Urban infrastructure digital twinning system based on live-action three dimensions

Abstract

The invention discloses an urban infrastructure digital twinning system based on live-action three dimensions, and relates to the technical field of digital twinning, and the system comprises a data fusion module, a data credibility evaluation module and an augmented reality guide module. The data credibility evaluation module calculates a quantitative credibility grade for the multi-source heterogeneous data. The augmented reality guiding module can adaptively select the guiding mode corresponding to the grade to display after obtaining the target infrastructure component and the data credibility grade associated with the target infrastructure component. Under the high credibility level, the system provides accurate point guidance; and under the low credibility level, switching to a fuzzy region range indication or high-risk warning mode. According to the method, invisible data quality risks are converted into differentiated and visual AR visual guidance, so that safety accidents caused by accurate execution of wrong instructions are avoided, and the safety of field operation and maintenance and the reliability of decision making are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of digital twinning, in particular to a city infrastructure digital twinning system based on real scene three dimensions. BACKGROUND

[0002] With the development of information technology, the city infrastructure digital twinning system based on real scene three dimensions has become an important direction of smart city construction. Such a system constructs a three-dimensional city model corresponding to the physical world through technologies such as unmanned aerial vehicle oblique photography and laser scanning, and fuses multi-source information such as Internet of Things (IoT) sensors and business data, aiming to realize visualized management and decision support for the whole life cycle of city infrastructure.

[0003] At the same time, augmented reality (AR) technology, as a key bridge connecting the digital world and the physical world, has been widely used in field operation and maintenance scenarios. Existing AR operation and maintenance applications are generally committed to improving the visual alignment accuracy of digital information (such as equipment drawings, pipeline paths, and operation instructions) and the real physical environment, in order to provide "what you see is what you get" intuitive guidance for front-line operation and maintenance personnel, thereby improving work efficiency.

[0004] However, the technical development in this field in the process of pursuing high-precision three-dimensional modeling and high-precision AR visual alignment is generally based on an implicit but often difficult to establish premise in engineering practice: that is, the underlying digital twinning data serving as the source of AR guidance information is completely accurate and reliable. In actual city infrastructure management, due to the diversity of data sources (for example, design drawings of different years, as-built drawings after construction changes, and multiple survey data), the timeliness of data updates, and the inherent errors of measurement methods, there is often a spatial position deviation between the digital model and the physical reality that cannot be completely eliminated.

[0005] When such a high-precision display tool is used to present data with uncertainty, a deep technical contradiction arises. The system transmits a kind of "accurate illusion" to front-line operation and maintenance personnel, hiding the uncertainty risk of underlying data under the seemingly accurate visual effect. Such misleading accurate instructions are prone to cause safety accidents due to "accurately executing incorrect instructions" in high-risk operation scenarios (for example, excavation repair of gas pipe network and operation of high-voltage power). SUMMARY

[0006] The application provides a real scene three-dimensional based urban infrastructure digital twin system, which aims to solve the technical problems of misleading instructions and potential operation risks caused by high-precision augmented reality (AR) guidance of uncertain data in the prior art.

[0007] In view of the above problems, the application provides a real scene three-dimensional based urban infrastructure digital twin system, which comprises: a data fusion module configured to collect and process multi-source data to construct and update a digital twin model of urban infrastructure, wherein the digital twin model at least comprises a real scene three-dimensional model recording geometric information and spatial position of infrastructure components; a data credibility evaluation module configured to calculate a data credibility level for part of the multi-source data; an augmented reality guidance module configured to display digital information related to a target infrastructure component in an augmented reality manner on a mobile terminal, wherein the augmented reality guidance module is further configured to: obtain the data credibility level associated with the target infrastructure component; and select an augmented reality guidance mode according to the data credibility level to render the digital information.

[0008] Preferably, the augmented reality guidance module is configured to: if the data credibility level meets a first preset condition, a first guidance mode is selected, in which the digital information is displayed in the form of a three-dimensional virtual object indicating the spatial position of the target infrastructure component; if the data credibility level meets a second preset condition, a second guidance mode is selected, in which the digital information is displayed in the form of a virtual area with a preset spatial range covering the target infrastructure component.

