A Digital Twin Model Construction System and Method Based on UE4

Abstract

This application discloses a digital twin model construction system and method based on UE4, relating to the field of computer vision technology. The system includes: a geometric modeling unit, which generates 3D model data based on the geometric modeling data of the target physical entity and achieves dynamic LOD rendering level switching through virtualization processing; a real-time data interaction unit, which parses real-time sensor data streams and generates equipment status data by combining them with transmission operation and maintenance instructions; a simulation engine unit, which performs rigid body motion and collision simulation based on the equipment status data and outputs collision event sequences and state transition data using event triggers; a machine learning unit, which trains an LSTM model based on the collision event sequences and state transition data and optimizes the state covariance parameters through adaptive Kalman filtering; and a visualization interaction unit, which realizes data synchronization between the command screen and mobile terminals through multi-screen collaboration. This application constructs a digital model that can accurately reflect the behavior and characteristics of physical entities.

Description

Technical Field

[0001] This application relates to the field of computer vision technology, and in particular to a digital twin model construction system and method based on UE4. Background Technology

[0002] With the rapid development of smart industry, building digital twin models that can accurately reflect the behavior and characteristics of physical entities has become a core requirement in fields such as smart manufacturing and industrial operation and maintenance.

[0003] Traditional digital twin systems mostly rely on static CAD / BIM models, which can only describe the geometric shape of objects and lack the ability to dynamically represent physical properties. Moreover, existing digital twin systems mostly adopt offline batch processing mode, which is difficult to meet the dynamic data update requirements of real-time interactive scenarios. Therefore, existing digital twin models have the defect of not being able to accurately reflect the behavior and characteristics of physical entities.

[0004] Therefore, constructing a digital model that can accurately reflect the behavior and characteristics of physical entities has practical application value and significance. Summary of the Invention

[0005] In order to construct a digital model that can accurately reflect the behavior and characteristics of physical entities, this application provides a digital twin model construction system and method based on UE4.

[0006] Firstly, the objective of this invention is achieved through the following technical solution:

[0007] A UE4-based digital twin model construction system includes:

[0008] The geometric modeling unit generates 3D model data containing polygonal meshes, topological structures, and UV coordinate systems based on the geometric modeling data of the target physical entity, and achieves dynamic LOD rendering level switching through virtualization processing;

[0009] The real-time data interaction unit parses real-time sensor data streams, combines them with transmission and maintenance instructions, and generates equipment status data through Kalman filtering.

[0010] The simulation engine unit performs rigid body motion and collision simulation based on the device state data, generates environmental effects, and outputs collision event sequences and state transition data using event triggers.

[0011] The machine learning unit trains an LSTM model based on the collision event sequence and state transition data, and optimizes the state covariance parameter through adaptive Kalman filtering;

[0012] The visualization and interaction unit renders a 3D scene and generates a device health heat map, enabling data synchronization between the command screen and mobile terminals through multi-screen collaboration.

[0013] By adopting the above technical solutions, UE4, short for Unreal Engine 4, enables cross-platform real-time 3D creation; it provides a closed-loop architecture based on simulation, learning, visualization, and real-time simulation, realizing self-optimization technology for digital twin systems, which is conducive to building digital models that can accurately reflect the behavior and characteristics of physical entities; specifically, 3D model data serves as the underlying data source, providing spatial anchors for real-time data interaction; the LOD rendering level switching generated by virtualization processing directly affects the rendering load of real-time data interaction; device status data, including sensor data and maintenance instructions, can drive the physical parameters of the simulation engine, enabling real-time data correction of simulation initial conditions; noise suppression data preprocessing through Kalman filtering improves simulation convergence speed; and collision event sequences and state transition data are integrated. As training samples for machine learning, synthetic data generated by the simulation environment compensates for the scarcity of real data (such as simulation of rare fault scenarios), and combined with adaptive Kalman filter parameter optimization, it inversely improves simulation accuracy. Since the model convergence data determines the confidence weight of the visualization rendering, the LSTM prediction results generate a health prediction heatmap (such as visualization of equipment remaining life), which not only improves the accuracy of equipment fault prediction, but also facilitates real-time debugging feedback through the visualization interface. The multi-screen collaboration module transmits the control commands of the mobile terminal in reverse to the real-time data interaction unit. Through bidirectional data flow and driving functions, the reliability of the data binding of the set modeling (such as sensor position mapping and collision area definition) is determined, thereby improving the model accuracy of the digital twin model.

[0014] In a preferred embodiment of this application, the geometric modeling unit further includes:

[0015] Acquire basic geographic information data, process the basic geographic information data, and obtain a first modeling dataset and a second modeling dataset;

[0016] Based on the first modeling dataset and the second modeling dataset, environmental element data related to the digital twin model is obtained, wherein the environmental element data includes first environmental element data and second environmental element data;

[0017] Based on the environmental element data, modeling impact factors representing the model accuracy level are obtained, wherein the modeling impact factors include a first modeling impact factor and a second modeling impact factor;

[0018] Based on the modeling influence factors, the first modeling dataset and the second modeling dataset are divided;

[0019] Obtain the partitioning results, and based on the partitioning results, obtain real-time dynamic parameters.

[0020] By adopting the above technical solutions, it is beneficial to improve the efficiency and accuracy of model construction. This application can effectively distinguish the modeling needs of different regions by performing hierarchical classification processing on basic geographic information data, thereby improving the construction efficiency and spatial expression accuracy of digital twin models. This application introduces modeling influencing factors, which enable the model to adjust the level of detail and data density according to the actual application scenario, thereby enhancing the model's adaptability to complex environmental changes. Furthermore, by obtaining real-time dynamic parameters through the partitioning results, it provides a dynamic update basis for subsequent model optimization, rendering strategy adjustment, and state prediction, thereby improving the intelligence level of the system.

