Heterogeneous network resource virtualization modeling and intelligent arrangement method and system

Abstract

The invention provides a heterogeneous network resource virtualization modeling and intelligent arrangement method and system based on a knowledge graph, and relates to unified modeling, dynamic retrieval and intelligent resource arrangement of heterogeneous network equipment. The method specifically comprises: 1, a unified modeling method based on a knowledge graph: integrating protocol attributes, dynamic states and topological relationships of heterogeneous devices such as a 5G base station, an SDN switch, a router, an Internet of Things gateway and the like into a structured knowledge graph, breaking the barrier of a manufacturer private data model, and realizing semantic-level collaborative scheduling of cross-domain resources; 2, designing a semantic retrieval engine: querying dynamic conditional reasoning through a natural language, replacing traditional manual rule definition, and improving retrieval response speed and accuracy; and 3, developing a graph-driven intelligent arrangement framework: combining a graph neural network (GNN) and reinforcement learning (RL), automatically generating a resource allocation strategy according to a real-time network state, reducing manual intervention and improving the resource utilization rate.

Description

Technical Field

[0001] The present invention relates to the technical field of communication network management. Specifically, it relates to a method and system for heterogeneous network resource virtualization modeling and intelligent orchestration based on a knowledge graph. Background Art

[0002] With the rapid development of 5G, the Internet of Things, and the industrial Internet, the communication network is undergoing profound changes, showing heterogeneous and highly dynamic characteristics. In this complex network environment, the types of network devices are extremely diverse, covering 5G base stations, traditional routers, edge computing nodes, and low-power sensors, etc. There are huge differences in protocol stacks, interface standards, and resource management strategies among different types of network devices.

[0003] This heterogeneity and high dynamics pose severe challenges to the management of cross-device resources, among which efficient coordination and dynamic orchestration become the core problems to be solved. From the current actual situation, there are many bottlenecks in the existing technologies when dealing with these challenges:

[0004] First, the fragmentation of resource representation is serious: The devices of each manufacturer are built based on their own independent data models and use different control interfaces, resulting in a fragmented state of resource description methods. Taking traditional virtualization technologies (such as NFV) as an example, although the physical resources are pooled and integrated to a certain extent, there is a lack of unified semantic modeling for the functional attributes and dynamic behaviors of the devices. This defect makes it extremely difficult to retrieve resources in a cross-domain environment and even more difficult to achieve efficient collaborative scheduling, seriously hindering the full utilization and collaborative work of network resources.

[0005] Second, the resource retrieval efficiency is low: The current resource management technologies mainly rely on static databases or rule engines. Facing complex and changing network states, such as sudden node failures and sudden traffic surges, it is difficult to make a quick response. Moreover, resource retrieval often requires manual definition of query conditions, which is not only inefficient but also unable to achieve semantic-level dynamic retrieval.

[0006] Third, the lack of resource orchestration ability: In terms of resource management, the existing technologies focus more on the storage management of resources and only pool computing and storage resources. However, its allocation strategy is relatively fixed. In actual applications, once the network state changes, it is necessary for operation and maintenance personnel to manually adjust parameters, lacking the ability to perceive the real-time state of resources and the ability of dynamic optimization and allocation. This makes it impossible to reasonably slice and efficiently allocate network resources according to real-time requirements.

[0007] Through the retrieval of patent documents, it is found that the invention patent with the publication number CN113722501A discloses a method, device, and storage medium for constructing a knowledge graph based on deep learning. The method includes: obtaining the constructed knowledge graph and extracting the first entity information in the constructed knowledge graph; collecting the information related to the first entity information in the constructed knowledge graph to obtain a data set; obtaining multiple natural paragraph sentences related to the first entity information in the data set; classifying the multiple natural paragraph sentences according to a preset unsupervised deep learning model to obtain multiple classification results; obtaining the natural paragraph sentences whose classification results meet the preset conditions in the multiple classification results to obtain relevant sentences, and adding the second entity information and relationship information corresponding to the first entity information in the constructed knowledge graph according to the relevant sentences. This patent focuses on the construction of the knowledge graph and does not involve the management of heterogeneous network resources. The invention patent with the publication number CN114126064A discloses a method and system for resource management for a 5G heterogeneous network, including a management device. A pulling groove is opened on the upper surface of the management device, a power supply board is slidably installed inside the pulling groove, a pulling buckle is installed on the upper surface of the power supply board, a water-cooled heat dissipation box is installed on the rear surface of the power supply board, a water inlet is installed on one side of the water-cooled heat dissipation box, one end of the water inlet is connected to a cooling pipe, one end of the cooling pipe is connected to a heat exchange pipe, and one end of the heat exchange pipe is connected to a circulating water pipe. This patent focuses on the heat dissipation, dust-proof structure, and operation of the management device and does not involve the virtualization modeling of heterogeneous network resources.