[0009] Preferably, the augmented reality guidance module is further configured to: if the data credibility level meets a third preset condition, a third guidance mode is selected, in which a warning information containing a risk prompt text is displayed on the user interface of the mobile terminal.

[0010] Preferably, the data credibility evaluation module is configured to evaluate the data from at least one of the following dimensions to calculate the data credibility level, including data source, spatial accuracy, data timeliness, and data integrity.

[0011] Preferably, the system further comprises: a physical simulation engine configured to perform simulation calculation on physical responses of one or more infrastructure components under preset load conditions based on the digital twin model; wherein the physical simulation engine is further configured to adjust the model parameters according to the data credibility level associated with the model parameters when performing the simulation calculation.

[0012] Preferably, the system further comprises: an artificial intelligence-based analysis model configured to analyze or predict the operating state of the infrastructure based on the data contained in the digital twin model; wherein the training samples of the artificial intelligence analysis model are weighted according to the data credibility level associated with the sample data when the artificial intelligence analysis model is trained.

[0013] Preferably, the artificial intelligence analysis model adopts a graph neural network architecture to abstract the infrastructure components and their connection relationships into a graph structure, and performs the analysis or prediction based on the graph structure.

[0014] The technical scheme provided by the application has at least the following technical effects or advantages: The application establishes a data credibility evaluation mechanism, and forcibly and adaptively associates the evaluation result with the front-end AR guidance mode, thereby solving the problem of high-risk operation caused by blindly trusting uncertain data. When the system identifies that the underlying data has low credibility, it no longer provides accurate point guidance that may mislead, but switches to a fuzzy area indication or high-risk warning mode. This converts the invisible data quality risk into visible, differentiated, and behaviorally meaningful visual interaction, which can effectively prevent safety accidents caused by accurate execution of incorrect instructions.

[0015] The application enables front-line operation and maintenance personnel and back-end management personnel to intuitively and quantitatively perceive the credibility of each piece of data in the digital twin system. For data with high credibility, personnel can make quick decisions and operations with confidence, thereby shortening the operation time; for data with low credibility, the system explicitly prompts the risk and guides personnel to adopt a more prudent operation process that prioritizes physical review, thereby avoiding ineffective labor and repetitive work caused by inaccurate information. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 Figure 1 is a schematic diagram of a city infrastructure digital twin system architecture based on real three-dimensional data according to the application. DETAILED DESCRIPTION

[0017] To make the purposes, technical solutions and advantages of the present application clearer and more complete, a preferred embodiment of the present application based on the whole life cycle of the city comprehensive pipe gallery and the key municipal facilities (such as bridges and water supply networks) will be described below. It should be understood that the specific embodiments described herein are intended to illustrate the core technical ideas and specific implementation paths of the present application in an engineering end-to-end scenario, rather than any form of limitation on the protection scope of the present application. The system realizes accurate perception of the state of urban infrastructure, early prediction of risks, and intelligent support for operation and maintenance decisions by constructing a closed-loop workflow that combines high-fidelity mapping of the physical world, deep fusion of multi-source heterogeneous data, physical simulation deduction, and adaptive guidance of data credibility.

[0018] Please refer to Figure 1 A city infrastructure digital twin system based on real three-dimensional, the system comprises: A data fusion module configured to collect and process multi-source data to construct and update a digital twin model of urban infrastructure, wherein the digital twin model at least contains a real three-dimensional model recording the geometric information and spatial position of infrastructure components; A data credibility evaluation module configured to calculate a data credibility level for a part of the multi-source data; An augmented reality guidance module configured to display digital information related to a target infrastructure component on a mobile terminal in an augmented reality manner, Wherein, the augmented reality guidance module is further configured to: Obtain the data credibility level associated with the target infrastructure component; And according to the data credibility level, select an augmented reality guidance mode to render the digital information.

