A smart park electronic equipment asset digitization ledger management system

Abstract

This invention specifically relates to a digital ledger management system for electronic equipment assets in smart parks, belonging to the field of smart park asset management technology. It includes a status acquisition and update module, used to periodically collect runtime indicators of devices through sensing agent software deployed on edge gateways, and write the indicator data into the dynamic attribute layer of the corresponding digital twin instance. In this invention, a digital twin data model containing static attribute layers, dynamic attribute layers, and spatial attribute layers is established for each device. The ledger record is expanded from a static archive storing only fixed fields to a holographic record covering the device's operating status and physical location, thus broadening the information dimensions carried by the ledger.

Description

Technical Field

[0001] This invention relates to the field of smart park asset management technology, and in particular to a digital ledger management system for smart park electronic equipment assets. Background Technology

[0002] The smart park is equipped with a large number of heterogeneous electronic devices, including cameras, access control card readers, temperature and humidity sensors, edge gateways, wireless access points, and large information display screens. To facilitate operation and maintenance management, the park typically uses an asset ledger system to record these devices in a unified manner.

[0003] Existing asset ledger systems record equipment information statically, including only fixed fields such as purchase date, model, and installation location. The ledger content relies on manual entry and updates. When equipment locations are moved or operating status changes frequently, the discrepancy between the ledger records and the actual state of the equipment continues to widen, resulting in a long-term disconnect between the ledger and the physical state of the equipment.

[0004] Furthermore, in deployment scenarios involving multiple parks and multiple tenants, the equipment ledger data in each region involves tenant privacy and cannot be centrally aggregated into a unified platform for cross-domain analysis, making it difficult to achieve cross-park equipment anomaly detection and asset collaborative reuse. Summary of the Invention

[0005] The purpose of this invention is to provide a digital ledger management system for electronic equipment assets in smart parks in order to solve the above-mentioned problems.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A digital ledger management system for electronic equipment assets in a smart park includes: The digital twin model generation module is used to create digital twin data models for electronic devices in the park, including static attribute layers, dynamic attribute layers, and spatial attribute layers, and to generate digital twin instance records for each network-connected device. The status acquisition and update module is used to periodically collect the device's runtime metrics through the perception agent software deployed on the edge gateway, and write the metric data into the dynamic attribute layer of the corresponding digital twin instance. The location awareness and synchronization module is used to sense the physical location of the device in real time using the positioning infrastructure. When the deviation between the current location coordinates of the device and the location coordinates recorded in the spatial attribute layer exceeds a preset distance threshold, the location coordinates in the spatial attribute layer are updated to the current location coordinates. The federated collaboration and fusion module is used for each park to maintain its local knowledge graph. It adopts a federated knowledge graph framework for cross-park collaboration. Each park uploads its local model parameters to the cloud after differential privacy processing. The cloud performs federated averaging to generate global model parameters and distributes them to each park. The conflict detection and correction module is used to compare the recorded values ​​of the digital twin instances of each device with the real-time data reported by the perception agent software. When the two are inconsistent, it is marked as a ledger conflict event and correction is performed. The digital twin instance set is output as a dynamic holographic ledger.

[0007] Preferably, the static attribute layer records the inherent information of the device, including serial number, manufacturer name, purchase date and warranty period; the dynamic attribute layer records the device's runtime indicators, including online status, CPU utilization, memory utilization, signal strength and predicted remaining lifespan; the spatial attribute layer records the device's physical location information, including building number, floor number, room coordinates and business domain identifier.

[0008] Preferably, the digital twin model generation module further includes: defining the semantic associations between device instances using an ontology language to generate a ledger knowledge graph pattern; wherein the semantic associations include subordinate relationships and data supply relationships, the subordinate relationship indicating that one device belongs to another device in the network topology, and the data supply relationship indicating that one device provides data input to another device; the subordinate relationship and the data supply relationship are stored in the ledger knowledge graph in the form of directed edges.

[0009] Preferably, the perception agent software supports data collection using multiple communication protocols: for devices that support active reporting, it receives periodically pushed status indicators from the devices via SNMP, MQTT, or Modbus protocols; for dumb terminal devices that do not support active reporting, it obtains status information using active detection methods, including LLDP protocol discovery, ARP address scanning, and ICMP liveness detection. When the network connection between the edge gateway and the cloud is lost, the perception agent software writes the collected indicator data into the local cache queue in timestamp order; when the network connection is restored, the data is retransmitted to the cloud in the time order of the local cache queue, and the cloud writes it into the digital twin instance of the corresponding device in timestamp order.

