Computing power routing method based on multi-intelligence collaboration

Abstract

The invention relates to the technical field of network resource scheduling and artificial intelligence, and provides a computing power routing method based on multi-intelligence collaboration, which comprises a satellite, an unmanned aerial vehicle, a ground base station, a user terminal and other multi-layer heterogeneous nodes. Network nodes are dynamically networked through a wireless link, and node movement, topology change and multi-hop communication are supported. Each node has certain computing power and storage capability, and can cooperatively complete data processing and task distribution. The core routing nodes are deployed in a network backbone in a distributed manner and are responsible for intelligent routing and resource scheduling. The system integrates key technologies such as link quality quantification, distributed election, reinforcement learning of intelligent agents, virtual wireless interface simulation, flow control and route issuing. The method provided by the invention supports large-scale simulation, dynamic routing optimization and computing power resource collaboration, is suitable for scenes such as satellite internet, unmanned aerial vehicle clusters, emergency communication and intelligent transportation, and realizes wide-area, intelligent and high-reliability communication and computing power services.

Description

Technical Field

[0001] This invention relates to the fields of network resource scheduling and artificial intelligence technology, and more specifically, to a computing power routing method based on multi-intelligent collaboration. Background Technology

[0002] The integration of air-space-ground (AS / G) networks with computing power networks represents a significant trend in modern network architecture. AS / G networks achieve wide-area, seamless, multi-layered connectivity by merging satellites, drones, and ground base stations. Computing power networks emphasize the coordinated scheduling of computing power, storage, and data among network nodes. This combination enables the network to dynamically select optimal communication paths and intelligently distribute tasks and optimize resources based on link quality and node computing power. For example, satellite and drone nodes can act as high-computing-power core routers, handling data processing and forwarding tasks, while ground nodes participate in coordination based on real-time link and computing power status. Key technologies include heterogeneous networking, distributed routing, link and computing power optimization, and agent-driven adaptive decision-making. This integration drives the evolution of networks from simple communication to integrated intelligent services combining communication and computing power.

[0003] Ad hoc networking technologies based on traditional routing protocols (OSPF, AODV, OLSR) and link quality awareness typically employ static or semi-dynamic routing strategies, relying on local node information or limited global topology. Link quality assessment is often based on simple metrics such as latency, packet loss, and RSSI, and routing decisions are separated from computing resource allocation. This leads to the following problems: Traditional methods struggle to handle the frequently changing topology and link quality in space-air-ground networks, resulting in slow route convergence and a high risk of communication interruptions or performance degradation. Furthermore, existing solutions typically focus only on communication links, failing to incorporate node computing power and storage resources into routing decisions, thus hindering intelligent task and data distribution. Moreover, existing methods lack agent-driven adaptive decision-making mechanisms such as reinforcement learning, making it difficult to dynamically optimize multiple objectives. The selection of core routing nodes largely depends on global information, limiting the application of distributed election algorithms and hindering the construction of efficient and robust backbone networks. Summary of the Invention

[0004] In view of this, the present invention proposes a computing power routing method based on multi-intelligent collaboration to solve the problems existing in the prior art.

[0005] To achieve the above objectives, this invention proposes a computing power routing method based on multi-intelligence collaboration, characterized by the following steps: By managing the basic status of the container collection node and collecting the wireless link status through the communication container, composite container node status data is constructed, and the composite container node status data is standardized and encapsulated into a structured data stream. The GRU time-series analysis module processes historical state data to extract trends in computing power and link changes; shared Critic and independent Actors collaborate to generate routing and computing power schemes, and make computing power allocation decisions to build a joint decision-making intelligent decision layer. Perform computing power-routing collaborative verification. When the link bandwidth cannot meet the computing power transmission requirements, switch to the backup path and reallocate computing power. The system parses and converts decision commands and protocols, deploys routing rules, and monitors link status in real time through BFD sessions. When the primary path is interrupted, it automatically switches to the backup path. Perform joint simulation of computing power and communication, calculate core indicators and use Prometheus to store data, and generate visualization results; transmit bottleneck reports to the intelligent decision-making layer to correct the prediction model and optimize the weights of the value function.

[0006] Furthermore, during the construction of the composite container node status data, the management container collects the basic node status through Linux system tools, including satellite / drone / ground station equipment resources; The communication container interfaces with wmediumd enhanced version to collect wireless link status, including received signal strength, link transmission delay, and packet loss rate.

