet arranged in the order of transmission; T_m(S_k) and T_m(S_k−1) represent the measured transmission delays of two adjacent valid data packets, respectively; the larger the measured jitter data value, the stronger the fluctuation in the link arrival time. The measured transmission delay and measured jitter data are used as inputs to step 205 to obtain the pruning factor of the inference model in the edge computing node and adjust the control sampling period of the intelligent agent, thereby updating the resource allocation matrix.

[0221] In this way, during communication link switching, data packets carrying incremental sequence numbers are sent synchronously along the ground communication link and the satellite communication link. After clock alignment by the time synchronization protocol, a jitter compensation window is set according to the propagation delay difference between the two links. Within the jitter compensation window, duplicate data packets are removed and the first arriving valid data packets are retained. The measured transmission delay and measured jitter data are calculated based on the start and arrival times of the valid data packets. Thus, during cross-domain link cutover, transmission delay abrupt changes and batch packet loss are eliminated by using two replicas to back each other up. At the same time, a data basis reflecting the actual transmission status of the link is obtained, providing a measured basis for subsequent updates to the resource allocation matrix.

[0222] In some embodiments, when performing the steps of obtaining the pruning factor of the inference model in the edge computing node and adjusting the control sampling period of the intelligent agent based on the measured transmission delay and the measured jitter data, and updating the resource configuration matrix based on the pruning factor and the adjusted control sampling period to obtain the updated resource configuration matrix, the following implementation methods may also be adopted, but are not limited to: when the measured transmission delay exceeds the maximum tolerable delay of the service and there are no adjustable communication resources, the pruning factor is calculated based on the delay excess, the current remaining computing power of the edge computing node, and the current number of channels of the inference model; the channels of the inference model are pruned according to the pruning factor to obtain the pruned model, and the computational load required by the pruned model is reduced; the control sampling period for the intelligent agent to perform model prediction is increased by a preset step size based on the measured jitter data to obtain the adjusted control sampling period; the resource configuration matrix is ​​updated based on the computational load corresponding to the pruned model and the adjusted control sampling period to obtain the updated resource configuration matrix.

[0223] In some embodiments, the satellite-to-ground computing network resource configuration device receives the measured transmission delay and measured jitter data output in step 204, determines whether the measured transmission delay exceeds the maximum tolerable delay of the service and whether there are adjustable communication resources; when the measured transmission delay exceeds the maximum tolerable delay of the service and there are no adjustable communication resources, the pruning factor is calculated based on the delay excess, the current remaining computing power of the edge computing node, and the current number of channels of the inference model, which can be calculated by the following formula:

[0224]

[0225] Where η is the pruning factor, representing the proportion of pruned channels to the current number of channels, with a value between 0 and 1; T_m is the measured transmission delay, i.e., the actual time taken for a valid data packet from its start time to its arrival time; D_max is the maximum tolerable delay for the service; T_m−D_max is the delay excess, i.e., the difference between the measured transmission delay and the maximum tolerable delay for the service. The larger the delay excess, the more time is spent on air interface transmission, and the more time is required for compression calculations, and the larger the value of the pruning factor should be; F_avail is the edge... The remaining computing power of the current computing node represents the core computing power within the node that is not yet utilized. The less remaining computing power, the smaller the pruning range that can be supported, and the stronger the constraint on the pruning factor. C_ch represents the computational cost required for a single channel of the inference model to perform one inference operation. n_ch represents the current number of channels in the inference model, i.e., the current number of parallel convolutional units with similar functions in the model. C_ch·n_ch represents the overall computational cost required by the inference model. The ratio of F_avail to this term represents the pruning ratio corresponding to the upper limit of the model size that the current remaining computing power can support. The smaller of the two terms is taken as the pruning factor, so that the pruning range is simultaneously limited by both latency compression requirements and the remaining computing power support capacity.

[0226] In some embodiments, the unadjustable communication resource represents that the bandwidth and transmission power have been exhausted according to the current allocation results, with no remaining adjustable margin. That is, the pure communication side can no longer compress the transmission latency by reallocating the bandwidth allocation results and transmission power results. At this time, the total task latency can only be reduced from the computing side. The inference model is a neural network model deployed in the edge computing node to assist the intelligent agent in performing model prediction. The channel is a parallel convolutional unit with similar function in the inference model.

[0227] In some embodiments, the satellite-to-ground computing network resource allocation device prunes the channels of the inference model according to a pruning factor, resulting in a pruned model. Pruning involves removing the number of channels corresponding to the pruning factor from the network starting from the last layer. Convolution operations on the pruned channels are no longer performed. The computational cost of the pruned model can be determined by the following formula:

[0228]

[0229] Wherein, C′ represents the computational cost of the pruned model; C_ch represents the computational cost of a single channel; n_ch represents the current number of channels; η represents the pruning factor; and 1−η represents the proportion of channels retained. The computational cost required by the pruned model decreases as the number of pruned channels increases, and the computation time for edge computing nodes to execute model predictions is compressed accordingly. This compensates for the increase in air interface transmission time, bringing the total end-to-end latency of the task back to within the maximum tolerable latency of the service.

[0230] In some embodiments, when pruning the channels of the inference model, channels with a lower impact on the model prediction results can be pruned first, for example, pruned sequentially from low to high according to the statistical mean of the channel activation amplitude, so as to reduce the amount of computation while keeping the model prediction output within an acceptable range.

[0231] In some embodiments, the satellite-to-ground computing network resource allocation device expands the control sampling period predicted by the intelligent subject execution model according to the measured jitter data and a preset step size to obtain the adjusted control sampling period. The number of expansions and the result can be determined by the following formula:

[0232]

[0233] In this diagram, N_step represents the number of times the control sampling period is expanded, obtained by rounding up the measured jitter data relative to the baseline jitter; J_m represents the measured jitter data, i.e., the statistical average of the variation in the measured transmission delay between adjacent valid data packets; J0 represents the baseline jitter, i.e., the jitter level allowed for normal operation of the intelligent agent; Δ_c represents the control sampling period before adjustment, i.e., the time interval between two adjacent model predictions performed by the intelligent agent; Δ_step represents the preset step size, i.e., the fixed increment of a single expansion of the control sampling period; Δ_c′ represents the control sampling period after adjustment; and the symbol ⌈·⌉ represents the rounding up operation. The larger the measured jitter data, the more expansions are performed, and the longer the control sampling period after adjustment.