[0010] Preferably, the perception agent software further includes an LSTM model deployed at the edge. The processing of the LSTM model includes: standardizing the CPU utilization rate sequence, memory utilization rate sequence, and signal strength sequence of a single device within a preset time window, concatenating them into a runtime indicator time series, and inputting it into the LSTM model; the output layer of the LSTM model is a fully connected classification layer, and the output is a health anomaly flag of the device, with a value of normal or abnormal. Specifically, when the health anomaly indicator is normal, the perception agent software only uploads the compressed status fingerprint to the cloud; when the health anomaly indicator is abnormal, the perception agent software uploads the abnormal feature vector and the complete time series segment to the cloud together.

[0011] Preferably, the positioning infrastructure adopts a fusion positioning method, which comprehensively utilizes at least two positioning methods from Wi-Fi probes, Bluetooth RSSI ranging, UWB tag ranging, and camera visual recognition; each positioning method outputs a position estimate, and the edge positioning engine performs weighted fusion of multiple position estimates to generate the final device position coordinates; wherein, the weight of each positioning method is determined according to the positioning accuracy level of the positioning method in the current environment, the higher the accuracy level, the greater the weight, and the sum of all weights is one; Specifically, when it is determined that the device has moved, a digital twin update event is generated, and the location change record is written to the device's location change history list.

[0012] Preferably, the location awareness and synchronization module further includes automatic identification and recording of pluggable devices: when a device carrying an NFC electronic tag connects to the edge gateway, the awareness agent software reads the device identification information in its NFC tag. If the device identification does not have a corresponding digital twin instance in the current ledger knowledge graph, a new digital twin instance record is automatically created. When the device does not carry an NFC tag, the awareness agent software generates a device fingerprint based on the device's MAC address, firmware version, and hardware feature parameters, and matches it with the existing device fingerprints in the ledger knowledge graph. If the match is successful, it is associated with an existing digital twin instance; if the match fails, a new digital twin instance record is created.

[0013] Preferably, the federated knowledge graph framework employs a federated graph neural network algorithm, comprising a local graph encoding unit, a secure gradient aggregation unit, and a global parameter distribution unit; The local graph encoding unit is deployed at the edge nodes of each park. It uses device instances in the local knowledge graph as graph nodes and relationships as graph edges. It standardizes each dynamic attribute field to generate node feature vectors and executes the message passing process of the graph neural network. In each layer, the average of the representations of each node to its neighbor nodes and its own representation is calculated and then processed by the weight matrix and activation function to obtain the updated node representation. After multiple layers of message passing, the fully connected output layer is connected to the node representation of the final layer to output the anomaly probability of each node. The local model gradient is generated using the cross-entropy loss function. The security gradient aggregation unit is deployed in the cloud, receives local model parameters uploaded by each park after differential privacy processing, and calculates the average of the local model parameters of all parks to generate global model parameters. The global parameter distribution unit distributes the global model parameters to the local graph coding units of each park. Each local graph coding unit replaces its local parameters with the global model parameters and continues the next round of training.

[0014] Preferably, the handling of the ledger conflict event includes: when the ledger record shows that a device is online but the sensing agent software does not respond after multiple consecutive probes, the status of the device is marked as suspected asset loss or disconnection; if the device recovers its response within a preset time threshold, the mark is removed and the original ledger record is maintained; if it still does not recover after the preset time threshold, a manual inventory task instruction is generated and pushed to the operation and maintenance management terminal; when the spatial attribute of a device in the ledger record is inconsistent with the positioning result and the location change event has been written, the spatial attribute value in the ledger is automatically overwritten with the positioning result.

[0015] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. This invention establishes a digital twin data model for each device, which includes a static attribute layer, a dynamic attribute layer, and a spatial attribute layer. The ledger records are expanded from static archives that only store fixed fields to holographic records that cover the device's operating status and physical location, thereby expanding the information dimensions carried by the ledger.

[0016] 2. This invention deploys sensing agent software on the edge gateway to obtain device status in real time and synchronize it to the digital twin instance through a combination of multi-protocol collection and active detection. The ledger records can be continuously refreshed as the device status changes, overcoming the problem of long-term disconnect between the ledger and the actual status caused by the reliance on manual input in traditional ledgers. Attached Figure Description

[0017] Further details, features, and advantages of this application are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which: Figure 1 This is a system structure diagram of the present invention. Detailed Implementation

[0018] Several embodiments of this application will now be described in more detail with reference to the accompanying drawings to enable those skilled in the art to implement this application. This application may be embodied in many different forms and for various purposes and should not be limited to the embodiments set forth herein. These embodiments are provided to make this application thorough and complete, and to fully convey the scope of this application to those skilled in the art. The embodiments described do not limit this application.

