A method, device and electronic equipment for associating physical assets with logical resources

Abstract

This invention discloses a method, apparatus, and electronic device for associating physical assets and logical resources. The method includes: acquiring and standardizing physical asset data and logical resource data; performing business rule verification and consensus notarization through smart contracts to generate digital twins of assets; using a graph neural network model for representation learning, and then calculating the association confidence between physical asset nodes and logical resource nodes; generating association recommendation results and anomaly alarms and outputting them to a user terminal; receiving confirmation instructions for the association recommendation results and correction instructions for the anomaly alarms, triggering smart contracts for anchoring and notarization, and generating notarization records; collecting the confirmation instructions, correction instructions, and notarization records as feedback data, and using the feedback data to incrementally learn or fine-tune the graph neural network model, forming a model optimization closed loop. This invention can effectively solve the operational pain points of discrepancies between physical assets and records, and the lagging and difficult-to-trace nature of association maintenance.

Description

Technical Field

[0001] This invention relates to the field of network resource technology, and in particular to a method, apparatus, and electronic device for associating physical assets with logical resources. Background Technology

[0002] Telecommunications operators face the challenge of managing network infrastructure where physical assets (such as base station equipment and servers) and logical resources (such as network elements and IP addresses) belong to different management systems. Enterprise Resource Planning (ERP) systems manage asset procurement and financial attributes, while Operations Support Systems (OSS) manage resource allocation and operational status; however, there is a lack of an automatic and reliable mapping mechanism between the two. With the development of 5G, cloud-network convergence, and network virtualization technologies, the number and complexity of network elements have increased dramatically, making the accurate correspondence between physical assets and logical resources a key bottleneck affecting operational efficiency, cost control, and decision-making quality. Currently, the correlation relies heavily on manual maintenance or scattered records, resulting in practical problems such as update delays, information inconsistencies, and difficulty in traceability. Summary of the Invention

[0003] The technical problem this invention aims to solve is to address the aforementioned shortcomings of existing technologies by proposing a method, apparatus, and electronic device for associating physical assets and logical resources. This method enables effective association, reliable evidence storage, and closed-loop governance of physical assets and logical resources in telecommunications networks, thereby effectively resolving operational pain points such as discrepancies between physical assets and records, delayed and difficult-to-trace association maintenance.

[0004] In a first aspect, the present invention provides a method for associating physical assets with logical resources, the method comprising:

[0005] Acquire physical asset data and logical resource data;

[0006] Physical asset data is standardized to obtain standardized asset data; and logical resource data is standardized to obtain standardized resource data.

[0007] By deploying smart contracts on a permissioned blockchain, business rules are verified and consensus is notarized for changes in the preset state of standardized asset data, thereby generating digital twins of assets.

[0008] Based on the state change events of asset digital twins, a heterogeneous information graph of telecommunications networks is constructed by combining standardized resource data and network topology relationships; the heterogeneous information graph of telecommunications networks includes various types of nodes and predefined telecommunications engineering semantic relationships;

[0009] A graph neural network model is used to learn the representations of nodes in a heterogeneous information graph of a telecommunications network, and the association confidence between physical asset nodes and logical resource nodes is calculated based on the learned node representation vectors.

[0010] Generate association recommendation results based on association confidence, and generate anomaly alarms based on the inconsistency between the state of the asset digital twin and the state of standardized resource data, and output the association recommendation results and anomaly alarms to the user terminal;

[0011] Receive confirmation instructions for related recommendation results and correction instructions for abnormal alarms, trigger the smart contract to anchor and store the association between physical assets and logical resources, and generate storage records;

[0012] Confirmation instructions, correction instructions, and evidence records are collected as feedback data, and the feedback data is used to incrementally learn or fine-tune the graph neural network model to form a closed loop of model optimization, thereby realizing the association between physical assets and logical resources.

[0013] Furthermore, smart contracts deployed on the permissioned blockchain are used to verify business rules and establish consensus for the state changes of standardized asset data, specifically including:

[0014] Invoke the first smart contract deployed on the permissioned blockchain, and determine whether the status change request of standardized asset data meets the preset business rules based on the first smart contract;

[0015] If the status change request meets the preset business rules, then the type of the status change request is further identified;

[0016] When the state change request is an asset scrapping request, the second smart contract deployed on the permissioned chain is invoked to perform association conflict verification.

[0017] The state change request that passes the correlation conflict verification is submitted to the permissioned chain consensus node as a legitimate transaction for consensus confirmation. After consensus is reached, the state change is recorded in the distributed ledger, and the digital twin state of the standardized asset data in the first smart contract is updated synchronously.

[0018] The first smart contract is an asset digital twin management contract, which is used to define a finite state machine for standardized asset data and execute business rule verification; the second smart contract is an association anchoring contract, which is used to store the digital fingerprints and related metadata of the confirmed asset and resource association relationships.

[0019] Furthermore, a graph neural network model is used to learn representations of nodes in the heterogeneous information graph of the telecommunications network, specifically including:

[0020] A graph neural network model is built using a relational graph convolutional network;

[0021] The self-loop weight matrix of the graph neural network model is used to transform the self-loop features of the target node to obtain the self-loop features.

[0022] According to the predefined telecommunications engineering semantic relationship type, the target node's neighbor nodes are divided into multiple neighbor node sets;

[0023] For any set of neighboring nodes, the node features of each neighboring node in the set are transformed and aggregated using an independent weight matrix that matches the semantic relationship type of telecommunications engineering, so as to obtain the neighbor aggregation features corresponding to the semantic relationship type of telecommunications engineering.

[0024] Summarize the neighbor aggregation features to obtain the total neighbor aggregation features;

[0025] The self-loop features are fused with the total neighbor aggregation features, and then processed by a non-linear activation function to obtain the updated node representation vector.

[0026] Furthermore, upon receiving confirmation instructions for the associated recommendation results and correction instructions for abnormal alarms, the smart contract is triggered to anchor and store the association between physical assets and logical resources, generating a storage record, specifically including:

[0027] Receive and parse confirmation and correction instructions to obtain the mapping relationship between physical assets and logical resources, as well as status correction information;

[0028] Trigger the smart contract to generate a state update proposal, which includes the current state of the asset digital twin, the target state, and the corrected state of standardized resource data;

[0029] Consensus verification is performed on the state update proposal, and after consensus is passed, the mapping relationship and the state update proposal are jointly hashed to obtain the target hash value;

[0030] The smart contract is invoked to anchor and store the target hash value on the permissioned blockchain;

[0031] Based on the evidence storage results, the status identifier of the digital twin of the asset is updated to the associated calibration status, and an evidence storage record containing the on-chain transaction hash, timestamp, and status change log is generated.

[0032] Furthermore, based on the learned node representation vectors, the association confidence between physical asset nodes and logical resource nodes is calculated, specifically including:

[0033] Extract the first feature vector and the second feature vector of the physical asset node and the logical resource node respectively from the output layer of the graph neural network model;

[0034] Calculate the cosine similarity between the first feature vector of a node and the second feature vector of a node, and use the value of the cosine similarity as the association confidence.

[0035] Furthermore, generate anomaly alerts based on inconsistencies between the state of the asset digital twin and the state of standardized resource data, specifically including:

[0036] Compare the real-time status of standardized resource data with the real-time status of asset digital twins;

[0037] If the real-time status of the asset digital twin is inconsistent with the real-time status of the standardized resource data, or if there is a logical conflict in the relationship, an anomaly alarm will be generated.