[0234] In some embodiments, after the control sampling period is extended, the time interval between two adjacent model predictions performed by the intelligent subject is extended accordingly, and the frequency of prediction requests per unit time is reduced synchronously with the uplink traffic. The intelligent subject maintains stable operation by actively slowing down or increasing the safety distance, so as to perform defensive degradation on the physical security dimension and avoid jitter affecting the motion control effect.

[0235] For example, the maximum tolerable latency of a certain intelligent entity's business is 20 milliseconds. After dual-transmission and selective reception, the actual transmission latency is 35 milliseconds, exceeding the latency limit by 15 milliseconds. Furthermore, there is no adjustable margin in bandwidth and transmission power. If the current number of channels in the inference model is 64, and the computational load per channel can only support a reduction of one-quarter based on the remaining computing power, the satellite-ground computing network resource allocation device performs channel reduction on the inference model according to the reduction factor, resulting in a reduced model with lower computational load. At the same time, the measured jitter data is 2.3 times the baseline jitter. The control sampling period is increased from 10 milliseconds to 30 milliseconds according to the preset step size, resulting in the adjusted control sampling period.

[0236] In some embodiments, the space-to-ground computing network resource configuration device updates the resource configuration matrix based on the computational load corresponding to the trimmed model and the adjusted control sampling period, resulting in an updated resource configuration matrix. The update involves substituting the computational load C′ corresponding to the trimmed model into the computational requirement parameters and the adjusted control sampling period Δ_c′ into the control state parameters, and re-executing periodic resource configuration analysis under constraints of latency, rate, and control stability. Since the computational load C′ is lower than before trimming, the computing power allocation result in the updated resource configuration matrix is ​​correspondingly reduced, allowing edge computing nodes to release their occupied core computing power for other intelligent entities within the target area to compete for. Because the adjusted control sampling period is extended, the bandwidth allocation result and link offloading decision are also adjusted synchronously with the reduction in service load. Subsequently, the space-to-ground computing network resource configuration device reconfigures communication resources and edge computing power resources within the space-to-ground computing network based on the updated resource configuration matrix, and substitutes the computational load corresponding to the trimmed model and the adjusted control sampling period into the multi-dimensional intent vector of the next time slot for use in the next analysis cycle.

[0237] Thus, when the measured transmission latency exceeds the maximum tolerable latency of the service and there are no adjustable communication resources, a pruning factor is calculated based on the latency excess, the current remaining computing power, and the current number of channels. The channels of the inference model are pruned according to the pruning factor to reduce the required computation. At the same time, the control sampling period is expanded by a preset step size based on the measured jitter data to enable the intelligent agent to perform defensive degradation. Then, the resource configuration matrix is ​​updated based on the computational load corresponding to the pruned model and the adjusted control sampling period, and the communication resources and edge computing power resources in the satellite-ground computing network are reconfigured. In this way, when there is no adjustment margin for communication resources, the incremental transmission latency is compensated by compressing the computation time and the jitter impact is suppressed by relaxing the prediction frequency, thus ensuring the determinism and operational stability of the transmission process of latency-sensitive services.

[0238] In some embodiments, when constructing the control state parameters based on the velocity vector, the state error, and the network jitter, the following implementation methods may also be adopted, for example: obtaining the state error vector and control input vector of the intelligent agent, wherein the state error vector represents the deviation between the actual motion trajectory and the desired motion trajectory; introducing a positive definite weight matrix to perform quadratic form calculations on the state error vector and the control input vector respectively to obtain the basic loss of control instability; constructing a jitter penalty term based on the network jitter and the jitter penalty term coefficient corresponding to the network jitter, and fusing the jitter penalty term with the basic loss of control instability to obtain a second-order control instability loss function; and constructing the control state parameters by combining the second-order control instability loss function with the upper bound of the control instability deviation.

[0239] In some embodiments, the satellite-to-ground computing network resource allocation device acquires the state error vector and control input vector of the intelligent agent. The state error vector, representing the deviation between the actual trajectory and the desired trajectory, can be calculated using the following formula:

[0240]

[0241] In this vector, x represents the state error vector of the intelligent agent; x_act represents the actual motion state of the intelligent agent, which can be read by the agent's motion sensors, and its components include position and velocity components along each coordinate axis; x_ref represents the desired motion state of the intelligent agent, determined by the desired trajectory given by the upper-level planning layer. The desired trajectory is the spatiotemporal trajectory that the intelligent agent should follow according to the task arrangement. The velocity vector, as a component of the actual motion state, participates in the composition of the state error vector. The faster the intelligent agent moves and the more frequently the path changes, the stronger the fluctuation of the state error vector.

[0242] In some embodiments, the control input vector is a vector composed of control quantities applied to the intelligent agent, the components of which include the drive command values ​​corresponding to each actuator, which can be read from the motion control unit of the intelligent agent; the magnitude of the control input vector reflects the control intensity exerted by the intelligent agent to correct state errors.

[0243] In some embodiments, the satellite-to-ground computing network resource allocation device introduces a positive definite weight matrix to perform quadratic calculations on the state error vector and the control input vector respectively, to obtain the basic loss of control instability, which can be calculated by the following formula:

[0244]

[0245] In this diagram, L_base represents the basic loss due to control instability; x represents the state error vector of the intelligent agent; u represents the control input vector of the intelligent agent; T represents the matrix transpose operation, where the vector is weighted by a positive definite weight matrix and then multiplied by itself to obtain the quadratic scalar loss; Q represents the positive definite weight matrix that weights the state error vector, with all eigenvalues ​​greater than zero, used to measure the influence of each component of the state error on control stability; for example, for operations sensitive to position deviation, the position component can be assigned a larger weight value; and R represents the positive definite weight matrix that weights the control input vector, with all eigenvalues ​​greater than zero, used to measure the influence of each component of the control quantity on control stability; for example, for intelligent agents sensitive to energy consumption, the control quantity components can be assigned a larger weight value. The value of the basic loss due to control instability increases with the increase of the state error vector and the control input vector, and a larger value indicates a higher degree of deviation from stability in the intelligent agent's control process.

