A network digital twin topology dynamic incremental updating method based on a heterogeneous graph neural network

Abstract

This invention relates to the field of network digital twin technology, specifically to a method for dynamic incremental update of network digital twin topology based on heterogeneous graph neural networks. The method includes the following steps: constructing a corresponding heterogeneous topology base graph based on the physical network; mapping devices and connections in the network to nodes and edges with type identifiers according to a preset node and edge type system; writing a feature field for each node; obtaining the current network digital twin topology graph; performing topology change event perception based on the heterogeneous topology base graph; mapping the collected topology change events to incremental change units; and obtaining the corresponding set of nodes in the affected domain according to the category and scope of the change events. This invention, by grouping by edge type and aggregating neighbor node information, can efficiently update a specific part of the network topology when the topology structure changes, without reconstructing the entire topology, thereby significantly improving the efficiency of network management.

Description

Technical Field

[0001] This invention relates to the field of network digital twin technology, and in particular to a method for dynamic incremental update of network digital twin topology based on heterogeneous graph neural networks. Background Technology

[0002] With the rapid development of information technology, computer networks have become an indispensable infrastructure in various industries and fields of modern society. Network topology is an important model that describes the various device nodes in the network and the connection relationships between them. It is a core component of network architecture. Effective network topology management can not only improve network performance and enhance fault recovery capabilities, but also play a key role in network planning, optimization and security management.

[0003] Currently, network topology management faces the following challenges:

[0004] Dynamic changes in network topology: As the network scale continues to expand, the addition, removal, configuration modification, and changes in network connection status of devices occur frequently. These changes affect the overall structure and performance of the network and need to be reflected in the topology diagram in a timely and accurate manner. Therefore, how to dynamically update the network topology diagram and ensure the consistency and accuracy of the topology is an important issue.

[0005] The complexity of managing heterogeneous network environments: In modern networks, there are different types of devices, connection methods and communication protocols. These heterogeneous network elements make it difficult for traditional network topology management methods to cope with complex and ever-changing topologies. Therefore, how to integrate and manage different types of network elements through reasonable technical means is an urgent problem to be solved.

[0006] Incremental update problem: For large-scale networks, reconstructing or updating the entire network topology is a very complex and computationally intensive task. To improve efficiency, incremental updates are usually adopted, updating only the parts that have changed. However, incremental updates are often accompanied by issues such as consistency verification and conflict resolution. How to ensure that the network topology remains consistent during the incremental update process and avoid erroneous updates and conflicts is a key issue in the dynamic update of network topology.

[0007] Real-time performance and accuracy of network topology diagrams: The real-time performance and accuracy of network topology diagrams are crucial for network operation and management, especially during network failures, performance monitoring, and traffic analysis. Changes in network topology need to be updated and reflected in the system in a timely manner. This requires that network topology updates not only be fast but also ensure data consistency to avoid network problems from being unable to be resolved in a timely manner due to update delays or errors.

[0008] To address the aforementioned issues, solutions based on emerging technologies such as graph neural networks and heterogeneous graph neural networks have emerged in recent years. These advanced methods enable more effective incremental updates, feature aggregation, and state repair of network topology graphs, thereby achieving dynamic management of complex network environments.

[0009] Most existing technologies focus on how to perform static or full updates of the topology using traditional methods, but their support for incremental updates, conflict resolution, and heterogeneous network management is relatively limited. In order to cope with the dynamic changes and complexity in network topology updates and improve the intelligence level of network management, a new method is urgently needed to achieve efficient, accurate, and real-time dynamic updates of the topology. Summary of the Invention

[0010] This invention provides a method for dynamic incremental update of network digital twin topology based on heterogeneous graph neural networks.

[0011] A method for dynamic incremental update of digital twin topology based on heterogeneous graph neural networks includes the following steps:

[0012] Based on the physical network, a corresponding heterogeneous topology base graph is constructed. According to the preset node and edge type system, the devices and connection relationships in the network are mapped to nodes and edges with type identifiers, and a feature field is written for each node.

[0013] Obtain the current network digital twin topology;

[0014] Topology change event perception is performed based on heterogeneous topology base map, and the collected topology change events are mapped into incremental change units, and the change events are classified according to their category and scope of impact.

[0015] Obtain the corresponding set of influence domain nodes;

[0016] Based on the influence domain node set grouped by edge type, local neighbor sampling and heterogeneous graph neural network aggregation are performed;

[0017] Obtain the local update embedding result and write it back to the heterogeneous topology base graph to form the updated target topology graph.

[0018] Based on the updated target topology graph, perform version registration and consistency verification to obtain topology increment records:

[0019] If the verification passes, the updated topology version is output. If the verification fails, the conflicting nodes are replayed sequentially to fix the issues before the updated topology version is output.