[0007] Furthermore, in the construction steps of the intelligent decision-making layer joint decision-making, the GRU time series analysis module processes historical state data, extracts computing power and link change trends, and provides a predictive basis for decision-making; the shared Critic and independent Actors collaborate to calculate, and based on real-time state and time series prediction results, generate routing and computing power schemes, and make computing power allocation decisions.

[0008] Furthermore, in the step of parsing and converting decision instructions into protocols, the decision instructions are divided into routing sub-instructions and computing power sub-instructions, and converted into recognizable configuration commands; routing rules are deployed based on FRR, and the link status is monitored in real time through BFD sessions, automatically switching to the backup path when the primary path is interrupted.

[0009] Furthermore, in the step of performing joint simulation of computing power and communication, the execution layer data and air-space-ground topology configuration are input, the channel is simulated and the simulation data is output; the core indicators are calculated and the data is stored using Prometheus to generate visualization results; the bottleneck report is transmitted to the intelligent decision-making layer to correct the prediction model and optimize the weight of the value function.

[0010] Furthermore, the method supports heterogeneous networking of multi-level nodes such as satellites, drones, and ground base stations, and coordinates the scheduling of computing resources and communication links to achieve integrated communication and computing services.

[0011] Furthermore, the method improves the accuracy and adaptability of routing decisions by collecting link quality data in batches and using historical states as input to the agent.

[0012] Furthermore, the method integrates automated simulation, performance verification, and visualization modules, supports large-scale network testing and optimization, and verifies the optimization decision model through simulation.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention eliminates the reliance on global topology and employs a local link quality-based and distributed election algorithm to dynamically select core routing nodes. Core nodes can achieve efficient self-organization without requiring comprehensive network information, thus improving network scalability and robustness.

[0014] This invention employs reinforcement learning-driven collaborative optimization of routing and computing power. By deploying PPO agents on core nodes and quantifying the status of nodes across multiple dimensions, including historical link quality and node computing power, it achieves joint optimization of routing and computing resources. Through a distributed Actor and global Critic structure, the agents can adaptively select the optimal path and task distribution scheme, significantly improving network performance.

[0015] This invention supports heterogeneous networking of multi-level nodes such as satellites, drones, and ground base stations, with coordinated scheduling of computing resources and communication links. The system can dynamically adjust data processing and forwarding paths based on node computing power and link status, achieving integrated communication and computing services.

[0016] This invention improves the accuracy and adaptability of routing decisions by collecting link quality data in batches and using historical states as input to the intelligent agent. It integrates automated simulation, performance verification, and visualization modules to support large-scale network testing and optimization. Attached Figure Description

[0017] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. In the drawings: Figure 1 This is a schematic diagram of the computing power routing method based on multi-intelligent collaboration of the present invention. Detailed Implementation

[0018] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0019] This embodiment proposes a computing power routing method based on multi-intelligent collaboration, such as... Figure 1 As shown, it includes the following steps: By managing the basic status of the container collection node and collecting the wireless link status through the communication container, composite container node status data is constructed, and the composite container node status data is standardized and encapsulated into a structured data stream. The GRU time-series analysis module processes historical state data to extract trends in computing power and link changes; shared Critic and independent Actors collaborate to generate routing and computing power schemes, and make computing power allocation decisions to build a joint decision-making intelligent decision layer. Perform computing power-routing collaborative verification. When the link bandwidth cannot meet the computing power transmission requirements, switch to the backup path and reallocate computing power. The system parses and converts decision commands and protocols, deploys routing rules, and monitors link status in real time through BFD sessions. When the primary path is interrupted, it automatically switches to the backup path. Perform joint simulation of computing power and communication, calculate core indicators and use Prometheus to store data, and generate visualization results; transmit bottleneck reports to the intelligent decision-making layer to correct the prediction model and optimize the weights of the value function.

[0020] The implementation principle and detailed steps of this embodiment are explained in the following manner, based on the above steps: This invention's technical solution revolves around the collaborative optimization of an integrated air-space-ground computing network. Its core technologies include the multi-agent PPO algorithm and a containerized SDN platform. Through a closed-loop process of state input, intelligent decision-making, execution scheduling, and simulation verification, it achieves joint optimization of computing power and routing. The process is structured around an "intelligent decision-making layer - execution scheduling layer - simulation verification layer." Each layer flows data through standardized interfaces, including a computing network protocol stack and data interaction links, covering the entire link from air-space-ground node state collection to decision optimization.