[0246] In some embodiments, the satellite-to-ground computing network resource allocation device constructs a jitter penalty term based on network jitter and the jitter penalty term coefficient corresponding to network jitter, and fuses the jitter penalty term with the basic loss of control instability to obtain a second-order control instability loss function, which can be calculated by the following formula:

[0247]

[0248] In this diagram, L_C3 represents the second-order control instability loss function; x represents the state error vector of the intelligent agent; u represents the control input vector of the intelligent agent; Q and R are positive definite weight matrices that weight the state error vector and the control input vector, respectively; λ is the jitter penalty coefficient, a non-negative weight used to measure the impact of network jitter on control stability. For services with higher real-time control requirements, the jitter penalty coefficient is larger; δ represents network jitter, characterizing the degree of variation in data packet arrival time, which can be obtained from the historical arrival times of the link by the satellite-to-ground computing network resource configuration device. The jitter penalty λδ² increases with increasing network jitter. The greater the network jitter, the stronger the fluctuation in the arrival times of control commands and state feedback, and the higher the degree of disturbance to the intelligent agent's control process.

[0249] In some embodiments, the jitter penalty term and the basic loss of control instability are added together according to the above formula, so that the second-order control instability loss function simultaneously covers the three sources of instability: end-side motion deviation, control strength, and network-side jitter. The jitter factor on the network side is thus introduced into the quantitative evaluation of control stability.

[0250] For example, when an embodied intelligent robot performs a transport task in a park, its state error vector x increases due to path bends. At the same time, the network jitter δ of the link carrying the service increases, and the value of the second-order control instability loss function rises accordingly. If the robot's service has high requirements for real-time control, its jitter penalty term coefficient λ is larger, and the effect of network jitter on the rise of the second-order control instability loss function is correspondingly stronger.

[0251] In some embodiments, the satellite-to-ground computing network resource allocation device constructs the second-order control instability loss function and the upper bound of the control instability deviation as control state parameters. The upper bound of the control instability deviation characterizes the maximum deviation that the intelligent agent's control process can withstand, i.e., the upper limit that the value of the second-order control instability loss function is not allowed to exceed, and is determined according to the service's requirements for control stability. The control state parameters are jointly constructed by the current value L_C3 of the second-order control instability loss function and the upper bound of the control instability deviation; the former reflects the degree to which the intelligent agent currently deviates from stable operation, while the latter defines the boundary of stable operation.

[0252] In some embodiments, after the control state parameters are constructed, the upper bound of the control instability deviation carried by them is used to construct the control stability constraint in step 203, that is, to limit the value of the second-order control instability loss function to not exceed the upper bound of the control instability deviation, so that the allocation results of communication resources and edge computing power resources are subject to the reverse constraint of the control stability requirements of the intelligent subject's end side; the value of the second-order control instability loss function carried by them serves as the control dimension component of the multi-dimensional intent vector and participates in the rolling execution of periodic resource allocation parsing.

[0253] Thus, by obtaining the state error vector representing the deviation between the actual motion trajectory and the expected motion trajectory, as well as the control input vector, the basic loss of control instability is calculated through the quadratic form of the positive definite weight matrix. Then, the jitter penalty term, composed of network jitter and jitter penalty term coefficients, is fused to form a second-order control instability loss function. This function, together with the upper bound of the control instability deviation, constructs the control state parameters. This incorporates network-side jitter factors into the quantitative evaluation of control stability, enabling the resource allocation analysis of the satellite-ground computing network to perceive the intelligent subject's tolerance margin for transmission jitter. This provides a control dimension constraint basis for the coordinated allocation of communication resources and edge computing resources.

[0254] The following describes the satellite-to-ground computing network resource configuration method of this application through specific embodiments. It includes: the collaborative task guarantee method provided by this invention comprises five nested steps (STEP 1 to STEP 5) with a strictly logical progressive relationship. These are described in detail below with reference to embodiments and accompanying drawings. To facilitate understanding of the technical solution of this invention, the "3C bidirectional closed loop" involved in this invention is first defined through a cross-layer mapping: Communication dimension (Communication, C1): the air interface channel gain corresponding to the satellite-to-ground heterogeneous network (5G micro base station and NTN low-orbit satellite). , allocate bandwidth Transmission power and measured transmission delay With shaking Computation (C2): The frequency of computing power allocation to the corresponding edge MEC node. Task computation volume (FLOPs) and channel pruning factors for AI inference models Control dimension (C3): corresponds to the instantaneous quadratic control instability loss function on the embodied intelligent agent side. Model Predictive Control (MPC) Sampling Period and the prediction time-domain penalty weight matrix The parameters C1, C2, and C3 mentioned above form a two-way closed-loop flow in the forward resource calculation and the backward control feedback. Figure 5 This is a system overall hierarchical architecture diagram provided for an embodiment of this application. It illustrates the physical and logical layered architecture of the integrated space-air-ground computing network of the present invention, including a terminal / campus layer (multi-entity entities, Star Gateway), a network and edge computing power layer (5G base stations, Star Hub Platform, MEC), and a space layer (connecting low-orbit satellites of the galaxy). Figure 6 This application provides a 3C bidirectional closed-loop and collaborative control flowchart, illustrating the logical flow from STEP 1 to STEP 5, highlighting the closed-loop feedback chain (3C closed loop) of "business demand state prediction, resource optimization, DetNet execution, and adaptive reconstruction". It includes STEP 1 to STEP 5.