[0038] Furthermore, abnormal alarms specifically include: alarms for unassociated resources, alarms for assets with no listings, or alarms for associated conflicts;

[0039] Unassociated resource alerts indicate isolated logical resources not carried by the asset digital twin; Empty asset alerts indicate empty asset digital twins that do not carry valid logical resources; Association conflict alerts indicate logical contradictions in the association relationship between assets and resources.

[0040] Furthermore, the graph neural network model is incrementally learned or fine-tuned using feedback data, including:

[0041] The physical assets and logical resources associated with the user's confirmation command are marked as positive samples, and the correction commands entered by the user are marked as negative samples.

[0042] Positive and negative samples, along with their corresponding node features, are stored in the training sample library for accumulation.

[0043] When the preset triggering conditions are met, a target loss function is constructed based on the positive and negative samples accumulated in the training sample library. The target loss function aims to maximize the cosine similarity of positive sample pairs in the representation space and minimize the cosine similarity of negative sample pairs.

[0044] Using the parameters of the current graph neural network model as initial weights, a fine-tuned learning rate smaller than the initial training learning rate is adopted. The graph neural network model is updated by backpropagation based on the target loss function to obtain an optimized graph neural network model and replace the current model.

[0045] The preset triggering conditions include: the number of new samples reaching a preset threshold, or reaching a preset periodic time interval.

[0046] In a second aspect, the present invention provides an apparatus for associating physical assets and logical resources, the apparatus comprising:

[0047] The acquisition unit is used to acquire physical asset data and logical resource data.

[0048] The standardization unit, connected to the acquisition unit, is used to standardize physical asset data to obtain standardized asset data; and to standardize logical resource data to obtain standardized resource data.

[0049] The generation unit, connected to the standardization unit, is used to perform business rule verification and consensus notarization on the preset state changes of standardized asset data through smart contracts deployed on the permissioned blockchain, thereby generating digital twins of assets.

[0050] The construction unit, connected to the generation unit, is used to construct a heterogeneous information graph of the telecommunications network based on the state change events of the asset digital twin, combined with standardized resource data and network topology relationships; the heterogeneous information graph of the telecommunications network contains various types of nodes and predefined telecommunications engineering semantic relationships;

[0051] The first processing unit, connected to the construction unit, is used to perform representation learning on nodes in the heterogeneous information graph of the telecommunications network using a graph neural network model, and to calculate the association confidence between physical asset nodes and logical resource nodes based on the learned node representation vectors.

[0052] The output unit, connected to the first processing unit, is used to generate association recommendation results based on association confidence, and to generate anomaly alarms based on the inconsistency between the state of the asset digital twin and the state of standardized resource data, and to output the association recommendation results and anomaly alarms to the user terminal.

[0053] The receiving unit, connected to the output unit, is used to receive confirmation instructions for the associated recommendation results and correction instructions for abnormal alarms, triggering the smart contract to anchor and store the association between physical assets and logical resources, and generate storage records.

[0054] The second processing unit, connected to the receiving unit, is used to collect confirmation instructions, correction instructions, and evidence records as feedback data, and to use the feedback data to perform incremental learning or fine-tuning of the graph neural network model to form a model optimization closed loop, thereby realizing the association between physical assets and logical resources.

[0055] Thirdly, the present invention provides an electronic device including a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the method for associating physical assets and logical resources according to the first aspect.

[0056] This invention standardizes cross-domain data and constructs a heterogeneous information graph based on network topology. It utilizes graph neural networks for automatic reasoning, introduces permissioned blockchain smart contracts for state verification and anchored evidence storage, and drives model iteration based on human interaction commands. This allows for the automatic extraction of asset-resource mapping relationships from massive amounts of complex network data, ensuring the absolute credibility and full traceability of the correlation benchmark, and enabling the correlation discovery capability to continuously evolve in operational practice. Specific beneficial effects are as follows:

[0057] 1. By constructing a heterogeneous information graph and using graph neural network inference, this method achieves automatic discovery of relationships, eliminating reliance on manual maintenance. This approach combines standardized data with network topology to construct a heterogeneous information graph containing various nodes and predefined telecommunications engineering semantic relationships. A graph neural network model is then used to learn node representation vectors and calculate association confidence. This method can deeply understand complex network topology semantics, automatically and accurately uncovering potential relationships between physical assets and logical resources, effectively solving the inefficiencies and omissions caused by traditional methods that rely on manual maintenance and static rules.

[0058] 2. By generating digital twins of assets and comparing their states, proactive anomaly detection is achieved, eliminating discrepancies between physical and recorded data. This method uses smart contracts to generate digital twins of assets as a state benchmark, monitors state change events in real time, and compares them with standardized resource data. Once an inconsistency between the physical and logical states is detected, an anomaly alarm is immediately generated. This mechanism breaks down the latency barrier of data synchronization, transforming reactive post-event verification into proactive real-time detection, effectively solving the pain points of cross-domain data information contradictions and update delays.

[0059] 3. By verifying rules and anchoring evidence through smart contracts, a trusted benchmark for relationships is established, enabling precise traceability throughout the entire process. This method uses smart contracts deployed on a permissioned blockchain for business rule verification and consensus-based evidence storage at both the state change and final relationship establishment stages. Due to the immutable nature of blockchain, the generation and modification of any relationship carries a clear timestamp and logical constraints, providing an absolutely reliable data foundation for cross-departmental asset and resource mapping, completely solving the practical problems of scattered and difficult-to-trace relationship records.

[0060] 4. By collecting interactive feedback to drive incremental model learning, a closed-loop governance mechanism is formed, enabling continuous self-optimization of association capabilities. This method does not terminate the process after outputting recommendation results and alarms, but further receives user confirmation and correction instructions, which, along with the stored evidence, are used as feedback data directly for the incremental learning or fine-tuning of the graph neural network model. This design streamlines the process from discovery to processing and optimization, allowing the method to continuously absorb real-world operational experience to correct its own deviations, effectively solving the problems of traditional association maintenance being lagging and unable to self-improve.

[0061] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0062] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. The above and other features and advantages will become more apparent to those skilled in the art from the detailed description of exemplary embodiments with reference to the accompanying drawings, in which:

[0063] Figure 1 A schematic diagram illustrating the method for associating physical assets and logical resources as provided in an embodiment of the present invention;

[0064] Figure 2 A diagram illustrating the association between physical assets and logical resources provided in this embodiment of the invention;

[0065] Figure 3 A sequence diagram of the associated business process between physical assets and logical resources provided for embodiments of the present invention;

[0066] Figure 4 A schematic diagram of a device for associating physical assets and logical resources provided in an embodiment of the present invention;

[0067] Figure 5 A framework diagram of an electronic device provided in an embodiment of the present invention.

[0068] Figure descriptions: 10, Acquisition unit; 20, Standardization unit; 30, Generation unit; 40, Construction unit; 50, First processing unit; 60, Output unit; 70, Receiving unit; 80, Second processing unit; 100, Processor; 200, Memory. Detailed Implementation

[0069] It is understood that the specific embodiments and accompanying drawings described herein are merely for explaining the invention and are not intended to limit the invention.

[0070] It is understood that, without conflict, the various embodiments and features in the embodiments of the present invention can be combined with each other.

[0071] It is understood that, for ease of description, only the parts related to the present invention are shown in the accompanying drawings, while the parts unrelated to the present invention are not shown in the drawings.

[0072] It is understood that each unit or module involved in the embodiments of the present invention may correspond to only one entity structure, or may be composed of multiple entity structures, or multiple units or modules may be integrated into one entity structure.