[0021] In a preferred embodiment of this application: the step of acquiring basic geographic information data, processing the basic geographic information data, and obtaining a first modeling dataset and a second modeling dataset includes:

[0022] High-precision terrain data is obtained through drone oblique photography, and structural BIM data is obtained through laser scanning.

[0023] Spatially register the terrain data and BIM data to establish a unified coordinate system;

[0024] Based on the unified coordinate system, obtain the first modeling dataset and the second modeling dataset.

[0025] By adopting the above technical solution, UAV oblique photogrammetry is used to acquire terrain data, and laser scanning is used to acquire BIM structural data. Spatial registration is then performed to establish a unified coordinate system, thereby generating two modeling datasets. This application combines UAV oblique photogrammetry and laser scanning technology to achieve high-precision data acquisition of macroscopic terrain and microscopic structure, providing a reliable data source for high-quality 3D modeling. The establishment of a unified coordinate system through spatial registration ensures the consistency of multi-source heterogeneous data in spatial location, avoiding model misalignment and distortion problems.

[0026] In a preferred example, this application obtains a modeling influence factor representing the model accuracy level based on the environmental element data, including:

[0027] Obtain preset modeling specification parameters, and filter the first environmental element data according to the preset modeling specification parameters to obtain a first modeling parameter set, wherein the first modeling parameter set includes a first geological structure parameter, a first building structure parameter, and a first material texture parameter;

[0028] Based on the first modeling parameter set, obtain the first modeling influence factor;

[0029] According to the preset modeling specification parameters, the second environmental element data is filtered to obtain the second modeling parameter set, wherein the second modeling parameter set includes the second hydrological and meteorological parameters, the second transportation network parameters, and the second sensor parameters;

[0030] Based on the second modeling parameter set, obtain the second modeling influence factor.

[0031] By adopting the above technical solution, it is beneficial to control the accuracy level of the model precisely. This application selects key modeling parameters through modeling specification parameters, which can scientifically evaluate the modeling priority and accuracy requirements of different regions and realize refined hierarchical modeling. The first and second modeling influence factors correspond to the fine structure and lightweight region, respectively, so that the model can not only meet the high-fidelity display of the core region, but also take into account the overall performance and visualization smoothness.

[0032] In a preferred embodiment of this application: the step of dividing the first modeling dataset and the second modeling dataset according to the modeling impact factor includes:

[0033] Based on the first modeling influence factor, establish a LOD500 level fine-grained model partition, and divide the first modeling dataset according to the fine-grained model partition;

[0034] Based on the second modeling influence factor, a lightweight model partition of LOD200 level is established, and the second modeling dataset is divided according to the lightweight model partition.

[0035] By adopting the above technical solution, the model is divided into fine and lightweight partitions, which effectively balances the model accuracy and computing resource consumption, making it suitable for efficient rendering in different terminal devices and interactive scenarios. Under the premise of ensuring visual quality, this application reasonably configures the model complexity, reduces the graphics rendering pressure, and improves the system's operating efficiency and stability.

[0036] In a preferred embodiment of this application: the device status data is a device status matrix containing time-series features, including:

[0037] in, The fusion state matrix; For adaptive weighting coefficients (0.6≤ ≤0.9); Raw sensor data; For transmitting operation and maintenance instruction sets.

[0038] By adopting the above technical solution, the completeness and accuracy of equipment status description are enhanced. This application constructs a status matrix containing sensor data and operation and maintenance instructions, which can comprehensively reflect the equipment's operating status and its external control behavior; and adopts adaptive weighting coefficients (0.6≤ (≤0.9) The data fusion ratio can be dynamically adjusted under different operating conditions to make the state estimation closer to the actual operating conditions and improve the system's perception accuracy.

[0039] In a preferred embodiment of this application: the simulation engine unit executes a collision detection algorithm based on rigid body dynamics, when a collision event occurs:

[0040] The UE4 Niagara system is invoked to generate fluid dynamics effects;

[0041] The event generator outputs state transition data containing collision force vectors.

[0042] Input the collision time series into the LSTM prediction model and dynamically optimize the state covariance matrix:

[0043] in, To estimate the covariance matrix, is a symmetric matrix; The prior covariance matrix; This is the state transition matrix; for The transpose of the matrix; Let be the process noise covariance matrix.

[0044] By adopting the above technical solutions, the UE4 Niagara system generates fluid dynamics effects, enhancing the realism and interactive experience of the digital twin scene. This application outputs state transition data through event triggers, realizing real-time response to sudden physical events and improving the system's ability to identify abnormal operating conditions. This application combines the LSTM model and the covariance matrix update formula to dynamically adjust the Kalman filter parameters, effectively suppressing noise interference and improving the stability and prediction accuracy of state estimation.

[0045] Secondly, the objective of this invention is achieved through the following technical solution:

[0046] A method for constructing a digital twin model based on UE4, the method comprising:

[0047] Using a geometric modeling unit, three-dimensional model data containing polygonal meshes, topological structures, and UV coordinate systems is generated based on the geometric modeling data of the target physical entity. The three-dimensional model data is then virtualized to achieve dynamic LOD rendering level switching.

[0048] The real-time data interaction unit is used to parse the real-time sensor data stream and combine it with the transmission of operation and maintenance instructions to generate equipment status data through Kalman filtering.

[0049] The simulation engine unit performs rigid body motion and collision simulation based on the device state data, generates environmental effects, and outputs collision event sequences and state transition data through event triggers.