[0020] As a further technical solution of the present invention, device entity information and connection relationship information in the physical network to be maintained are collected, and the collected network topology is formally represented as a combination structure of node set and edge set, defined as: ;in, To represent the set of device entities in a network, To represent the set of connections between device entities, each device entity is tagged with a type according to a preset node type system to form a corresponding node set, and each connection relationship is tagged with a type according to a preset edge type system to form a corresponding edge set. Semantic tagging is then applied to the node set and edge set to obtain: ;in, This represents the overall heterogeneous topology structure in a network digital twin. A collection of node types A set of edge types;

[0021] A heterogeneous topological framework is constructed based on the node set and edge set, and a set of feature fields matching the type of each node is established. The node feature field set is defined as follows: ;in, Represents the set of node features. Represents a node eigenvectors, Represents a set of nodes For any node in the network, the set of feature fields is configured differently based on the node type.

[0022] As a further technical solution of the present invention, real-time status data corresponding to each node is obtained and written into its corresponding feature field to form a node feature representation, which is then bound to the heterogeneous topology framework to generate a complete heterogeneous topology base graph, which is represented as follows: ;in, This represents the generated heterogeneous topology base map of the network digital twin. Represents the set of node features.

[0023] As a further technical solution of the present invention, multi-source topology change signals are collected from the physical network corresponding to the heterogeneous topology base map. The multi-source topology change signals are represented as follows: ;in, Represents a set of multi-source topology change signals. Indicates the first A change signal, The number of signals is indicated by the topology change signals, which include at least forwarding path change signals, device status change signals, control plane adjacency relationship change signals, and virtual resource change signals. A unified event extraction is performed on each type of signal to generate a corresponding topology change event set, which is represented as follows:

[0024] ;in, Represents a set of topology change events. Indicates the first One topology change event, Indicates the number of events;

[0025] Each topology change event in the set of topology change events is mapped to an incremental change unit according to a preset rule. Each incremental change unit includes at least a change object identifier, a change type identifier, and state information before and after the change, used to describe the state changes of the corresponding node or edge in the heterogeneous topology base graph. The incremental change unit is represented as follows:

[0026] ;in, Indicates the first Incremental change unit Indicates the change of object identifier. Indicates the change type identifier. Indicates the state before the change. This indicates the status after the change.

[0027] As a further technical solution of the present invention, the incremental change units are classified according to the object category of the topology change event. The object category includes at least node existence change, edge topology change, and node attribute change. A corresponding processing method is determined based on different categories, and the classification mapping function is defined as follows: ;in, Represents the classification mapping function, Indicates the first The object category corresponding to each incremental change unit.

[0028] As a further technical solution of the present invention, based on the object category and influence range of the incremental change unit, an influence domain node set is determined in the heterogeneous topology base graph, and the influence domain node set is represented as follows: ;in, Indicates the first The set of influence domain nodes corresponding to each incremental change unit Let represent the set of nodes in the heterogeneous topological base graph, where;

[0029] When the incremental change unit is a node existence change, the set of nodes in the affected domain includes the changed node and its neighboring nodes within a preset hop range;

[0030] When the incremental change unit is an edge topology change, the set of nodes in the affected domain includes the nodes at both ends of the changed edge and its neighboring nodes within a preset hop count range;

[0031] When the incremental change unit is a node attribute change, the affected domain node set includes the set of nodes whose attributes have changed.

[0032] As a further technical solution of the present invention, based on the set of nodes in the influence domain, corresponding local neighbor information is extracted from the heterogeneous topological base graph, and the neighbors are grouped according to edge type to form a set of neighbor subsets distinguished by edge type, as shown below: ;in, Represents a node The set of neighbor subsets Represents nodes By edge type The set of connected neighboring nodes.

[0033] As a further technical solution of the present invention, the neighbor subset set Perform heterogeneous graph neural network aggregation operation, where:

[0034] For each neighbor subset Feature fusion is performed using the aggregation function of a heterogeneous graph neural network, and the local result is represented as follows: : ;in, Representing neighboring nodes eigenvectors, Represents a node Subset of neighbors Feature representation after aggregation;

[0035] The aggregation results of all neighbor subsets are merged to obtain the final local update embedding result, represented as follows: : ;in, This indicates that the aggregation results are being merged.

[0036] Embed the local update result Write back to the nodes in the heterogeneous topology base graph The feature fields are analyzed, and the node states are updated to form the updated target heterogeneous topology graph. .

[0037] As a further technical solution of the present invention, the incremental change units and their corresponding local update embedding results in the target heterogeneous topology graph are timestamped to generate a topology incremental record with timestamps. The timestamps represent the time when the incremental change occurs and are used to record the historical order of topology changes. The time when an incremental change occurs is represented as follows: ;in, Incremental change unit The corresponding timestamp, This indicates the time of change in the current system time or network status.

[0038] As a further technical solution of the present invention, the topology incremental record is subjected to consistency verification to check whether the incremental change unit and the local update embedding result are consistent with the existing topology version. If the consistency verification passes, the incremental change unit is written into the topology storage module to update the target heterogeneous topology graph; if the consistency verification fails, a conflict repair operation is performed.

[0039] If the consistency check fails, a sequential replay repair operation is performed based on the state information before and after the change in the topology incremental record. By replaying the change history and adjusting the state of nodes or edges, conflicts are resolved and consistency is restored. Finally, the updated target heterogeneous topology graph is output.

