Virtual-real combined sky-ground multi-domain heterogeneous network cooperative computing simulation system and method

Abstract

The application discloses a kind of virtual and real sky-ground multi-domain heterogeneous network cooperative computing simulation system and method, system includes: heterogeneous computing power prototype subsystem, virtualization simulation subsystem, spaceborne computing terminal, programmable network link simulator, heterogeneous resource management subsystem and simulation verification system interface.Real hardware computing power characteristics are introduced by physical board card, large-scale expansion is realized by virtual node, the authenticity and scale of simulation are considered;Link parameters are generated according to high-precision orbit dynamics model, which greatly improves the fidelity of sky-ground network simulation;Based on hybrid container orchestration technology, unified management and intelligent scheduling of sky, space and ground heterogeneous resources are realized, providing a powerful platform for verifying advanced cross-domain cooperative algorithms;A set of system can support networking routing, task offloading, distributed cooperative computing, constellation cluster control and other simulation verification requirements, reducing development cost and complexity.

Description

Technical Field

[0001] This invention belongs to the technical field of integrated sky-ground networks, and particularly relates to a collaborative computing simulation system and method for multi-domain heterogeneous networks that combine virtual and real elements. Background Technology

[0002] Information globalization is accelerating the transformation of the information service industry, and the world has entered a new stage of large-scale satellite deployment and intelligent upgrading in parallel. With the rapid integration and development of space-air-ground integrated networks, application scenarios are expanding to multi-domain space networking and deep multi-user linkages across land, sea, air, and space, potentially providing ubiquitous access and on-orbit edge computing services to global users. However, the heterogeneity, dynamism, and resource dispersion of space-air-ground integrated networks pose significant challenges to network protocol design and task scheduling algorithm verification.

[0003] The publicly disclosed related patents include CN111447025B, "A System-Level Simulation Platform and Construction Method for Satellite Mobile Communication" (focusing on pure software network simulation). Pure software simulation cannot accurately reflect the computing characteristics of heterogeneous hardware, while pure hardware testing has limited scale. CN202210344692.6, "A Space-Ground Integrated Network Simulation Test System and Test Method" (focusing on network performance evaluation); and CN117082551B, "Simulation System, Method, and Computer-Readable Medium for Space-Ground Integrated Satellite Network" (focusing on high-concurrency service network performance simulation of satellite networks, but lacking specific features). Existing technologies are mostly focused on single-domain simulation, emphasizing either space-ground network performance or only computational offloading or satellite-ground micro-clouds. They cannot comprehensively support integrated simulation of space-based, air-based, and ground-based multi-domain systems, and lack a unified verification platform for cross-domain computing power scheduling and heterogeneous network collaborative computing. Examples include: CN114500579B "A Subscription-Publish-Based Autonomous Service Discovery and Collaboration Method for Space-Based Computation" (focusing on compute offloading, lacking integrated space-ground network collaboration); CN119225209B "A Hardware-in-the-Board Simulation System and Method for Low-Earth Orbit Satellite Cluster Edge Micro-Clouds" (focusing on satellite cluster edge micro-cloud performance evaluation, lacking network topology simulation and collaborative computing); and CN19557067B "Method, Device, and Storage Medium for Computing Resource Scheduling in Multi-Cloud Scenarios" (focusing on computing power scheduling, lacking heterogeneous computing power characteristics and network topology simulation).

[0004] Therefore, there is an urgent need in this field for a collaborative computing simulation system that can combine real heterogeneous computing power with virtual large-scale simulation and accurately simulate time-varying network environments between the sky and the ground. Summary of the Invention

[0005] The purpose of this invention is to provide a virtual-physical multi-domain heterogeneous network collaborative computing simulation system and method that can effectively integrate physical and virtual nodes, simulate dynamic time-varying networks, and realize cross-domain collaborative computing scheduling.

