A digital twin water conservancy scene dynamic building and scheduling system and method

Abstract

The application discloses a kind of digital twin water conservancy scene dynamic building and scheduling system and method, comprising: data processing module is used to obtain standard geospatial data service and standard three-dimensional model data service;Preliminary scene building module is used to obtain preliminary scene according to standard geospatial data service in the scene template standard three-dimensional model data service data;Scene rendering scheduling module is used to obtain triangular mesh data after vector data is obtained by asynchronous computing thread, the triangular mesh data obtained by main thread is framed and pushed to rendering thread to execute rendering operation, obtains basic scene;Scene editing and storage module is used to edit and structured storage to basic scene.The application is identified by AI remote sensing, asynchronous rendering scheduling and structured storage, solves the problems that traditional digital twin water conservancy scene construction exists low customization development efficiency, scene is difficult to reuse and large-scale data rendering lag, improves water conservancy management efficiency and decision-making accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of digital twinning, in particular to a digital twinning water conservancy scene dynamic building and scheduling system and method. BACKGROUND

[0002] Digital twinning water conservancy is a digital virtual body formed by mapping and modeling all factors, all aspects and multi-scale of physical water conservancy systems such as river basins and water conservancy projects through new generation information technologies such as Internet of Things, cloud computing, big data and artificial intelligence, and interacting with physical systems in real time and dynamically fusing. It can perceive the state of physical water conservancy systems, simulate and predict their trends, and provide intelligent decision support for water conservancy management and governance.

[0003] The digital twinning scene is built by using geographic spatial data and BIM models, and visualized by using high-performance visualization engines as a mapping of physical space. At present, most digital twinning systems are developed in a customized manner, and the scene is built manually. This method is not only tedious, time-consuming and labor-intensive, but also susceptible to human factors, resulting in inaccurate scene elements. Moreover, the results are not reusable and prone to repeated development, wasting manpower. SUMMARY

[0004] The purpose of the present application is to provide a digital twinning water conservancy scene dynamic building and scheduling system and method. The present application solves the problems of low efficiency of customized development, high cost and poor accuracy of manual building, difficulty in reusing scene results, and lag in rendering large-scale data in traditional digital twinning water conservancy scene construction through data standardization services, programmatic automatic generation, asynchronous multi-threaded computing and high-performance rendering optimization, and structured storage management based on scene graphs.

[0005] To achieve this purpose, the present application designs a digital twinning water conservancy scene dynamic building and scheduling system, which comprises:

[0006] The data processing module is used to preprocess the collected geographic spatial data and three-dimensional model data, and then publish the preprocessed geographic spatial data and three-dimensional model data as standard geographic spatial data services and standard three-dimensional model data services;

[0007] The preliminary scene building module is used to select a preset scene template according to business requirements, build a scene baseplate in the scene template based on standard geographic spatial data services, identify and classify digital orthophoto maps through AI remote sensing image recognition technology, access the scene template with the classified results corresponding to the feature model of the scene material library and the BIM model loaded based on the standard three-dimensional model data services, and obtain a preliminary scene.

[0008] The scene rendering scheduling module is used to acquire vector data through a standard geospatial data service, sequentially complete coordinate resolution and triangulation by an asynchronous computing thread, and push the obtained triangulation data to a rendering thread by a main thread in frames, so that the rendering thread performs a rendering operation on the basis of a preliminary scene according to the triangulation data, and a basic scene is obtained.

[0009] The scene editing and storage module is used to edit attributes and styles in the basic scene, and store scene information in a structured manner according to a tree structure configuration file specification.

[0010] The beneficial effects of the application are as follows:

[0011] 1. The application provides a system for dynamically building and scheduling a digital twin water conservancy scene, adopts an AI to build a basic scene and a manual fine-tuning mode, realizes dynamic and rapid building of a water conservancy digital twin three-dimensional scene, improves building efficiency, and reduces labor and time costs. The scene file specification is constructed to realize scene editing, saving and loading, scene reuse, and reduction of repeated development work.

[0012] 2. The data processing module is used to perform unified time and space reference and format standardization processing on multi-source heterogeneous geospatial data and three-dimensional model data, and publish the data based on an open geospatial information consortium standard service, so that the data interoperability and integration efficiency are enhanced, and multi-source data fusion and cross-platform calling are supported.

[0013] 3. The distributed time and space database is used in combination with an R-tree or quadtree spatial index to significantly improve the storage and retrieval performance of large-scale geospatial data, and effectively support real-time scheduling and visualization of massive data.

[0014] 4. The three-dimensional model data is used to construct a level of detail (LOD) model and perform lightweight processing, so that efficient transmission and rendering of large-volume models are realized, and visual effects and system performance are taken into account.

[0015] 5. The preliminary scene building module is used in combination with an AI remote sensing image recognition technology to automatically identify and classify ground objects, and intelligently match scene materials, so that the automation degree and accuracy of scene construction are greatly improved, and the cost of manual modeling is reduced.

[0016] 6. The scene rendering scheduling module is used to process vector data triangulation by an asynchronous computing thread, and push the data to a rendering thread by a main thread in frames, so that interface lag is effectively avoided, and system responsiveness and user experience are improved.