[0008] In summary, aiming at the problems of the above-mentioned existing technologies, researching a method and system for virtualization modeling and intelligent orchestration of heterogeneous network resources based on a knowledge graph has become a key task that needs to be solved urgently at present. Summary of the Invention

[0009] Aiming at the defects in the prior art, the purpose of the present invention is to provide a method and system for virtualization modeling and intelligent orchestration of heterogeneous network resources based on a knowledge graph.

[0010] According to a method for virtualization modeling and intelligent orchestration of heterogeneous network resources based on a knowledge graph provided by the present invention, the method includes the following steps:

[0011] Step S1, collecting multi-modal network resource data;

[0012] Step S2, based on the multi-modal network resource data, generating standardized data in JSON-LD format through protocol adaptation and semantic annotation;

[0013] Step S3, based on the standardized data, performing knowledge graph modeling to generate a dynamic knowledge graph;

[0014] Step S4, performing dynamic resource retrieval and conflict detection based on semantic reasoning, and outputting a matching resource list;

[0015] Step S5: Based on the matching resource list and the dynamic knowledge graph, perform graph-driven intelligent resource orchestration.

[0016] Preferably, step S1 includes the following sub-steps: deploying lightweight collection agents in 5G base stations, SDD switches and sensor heterogeneous devices to collect static attributes, dynamic status and topological relationships in real time. Static attributes include hardware specifications, protocol support list and interface type. Dynamic status includes real-time link bandwidth occupancy, end-to-end delay and packet loss rate. Topological relationships include physical connection information and logical dependencies.

[0017] Preferably, step S3 includes the following sub-steps:

[0018] Step S3.1, setting the network resource ontology model including entity types, relationship types and dynamic attribute extensions;

[0019] Step S3.2, constructing a dynamic knowledge graph based on the graph database;

[0020] Step S3.3: Set the incremental update algorithm to trigger local graph reconstruction when the device status change exceeds a threshold or the topological relationship is updated.

[0021] Preferably, step S4 includes the following sub-steps:

[0022] Step S4.1: Input a query command in natural language and generate a graph traversal command through the semantic parser;

[0023] Step S4.2: Call the time series prediction model to generate device state prediction, and embed the device state prediction into the graph traversal instruction to obtain an enhanced query statement;

[0024] Step S4.3: Execute the enhanced query statement in the graph database and return a list of matching resources.

[0025] Preferably, step S5 includes the following sub-steps:

[0026] Step S5.1: Build a resource-service association graph based on the dynamic knowledge graph and use a graph neural network to generate low-dimensional embedding vectors of devices, links, and services;

[0027] Step S5.2: Generate a dynamic scheduling strategy based on the low-dimensional embedding vector and real-time load data through reinforcement learning pre-training and multi-objective optimization;

[0028] In step S5.3, based on the dynamic orchestration strategy, resource configuration parameters are delivered through the southbound interface controller, and execution indicators are monitored.

[0029] Preferably, in step S5.1, the graph neural network adopts a 3-layer GraphSAGE model, the aggregation function is MEAN, the output dimension is 128, and the loss function is to maximize the embedding similarity of associated devices.