[0255] STEP 1: The system for recognizing the intent of multiple intelligent agents and constructing a cross-domain collaborative efficiency matrix within a time limit Periodically collect data from various entities within the park. The operational status transforms unstructured business intent graphs into multi-dimensional intent vectors containing computational, communication, and control attributes. :

[0256]

[0257] In the formula, and The main body exist The three-dimensional spatial coordinates and velocity vector at any given moment; The maximum end-to-end latency limit that the business can tolerate; Minimum guaranteed rate; The computational cost required to process a single task (in FLOPs). This is the upper bound of the maximum deviation that the control loop can tolerate; Prioritize business operations (1: Embodied Control, 2: High Awareness, 3: Conventional IoT). This step involves using multi-dimensional intent vectors. Deconstructed into 3C three-dimensional attributes, it realizes the parameterized representation of business requirements: 1. Communication constraints (C1): by (Maximum tolerable delay) and (Minimum rate) constitutes; 2. Computational attribute (C2): composed of... (Single task FLOPs computing power requirement) constitutes; 3. Control state (C3): composed of state error-based... With network jitter Constructed control instability loss function Composition. The above-mentioned collaborative efficiency function The weighted overheads of C1, C2, and C3 are explicitly integrated, providing a unified objective function for the state prediction in STEP 2 and the joint solution in STEP 3. For subjects with strong control properties, a second-order control instability loss function is constructed. :

[0258]

[0259] In the formula, and These are the agent's state error and the control input vector, respectively. , It is a positive definite weight matrix; For network latency jitter; Let the jitter penalty coefficient be used. Construct a normalized collaborative performance function. As a baseline target for subsequent optimization:

[0260]

[0261] In the formula, According to Dynamically adapted non-negative weights, and ; STEP 1 completes the mapping from "business requirements to mathematical constraints." Its output set of hard constraints is matched with real-time spatiotemporal coordinates. It will be directly injected into STEP 2 as an index variable for the physical channel prediction and computing power queuing model.

[0262] STEP 2: Spatiotemporal Evolution of Space-Air-Ground Topology and Joint Prediction of Channel-Computing Power State Based on Subject Three-Dimensional Coordinates Simultaneously acquire the set of ground micro base stations With the set of visible low-Earth orbit satellites in the ephemeris 5G terrestrial channel gain The model is as follows:

[0263]

[0264] In the formula, For reference path loss, This is the road damage index. For shadow fading amount, Base station coordinates. Satellite NTN channel gain. Using tilt angle Rice model:

[0265]

[0266] In the formula, At the speed of light, For carrier frequency, This represents the Earth-Satellite tilt distance. The Rice factor is modulated by the elevation angle.

[0267] At the same time, for the MEC nodes in the park Establish an M / M / c queuing model and predict the total task processing latency. for:

[0268]

[0269] In the formula, For the assigned calculation frequency, As the core number, For arrival rate, For service rate, The queuing probability is calculated. STEP 2 uses the physical layer model to solve for the system state parameters of the next time slot. The predicted channel gain matrix is ​​then calculated. With computing power queue latency It will be input as a constant matrix into the global optimization equation of STEP 3.

[0270] STEP 3: Joint solution of satellite-ground collaborative resources under control and management closed-loop constraints. Construct a mixed-integer nonlinear programming (MINLP) problem. Decision variables include: unloading path ratio. Bandwidth allocation Computing power allocation Transmission power The optimization objective is to maximize the overall system efficiency of the campus while minimizing total energy consumption.

[0271]

[0272] Constraints (partial complete mathematical expression): (Time Delay): (bandwidth): , (Computing power): (stability): This embodiment uses a Lagrange dual decomposition combined with a multi-agent deep reinforcement learning (MADDPG) algorithm to solve the aforementioned NP-hard problem, outputting the optimal resource allocation matrix within 3 milliseconds. The MINLP optimization equations in this step achieve a deep joint binding solution across the 3C dimensions:

[0273] Delayed Link (C1+C2): Total end-to-end delay is determined by the air interface transmission delay. Queuing and processing latency in edge computing Together they are superimposed;

[0274] Control safety limits (C3): The optimization process sets an upper bound on the instability. As a hard constraint, the optimal communication bandwidth is output through a unified algorithm solution. (C1) Transmission power (C1) and edge computing frequency (C2) ensures that air interface resources and edge computing resources always serve the control stability of the edge intelligent agent (C3).

[0275] The theoretically optimal resource allocation and routing were calculated. This included the offloading decision. The physical switching requirement of the link was triggered, which needs to be seamlessly executed by the underlying deterministic network protocol of STEP 4 to ensure that the theoretically optimal solution does not produce jitter at the physical layer.

[0276] STEP 4: DetNet-based satellite-ground cooperative dual-transmission selection and deterministic control with optimized instructions Determining the subject When switching from 5G to NTN or enabling dual-channel communication, a deterministic network (DetNet) dual-transmission and reception (PRE, Packet Replication and Elimination) mechanism is activated between the campus-side gateway and the edge-side platform. Packet replication and serialization: The gateway adds an incrementing sequence number to downlink / uplink IP packets. It is transmitted synchronously along both 5G and satellite paths. Clock compensation buffer: The satellite-to-ground gateway clock is synchronized using the IEEE 1588v2 protocol. This is due to satellite propagation delay. Greater than 5G latency Set a jitter compensation window at the receiving end. Elimination logic: The receiving end maintains a sliding deduplication window. For the arriving data packets :

[0277]

[0278] This completely masks air interface packet loss during satellite-to-ground physical link switching, outputting the measured deterministic residual transmission delay. With shaking Although STEP 4 ensures a smooth physical layer to the greatest extent possible, the absolute latency is limited by the physical distance between satellites. It may still rise. Therefore, STEP 4 will use the underlying measured indicators. and Feedback is sent to the upper-layer application, triggering the closed-loop reconstruction in STEP 5.