[0019] The construction and operation of the system starts from a data collection and preprocessing stage covering the whole process, and the goal is to obtain high-precision, multi-modal raw data required for constructing a digital twin model.

[0020] In one embodiment, the system first performs a collaborative collection of multi-source data for a target operation and maintenance area, such as a five-kilometer-long urban comprehensive pipe gallery and its surrounding surface environment. The system plans and executes a three-dimensional collection scheme composed of unmanned aerial vehicles, ground mobile measurement vehicles, and fixed Internet of Things sensors. Among them, an industrial-grade unmanned aerial vehicle equipped with a five-lens tilt camera (for example, the effective pixels of the central camera can be no less than forty-five million, and the effective pixels of the four tilt cameras can be no less than two million five hundred thousand) and a high-density laser radar scanner (for example, the scanning frequency can be no less than three hundred thousand points per second) flies at a constant ground speed (for example, five meters per second) aiming to balance efficiency and data quality according to a three-dimensional route optimized by a terrain simulation software. To ensure the geometric accuracy and texture continuity of subsequent three-dimensional reconstruction, the forward overlap rate of the image should be no less than eighty percent, and the lateral overlap rate should be no less than seventy percent when planning the route. At the same time, one or more ground mobile measurement vehicles operate synchronously along the road or inspection path above the pipe gallery. The measurement vehicle is integrated with a panoramic camera, a dual-frequency global navigation satellite system receiver, and a higher line number ground laser radar (for example, one hundred twenty-eight lines) for collecting high-precision point cloud data and image data of the ground view of the area blocked by the unmanned aerial vehicle view, and the absolute positioning accuracy of the point cloud.

[0021] In a preferred embodiment, it can be controlled within three centimeters after joint adjustment with pre-set control points on the ground. Inside the pipe gallery and at the predetermined monitoring points of key infrastructure (such as bridge supports, main cables, key valves of water supply pipe network, and pressurized pump stations), a large number of diversified Internet of Things sensors are deployed in advance, such as distributed fiber grating sensors for monitoring structural stress and strain, static water levels for monitoring micro-displacement, sensors for monitoring environmental temperature and humidity, sensors for monitoring fluid pressure and flow inside the pipeline, and high-fidelity pickups for capturing abnormal sound. The data collection frequency of these sensors is a key parameter that can be adaptively adjusted according to the rate of change of the monitored physical quantity, for example, for high-frequency vibration monitoring of bridges under vehicle load, the sampling frequency can be set to one hundred hertz, and for environmental temperature, which is a slowly varying parameter, the sampling frequency can be set to once per minute.

[0022] All raw data collected on site, including massive three-dimensional point cloud and image data generated by unmanned aerial vehicles and ground vehicles, and time series data streams generated by Internet of Things sensors, are transmitted back to the data processing center in real time or quasi-real time through a high-bandwidth, low-latency hybrid network architecture.

[0023] In one network deployment scenario, the fifth generation mobile communication network (5G) is preferred for backbone data transmission; in areas with poor signal coverage, such as deep pipe corridors or underground spaces, it is automatically switched to a pre-configured industrial wireless local area network or low-power wide-area network (e.g., LoRaWAN) as a supplementary link. To ensure the integrity and reliability of data during transmission, a lightweight edge computing gateway is deployed at the edge (e.g., a mobile measurement vehicle or a fixed aggregation node close to the data collection source). The gateway has a built-in distributed message queue service (e.g., an Apache Kafka cluster consisting of three nodes) with high throughput and persistence capability, which performs peak shaving and buffer storage on all incoming raw data streams. Before entering the queue, the data is automatically encapsulated into a standardized data packet, which contains not only the raw data payload but also a series of key metadata, including the unique identifier of the data source device, the batch number of the collection task, the precise collection timestamp conforming to the International Organization for Standardization (ISO) 8601 standard with time zone information, and a monotonically increasing sequence number to ensure data packet order. The edge computing gateway also performs one or more immediate and lightweight data cleaning and preprocessing tasks, aiming to filter out obvious noise data and reduce the burden of subsequent transmission and storage.