[0030] Preferably, step S5.2 includes the following sub-steps:

[0031] Step S5.2.1, construct a reinforcement learning model and train the reinforcement learning model based on fixed weights to generate a preliminary resource allocation strategy;

[0032] Step S5.2.2, use the NSGA-II algorithm with the preliminary resource allocation strategy as the initial population to generate Pareto front solutions, and convert the Pareto front solutions into device configuration parameters and store them in the policy switching table.

[0033] Preferably, in step S5.2.1, the state space of the reinforcement learning model is a combination of the embedding vector generated by the GNN and the real-time load matrix, the action space is the resource allocation decision; the reward function is reward = α*(1-delay_ratio)+β*energy_efficiency+γ*slice_priority, where delay_ratio is the ratio of the delay to the threshold, and α, β, γ are weight coefficients.

[0034] Preferably, step S5.3 includes the following sub-steps:

[0035] Step S5.3.1, send the resource configuration parameters corresponding to the Pareto front solutions to the devices through the southbound interface controller;

[0036] Step S5.3.2, monitor the execution metrics of the link bandwidth occupancy rate, end-to-end delay, and packet loss rate in real time. If the deviation of the execution metrics exceeds the threshold, trigger the update of the dynamic knowledge graph to form a "perception - decision - execution - optimization" closed loop.

[0037] The present invention also provides a heterogeneous network resource virtualization modeling and intelligent orchestration system based on a knowledge graph, including:

[0038] Module M1, which collects multi-modal network resource data;

[0039] Module M2, which generates standardized data in JSON-LD format based on the multi-modal network resource data through protocol adaptation and semantic annotation;

[0040] Module M3, which performs knowledge graph modeling based on the standardized data to generate a dynamic knowledge graph;

[0041] Module M4, which performs dynamic resource retrieval and conflict detection based on semantic reasoning and outputs a matching resource list;

[0042] Module M5 performs graph-driven intelligent resource orchestration based on a matching resource list and a dynamic knowledge graph.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] 1. The present invention unifies semantic modeling and breaks down the heterogeneous resource barriers: By constructing a heterogeneous network resource virtualization model based on a knowledge graph, the static attributes, dynamic states, and topological relationships of network devices such as 5G base stations, SDN switches, routers, and IoT gateways are unified and integrated into a structured knowledge graph. The use of semantic tag embedding and protocol adaptation plugins in JSON-LD format solves the defect that traditional virtualization technologies only pool physical resources but lack semantic modeling, breaks through the limitations of vendor-specific data models, realizes semantic-level collaborative scheduling of cross-domain resources, and improves the unified representation and collaborative capabilities of heterogeneous network resources.

[0045] 2. The present invention realizes dynamic semantic retrieval and resource matching: By designing a multimodal semantic retrieval engine that supports dynamic conditional reasoning for natural language queries, extracting semantic keywords through a BERT model and converting them into Cypher query statements to replace traditional manual rule definitions, the problem of low retrieval efficiency in static databases is solved. At the same time, combined with a time series prediction model, the prediction results are embedded in the retrieval conditions to achieve resource matching, avoiding resource allocation failures caused by sudden changes in device states and improving the real-time performance and accuracy of resource retrieval.

[0046] 3. The present invention realizes graph-driven multi-objective dynamic orchestration optimization: Based on a knowledge graph, a resource-service association graph is constructed, and a three-layer GraphSAGE model is used to generate low-dimensional embedding vectors of devices, links, and services. Through the collaborative mechanism of reinforcement learning and the NSGA-II multi-objective optimization algorithm, the problem that resource orchestration depends on a fixed-weight reward function and thus falls into a local optimum is solved. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Other features, objects, and advantages of the present invention will become more apparent by reading the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0048] Figure 1 It is the heterogeneous network resource virtualization modeling and intelligent orchestration framework based on a knowledge graph in an embodiment of the present invention;

[0049] Figure 2 It is an example of the JSON-LD data structure (SDN switch) in an embodiment of the present invention;

[0050] Figure 3 It is a schematic diagram of the knowledge graph modeling process in an embodiment of the present invention;

[0051] Figure 4 Schematic diagram of the semantic input parsing process in the embodiments of the present invention;

[0052] Figure 5 Schematic diagram of the graph-driven intelligent resource orchestration process in the embodiments of the present invention;