[0021] As a preferred embodiment, the steps of the method in this embodiment are explained in detail in a phased manner, as follows: Phase 1: Data Acquisition and Input of Space-Air-Ground Node Status Step 1.1 Composite Container Node Status Acquisition Operation objects: Composite container nodes that execute the scheduling layer and their subordinate management containers, communication containers, and computing power containers.

[0022] Specific steps: The container management system collects basic node status data through Linux system tools, including resources of devices such as satellites, drones, and ground stations. The communication container interfaces with the enhanced version of wmediumd to collect wireless link status: received signal strength, link transmission delay, and packet loss rate, providing a basis for routing decisions. Step 1.2 State Standardization and Protocol Encapsulation Operation target: State processing unit that connects the execution scheduling layer and the intelligent decision-making layer.

[0023] Specific steps: Standardized protocol encapsulation: Integrate "node ID - computing power status - link quality - business requirements" into a structured data stream to reduce the transmission overhead of air-space-ground links and adapt to the "satellite link bandwidth optimization" requirement.

[0024] Phase 2: Joint Decision Generation by the Intelligent Decision-Making Layer Step 2.1 Time Series Trend Analysis Target audience: Enhanced version of the GRU time series analysis module in the intelligent decision-making layer.

[0025] Specific steps: Input historical status data, including past satellite load changes and space-air-ground network link delay fluctuation sequences; The GRU module processes timing information and extracts computing power and link change trends through update and reset gates to avoid decision lag, which is in line with the goal of "proactively predicting network changes". Output a time series forecast report and transmit it to the joint value function calculation module to provide a forecast basis for decision-making.

[0026] As a preferred embodiment, GRU can be replaced with LSTM to further enhance long-term predictions, making it suitable for satellite orbit-level planning, or "experience replay" can be used to correct prediction biases.

[0027] Step 2.2 Shared Critic and Independent Actor Collaborative Computation Operation targets: Shared Critic Enhanced Version and Independent Actors in the Intelligent Decision-Making Layer.

[0028] Specific steps: Receive real-time status and time-series prediction results, and establish a comprehensive scoring model with computing power utilization and link stability as objectives according to the "global collaboration" logic; Independent Actor-Route generates routing and computing power schemes based on local state, makes computing power allocation decisions, and shares Critic to provide value feedback, thus avoiding local decisions from deviating from the global optimum. Receive the primary / backup path decision and time series prediction results, calculate the comprehensive value of the scheme according to the "multi-objective optimization" logic in the document, and select the Top N optimal candidate schemes.

[0029] In a preferred embodiment, in the above steps, the independent Actor can be split into a combination of routing Actor and computing Actor to form the original multi-agent architecture of the document; the shared Critic value function can be replaced with a weighted summation model to dynamically adjust the weights.

[0030] Step 2.3 Computing power-routing collaborative verification Operation target: the computing power-routing collaborative decision-making unit of the intelligent decision-making layer.

[0031] Specific steps: When the link bandwidth cannot meet the computing power transmission requirements, switch to the backup path, switch the satellite path to the UAV path and reallocate computing power; output the final decision instruction containing the primary / backup routing table and computing power allocation details, and adapt to the execution scheduling layer interface.

[0032] Equivalent alternative: Collaborative verification can be replaced by a hybrid mechanism of "rules + learning". Preset rules are used for simple scenarios, and reinforcement learning is used for optimization in complex scenarios.

[0033] Phase 3: Implementation of Scheduling Layer Decisions Step 3.1 Decision Instruction Parsing and Protocol Conversion Operation target: The computing network protocol stack that executes the scheduling layer.

[0034] Specific steps: The decision instructions are divided into routing sub-instructions, which include primary / backup paths and BFD parameters, and computing power sub-instructions, which include target node ID and resource quota. The routing sub-instructions are converted into configuration commands that FRR can recognize, and the computing power sub-instructions are converted into Docker resource adjustment commands to ensure that the execution layer can recognize them.

[0035] In a preferred embodiment, the protocol stack in the above steps can be replaced with a satellite-specific communication protocol to optimize long-latency transmission, or FlaskRESTfulAPI can be used for ground node interaction.

[0036] Step 3.2 Route Execution and Fault Detection Operation targets: the routing execution units BFD and FRR of the execution scheduling layer, as well as composite container nodes.