[0019] Unless otherwise defined, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It will be further understood that terms such as those defined in commonly used dictionaries shall be interpreted as having a meaning consistent with their meaning in the relevant field and / or the context of this specification, and shall not be interpreted in an idealized or overly formal sense unless expressly defined herein.

[0020] Example 1

[0021] Its specific implementation method is combined with the appendix Figure 1 Please provide a detailed explanation.

[0022] In this embodiment, it includes: Step 1: Establish a digital twin data model of the equipment and generate a semantic ledger ontology.

[0023] For each type of electronic device in the park, a standardized digital twin data model is established. The digital twin data model includes three attribute layers: static attribute layer, dynamic attribute layer, and spatial attribute layer.

[0024] The static attribute layer records the device's inherent information, including serial number, manufacturer name, purchase date, and warranty period. The dynamic attribute layer records the device's runtime metrics, including online status, CPU utilization, memory utilization, signal strength, and predicted remaining lifespan. The spatial attribute layer records the device's physical location information, including building number, floor number, room coordinates, and the business domain identifier.

[0025] Based on the above three attribute layers, a digital twin instance record is generated for each network-connected device, serving as the basic data entry for the dynamic holographic ledger.

[0026] It should be noted that semantic relationships exist between the device instances in the aforementioned digital twin data model. The semantic relationships between devices are defined using the OWL ontology language, generating a ledger knowledge graph schema. These relationships include dependency relationships and data supply relationships. A dependency relationship indicates that one device belongs to another in the network topology; for example, an IP camera belongs to the subnet of a wireless access point. A data supply relationship indicates that one device provides data input to another; for example, a temperature and humidity sensor provides environmental parameters to an air conditioner controller. These relationships are stored in the ledger knowledge graph as directed edges, giving the ledger a reasonable semantic structure.

[0027] Step 2: Collect real-time device status data and update the dynamic attribute layer of the digital twin.

[0028] Perception agent software is deployed on the edge gateway of each park. The perception agent software periodically collects the runtime metrics of the devices under its jurisdiction and writes the collected metric data into the dynamic attribute layer of the corresponding device's digital twin instance to complete the real-time status synchronization of the ledger.

[0029] It should be noted that the aforementioned sensing agent software supports data collection using multiple communication protocols. For devices that support proactive reporting, the sensing agent software receives periodically pushed status indicators from the devices via SNMP, MQTT, or Modbus protocols. For dumb terminal devices that do not support proactive reporting, the sensing agent software uses proactive probing methods to obtain their status information. Proactive probing methods include LLDP protocol discovery, ARP address scanning, and ICMP liveness detection. Through proactive probing, the software obtains the liveness status, IP address changes, and connection topology information of dumb terminal devices.

[0030] In this embodiment, to achieve preliminary assessment of device health at the edge and reduce the amount of data transmitted to the cloud, the perception agent software also includes a locally running LSTM model. The CPU utilization sequence, memory utilization sequence, and signal strength sequence of a single device within a preset time window are Z-score standardized, concatenated into a runtime indicator time series, and input into the LSTM model. The output layer of the LSTM model is a fully connected classification layer, outputting a health anomaly flag for the device, with values ​​either normal or abnormal. During training, historical runtime indicator time series are used as training samples, manually labeled health status tags are used as supervision signals, a binary cross-entropy loss function is employed, and the Adam optimization algorithm is used for parameter updates. When the health anomaly flag is normal, the perception agent software only uploads the compressed status fingerprint to the cloud; when the health anomaly flag is abnormal, the perception agent software uploads the abnormal feature vector along with the complete time series segment. This method reduces the transmission overhead of conventional status data in bandwidth-constrained edge environments.

[0031] In this embodiment, to ensure that state data is not lost during network outages, the perception agent software also has the capability of network outage caching and retransmission. When the network connection between the edge gateway and the cloud is lost, the perception agent software writes the collected indicator data into a local cache queue in timestamp order. When the network connection is restored, the perception agent software retransmits the data to the cloud in the time order of the local cache queue, and the cloud then rewrites the data into the digital twin instance of the corresponding device according to the timestamp.

[0032] Step 3: Perform dynamic asset discovery and real-time location synchronization, and update the digital twin spatial attribute layer.