[0053] Figure 6 Schematic diagram of the graph-driven intelligent resource orchestration process in the embodiments of the present invention. Detailed implementation manners

[0054] The present invention will be described in detail below with reference to specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but do not limit the present invention in any form. It should be noted that those of ordinary skill in the art can make several changes and improvements without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0055] The present invention provides a heterogeneous network resource virtualization modeling and intelligent orchestration method and system based on a knowledge graph, which relates to unified modeling, dynamic retrieval, and intelligent resource orchestration of heterogeneous network devices. Specifically, it includes:

[0056] (1) Unified modeling method based on a knowledge graph: Integrate the protocol attributes, dynamic states, and topological relationships of heterogeneous devices such as 5G base stations, SDN switches, routers, and IoT gateways into a structured knowledge graph, break the barriers of vendor-specific data models, and achieve semantic-level collaborative scheduling of cross-domain resources.

[0057] (2) Design a semantic retrieval engine: Through natural language query dynamic condition reasoning, replace traditional manual rule definition, and improve the retrieval response speed and accuracy.

[0058] (3) Develop a graph-driven intelligent orchestration framework: Combine graph neural networks (GNNs) with reinforcement learning (RL), automatically generate resource allocation strategies according to the real-time network state, reduce manual intervention, and improve resource utilization.

[0059] Embodiment 1:

[0060] Figure 1 Heterogeneous network resource virtualization modeling and intelligent orchestration framework based on a knowledge graph in the embodiments of the present invention.

[0061] As Figure 1 shown, this embodiment provides a heterogeneous network resource virtualization modeling and intelligent orchestration method based on a knowledge graph, including the following steps:

[0062] Step S1, collect multi-modal network resource data.

[0063] Specifically, step S1 includes the following sub-steps: Deploy lightweight acquisition agents on 5G base stations, SDD switches, and sensor heterogeneous devices to collect static attributes, dynamic states, and topological relationships in real time. The static attributes include hardware specifications, protocol support lists, and interface types. The dynamic states include real-time rows, and the topological relationships include physical connection information and logical dependency relationships.

[0064] Step S2, based on the multi-modal network resource data, generate standardized data in JSON-LD format through protocol adaptation and semantic annotation.

[0065] Figure 2 This is an example of the JSON-LD data structure in the embodiments of the present invention (SDN switch).

[0066] As Figure 2 shown, through the protocol adaptation plugin, define unified data conversion rules for different device interfaces, convert the collected data into JSON-LD format and embed semantic tags.

[0067] Step S3, based on the standardized data, perform knowledge graph modeling to generate a dynamic knowledge graph.

[0068] Figure 3 This is a schematic diagram of the knowledge graph modeling process in the embodiments of the present invention.

[0069] As Figure 3 shown, step S3 includes the following sub-steps:

[0070] Step S3.1, set the network resource ontology model including entity types, relationship types, and dynamic attribute extensions.

[0071] In this embodiment, the entity types include Device (device), Interface (interface), Service (service), and Slice (slice); the relationship types include connectedTo (physical connection), dependsOn (logical dependency), and conflictsWith (resource conflict); the dynamic attribute extensions include adding temporal attributes (such as device.cpu_utilization[timestamp]) and prediction attributes (such as link.future_congestion_probability) to the entities.

[0072] Step S3.2, construct a dynamic knowledge graph based on the graph database.

[0073] In this embodiment, the node attributes are as follows:

[0074] CREATE(:Device{

[0075] id: "EdgeNode_01",

[0076] type: "EdgeServer",

[0077] vendor: "Huawei",

[0078] cpu_cores: 16,

[0079] mem_capacity: "64GB",

[0080] protocols: ["IPv6", "MQTT"],

[0081] realtime_load: 45.3 / / Dynamic update field

[0082] [[ID=21}}

[0083] In this embodiment, the Neo4j graph database is used to store the dynamic knowledge graph.

[0084] Step S3.3: Set the incremental update algorithm. When the device status changes exceed the threshold or the topological relationship is updated, trigger the local graph reconstruction to ensure the dynamic update of the graph.