[0024] Specifically, the preprocessing procedure can include: for raw image data, applying a frequency domain filtering algorithm based on fast Fourier transform to remove periodic noise, or applying a lightweight convolutional neural network model to enhance image clarity; for massive raw point cloud data, performing a downsampling algorithm based on voxel grid, where the edge length of the voxel grid is a key parameter that can be flexibly configured according to the accuracy requirements of subsequent application scenarios, for example, it can be set to ten centimeters in urban macro scenario modeling, and reduced to one centimeter in component-level fine modeling; for time series data generated by sensors, a moving window-based outlier rejection algorithm is applied, for example, to reject isolated data points that deviate from the mean value of the window by more than three standard deviations, or to use a more robust isolation forest algorithm to identify small clusters of abnormal data. The preprocessed data is securely pushed to the central data storage platform deployed in the cloud or local data center in a unified, self-describing data format (e.g., Apache Avro or ProtocolBuffers).

[0025] At the central data platform, the system performs a rigorous, multi-dimensional quality assessment and coordinate system alignment process for all ingested data. The system automatically assesses the completeness, logical consistency, timeliness, and spatial accuracy of the data, in reference to geospatial metadata standards. For example, completeness can be assessed by checking whether the actual number of files received in a collection batch matches a predefined list; logical consistency can be checked by verifying that the value range of sensor data is within the reasonable range specified in the device manual. If any of the assessment indicators falls below a pre-defined quality threshold (e.g., data packet completeness is below 99%), the system automatically generates a data quality anomaly work order and assigns it to the data collection team through an integrated project management system for root cause analysis and re-collection. Subsequently, the system enters the unified spatial reference processing stage. All data with spatial attributes, including 3D point clouds, camera pose parameters of oblique imagery, deployment coordinates of IoT sensors, and subsequent imported building information models (BIM) or design drawings with geographic reference points, are uniformly converted to a high-precision, regionally appropriate, city-independent coordinate system or national geodetic coordinate system. For building information models from diverse sources, which often use relatively independent local coordinate systems, the system provides a semi-automatic registration tool based on multi-point affine transformation. Operators only need to pick up at least three homonymous points (e.g., corner points of a building or base centers of streetlights) on the building information model and high-precision point clouds or real-world maps, and the system can automatically calculate the optimal transformation matrix (including rotation, translation, and scaling parameters) to accurately map the building information model to the unified geospatial coordinate system. The final mapping root mean square error is required to be controlled within five centimeters in a typical embodiment.

[0026] After data cleaning and spatial alignment, the system begins the task of multi-source heterogeneous data deep fusion and digital twin 3D model construction. This process aims to organically integrate data from different sources and modalities, forming a digital twin that is geometrically accurate, texturally realistic, semantically rich, and physically complete. First, the system performs an automatic matching process based on geometry and semantics, fusing the structured information of building information models (BIM) with the unstructured information of real-world 3D point clouds. Specifically, the system traverses each component in the building information model (e.g., a beam, a column, a section of pipe) to obtain its precise geometric bounding box. Then, using an optimized spatial indexing structure (e.g., 3D R-tree or K-dimensional tree), it quickly retrieves all point cloud subsets falling within the bounding box from the vast amount of point cloud data. By calculating the average distance between the point cloud subset and the surface of the building information model component, as well as the coverage rate of the point cloud, the matching degree is determined. If the matching is successful, the real color or laser radar reflectivity intensity information carried by the point cloud can be mapped as texture to the corresponding building information model component surface to compensate for its appearance defects.