[0050] An LSTM model is trained using a machine learning unit based on the collision event sequence and state transition data, and the state covariance parameter is optimized by adaptive Kalman filtering.

[0051] The system utilizes a visual interactive unit to render a 3D scene and generate a device health heatmap. Multi-screen collaboration technology enables data synchronization between the command screen and mobile terminals.

[0052] In a preferred embodiment of this application, the modeling steps of the geometric modeling unit include:

[0053] Acquire basic geographic information data;

[0054] The basic geographic information data is processed to obtain a first modeling dataset and a second modeling dataset.

[0055] Based on the first modeling dataset and the second modeling dataset, environmental element data related to the digital twin model is obtained, wherein the environmental element data includes first environmental element data and second environmental element data;

[0056] Based on the environmental element data, modeling impact factors representing the model accuracy level are obtained, wherein the modeling impact factors include a first modeling impact factor and a second modeling impact factor;

[0057] Based on the modeling influence factors, the first modeling dataset and the second modeling dataset are divided;

[0058] Obtain the partitioning results, and use the partitioning results to obtain real-time dynamic parameters for model optimization and updating.

[0059] By adopting the above technical solutions, the modeling task is decomposed and parameters are screened based on a standardized process, ensuring that the modeling process conforms to industry standards and improving model quality and consistency. By evaluating the influencing factors of modeling, the modeling area and level of detail are reasonably divided to avoid resource waste and improve rendering efficiency and system performance. Based on the division results, this application obtains real-time dynamic parameters, enabling the model to have self-evolution capabilities and adapt to the constantly changing physical entity state and business needs.

[0060] Thirdly, the objective of this invention is achieved through the following technical solution:

[0061] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method for constructing a digital twin model based on UE4.

[0062] In summary, this application includes at least one of the following beneficial technical effects:

[0063] 1. Covering the entire lifecycle management from physical entity modeling, data acquisition, state simulation, intelligent analysis to visualization, forming a complete digital twin closed-loop system, this application constructs a complete closed-loop digital twin system framework. Through the collaborative work of each module, it is conducive to improving the intelligence level and system integration of the digital twin model construction system;

[0064] 2. Standardizing the process for modeling task decomposition and parameter selection helps improve the scientific rigor and standardization of 3D modeling; this application achieves hierarchical modeling and optimal resource allocation through the evaluation of modeling impact factors. Attached Figure Description

[0065] Figure 1 This is a framework diagram of a UE4-based digital twin model construction system according to one embodiment of this application;

[0066] Figure 2 This is a step diagram of a method for a geometric modeling unit in a UE4-based digital twin model construction system according to an embodiment of this application;

[0067] Figure 3 This is a flowchart of a method for constructing a digital twin model based on UE4 in one embodiment of this application. Detailed Implementation

[0068] The present application is further described in detail below with reference to the accompanying drawings.

[0069] In one embodiment, such as Figure 1 As shown, this application discloses a digital twin model construction system based on UE4. The digital twin model construction system includes a geometric modeling unit, a real-time data interaction unit, a simulation engine unit, a machine learning unit, and a visualization interaction unit. The geometric modeling unit generates 3D model data containing polygonal meshes, topological structures, and UV coordinate systems based on the geometric modeling data of the target physical entity. The data is then processed by virtualization to achieve dynamic LOD rendering level switching. The polygonal mesh is a 3D model surface representation composed of vertices, edges, and faces, which determines the geometric shape of the object. The topological structure is used to describe the connection relationship of the object's faces / edges / vertices and is used for collision detection and physical simulation. The UV coordinate system is a 2D texture mapping coordinate system that determines the texture mapping position of the material on the 3D model surface. Dynamic LOD is the real-time switching of the model detail level according to the view distance / performance requirements.

[0070] The real-time data interaction unit parses the real-time sensor data stream, combines it with the transmission of maintenance instructions, and generates equipment status data through Kalman filtering. The real-time data interaction unit interacts with the sensors used to detect equipment status using the Modbus TCP protocol, parses the Modbus TCP protocol, extracts the equipment status code (such as fault code 0x8001), and obtains the transmission of maintenance instructions through the OPC-UA protocol. (In practical applications, it is necessary to set an abnormal threshold for triggering the transmission of maintenance instructions for abnormal equipment status. Different types or categories of abnormal equipment statuses need to be set with corresponding thresholds to immediately detect abnormal equipment statuses and trigger the transmission of maintenance instructions as soon as possible, such as the shutdown threshold setting value.)

[0071] Specifically, the Kalman filtering implementation steps of the real-time data interaction unit include:

[0072] Based on state equations: ,in, and These are n-dimensional vectors at time k and k-1, respectively, containing all the key state variables describing the dynamic characteristics of the system, such as mechanical systems (position, velocity, acceleration, and angle, etc.), equipment monitoring systems (temperature, vibration amplitude, rotational speed, and fault flags, etc.), and digital twin systems (physical entity coordinates, material state, and environmental parameters, etc.). The state transition matrix (including time step Δt = 0.1s) is shown. To control the input matrix, control commands will be sent. Convert to a gain matrix that reflects state changes; This is a vector for operation and maintenance instructions; For process noise; configure the noise covariance matrix. This identifies the process noise variance for two state variables (such as position and state); high The value indicates low model confidence, meaning the filter relies more heavily on sensor measurements; low The value indicates high model confidence, meaning the filter trusts the predicted values ​​more; configure the measurement noise covariance. ,high A value of 1 indicates low sensor accuracy, meaning the filter relies more on model predictions; conversely, a value of 0 indicates high sensor accuracy. Matrix operations are performed using the Eigen library, with an update cycle of 100ms.