[0279] Figure 7 This application provides a schematic diagram of a DetNet-based satellite-to-ground dual-transmission and selective reception (PRE) mechanism, detailing the execution process of the STEP 4 micro-protocol layer. It demonstrates packet serialization and tagging at the gateway, dual-path transmission of 5G and satellite (with different latency), and the deduplication sliding window and clock alignment mechanism at the receiver.

[0280] STEP 5: The 3C closed-loop driven adaptive model dimensionality reduction and control reconstruction system initiates inverse adaptation of the application layer and control layer based on the real-time physical network state feedback from STEP 4: 1. Edge computing layer model pruning: When Furthermore, when there are no resources available for adjustment on the pure communication side, the edge platform applies channel pruning to the large model executing inference, with the pruning factor... :

[0281]

[0282] The required computing power is reduced after trimming. This reduces the time consumption of edge computing and compensates for the increase in air interface latency. The measured transmission latency of the DetNet protocol deployed in STEP 4 is... Severe network jitter may occur. When (C1 dimension degradation), the system triggers "C1". Reverse closed-loop adjustment of "C2 / C3":

[0283] Communication feedback computing (C1) The edge platform calculates the AI ​​large model channel pruning factor based on the measured transmission timeout. Forcibly compress the required computing power to By reducing computation time Hard compensation for air interface transmission time The increase brought the total latency back to [previous value]. Within the safety line. Communication feedback control (C1) The edge-side embodied intelligent agent is based on network residual jitter. Adaptive extension of MPC control sampling period And reconstruct the penalty weight matrix. Intelligent agents achieve a defensive degradation in the physical security dimension by actively slowing down or increasing the safe distance. (Adjusted) and Reinsert into the next time slot Intent vector In the middle (returning to STEP 1), "state measurement" was implemented. Dynamic intervention The complete 3C closed loop of "parameter update". Control-end MPC reconstruction: The Model Predictive Controller (MPC) of the embodied intelligent agent adaptively extends the control sampling period according to the channel conditions. Simultaneously, the penalty weights in the prediction time domain are reconstructed. This prompts the agent to proactively slow down to maintain global stability. Finally, the updated... and Resubstitute the intent vector for the next time slot Then, return to STEP 1 to form a tight closed loop of communication, computing, and control.

[0284] It should be noted that the above-described method embodiments, or the various possible implementations of the method embodiments, can be executed individually, or, provided there is no conflict, they can be combined with each other. The specific implementation can be determined according to actual usage requirements, and this application embodiment does not impose any restrictions on this.

[0285] As can be seen, the above mainly describes the solutions provided by the embodiments of this application from a methodological perspective. To achieve the above functions, the embodiments of this application provide corresponding hardware structures and / or software modules for executing each function. Those skilled in the art should readily recognize that, in conjunction with the modules and algorithm steps of the various examples described in the embodiments disclosed herein, the embodiments of this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0286] This application embodiment can divide the satellite-to-ground computing network resource configuration method into functional modules based on the above method example. For example, each function can be divided into its own functional module, or two or more functions can be integrated into one processing module. The integrated module can be implemented in hardware or as a software functional module. Optionally, the module division in this application embodiment is illustrative and only represents one logical functional division; other division methods may be used in actual implementation.

[0287] In some embodiments, this application also provides a satellite-to-ground computing network resource configuration apparatus. This apparatus may include one or more functional modules for implementing the satellite-to-ground computing network resource configuration method described in the above embodiments.

[0288] For example, Figure 8 This is a schematic diagram of a satellite-to-ground computing network resource allocation device provided in an embodiment of this application. Figure 8 As shown, the satellite-to-ground computing network resource allocation device 900 includes:

[0289] The acquisition module 901 is used to acquire the multi-dimensional intent vector corresponding to the intelligent subject within the target area. The multi-dimensional intent vector includes three-dimensional spatial coordinates, communication constraint parameters, computational requirement parameters, and control state parameters.

[0290] The calculation module 902 is used to calculate the channel gain of the terrestrial communication link and the satellite communication link based on the above-mentioned multi-dimensional intent vectors, and to calculate the task processing delay of the edge computing node.

[0291] The parsing module 903 is used to perform periodic resource configuration parsing based on the above-mentioned communication constraint parameters, the above-mentioned control state parameters, the above-mentioned channel gain and the above-mentioned task processing delay to obtain a resource configuration matrix. The resource configuration matrix is ​​used to control the coordinated configuration of communication resources and edge computing resources of the satellite-ground computing network in the target area. The resource configuration matrix includes link offloading decision, bandwidth allocation result, computing power allocation result and transmission power result.

[0292] Processing module 904 is used to perform dual-transmission selective reception and delay compensation processing on the ground communication link and the satellite communication link when the above-mentioned link offloading decision instructs the intelligent subject to switch the communication link between the above-mentioned ground communication link and the above-mentioned satellite communication link, so as to obtain the measured transmission delay and measured jitter data.

[0293] The update module 905 is used to update the resource configuration matrix based on the measured transmission delay and the measured jitter data.

[0294] Configuration module 906 is used to configure the communication resources and edge computing resources within the aforementioned satellite-to-ground computing network based on the updated resource configuration matrix.