[0040] The beneficial effects of this invention are:

[0041] This invention employs a heterogeneous graph neural network for incremental updates of the network topology graph. It can perform precise local updates based on the characteristics of different types of edges and nodes. By grouping by edge type and aggregating neighbor node information, it can efficiently update specific parts of the network topology when the topology changes, without having to reconstruct the entire topology, thereby significantly improving the efficiency of network management.

[0042] This invention, through the conflict repair mechanism of the incremental change unit, can effectively handle the situation where the consistency check fails during the incremental update of the topology. By tracing back the change history, adjusting the state of nodes or edges, and performing sequential replay repair, it ensures that the network topology remains consistent during the incremental update process, avoids inconsistent or erroneous updates of topology data, and thus guarantees the accuracy and reliability of the network topology graph. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 This is a schematic diagram of the process method of the present invention;

[0045] Figure 2 This is a schematic diagram of the process method S1 of the present invention;

[0046] Figure 3 This is a schematic diagram of the process method S2 of the present invention. Detailed Implementation

[0047] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. For some well-known technologies, those skilled in the art may also use other alternative methods to implement the invention. Moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0048] like Figure 1-3 As shown, a dynamic incremental update method for network digital twin topology based on heterogeneous graph neural network is described. In step S1, a heterogeneous topology base graph corresponding to the network digital twin is constructed for the physical network to be maintained. According to the preset node type system and edge type system, different device entities and different connection relationships in the network are mapped to nodes and edges with type identifiers, and corresponding feature fields are written for each node to form the heterogeneous topology base graph of the current network digital twin.

[0049] like Figure 2 As shown in S11, firstly, in terms of obtaining device entity information, various types of devices in the network are uniformly identified and registered. Specifically, the device list is obtained through the network controller interface, and the instance registration information in the address allocation record is combined to identify physical devices, virtual devices, and service instances in the network. For each identified device entity, its unique identifier, such as device ID, management address or instance ID, and basic attribute information are extracted to form a complete set of device entities.

[0050] In terms of acquiring connection relationship information, it is necessary to identify the actual connection or logical association between devices. For physical layer connection relationships, the direct connection between devices is determined through the link layer discovery mechanism or port peer information. For network layer connection relationships, the path association between devices is determined through the routing protocol adjacency relationship. For connection relationships in virtual networks or cloud environments, they are obtained through tunnel configuration, virtual switch mapping relationship or service registration and call relationship. In other words, connection relationships with different semantics can be identified, and the starting device and target device of the connection can be identified, thereby forming a set of connection relationships.

[0051] Finally, the device entity information and connection relationship information mentioned above are organized in a unified manner. Each device entity is abstracted as a node, and each connection relationship is abstracted as an edge. An association mapping relationship between nodes and edges is established. Specifically, each edge records the starting node and ending node of its connection, and the edges associated with it are registered in the corresponding node. This allows the corresponding node to be found through the edge, and all related connections can be quickly found through the node, thus forming a complete topological association structure. This transforms the device and connection structure in the real network into a structured network topology representation.

[0052] The collected network topology is then formally represented as a combination of node sets and edge sets, defined as: ;in, To represent the set of device entities in a network, This represents the set of connection relationships between device entities.

[0053] By introducing a node type system and an edge type system, and semantically tagging the node set and edge set, we obtain: That is, when constructing the node set, each node is assigned to a corresponding node type based on the actual attributes of the device; when constructing the edge set, each edge is assigned to a corresponding edge type based on the actual semantics of the connection relationship; and the type label is recorded in the data structure of the node or edge, so that each node carries its type identifier and each edge also carries its type identifier, thereby completing the semantic labeling of the node set and edge set; among which, This represents the overall heterogeneous topology graph structure in a network digital twin, used to describe the network's nodes, connections, and their types. To create a set of node types, we first identify all types of devices actually existing in the network and categorize them according to their functional roles, deployment methods, or operating environments. These classifications are then deduplicated and organized to form a finite and enumerable set of types, i.e., the node type set, used to identify different device entity categories. As a set of edge types, we analyze the connection methods between different nodes and their actual functions. We divide the connections according to their source and semantics, and summarize and organize these connection types with different semantics to form a standardized set of connection relationship types, namely the edge type set, which is used to identify the semantics of different connection relationships.

[0054] To represent nodes Type identifier, To represent the edge The process involves several steps: First, each node or edge is attribute-based. For nodes, their category is determined based on their actual attributes, such as whether they are physical devices, virtual instances, or functional components. For edges, their type is determined based on the semantic meaning of their connections, such as whether they are physical connections, routing relationships, or service calls. Then, type assignment is performed based on the identification results. This involves selecting the type that best matches the node's attributes or semantics from a predefined set of node or edge types and using this type as the unique type identifier for that node or edge. Finally, this type identifier is written into the corresponding data structure for recording. Specifically, a type field is added to the node's data record to store the node's type identifier, and a type field is also added to the edge's data record to store the edge's type identifier. Thus, each node and each edge carries explicit type information. Through this process of identification, classification, and writing, the specific elements in the node and edge sets are mapped to specific types in the corresponding type sets.