[0017] 7. Technologies such as view frustum culling, GPU instantiation and LOD are comprehensively used in the rendering process, so that GPU load is significantly reduced, and rendering frame rate and the fluency of large-scale scenes are improved.

[0018] 8. The scene editing and storage module provides flexible attribute and style editing functions, and adopts a structured tree-shaped JSON configuration file specification to support scene version management, rapid recovery and sharing reuse, and improve system maintainability and expandability;

[0019] 9. The application constructs a full-process solution from data access, intelligent construction, efficient rendering to visual editing and storage, has good engineering applicability and promotion value, and is especially suitable for complex application scenarios such as smart water conservancy and digital twin river basin. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 A module workflow diagram of the digital twin water conservancy scene dynamic building and scheduling system of the application;

[0021] Figure 2 A scene data preparation and preprocessing flowchart for an embodiment of the application;

[0022] Figure 3 A scene building flowchart for an embodiment of the application;

[0023] Figure 4 A scene editing and saving flowchart for an embodiment of the application;

[0024] Figure 5 A scene rendering and scheduling flowchart for an embodiment of the application;

[0025] Figure 6 A scene organization structure diagram for an embodiment of the application. DETAILED DESCRIPTION

[0026] The application will be further described in detail below in combination with the drawings and specific embodiments: Embodiment 1

[0027] As shown in Figures 1-6 A digital twin water conservancy scene dynamic building and scheduling system includes:

[0028] The data processing module is used for preprocessing the collected geographic space data and three-dimensional model data respectively, and publishing the preprocessed geographic space data and three-dimensional model data as standard geographic space data services and standard three-dimensional model data services;

[0029] The preliminary scene building module is used for selecting a preset scene template according to a business requirement, constructing a scene baseplate in the scene template based on the standard geographic space data services, identifying and classifying a digital orthographic image through an AI remote sensing image recognition technology, accessing a ground object model of the scene material library corresponding to the classification result and a BIM model loaded based on the standard three-dimensional model data services into the scene template, and obtaining a preliminary scene.

[0030] The scene rendering scheduling module is used to obtain vector data through a standard geospatial data service, sequentially complete coordinate resolution and triangulation by an asynchronous computing thread, and then push the obtained triangulation data to a rendering thread by a main thread, so that the rendering thread performs a rendering operation on the basis of a preliminary scene according to the triangulation data, and a basic scene is obtained.

[0031] The scene editing and storage module is used to edit attributes and styles in the basic scene, and store scene information in a structured manner according to a tree structure configuration file specification.

[0032] The application constructs a complete digital twin water conservancy scene dynamic building and scheduling system, realizes full-process automation, high efficiency and standardization from data to application. The data processing module in the system is responsible for cleaning, standardizing and publishing the original data of multiple sources and heterogeneous data as a unified standard service, solving the problem of data fusion and providing high-quality and easy-to-call data basis for subsequent modules; the preliminary scene building module adopts an AI automatic recognition and preset template mode, quickly builds a scene base based on a standard service, then intelligently classifies ground objects by using an AI remote sensing image recognition technology, and automatically matches and accesses models and BIM models in a material library, so as to quickly generate a preliminary preliminary scene. This greatly improves the automation level and efficiency of scene construction, and liberates people from heavy manual modeling work; the scene rendering scheduling module solves the performance bottleneck problem caused by large-scale vector data rendering, performs time-consuming coordinate resolution and triangulation calculation by an asynchronous computing thread, and then pushes the calculation results to a rendering thread by a main thread. This multi-thread cooperative scheduling mechanism effectively avoids interface lag, guarantees the smoothness and real-time interactivity of the system under large data, and finally outputs a smooth basic scene. Finally, the scene editing and storage module provides flexible post-editing functions, and stores the scene information in a structured manner according to a tree structure configuration file specification, which not only meets the personalized customization needs of users for scene element attributes and styles, but also more importantly realizes the serialization, reusability and shareability of scene information, greatly reduces repeated development work, and improves the maintainability and expandability of the project.

[0033] In the above technical solution, the method for publishing the preprocessed geospatial data and three-dimensional model data as a standard geospatial data service and a standard three-dimensional model data service after preprocessing the collected geospatial data and three-dimensional model data respectively is as follows:

[0034] The process of preprocessing the geospatial data and publishing the data is as follows:

[0035] Unifying the time and space reference of geospatial data, including time reference, space reference and height reference, obtaining geospatial data obtained by unifying the time and space reference;

[0036] Standardizing the format of the geospatial data obtained by unifying the time and space reference, obtaining preprocessed geospatial data;

[0037] Storing the preprocessed geospatial data according to the geospatial position by using a distributed time and space database, and storing time as a data attribute, obtaining a geospatial database;

[0038] Respectively constructing R-tree or quadtree space index for each partition in the geospatial database;

[0039] Publishing the geospatial data as an open geospatial information consortium standard service;

[0040] The open geospatial information consortium standard service is abbreviated as OGC standard service.