[0073] It is understood that, without conflict, the functions and steps marked in the flowcharts and block diagrams of this invention may occur in a different order than that marked in the accompanying drawings.

[0074] It is understood that the flowcharts and block diagrams of this invention illustrate the possible architecture, functions, and operations of systems, apparatuses, devices, and methods according to various embodiments of this invention. Each block in the flowchart or block diagram may represent a unit, module, program segment, or code, containing executable instructions for implementing the specified function. Furthermore, each block or combination of blocks in the block diagram and flowchart can be implemented using a hardware-based system to achieve the specified function, or using a combination of hardware and computer instructions.

[0075] It is understood that the units and modules involved in the embodiments of the present invention can be implemented by software or by hardware. For example, the units and modules can be located in a processor.

[0076] Example 1:

[0077] The inventors discovered that while IoT-based asset tracking and visualization technologies (such as RFID, sensors, and digital twin solutions) can improve the location and status monitoring capabilities of individual physical entities, their scope is limited to the physical layer. Faced with complex scenarios such as the dynamic mapping between virtualized network elements and underlying physical servers in 5G core networks, and the elastic scheduling of container instances and host machines in cloud resource pools, they cannot automatically discover the attribution and connection relationships between physical carriers and logical functions, and are even less able to cope with the dynamic changes in relationships brought about by the migration of virtual resources across physical devices. Decision optimization technologies based on complex system modeling and analysis (such as causal knowledge graphs and graph neural network analysis) possess powerful relational reasoning and optimization decision-making capabilities, their effective operation presupposes that the basic relationships between entities have been accurately defined. Essentially, they perform high-order deductions based on "known relationships," rather than solving the more fundamental data governance problem of "how to establish relationships." When the underlying relational data itself is missing, erroneous, or has version conflicts, the upper-level analysis only amplifies the bias. The application of blockchain technology in the asset field is mostly focused on external circulation scenarios such as supply chain finance and asset tokenization. The core goal is to achieve credible evidence of ownership transfer and market transaction matching. However, the core requirements of the internal network operation and maintenance scenario of operators are completely different. It is necessary to use the immutability and consensus mechanism of blockchain to build a credible benchmark for the core status and relationship of assets across ERP, OSS and professional network management systems, and solve the consistency and traceability problems of internal multi-source data. The existing external circulation-oriented blockchain application model cannot directly adapt to the internal governance requirements of access control, high-frequency updates and privacy protection.

[0078] The aforementioned technical limitations manifest as three major systemic defects in telecommunications network scenarios: First, the intelligence level of the correlation discovery process is insufficient. Manually configured rules are difficult to cover the complex patterns of massive heterogeneous data, and static mapping cannot adapt to the elastic changes of cloud network resources, resulting in a large number of potential correlations being hidden in data silos. Second, the data consistency guarantee mechanism is weak. Data synchronization between ERP and OSS relies on timed batch processing or interface docking, which has inherent delays and error accumulation. The lack of a single reliable source to prevent tampering makes "who should be the standard" a point of contention in cross-departmental collaboration. Third, the management process is fragmented. Correlation maintenance, anomaly handling, and knowledge accumulation are disconnected from each other, making it impossible to continuously learn and optimize from operational feedback. The same type of correlation error recurs and is difficult to eradicate.

[0079] This embodiment addresses the aforementioned technological gaps by providing a systematic and closed-loop solution for associating physical assets with logical resources. Its typical application scenarios span the entire lifecycle governance of telecommunications operator network infrastructure. During the 5G network construction phase, facing the parallel operations of warehousing massive physical assets such as base station main equipment, antenna feeder systems, and power supply, as well as resource activation such as BBU / AAU logical configuration, cell parameters, and IP address allocation, the traditional model often involves project acceptance and resource entry belonging to different departments and systems. This frequently results in discrepancies between physical equipment being powered on and running but ERP asset cards not yet created, or logical resources being allocated but on-site physical labels being incorrectly or missingly affixed. This embodiment deploys an intelligent association engine to automatically collect physical attributes such as equipment serial numbers, equipment room locations, and port connections. It then interfaces with network element models and configuration data in OSS and uses graph neural networks to learn the implicit association patterns between physical features and logical configurations in historical project data. This enables automatic matching and binding of assets and resources for newly built sites. Simultaneously, the association results are uploaded to the blockchain in real time for evidence storage, ensuring data consistency throughout the entire process from physical acceptance and logical activation to financial accounting. This eradicates asset loss and depreciation accrual distortion caused by using the equipment first and then adding new data. During the cloud-network convergence operation and maintenance period, virtual machines and container instances in the virtualization resource pool frequently migrate across physical servers, and NFV network functions dynamically scale up and down on demand, rendering traditional static association tables completely ineffective. This embodiment continuously monitors the event stream and network traffic characteristics of the virtualization platform, and combines contextual information such as the resource load and network topology of the physical server to infer the dynamic mapping relationship between virtual network elements and the host physical machine in real time. Whenever a migration event is detected, the association relationship is updated, and the change record is written to the blockchain to form a traceable association change history. This supports the rapid location of physical carriers when a fault occurs, the accurate tracing of attack paths during security incidents, and the accurate assessment of hardware lifecycle during capacity planning. During the asset decommissioning and disposal period, facing scenarios such as the decommissioning of massive 4G devices and the removal of old servers due to 5G upgrades, the traditional model suffers from a disconnect between physical dismantling and system write-off, leading to long-term discrepancies between accounts and actual assets and posing a risk of asset loss. This embodiment establishes an on-chain linkage mechanism for physical status changes, logical resource release, and financial asset scrapping. When the intelligent association engine detects decommissioning characteristics such as physical device power-off and zero port traffic, it automatically triggers the verification of association failure. After confirmation by maintenance personnel, the OSS resource status and ERP asset status are updated synchronously, and the decommissioning decision basis and disposal process are recorded on the chain, realizing closed-loop governance and audit traceability of the entire asset lifecycle.In cross-domain collaborative optimization scenarios, facing complex tasks requiring multi-departmental collaboration such as network fault location, customer complaint handling, and network cutover scheduling, the trusted associated data foundation provided in this embodiment breaks down the data barriers of ERP, OSS, and various professional network management systems. The intelligent association engine quickly constructs a complete view of customer services, logical channels, and physical paths based on real-time topology reasoning. Blockchain evidence storage ensures that various departments collaborate on the same version of data. The experience and knowledge gained during anomaly handling, such as identification rules for specific vendor equipment and association patterns in special networking scenarios, are continuously optimized through feedback learning to form a self-optimizing closed loop of data trustworthiness, intelligent discovery, collaborative handling, and knowledge evolution. This fundamentally improves the automation level and trustworthiness assurance capabilities of telecommunications operators' network asset and resource governance.

[0080] like Figure 1 As shown, the method for associating physical assets and logical resources in this embodiment specifically includes steps S1 to S8.

[0081] Step S1: Obtain physical asset data and logical resource data.

[0082] As the initial step in building a digital twin foundation, this step specifically covers physical asset data that accurately reflects the state of the underlying hardware and logical resource data that reflects the virtualization orchestration state of the upper-layer network. The physical asset data meticulously depicts static entity attributes such as the distribution of data center space, physical location of racks, equipment board models, port connection relationships, and unique serial numbers, while the logical resource data accurately describes dynamic configuration parameters such as IP address pool allocation, VLAN segment division, routing protocol instances, VPN bearer channels, and computing and storage resource quotas. By deeply extracting and integrating these two types of data that cross the physical and logical boundaries, data support is provided for achieving accurate mapping between real facilities and virtual models.