[0006] To solve the above problems, the technical solution of the present invention is as follows: A virtual-real hybrid multi-domain heterogeneous network collaborative computing simulation system includes: The heterogeneous computing power prototype subsystem consists of several physical heterogeneous computing power boards, which are used to represent the actual computing power nodes in space, air, and ground respectively. The virtualization simulation subsystem runs on the workstation and is used to generate virtual nodes corresponding to the sky, air, and ground nodes through virtual machines / containers to expand the simulation scale. The onboard computing terminal is an on-orbit heterogeneous computing node deployed on a real satellite or satellite simulation module, used to inject real onboard computing power into the simulation network. The programmable network link simulator is interconnected with the heterogeneous computing power prototype subsystem, the virtualization simulation subsystem, and the spaceborne computing terminal. It is used to generate link delay, bit error rate, packet loss rate, and topology change parameters in real time based on orbital dynamics or dynamic scenario scripts, so as to simulate the highly dynamic time-varying links and network topology between the sky and the ground with high realism. The heterogeneous resource management subsystem is deployed across ground cloud centers, virtual nodes, and physical computing nodes. It is used to manage physical and virtual nodes into a unified cloud-edge-device resource pool through a multi-level container orchestration engine, and to complete cross-domain computing power scheduling, task unloading, and container migration based on network status and task requirements. The simulation verification system interface interacts with the heterogeneous resource management subsystem and the programmable network link simulator. It is used to input simulation task scripts, output performance indicators, and support closed-loop verification of networking routing protocols, distributed computing offloading strategies, and constellation cluster management algorithms.

[0007] According to an embodiment of the present invention, a single physical node of the heterogeneous computing power prototype subsystem includes: Container engine; At least one heterogeneous computing board selected from D2000, RK3588, FT7004, neuromorphic chip, Atlas200, and Raspberry Pi; Wired / wireless high-speed bus interface for communication with a programmable network link simulator; Each physical computing node is configured differently using container images to play the roles of space-based, air-based, or ground-based systems.

[0008] According to one embodiment of the present invention, the virtualization simulation subsystem uses the VMware virtualization platform and runs within each virtual machine: Virtual computing power unit; Virtual network adapter; And a container engine for quickly cloning and generating large-scale virtual nodes in the sky, air, and ground.

[0009] According to an embodiment of the present invention, the spaceborne computing terminal includes: The heterogeneous computing module integrates at least two edge heterogeneous processors: the RK3588 and a neuromorphic chip. The primary and backup management module adopts a dual-processor cold / hot backup architecture for on-orbit fault switching; The routing protocol module is used to load the inter-satellite / satellite-to-ground dynamic routing protocol stack; The communication interface module provides an Ethernet interface to enable data backhaul with the programmable network link simulator. The primary and backup management module deploys a container engine, in which trajectory prediction and task unloading decision-making business applications run.

[0010] According to an embodiment of the present invention, the programmable network link simulator includes: The input terminal is used to receive orbital ephemeris, maneuvering strategies, or user-defined dynamic scenario scripts. The spatiotemporal link parameter database is used to store prior or real-time measurement parameters of inter-satellite, satellite-to-ground, and air-to-ground links. The calculation module is used to calculate the link distance, Doppler shift, and visibility matrix based on the orbital dynamics model; The topology and link status configuration module is used to convert the calculated link parameters into link latency, bandwidth, bit error rate, packet loss rate, and interruption events. The link state simulation and network topology simulation module is used to apply real-time impairment and forwarding rules to data packets between real / virtual nodes based on the parameters output by the topology and link state configuration module. The scene display module is used to graphically output dynamic topology and link quality curves.

[0011] According to an embodiment of the present invention, the heterogeneous resource management subsystem further includes: Resource scheduling middleware is used to parse task requirement constraints, node remaining computing power, and network link quality to generate cross-domain offloading decisions; The cloud-edge collaborative controller is used to upload computing tasks to satellite or airborne nodes when the link quality is higher than a threshold, and to trigger container migration back to the ground when the link quality is lower than a threshold. Edge-to-edge collaboration controllers are used to implement P2P distribution and load balancing of container images between adjacent nodes within a constellation or cluster; Edge-to-device collaborative controllers are used to offload edge computing subtasks to drones or ground terminals that are directly connected to them.

[0012] According to an embodiment of the present invention, the simulation verification system interface is configured as follows: Provide RESTful API and / or gRPC services for script injection based on orbital dynamics or dynamic scenarios; Performance probes are used to collect throughput, end-to-end latency, task completion time, CPU / GPU utilization, and energy consumption metrics. A visual evaluation panel is used to compare performance curves under different routing, scheduling, or offloading strategies and output test reports.