[0041] The process of preprocessing and data publishing of the three-dimensional model data is:

[0042] Performing lightweight processing on the three-dimensional model data by using a facet reduction or vertex thinning method, constructing a level of detail model, and standardizing the format of the three-dimensional model data in the level of detail model, obtaining preprocessed three-dimensional model data;

[0043] Dividing the preprocessed three-dimensional model data according to the level in the level of detail model, publishing the three-dimensional model with a large volume order as a 3DTiles format service, and publishing the lightweight three-dimensional model in the form of a file.

[0044] The above scheme describes a specific implementation method of the data processing module, which ensures the consistency, efficient management and efficient service of massive and multi-source data. For geospatial data, the misalignment and deviation between different source data are eliminated by unifying the time reference, spatial reference and elevation reference, laying a solid foundation for building accurate digital twins. The distributed spatio-temporal database is used for storage by geographical spatial location, combined with R-tree or quad-tree spatial index, which greatly optimizes the storage efficiency and retrieval speed of massive spatio-temporal data, enabling fast query and call of data by region or range. Publishing data as OGC standard services ensures the openness and interoperability of data, which can be widely called by various clients and platforms. For three-dimensional model data, face reduction or vertex thinning is used for lightweight processing and construction of level of detail (LOD) model, which is a key technology to ensure smooth loading and rendering of large-scale three-dimensional scene. It automatically calls models of different precision according to the viewpoint distance, significantly reducing the rendering pressure of GPU and network transmission load. According to the model volume, services are published in 3DTiles format, and light weight is published in file format. 3DTiles format is designed for streaming large amounts of three-dimensional geographic data, supporting LOD and block loading, and realizing efficient and smooth transmission and visualization of massive three-dimensional models in network environment.

[0045] Preferably, the time reference adopts Beijing time, the spatial reference adopts CGCS2000 coordinate system, and the elevation reference adopts 1985 national elevation reference.

[0046] In the above technical solution, the process of standardizing the format of geospatial data and three-dimensional model data of the data processing module is:

[0047] The geospatial data includes digital orthophoto map, digital elevation model, digital surface model, oblique photography model, laser point cloud, underwater topography, and vector data.

[0048] The digital orthophoto map is abbreviated as DOM; the digital elevation model is abbreviated as DEM; and the digital surface model is abbreviated as DSM.

[0049] The digital orthophoto map, digital elevation model and digital surface model adopt tagged image file format data; and the vector data is converted into SHP or GeoJSON format data.

[0050] The tagged image file format data is abbreviated as TIFF format data.

[0051] The BIM model is converted into 3DTiles format data.

[0052] The three-dimensional model data includes landscape model and character model.

[0053] The three-dimensional model data is converted into FBX or OBJ format data.

[0054] The above scheme further refines the format standardization of geospatial data and three-dimensional model data, and realizes seamless integration and efficient processing of multi-source heterogeneous data through unified format specifications. Various data types supported (such as DOM, DEM, DSM, laser point cloud, BIM, vector, etc.) and their corresponding standardized formats. Digital orthophoto map (DOM), digital elevation model (DEM), and digital surface model (DSM) are unified into TIFF format, which can well store geographic reference information and high-bit-depth pixel data, and is the standard raster data format in the field of geographic information system (GIS), ensuring the accuracy and universality of the data. Vector data is converted into SHP or GeoJSON format, SHP is the most widely supported vector format by GIS software, and GeoJSON is very suitable for data transmission and analysis in Web environment, both of which meet the application requirements of desktop and Web. Converting BIM model to 3DTiles format is a key step to integrate BIM and GIS, 3DTiles format enables BIM model, which is originally used for fine component expression, to be integrated into large-scale geographic scene in the form of spatial index and LOD, realizing the organic integration of macro geographic environment and micro building information. Three-dimensional model data (landscape, character model) is converted into FBX or OBJ format, which is a common exchange format in the field of three-dimensional modeling and game engine. These two formats support complex materials, animations and skin information, ensuring that the model can be correctly and high-quality rendered in the rendering engine. Preferably, the digital orthophoto map, digital elevation model and digital surface model of the data processing module can be published as a tile map service.

[0055] The tile map service is referred to as TMS.

[0056] The above scheme gives specific geographic spatial data for digital orthophoto map (DOM), digital elevation model (DEM) and digital surface model (DSM) data, which can be published as a tile map service, and the transmission efficiency and display performance of the raster data service can be optimized. The tile map service is a standard tile map service protocol, and its working principle is to cut the large raster data into a large number of small size tile pictures (such as 256x256 pixels) under different scales (LOD) in advance. When the client requests data, it only needs to dynamically load the required tiles according to the range and zoom level of the current map view. This method has great advantages compared to loading the entire large image at once, greatly reducing the amount of data transmitted over the network and speeding up the data loading speed; it realizes progressive loading and rendering of data, providing a smooth user experience; it allows caching of tiles to further improve the efficiency of repeated access; and since the tiles are static pictures, the rendering pressure on the server side is greatly reduced. Therefore, using the tile map service to publish the core raster data is a key technical means to ensure that the digital twin large-scale scene can be quickly and smoothly visualized.

[0057] In the above technical solution, the method for obtaining the preliminary scene in the preliminary scene building module is:

[0058] Obtain a preset scene template library and a preset scene material library;

[0059] According to the business requirements, select a scene template from the preset scene template library;

[0060] According to the business requirements, if the twin range is a large-scale river basin level, select a digital twin river basin template, if it is an engineering or park level, select a corresponding template in the digital twin engineering template and the digital twin park template.