[0033] For devices that have completed step 2 (status acquisition), step 3 can be performed. Step 3 may specifically include: Utilizing positioning infrastructure deployed within the park, the physical location of electronic devices is sensed in real time. When the deviation between the current location coordinates of a device and the location coordinates recorded in the spatial attribute layer of its digital twin instance exceeds a preset distance threshold, it is determined that the device has moved. At this time, a digital twin update event is generated, updating the location coordinates in the spatial attribute layer of the device to the current location coordinates, and this location change record is written to the device's location change history list.

[0034] It should be noted that the aforementioned positioning infrastructure employs a fusion positioning method. This method integrates at least two positioning techniques from Wi-Fi probes, Bluetooth RSSI ranging, UWB tag ranging, and camera visual recognition. Each positioning method outputs a location estimate, which is then weighted and fused by the edge positioning engine to generate the final device location coordinates. The formula for weighted fusion is as follows: ; in, The total number of positioning methods participating in the integration. For the index of positioning methods, ; For the first The position estimate output by each positioning method is represented by a coordinate vector in a unified coordinate system; For the first The weight of each positioning method is determined based on its positioning accuracy level in the current environment; the higher the accuracy level, the greater the weight, and the following conditions must be met. ; These are the final fusion position coordinates. Each All are represented by coordinate vectors in the same coordinate system, with consistent dimensions, and the weighted summation operation is valid in terms of dimensions.

[0035] In this embodiment, to achieve automatic identification and accounting of pluggable devices, NFC electronic tag identification or device fingerprint identification technology is used for replaceable sensor modules, portable terminals, and other pluggable devices. When a device carrying an NFC electronic tag connects to the edge gateway, the perception agent software reads the device identification information from its NFC tag. If the device identification does not have a corresponding digital twin instance in the current ledger knowledge graph, a new digital twin instance record is automatically created, static attribute layer information is filled in, and the dynamic attribute layer and spatial attribute layer are initialized to the currently collected values. When the device does not carry an NFC tag, the perception agent software generates a device fingerprint based on the device's MAC address, firmware version, and hardware characteristic parameters, and matches it with existing device fingerprints in the ledger knowledge graph. If the match is successful, it is associated with an existing digital twin instance; if the match fails, a new digital twin instance record is created.

[0036] Step 4: Achieve cross-park ledger collaboration and integration based on federated knowledge graphs

[0037] Building upon steps 1 through 3, each park has maintained its own local knowledge graph. This local knowledge graph contains digital twin instances of all devices within the park, the edges connecting these instances, and dynamically updated attribute values. When multiple smart parks exist or multiple tenants reside within the same park, each region does not upload its original device ledger data to the cloud; instead, it collaborates using a federated knowledge graph framework.

[0038] A global schema layer and aggregated statistical information are maintained in the cloud. The global schema layer defines a unified equipment type classification system and semantic specifications for relationships. The aggregated statistical information includes anonymized statistical indicators such as the quantity distribution of each equipment type across all participating parks, average failure rate, and average online rate.

[0039] It should be noted that in the above federated knowledge graph framework, the collaborative model training across campuses adopts the federated graph neural network algorithm. The federated graph neural network algorithm consists of three processing units: a local graph encoding unit, a secure gradient aggregation unit, and a global parameter distribution unit. The data transfer relationships between the units are as follows.

[0040] Local graph encoding units are deployed on edge nodes of each park. Each park's local graph encoding unit takes its park's local knowledge graph as input, with device instances as graph nodes and relationships as graph edges. Z-score normalization is applied to each dynamic attribute field to generate node feature vectors. ,in Let be the index of the node in the graph. As input features for each node, the local graph encoding unit executes the message passing process of the graph neural network. The node representation update formula for the layer is: ; in, For nodes In the Hidden representation of layers; This refers to the layer index of a graph neural network. This is the index of the current node; The index of the neighboring node; initial value ; For nodes The set of neighboring nodes; For the first The trainable weight matrix of the layer; This is a neighborhood aggregation function that calculates the average of the neighbor node representations and the node's own representation. This is the activation function. The specific calculation formula is as follows: ; in, For nodes The number of neighboring nodes. After After layer message passing, The total number of message passing layers is represented by the local graph coding unit at the final layer node. A fully connected output layer is connected above this to output the anomaly probability of each node. With device anomaly detection as the training objective, the cross-entropy loss function is used to generate the local model gradient. ,in This represents gradient operation. For the first The loss function value for each park, This is the park's number.

[0041] The security gradient aggregation unit is deployed in the cloud. After the local graph encoding units in each park complete their local gradient updates, they will update the local model parameters. After differential privacy processing, the data is uploaded to the cloud. For the first Local model parameters for each park, This serves as an index for the park. The safety gradient aggregation unit receives local model parameters uploaded by each park, performs federated averaging, and generates global model parameters. The federal average formula is: ; in, The total number of campuses participating in federal training. For the park's index, For the first Local model parameters uploaded by each park These are the aggregated global model parameters.