[0055] The node types include at least one or more of the following: physical switching device nodes, physical routing device nodes, virtual switching nodes, network function virtualization instance nodes, terminal server nodes, and containerized application instance nodes; the edge types include at least one or more of the following: physical link relationships, routing adjacency relationships, virtual tunnel relationships, and service call relationships; through the above-mentioned typified tags, the heterogeneous semantic expression of the original network topology is completed, forming a set of nodes and a set of edges with type labels.

[0056] S12, after completing the type labeling of the node set and edge set, a heterogeneous topology framework is constructed based on the node set and edge set. Specifically, each node in the node set is regarded as a structural unit in the topology, and each edge in the edge set is regarded as a connection relationship between nodes. According to the starting node and ending node recorded by the edge, these nodes are connected in pairs to form an overall network structure. At the same time, the type identifier of the node, the type identifier of the edge, and the corresponding node feature information are bound to the structure. Finally, a complete heterogeneous topology framework containing connection relationships, type semantics, and node attribute information is obtained, and a feature field set matching its type is established for each node.

[0057] Define the set of node feature fields as follows: ;in, This represents the node feature set, used to store the feature information of all nodes. Represents a node The feature vector is used to describe the attributes of the node. Represents a set of nodes Any node in the array; different types of nodes correspond to different sets of feature fields with different dimensions and semantics.

[0058] Configure differentiated feature fields for different node types:

[0059] For physical switching device nodes, the characteristic fields include device identifier, total number of ports, number of active ports, device model category code, management level code, CPU utilization, memory utilization, and in-band telemetry version tag.

[0060] For virtual nodes or container instance nodes, the characteristic fields include instance identifier, service to which it belongs, resource usage, and lifecycle status;

[0061] For network function virtualization instance nodes, the characteristic fields include function type, processing capacity indicators, and running status information.

[0062] Implement type-related configurations for node feature fields to ensure that node feature representations are consistent with their physical or logical attributes.

[0063] S13, after completing the feature field definition, the real-time operation information generated by various devices, protocols and operating systems in the network is continuously collected and aggregated according to nodes to obtain the real-time status data corresponding to each node, and written into its corresponding feature field. Specifically, after obtaining the real-time status data, it is matched to the corresponding field position according to the content type and source of the data, such as writing resource occupancy information into the resource class field, writing the running status into the status class field, etc. Finally, on a node-by-node basis, these matched data are filled into its feature structure, so that all feature fields of the node are updated by the current status data, forming a complete node feature representation.

[0064] The heterogeneous topology framework is bound to a set of node features. The binding process uses the node's unique identifier as an index to match the node's structural information with its corresponding feature data, and then appends this feature data to the node. This ensures that a node not only represents a point in the structure but also carries its state and attribute information, resulting in a complete heterogeneous topology base graph, defined as: ;in, This represents the generated heterogeneous topology base map of the network digital twin. It represents the set of node features, which consists of all node feature vectors.

[0065] The above binding process enables the topology information and node status information to form a unified representation, thereby constructing a digital twin heterogeneous topology base map that can reflect the current operating status of the physical network.

[0066] S2, perform topology change event perception on the physical network corresponding to the heterogeneous topology base map, map the collected topology change events to incremental change units for the heterogeneous topology base map, and determine the set of influence domain nodes corresponding to the incremental change units based on the object category and influence level of the topology change events.

[0067] like Figure 3 As shown in step S21, during the operation of the physical network corresponding to the heterogeneous topology base map, multi-source topology change signals are continuously collected. Specifically, the state changes during data forwarding are monitored in real time, changes in device operating status are continuously received, changes in adjacency relationships in the control plane are tracked, and virtual resource scheduling or instance changes are subscribed to. The multi-source signals are then processed by unified event extraction to generate a set of topology change events. Specifically, for each collected change signal, key elements related to the topology change are extracted, such as the nodes or edges involved, the type of change, and the state before and after the change. These elements are then standardized and organized according to the rules of unified extraction content, unified classification semantics, unified data format, and ensuring time consistency, and converted into a topology change event in a unified format. Finally, all processed topology change events are summarized to form a topology change event set with a consistent structure.

[0068] The acquired multi-source signals are formally represented as follows: ;in, Represents a set of multi-source topology change signals. Indicates the first A change signal, The number of signals is represented by counting all acquired signals.

[0069] The topology change signal includes at least:

[0070] a. Forwarding path change signals are used to reflect changes in the next-hop node in the data plane;

[0071] b. Device status change signals are used to reflect changes in the online / offline status or port status of the device;

[0072] c. Control plane adjacency relationship change signal, used to reflect changes in routing neighbor relationships;

[0073] d. Virtual resource change signals, used to reflect the creation and deletion of containers or virtual instances.