[0073] Furthermore, the device status data is a device status matrix containing time-series features, including: in, The fusion state matrix; For adaptive weighting coefficients (0.6≤ ≤0.9); Raw sensor data; For transmitting operation and maintenance instruction sets.

[0074] The simulation engine unit performs rigid body motion and collision simulations based on device state data and generates environmental effects. It uses event triggers to output collision event sequences and state transition data. Rigid body dynamics is used to study the motion laws of objects under force (ignoring deformation). Collision detection is used to determine whether geometric penetration occurs between objects. Event triggers generate asynchronous notifications when specific conditions are detected. Specifically, the physics simulation uses the UE4 Chaos physics system with SIMD acceleration enabled. The integral algorithm for rigid body motion simulation adopts the Verlet integral method with a time step Δt = 0.016s (60FPS). The collision detection frequency is two consecutive collision detections per frame. The CAD model is converted into Convex Collision (convex hull simplification rate ≥ 85%). Complex surfaces are generated using the Sphere Sweep algorithm to generate approximate collision bodies. For example, a car bumper collision body consists of 32 convex hulls, with a total number of facets ≤ 512. The constraint solving algorithm uses the PGS (Progressive Gradient Descent) algorithm to solve contact constraints, with an error tolerance of 1e-5.

[0075] Furthermore, the simulation engine unit executes a collision detection algorithm based on rigid body dynamics when a collision event occurs:

[0076] The UE4 Niagara system is invoked to generate fluid dynamics effects; state transition data containing collision force vectors is output through the event generator; the collision time series is input into the LSTM prediction model to dynamically optimize the state covariance matrix.

[0077] in, To estimate the covariance matrix, is a symmetric matrix; The prior covariance matrix; This is the state transition matrix; for The transpose of the matrix; Let be the process noise covariance matrix.

[0078] When a collision event occurs, if the contact force is greater than 1000N, the event trigger determines the time of the collision, records the coordinates of the collision point (accuracy ±1cm), and outputs the event in JSON format.

[0079] The environmental effects are generated using a particle system, which generates explosion effects based on parameters of 2000 particles / second and a lifespan of 0.5-2 seconds. The environmental effects also include smoke simulation, which uses the SPH (Smooth Particle Hydrodynamics) algorithm for smoke simulation, and collision events are linked to the Wwise audio engine for sound effect simulation.

[0080] The machine learning unit trains an LSTM model based on collision event sequences and state transition data, and optimizes the state covariance parameter through adaptive Kalman filtering. Adaptive Kalman filtering is a filtering algorithm that dynamically adjusts noise parameters according to the system state. It uses recursive least squares (RLS) to update noise statistics with an update period of 5 seconds. The time window length of the collision event sequence is 10 seconds, and the sampling rate is 100 Hz. The state transition data includes device mode switching records. The loss function of the LSTM model is the cross-entropy loss function, the optimizer is Adam (lr=0.001), the epochs are 100, and early stopping is used to monitor the validation set loss.

[0081] The visualization and interactive unit renders a 3D scene and generates a device health heatmap. Multi-screen collaboration enables data synchronization between the command center and mobile terminals. The device health heatmap maps device status parameters to a color gradient map, and multi-screen collaboration ensures synchronized data display between the command center and mobile terminals. Specifically, the heatmap maps the device health index HPI (0-100) to the HSV color space (H=0°~360°), using bilinear interpolation to generate a smooth gradient. The WebGL shader updates the heatmap every frame (frame rate ≥30FPS). Multi-screen collaboration is based on a data synchronization protocol, using WebSocket to transmit JSON data (compression rate ≥60%), combined with client-side prediction algorithms for latency compensation.

[0082] Furthermore, the formula for calculating the Equipment Health Index (HPI) is as follows: ,in, The number of parameters involved in the calculation typically corresponds to the number of key monitoring indicators of the equipment. For parameter index; The parameter baseline value (ideal state or maximum allowable value) is used as a reference standard to measure the current state; The current measured value of the i-th parameter reflects the real-time operating status of the equipment.

[0083] In one embodiment, such as Figure 2 As shown, the geometric modeling unit also includes the following method steps:

[0084] S1: Acquire basic geographic information data, process the basic geographic information data, and obtain the first modeling dataset and the second modeling dataset.

[0085] In this embodiment, the basic geographic information data includes digital data of topography and landform features, such as drone images, satellite images, DEM (digital elevation model), and DOM (digital orthophoto).

[0086] Specifically, step S1 includes:

[0087] S11: High-precision terrain data is obtained through drone oblique photography, and structural BIM data is obtained through laser scanning.

[0088] Specifically, the flight path settings for the UAV include a direct flight path (120m altitude) and an oblique flight path (100m altitude, 70% lateral overlap); the laser scanning strategy includes structure scanning (point cloud density ≥1000 points / m²) and terrain scanning (point cloud density ≥200 points / m²).

[0089] S12: Spatial registration of terrain data and BIM data to establish a unified coordinate system.

[0090] Specifically, the ORB (Oriented FAST and Rotated BRIEF) algorithm is first used to extract key points and image features, followed by point cloud feature matching. The same coordinate system includes at least 10 control points (GCPs) deployed on the ground. Then, a seven-parameter Bursa-Wolf model is used for coordinate system transformation. The transformation parameters are calculated using the following formula:

[0091] in, , , These are the transformed target coordinate parameters; , , ΔX, ΔY, and ΔZ are the original coordinate parameters before transformation; k is the scale parameter; and c is the rotation parameter.

[0092] S13: Obtain the first modeling dataset and the second modeling dataset according to the unified coordinate system.