[0085] In this embodiment, the incremental update algorithm only reconstructs the affected subgraph. For example, when the bandwidth utilization rate of a certain link changes suddenly, update the connectedTo edge attribute associated with it and the realtime_load attribute of the adjacent nodes.

[0086] Step S4: Perform dynamic resource retrieval and conflict detection based on semantic reasoning, and output the matching resource list.

[0087] Figure 4 It is a schematic diagram of the semantic input parsing process in the embodiment of the present invention.

[0088] As Figure 4 shown, step S4 includes the following sub-steps:

[0089] Step S4.1: Input a natural language query instruction, and generate a graph traversal instruction through a semantic parser.

[0090] Specifically, use the BERT model to extract the keywords of the natural language query, and then convert the keywords into Cypher query statements through conditional mapping;

[0091] Input example: "Find edge nodes that support SRv6 and have remaining memory > 20%, and sort them in ascending order of delay"

[0092] Parsing process:

[0093] 1. Semantic parsing: Use the BERT model to extract keywords (SRv6, remaining memory > 20%, latency sorting).

[0094] 2. Conditional mapping: Convert keywords into Cypher query statements:

[0095] MATCH(n:Device{type:"EdgeServer"})

[0096] WHERE"SRv6"IN n.protocols AND n.mem_capacity - n.mem_usage > 20%

[0097] RETURN n ORDER BY n.avg_latency ASC

[0098] Step S4.2, call the time series prediction model to generate device status predictions, and embed the device status predictions into the graph traversal instruction to obtain an enhanced query statement.

[0099] The enhanced query statement realizes forward-looking matching by fusing real-time status and device status predictions.

[0100] For example: Predict the probability that the load of a certain base station exceeds 80% in the next 5 minutes. If the probability > 70%, then exclude this node from the retrieval results.

[0101] Step S4.3, execute the enhanced query statement in the graph database and return the list of matching resources.

[0102] Step S5, based on the list of matching resources and the dynamic knowledge graph, perform graph-driven intelligent resource orchestration.

[0103] Figure 5 It is a schematic diagram of the graph-driven intelligent resource orchestration process in the embodiments of the present invention.

[0104] As Figure 5 shown, step S5 includes the following sub-steps:

[0105] Step S5.1, construct a resource-service association graph based on the dynamic knowledge graph, and use a graph neural network (GNN) to generate low-dimensional embedding vectors of devices, links, and services.

[0106] Further, in step S5.1, the graph neural network adopts a 3-layer GraphSAGE model, the aggregation function is MEAN, the output dimension is 128, and the loss function is to maximize the embedding similarity of associated devices (such as belonging to the same Slice).

[0107] Step S5.2: Based on the low-dimensional embedding vectors and real-time load data, generate a dynamic orchestration policy through reinforcement learning pre-training and multi-objective optimization.

[0108] Furthermore, step S5.2 includes the following sub-steps:

[0109] Step S5.2.1: Build a reinforcement learning model and train the reinforcement learning model based on fixed weights to generate a preliminary resource allocation policy, which can quickly adapt to the dynamic environment.

[0110] In this embodiment, in step S5.2.1, the state space (State) of the reinforcement learning model is the combination of the embedding vectors generated by the GNN and the real-time load matrix, and the action space (Action) is the resource allocation decision (such as allocating 10 MHz spectrum to slice A); the reward function is reward = α * (1 - delay_ratio) + β * energy_efficiency + γ * slice_priority, where delay_ratio is the ratio of the delay to the threshold, and α, β, and γ are weight coefficients.

[0111] Step S5.2.2: Use the NSGA-II algorithm to generate Pareto front solutions with the preliminary resource allocation policy as the initial population, and convert the Pareto front solutions into device configuration parameters and store them in the policy switching table.

[0112] Specifically, each solution of the Pareto front solutions represents a trade-off scheme for objectives such as delay and energy efficiency, and the resource allocation policy can be dynamically selected from them according to real-time requirements.