[0013] A simulation method for collaborative computing of heterogeneous networks across multiple domains (sky and ground) that combines virtual and real-world applications includes: System construction steps: Interconnect the heterogeneous computing power prototype subsystem, virtualization simulation subsystem and spaceborne computing terminal through a programmable network link simulator, configure the initial link parameters according to orbit data or dynamic scenario scripts, and form a unified sky-ground multi-domain simulation network environment. Resource management steps: Deploy container orchestration engines on ground cloud centers, virtual nodes, and physical / spaceborne nodes, register all nodes to the same cluster, and build a unified cloud-edge-device resource pool; Scenario configuration steps: Input the network topology change script, computation task dependency graph and resource requirement constraints through the simulation verification system interface to generate the collaborative computing scenario to be verified; Dynamic simulation and scheduling steps: Start the programmable network link simulator and update the link status in real time according to the script; the heterogeneous resource management subsystem dynamically executes cross-domain task scheduling, container migration and compute unloading based on the real-time network status and task requirements; Data collection and analysis steps: Collect network performance and computing performance indicators, perform statistical and visual comparisons, and output an algorithm verification report.

[0014] According to an embodiment of the present invention, the dynamic simulation and scheduling step further includes: When an inter-satellite link interruption or latency mutation exceeding a set threshold is detected, a container snapshot is generated and the system is relocated back to the ground cloud center or other visible satellite nodes via a backup link. When the computational load of the airborne node is detected to exceed the limit, some inference subtasks will be offloaded to the neighboring drone node in an edge-to-edge collaborative manner. When the quality of the link between the ground terminal and the airborne node is detected to be better than that of the satellite-to-ground link, the calculation results are prioritized through the airborne node.

[0015] According to an embodiment of the present invention, the data collection and analysis step further includes: Use time-series database storage performance metrics; Using machine learning models to perform regression analysis on the collected data, we can predict bottleneck nodes under different constellation sizes or task intensities. The prediction results are fed back to the scenario configuration step to optimize the next round of simulation scripts and initial resource deployment strategies.

[0016] Because the present invention adopts the above technical solution, it has the following advantages and positive effects compared with the prior art: The virtual-physical hybrid space-air-ground multi-domain heterogeneous network collaborative computing simulation system in one embodiment of the present invention introduces the characteristics of real hardware computing power through physical boards and achieves large-scale expansion through virtual nodes, thus balancing the realism and scalability of the simulation. Link parameters are generated based on a high-precision orbital dynamics model, greatly improving the realism of the space-air-ground network simulation. Based on hybrid container orchestration technology, unified management and intelligent scheduling of heterogeneous resources in space, air, and ground are achieved, providing a powerful platform for verifying advanced cross-domain collaborative algorithms. A single system can support various simulation verification needs such as network routing, task offloading, distributed collaborative computing, and constellation cluster management, reducing R&D costs and complexity. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall architecture of a virtual-real combined sky-ground multi-domain heterogeneous network collaborative computing simulation system according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the composition of a heterogeneous computing power prototype subsystem in one embodiment of the present invention; Figure 3 This is a schematic diagram of node mapping in a virtualization simulation subsystem according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the composition of a spaceborne computing terminal in one embodiment of the present invention; Figure 5 This is a schematic diagram illustrating the dynamic topology configuration principle of a programmable network link simulator in one embodiment of the present invention. Figure 6 This is a flowchart of a simulation method according to an embodiment of the present invention. Detailed Implementation

[0018] The following detailed description, in conjunction with the accompanying drawings and specific embodiments, provides a further detailed explanation of the proposed invention: a virtual-real combined sky-ground multi-domain heterogeneous network collaborative computing simulation system and method. The advantages and features of the invention will become clearer from the following description and claims.