[0074] Each type of signal is analyzed to extract core information reflecting topology changes, such as which node or connection is involved, what changes occurred, and the state before and after the change. The extracted information is then reorganized according to uniformly defined fields. Regardless of whether the signal originates from the data plane, device reporting, or control plane, the same set of fields is used to express this information, such as describing it using the changed object, change type, state before change, and state after change. Next, semantic alignment is performed on changes from different sources but with the same meaning. For example, link disconnection, adjacency loss, and next-hop changes are uniformly grouped into a certain type of topology change to ensure logical consistency. Finally, the processed results are output in a unified structure, making each change a topology change event with a consistent format, and summarizing them into a unified event set. Therefore, signals from different sources are converted into topology change events in a unified format as follows:

[0075] ;in, Represents a set of topology change events. Indicates the first One topology change event, This indicates the number of events, representing the total number of topology change events generated within the processing cycle.

[0076] S22, for each topology change event, read the key information that has been uniformly extracted, including the objects involved in the change, what changes occurred, and the state before and after the change. Then, according to the preset mapping rules, the information is converted into corresponding fields. Specifically, the preset mapping rules are to map the involved objects to the change object identifier, the change type to the change type identifier, and to extract the state before and after the change recorded in the event to form a pair of before and after state descriptions. Then, the above elements are encapsulated according to a fixed structure and combined into a unified data unit. A time stamp is added to each mapped unit to ensure that multiple changes to the same object are traceable and distinguishable in time sequence, so that each event is expressed as a record in the same format. Finally, the above conversion process is performed on all events one by one, thereby transforming the original semantic topology change events into structured incremental change units to describe the changes of nodes or edges in the heterogeneous topology base graph.

[0077] The incremental change unit is defined as follows: ;in, Indicates the first Incremental change unit This represents the identifier of the object being changed, a unique identifier for the corresponding node or edge. Indicates the change type identifier. Indicates the state before the change. This indicates the status after the change.

[0078] The original topology change events are uniformly transformed into standardized data units, which facilitates subsequent unified processing and calculation.

[0079] S23, Based on the characteristics of the change objects in the incremental change unit, the incremental change unit is classified into object categories.

[0080] The classification mapping function is defined as follows: ;in, Represents the classification mapping function, Indicates the first The object category corresponding to each incremental change unit.

[0081] Based on the characteristics of the incremental change units, the following rules are used for classification:

[0082] A change in node existence indicates the addition or deletion of a node, such as the addition or removal of a device.

[0083] Edge topology changes refer to the addition, deletion, or type change of connection relationships, such as the establishment or disconnection of connections, or a change in the type of connection.

[0084] Changes in node attributes refer to changes in the node's characteristic fields, which are usually updates to the node's internal state without altering the network structure.

[0085] Based on different categories, a classification basis is provided for the subsequent determination of the area of ​​influence.

[0086] S24. After completing the classification of incremental change units, the corresponding set of influence domain nodes is determined in the heterogeneous topology base map based on their object category and influence range.

[0087] The influence domain node set is defined as follows: ;in, Indicates the first The set of influence domain nodes corresponding to each incremental change unit This represents the set of nodes in a heterogeneous topological base graph.

[0088] The specific rules for determination are as follows:

[0089] a. When When the existence of a node is changed: ;in, Indicates the node that has changed. Represents a node In the topology diagram The set of neighboring nodes within the jump range, starting from node Start by marking the node as the 0th hop, representing the node itself, and then traverse the nodes. The direct neighbors, all directly connected The node belongs to a 1-hop neighbor, for The 1-hop neighbors are further traversed to obtain the nodes connected to these neighbors, forming 2-hop neighbors. For higher hop numbers, such as K = 3, the same operation is performed on the 2-hop neighbors, finding their neighbors, marking the hop number each time, and continuing to traverse according to the connection relationship of the nodes. When traversing to When jumping, stop searching and get all in The set of nodes within the jump range , This represents the preset hop count parameter, which is set according to the size and frequency of change of the network. For example, a smaller network may only need 1 or 2 hops to cover the impact range, while a large-scale network may need a higher hop count to cover more distant nodes that may be affected. It is usually selected based on the structural complexity and granularity of the network.

[0090] b. When When performing edge topology changes: ;in, This represents the two endpoints of the changed edge.

[0091] c. When When node attributes change: That is, it only includes the nodes themselves that have undergone attribute changes.

[0092] It enables differentiated modeling of the impact range of different types of topology changes, so that subsequent incremental updates only affect the necessary node regions.

[0093] S3. Based on the set of nodes in the influence domain, perform local neighbor sampling and heterogeneous graph neural network aggregation by edge type in the heterogeneous topology base graph to obtain the local update embedding result corresponding to the incremental change unit, and write the local update embedding result back to the heterogeneous topology base graph to form the updated target heterogeneous topology graph.