[0113] In this embodiment, the reinforcement learning model learns the policy by maximizing the cumulative reward. However, if there are conflicts among multiple objectives (such as delay, energy efficiency, and cost), a single linear weighted reward function (such as reward = αA + βB) may cause the model to fall into a local optimum and fail to find the global equilibrium solution. Moreover, the network requirements may change over time, and it is necessary to frequently adjust α, β, and γ manually, lacking flexibility. Therefore, on this basis, the NSGA-II algorithm is used to generate a set of non-dominated solutions (Pareto front), and each solution represents the trade-off result between different objectives. The operation and maintenance personnel can select the most suitable solution according to the current scenario instead of relying on fixed weights.

[0114] Step S5.3: Based on the dynamic orchestration policy, send the resource configuration parameters through the southbound interface controller and monitor the execution metrics.

[0115] Specifically, step S5.3 includes the following sub-steps:

[0116] Step S5.3.1: Send the resource configuration parameters corresponding to the Pareto front solutions to the device through the southbound interface controller.

[0117] Step S5.3.2: Monitor execution metrics such as link bandwidth occupancy rate, end-to-end delay, and packet loss rate in real time. If the deviation of the execution metrics exceeds the threshold, trigger the update of the dynamic knowledge graph to form a "perception - decision - execution - optimization" closed loop.

[0118] Furthermore, when a network demand change (such as a change in Slice priority) is detected, automatically select an appropriate Pareto front solution according to the policy switching table without retraining the reinforcement learning model.

[0119] Embodiment 2:

[0120] In the implementation of this example, an independently developed resource orchestration system is adopted to construct a dynamic knowledge graph based on the Neo4j graph database. The network topology is shown as Figure 6 shown, including the following heterogeneous devices:

[0121] Communication Node 1: Edge computing node (connected to sensors, deploying lightweight data collection agents)

[0122] Communication Node 2: 5G base station (providing 5G terminal access, deploying lightweight data collection agents)

[0123] Communication Node 3: Traditional router (providing wide - area interconnection, deploying lightweight data collection agents)

[0124] Connection method:

[0125] Inter - device interconnection: Achieve intercommunication through optical fiber / Ethernet, supporting protocol collaboration.

[0126] Southbound interface: The intelligent resource orchestration system communicates with all devices through the southbound interface to achieve data collection and orchestration policy distribution.

[0127] A heterogeneous network resource virtualization modeling and intelligent orchestration method based on a knowledge graph provided in this embodiment includes the following steps:

[0128] Step 1: Multimodal data collection and knowledge graph construction

[0129] 1. Data collection:

[0130] 5G base station: Collect wireless channel status (such as 5G signal strength, spectrum utilization rate, slice status, etc.).

[0131] IoT gateway: Collect sensor data traffic, protocol type, end - to - end delay, etc.

[0132] Traditional router: Collect routing tables, link status information, etc.

[0133] 2. Protocol adaptation

[0134] The intelligent resource orchestration system uniformly converts the collected heterogeneous device status data into the JSON-LD format through protocol plug-ins.

[0135] 3. Knowledge graph construction

[0136] Define entity relationships in Neo4j using the Cypher language. For example:

[0137]

[0138] That is: create a 5G base station node with the identifier "5G_BaseStation_01", the type "5G_BS", supporting the SRv6 protocol, with a spectrum utilization rate of 65% and a latency of 25 ms.

[0139] Step 2: Dynamic semantic retrieval and resource matching

[0140] 1. Natural language query: "Find all device nodes of type '5G_BS' that support the SRv6 protocol and have a spectrum utilization rate lower than 70%", and the query results will be sorted in ascending order of latency.

[0141] 2. Parse into a Cypher query:

[0142] MATCH(n:Device{type:"5G_BS"})

[0143] WHERE"SRv6"IN n.protocols AND n.realtime_load.spectrum_utilization<70

[0144] RETURN n ORDER BY n.latency ASC

[0145] Step 3: Graph-driven intelligent orchestration

[0146] 1. Embedding vector generation:

[0147] Use the GraphSAGE model to generate 128-dimensional embedding vectors for devices and links, and associate network resources (such as binding low-spectrum-utilization nodes with low-latency links).