[0083] Step S2: Standardize the physical asset data to obtain standardized asset data; and standardize the logical resource data to obtain standardized resource data.

[0084] This step specifically includes: conducting in-depth standardization transformation of the acquired multi-source heterogeneous basic information to completely eliminate data barriers. For physical asset data, this involves transforming it into highly consistent standardized asset data by unifying naming conventions, converting data formats, removing redundant records, and supplementing missing entity attribute information. At the same time, for logical resource data, this involves refining it into standardized resource data with a clear structure and easy interaction by relying on operations such as format normalization, semantic alignment, and cleaning up invalid configuration items. This ensures that both physical space and logical dimensions follow a unified measurement scale and standardized description framework at the data expression level.

[0085] Step S3: Through smart contracts deployed on the permissioned blockchain, business rules are verified and consensus is notarized for the preset state changes of standardized asset data, generating a digital twin of the asset.

[0086] As a specific implementation method, smart contracts deployed on a permissioned blockchain are used to perform business rule verification and consensus notarization of preset state changes of standardized asset data, specifically including:

[0087] Invoke the first smart contract deployed on the permissioned blockchain, and determine whether the status change request of standardized asset data meets the preset business rules based on the first smart contract;

[0088] If the status change request meets the preset business rules, then the type of the status change request is further identified;

[0089] When the state change request is an asset scrapping request, the second smart contract deployed on the permissioned chain is invoked to perform association conflict verification.

[0090] The state change request that passes the correlation conflict verification is submitted to the permissioned chain consensus node as a legitimate transaction for consensus confirmation. After consensus is reached, the state change is recorded in the distributed ledger, and the digital twin state of the standardized asset data in the first smart contract is updated synchronously.

[0091] The first smart contract is an asset digital twin management contract, which is used to define a finite state machine for standardized asset data and execute business rule verification; the second smart contract is an association anchoring contract, which is used to store the digital fingerprints and related metadata of the confirmed asset and resource association relationships.

[0092] Step S4: Based on the state change events of the asset digital twin, construct a heterogeneous information graph of the telecommunications network by combining standardized resource data and network topology relationships; wherein, the heterogeneous information graph of the telecommunications network contains various types of nodes and predefined telecommunications engineering semantic relationships.

[0093] Specifically, based on the state change events of digital twin assets, a heterogeneous information graph of the telecommunications network is constructed by combining standardized resource data and network topology relationships. The aim is to dynamically fuse and generate a graph containing multi-dimensional relationships of events, physical and logical connections, capable of real-time mapping of fault propagation and state evolution. First, using standardized resource data that has undergone time-series alignment, anomaly cleaning, and feature derivation processing as a foundation, a basic skeleton containing multiple types of nodes, including physical network elements and logical networks, is built in the graph database. Multi-layered topology relationships, including physical layer fiber optic connections, link layer Ethernet links, network layer IP routing adjacencies, service layer service bearers, and logical layer resource ownership, are extracted from network resource management as the connecting network. Simultaneously, state change events of the digital twin are collected in real-time and standardized for parsing, accurately extracting event identifiers, timestamps, event types, associated assets, and state parameters before and after the change. These are then instantiated after spatiotemporal mapping with resource data. Event nodes are mounted into the graph, and the host node's status attributes are updated synchronously by establishing triggering association edges. Then, the graph is traversed and causal reasoning is carried out using preset multi-layer topology relationships, and cross-level influence edges are automatically derived along physical links and logical carrying paths. Finally, with asset nodes, event nodes, resource indicator nodes, and topology connection nodes as entities, and twin mapping edges, resource description edges, topology connection edges, event causal edges, and time association edges as links, a complete graph integrating network topology, resource status time sequence, and event causal relationships is constructed. It fully supports its real-time query and historical backtracking needs by relying on graph database storage and partitioned indexing strategies.

[0094] Step S5: Use a graph neural network model to learn the representations of nodes in the heterogeneous information graph of the telecommunications network, and calculate the association confidence between physical asset nodes and logical resource nodes based on the learned node representation vectors.

[0095] As a specific implementation method, a graph neural network model is used to learn the representations of nodes in a heterogeneous information graph of a telecommunications network, specifically including:

[0096] A graph neural network model is built using a relational graph convolutional network;

[0097] The self-loop weight matrix of the graph neural network model is used to transform the self-loop features of the target node to obtain the self-loop features.

[0098] According to the predefined telecommunications engineering semantic relationship type, the target node's neighbor nodes are divided into multiple neighbor node sets;

[0099] For any set of neighboring nodes, the node features of each neighboring node in the set are transformed and aggregated using an independent weight matrix that matches the semantic relationship type of telecommunications engineering, so as to obtain the neighbor aggregation features corresponding to the semantic relationship type of telecommunications engineering.

[0100] Summarize the neighbor aggregation features to obtain the total neighbor aggregation features;

[0101] The self-loop features are fused with the total neighbor aggregation features, and then processed by a non-linear activation function to obtain the updated node representation vector.

[0102] As a specific implementation method, based on the learned node representation vectors, the association confidence between physical asset nodes and logical resource nodes is calculated, specifically including:

[0103] Extract the first feature vector and the second feature vector of the physical asset node and the logical resource node respectively from the output layer of the graph neural network model;

[0104] Calculate the cosine similarity between the first feature vector of a node and the second feature vector of a node, and use the value of the cosine similarity as the association confidence.

[0105] Step S6: Generate association recommendation results based on association confidence, and generate anomaly alarms based on the inconsistency between the state of the asset digital twin and the state of standardized resource data, and output the association recommendation results and anomaly alarms to the user terminal.

[0106] As a specific implementation method, an anomaly alert is generated based on the inconsistency between the state of the asset digital twin and the state of standardized resource data, specifically including:

[0107] Compare the real-time status of standardized resource data with the real-time status of asset digital twins;

[0108] If the real-time status of the asset digital twin is inconsistent with the status of the standardized resource data or if there is a logical conflict in the relationship, an anomaly alarm will be generated.

[0109] As a more specific implementation method, abnormal alarms specifically include: alarms for unassociated resources, alarms for assets with no listings, or alarms for associated conflicts;

[0110] Unassociated resource alerts are used to indicate isolated logical resources that are not carried by any asset digital twin. The conditions for their generation are: the status of standardized resource data indicates the existence of logical resources that are not carried by any asset digital twin, and the logical resources continuously generate traffic or alert data within a preset time window.

[0111] The asset idle alarm is used to indicate an idle asset digital twin that does not carry valid logical resources. The conditions for its generation are: the status of the asset digital twin is in use or standby, and the status of the standardized resource data indicates that no logical resource in operation corresponding to the asset digital twin is found.

[0112] Association conflict alerts are used to indicate that there is a logical contradiction in the association between assets and resources. The conditions for generating an alert are: the association recommendation result is logically mutually exclusive with the existing association relationship, or the same logical resource is recommended to establish associations with multiple physical assets that reach a preset confidence threshold.

[0113] Step S7: Receive confirmation instructions for the associated recommendation results and correction instructions for abnormal alarms, trigger the smart contract to anchor and store the association between physical assets and logical resources, and generate storage records.