[0094] S31, from the heterogeneous topology base graph, based on the influence domain node set Extract the corresponding local neighbor information for each node. Each node is matched with its neighboring nodes based on its edge type, forming a subset of neighbors distinguished by edge type. Specifically, this is the set of nodes in the influence domain. It includes nodes in the network that need to be updated due to topology changes; each node... Each node in the topology has a set of neighboring nodes connected to it, and these neighboring nodes are connected to the nodes through different types of edges. Connections; to obtain local neighbor information for each node, it is necessary to find all neighboring nodes directly connected to that node and further classify these neighboring nodes; each node's neighbors are connected to it through different types of edges, such as physical links, virtual tunnels, and route adjacencies. To accurately classify neighbors, it is necessary to classify them according to the type of edge. Match these neighbor nodes; by classifying the neighbors of each node, obtain a subset of neighbors for each node. A combination of subsets formed by different edge types, each subset Includes nodes By edge type The set of connected neighboring nodes.

[0095] The set of neighbor subsets is represented as: ;in, Represents a node The set of neighboring subsets.

[0096] Through the grouping process, neighbors of different edge types are extracted and organized into different subsets for subsequent personalized processing.

[0097] S32, for the set of neighbor subsets For each subset Perform heterogeneous graph neural network aggregation operation, and fuse and update the features of neighboring nodes of each subset through the heterogeneous graph neural network aggregation function.

[0098] The specific steps are as follows:

[0099] a. Neighbor subset aggregation: For each subset The aggregation function in heterogeneous graph neural networks is used. The aggregated representation of the features of neighboring nodes is as follows: ;in, Representing neighboring nodes In graph neural networks, each node, including its neighbors, has a feature vector. This feature vector is extracted from the node's attributes or other data during network initialization. The feature vector can represent various information about the node, such as device status, resource usage, and connection information. Neighbor nodes It is with nodes In the topology graph, for nodes directly connected by edge types such as physical links and virtual tunnels, neighboring nodes are defined and represented through edge connections, thus determining the node's position. The set of neighboring nodes allows us to extract corresponding feature vectors from the neighboring nodes. The feature vectors are stored in the node attributes or feature matrix of the network; Represents a node Subset of neighbors The aggregated feature representation.

[0100] aggregate functions Its main function is to fuse the feature information of a node's neighboring nodes to obtain a new feature representation representing the node and its neighbors. The specific form of the aggregation function is summation aggregation, which adds the feature vectors of all neighboring nodes, specifically expressed as: .

[0101] b. Merge the aggregation results of each subset: Aggregate the results of all neighboring subsets. The merging process yields the final locally updated embedding result, represented as follows: : ;in, This indicates that the average of the various aggregation results is calculated.

[0102] S33, Finally, embed the local update into the result. Write back to the nodes in the heterogeneous topology base graph The feature fields are updated, and the node status is updated, ensuring that the features and status of each node in the target graph can reflect the latest network topology changes.

[0103] The specific steps are as follows:

[0104] Updated embedding results Write node Feature fields;

[0105] Update node states to form the target heterogeneous topology graph. .

[0106] Specifically, each node Each node has an associated feature field, which stores the node's feature vector, such as its state and attributes. In graph neural networks, the feature field of a node is used to represent its internal state or external features, resulting in an updated embedding. It is a new feature vector obtained by aggregating the features of neighboring nodes. It reflects the changes in the relationship between the node and its neighboring nodes. Write back to node In the feature fields, ensure that the node features are up-to-date; assume the node In heterogeneous topology graph There is a feature field that stores the original feature vector. When the network topology changes, the nodes... The feature vectors are updated with the new aggregated features. The data is then stored back in the original feature fields. Besides updating the feature fields, the node's state, such as its activity and running status, also needs to be updated. State updates typically involve changing some additional identifiers or parameters in the node's data structure to reflect the node's latest running status or role in the topology graph. After updating the node's features and state, the target heterogeneous topology graph is obtained. .

[0107] Finally, the updated target heterogeneous topology graph This is the latest topology map that includes incremental changes, reflecting the changes in network status.

[0108] S4. Perform time-stamped version registration and consistency verification on the target heterogeneous topology graph. Generate a topology incremental record by combining the incremental change unit, the local update embedding result, and the corresponding state before and after the change. When the consistency verification passes, output the updated network digital twin topology version. When the consistency verification fails, perform sequential replay repair on the conflicting nodes according to the topology incremental record and then output the updated network digital twin topology version.

[0109] S41, First, timestamp the incremental change units and their corresponding local update embedding results in the target heterogeneous topology graph to generate a topology incremental record with timestamps.

[0110] Specifically, whenever the topology changes, the network management system generates a timestamp for each incremental change unit, such as the addition or deletion of nodes, or topological changes to edges, to record the time of the change. Each incremental change unit will then have a unique timestamp indicating the time of the change event. The timestamp is generated using the current time of the network management system. The network management system obtains the current time when a topology change occurs by calling the network management system clock or the time of network status update, such as when it can be used... The function retrieves the current timestamp of the network management system; each incremental change unit, such as node addition / deletion and edge modification, will be compared with its current timestamp. Association, timestamp It is an identifier used to record the specific time when the change unit occurred; it is used to embed the result of each incremental change unit and its corresponding local update. Generate a topology increment record with a timestamp. The topology increment record not only contains the change information of the increment unit, such as the identifiers of nodes and edges, and the state before and after the change, but also includes a timestamp. This allows for the tracking of changes in chronological order. The purpose of timestamps is to ensure that the order of topology changes is accurately recorded, providing a basis for determining the order of subsequent version updates, historical retrospectives, and incremental updates. Through timestamp marking, the network management system can ensure that each change event can be tracked in chronological order during multiple topology changes and can restore historical states. The network management system is a tool or platform used to monitor, manage, and maintain the hardware, software, communication protocols, and network services of computer networks. It helps network administrators monitor and manage the network's status, performance, security, and faults in real time, ensuring the efficient operation of the network.