[0061] Index the geographic spatial data service according to the latitude and longitude range information of the preset scene template to obtain the geographic spatial data of the scene;

[0062] Construct a scene floor within the scene template using the digital orthophoto map, digital elevation model and digital surface model in the geographic spatial database obtained by the standard geographic spatial data service;

[0063] Extract features from the digital orthophoto map through AI remote sensing image recognition technology, identify and classify the features according to the features, obtain a classification result, match the classification result with the corresponding ground object model in the preset scene material library, and connect the ground object model in the preset scene material library to the scene template;

[0064] Load and call the BIM model obtained by the standard three-dimensional model data service in a streaming loading mode, connect the BIM model to the scene template, and complete the preliminary scene building.

[0065] The above process describes the completion process of the preliminary scene building module, which realizes the templating, automation and intelligentization of scene construction, and significantly improves the building efficiency. Through the preset scene template library and the preset scene material library, the template library provides the initial framework (such as latitude and longitude range, initial camera position, etc.) of different business scenes (such as river, reservoir, city flood control), realizes the rapid start and standardization of different scene building. According to the template range index geographic space data service constructs the scene bottom plate on the scene template, ensures the accurate matching of the scene bottom plate and the business demand, and the data is dynamically obtained, which guarantees the real-time of the scene bottom plate. Through the AI remote sensing image recognition technology, the digital orthographic image is extracted and classified, and the AI is introduced to automatically realize the work of manual modeling and visual interpretation. The system can intelligently identify water, vegetation, roads, buildings and other ground object categories. Using the classification result to match the corresponding ground object model in the preset scene material library realizes the leap from recognition to generation, automatically places the symbolic three-dimensional model (such as standardized trees, house models) to the correct geographical position, greatly reducing the workload of manual layout. Finally, the BIM model is loaded in a streaming manner, ensuring that large BIM models can be loaded in blocks as needed without blocking the main thread, and smoothly integrated into the macro scene.

[0066] The preset scene template library and the preset scene material library are existing templates and ground object models pre-constructed in the system.

[0067] Preferably, the scene template includes: a digital twin river basin template, a digital twin engineering template, and a digital twin park template.

[0068] Preferably, the preset scene material library includes basic open geographic space data and basic three-dimensional model data of water conservancy and hydropower projects.

[0069] Preferably, by connecting real-time shared data such as ecological environment and weather into the scene template, the preliminary scene changes with the real-time updated shared data.

[0070] The convolutional neural network in the field of deep learning is used for remote sensing image classification and recognition. The two-dimensional convolution formula is as follows, where O is the output matrix, I is the input matrix, w is the convolution kernel, kh and kw are the height and width of the convolution kernel, respectively, representing the size of the convolution kernel. The convolution kernel is the core component of the convolutional neural network, which is used to extract features from input data. The size of the convolution kernel is generally 1x1, 3x3, 5x5, etc.

[0071] ;

[0072] where kh represents the height of the convolution kernel and kw represents the width of the convolution kernel. element in the i-th row and j-th column of the matrix.

[0073] ;

[0074] ;

[0075] where, denotes the height of the output feature map, denotes the width of the output feature map, H denotes the height of the input picture, W denotes the width of the input picture, Ph denotes the number of padding in the height direction, Pw denotes the number of padding in the width direction, padding refers to padding elements on both sides of the input height and width, usually 0 elements, Sh stride in the height direction, Sw denotes the stride in the width direction, stride refers to the number of rows and columns that the convolution window slides on the input array each time.

[0076] The process of feature extraction and ground object classification of digital orthophoto map using AI remote sensing image recognition technology is as follows:

[0077] Different ground objects are distinguished by their spectral characteristics in different wavebands, and classified, for example:

[0078] Vegetation has high reflectivity in the near-infrared band, and healthy vegetation usually appears bright red;

[0079] Water absorbs strongly in the near-infrared band, appearing dark;

[0080] Buildings have high reflectivity in the visible light band, usually appearing gray or white.

[0081] Different ground objects are distinguished by their spatial characteristics such as shape, size and texture, and classified, for example:

[0082] Shape and size: buildings usually have regular geometric shapes, roads are linear, and farmland is block-shaped.

[0083] Texture: texture roughness is analyzed by gray level co-occurrence matrix (GLCM). For example, the texture of forests is usually rough, while the texture of water bodies is smooth.

[0084] Spatial relationship: combined with contextual information, such as rivers adjacent to bridges, vegetation along roads, etc.

[0085] The result of feature extraction and ground object classification and recognition of digital orthophoto map using AI remote sensing image recognition technology is as follows: buildings, vegetation, roads, water systems, and bridges.

[0086] Preferably, spectral indices (such as vegetation index NDVI, water index NDWI) can be calculated to enhance the characteristics of specific features.