[0027] Further, for the objects that are not included in the BIM but actually exist in the field (e.g. newly added equipment, surrounding trees or temporary piles), the system can use point cloud clustering algorithms (e.g. DBSCAN algorithm) to segment them out and automatically or semi-automatically create new geometric models and semantic labels for them, thus achieving dynamic updating and enrichment of the original design model. In order to ensure visual fidelity while meeting the requirement of smooth real-time rendering on different terminal devices (from high-performance workstations to mobile augmented reality glasses), the system uses multi-level of detail (LOD) technology to organize the fused model. The system automatically generates models at multiple levels from LOD0 to LOD4: LOD0 is a full-factor model containing all original point clouds and the highest precision geometric information, used for offline high-precision engineering analysis; LOD1 to LOD3 gradually reduce the number of model facets and texture resolution through a series of mesh simplification algorithms (e.g. quadratic error metric method) and texture compression techniques (e.g. block-based texture compression), respectively suitable for near, medium and far real-time rendering; LOD4 may be an extremely simplified block model that only retains the main outline, used for city-level bird's eye view. When rendering, the system will automatically and seamlessly switch between different LOD levels according to the distance between the virtual camera and the model object.

[0028] On the basis of geometric and visual model construction, the system further integrates a physical simulation engine. The simulation engine can be based on a variety of mature numerical calculation methods, for example, for structural mechanics analysis, finite element method can be used; for fluid dynamics analysis, computational fluid dynamics can be used.

[0029] In a preferred embodiment applied to bridge structural health monitoring, the system first extracts the geometric models of key load-bearing components such as beams, slabs, cables, piers, etc. from the fused digital twin model, and automatically or with the assistance of engineers, generates high-quality finite element mesh models. The material properties (e.g. elastic modulus, Poisson's ratio) of the models can be inherited from the associated building information model or design documents. Then, the system loads real-time or simulated load cases, such as real-time traffic volume and vehicle distribution analyzed from traffic monitoring camera data, or strong wind load and temperature stress simulated from weather forecast data. By solving large-scale partial differential equations, the simulation engine can calculate the stress, strain, and displacement response contours of the bridge under various loads, and visualize them in 3D overlaid on the digital twin model. For gas or water supply pipelines in a utility tunnel, the system can build fluid dynamics models to simulate pressure distribution, flow rate changes, and the diffusion range and concentration distribution of leaked substances under different operating conditions such as normal operation or pipeline damage. In a high-requirement embodiment, the accuracy of the simulation results is required to be within 5% of the actual monitoring data.

[0030] To achieve comprehensive monitoring and intelligent analysis of facility operation status, the system establishes a mechanism that strongly binds physical facilities, digital twin models, and real-time sensor data. The system gives each independent infrastructure component in the real world (from a valve, a street lamp, to a bridge, a tunnel) a unique digital identity through a globally unique identifier generation mechanism. In the background graph database, each component exists as a node, and its attributes include static design information, materials, suppliers, historical maintenance records, and dynamic associations to real-time data. For example, a node representing "No. 2 water pump in X pump station" is connected to nodes representing its 3D model, its corresponding unit in the physical simulation model, and all IoT sensor nodes monitoring its operating status (such as vibration, temperature, current). This graph structure storage method facilitates fast, multi-level association information queries from any data point.

[0031] The system introduces a set of data credibility adaptive guidance and artificial intelligence decision optimization mechanisms. The system conducts periodic and automated credibility assessment on all real-time and historical data entering the platform at a configurable time granularity (e.g., every hour). The credibility is quantified as a comprehensive index between zero and one, which is calculated by weighting the evaluation scores of multiple sub-dimensions. In a preferred embodiment, these sub-dimensions can include: the authority of the data source (e.g., an officially calibrated sensor scores 1.0, a user's mobile phone uploaded estimated data scores 0.4), spatial location accuracy (different scores are given according to their positioning methods, such as differential global positioning system or base station positioning), data time freshness (e.g., data within one minute scores 1.0, data more than one day old scores exponentially decayed), data integrity and consistency (evaluated by comparing with historical data patterns or physical laws), and user feedback credibility from operation and maintenance personnel (cross-verified data credibility scores by multiple senior engineers will be improved). The weight coefficients of each dimension can be calibrated by system administrators according to specific application scenarios (e.g., in emergency rescue scenarios, the weight of time freshness will be dynamically adjusted). This credibility index is deeply integrated into every aspect of the system: in the three-dimensional visualization interface, low credibility models or data will be rendered with semi-transparent, dashed lines or specific warning colors, with a prompt "data not verified, for reference only"; when performing physical simulation, model parameters can be adjusted according to the credibility of their source data, for example, using high-credibility material strength data for accurate calculation, while low-credibility load data is simulated by Monte Carlo simulation to evaluate its uncertainty range; when training artificial intelligence models for fault prediction or anomaly detection, each training sample is assigned its corresponding credibility index as a sample weight, which makes the model pay more attention to the rules contained in high-quality, high-credibility data during the learning process, and automatically ignore or reduce the impact of low-credibility noise data, thereby improving the robustness and accuracy of artificial intelligence decision-making.