[0295] The satellite-to-ground computing network resource allocation device provided in this application acquires multi-dimensional intent vectors corresponding to intelligent entities within a target area. These multi-dimensional intent vectors include three-dimensional spatial coordinates, communication constraint parameters, computing requirement parameters, and control state parameters. This parameterizes and structures the business requirements of intelligent entities in the three dimensions of communication, computing, and control, providing a unified requirement input for subsequent resource allocation and avoiding coarse-grained perception of business requirements. Based on this, the device calculates the channel gain of the ground communication link and the satellite communication link, as well as the task processing delay of the edge computing node, based on the aforementioned multi-dimensional intent vectors. This involves jointly predicting the satellite-to-ground channel state and computing queue state for the next time slot, ensuring that resource allocation is based on the real-time state of the physical and computing layers, thus improving the matching accuracy between the calculation results and the actual operating environment. Furthermore, based on the aforementioned communication constraint parameters, control state parameters, channel gain, and task processing delay, the device performs periodic resource allocation analysis to obtain a resource allocation matrix including link offloading decisions, bandwidth allocation results, computing power allocation results, and transmission power results. This enables the integration of satellite-to-ground communication resources and edge computing resources. The coordinated allocation ensures that resource allocation decisions are simultaneously constrained by service latency tolerance and control stability requirements, guaranteeing the match between resource allocation results and the intelligent agent's service needs. When the aforementioned link offloading decision instructs the intelligent agent to switch communication links between ground and satellite links, dual-transmission and selective reception with latency compensation are performed on both ground and satellite communication links. This ensures that both ground and satellite links synchronously carry data during the switchover and achieve seamless connection through deduplication and latency compensation. This eliminates the inevitable transmission latency spikes and batch packet losses during cross-domain link cutovers caused by the traditional "break-then-switch" method. In this process, measured transmission latency and jitter data are obtained. Finally, based on the measured transmission latency and jitter data, the resource allocation matrix is ​​updated, and the communication resources and edge computing resources within the satellite-ground computing network are configured based on the updated matrix. This allows the resource allocation to be dynamically adjusted according to the actual transmission status of the links, ensuring that the resource allocation continuously matches the actual link status and guaranteeing the continuity and determinism of the intelligent agent's transmission process before and after the communication link switchover.

[0296] In other embodiments, the update module 905 is further configured to:

[0297] Based on the measured transmission delay and jitter data, the pruning factor of the inference model in the edge computing node is obtained and the control sampling period of the intelligent agent is adjusted. The resource configuration matrix is ​​updated according to the pruning factor and the adjusted control sampling period to obtain the updated resource configuration matrix. The communication resources and edge computing resources within the satellite-ground computing network are then reconfigured based on the updated resource configuration matrix.

[0298] In other embodiments, the acquisition module 901 described above is specifically used for:

[0299] Collect the three-dimensional spatial coordinates, velocity vector, maximum tolerable latency, minimum guaranteed rate, single task computation amount, upper limit of control instability deviation, and service priority identifier of the intelligent subject;

[0300] The communication constraint parameters are constructed based on the maximum tolerable latency and the minimum guaranteed rate of the above services. The computational requirement parameters are constructed based on the computational amount of a single task. The control state parameters are constructed based on the velocity vector, the state error and the network jitter.

[0301] By combining the aforementioned three-dimensional spatial coordinates, communication constraint parameters, computational requirement parameters, and control state parameters according to preset dimensions, a structured multi-dimensional intent vector is obtained.

[0302] In other embodiments, the computing module 902 is specifically used for:

[0303] Based on the aforementioned multi-intention vectors and ground base station coordinates, combined with reference path loss, path loss index, and shadow fading based on measured channel sample statistics, the channel gain of the aforementioned ground communication link relative to the aforementioned multi-agent is calculated.

[0304] Based on the aforementioned multi-intention vectors and satellite ephemeris data, combined with carrier frequency, satellite-to-ground tilt distance, and Rice factor modulated by elevation angle, the channel gain of the aforementioned satellite communication link relative to the aforementioned multi-intelligence agent is calculated using the Rice model.

[0305] Based on the number of cores, task arrival rate, and service rate of the aforementioned edge computing nodes, the queuing probability is calculated. Based on the queuing probability, the waiting latency and processing latency of the aforementioned intelligent subject's task on the edge computing nodes are calculated. Based on the waiting latency and processing latency, the task processing latency is determined.

[0306] In other embodiments, the parsing module 903 is specifically used for:

[0307] Decision variables are constructed based on link offload ratio, bandwidth allocation value, computing power allocation value, and transmit power value;

[0308] A constant matrix is ​​constructed based on the aforementioned channel gain and task processing delay.

[0309] Based on the maximum tolerable delay of the above-mentioned service in the above-mentioned communication constraint parameters, a delay constraint is constructed; based on the minimum guaranteed rate in the above-mentioned communication constraint parameters, a rate constraint is set; and based on the upper bound of the control instability deviation in the above-mentioned control state parameters, a control stability constraint is constructed.

[0310] Under the constraints of the aforementioned time delay, rate, and control stability, periodic resource allocation analysis is performed on the aforementioned decision variables and constant matrix to obtain the aforementioned resource allocation matrix.

[0311] In other embodiments, the processing module 904 described above is specifically used for:

[0312] The data packets carrying the incrementing sequence number are sent synchronously along the aforementioned ground communication link and the aforementioned satellite communication link, resulting in two arriving data packets;

[0313] Based on the time synchronization protocol, clock synchronization is performed on the two incoming data packets to obtain the propagation delay difference between the two links, and a jitter compensation window is set based on the propagation delay difference.

[0314] Within the jitter compensation window mentioned above, deduplication is performed on the data packets arriving from both links, retaining the first valid data packet that arrives.

[0315] The measured transmission delay and jitter data were calculated based on the start and arrival times of the end-to-end transmission of the aforementioned valid data packets.

[0316] In other embodiments, the update module 905 is specifically used for:

[0317] When the measured transmission delay exceeds the maximum tolerable delay of the service and there are no adjustable communication resources, the pruning factor is calculated based on the delay excess, the current remaining computing power of the edge computing node, and the current number of channels in the above inference model.

[0318] The channels of the above inference model are pruned according to the above pruning factor to obtain the pruned model, which requires less computation.

[0319] Based on the measured jitter data, the control sampling period predicted by the above-mentioned intelligent subject execution model is increased by a preset step size to obtain the above-mentioned adjusted control sampling period.

[0320] Based on the computational cost of the trimmed model and the adjusted control sampling period, the resource configuration matrix is ​​updated to obtain the updated resource configuration matrix.

[0321] In other embodiments, the acquisition module 901 described above is specifically used for:

[0322] Obtain the state error vector and control input vector of the intelligent agent. The state error vector represents the deviation between the actual motion trajectory and the desired motion trajectory.