[0019] Please refer to Figure 1 This embodiment provides a virtual-real integrated multi-domain heterogeneous network collaborative computing simulation system, including: The heterogeneous computing power prototype subsystem consists of several physical heterogeneous computing power boards, which are used to represent the actual computing power nodes in space, air, and ground respectively. The virtualization simulation subsystem runs on the workstation and is used to generate virtual nodes corresponding to the sky, air, and ground nodes through virtual machines / containers to expand the simulation scale. The onboard computing terminal is an on-orbit heterogeneous computing node deployed on a real satellite or satellite simulation module, used to inject real onboard computing power into the simulation network. The programmable network link simulator is interconnected with the heterogeneous computing power prototype subsystem, the virtualization simulation subsystem, and the spaceborne computing terminal. It is used to generate link delay, bit error rate, packet loss rate, and topology change parameters in real time based on orbital dynamics or dynamic scenario scripts, so as to simulate highly dynamic time-varying links and network topologies between the sky and the ground with high realism. The heterogeneous resource management subsystem is deployed across ground cloud centers, virtual nodes, and physical computing nodes. It is used to manage physical and virtual nodes into a unified cloud-edge-device resource pool through a multi-level container orchestration engine, and to complete cross-domain computing power scheduling, task unloading, and container migration based on network status and task requirements. The simulation verification system interface interacts with the heterogeneous resource management subsystem and the programmable network link simulator. It is used to input simulation task scripts, output performance indicators, and support closed-loop verification of networking routing protocols, distributed computing offloading strategies, and constellation cluster management algorithms.

[0020] This system combines virtual and real technologies, integrating real heterogeneous hardware with a large number of virtual nodes into a programmable dynamic network environment. It enables one-stop high-fidelity simulation of sky-ground links and cross-domain cloud-edge-device computing power scheduling, and achieves closed-loop verification of multiple algorithms such as routing, offloading, and cluster management, significantly reducing the cost and risk of field trials.

[0021] For details, please refer to Figure 2 The single physical node of the heterogeneous computing power prototype subsystem includes: Container engine; At least one heterogeneous computing board selected from D2000, RK3588, FT7004, neuromorphic chip, Atlas200, and Raspberry Pi; A high-speed bus interface for communicating with a programmable network link simulator; Each physical node is configured differently using container images to play the roles of space-based, air-based, or ground-based systems.

[0022] This heterogeneous computing prototype subsystem uses physical heterogeneous boards, a container engine, and a high-speed bus as its smallest units. It abstracts real hardware with different architectures, computing power levels, and power consumption characteristics into nodes that can be uniformly scheduled, dynamically scaled, and mapped to the air / ground domains according to their roles. In terms of resource management, it virtually quantizes the CPU / GPU / NPU core count, frequency, cache capacity, and power consumption curves of each physical heterogeneous board and writes them into node tags. The container orchestrator (KubeEdge / K3s / ECS) uses node tags and an extended scheduler to bind container instances to the most suitable physical heterogeneous boards according to QoS policies. During operation, the board-side container engine exposes hardware-level indicators such as computing power utilization, power consumption, temperature, and ECC error count in real time through the cGroup driver, and transmits them back to the programmable network link simulator and the heterogeneous resource management subsystem via high-speed Ethernet for task hot migration and backup.

[0023] Through the above mechanisms, the heterogeneous computing power prototype subsystem maintains the original performance and power consumption characteristics of various chips while completing the mapping of roles with the sky / air / ground, container-level lifecycle management and closed-loop health monitoring, so as to achieve high-fidelity operation of real hardware, real computing power and real faults.

[0024] Please refer to Figure 3 In this embodiment, the virtualization simulation subsystem uses the VMware virtualization platform and runs within each virtual machine. Virtual computing power unit; Virtual network adapter; And a container engine for quickly cloning and generating large-scale virtual nodes in the sky, air, and ground.

[0025] This virtualization simulation subsystem uses a virtualization layer, container runtime, virtual network interface card, and scenario-driven scripts as its framework. It clones VM instances corresponding to the roles of space, air, and ground nodes in batches on one or more high-performance x86 servers. During the startup phase, each VM instance is automatically equipped with differentiated role tags (satellite / airship / drone / ground station), virtual hardware configuration (vCPU cores, memory, virtual accelerator card), and initial network topology parameters through an OVA template. Then, it launches a KubeEdge, K3s, or ECS container engine inside the VM instance, enabling the VM to have both virtual machine isolation and lightweight container runtime. During system operation, a programmable network link simulator is accessed via a virtual network interface card. The simulator injects parameters such as link latency, packet loss, and bandwidth jitter in real time based on orbital dynamics or maneuvering scripts, enabling VM-level nodes to synchronize with real links. Simultaneously, in terms of resource management, metrics such as vCPU utilization, memory bandwidth, and container image startup time for each VM instance are collected and uploaded to a unified scheduler along with node tags. This data is used to dynamically adjust the number of container replicas, task migration, or VM backups during simulation to verify constellation expansion or node failure scenarios. Through templated batch cloning, differentiated role configuration, real-time link status injection, and metric collection, the virtualization simulation subsystem can generate hundreds to thousands of space-air-ground virtual nodes within minutes, achieving low-cost, highly flexible, and large-scale space-air-ground collaborative computing simulation synchronized with real links.