[0114] As a specific implementation method, the system receives confirmation instructions for related recommendation results and correction instructions for abnormal alarms, triggering a smart contract to anchor and store the association between physical assets and logical resources, generating a storage record. Specifically, this includes:

[0115] Receive and parse confirmation and correction instructions to obtain the mapping relationship between physical assets and logical resources, as well as status correction information;

[0116] Trigger the smart contract to generate a state update proposal, which includes the current state of the asset digital twin, the target state, and the corrected state of standardized resource data;

[0117] Consensus verification is performed on the state update proposal, and after consensus is passed, the mapping relationship and the state update proposal are jointly hashed to obtain the target hash value;

[0118] The smart contract is invoked to anchor and store the target hash value on the permissioned blockchain;

[0119] Based on the evidence storage results, the status identifier of the digital twin of the asset is updated to the associated calibration status, and an evidence storage record containing the on-chain transaction hash, timestamp, and status change log is generated.

[0120] Step S8: Collect confirmation instructions, correction instructions, and evidence records as feedback data, and use the feedback data to incrementally learn or fine-tune the graph neural network model to form a model optimization closed loop, thereby realizing the association between physical assets and logical resources.

[0121] As a specific implementation method, incremental learning or fine-tuning of the graph neural network model using feedback data includes:

[0122] The physical asset and logical resource association pairs corresponding to the user's confirmation instruction are marked as positive samples, and the association pairs corresponding to the user's correction instruction or rejection instruction for the association recommendation are marked as negative samples.

[0123] Positive and negative samples, along with their corresponding node features, are stored in the training sample library for accumulation.

[0124] When the preset triggering conditions are met, a target loss function is constructed based on the positive and negative samples accumulated in the training sample library. The target loss function aims to maximize the cosine similarity of positive sample pairs in the representation space and minimize the cosine similarity of negative sample pairs.

[0125] Using the parameters of the current graph neural network model as initial weights, a fine-tuned learning rate smaller than the initial training learning rate is adopted. The graph neural network model is updated by backpropagation based on the target loss function to obtain an optimized graph neural network model and replace the current model.

[0126] The preset triggering conditions include: the number of new samples reaching a preset threshold, or reaching a preset periodic time interval.

[0127] In the association determination stage, this embodiment introduces a configurable confidence threshold mechanism to achieve accurate association decisions. Specifically, a preset judgment threshold θ is set. This threshold is typically chosen based on the accuracy requirements and fault tolerance of the actual application scenario, generally selecting 0.75 or 0.85 as the initial baseline value. When the graph neural network model outputs an association probability, it is determined that there is a credible association between asset node a and resource node b, triggering the subsequent association confirmation and on-chain anchoring process; otherwise, it is determined that there is no association between the two, and they are excluded or transferred to the observation queue. It is worth noting that this threshold is not fixed, but supports dynamic adjustment and optimization based on the long-term operating effect of the system—when the association false positive rate is too high in actual application, the threshold can be appropriately increased to improve accuracy; when association false negatives lead to insufficient data coverage, the threshold can be appropriately decreased to enhance recall capability, thereby achieving an adaptive balance between association discovery accuracy and coverage.

[0128] This embodiment proposes a method for associating physical assets with logical resources, which is an intelligent management mechanism based on blockchain and graph neural networks. The scheme establishes the core design concept of "trusted data driving intelligent discovery, and intelligent discovery results feeding back into a trusted closed loop," aiming to fundamentally solve the complex governance challenges of associating network assets and logical resources with operators. The following section, with reference to the accompanying drawings, will elaborate on the overall architecture, core processing flow, key algorithms, and specific implementation.

[0129] 1. Four-layer, two-ring collaborative architecture

[0130] This embodiment is based on a "four-layer, two-ring" collaborative architecture (such as...). Figure 2 As shown): The system comprises a data acquisition and preprocessing layer, a blockchain trusted data layer, a GNN intelligent analysis layer, and an application service layer, including a data consistency closed loop and a model optimization closed loop. Through clearly defined layered responsibilities and data flow, closed-loop governance is achieved from data trustworthiness and intelligent analysis to decision execution.

[0131] 1) Data Acquisition and Preprocessing Layer (Layer 1): Serving as the external data interface, this layer collects raw data from multiple systems through configurable adapters (ERP adapter, OSS adapter, network management adapter, IoT adapter). After processing through standardized pipelines (cleaning, transformation, normalization, and encapsulation), it outputs a data stream in a unified format. Key Design: For event data from physical operations (such as device barcode scanning), this layer provides a standardized data encapsulation format to prepare for on-chain processing.

[0132] 2) Blockchain Trusted Data Layer (Second Layer): A trust foundation built on a permissioned blockchain, with key departments within the operator (such as network, finance, and materials departments) serving as consensus nodes. Two types of smart contracts are deployed at its core:

[0133] The asset digital twin management contract defines a finite state machine for physical assets (e.g., inactive, in use, standby, under maintenance, scrapped). Any request to change the asset's state must invoke the corresponding function in this contract. The contract verifies business rules (e.g., confirming no active business associations before scrapping an asset) before submitting the state change as a transaction to consensus and recording it in an immutable distributed ledger. This codifies management rules, enforces cross-departmental data consistency.

[0134] Linkage Anchoring Contract: This contract provides a public association registry function. When the association between an asset and a resource is discovered intelligently by the system or confirmed manually, the anchoring function of this contract is called to permanently store the hash values ​​of the asset's digital twin ID and the resource's unique ID, along with metadata (timestamp, operator). This provides a globally verifiable and non-repudiable "birth certificate" for all associations.

[0135] 3) GNN Intelligent Analysis Layer (Third Layer): The system's intelligent engine. This layer listens for asset state change events emitted by the blockchain layer and, combined with near real-time resource state and configuration snapshots obtained from OSS, dynamically constructs and maintains a heterogeneous information graph of the telecommunications network. This graph serves as the carrier of domain knowledge.

[0136] Node types include physical devices (such as base stations and servers), device components (such as boards and ports), logical resources (such as network elements and IP addresses), and spatial locations (such as computer rooms and racks).

[0137] Edge relationship types: Predefined relationships in telecommunications engineering semantics, such as hosted_on, physically connected_to, logically bound_to, located_in, etc.

[0138] Node features: Each node has a feature vector that contains not only common attributes (ID, name) but also deeply embedded domain features, such as encoding "manufacturer model standard code" for device nodes and "highest supported protocol type" for port nodes.

[0139] The graph neural network model deployed at this layer (preferably the relational graph convolutional network R-GCN) continuously learns from this heterogeneous information graph. Through a message-passing mechanism, the model allows node features to propagate and aggregate on a predefined relational network, ultimately enabling each node to obtain a representation vector containing its network context. Based on this, the system can calculate the similarity between the representations of any two nodes in the graph (such as an asset node and a resource node), serving as a confidence score for their association.

[0140] 4) Application Service Layer: A unified user interface for operations, management, and auditing personnel. The core functions of this layer include:

[0141] Panoramic Relationship View: Visualizes the dynamic relationship graph output by the GNN layer.

[0142] Intelligent Recommendation and Alarm Console: Lists high-confidence unconfirmed associations, as well as data inconsistency alarms detected by the anomaly detection module (such as "Asset is offline, but associated resources are still active").

[0143] Closed-loop work order engine: Automatically converts alarms or recommendations into standardized work orders, which are then routed to the relevant responsible persons and their status is tracked.

[0144] Open API Gateway: Provides standardized data and service interfaces for upper-layer applications such as financial analysis and resource optimization to call.