[0042] The global parameter distribution unit is also deployed in the cloud. The global parameter distribution unit distributes global model parameters... Local map coding units are distributed to each park, and each local map coding unit uses global model parameters. After replacing the local parameters, continue to the next round of training.

[0043] After multiple rounds of iterative training, each park has acquired a shared global capability for detecting device anomalies, which can utilize global model parameters. Anomaly detection is performed on local devices. For example, a federated knowledge graph can answer whether a particular device model has a common high-temperature failure risk across multiple campuses without exposing the IP address or installation location of the specific devices in each campus.

[0044] In this embodiment, to assist in decision-making for spare parts allocation and asset reuse in cross-park scenarios, a cross-park asset matching model is trained using a federated knowledge graph framework. The input to the cross-park asset matching model is the graph embedding vectors of idle equipment in each park. The graph embedding vectors describing the demand of similar equipment in other parks are used as matching targets. The output layer is a fully connected scoring layer, outputting an asset matching similarity score. During training, historical allocation records with known matching results are used as supervision signals. A contrastive loss function is employed, and the Adam optimization algorithm is used for parameter updates. The matching process is completed in the cloud based on the aggregated graph embeddings, without involving the exchange of original equipment data.

[0045] Step 5: Perform ledger conflict detection and automatic correction to generate a dynamic holographic ledger.

[0046] The system continuously compares the recorded values ​​of the digital twin instances of each device with the real-time data reported by the edge-aware agent. When inconsistencies occur between the two, they are marked as ledger conflict events.

[0047] When the ledger record shows that a device is online, but the sensing agent software fails to respond after multiple consecutive probes, the device's status is marked as suspected asset loss or disconnection. If the device recovers its response within a preset time threshold, the mark is removed and the original ledger record is maintained; if it still fails to recover after the preset time threshold, a manual inventory task instruction is generated and pushed to the operation and maintenance management terminal.

[0048] When the spatial attribute of a device in the ledger record is inconsistent with the location result in step 3, and the location change event in step 3 has been written, the spatial attribute value in the ledger will be automatically overwritten based on the result in step 3.

[0049] After the aforementioned conflict detection and correction processes, a set of digital twin instances for each device is output, which constitutes the dynamic holographic ledger. Each record in the dynamic holographic ledger contains static, dynamic, and spatial attributes that have been verified in real time.

[0050] In this embodiment, to improve the accuracy of the remaining lifespan prediction values ​​in the ledger, the system continuously collects actual operating lifespan data of the equipment. The actual operating lifespan data is used as feedback labels to perform online learning and updates to the remaining lifespan prediction values ​​in the dynamic attribute layer. For example, when the actual operating lifespan of a device exceeds the originally recorded estimated lifespan in the ledger, this deviation information is used as a training sample and input into the online learning process for remaining lifespan prediction, making the lifespan prediction values ​​of subsequent devices of the same model closer to their actual performance.

[0051] Example 2

[0052] A smart industrial park (hereinafter referred to as "Park A") has deployed hundreds of heterogeneous devices, including cameras, access control card readers, temperature and humidity sensors, edge gateways, wireless access points, and information display screens. Park A and the adjacent Park B belong to the same operating entity, but the device ledger data of the two parks belong to different tenants and cannot directly share the original data. In QB of 20XX, the operation and maintenance team found a significant discrepancy between the traditional static ledger and the actual status of the devices. Some devices were marked as online in the ledger, but had actually been offline for several days; in addition, several sensor modules had been relocated, but the original room coordinates were still retained in the ledger. To address this, Park A and Park B jointly adopted this method to implement dynamic holographic ledger management for all heterogeneous devices.

[0053] Step 1 implementation: Establish a digital twin data model of the equipment and generate a semantic ledger ontology.

[0054] The park's maintenance platform A established digital twin data model instances for six main types of equipment. Taking five representative devices as examples, the initialization records of the static attribute layer, dynamic attribute layer, and spatial attribute layer are shown in Table 1 below.

[0055] Table 1 Initialization Record of Device Digital Twin Instance

[0056] In the knowledge graph schema layer, DEV001 (IP camera) points to DEV003 (wireless access point) via a directed edge indicating a dependency relationship, signifying that the camera belongs to the subnet of DEV003 in the network topology. DEV002 (temperature and humidity sensor) points to the air conditioning controller on the 3rd floor of Building B via a directed edge indicating a data supply relationship, signifying that it provides parameter input for environmental control. These relationships are stored in the local knowledge graph using the OWL ontology language, providing the graph structure foundation for subsequent message passing in the graph neural network.