[0148] 2. Multi-objective optimization:

[0149] Scenario: Allocate resources for the 4K online video service, which needs to simultaneously meet the latency < 20 ms and the minimum bandwidth ≥ 50 Mbps.

[0150] Policy generation:

[0151] 1) RL pre-training generates a basic policy (such as allocating 10 MHz spectrum resources for 5G base station 1).

[0152] 2) NSGA-II generates a Pareto solution set, providing options of "delay priority" or "bandwidth priority".

[0153] Step 4: Policy distribution and closed-loop optimization

[0154] 1. Policy distribution:

[0155] Send the slice configuration (xml) to the 5G base station through the southbound interface:

[0156] <config>

[0157] <slice id="Video_Slice">

[0158] <bandwidth>10MHz< / bandwidth>

[0159] <priority>high< / priority>

[0160]

[0161] < / config>

[0162] 2. Closed-loop monitoring:

[0163] Real-time monitor the delay and bandwidth. If the delay > 25 ms, trigger the update of the knowledge graph and re-allocate resources (such as switching to 5G base station 2).

[0164] Embodiment 3:

[0165] The present invention also provides a heterogeneous network resource virtualization modeling and intelligent orchestration system based on a knowledge graph. The heterogeneous network resource virtualization modeling and intelligent orchestration system based on a knowledge graph can be implemented by executing the process steps of the heterogeneous network resource virtualization modeling and intelligent orchestration method based on a knowledge graph. That is, those skilled in the art can understand the heterogeneous network resource virtualization modeling and intelligent orchestration method based on a knowledge graph as the preferred embodiment of the heterogeneous network resource virtualization modeling and intelligent orchestration system.

[0166] Specifically, the heterogeneous network resource virtualization modeling and intelligent orchestration system based on a knowledge graph includes:

[0167] Module M1, which collects multi-modal network resource data;

[0168] Module M2, which generates standardized data in JSON-LD format based on the multi-modal network resource data through protocol adaptation and semantic annotation;

[0169] Module M3, which performs knowledge graph modeling based on the standardized data to generate a dynamic knowledge graph;

[0170] Module M4, which performs dynamic resource retrieval and conflict detection based on semantic reasoning and outputs a matching resource list;

[0171] Module M5, which performs graph-driven intelligent resource orchestration based on the matching resource list and the dynamic knowledge graph.

[0172] Those skilled in the art know that, in addition to implementing the system and its various devices, modules, and units provided by the present invention in the form of pure computer-readable program code, the method steps can be logically programmed to enable the system and its various devices, modules, and units provided by the present invention to be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, embedded microcontrollers, etc. to achieve the same functions. Therefore, the system and its various devices, modules, and units provided by the present invention can be considered as a kind of hardware component, and the devices, modules, and units included therein for implementing various functions can also be regarded as the structures within the hardware component; the devices, modules, and units for implementing various functions can also be regarded as either software modules for implementing the method or structures within the hardware component.

[0173] The specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. Without conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A heterogeneous network resource virtualization modeling and intelligent orchestration method based on a knowledge graph, characterized in that, The following steps are involved: Step S1, collecting multimodal network resource data; Step S2: Based on the multimodal network resource data, standardized data in JSON-LD format is generated through protocol adaptation and semantic annotation; Step S3: Based on the standardized data, knowledge graph modeling is performed to generate a dynamic knowledge graph; Step S4: dynamic resource retrieval and conflict detection based on semantic reasoning, and outputting a matching resource list; Step S5: Perform graph-driven intelligent resource orchestration based on the matching resource list and the dynamic knowledge graph.

2. The heterogeneous network resource virtualization modeling and intelligent orchestration method based on a knowledge graph according to claim 1, characterized in that The step S1 includes the following sub-steps: deploying lightweight collection agents on 5G base stations, SDD switches and sensor heterogeneous devices to collect static attributes, dynamic status and topological relationships in real time. The static attributes include hardware specifications, protocol support list and interface type; the dynamic status includes real-time link bandwidth occupancy, end-to-end delay and packet loss rate; the topological relationship includes physical connection information and logical dependency.