[0111] The timestamp mark The time when an incremental change occurs is represented as follows: ;in, Incremental change unit The corresponding timestamp, This indicates the time of change in the current system time or network status.

[0112] The purpose of time stamps is to record the historical sequence of topology changes, providing an accurate timeline for subsequent topology version updates and backtracking.

[0113] S42, perform a consistency check on the topology incremental record to ensure that the incremental change unit and the local update embedding result are consistent with the existing topology version.

[0114] When performing consistency checks, the network management system first obtains the current target heterogeneous topology graph. This is the latest version of the network topology, containing the current state information of all nodes and edges. The network management system then increments the change unit... and local update embedding results The verification process includes: verifying the existence of nodes, the rationality of edge connections, and whether the attributes of nodes or edges meet the expected update criteria. If the incremental change unit and update results are compatible with the existing topology version, the verification passes. The incremental change unit is then written to the topology storage module, and the following updates are performed:

[0115] ;in, This represents the updated target heterogeneous topology graph. Indicates an incremental change unit. This indicates a partial update of the embedded result.

[0116] If the consistency check fails, a conflict resolution operation is performed.

[0117] First, identify the root cause of the conflict, such as a conflict in node states or inconsistent edge connections. Using the state information before and after the change in the topology incremental record, trace back to the last consistent network state to view previous topology versions and understand how the network topology evolved before the conflict occurred. During the process of tracing back the change history, different repair strategies are adopted based on the type of conflict.

[0118] Node conflict resolution: If the node states are inconsistent, select the latest valid node state based on the timestamp and update the node's feature fields;

[0119] Edge conflict repair: If a conflict occurs in the connection relationship of an edge, the connection status of the edge is adjusted according to topological rules, such as prioritizing the shortest path.

[0120] Sequential replay repair: If the conflict still exists, the previous incremental changes are reapplied through sequential replay to gradually adjust the network state and ensure that consistency is eventually restored. The replay process ensures that each part of the topology is updated in a predetermined order by reapplying the change records sequentially, thus preventing the conflict from happening again.

[0121] After conflict resolution and restoration of topology consistency, a consistency check is re-executed to ensure that the repaired topology is completely consistent with the network state. If the consistency check passes, the repair operation is successful, and the target heterogeneous topology is updated. This ensures that the state of all nodes and edges reflects the latest network state.

[0122] Finally, after sequential replay repair, the network management system will output an updated target heterogeneous topology map. At this point, the states of all nodes and edges in the graph have been restored to a consistent state.

[0123] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0124] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for dynamic incremental update of network digital twin topology based on heterogeneous graph neural networks, characterized in that, Includes the following steps: Based on the physical network, a corresponding heterogeneous topology base graph is constructed. According to the preset node and edge type system, the devices and connection relationships in the network are mapped to nodes and edges with type identifiers, and a feature field is written for each node. Obtain the current network digital twin topology; Topology change event perception is performed based on heterogeneous topology base map, and the collected topology change events are mapped into incremental change units, and the change events are classified according to their category and scope of impact. Obtain the corresponding set of influence domain nodes; Based on the influence domain node set grouped by edge type, local neighbor sampling and heterogeneous graph neural network aggregation are performed; Obtain the local update embedding result and write it back to the heterogeneous topology base graph to form the updated target topology graph. Based on the updated target topology graph, perform version registration and consistency verification to obtain topology increment records: If the verification passes, the updated topology version is output. If the verification fails, the conflicting nodes are replayed sequentially to fix the issues before the updated topology version is output.

2. The method for dynamic incremental update of network digital twin topology based on heterogeneous graph neural networks according to claim 1, characterized in that, Collect device entity information and connection relationship information in the physical network to be maintained. Formalize the collected network topology as a combination of node sets and edge sets, defined as: ;in, To represent the set of device entities in a network, To represent the set of connections between device entities, each device entity is tagged with a type according to a preset node type system to form a corresponding node set, and each connection relationship is tagged with a type according to a preset edge type system to form a corresponding edge set. Semantic tagging is then applied to the node set and edge set to obtain: ;in, This represents the overall heterogeneous topology structure in a network digital twin. A collection of node types A set of edge types; A heterogeneous topological framework is constructed based on the node set and edge set, and a set of feature fields matching the type of each node is established. The node feature field set is defined as follows: ;in, Represents the set of node features. Represents a node eigenvectors, Represents a set of nodes For any node in the network, the set of feature fields is configured differently based on the node type.