[0026] Please refer to Figure 4 The spaceborne computing terminal in this embodiment includes: The heterogeneous computing module integrates at least two heterogeneous processors: the RK3588 and a neuromorphic chip. The primary and backup management module adopts a dual-processor cold / hot backup architecture for on-orbit fault switching; The routing protocol module is used to load the inter-satellite / satellite-to-ground dynamic routing protocol stack; The communication interface module provides an Ethernet interface to enable data backhaul with the programmable network link simulator. The primary and backup management module deploys a container engine to predict running trajectories and make task unloading decisions within containers.

[0027] This spaceborne computing terminal employs an architecture of dual-mode cold backup main control, heterogeneous computing array, hard real-time bus, and on-orbit container engine to achieve on-orbit edge computing and routing protocol verification within a real satellite or satellite simulation module. Through a collaborative mechanism of primary / backup heterogeneous main control, hard acceleration array, containerized scheduling, and real-time fault injection, the spaceborne computing terminal achieves high-fidelity synchronous operation of real spaceborne computing power, real space environment, and ground simulation network in orbit.

[0028] Please refer to Figure 5The programmable network link simulator in this embodiment includes: The input terminal is used to receive orbital ephemeris, maneuvering strategies, or user-defined dynamic scenario scripts. The spatiotemporal link parameter database is used to store prior or real-time measurement parameters of inter-satellite, satellite-to-ground, and air-to-ground links. The calculation module is used to calculate the link distance, Doppler shift, and visibility matrix based on the orbital dynamics model; The topology and link status configuration module is used to convert the calculated link parameters into link latency, bandwidth, bit error rate, packet loss rate, and interruption events. The link state simulation and network topology simulation module is used to apply real-time impairment and forwarding rules to data packets between real / virtual nodes based on the parameters output by the topology and link state configuration module. The scene display module is used to graphically output dynamic topology and link quality curves.

[0029] This programmable network link simulator, through an orbital dynamics engine, a scene script parser, and a link parameter database, transforms the position, velocity, and attitude of highly dynamic constellations / UAV swarms into configurable link-level parameters. It also applies nanosecond-level precision to each frame of packets between real and virtual nodes, including latency, bit error rate, packet loss, jitter, bandwidth limitations, and interruptions, to achieve high-fidelity simulation of network topology and link status.

[0030] The input terminal supports two modes: Orbit Ephemeris Mode: Import TLE / SDP4 or custom RINEX files, and the SGP4 / HPOP extrapolation module calculates the ECI / ECEF coordinates, relative velocities, and Doppler shifts of each node at a simulation clock Δt=100 ms; Script mode: Users write highly maneuverable trajectories in Python, and the parser generates xyz+quaternion sequences in real time.

[0031] Its scene clock is globally synchronized with the container cluster (PTP / gPTP, deviation <1 μs), ensuring that link changes are aligned with container scheduling events.

[0032] Real-time calculation of link parameters includes: Visibility matrix: based on spherical Earth + terrain occlusion + antenna pointing semi-cone angle, the visibility window is updated every 100 ms; Link-state database (time-series), its table structure: {time, src_id, dst_id, visible, distance_m, fspl_dB, snr_dB, delay_ns, ber_base, bw_Hz, jitter_ns, link_state} It employs an in-memory time-series database (InfluxDB) + a circular buffer, supporting continuous writing and random reading of 200 k link statuses per second, with a latency of < 50μs.

[0033] The aforementioned programmable network link simulator, through orbit-clock-link ternary drive and nanosecond-level damage pipeline, completely maps the highly dynamic topology and discontinuous link connectivity of real satellites / UAVs onto a single network cable in the laboratory, providing a high-fidelity network environment for sky-ground collaborative computing simulation.