[0057] Step 2 implementation: Collect real-time device status data and update the dynamic attribute layer of the digital twin.

[0058] The sensing agent software on the edge gateway of the park performs multi-protocol data collection on the devices under its jurisdiction with a collection cycle of 5 minutes. DEV001, DEV002, and DEV003 support active reporting, pushing status indicators via MQTT, Modbus, and SNMP protocols, respectively; DEV004 is a dumb terminal device, and its status information is obtained by the sensing agent software through ICMP liveness detection and ARP address scanning.

[0059] During a certain data collection period, the perception agent software obtained the following raw data of runtime indicators and input them into the local LSTM model for health anomaly detection. The relevant data is shown in Table 2.

[0060] Table 2. Data collected on equipment operating indicators and results of health anomaly identification.

[0061] DEV003 exhibited high CPU and memory usage, leading the LSTM model to identify it as an anomaly. The perception agent software uploaded DEV003's anomaly feature vector along with the complete time series segment to the cloud. Based on this, the cloud updated the health status field in DEV003's dynamic attribute layer to anomaly. The other four devices only uploaded compressed status fingerprints, effectively reducing bandwidth consumption from the edge to the cloud.

[0062] Step 3 implementation: Perform dynamic asset discovery and real-time location synchronization, and update the digital twin spatial attribute layer.

[0063] Park A deployed three positioning methods in Buildings B and C: Wi-Fi probes, Bluetooth RSSI ranging, and UWB tag ranging. The edge positioning engine assigns weights based on the accuracy level of each method in the current indoor environment: UWB has the highest accuracy and the highest weight. Wi-Fi probes are the next most important, with higher weight. Bluetooth RSSI has the lowest accuracy and is the most weighted. ,satisfy .

[0064] During this positioning cycle, the three positioning methods of DEV002 (temperature and humidity sensor) each output position estimates, which are then fused to calculate the final coordinates: ; Table 3 shows the location fusion process and spatial attribute layer update records for DEV002.

[0065] Fusion coordinates Coordinates of the original ledger record The deviation between them is 7.32 meters, exceeding the preset distance threshold (3 meters). The system determines that DEV002 has moved, automatically generates a digital twin update event, and overwrites the position coordinates in the spatial attribute layer with the new coordinates. This change will be added to the location change history list of DEV002.

[0066] During the same data collection cycle, a portable temperature and humidity acquisition module (tag identification: NFC-TMP-0093) carrying an NFC tag connects to the edge gateway. The perception agent software reads its NFC tag, and if no corresponding digital twin instance is found in the local knowledge graph, it automatically creates a new instance DEV006, fills in the static attribute layer information, and initializes the dynamic attribute layer and spatial attribute layer to the current collection values, thus completing the automatic accounting.

[0067] Implementation of Step 4: Achieving cross-park ledger collaboration and integration based on federated knowledge graphs.

[0068] Park A and Park B each maintain their local knowledge graphs and jointly participate in the training of the federated graph neural network. Taking node DEV003 in the local knowledge graph of Park A as an example, its initial node feature vector It is composed of CPU utilization, memory utilization, and signal strength after being standardized by Z-score.

[0069] In the first round of federated training, the local graph coding units in each campus perform message passing. The set of neighboring nodes for node DEV003. Including DEV001 (connected via dependency), number of neighbors In the first-level node representation update, the aggregation function calculates the average of the initial representations of DEV001 and DEV003 themselves, and then applies the result to the weight matrix. Transform and pass through the activation function to obtain .go through After the layer message is passed, the output layer gives an abnormal probability of 0.87 for DEV003, which is higher than the discrimination threshold and consistent with the LSTM discrimination result in step 2, further confirming that DEV003 is in an abnormal state.

[0070] After each park's local map coding unit completes its local gradient calculation, it then processes the local model parameters. Differential privacy noise is applied, and the data is uploaded to the cloud-based secure gradient aggregation unit. Federated averaging is then performed in the cloud, where... and These are the local model parameters uploaded by Park A and Park B, respectively. The aggregated global model parameters: ; The aggregated records of parameters for each federated training round are shown in Table 4. Table 4 Aggregated Records of Federated Training Round Parameters

[0071] After eight rounds of iteration, the local losses for both parks converged to below 0.25, and the global model parameters... The data was distributed back to the local graph encoding units of each campus. Based on the global model, the federated knowledge graph identified that the wireless access point of model DEV003 exhibited an abnormal high memory usage pattern in both campuses, indicating a common firmware memory leak vulnerability in this model of device. Throughout the entire process, the device's IP address and installation coordinates remained locally in both campuses.