3. The method for dynamic incremental update of network digital twin topology based on heterogeneous graph neural networks according to claim 2, characterized in that, The real-time status data of each node is acquired and written into its corresponding feature field to form a node feature representation. This representation is then bound to the heterogeneous topology framework to generate a complete heterogeneous topology base graph, which is represented as follows: ;in, This represents the generated heterogeneous topology base map of the network digital twin. Represents the set of node features.

4. The method for dynamic incremental update of network digital twin topology based on heterogeneous graph neural networks according to claim 1, characterized in that, Multi-source topology change signals are collected from the physical network corresponding to the heterogeneous topology base map. The multi-source topology change signals are represented as follows: ;in, Represents a set of multi-source topology change signals. Indicates the first A change signal, The number of signals is indicated by the topology change signals, which include at least forwarding path change signals, device status change signals, control plane adjacency relationship change signals, and virtual resource change signals. A unified event extraction is performed on each type of signal to generate a corresponding topology change event set, which is represented as follows: ;in, Represents a set of topology change events. Indicates the first One topology change event, Indicates the number of events; Each topology change event in the set of topology change events is mapped to an incremental change unit according to a preset rule. Each incremental change unit includes at least a change object identifier, a change type identifier, and state information before and after the change, used to describe the state changes of the corresponding node or edge in the heterogeneous topology base graph. The incremental change unit is represented as follows: ;in, Indicates the first Incremental change unit Indicates the change of object identifier. Indicates the change type identifier. Indicates the state before the change. This indicates the status after the change.

5. The method for dynamic incremental update of network digital twin topology based on heterogeneous graph neural networks according to claim 4, characterized in that, The incremental change units are classified according to the object category of the topology change event. The object category includes at least node existence changes, edge topology changes, and node attribute changes. A corresponding processing method is determined based on different categories, and the classification mapping function is defined as follows: ;in, Represents the classification mapping function, Indicates the first The object category corresponding to each incremental change unit.

6. The method for dynamic incremental update of network digital twin topology based on heterogeneous graph neural networks according to claim 5, characterized in that, Based on the object category and influence range of the incremental change unit, an influence domain node set is determined in the heterogeneous topology base graph. The influence domain node set is represented as follows: ;in, Indicates the first The set of influence domain nodes corresponding to each incremental change unit Let represent the set of nodes in the heterogeneous topological base graph, where; When the incremental change unit is a node existence change, the set of nodes in the affected domain includes the changed node and its neighboring nodes within a preset hop range; When the incremental change unit is an edge topology change, the set of nodes in the affected domain includes the nodes at both ends of the changed edge and its neighboring nodes within a preset hop count range; When the incremental change unit is a node attribute change, the affected domain node set includes the set of nodes whose attributes have changed.

7. The method for dynamic incremental update of network digital twin topology based on heterogeneous graph neural networks according to claim 1, characterized in that, Based on the set of nodes in the influence domain, the corresponding local neighbor information is extracted from the heterogeneous topological base graph, and the neighbors are grouped according to edge type to form a set of neighbor subsets distinguished by edge type, as shown below: ;in, Represents a node The set of neighbor subsets Represents nodes By edge type The set of connected neighboring nodes.

8. The method for dynamic incremental update of network digital twin topology based on heterogeneous graph neural networks according to claim 7, characterized in that, For the set of neighbor subsets Perform heterogeneous graph neural network aggregation operation, where: For each neighbor subset Feature fusion is performed using the aggregation function of a heterogeneous graph neural network, and the local result is represented as follows: : ;in, Representing neighboring nodes eigenvectors, Represents a node Subset of neighbors The aggregated feature representation; The aggregation results of all neighbor subsets are merged to obtain the final local update embedding result, represented as follows: : ;in, This indicates that the aggregation results are being merged. Embed the local update result Write back to the nodes in the heterogeneous topology base graph The feature fields are analyzed, and the node states are updated to form the updated target heterogeneous topology graph. .

9. The method for dynamic incremental update of network digital twin topology based on heterogeneous graph neural networks according to claim 1, characterized in that, The incremental change units and their corresponding local update embedding results in the target heterogeneous topology graph are timestamped to generate time-stamped topology incremental records. The timestamps represent the timestamps that occur during the incremental changes and are used to record the historical order of topology changes. The time when an incremental change occurs is represented as follows: ;in, Incremental change unit The corresponding timestamp, This indicates the time of change in the current system time or network status.

10. The method for dynamic incremental update of network digital twin topology based on heterogeneous graph neural networks according to claim 9, characterized in that, The consistency check is performed on the incremental topology record to check whether the incremental change unit and the local update embedding result are consistent with the existing topology version. If the consistency check passes, the incremental change unit is written into the topology storage module to update the target heterogeneous topology graph. If the consistency check fails, a conflict resolution operation is performed. If the consistency check fails, a sequential replay repair operation is performed based on the state information before and after the change in the topology incremental record. By replaying the change history and adjusting the state of nodes or edges, conflicts are resolved and consistency is restored. Finally, the updated target heterogeneous topology graph is output.

Images