[0093] Specifically, the first modeling dataset corresponds to the LOD500 level fine model dataset grading standard, which is suitable for core buildings (such as nuclear power plant reactor buildings), with a point cloud density of ≥5000 points / m² and a texture resolution of ≥5mm / pixel; the second modeling dataset corresponds to the LOD200 level lightweight model dataset grading standard, which is suitable for auxiliary facilities (such as plant roads and green belts), with a data requirement of a point cloud density of ≥500 points / m² and a texture resolution of ≥20mm / pixel.

[0094] In this embodiment, the optimization of the dataset and the processing of outlier data include: using a Voxel Grid filter (0.1m voxel size) for noise reduction, performing statistical filtering to remove outliers (standard deviation threshold 3σ), then generating a triangular mesh using the Poisson surface reconstruction algorithm, optimizing the mesh using Laplacian smoothing, and using Spherical Mapping to project the image onto the mesh surface, and performing seamless fusion based on SIFT feature matching to stitch the textures together to improve modeling accuracy.

[0095] Furthermore, spatial registration of multi-source data can be improved using an enhanced Icp-SVD algorithm: first, a normal vector constraint term is added, which improves the registration success rate. The calculation formula is as follows:

[0096] Where R is the rotation matrix, This represents a rigid body transformation, which means transforming the point cloud using only rotation and translation operations while preserving its geometric structure; j is the point cloud data index; Let j be the j-th 3D coordinate point in the target point cloud; Let be the j-th 3D coordinate point in the source point cloud; R is a 3x3 rotation matrix used to rotate the source point cloud to a pose aligned with the target point cloud. The rotation matrix needs to satisfy the conditions of orthogonality (R^TR=I) and determinant of 1. It is a 3×1 translation vector, such as N represents the number of point pairs participating in the registration.

[0097] S2: Based on the first modeling dataset and the second modeling dataset, obtain environmental element data related to the digital twin model, wherein the environmental element data includes the first environmental element data and the second environmental element data.

[0098] In this embodiment, the first environmental element data reflects the environmental parameters of the region's static natural attributes, including geological structure, topographic features, soil type, etc.; the second environmental element data includes environmental parameters of dynamic changes and human intervention, such as hydrological conditions, vegetation coverage, meteorological disaster risk, etc.

[0099] Specifically, the geological structure data is derived from regional geological survey reports. For topographic feature extraction, the data features obtained by fusing DEM data (30m resolution) with UAV oblique photogrammetry DOM (0.5m resolution) are first analyzed. GIS hydrological analysis tools are then used to extract areas with slopes ≥25° for slope classification. Tributary density is calculated based on morphological operations to obtain the topographic valley density. Hydrological conditions are obtained from sensor monitoring data from water level gauges, flow meters, and velocity profilers deployed at key river face sections. Multi-source environmental element data are fused and registered: sensor clocks are calibrated using the NTP protocol for time synchronization, and spatial registration is performed using GCP control points to unify the spatiotemporal reference of multi-source data. Meteorological and ecological data undergo data interpolation and parameter standardization. Meteorological data standardization includes converting meteorological data to the CF (Climate and Forecast) standard format, while ecological data standardization includes mapping ecological data to the ecological red line evaluation index system (GB 35571-2017).

[0100] S3: Based on environmental element data, obtain the modeling impact factors that represent the model accuracy level. The modeling impact factors include the first modeling impact factor and the second modeling impact factor.

[0101] In this embodiment, step S3 includes:

[0102] S31: Obtain preset modeling specification parameters, filter the first environmental element data according to the preset modeling specification parameters, and obtain the first modeling parameter set, wherein the first modeling parameter set includes the first geological structure parameters, the first building structure parameters, and the first material texture parameters.

[0103] In this embodiment, the preset modeling specification parameters are data filtering rules formulated according to the "Technical Standard for Urban Information Modeling (CIM) Basic Platform" (GB / T 38667-2020), which includes accuracy level, coordinate system (CGCS2000) and semantic label specifications.

[0104] Specifically, the specification documents are obtained, and the accuracy levels in the specification documents are converted to the StaticMeshLOD grouping strategy of UE4 (e.g., LOD0 corresponds to a 1cm accuracy model). The first environmental elements are then screened and extracted. The first geological structural parameters are a set of physical characteristic parameters describing the geological structure of the target area, including soil and rock types (soil / rock layer distribution obtained through geological borehole data (e.g., clay, sandstone, gravel layers)), geological age (referring to the tectonic activity classification based on seismic intensity zoning maps (e.g., Quaternary sedimentary layers, Cretaceous bedrock)), foundation bearing capacity (foundation bearing capacity index calibrated through static cone penetration test (CPT) or standard penetration test (SPT)), and groundwater level (extracted by combining aquifer depth and permeability coefficient from a hydrogeological model). The first set of structural parameters characterizes the physical properties of man-made structures, including building type (building function classification (residential / commercial / industrial) parsed from BIM data), structural system (such as load-bearing system parameters such as frame structure / shear wall structure / steel structure), material properties (concrete grade such as C30 / C40, steel yield strength such as Q235 / Q355), and geometric topology (coordinates of beam-column nodes, floor slab thickness, and dimensions of door and window openings).

[0105] S32: Obtain the first modeling influence factor based on the first modeling parameter set.

[0106] In this embodiment, the first modeling influence factor controls the core parameters of the geometric model accuracy, including data integrity (missing rate ≤5%), parameter accuracy (coordinate error ≤0.5m), and topological consistency (faces cannot overlap ≤0.1%).