[0323] By introducing a positive definite weight matrix, quadratic form calculations are performed on the aforementioned state error vector and the aforementioned control input vector respectively to obtain the basic loss of control instability.

[0324] Based on the above network jitter and the jitter penalty term coefficients corresponding to the above network jitter, a jitter penalty term is constructed. The jitter penalty term is then fused with the above control instability basic loss to obtain the second-order control instability loss function.

[0325] The above-mentioned second-order control instability loss function and the above-mentioned upper bound of control instability deviation are used to construct the above-mentioned control state parameters.

[0326] It should be noted that the satellite-to-ground computing network resource configuration device can implement all the processes implemented in the above method embodiments and achieve the same beneficial effects. To avoid repetition, it will not be described again here.

[0327] In the case where the functions of the integrated modules described above are implemented in hardware, this application provides a possible structural schematic diagram of the electronic device involved in the above embodiments. For example... Figure 9 As shown, the electronic device 90 includes: a processor 92, a communication interface 93, and a bus 94. Optionally, the electronic device 90 may also include a memory 91.

[0328] Processor 92 may implement or execute various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 92 may be a central processing unit, a general-purpose processor, a digital signal processor, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It may implement or execute various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 92 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0329] Communication interface 93 is used to connect with other devices via a communication network. This communication network can be Ethernet, wireless access network, wireless local area network (WLAN), etc.

[0330] The memory 91 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto.

[0331] As one possible implementation, the memory 91 can exist independently of the processor 92. The memory 91 can be connected to the processor 92 via a bus 94 and is used to store instructions or program code. When the processor 92 calls and executes the instructions or program code stored in the memory 91, it can implement the satellite-to-ground computing network resource configuration method provided in this application embodiment.

[0332] In another possible implementation, memory 91 can also be integrated with processor 92.

[0333] Bus 94 can be an Extended Industry Standard Architecture (EISA) bus, etc. Bus 94 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 9 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0334] Through the above description of the implementation methods, those skilled in the art can clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the service calling device can be divided into different functional modules to complete all or part of the functions described above.

[0335] This application embodiment also provides a chip, which includes a processor and a communication interface. The communication interface and the processor are coupled. The processor is used to run programs or instructions to implement the various processes of the above-described satellite-ground computing network resource configuration method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0336] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0337] This application also provides a readable storage medium storing a program or instructions that, when executed by a computer, implement the satellite-to-ground computing network resource configuration method provided in the above embodiments. It is understood that all or part of the processes in the above method embodiments can be executed by computer instructions instructing related hardware; the readable storage medium can be any of the foregoing embodiments or memory; the readable storage medium can also be an external storage device of the service invocation device, such as a pluggable hard drive, SmartMedia Card (SMC), Secure Digital (SD) card, flash card, etc., equipped on the service invocation device. Further, the readable storage medium can include both internal storage units of the service invocation device and external storage devices. The readable storage medium is used to store the computer program and other programs and data required by the service invocation device. The readable storage medium can also be used to temporarily store data that has been output or will be output.

[0338] This application also provides a computer program product, which is stored in a storage medium and, when executed by a computer, implements the satellite-to-ground computing network resource configuration method provided in the above embodiments.

[0339] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0340] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of 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 this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a computer 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 (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0341] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method for allocating resources in a satellite-to-ground computing network, characterized in that, include: Obtain the multi-dimensional intent vector corresponding to the intelligent subject within the target area. The multi-dimensional intent vector includes three-dimensional spatial coordinates, communication constraint parameters, computational requirement parameters, and control state parameters. Based on the multivariate intent vectors, calculate the channel gain of the terrestrial communication link and the satellite communication link respectively, and calculate the task processing latency of the edge computing node; Based on the communication constraint parameters, the control state parameters, the channel gain, and the task processing delay, periodic resource configuration analysis is performed to obtain a resource configuration matrix. The resource configuration matrix is ​​used to control the coordinated configuration of communication resources and edge computing resources of the satellite-ground computing network within the target area. The resource configuration matrix includes link offloading decisions, bandwidth allocation results, computing power allocation results, and transmit power results. When the link offloading decision instructs the intelligent agent to switch communication links between the ground communication link and the satellite communication link, dual-transmission selective reception and delay compensation processing is performed on the ground communication link and the satellite communication link to obtain measured transmission delay and measured jitter data; The resource configuration matrix is ​​updated based on the measured transmission delay and the measured jitter data; Configure the intra-satellite network communication resources and edge computing resources based on the updated resource configuration matrix.

2. The satellite-to-ground computing network resource allocation method according to claim 1, characterized in that, The step of updating the resource configuration matrix based on the measured transmission delay and the measured jitter data includes: Based on the measured transmission delay and the measured jitter data, the pruning factor of the inference model in the edge computing node is obtained and the control sampling period of the intelligent agent is adjusted. The resource configuration matrix is ​​updated according to the pruning factor and the adjusted control sampling period to obtain the updated resource configuration matrix. The communication resources and edge computing resources within the satellite-ground computing network are reconfigured based on the updated resource configuration matrix.

3. The satellite-to-ground computing network resource allocation method according to claim 1, characterized in that, The acquisition of the multi-dimensional intent vectors corresponding to intelligent agents within the target area includes: Collect the three-dimensional spatial coordinates, velocity vector, maximum tolerable latency, minimum guaranteed rate, single task computation amount, upper limit of control instability deviation, and service priority identifier of the intelligent subject; The communication constraint parameters are constructed based on the maximum tolerable latency of the service and the minimum guaranteed rate; the computational requirement parameters are constructed based on the computational amount of a single task; and the control state parameters are constructed based on the velocity vector, the state error, and the network jitter. By combining the three-dimensional spatial coordinates, the communication constraint parameters, the calculation requirement parameters, and the control state parameters according to preset dimensions, a structured multi-dimensional intent vector is obtained.