[0034] The heterogeneous resource management subsystem in this embodiment also includes: Resource scheduling middleware is used to parse task requirement constraints, node remaining computing power, and network link quality to generate cross-domain offloading decisions; The cloud-edge collaborative controller is used to upload computing tasks to satellite or airborne nodes when the link quality is higher than a threshold, and to trigger container migration back to the ground when the link quality is lower than a threshold. Edge-to-edge collaboration controllers are used to implement P2P distribution and load balancing of container images between adjacent nodes within a constellation or cluster; Edge-to-device collaborative controllers are used to offload edge computing subtasks to drones or ground terminals that are directly connected to them.

[0035] This heterogeneous resource management subsystem uses a unified cloud-edge-device resource pool as its top-level abstraction. Through a multi-orchestrator hybrid deployment + cross-domain collaborative scheduling middleware architecture, it unifies the management of heterogeneous computing power prototype subsystem, virtualization simulation subsystem, and spaceborne computing terminal heterogeneous computing power, enabling millisecond-level cross-domain unloading, elastic scaling, fault migration, and energy consumption optimization at the task granularity (container / Pod).

[0036] Among them, a Kubernetes orchestrator is deployed in the ground cloud center to be responsible for task planning and computing, AI training, and big data batch processing; Deploy KubeEdge orchestrators / K3s orchestrators on spaceborne computing terminals and UAV pods to support offline autonomy of spaceborne / airborne nodes, orbit prediction, and task offloading; Deploy ECS orchestrators on handheld terminals, ground vehicles, neuromorphic chips, and RK3588 to run offload tasks and monitor resources in a lightweight manner. In this embodiment, the simulation verification system interface is configured as follows: Provide RESTful API and / or gRPC services for external algorithm plugin injection; Performance probes are used to collect throughput, end-to-end latency, task completion time, CPU / GPU utilization, and energy consumption metrics. A visual evaluation panel is used to compare performance curves under different routing, scheduling, or offloading strategies and output test reports.

[0037] The aforementioned hybrid space-ground multi-domain heterogeneous network collaborative computing simulation system, combining virtual and physical elements, simultaneously connects real heterogeneous hardware (D2000, RK3588, neuromorphic chips, etc.) and a large number of virtual nodes to the same simulation network. This preserves the real computational characteristics of physical boards in terms of instruction sets, power consumption, and acceleration libraries, while leveraging virtualization to overcome the limitations of the number of physical devices. It enables large-scale scenarios involving "thousands of satellites + a massive number of ground terminals," achieving a balance between "realism" and "scalability" that was previously impossible with purely software or hardware systems. Its programmable network link simulator, based on orbital dynamics / maneuvering scripts, outputs link latency, bit errors, packet loss, and interruption events in real time. It can "reproduce" extreme network conditions such as changes in high-dynamic constellation visibility to Earth, inter-satellite link switching, and high-speed maneuvers of UAV swarms in the laboratory, addressing the pain point of large discrepancies between traditional static topology simulations and real-world scenarios, and significantly improving the credibility of protocol and algorithm verification. Furthermore, by adopting a hybrid container orchestration of Kubernetes+KubeEdge / K3s / ECS, computing resources with different architectures, operating systems, and performance levels in space, air, and ground are abstracted into a unified cloud-edge-device resource pool. With the help of resource scheduling middleware, task migration and offloading between satellite and ground / between satellites / between satellites can be completed, realizing unified scheduling of cross-domain heterogeneous computing power.

[0038] In addition, this embodiment also provides a method for a collaborative computing simulation system of a multi-domain heterogeneous network combining virtual and real elements, please refer to [link / reference]. Figure 6 The method includes the following steps: System construction steps: Interconnect the heterogeneous computing power prototype subsystem, virtualization simulation subsystem and spaceborne computing terminal through a programmable network link simulator, configure the initial link parameters according to orbit data or dynamic scenario scripts, and form a unified sky-ground multi-domain simulation network environment. Resource management steps: Deploy container orchestration engines on ground cloud centers, virtual nodes, and physical / spaceborne nodes, register all nodes to the same cluster, and build a unified cloud-edge-device resource pool; Scenario configuration steps: Input the network topology change script, computation task dependency graph and resource requirement constraints through the simulation verification system interface to generate the collaborative computing scenario to be verified; Dynamic simulation and scheduling steps: Start the programmable network link simulator and update the link status in real time according to the script; the heterogeneous resource management subsystem dynamically executes cross-domain task scheduling, container migration and compute unloading based on the real-time network status and task requirements; Data collection and analysis steps: Collect network performance and computing performance indicators, perform statistical and visual comparisons, and output an algorithm verification report.