[0107] Specifically, the PDAL library was used for coordinate system transformation (WGS84→CGCS2000), and the spatial registration method of the ICP algorithm was used to achieve point cloud registration; data integrity was calculated by scoring the point cloud data integrity of geological / architectural / material data to obtain the data missing rate. The data missing rate is greater than 5%, and data integrity is verified by using CloudCompare for point cloud density analysis.

[0108] ,in, The total amount of raw, unprocessed point cloud data (unit: number of points); This represents the actual amount of valid point cloud data used in the UE4 model (unit: number of points). If the data missing rate is >5%, a data replenishment process is triggered (through UAV reflight or lidar rescanning).

[0109] The parameter accuracy was verified using a coordinate fusion algorithm based on Kalman filtering.

[0110] in, For the state estimate at time k (such as coordinate position, velocity, acceleration); Kalman gain; The original sensor observation at time k; The output coordinate error standard deviation σ ≤ 0.3m (confidence level 95%) is the state prediction value based on time k−1.

[0111] Topology consistency can be checked using the CGAL library to detect overlapping surfaces, and the output results can be normalized.

[0112] The formula for calculating the first modeling impact factor is: ,in, 1 is the first influencing factor for modeling; D is data integrity (missing rate ≤ 5%); P is parameter accuracy (coordinate error ≤ 0.5m); T is topological consistency (surface overlap ≤ 0.1%). , , The weights are determined using the AHP (Analytic Hierarchy Process) method (example: =0.42, =0.35, =0.23).

[0113] S33: Based on the preset modeling specification parameters, filter the second environmental element data to obtain the second modeling parameter set, which includes the second hydrological and meteorological parameters, the second transportation network parameters, and the second sensor parameters.

[0114] In this embodiment, the second meteorological data is output through the SWAT model (time resolution 15min), the second traffic network data is extracted with the road centerline through OpenStreetMap (accuracy ≤1m), and the second sensor data is transmitted with vibration monitoring values ​​through the LoRaWAN protocol (frequency 10Hz).

[0115] Specifically, the second set of hydro-meteorological parameters is a dynamic environmental data set describing the hydrological cycle and meteorological characteristics of the target area, including rainfall, river water level (river water level change curves based on hydrological models), wind speed and direction (near-surface wind field data with a 10-minute resolution), and snow cover (winter snow thickness and distribution). The second set of transportation network parameters is a dynamic set of parameters characterizing the real-time operating status of the road network, including vehicle data, average speed, traffic light timing, and accident hotspots. The second set of sensor parameters is a set of engineering parameters used to describe the physical characteristics and deployment scheme of monitoring equipment, including installation coordinates, sampling frequency, data type of detection, measurement range, and data transmission standards.

[0116] S34: Obtain the second modeling influence factor based on the second modeling parameter set.

[0117] Specifically, the second modeling influencing factor is a set of parameters that affect the accuracy of dynamic simulation, including real-time performance (data update frequency ≥ 1Hz), coverage (sensor blind zone ≤ 10%), and noise ratio (signal-to-noise ratio ≥ 20dB).

[0118] Specifically, ,in, The second modeling influencing factor is R, which represents real-time performance (data update frequency ≥ 1Hz), C represents coverage (sensor blind zone ≤ 10%), and N represents noise ratio (signal-to-noise ratio ≥ 20dB). , and The weights are determined using grey relational analysis (example: =0.38, =0.32, =0.30).

[0119] In this embodiment, real-time performance is evaluated using end-to-end latency testing:

[0120] T_delay = T_sensor + T_edge + T_cloud, where T_delay is the end-to-end delay time; T_sensor is the sensor data acquisition time; T_edge is the edge computing node data processing time; and T_cloud is the cloud data transmission and processing time.

[0121] Coverage was analyzed and quantified using Voronoi diagram coverage: ,in, The area effectively covered by the sensor; The target area is defined as its geometric area. The DBSCAN clustering algorithm is used to identify weak signal areas (spatial distance threshold = 5m, minimum number of neighborhood points of the core point = 5) to determine the detection blind zone. QoS-tiered transmission is implemented via the MQTT protocol (traffic data priority P=2, meteorological data P=1).

[0122] S4: Based on the modeling impact factors, divide the first modeling dataset into the second modeling dataset.

[0123] In this embodiment, step S4 includes:

[0124] S41: Based on the first modeling impact factor, establish a LOD500 level fine-grained model partition, and divide the first modeling dataset according to the fine-grained model partition.

[0125] In this embodiment, the LOD500 level fine model partition is a model area that meets the accuracy requirements of a 1:100 scale (such as precision mechanical parts, geological fault zones), with a polygonal facet number ≥10^6, and includes detailed information such as material textures and normal maps.

[0126] Specifically, based on the first modeling impact factor score, LOD500 level fine model regions are divided. A dynamic partitioning threshold is established to divide fine models of different fineness levels. The dynamic partitioning threshold is a partition boundary parameter that is automatically adjusted based on the modeling impact factor score. For example, regions with a geological structure integrity score ≥ 0.8 are classified into LOD500 partitions.

[0127] S42: Based on the second modeling influence factor, establish a LOD200 level lightweight model partition, and divide the second modeling dataset according to the lightweight model partition.

[0128] Specifically, based on the score of the second modeling influence factor, LOD200 level refined model regions are divided; LOD200 level lightweight model regions are model regions that meet the accuracy requirements of a 1:500 scale (such as urban roads and building outlines), with ≤10^4 polygon faces, and employing mesh simplification algorithms (such as Quadric Error Metrics); in practical applications, a dynamic adjustment mechanism for the digital twin model and high-priority regions also need to be established to achieve model lightweighting while improving model refinement. High-priority regions, such as areas with a geological structure score ≥0.8, are forcibly classified into LOD500 regions; the dynamic adjustment mechanism automatically upgrades the LOD level of local areas when sensors detect geological activity (such as earthquake waveform amplitude > threshold).