[0037] Specific steps: Based on FRR, routing rules are deployed, and the primary / backup routing tables are distributed to the kernel routing module of the composite container node through the Netlink interface; A BFD session is established between the main path nodes, and a detection message is sent every 500ms to start fault detection and monitor the link in real time. When the primary path is interrupted, FRR automatically switches to the backup path and feeds back the results to the intelligent decision-making layer, which meets the goal of ensuring data plane continuity.

[0038] In a preferred embodiment, BFD can be replaced with the low dynamic terrestrial network VRRP and FRR can be replaced with the small-scale network Quagga in the above steps to achieve functional compatibility.

[0039] Phase 4: Simulation Verification and Decision Optimization Step 4.1 Co-simulation of computing power and communication Operation targets: the computing power simulation module of the simulation verification layer and the wmediumd enhanced version.

[0040] Specific steps: Input execution layer data, decision-making includes execution logs of route switching counts and task completion rates, and air-ground topology configuration, including node coordinates, trajectories, and channel parameters, including Ka-band rain attenuation, to adapt to three-dimensional wireless network simulation scenarios; wmediumd simulates the channel in conjunction with tc: it calculates SNR based on node coordinates, simulates frame loss and delay, correlates computing power transmission, limits the rate when bandwidth is insufficient, and outputs simulation data, including computing power utilization and routing delay every 10 seconds.

[0041] In a preferred embodiment, wmediumd can be replaced with NS-3 high-precision satellite channel, and computing power simulation can be replaced with CloudSim customized interface adapter container.

[0042] Step 4.2 Joint Assessment and Bottleneck Identification Operation targets: Joint evaluation unit of simulation verification layer, Prometheus monitoring.

[0043] Specific steps: Calculate core metrics, including computing power utilization, task completion rate, routing latency, and switchover success rate, to conform to multi-dimensional evaluation logic; Use Prometheus to store data and combine it with visualization tools to generate time series curves and delay histograms to present the results intuitively.

[0044] Step 4.3 Decision Optimization Feedback Operation target: Feedback interface between simulation verification layer and intelligent decision-making layer.

[0045] Specific steps: The bottleneck report is transmitted to the intelligent decision-making layer; the GRU time series analysis module corrects the prediction model and shares the weights of the Critic optimized value function. The adjusted model is used for the next round of decision-making, realizing a loop, improving adaptability to dynamic scenarios, and thus achieving the goal of optimizing the efficiency of matching document reinforcement learning samples.

[0046] As a preferred embodiment, the feedback in the above steps can be implemented using an experience replay approach, prioritizing the training of high-value simulation samples.

[0047] In the method steps of this embodiment, the key device interaction process is as follows: Satellite-UAV interaction: The satellite transmits computing power task data to the UAV through a dedicated link in the communication container. After processing, the UAV confirms the link status via BFD and provides feedback. When the link is interrupted, FRR switches to the backup line to ensure transmission, which conforms to the logic of "dynamic topology adaptation".

[0048] UAV-Ground Station Interaction: The UAV reports its network status to the ground station via the wmediumd link. The ground station prioritizes assigning long-cycle tasks to high-computing-power nodes to match computing power with task performance.

[0049] Intelligent Decision Layer: Core of Joint Decision-Making Based on Computing Power and Routing 1. Shared Critic (pure software) Functionally, it expands the value function estimation capability of the "shared commentator" in the document, integrates routing benefits into a joint value function, provides a unified value baseline for decision-making, reduces training oscillations, and aligns with the document's goal of "improving convergence efficiency." It can be replaced with a dual commentator mechanism to reduce value estimation bias.

[0050] Joint value function calculation The defined link quality, node cache queue status, and newly added computing power status are used to output a joint value score as a basis for decision priority. The routing decision function of the integrated document "Independent Actor" is combined with the newly added computing power decision function, and the primary / backup path and computing power allocation instructions are output based on the local computing power-communication status.

[0051] The primary / backup path prioritizes links connecting to high-performance nodes, and uses BFD (Browser Detection and Failure Prevention) for rapid fault detection to enable automatic failover in case of failure, ensuring data plane continuity. It has no independent hardware dependencies, relying on the container node's virtual network interface (mac80211_hwsim) for execution, and can be replaced with multi-path decision-making for enhanced robustness.

[0052] In the GRU time series analysis module, the function of extracting link time series trends from document GRU has been expanded, and computing power trend prediction has been added. By relying on the ability to "predict network conditions", decisions can be adjusted in advance to reduce recomputation overhead.