[0087] In the technical solution, the asynchronous computing thread of the rendering module acquires the triangular mesh data and pushes the triangular mesh data to the rendering thread through frame division, and the process of the rendering thread performing the rendering operation is as follows:

[0088] The vector data of the geographic space is acquired by using a standard geographic spatial data service, and the vector data is transmitted to the asynchronous computing thread;

[0089] The asynchronous computing thread performs coordinate analysis on the vector data, and performs triangular meshing calculation on the line vector data and the surface vector data based on the analysis result, to generate triangular mesh data;

[0090] The main thread listens to the state of the asynchronous computing thread in real time, and when it is found that the asynchronous computing thread has completed the calculation, the triangular mesh data is acquired;

[0091] The main thread pushes the acquired triangular mesh data to the rendering thread through frame division;

[0092] The rendering thread performs the rendering operation on the basis of the preliminary scene according to the received triangular mesh data, to obtain a basic scene. The triangular mesh data that has not been pushed by the main thread within the current rendering frame time is continuously pushed in the subsequent rendering frame until all the data is pushed.

[0093] The above process describes a multi-thread cooperation mechanism for processing vector data in the scene rendering scheduling module, which solves the performance bottleneck problem in large-scale vector data visualization and ensures the fluency and interactivity of the system. Vector data (such as river boundaries and water conservancy facility distribution maps) usually contains a large number of coordinate points, and its rendering needs to be calculated by triangular meshing first, which converts lines and polygons into triangular patches that can be drawn by GPU. This process requires a huge amount of calculation, and if it is completed in the main thread or rendering thread, it will inevitably block user interaction and cause interface lag. By designing an asynchronous calculation thread to perform time-consuming coordinate analysis and triangular meshing calculation, the main thread and rendering thread are freed from heavy calculation tasks, allowing them to continuously respond to user operations and perform rendering output, maintaining the fluency of the UI. The main thread listens to the state of the asynchronous thread in real time and obtains the results after the calculation is completed, which reflects the efficient communication and synchronization between threads. The main thread pushes the obtained triangular mesh data to the rendering thread frame by frame, rather than feeding all the large amount of mesh data calculated at once to the GPU. The data is split and submitted gradually in multiple rendering frames, avoiding GPU instantaneous overload and frame rate drop caused by excessive data volume in a single frame. This asynchronous calculation and frame-by-frame submission scheduling strategy is a key technical guarantee for smooth and non-lagging visualization of large volume of vector data in three-dimensional scenes.

[0094] The triangular mesh data includes triangular vertex coordinates, indices, UV coordinates, and normal information.

[0095] In the above technical solution, during the execution of the rendering operation by the rendering thread, at least one of the following is used to reduce the GPU rendering load: view frustum culling, a graphics processor instance, and a level of detail model.

[0096] The above process is a further optimization scheme for the rendering process, which greatly reduces the rendering load of the GPU through various graphics techniques, further improves the rendering performance and frame rate of large-scale complex scenes. The view frustum culling can ensure that the GPU only renders the objects within the current camera view range (frustum), and directly skips the rendering process for objects outside the view. This technique has a significant effect on large-scale water conservancy scenes, as the user can only see a small part of the entire scene at any time. This technique can instantly eliminate most of the invisible models and triangular faces, greatly reducing the number of primitives that the GPU needs to process. The GPU Instancing technique targets a large number of repeated objects in the scene (such as trees, railings, bolts, etc. of the same type), allowing the GPU to render multiple identical objects using a single draw call, only passing different position, scaling, and other transformation information. This avoids initiating an independent draw call for each repeated object, greatly reducing the communication overhead between the CPU and the GPU and improving rendering efficiency. The Level of Detail (LOD) model can automatically switch between different precision models based on the distance between the object and the camera. When the distance is far, a coarse model with fewer faces is displayed, and when the distance is close, a fine model with more faces is displayed. This keeps the number of faces that the GPU needs to process within a reasonable range while maintaining visual effects. These three techniques (usually used in combination) optimize the rendering pipeline from three dimensions: eliminating invisible objects, optimizing draw calls, and reducing the number of faces per model. They are the core technology combination for maintaining high-performance and high-frame-rate rendering of large-scale digital twin scenes.

[0097] In the above technical solution, the method for editing and storing the basic scene in the scene editing and storage module is:

[0098] The display style of the scene baseplate in the basic scene is edited.

[0099] The ground object model and BIM model in the basic scene are translated, rotated, scaled, and edited for attribute parameters.

[0100] The material, color, texture, and data coordinates of the spatial vector element when the rendering thread performs a rendering operation are edited.

[0101] According to the preset scene configuration file specification, the structured information of the edited basic scene is stored in the service-side scene database or local file.

[0102] The scene configuration file specification is based on a JSON format serialized scene graph structure, organized using a tree-shaped scene graph structure, including a root node, multiple intermediate group nodes, and multiple scene entity elements as leaf nodes.

[0103] The scene entity elements correspond to the ground object model, BIM model, and spatial vector element in the scene.

[0104] The space vector element refers to a geographical entity represented in the form of point, line, and surface vector data.