[0129] In this embodiment, partition conflict arbitration also needs to be established.

[0130] When the boundaries of LOD500 and LOD200 partitions are inconsistent, the following arbitration rules are activated: if the first modeling impact factor score is greater than 0.7, it is classified as a LOD500 level refined model; if the second modeling impact factor score is less than 0.4, it is classified as a LOD200 level refined model; otherwise, a LOD300 level refined model is set up to model the partition conflict area, and a partition modeling command is triggered based on the modeling information and conflict situation of the partition conflict area to prompt the user with modeling details.

[0131] S5: Obtain the partitioning results and, based on the partitioning results, obtain real-time dynamic parameters.

[0132] In this embodiment, the partitioning result is the model partitioning result (LOD500 / LOD200); the real-time dynamic parameters are a set of system parameters that are dynamically adjusted according to the model partitioning result (LOD500 / LOD200), such as the rendering level switching threshold, sensor fusion weight, and collision detection sensitivity.

[0133] Specifically, dynamic parameter vectors are set, such as sensor weights and filter coefficients; and partition weights are associated with LOD500 and LOD200 respectively. Then, sensor weight adjustment adopts sensor weighted fusion based on partition priority. For example, sensors in high-precision areas are given higher weights. The adjustment of Kalman filter covariance is as follows: the process noise covariance for high-precision requirements is reduced, and the process noise covariance for low-precision requirements is increased. The rendering level switching threshold adopts the dynamic LOD switching distance threshold calculated according to priority: the higher the priority (LOD500), the closer the switching distance (precision priority).

[0134] In one embodiment, such as Figure 3 As shown, a method for constructing a digital twin model based on UE4 is provided. This method is applied to a UE4-based digital twin model construction system. The specific steps of the UE4-based digital twin model construction method are as follows:

[0135] S10: Using the geometric modeling unit, generate 3D model data containing polygonal meshes, topological structures and UV coordinate systems based on the geometric modeling data of the target physical entity, and perform virtualization processing on the 3D model data to achieve dynamic LOD rendering level switching.

[0136] S20: Utilizes the real-time data interaction unit to parse real-time sensor data streams, and combines them with transmission maintenance instructions to generate equipment status data through Kalman filtering.

[0137] S30: Utilizes the simulation engine unit to perform rigid body motion and collision simulation based on device state data, generates environmental effects, and outputs collision event sequences and state transition data through event triggers;

[0138] S40: Use machine learning units to train an LSTM model based on collision event sequences and state transition data, and optimize the state covariance parameter through adaptive Kalman filtering;

[0139] S50: Utilizes a visual interactive unit to render a 3D scene and generate a device health heatmap, and achieves data synchronization between the command screen and mobile terminals through multi-screen collaboration technology.

[0140] In this embodiment, the modeling steps of the geometric modeling unit include:

[0141] Acquire basic geographic information data;

[0142] Process the basic geographic information data to obtain the first modeling dataset and the second modeling dataset;

[0143] Based on the first modeling dataset and the second modeling dataset, environmental element data related to the digital twin model is obtained, wherein the environmental element data includes the first environmental element data and the second environmental element data.

[0144] Based on environmental element data, modeling impact factors representing the model accuracy level are obtained, including a first modeling impact factor and a second modeling impact factor.

[0145] Based on the modeling impact factors, the first modeling dataset and the second modeling dataset are divided.

[0146] Obtain the partitioning results and use the real-time dynamic parameters obtained from the partitioning results for model optimization and updating.

[0147] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0148] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0149] S10: Using the geometric modeling unit, generate 3D model data containing polygonal meshes, topological structures and UV coordinate systems based on the geometric modeling data of the target physical entity, and perform virtualization processing on the 3D model data to achieve dynamic LOD rendering level switching.

[0150] S20: Utilizes the real-time data interaction unit to parse real-time sensor data streams, and combines them with transmission maintenance instructions to generate equipment status data through Kalman filtering.

[0151] S30: Utilizes the simulation engine unit to perform rigid body motion and collision simulation based on device state data, generates environmental effects, and outputs collision event sequences and state transition data through event triggers;

[0152] S40: Use machine learning units to train an LSTM model based on collision event sequences and state transition data, and optimize the state covariance parameter through adaptive Kalman filtering;

[0153] S50: Utilizes a visual interactive unit to render a 3D scene and generate a device health heatmap, and achieves data synchronization between the command screen and mobile terminals through multi-screen collaboration technology.