[0072] Step 5 implementation: Perform ledger conflict detection and automatic correction to generate a dynamic holographic ledger.

[0073] The system continuously compares the recorded values ​​of the digital twin instances of all devices with the real-time reported data of the perception agent. During this processing cycle, three types of ledger conflict events were detected, and the relevant processing results are shown in Table 5.

[0074] Table 5 Results of Conflict Detection and Handling in Ledgers

[0075] The location conflict of DEV002 is automatically covered by the location result of step 3 without manual intervention; the health abnormality of DEV003 has been confirmed by both steps 2 and 4, and the abnormality mark is synchronously written into the dynamic attribute layer and an operation and maintenance alarm is pushed; the online status conflict of DEV007 exceeds the preset time threshold (30 minutes), the system generates a manual inventory task instruction and pushes it to the operation and maintenance management terminal.

[0076] After the above conflict detection and correction, Park A outputs a dynamic holographic ledger containing DEV001 to DEV007 (including DEV006 which is automatically entered into the ledger). Each record contains three layers of complete data: static attributes, dynamic attributes, and spatial attributes, which are verified in real time.

[0077] The logical chain of the entire data flow is as follows: Step 1 establishes a three-layer digital twin instance for each device, providing a structured container for all subsequent data writing; Step 2's perception agent writes real-time runtime indicators into the dynamic attribute layer, and the LSTM model completes edge health judgment to determine the upload granularity; Step 3's fusion positioning results are written into the spatial attribute layer, and the DEV002's position movement event is generated, becoming the basis for spatial conflict detection in Step 5; Step 4 uses the local knowledge graph accumulated from Steps 1 to 3 as input, and completes cross-park abnormal pattern sharing through federated graph neural networks without exchanging original data. Its output global model parameters further enhance the local anomaly judgment capability of each park; Step 5 aggregates the outputs of all the above steps, performs automatic correction or manual intervention triggering for the three types of conflicts: dynamic attributes, spatial attributes, and online status, and finally outputs a dynamic holographic ledger that is continuously consistent with the physical world state of the device.

[0078] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0079] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0080] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A digital ledger management system for electronic equipment assets in a smart park, characterized in that, include: The digital twin model generation module is used to create digital twin data models for electronic devices in the park, including static attribute layers, dynamic attribute layers, and spatial attribute layers, and to generate digital twin instance records for each network-connected device. The status acquisition and update module is used to periodically collect the device's runtime metrics through the perception agent software deployed on the edge gateway, and write the metric data into the dynamic attribute layer of the corresponding digital twin instance. The location awareness and synchronization module is used to sense the physical location of the device in real time using the positioning infrastructure. When the deviation between the current location coordinates of the device and the location coordinates recorded in the spatial attribute layer exceeds a preset distance threshold, the location coordinates in the spatial attribute layer are updated to the current location coordinates. The federated collaboration and fusion module is used for each park to maintain its local knowledge graph. It adopts a federated knowledge graph framework for cross-park collaboration. Each park uploads its local model parameters to the cloud after differential privacy processing. The cloud performs federated averaging to generate global model parameters and distributes them to each park. The conflict detection and correction module is used to compare the recorded values ​​of the digital twin instances of each device with the real-time data reported by the perception agent software. When the two are inconsistent, it is marked as a ledger conflict event and correction is performed. The digital twin instance set is output as a dynamic holographic ledger.

2. The smart park electronic equipment asset digital ledger management system according to claim 1, characterized in that, The static attribute layer records the inherent information of the device, including serial number, manufacturer name, purchase date and warranty period; the dynamic attribute layer records the device's runtime indicators, including online status, CPU utilization, memory utilization, signal strength and predicted remaining lifespan; the spatial attribute layer records the device's physical location information, including building number, floor number, room coordinates and business domain identifier.

3. The smart park electronic equipment asset digital ledger management system according to claim 1, characterized in that, The digital twin model generation module further includes: defining the semantic associations between device instances using an ontology language to generate a ledger knowledge graph pattern; wherein, the semantic associations include subordinate relationships and data supply relationships, the subordinate relationship indicating that one device belongs to another device in the network topology, and the data supply relationship indicating that one device provides data input to another device; the subordinate relationship and the data supply relationship are stored in the ledger knowledge graph in the form of directed edges.