3. A heterogeneous network resource virtualization modeling and intelligent orchestration method based on a knowledge graph according to claim 1, characterized in that The step S3 includes the following sub-steps: Step S3.1, setting the network resource ontology model including entity types, relationship types and dynamic attribute extensions; Step S3.2, constructing a dynamic knowledge graph based on the graph database; Step S3.3: Set the incremental update algorithm to trigger local graph reconstruction when the device status change exceeds a threshold or the topological relationship is updated.

4. A method for virtualization modeling and intelligent orchestration of heterogeneous network resources based on a knowledge graph according to claim 1, characterized in that, The step S4 includes the following sub-steps: Step S4.1: Input a query command in natural language and generate a graph traversal command through the semantic parser; Step S4.2, calling the time series prediction model to generate a device state prediction, and embedding the device state prediction into the graph traversal instruction to obtain an enhanced query statement; Step S4.3: execute the enhanced query statement in the graph database and return a list of matching resources.

5. A method for virtualization modeling and intelligent orchestration of heterogeneous network resources based on a knowledge graph according to claim 1, characterized in that, The step S5 includes the following sub-steps: Step S5.1, constructing a resource-service association graph based on the dynamic knowledge graph, and using a graph neural network to generate low-dimensional embedding vectors of devices, links, and services; Step S5.2, generating a dynamic scheduling strategy based on the low-dimensional embedding vector and real-time load data through reinforcement learning pre-training and multi-objective optimization; Step S5.3: Based on the dynamic orchestration strategy, resource configuration parameters are issued through the southbound interface controller, and execution indicators are monitored.

6. The heterogeneous network resource virtualization modeling and intelligent orchestration method based on a knowledge graph according to claim 5, characterized in that, In step S5.1, the graph neural network adopts a 3-layer GraphSAGE model, the aggregation function is MEAN, the output dimension is 128, and the loss function is to maximize the embedding similarity of associated devices.

7. A method for virtualized modeling and intelligent orchestration of heterogeneous network resources based on a knowledge graph according to claim 5, characterized in that, The step S5.2 includes the following sub-steps: Step S5.2.1, constructing a reinforcement learning model, and training the reinforcement learning model based on fixed weights to generate a preliminary resource allocation strategy; Step S5.2.2, using the preliminary resource allocation strategy as the initial population, using the NSGA-II algorithm to generate a Pareto frontier solution, and converting the Pareto frontier solution into device configuration parameters, and storing them in the strategy switching table.

8. A method for virtualization modeling and intelligent orchestration of heterogeneous network resources based on a knowledge graph according to claim 7, characterized in that, In the step S5.2.1, the state space of the reinforcement learning model is a combination of the embedding vector generated by the GNN and the real-time load matrix, the action space is the resource allocation decision; the reward function is reward = α*(1-delay_ratio)+β*energy_efficiency+γ*slice_priority, where delay_ratio is the ratio of the delay to the threshold, and α, β, and γ are weight coefficients.

9. A method for virtualization modeling and intelligent orchestration of heterogeneous network resources based on a knowledge graph according to claim 5, characterized in that, Step S5.3 includes the following sub-steps: Step S5.3.1, send the resource configuration parameters corresponding to the Pareto front solution to the device through the southbound interface controller; Step S5.3.2, monitor the execution metrics of the link bandwidth occupancy rate, end-to-end delay, and packet loss rate in real time. If the deviation of the execution metrics exceeds the threshold, trigger the update of the dynamic knowledge graph to form a "perception - decision - execution - optimization" closed loop.

10. A heterogeneous network resource virtualization modeling and intelligent orchestration system based on a knowledge graph, characterized in that, It includes: Module M1, which collects multi-modal network resource data; Module M2, based on the multi-modal network resource data, generates standardized data in JSON-LD format through protocol adaptation and semantic annotation; Module M3, based on the standardized data, conducts knowledge graph modeling to generate a dynamic knowledge graph; Module M4, conducts dynamic resource retrieval and conflict detection based on semantic reasoning, and outputs a matching resource list; Module M5, based on the matching resource list and the dynamic knowledge graph, conducts graph-driven intelligent resource orchestration.

Images