[0145] Dual closed-loop drive mechanism:

[0146] 1) Inner Loop: The data consistency closed loop, acting as the inner loop mechanism, begins its operation with actual changes in the physical world or the occurrence of internal system events, triggering data collection and preprocessing. Subsequently, this embodiment calls the blockchain smart contract to complete the on-chain consensus operation of the state. Based on this, on-chain events activate the graph neural network layer, prompting it to update the heterogeneous information graph and perform anomaly detection. Once an anomaly is identified, it is pushed to the application layer to generate a corresponding handling order. The new data generated after on-site personnel complete the handling action will flow back into the preprocessing process, thus forming a complete loop. The core purpose of this closed loop design is to ensure that the system data and the physical reality and logical state of the network always maintain a high degree of strong consistency.

[0147] 2) Outer Loop: The model optimization closed loop serves as the outer loop mechanism. Its operation begins with the association recommendation results output by the graph neural network model. After receiving this result at the application layer, operations personnel confirm or reject it, providing clear feedback. Next, feedback data containing positive and negative examples is collected and stored in the training sample library. In this embodiment, these samples periodically trigger the model's incremental learning or fine-tuning process. The iteratively optimized model can generate more accurate recommendation results in subsequent analysis stages. This closed-loop mechanism ensures that the system's intelligence level can continuously evolve with actual application, completely overcoming the inherent limitations of traditional static models.

[0148] This embodiment adopts a four-layer vertical hierarchical structure. From top to bottom, it consists of: an application interaction layer containing visualization, a governance workbench, and a management configuration center; an intelligent analysis layer containing a correlation analysis engine, a model training module, and an anomaly detection module; a blockchain service layer containing smart contracts, consensus nodes, and a distributed ledger; and a data awareness layer containing physical asset acquisition, logical resource acquisition, and data preprocessing. Each component within each layer collaborates horizontally to complete specific functions. Between layers, bidirectional data flow is achieved through the main data flow direction and feedback control flow, forming a closed-loop mechanism. This mechanism manifests in two core closed loops. The first is the data consistency inner loop, where the data awareness layer collects raw data from physical assets and logical resources, which is then stored on the blockchain by the blockchain service layer. Next, the intelligent analysis layer performs correlation analysis and generates work orders. Finally, the application interaction layer issues operation instructions, which are then returned to the data awareness layer for execution. This constitutes a complete cycle covering physical operations, on-chain data storage, analysis, work orders, and operations. The second is the outer closed loop of model optimization. This process involves the intelligent analysis layer outputting related recommendation results, which, after receiving human feedback from the application interaction layer, are returned to the model training module for incremental training and optimization. This improves the quality of subsequent recommendations, thus forming an evolutionary cycle encompassing recommendation, feedback, training, and recommendation optimization. Through this dual-loop mechanism, a complete trust transfer is ultimately achieved from the underlying trusted data source to the top-level trusted decision-making.

[0149] 2. Core Business Processes and Collaboration Examples

[0150] The core process of this embodiment embodies the governance logic of "mandatory on-chaining, trust-first; intelligent discovery, closed-loop feedback." For example... Figure 3 As shown, this diagram, presented in sequence, uses the example of "replacing a faulty core network router board" to illustrate how the various components of the system work together to complete a full closed loop.

[0151] 1) Trigger: The network management system (OSS) reports a port failure alarm for a certain core network router element (resource R).

[0152] 2) Perception and Intelligent Diagnosis: The GNN intelligent analysis layer captures the alarm in real time and immediately queries the blockchain to obtain the status of the physical board asset A currently marked as carrying this network element R (displayed as "in use"). The GNN model completes the comparison within seconds and judges it as "resource status does not match asset registration status", and then generates a high-priority abnormal event of "suspected hardware failure" and pushes it to the application service layer.

[0153] 3) Decision-making and Standardized Execution: The application service layer automatically creates fault handling work orders and assigns them to on-site engineers. The engineer arrives at the data center, locates the faulty board (Asset A), and replaces it with a new board (Asset B). After installation, the engineer uses a mobile terminal to scan the barcode of Asset B. This scanning action is not a simple recording, but triggers a structured data packet. After preprocessing, it calls the asset digital twin management contract on the blockchain to execute an atomic transaction containing two operations: a) update the status of Asset A to "standby"; b) register the new Asset B, with an initial status of "in use". This transaction is confirmed after consensus by the blockchain network.

[0154] 4) Intelligent Reconstruction and Recommendation: Events updating the asset status on the blockchain are subscribed to by the GNN layer. Based on the latest network topology (new board B has been inserted into the original slot), asset attributes, and historical association patterns, the GNN model performs inference in real time, calculating a confidence level of up to 98.7% that there is a correlation between asset B and resource R. This recommended association is immediately pushed to the engineer's mobile terminal interface.

[0155] 5) Confirmation and Authoritative Verification: After the engineer verifies that everything is correct, they click "Confirm Association" on the terminal interface. This operation triggers a call to the blockchain association anchoring contract, permanently writing the digital fingerprint hash value of the association pair consisting of the twin ID of asset B and the R ID, along with the operation time and work order number, into the blockchain.

[0156] 6) Closed-loop and state synchronization: Within minutes, the states of assets A / B, resource R, and the relationship between asset B and resource R are all globally trusted and consistent on the blockchain. The network management system can update the spare parts information for resource R based on this, and the financial system can simultaneously initiate depreciation changes for asset A. The entire process achieves full automation and trustworthiness from fault detection, root cause location, standardized parts replacement, relationship reconstruction to global synchronization.

[0157] To clearly demonstrate the specific implementation path of the governance logic of "forced on-chain, trust priority; intelligent discovery, closed-loop feedback" in this embodiment, the following uses "core network router board failure replacement" as a typical scenario, and elaborates in detail the entire process of each component working together to complete a complete closed loop in the form of a sequence diagram. The specific steps include:

[0158] 1. The OSS system reports a resource failure;

[0159] 2. The GNN layer detected an anomaly of state inconsistency;

[0160] 3. The application layer automatically generates processing work orders;

[0161] 4. Replace and scan the code at the maintenance site;

[0162] 5. Call AssetLifecycleContract to update the state;

[0163] 6. The blockchain confirms the transaction and publishes the event;

[0164] 7. GNN layer subscription events trigger recalculation;

[0165] 8. Output new association recommendations (98.7% confidence level);

[0166] 9. Maintenance personnel confirm the recommended association;

[0167] 10. Call AssociationAnchorContract to anchor;

[0168] 11. The interface displays that the closed loop is complete;

[0169] 12. External system (OSS / ERP) synchronization update status;

[0170] 3. Specific technical implementation details;

[0171] 3.1 A dedicated GNN model for discovering telecommunications associations.

[0172] In this embodiment, the graph neural network model employs a relational graph convolutional network (R-GCN) because it can explicitly handle various types of relationships in heterogeneous graphs. For each node i in the graph, its feature representation at layer l+1... Updates are achieved by aggregating information from itself and its neighbors, specifically defined by the following formula: This aggregation of differentiated information along edges with specific relationships is implemented as follows:

[0173] The feature vector of node i in layer l+1 The calculation formula is:

[0174]

[0175] in:

[0176] To represent a nonlinear activation function, the ReLU function is typically used, i.e. ;

[0177] Let d represent the feature vector of node i in the l-th layer, and let its dimension be d (usually 64, 128 or 256).

[0178] when hour, The initial node feature vector is the input.