4. The smart park electronic equipment asset digital ledger management system according to claim 1, characterized in that, The perception agent software supports data collection using multiple communication protocols: for devices that support active reporting, it receives status indicators periodically pushed by the device via SNMP, MQTT, or Modbus protocols; for dumb terminal devices that do not support active reporting, it obtains status information using active detection methods, including LLDP protocol discovery, ARP address scanning, and ICMP liveness detection. When the network connection between the edge gateway and the cloud is lost, the perception agent software writes the collected indicator data into the local cache queue in timestamp order; when the network connection is restored, the data is retransmitted to the cloud in the time order of the local cache queue, and the cloud writes it into the digital twin instance of the corresponding device in timestamp order.

5. The smart park electronic equipment asset digital ledger management system according to claim 4, characterized in that, The perception agent software also includes an LSTM model deployed at the edge. The processing of the LSTM model includes: standardizing the CPU utilization rate sequence, memory utilization rate sequence and signal strength sequence of a single device within a preset time window and concatenating them into a runtime indicator time series, which is then input into the LSTM model; the output layer of the LSTM model is a fully connected classification layer, and the output is a health anomaly flag of the device, with a value of normal or abnormal. Specifically, when the health anomaly indicator is normal, the perception agent software only uploads the compressed status fingerprint to the cloud; when the health anomaly indicator is abnormal, the perception agent software uploads the abnormal feature vector and the complete time series segment to the cloud together.

6. The smart park electronic equipment asset digital ledger management system according to claim 1, characterized in that, The positioning infrastructure adopts a fusion positioning approach, which integrates at least two positioning methods from Wi-Fi probes, Bluetooth RSSI ranging, UWB tag ranging, and camera visual recognition. Each positioning method outputs a location estimate, and the edge positioning engine performs weighted fusion of multiple location estimates to generate the final device location coordinates. The weight of each positioning method is determined according to the positioning accuracy level of that method in the current environment. The higher the accuracy level, the greater the weight, and the sum of all weights is one. Specifically, when it is determined that the device has moved, a digital twin update event is generated, and the location change record is written to the device's location change history list.

7. The smart park electronic equipment asset digital ledger management system according to claim 1, characterized in that, The location awareness and synchronization module also includes automatic identification and recording of pluggable devices: When a device carrying an NFC electronic tag connects to the edge gateway, the awareness agent software reads the device identification information in its NFC tag. If the device identification does not have a corresponding digital twin instance in the current ledger knowledge graph, a new digital twin instance record is automatically created. When the device does not carry an NFC tag, the awareness agent software generates a device fingerprint based on the device's MAC address, firmware version, and hardware feature parameters, and matches it with the existing device fingerprints in the ledger knowledge graph. If the match is successful, it is associated with an existing digital twin instance; if the match fails, a new digital twin instance record is created.

8. The smart park electronic equipment asset digital ledger management system according to claim 1, characterized in that, The federated knowledge graph framework employs a federated graph neural network algorithm, which includes a local graph encoding unit, a secure gradient aggregation unit, and a global parameter distribution unit. The local graph encoding unit is deployed at the edge nodes of each park. It uses device instances in the local knowledge graph as graph nodes and relationships as graph edges. It standardizes each dynamic attribute field to generate node feature vectors and executes the message passing process of the graph neural network. In each layer, the average of the representations of each node to its neighbor nodes and its own representation is calculated and then processed by the weight matrix and activation function to obtain the updated node representation. After multiple layers of message passing, the fully connected output layer is connected to the node representation of the final layer to output the anomaly probability of each node. The local model gradient is generated using the cross-entropy loss function. The security gradient aggregation unit is deployed in the cloud, receives local model parameters uploaded by each park after differential privacy processing, and calculates the average of the local model parameters of all parks to generate global model parameters. The global parameter distribution unit distributes the global model parameters to the local graph coding units of each park. Each local graph coding unit replaces its local parameters with the global model parameters and continues the next round of training.

9. The smart park electronic equipment asset digital ledger management system according to claim 1, characterized in that, The handling of ledger conflict events includes: when the ledger record shows that a device is online but the sensing agent software does not respond after multiple consecutive probes, the device's status is marked as suspected asset loss or disconnection. If the device recovers its response within a preset time threshold, the mark is removed and the original ledger record is maintained. If it still does not recover after the preset time threshold, a manual inventory task instruction is generated and pushed to the operation and maintenance management terminal. When the spatial attribute of a device in the ledger record is inconsistent with the location result and the location change event has been written, the spatial attribute value in the ledger is automatically overwritten with the location result.

Images