[0105] The scene editing and storage module provides powerful scene customization capabilities and realizes the structuring, persistence, and reusability of scene information. This module allows users to conduct comprehensive and fine editing on the constructed basic scene, including modifying the display style (such as color tone and brightness) of the scene baseboard to adapt to different themes; adjusting the terrain model and BIM model for translation, rotation, scaling, and attribute parameters, including material, shape, and position information, to make them more consistent with the actual situation or highlight specific information; editing the visual representation and data of space vector elements to achieve dynamic effects or data-driven visualization. In addition, this module stores based on the scene configuration file specification of the tree-shaped scene graph structure in JSON format. JSON format is a lightweight, easy-to-read and write text format with good cross-platform compatibility. The tree-shaped scene graph structure is a very efficient organization method, with the root node as the entry, the middle group nodes logically organizing scene elements (such as putting all trees into one group and all houses into another group), and the leaf nodes corresponding to specific entity elements (terrain and BIM model). This structure has clear levels and is very convenient for program traversal, management, and serialization. Storing the scene in such a structured manner in the database or file allows the entire scene state to be saved and reproduced completely, realizing true scene reuse, and users can continue editing on the basis of previous work or quickly apply the same scene template to different projects, greatly improving development efficiency, reducing maintenance costs, and facilitating team collaboration and version management.

[0106] As shown in Figure 2 , it is the scene data preparation and preprocessing stage. The data commonly used in digital twinning includes geographic spatial data and three-dimensional model data. Geographic spatial data mainly comes from aerial photography and machine measurement. BIM models mainly come from fine hand modeling.

[0107] The collected geospatial data is first converted into CGCS2000 geodetic coordinate system. The geospatial database is constructed according to the geospatial position, and the data is indexed in space at the same time. Commonly used spatial indexes include R-tree index, quadtree index, KD-tree, etc. Based on the advantages and disadvantages of R-tree and quadtree, QR-tree is adopted, which is a spatial index structure combining R-tree and quadtree. Starting from the characteristics of R-tree, in order to improve the search performance, reduce the overlap of index space, and avoid or reduce the search branches, the "quadtree" hierarchical division method of index space is introduced, which divides the entire index space into multiple levels of sub-index space, and then indexes each level of sub-index space with R-tree. The query is limited to the local space as much as possible, thereby improving the search performance.

[0108] For vector data, its format is converted into shp format or GeoJson format, and after being stored in the database, it is published as a WFS service based on OGC standard through a program. For manual models, they can be converted into 3DTiles format data, and after being published as a service, they are loaded in the scene in a streaming manner, or they are built into LOD and loaded in the scene in a lightweight manner.

[0109] As shown in Figure 3 The scene building process is shown in the figure. When building a three-dimensional scene, first determine the scene space range, and perform spatial query in the geospatial database to retrieve the data within the range. DEM data and DOM data are used as the data basis of the scene. There are generally two loading methods. The first method is to convert DEM data into model data and place it in the scene, and DOM data is used as a material map on the model surface. The second method is to slice DEM data and DOM data, publish them as tile map services, WMS, WMTS, and other format data services, and the scene loads data in real time through the data service. Based on high-precision DOM data, AI technology is used for digital orthophoto map analysis and feature extraction, which generally includes spectral features, texture features, and shape features. Spectral feature extraction extracts the spectral features of ground objects by analyzing the spectral reflectance or emissivity of the digital orthophoto map. This method is suitable for digital orthophoto maps with high spectral resolution, such as hyperspectral images. Texture features reflect the complexity and organization of the surface structure of ground objects, and are an important basis for ground object recognition. Shape features describe the geometric shape and spatial distribution of ground objects, and are of great significance for ground object classification. After identifying buildings, water systems, roads, vegetation, and other ground objects through the above methods, corresponding model materials are matched in the scene material library and placed in the corresponding position of the scene.

[0110] Vector data includes two formats of OGC standard-based vector WFS service and GeoJson. In order to facilitate data editing, the visualization of vector data adopts the method of converting vector data into model patches and assigning materials. The whole process adopts the state machine mode. First, define the data structure for storing geographic coordinates, triangle index, UV coordinates, normal, etc. The second step is to analyze the data coordinates. There are different analysis and calculation methods for point, line and surface data formats. In order to reduce scene lag and improve the fluency of three-dimensional scene, push the vector data to a newly opened task thread for data analysis, triangle division, UV coordinate, normal calculation, etc. The task thread enters the calculation state, and after all the calculation tasks are completed, it enters the completion state, and returns the data to the main thread. The main thread listens to the running state of the task thread while performing real-time rendering of the scene. After the task thread enters the completion state, the main thread can obtain the calculation result and push it to the rendering thread in steps. In order to avoid causing scene lag and keep the scene above 30 frames with a frame time of no more than 0.03 seconds, the data not completed in the current frame is pushed to the next frame, and so on, until the data pushing is completed.

[0111] As shown in Figure 4 The scene editing and saving process is shown. Scene editing is the optimization of the basic scene built by the program intelligently. It supports replacing the material model with a more detailed or more realistic handmade model, and editing the shape, position, material and other attribute elements of the model. For vector data, it supports editing its material, color, shape, data coordinates and other attribute elements. The edited scene supports saving it as a scene configuration file, and supports saving it to the scene database on the server and the local file in two modes.