[0032] Based on the above credibility-weighted, high-quality fusion data set, the system constructs one or more artificial intelligence analysis models to realize the transition from passive response to active prediction in operation and maintenance mode.

[0033] In a preferred embodiment for pipe network leakage prediction, the system can employ a deep learning architecture based on graph neural networks. The model abstracts the entire pipe network system as a complex graph structure, where each sensor, valve, and pipe segment is a node in the graph, and their physical connections and logical relationships form the edges of the graph. The input features of the model include not only the real-time and historical time series data of each node itself (such as pressure, flow, and sound wave signals), but also the state information of its neighbor nodes aggregated through the message passing mechanism of graph neural networks. This architecture enables the model to learn the complex spatial correlation patterns of fault propagation in the pipe network. The model learns an end-to-end mapping from the overall state of the pipe network to the leakage probability of each pipe segment by training on a large amount of historical normal operation data and a small number of real fault data (also weighted by credibility). The output of the model is a dynamically updated risk heat map of the entire pipe network, which visually highlights the high-risk pipe segments that are most likely to leak in the near future (e.g., within twenty-four hours). When the predicted leakage probability of any pipe segment exceeds a pre-set alarm threshold (e.g., eighty-five percent), the system automatically generates a high-priority warning work order.

[0034] The system extends the capabilities of digital twin models to the field through a mobile operation terminal integrated with augmented reality (AR) technology. When the operation personnel wear AR smart glasses or hold an industrial tablet computer to the scene, the simultaneous localization and mapping (SLAM) algorithm on the device, combined with the fusion of high-precision differential global positioning system signals, can achieve centimeter-level accurate registration of the model and the real world. In the field of view of the operation personnel, underground pipelines, equipment in walls, and various invisible facilities can be displayed in three-dimensional perspective in the real environment. The daily inspection task list can be navigated in the form of virtual road signs to guide personnel to each inspection point. At the equipment side, relevant real-time monitoring data, historical maintenance records, standard operation procedure (SOP) documents, and three-dimensional disassembly animations can be called out and suspended in the air through voice or gesture instructions. When an anomaly is found, the operation personnel can directly create a virtual marker circle at the scene location through voice instructions (e.g., saying "create a level three water seepage fault work order here"), and the system can automatically create a structured electronic work order containing accurate geographic coordinates, first-person perspective photos or videos, and voice-to-text descriptions. When encountering complex problems, a remote expert collaboration session can be initiated with one click. The remote expert can see the first-person perspective of the on-site personnel in real time and can perform three-dimensional annotations, project drawings, or precise indications with a virtual laser pointer to guide the on-site personnel's operation.

[0035] All the early warning and on-site reporting are finally integrated into a closed-loop work order and task intelligent scheduling management module. The system supports multiple work order generation modes and sorts and dispatches all pending work orders according to a dynamic, multi-factor priority sorting algorithm. In a preferred embodiment, the algorithm is a comprehensive scoring model whose input factors include: the severity level of the fault, the importance level of the affected business, the real-time geographic location of the maintenance personnel and the distance to the pending point, the matching degree of the personnel's skill certification and the task demand, and the current task saturation of the personnel. The system calculates the matching score of each "personnel-task" combination through a weighting function configurable by the manager, and automatically recommends or assigns the task to the personnel or team with the highest score. The whole process of task execution, its state (such as "has accepted the order", "has departed", "is processing", "has completed") will be synchronized in real time on the AR terminal and the center platform. A fault is not considered "closed-loop" by the system until the review personnel confirm that the relevant monitoring indicators have returned to normal and this confirmation information is recorded.