4. The satellite-to-ground computing network resource allocation method according to claim 1, characterized in that, The calculation of channel gain for terrestrial and satellite communication links based on the multivariate intent vectors, and the calculation of task processing latency for edge computing nodes, include: Based on the multi-intention vector and the coordinates of the ground base station, combined with the reference path loss, path loss index and shadow fading based on measured channel sample statistics, the channel gain of the ground communication link relative to the multi-intelligence agent is calculated. Based on the multi-entity intent vector and satellite ephemeris data, combined with carrier frequency, satellite-to-ground tilt distance and Rice factor modulated by elevation angle, the channel gain of the satellite communication link relative to the multi-entity agent is calculated using the Rice model; The queuing probability is calculated based on the number of cores, task arrival rate, and service rate of the edge computing node. The waiting latency and processing latency of the intelligent subject's task on the edge computing node are calculated based on the queuing probability. The task processing latency is determined based on the waiting latency and the processing latency.

5. The satellite-to-ground computing network resource allocation method according to claim 1, characterized in that, The process of performing periodic resource allocation parsing based on the communication constraint parameters, the control state parameters, the channel gain, and the task processing delay yields a resource allocation matrix, including: Decision variables are constructed based on link offload ratio, bandwidth allocation value, computing power allocation value, and transmit power value; A constant matrix is ​​constructed based on the channel gain and the task processing delay; A delay constraint is constructed based on the maximum tolerable delay of the service in the communication constraint parameters; a rate constraint is set based on the minimum guaranteed rate in the communication constraint parameters; and a control stability constraint is constructed based on the upper bound of the control instability deviation in the control state parameters. Under the constraints of the time delay constraint, the rate constraint, and the control stability constraint, periodic resource allocation analysis is performed on the decision variables and the constant matrix to obtain the resource allocation matrix.

6. The satellite-to-ground computing network resource allocation method according to claim 1, characterized in that, The process of performing dual-transmission selective reception and delay compensation processing on the ground communication link and the satellite communication link to obtain measured transmission delay and measured jitter data includes: The data packets carrying incremental sequence numbers are sent synchronously along the ground communication link and the satellite communication link to obtain two incoming data packets; The two incoming data packets are clock-synchronized based on a time synchronization protocol to obtain the propagation delay difference between the two links, and a jitter compensation window is set based on the propagation delay difference. Within the jitter compensation window, deduplication is performed on data packets arriving from both links, retaining the first arriving valid data packet; The measured transmission delay and the measured jitter data are calculated based on the start time and arrival time of the end-to-end transmission of the valid data packets.

7. The satellite-to-ground computing network resource allocation method according to claim 2, characterized in that, The step involves obtaining a pruning factor for the inference model in the edge computing node based on the measured transmission delay and the measured jitter data, adjusting the control sampling period of the intelligent agent, and updating the resource configuration matrix based on the pruning factor and the adjusted control sampling period to obtain the updated resource configuration matrix, including: When the measured transmission delay exceeds the maximum tolerable delay of the service and there are no adjustable communication resources, a pruning factor is calculated based on the delay excess, the current remaining computing power of the edge computing node, and the current number of channels in the inference model. The channels of the inference model are pruned according to the pruning factor to obtain the pruned model, which requires less computation. Based on the measured jitter data, the control sampling period predicted by the intelligent subject execution model is increased by a preset step size to obtain the adjusted control sampling period. The resource configuration matrix is ​​updated based on the computational cost of the trimmed model and the adjusted control sampling period to obtain the updated resource configuration matrix.

8. The satellite-to-ground computing network resource allocation method according to claim 3, characterized in that, The construction of the control state parameters based on the velocity vector, the state error, and the network jitter includes: The state error vector and control input vector of the intelligent agent are obtained, wherein the state error vector represents the deviation between the actual motion trajectory and the desired motion trajectory; By introducing a positive definite weight matrix, quadratic form calculations are performed on the state error vector and the control input vector respectively to obtain the basic loss of control instability; A jitter penalty term is constructed based on the network jitter and the jitter penalty term coefficient corresponding to the network jitter. The jitter penalty term is then fused with the basic loss of control instability to obtain a second-order control instability loss function. The second-order control instability loss function and the upper bound of the control instability deviation are used to construct the control state parameters.

9. A satellite-to-ground computing network resource allocation device, characterized in that, include: The acquisition module is used to acquire the multi-dimensional intent vector corresponding to the intelligent subject within the target area. The multi-dimensional intent vector includes three-dimensional spatial coordinates, communication constraint parameters, computational requirement parameters, and control state parameters. The calculation module is used to calculate the channel gain of the terrestrial communication link and the satellite communication link based on the multivariate intent vectors, and to calculate the task processing latency of the edge computing node. The parsing module is used to perform periodic resource configuration parsing based on the communication constraint parameters, the control state parameters, the channel gain and the task processing delay to obtain a resource configuration matrix. The resource configuration matrix is ​​used to control the coordinated configuration of communication resources and edge computing resources of the satellite-ground computing network in the target area. The resource configuration matrix includes link offloading decisions, bandwidth allocation results, computing power allocation results and transmission power results. The processing module is used to perform dual-transmission selective reception and delay compensation processing on the ground communication link and the satellite communication link when the link offloading decision instructs the intelligent subject to switch the communication link between the ground communication link and the satellite communication link, so as to obtain the measured transmission delay and measured jitter data. The update module is used to update the resource configuration matrix based on the measured transmission delay and the measured jitter data; The configuration module is used to configure the communication resources and edge computing resources within the satellite-to-ground computing network based on the updated resource configuration matrix.

10. An electronic device, characterized in that, It includes a processor and a memory, the memory storing programs or instructions that can run on the processor, the programs or instructions being executed by the processor to implement the satellite-to-ground computing network resource configuration method as described in any one of claims 1-8.

11. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions, which, when executed by a computer, implement the satellite-to-ground computing network resource configuration method as described in any one of claims 1-8.

12. A computer program product, characterized in that, The computer program product is stored in a storage medium, and when the computer program product is executed by a computer, it implements the satellite-to-ground computing network resource configuration method as described in any one of claims 1-8.