[0179] and Let r and r represent the trainable weight matrices of the self-loop links and their corresponding relationships in the l-th layer, respectively, with dimensions 1 and 2. The parameter values ​​are obtained through model training;

[0180] A predefined set of relation types in an R table view. The total number of relation types (usually set to 3 to 10 based on telecommunications engineering practice). r is the index of the relation type, with a value range of... ,For example It can represent a "hosted_on" relationship. It can represent a "physical connection (connected_to)" relationship;

[0181] Let j represent the set of neighboring nodes of node i under relation type r. The elements in this set are the indices of the neighboring nodes, and it can be an empty set. j is the index of a neighboring node in this set.

[0182] Let be the normalization constant for node i under relation r, typically taking the value of . (i.e., the number of neighboring nodes), or you can take It is used to balance the influence of nodes with different degrees.

[0183] This formula is the core of this model, enabling nodes to differentially aggregate contextual information from different types of connections. The association confidence score between asset node a and resource node r is obtained by calculating the cosine similarity of their final layer feature representations:

[0184] After L layers of propagation, asset node a and resource node b respectively obtain their final feature vectors. and .

[0185] The confidence level of the association between the two This is obtained by calculating the cosine similarity between the two vectors.

[0186]

[0187] in:

[0188] The dot product (inner product) of two eigenvectors is the sum of the products of their corresponding components. ;

[0189] and Let L and L be the magnitudes (L2 norms) of the two vectors, respectively. and , used to normalize vectors;

[0190] d is the dimension of the feature vector, which is consistent with the dimension of the node feature vector mentioned above, and is usually 64, 128 or 256.

[0191] and These represent the k-th dimension components of the final feature vectors of asset nodes and resource nodes, respectively, and their values ​​are calculated by forward propagation of the model.

[0192] Calculated association confidence The value range is [0,1]. The closer the value is to 1, the higher the probability that there is a relationship between asset node a and resource node b.

[0193] Model training and continuous learning:

[0194] Initial training: Supervised training is conducted using verified <asset, resource> association pairs extracted from historical compliance work orders and configuration libraries as monitoring signals.

[0195] Online learning: During system operation, each "confirmation" or "rejection" operation by maintenance personnel on the recommendation association is treated as a new labeled sample. The system periodically (e.g., weekly) uses the accumulated new samples to fine-tune the model, achieving continuous performance evolution.

[0196] 3.2 Design of smart contracts to implement business rule coding:

[0197] (1) AssetLifecycleContract: This contract encodes management rules into executable logic. For example, its decommissionAsset function includes the following steps: 1) Permission verification; 2) Querying AssociationAnchorContract to confirm that the asset has no active associations; 3) Executing the status change. This design technically prevents the illegal operation of "decommissioning assets with business associations".

[0198] (2) AssociationAnchorContract: Its anchor function records the timestamp, operator digital signature and triggering transaction hash when storing association hash, forming a complete, verifiable and non-repudiable evidence chain, providing the ultimate basis for auditing.

[0199] 3.3 Data Model and Interface Specifications:

[0200] This embodiment defines a unified asset-resource association data model. This model mandates that any asset record circulating within the system must contain a unique reference ID pointing to its corresponding digital twin on the blockchain. Simultaneously, all data services accessed externally through the API gateway, when returning association information, will include the notarized transaction hash of that association on the blockchain, allowing external systems to independently verify it.

[0201] This embodiment proposes a closed-loop association method based on blockchain and graph neural networks, aiming to address the pain points of missing associations and data inconsistencies between physical assets and logical resources in telecommunications networks. This method relies on a "four-layer, two-ring" architecture, using smart contracts to perform rule verification and on-chain notarization of asset state changes, constructing a trustworthy digital twin of assets. Simultaneously, it combines resource data to construct a heterogeneous information graph, utilizing a relational graph convolutional network (R-GCN) to calculate association confidence, achieving intelligent discovery and anomaly alerts. Based on this, manual feedback triggers on-chain anchoring of association relationships, and the GNN model is continuously fine-tuned using feedback data, forming a self-optimizing mechanism of "trustworthy data driving intelligent discovery, and intelligent discovery feeding back into a trustworthy closed loop." This comprehensively empowers the automated and trustworthy governance of telecommunications network assets throughout their entire lifecycle, from 5G construction and cloud network operation and maintenance to network decommissioning.

[0202] Example 2:

[0203] like Figure 4 As shown, this embodiment provides a device for associating physical assets and logical resources. The device includes:

[0204] Acquisition unit 10 is used to acquire physical asset data and logical resource data;

[0205] The standardization unit 20, connected to the acquisition unit 10, is used to standardize physical asset data to obtain standardized asset data; and to standardize logical resource data to obtain standardized resource data.

[0206] The generation unit 30, connected to the standardization unit 20, is used to perform business rule verification and consensus notarization on the preset state changes of standardized asset data through smart contracts deployed on the permissioned blockchain, thereby generating a digital twin of the asset.

[0207] The construction unit 40, connected to the generation unit 30, is used to construct a heterogeneous information graph of the telecommunications network based on the state change events of the asset digital twin, combined with standardized resource data and network topology relationships; wherein, the heterogeneous information graph of the telecommunications network contains various types of nodes and predefined telecommunications engineering semantic relationships;

[0208] The first processing unit 50, connected to the construction unit 40, is used to perform representation learning on nodes in the heterogeneous information graph of the telecommunications network using a graph neural network model, and to calculate the association confidence between physical asset nodes and logical resource nodes based on the learned node representation vectors.

[0209] The output unit 60 is connected to the first processing unit 50 and is used to generate association recommendation results based on association confidence, and to generate anomaly alarms based on the inconsistency between the state of the asset digital twin and the state of the standardized resource data, and output the association recommendation results and anomaly alarms to the user terminal.

[0210] The receiving unit 70, connected to the output unit 60, is used to receive confirmation instructions for the associated recommendation results and correction instructions for abnormal alarms, triggering the smart contract to anchor and store the association between physical assets and logical resources, and generate storage records.

[0211] The second processing unit 80, connected to the receiving unit 70, is used to collect confirmation instructions, correction instructions, and evidence records as feedback data, and to use the feedback data to perform incremental learning or fine-tuning of the graph neural network model to form a model optimization closed loop, thereby realizing the association between physical assets and logical resources.

[0212] The apparatus in this embodiment is capable of performing the method in Embodiment 1.

[0213] Example 3:

[0214] like Figure 5 As shown, this embodiment provides an electronic device, which includes a memory 200 and a processor 100. The memory 200 stores a computer program. When the processor 100 runs the computer program stored in the memory 200, the processor 100 executes the method for associating physical assets and logical resources according to Embodiment 1.

[0215] It is understood that the above embodiments are merely exemplary implementations used to illustrate the principles of the present invention, and the present invention is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A method for associating physical assets with logical resources, characterized in that, include: Acquire physical asset data and logical resource data; The physical asset data is standardized to obtain standardized asset data. Furthermore, the logical resource data is standardized to obtain standardized resource data; By using smart contracts deployed on the permissioned blockchain to verify business rules and consensus notarization of the state changes of the standardized asset data, a digital twin of the asset is generated. Based on the state change events of the asset digital twin, a heterogeneous information graph of the telecommunications network is constructed by combining the standardized resource data and network topology relationships; wherein, the heterogeneous information graph of the telecommunications network includes multiple types of nodes and predefined telecommunications engineering semantic relationships; A graph neural network model is used to learn the representations of nodes in the heterogeneous information graph of the telecommunications network, and the association confidence between physical asset nodes and logical resource nodes is calculated based on the learned node representation vectors. Generate association recommendation results based on the association confidence level, and generate an anomaly alarm based on the inconsistency between the state of the asset digital twin and the state of the standardized resource data, and output the association recommendation results and the anomaly alarm to the user terminal; Upon receiving confirmation instructions for the associated recommendation results and correction instructions for the abnormal alarms, the smart contract is triggered to anchor and store the association between physical assets and logical resources, generating a storage record. The confirmation instruction, the correction instruction, and the evidence record are collected as feedback data, and the feedback data is used to incrementally learn or fine-tune the graph neural network model to form a model optimization closed loop, thereby realizing the association between physical assets and logical resources.