[0112] As shown in Figure 5 The scene rendering scheduling process is shown. The rendering process of the rendering engine has the following steps:

[0113] 1. Resource loading, loading local models, materials and other resources;

[0114] 2. Scene building, organizing objects in the scene into a specific data structure;

[0115] 3. View frustum culling and depth detection, used to remove objects not in the current view range;

[0116] 4. Visibility judgment, remove objects that are occluded in the current view;

[0117] 5. Rendering pipeline, describes how to render, set rendering state, allocate GPU resources, etc.;

[0118] 6. GPU rendering, start rendering.

[0119] In order to reduce the GPU rendering load, there are mainly the following ways: the first is the view frustum culling and visibility determination, the content finally rendered to the two-dimensional screen of the three-dimensional scene is obtained through the projection transformation, the objects outside the camera view frustum range are the contents invisible in the current screen, and the objects blocked by the objects are also invisible contents. For the invisible contents, the rendering process is not added, which can effectively reduce the GPU rendering load. The second is the GPU Instancing batch rendering technology. For the models in the scene that repeatedly appear and have consistent appearance shapes except position, scaling and other information, the batch rendering method is used to reduce the Draw Call, that is, the number of rendering submitted by CPU to GPU, so as to reduce the GPU rendering load. For complex and large models, vertex thinning, model reduction and other model simplification algorithms are used to establish a level of detail model (LOD), and different precision models are automatically switched according to the camera distance, so as to further reduce the GPU rendering load.

[0120] As shown in Figure 6 The scene structure is shown. The objects in the scene are stored and managed in a tree structure. There is only one root node, and the leaf node is an entity, and the child node is a data group. The scene tree structure is saved and analyzed by a recursive method.

[0121] As shown in Table 1, taking the flood storage area and beach village in the Yangtze River Basin as an example, more than 2000 scene elements, million-level vertices and triangular mesh quantities, the integrated scene construction method has a significant improvement in performance and time.

[0122] Table 1

[0123] Embodiment 2

[0124] A digital twin water conservancy scene dynamic building and scheduling method, comprising the following steps:

[0125] Step 1, after the collected geographic space data and three-dimensional model data are preprocessed, the preprocessed geographic space data and three-dimensional model data are published as standard geographic space data services and standard three-dimensional model data services;

[0126] Step 2, according to the business requirements, a preset scene template is selected, a scene base plate is constructed in the scene template based on the standard geographic space data service, an AI remote sensing image recognition technology is used to identify and classify the digital orthophoto map, the classification results are connected to the ground object model of the scene material library and the BIM model loaded based on the standard three-dimensional model data service, and the preliminary scene is obtained.

[0127] Step 3, the vector data is acquired through a standard geospatial data service, the coordinate resolution and the triangular meshing are sequentially completed by an asynchronous computing thread, the obtained triangular mesh data is pushed to a rendering thread by a main thread in frames, and the rendering thread performs a rendering operation on the basis of a preliminary scene according to the triangular mesh data to obtain a basic scene.

[0128] Step 4, the attributes and styles in the basic scene are edited, and scene information is stored in a structured manner according to a tree structure configuration file specification. Embodiment 3

[0129] A computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the above method. Embodiment 4

[0130] A computer program product includes a computer program, and the computer program is executed by a processor to implement the steps of the above method.

[0131] The contents not described in detail in the specification belong to the prior art known to those skilled in the art.

Claims

1. A dynamic construction and scheduling system for digital twin water conservancy scenarios, characterized in that, include: The data processing module is used to preprocess the collected geospatial data and 3D model data respectively, and then publish the preprocessed geospatial data and 3D model data as standard geospatial data services and standard 3D model data services. The preliminary scene building module is used to select a preset scene template according to business needs, build a scene base on the scene template based on standard geospatial data services, identify and classify digital orthophoto maps through AI remote sensing image recognition technology, and connect the land feature models in the scene material library corresponding to the classification results and the BIM models loaded based on standard 3D model data services to the scene template to obtain a preliminary scene. The scene rendering scheduling module is used to obtain vector data through standard geospatial data services. After the asynchronous calculation thread completes the coordinate parsing and triangulation in sequence, the main thread pushes the obtained triangular mesh data to the rendering thread in frames. The rendering thread performs rendering operations based on the triangular mesh data to obtain the basic scene. The scene rendering scheduling module is also used for: The vector data of the geospatial space is obtained using standard geospatial data services, and the vector data is transmitted to the asynchronous computing thread; The asynchronous computing thread performs coordinate analysis on the vector data and performs triangulation calculation on the line vector data and surface vector data based on the analysis results to generate triangular mesh data; The main thread monitors the status of the asynchronous computing thread in real time, and obtains the triangular mesh data when it detects that the asynchronous computing thread has completed the calculation. The main thread pushes the acquired triangular mesh data to the rendering thread frame by frame; The rendering thread performs rendering operations based on the received triangular mesh data to obtain the basic scene; The scene editing and storage module is used to edit the attributes and styles in the basic scene and store the scene information in a structured manner according to the tree structure configuration file specification; The method for obtaining the preliminary scene in the preliminary scene construction module is as follows: Obtain the preset scene template library and preset scene material library; Based on business needs, select a scene template from the preset scene template library; Based on the latitude and longitude range information of the preset scene template, the geospatial data service is indexed to obtain the geospatial data of the scene; The scene base is constructed within the scene template using digital orthophoto maps, digital elevation models, and digital surface models obtained from the geospatial database using standard geospatial data services. AI remote sensing image recognition technology is used to extract features from digital orthophoto maps, and ground features are identified and classified according to the features to obtain classification results. The classification results are then used to match the corresponding ground feature models in a preset scene material library, and the ground feature models in the preset scene material library are then connected to the scene template. The BIM model is obtained by loading and calling the standard 3D model data service in a streaming manner, and then the BIM model is connected to the scene template to complete the initial scene construction.