[0154] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0155] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0156] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A digital twin model construction system based on UE4, characterized in that, include: The geometric modeling unit generates 3D model data containing polygonal meshes, topological structures, and UV coordinate systems based on the geometric modeling data of the target physical entity, and achieves dynamic LOD rendering level switching through virtualization processing; The real-time data interaction unit parses real-time sensor data streams, combines them with transmission and maintenance instructions, and generates equipment status data through Kalman filtering. The simulation engine unit performs rigid body motion and collision simulation based on the device state data, generates environmental effects, and outputs collision event sequences and state transition data using event triggers. The machine learning unit trains an LSTM model based on the collision event sequence and state transition data, and optimizes the state covariance parameter through adaptive Kalman filtering; The visualization and interaction unit renders a 3D scene and generates a device health heat map, enabling data synchronization between the command screen and mobile terminals through multi-screen collaboration; The geometric modeling unit also includes: Acquire basic geographic information data, process the basic geographic information data, and obtain a first modeling dataset and a second modeling dataset; Based on the first modeling dataset and the second modeling dataset, environmental element data related to the digital twin model is obtained, wherein the environmental element data includes first environmental element data and second environmental element data; Based on the environmental element data, modeling impact factors representing the model accuracy level are obtained, wherein the modeling impact factors include a first modeling impact factor and a second modeling impact factor; Based on the modeling influence factors, the first modeling dataset and the second modeling dataset are divided; Obtain the partitioning results, and based on the partitioning results, obtain real-time dynamic parameters; Based on the environmental element data, modeling influence factors representing the model accuracy level are obtained, including: Obtain preset modeling specification parameters, and filter the first environmental element data according to the preset modeling specification parameters to obtain a first modeling parameter set, wherein the first modeling parameter set includes a first geological structure parameter, a first building structure parameter, and a first material texture parameter; Based on the first modeling parameter set, obtain the first modeling influence factor; According to the preset modeling specification parameters, the second environmental element data is filtered to obtain the second modeling parameter set, wherein the second modeling parameter set includes the second hydrological and meteorological parameters, the second transportation network parameters, and the second sensor parameters; Based on the second modeling parameter set, obtain the second modeling influence factor; The device status data is a device status matrix containing time-series features, including: S(t)=α×S sen (t)+(1-α)×S cmd (t) Where S(t) is the fusion state matrix; α is the adaptive weight coefficient (0.6≤α≤0.9); S sen (t) represents the raw sensor data; S cmd (t) represents the transmission operation and maintenance instruction set.

2. The digital twin model construction system based on UE4 according to claim 1, characterized in that, The process of acquiring basic geographic information data and processing the basic geographic information data to obtain a first modeling dataset and a second modeling dataset includes: High-precision terrain data is obtained through drone oblique photography, and structural BIM data is obtained through laser scanning. Spatially register the terrain data and BIM data to establish a unified coordinate system; Based on the unified coordinate system, obtain the first modeling dataset and the second modeling dataset.

3. The digital twin model construction system based on UE4 according to claim 1, characterized in that, The step of dividing the first modeling dataset and the second modeling dataset according to the modeling influence factor includes: Based on the first modeling influence factor, establish a LOD500 level fine-grained model partition, and divide the first modeling dataset according to the fine-grained model partition; Based on the second modeling influence factor, a lightweight model partition of LOD200 level is established, and the second modeling dataset is divided according to the lightweight model partition.

4. The digital twin model construction system based on UE4 according to claim 1, characterized in that, The simulation engine unit executes a collision detection algorithm based on rigid body dynamics. When a collision event occurs: The UE4Niagara system is invoked to generate fluid dynamics effects; The event generator outputs state transition data containing collision force vectors. Input the collision time series into the LSTM prediction model and dynamically optimize the state covariance matrix: Where, ∑ k|k To estimate the covariance matrix, is a symmetric matrix; ∑ k-1|k-1 F is the prior covariance matrix; k This is the state transition matrix; For F k The transpose of Q; k Let be the process noise covariance matrix.

5. A method for constructing a digital twin model based on UE4, characterized in that, The method includes: Using a geometric modeling unit, three-dimensional model data containing polygonal meshes, topological structures, and UV coordinate systems is generated based on the geometric modeling data of the target physical entity. The three-dimensional model data is then virtualized to achieve dynamic LOD rendering level switching. The real-time data interaction unit is used to parse the real-time sensor data stream and combine it with the transmission of operation and maintenance instructions to generate equipment status data through Kalman filtering. The simulation engine unit performs rigid body motion and collision simulation based on the device state data, generates environmental effects, and outputs collision event sequences and state transition data through event triggers. An LSTM model is trained using a machine learning unit based on the collision event sequence and state transition data, and the state covariance parameter is optimized by adaptive Kalman filtering. The system utilizes a visual interactive unit to render a 3D scene and generate a device health heatmap. It also uses multi-screen collaboration technology to achieve data synchronization between the command screen and mobile terminals. The modeling steps of the geometric modeling unit include: Acquire basic geographic information data; The basic geographic information data is processed to obtain a first modeling dataset and a second modeling dataset. Based on the first modeling dataset and the second modeling dataset, environmental element data related to the digital twin model is obtained, wherein the environmental element data includes first environmental element data and second environmental element data; Based on the environmental element data, modeling impact factors representing the model accuracy level are obtained, wherein the modeling impact factors include a first modeling impact factor and a second modeling impact factor; Based on the modeling influence factors, the first modeling dataset and the second modeling dataset are divided; Obtain the partitioning results, and obtain real-time dynamic parameters based on the partitioning results for model optimization and updating; Based on the environmental element data, modeling influence factors representing the model accuracy level are obtained, including: Obtain preset modeling specification parameters, and filter the first environmental element data according to the preset modeling specification parameters to obtain a first modeling parameter set, wherein the first modeling parameter set includes a first geological structure parameter, a first building structure parameter, and a first material texture parameter; Based on the first modeling parameter set, obtain the first modeling influence factor; According to the preset modeling specification parameters, the second environmental element data is filtered to obtain the second modeling parameter set, wherein the second modeling parameter set includes the second hydrological and meteorological parameters, the second transportation network parameters, and the second sensor parameters; Based on the second modeling parameter set, obtain the second modeling influence factor; The device status data is a device status matrix containing time-series features, including: S(t)=α×S sen (t)+(1-ɑ)×S cmd (t) Where S(t) is the fusion state matrix; α is the adaptive weight coefficient (0.6≤α≤0.9); S sen (t) represents the raw sensor data; S cmd (t) represents the transmission operation and maintenance instruction set.

6. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the UE4-based digital twin model construction method as described in claim 5.

Images