[0039] The dynamic simulation and scheduling steps further include: When an inter-satellite link interruption or latency mutation exceeding a set threshold is detected, a container snapshot is generated and the system is relocated back to the ground cloud center or other visible satellite nodes via a backup link. When the computational load of the airborne node is detected to exceed the limit, some inference subtasks will be offloaded to the neighboring drone node in an edge-to-edge collaborative manner. When the quality of the link between the ground terminal and the airborne node is detected to be better than that of the satellite-to-ground link, the calculation results are prioritized through the airborne node.

[0040] The data collection and analysis steps further include: Use time-series database storage performance metrics; Using machine learning models to perform regression analysis on the collected data, we can predict bottleneck nodes under different constellation sizes or task intensities. The prediction results are fed back to the scenario configuration step to optimize the next round of simulation scripts and initial resource deployment strategies.

[0041] The above method can realize the function of a collaborative computing simulation system for heterogeneous networks in multiple domains, combining virtual and real elements, and its implementation method is similar, so it will not be described in detail here.

[0042] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings, but the present invention is not limited to the above embodiments. Even if various changes are made to the present invention, if these changes fall within the scope of the claims of the present invention and their equivalents, they shall still fall within the protection scope of the present invention.

Claims

1. A virtual-real combined sky-ground multi-domain heterogeneous network cooperative computing simulation system, characterized in that, include: The heterogeneous computing power prototype subsystem consists of multiple physical heterogeneous computing power boards, which are used to represent real computing power nodes in space, air, and ground respectively. The virtualization simulation subsystem runs on the workstation and is used to generate virtual nodes corresponding to the sky, air, and ground nodes through virtual machines / containers to expand the simulation scale. The onboard computing terminal is an on-orbit heterogeneous computing node deployed on a real satellite or a heterogeneous computing node deployed in a satellite simulation module, used to inject real onboard computing power into the simulation network. The programmable network link simulator is interconnected with the heterogeneous computing power prototype subsystem, the virtualization simulation subsystem, and the spaceborne computing terminal. It is used to generate link delay, bit error rate, packet loss rate, and topology change parameters in real time based on orbital dynamics or dynamic scenario scripts, so as to simulate the highly dynamic time-varying links and network topology between the sky and the ground with high realism. The heterogeneous resource management subsystem is deployed across ground cloud centers, virtual nodes, and physical computing nodes. It is used to manage physical and virtual nodes into a unified cloud-edge-device resource pool through a multi-level container orchestration engine, and to complete cross-domain computing power scheduling, task unloading, and container migration based on network status and task requirements. The simulation verification system interface interacts with the heterogeneous resource management subsystem and the programmable network link simulator. It is used to input simulation task scripts, output performance indicators, and support closed-loop verification of networking routing protocols, distributed computing offloading strategies, and constellation cluster management algorithms.

2. The virtual-real combined sky-ground multi-domain heterogeneous network cooperative computing simulation system of claim 1, wherein, The single physical node of the heterogeneous computing power prototype subsystem includes: Container engine; At least one physical heterogeneous computing power board; Wired / wireless high-speed bus interface for communication with a programmable network link simulator; Each physical computing node is configured as a space-based, air-based, or ground-based computing node through container image differentiation.

3. The virtual-real combined sky-ground multi-domain heterogeneous network cooperative computing simulation system of claim 1, wherein, The virtualization simulation subsystem uses a virtualization platform and runs within each virtual machine: Virtual computing power unit; Virtual network adapter; And a container engine for quickly cloning and generating large-scale virtual nodes in the sky, air, and ground.

4. The sky-ground multi-domain heterogeneous network collaborative computing simulation system of any one of claims 1-3, wherein, The onboard computing terminal includes: Heterogeneous computing power module, integrating multiple edge heterogeneous processors; The primary and backup management module adopts a dual-processor cold / hot backup architecture for on-orbit fault switching; The routing protocol module is used to load the inter-satellite / satellite-to-ground dynamic routing protocol stack; The communication interface module provides an Ethernet interface to enable data backhaul with the programmable network link simulator. The primary and backup management module deploys a container engine, in which trajectory prediction and task unloading decision-making business applications run.