2. The method for associating physical assets and logical resources according to claim 1, characterized in that, The process of verifying business rules and establishing consensus for the state changes of the standardized asset data through smart contracts deployed on the permissioned blockchain specifically includes: Invoke the first smart contract deployed on the permissioned blockchain, and determine whether the status change request of the standardized asset data meets the preset business rules based on the first smart contract; If the status change request meets the preset business rules, then the type of the status change request is further identified; When the status change request is an asset scrapping request, the second smart contract deployed on the permissioned chain is invoked to perform association conflict verification. The state change request that passes the correlation conflict verification is submitted to the permissioned chain consensus node as a legitimate transaction for consensus confirmation. After consensus is reached, the state change is recorded in the distributed ledger, and the digital twin state of the standardized asset data in the first smart contract is updated synchronously. The first smart contract is an asset digital twin management contract, used to define the finite state machine of the standardized asset data and execute business rule verification; the second smart contract is an association anchoring contract, used to store the digital fingerprint of the confirmed asset and resource association relationship and related metadata.

3. The method for associating physical assets and logical resources according to claim 1, characterized in that, The method of using a graph neural network model to learn the representation of nodes in the heterogeneous information graph of the telecommunications network specifically includes: The graph neural network model is constructed using a relational graph convolutional network; The self-loop weight matrix preset by the graph neural network model is used to transform the self-loop features of the target node to obtain the self-loop features; According to the predefined telecommunications engineering semantic relationship type, the neighbor nodes of the target node are divided into multiple neighbor node sets; For any of the neighbor node sets, the node features of each neighbor node in the neighbor node set are transformed and aggregated using an independent weight matrix that matches the semantic relationship type of telecommunications engineering, so as to obtain the neighbor aggregated features corresponding to the semantic relationship type of telecommunications engineering. By summarizing the aforementioned neighbor aggregation features, the total neighbor aggregation features are obtained. The self-looping features are fused with the total neighbor aggregation features, and then processed by a nonlinear activation function to obtain the updated node representation vector.

4. The method for associating physical assets and logical resources according to claim 1, characterized in that, Receiving confirmation instructions for the associated recommendation results and correction instructions for the abnormal alarms triggers the smart contract to anchor and store the association between physical assets and logical resources, generating a storage record, specifically including: Receive and parse the confirmation instruction and the correction instruction to obtain the mapping relationship between physical assets and logical resources and the status correction information; The smart contract is triggered to generate a state update proposal, which includes the current state of the asset digital twin, the target state, and the corrected state of the standardized resource data. The state update proposal is validated through consensus, and after consensus is reached, the mapping relationship and the state update proposal are subjected to joint hashing to obtain the target hash value. The smart contract is invoked to anchor and store the target hash value on the permissioned blockchain. Based on the evidence storage results, the status identifier of the digital twin of the asset is updated to the associated calibration status, and an evidence storage record containing the on-chain transaction hash, timestamp, and status change log is generated.

5. The method for associating physical assets and logical resources according to claim 1, characterized in that, The calculation of the association confidence between physical asset nodes and logical resource nodes based on the learned node representation vectors specifically includes: Extract the first feature vector and the second feature vector of the physical asset node and the logical resource node respectively from the output layer of the graph neural network model; Calculate the cosine similarity between the first feature vector of the node and the second feature vector of the node, and use the value of the cosine similarity as the association confidence.

6. The method for associating physical assets and logical resources according to claim 1, characterized in that, The generation of an anomaly alert based on the inconsistency between the state of the asset digital twin and the state of the standardized resource data specifically includes: Compare the real-time status of the standardized resource data with the real-time status of the asset digital twin; If the real-time status of the asset digital twin is inconsistent with the real-time status of the standardized resource data or if there is a logical conflict in the relationship, an anomaly alarm will be generated.

7. The method for associating physical assets and logical resources according to claim 6, characterized in that, The abnormal alarms specifically include: alarms for unassociated resources, alarms for assets with no listings, or alarms for associated conflicts; The unassociated resource alert is used to indicate isolated logical resources that are not carried by the asset digital twin; The asset idle alarm is used to indicate an empty digital twin of an asset that does not carry valid logical resources. The association conflict alarm is used to indicate that there is a logical contradiction in the association between assets and resources.

8. The method for associating physical assets and logical resources according to any one of claims 1 to 7, characterized in that, The incremental learning or fine-tuning of the graph neural network model using the feedback data includes: The physical asset and logical resource association pairs corresponding to the confirmation command entered by the user are marked as positive samples, and the correction command entered by the user is marked as a negative sample. The positive and negative samples, along with their corresponding node features, are stored in the training sample library for accumulation. When the preset triggering conditions are met, a target loss function is constructed based on the positive and negative samples accumulated in the training sample library. The target loss function aims to maximize the cosine similarity of positive sample pairs in the representation space and minimize the cosine similarity of negative sample pairs. Using the parameters of the current graph neural network model as initial weights, a fine-tuned learning rate less than the initial training phase learning rate is adopted. The graph neural network model is then updated by backpropagation based on the target loss function to obtain an optimized graph neural network model and replace the current model. The preset triggering conditions include: the number of new samples reaching a preset threshold, or reaching a preset periodic time interval.

9. A device for associating physical assets with logical resources, characterized in that, include: The acquisition unit is used to acquire physical asset data and logical resource data. A standardization unit, connected to the acquisition unit, is used to standardize the physical asset data to obtain standardized asset data. Furthermore, the logical resource data is standardized to obtain standardized resource data; The generation unit, connected to the standardization unit, is used to perform business rule verification and consensus notarization on the preset state changes of the standardized asset data through smart contracts deployed on the permissioned blockchain, thereby generating a digital twin of the asset. A construction unit, connected to the generation unit, is used to construct a heterogeneous information graph of a telecommunications network based on the state change events of the asset digital twin, combined with the standardized resource data and network topology relationships; wherein, the heterogeneous information graph of the telecommunications network includes multiple types of nodes and predefined telecommunications engineering semantic relationships; The first processing unit, connected to the construction unit, is used to perform representation learning on the nodes in the heterogeneous information graph of the telecommunications network using a graph neural network model, and to calculate the association confidence between physical asset nodes and logical resource nodes based on the learned node representation vectors. An output unit, connected to the first processing unit, is used to generate association recommendation results based on the association confidence level, and to generate an anomaly alarm based on the inconsistency between the state of the asset digital twin and the state of the standardized resource data, and to output the association recommendation results and the anomaly alarm to the user terminal. The receiving unit, connected to the output unit, is used to receive confirmation instructions for the associated recommendation results and correction instructions for the abnormal alarms, triggering the smart contract to anchor and store the association relationship between physical assets and logical resources, and generate a storage record. The second processing unit, connected to the receiving unit, is used to collect the confirmation instruction, the correction instruction, and the evidence record as feedback data, and to use the feedback data to perform incremental learning or fine-tuning on the graph neural network model to form a model optimization closed loop, thereby realizing the association between physical assets and logical resources.

10. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the method for associating physical assets and logical resources according to any one of claims 1 to 8.

Images