[0053] Computing power-routing collaborative decision-making unit To address the challenges of global collaboration and local autonomous balancing, a computing power demand-routing path matching rule is established. When a link is interrupted, computing power reallocation is triggered to meet the "fast route switching" requirement. This relies on container node CPU computation and utilizes Linux namespace isolation.

[0054] This embodiment is applicable to an integrated air-space-ground computing network environment, including multi-layered heterogeneous nodes such as satellites, drones, ground base stations, and user terminals. Network nodes dynamically network via wireless links, supporting node movement, topology changes, and multi-hop communication. Each node possesses certain computing and storage capabilities, enabling collaborative data processing and task distribution. Core routing nodes are distributed across the network backbone, responsible for intelligent routing and resource scheduling. The system integrates key technologies such as link quality quantization, distributed election, reinforcement learning agents, virtual wireless interface simulation (e.g., mac80211_hwsim), flow control, and route distribution. This environment supports large-scale simulation, dynamic route optimization, and collaborative computing resources, making it suitable for scenarios such as satellite internet, drone swarms, emergency communications, and intelligent transportation, achieving wide-area, intelligent, and highly reliable communication and computing services.

[0055] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A computing power routing method based on multi-intelligence collaboration, characterized in that, Includes the following steps: By managing the basic status of the container collection node and collecting the wireless link status through the communication container, composite container node status data is constructed, and the composite container node status data is standardized and encapsulated into a structured data stream. The GRU time-series analysis module processes historical state data to extract trends in computing power and link changes; shared Critic and independent Actors collaborate to generate routing and computing power schemes, and make computing power allocation decisions to build a joint decision-making intelligent decision layer. Perform computing power-routing collaborative verification. When the link bandwidth cannot meet the computing power transmission requirements, switch to the backup path and reallocate computing power. The system parses and converts decision commands and protocols, deploys routing rules, and monitors link status in real time through BFD sessions. When the primary path is interrupted, it automatically switches to the backup path. Perform joint simulation of computing power and communication, calculate core indicators and use Prometheus to store data, and generate visualization results; transmit bottleneck reports to the intelligent decision-making layer to correct the prediction model and optimize the weights of the value function.

2. The computing power routing method based on multi-intelligence collaboration according to claim 1, characterized in that, During the construction of the composite container node status data, the management container collects the basic status of the nodes through Linux system tools, including satellite / drone / ground station equipment resources; The communication container interfaces with wmediumd enhanced version to collect wireless link status, including received signal strength, link transmission delay, and packet loss rate.

3. The computing power routing method based on multi-intelligence collaboration according to claim 1, characterized in that, In the construction steps of the intelligent decision-making layer joint decision-making, the GRU time series analysis module processes historical state data, extracts computing power and link change trends, and provides a predictive basis for decision-making; the shared Critic and independent Actors collaborate to calculate, generate routing and computing power schemes based on real-time state and time series prediction results, and make computing power allocation decisions.

4. The computing power routing method based on multi-intelligence collaboration according to claim 1, characterized in that, In the step of parsing and converting decision instructions into protocols, the decision instructions are divided into routing sub-instructions and computing sub-instructions, and converted into recognizable configuration commands; routing rules are deployed based on FRR, and the link status is monitored in real time through BFD sessions, and the system automatically switches to the backup path when the primary path is interrupted.

5. The computing power routing method based on multi-intelligence collaboration according to claim 1, characterized in that, In the process of performing joint simulation of computing power and communication, the execution layer data and air-space-ground topology configuration are input, the channel is simulated and the simulation data is output; the core indicators are calculated and the data is stored using Prometheus to generate visualization results; the bottleneck report is transmitted to the intelligent decision-making layer to correct the prediction model and optimize the weight of the value function.

6. The computing power routing method based on multi-intelligence collaboration according to any one of claims 1 to 5, characterized in that, The method supports heterogeneous networking of multi-level nodes such as satellites, drones, and ground base stations, and coordinates the scheduling of computing resources and communication links to achieve integrated communication and computing services.

7. The computing power routing method based on multi-intelligence collaboration according to any one of claims 1 to 5, characterized in that, The method improves the accuracy and adaptability of routing decisions by collecting link quality data in batches and using historical states as input to the agent.

8. The computing power routing method based on multi-intelligence collaboration according to any one of claims 1 to 5, characterized in that, The method integrates automated simulation, performance verification, and visualization modules, supports large-scale network testing and optimization, and verifies the optimization decision model through simulation.

Images