[0036] Various alternative embodiments and extended application scenarios of the present application also fall within the scope of protection of the present application. For example, in the selection of artificial intelligence models, in addition to graph neural networks, for some specific time series anomaly detection tasks, models based on long short-term memory networks (LSTM) or Transformer architecture can also be used. Those skilled in the art can understand that the selection of these models can be adjusted according to the characteristics of the specific task and the dimension of the data. In terms of data security, the system can use distributed ledger technology based on blockchain to record key maintenance operations (such as major equipment start and stop, permission changes) in an unalterable manner to enhance the audit security of the system. The system architecture and core methodology described in the present application are not only suitable for the city comprehensive pipe gallery and municipal facilities, but also can be widely extended and applied to smart parks, intelligent manufacturing factories, large airports, ports, data centers and other maintenance management fields with complex physical assets and dense data streams after appropriate industry knowledge base and model adaptation.

Claims

1. A real-scene three-dimensional based digital twin system of urban infrastructure, characterized in that, The system comprises: a data fusion module configured to collect and process multi-source data to build and update a digital twin model of a city infrastructure, wherein the digital twin model at least comprises a real-scene three-dimensional model recording geometric information and spatial position of infrastructure components; a data credibility assessment module configured to calculate a data credibility level for a part of the multi-source data; an augmented reality guidance module configured to display digital information related to a target infrastructure component on a mobile terminal in an augmented reality manner, wherein the augmented reality guidance module is further configured to: obtain the data credibility level associated with the target infrastructure component; and select an augmented reality guidance mode to render the digital information according to the data credibility level.

2. A real-scene three-dimensional based digital twin system of urban infrastructure as claimed in claim 1, wherein, The augmented reality guidance module is configured to: if the data credibility level meets a first preset condition, select a first guidance mode in which the digital information is displayed in the form of a three-dimensional virtual object indicating the spatial position of the target infrastructure component; if the data credibility level meets a second preset condition, select a second guidance mode in which the digital information is displayed in the form of a virtual area with a preset spatial range covering the target infrastructure component.

3. A real-scene three-dimensional based digital twin system of urban infrastructure as claimed in claim 1, wherein, The augmented reality guidance module is further configured to: if the data credibility level meets a third preset condition, select a third guidance mode in which an alert information containing a risk prompt text is displayed on the user interface of the mobile terminal.

4. The real-scene three-dimensional based digital twin system of urban infrastructure according to claim 1, wherein, The data credibility assessment module is configured to evaluate the data from at least one of the following dimensions to calculate the data credibility level, including: data source, spatial accuracy, data timeliness, and data integrity.

5. A real-scene three-dimensional based digital twin system of urban infrastructure as claimed in claim 1, wherein, The system further comprises: a physical simulation engine configured to perform simulation calculation on the physical response of one or more infrastructure components under a preset load condition based on the digital twin model; wherein the physical simulation engine is further configured to adjust the model parameters according to the data credibility level associated with the model parameters during the simulation calculation.

6. A real-scene three-dimensional based digital twin system of urban infrastructure as claimed in claim 1, wherein, The system further comprises: an artificial intelligence-based analysis model configured to analyze or predict the operating state of the infrastructure based on the data contained in the digital twin model; wherein the training samples of the artificial intelligence analysis model are weighted according to the data credibility level associated with the sample data during training.

7. A real-scene three-dimensional based digital twin system of urban infrastructure as claimed in claim 1, wherein, The artificial intelligence analysis model uses a graph neural network architecture to abstract the infrastructure components and their connection relationships into a graph structure and performs the analysis or prediction based on the graph structure.

Images