5. The hybrid sky-ground multi-domain heterogeneous network coordinated computing emulation system of claim 4, wherein, The programmable network link simulator includes: The input terminal is used to receive orbital ephemeris, maneuvering strategies, or user-defined dynamic scenario scripts. The spatiotemporal link parameter database is used to store prior or real-time measurement parameters of inter-satellite, satellite-to-ground, and air-to-ground links. The calculation module is used to calculate the link distance, Doppler shift, and visibility matrix based on the orbital dynamics model; The topology and link status configuration module is used to convert the calculated link parameters into link latency, bandwidth, bit error rate, packet loss rate, and interruption events. The link state simulation and network topology simulation module is used to apply real-time impairment and forwarding rules to data packets between real / virtual nodes based on the parameters output by the topology and link state configuration module. The scene display module is used to graphically output dynamic topology and link quality curves.

6. The sky-ground multi-domain heterogeneous network cooperative computing emulation system of claim 1 or 2, wherein, The heterogeneous resource management subsystem also includes: Resource scheduling middleware is used to parse task requirement constraints, node remaining computing power, and network link quality to generate cross-domain offloading decisions; The cloud-edge collaborative controller is used to upload computing tasks to satellite or airborne nodes when the link quality is higher than a threshold, and to trigger container migration back to the ground when the link quality is lower than a threshold. Edge-to-edge collaboration controllers are used to implement P2P distribution and load balancing of container images between adjacent nodes within a constellation or cluster; Edge-to-device collaborative controllers are used to offload edge computing subtasks to drones or ground terminals that are directly connected to them.

7. The virtual-real combined sky-ground multi-domain heterogeneous network cooperative computing simulation system of claim 1, wherein, The simulation verification system interface is configured as follows: Provide RESTful API and / or gRPC services for script injection based on orbital dynamics or dynamic scenarios; Performance probes are used to collect throughput, end-to-end latency, task completion time, CPU / GPU utilization, and energy consumption metrics. A visual evaluation panel is used to compare performance curves under different routing, scheduling, or offloading strategies and output test reports.

8. A virtual-real combined sky-ground multi-domain heterogeneous network collaborative computing simulation method based on the system described in any one of claims 1-7, characterized in that, include: System construction steps: Interconnect the heterogeneous computing power prototype subsystem, virtualization simulation subsystem and spaceborne computing terminal through a programmable network link simulator, configure the initial link parameters according to orbit data or dynamic scenario scripts, and form a unified sky-ground multi-domain simulation network environment. Resource management steps: Deploy container orchestration engines on ground cloud centers, virtual nodes, and physical / spaceborne nodes, register all nodes to the same cluster, and build a unified cloud-edge-device resource pool; Scenario configuration steps: Input the network topology change script, computation task dependency graph and resource requirement constraints through the simulation verification system interface to generate the collaborative computing scenario to be verified; Dynamic simulation and scheduling steps: Start the programmable network link simulator and update the link status in real time according to the script; the heterogeneous resource management subsystem dynamically executes cross-domain task scheduling, container migration and compute unloading based on the real-time network status and task requirements; Data collection and analysis steps: Collect network performance and computing performance indicators, perform statistical and visual comparisons, and output an algorithm verification report.

9. The virtual-real combined sky-ground multi-domain heterogeneous network collaborative computing simulation method as described in claim 8, characterized in that, The dynamic simulation and scheduling steps further include: When an inter-satellite link interruption or latency mutation exceeding a set threshold is detected, a container snapshot is generated and the system is relocated back to the ground cloud center or other visible satellite nodes via a backup link. When the computational load of the airborne node is detected to exceed the limit, some inference subtasks will be offloaded to the neighboring drone node in an edge-to-edge collaborative manner. When the quality of the link between the ground terminal and the airborne node is detected to be better than that of the satellite-to-ground link, the calculation results are prioritized through the airborne node.

10. The virtual-real combined sky-ground multi-domain heterogeneous network collaborative computing simulation method as described in claim 8 or 9, characterized in that, The data collection and analysis steps further include: Use time-series database storage performance metrics; Machine learning models are used to perform regression analysis on the collected data to predict bottleneck nodes under different constellation sizes or task intensities; the prediction results are fed back to the scenario configuration step to optimize the next round of simulation scripts and initial resource deployment strategies.

Images