2. The digital twin water conservancy scenario dynamic construction and scheduling system according to claim 1, characterized in that, In the data processing module, the method for preprocessing the collected geospatial data and 3D model data, and then publishing the preprocessed geospatial data and 3D model data as standard geospatial data services and standard 3D model data services is as follows: The process of preprocessing and publishing geospatial data is as follows: Unify the spatiotemporal reference of geospatial data to obtain geospatial data with a unified spatiotemporal reference. The spatiotemporal reference includes a time reference, a spatial reference, and an elevation reference; The geospatial data after the unified spatiotemporal reference is standardized in format to obtain preprocessed geospatial data. A distributed spatiotemporal database is used to partition and store the preprocessed geospatial data according to geospatial location to obtain a geospatial database. Each partition in the geospatial database is indexed using an R-tree or quadtree spatial index. Publish geospatial data as a service based on the Open Geospatial Information Consortium (OGC) standards; The process of preprocessing and publishing 3D model data is as follows: The 3D model data is lightweighted by using face reduction or vertex thinning methods to construct a level-of-detail model. The 3D model data in the level-of-detail model is then standardized to obtain the pre-processed 3D model data. The preprocessed 3D model data is divided according to the levels in the level of detail model. Large-scale 3D models are published as 3DTiles format services, while lightweight 3D models are published as files.

3. The digital twin water conservancy scenario dynamic construction and scheduling system according to claim 2, characterized in that, In the data processing module, the process of standardizing the format of geospatial data and 3D model data is as follows: Geospatial data includes digital orthophoto maps, digital elevation models, digital surface models, oblique photogrammetry models, laser point clouds, underwater topography, and vector data; Among them, the digital orthophoto map, digital elevation model and digital surface model adopt the labeled image file format data; the vector data is converted into SHP or GeoJSON format data; The BIM model is converted into 3DTiles format data; The 3D model data includes landscape models and character models; Convert 3D model data to FBX or OBJ format.

4. The digital twin water conservancy scenario dynamic construction and scheduling system according to claim 3, characterized in that, During the rendering operation, the rendering thread employs at least one of the following methods: view frustum clipping, graphics processor instance, and level of detail model.

5. The digital twin water conservancy scenario dynamic construction and scheduling system according to claim 4, characterized in that, The method for editing and storing basic scenes in the scene editing and storage module is as follows: Edit the display style of the scene base in the basic scene; The terrain feature model and BIM model in the basic scene are translated, rotated, scaled, and their attribute parameters are edited. Edit the material, color, texture, and data coordinates of the spatial vector elements when the rendering thread performs rendering operations; According to the preset scenario configuration file specifications, the structured information of the edited basic scenario is stored in the server-side scenario database or local file; The scene configuration file specification is based on a scene graph structure serialized in JSON format. It is organized using a tree-like scene graph structure, including a root node, multiple intermediate group nodes, and multiple scene entity elements as leaf nodes. The scene entity elements correspond to the terrain model, BIM model, and spatial vector elements in the scene.

6. A method for dynamically constructing and scheduling a digital twin water conservancy scenario, characterized in that, Includes the following steps: After preprocessing the collected geospatial data and 3D model data respectively, the preprocessed geospatial data and 3D model data are published as standard geospatial data services and standard 3D model data services. Select a preset scene template according to business needs, build a scene baseboard within the scene template based on standard geospatial data services, identify and classify digital orthophoto maps using AI remote sensing image recognition technology, and connect the ground feature models in the scene material library corresponding to the classification results and the BIM models loaded based on standard 3D model data services into the scene template to obtain a preliminary scene. Vector data is obtained through standard geospatial data services. After coordinate parsing and triangulation are completed sequentially by an asynchronous computing thread, the main thread pushes the obtained triangular mesh data to the rendering thread in frames. The rendering thread performs rendering operations based on the triangular mesh data to obtain the basic scene. The asynchronous calculation thread acquires triangular mesh data and pushes it to the rendering thread frame by frame. The rendering thread then performs the rendering operation as follows: The vector data of the geospatial space is obtained using standard geospatial data services, and the vector data is transmitted to the asynchronous computing thread; The asynchronous computing thread performs coordinate analysis on the vector data and performs triangulation calculation on the line vector data and surface vector data based on the analysis results to generate triangular mesh data; The main thread monitors the status of the asynchronous computing thread in real time, and obtains the triangular mesh data when it detects that the asynchronous computing thread has completed the calculation. The main thread pushes the acquired triangular mesh data to the rendering thread frame by frame; The rendering thread performs rendering operations based on the received triangular mesh data to obtain a basic scene; it then edits the attributes and styles in the basic scene and stores the scene information in a structured manner according to the tree structure configuration file specification. The method for obtaining the preliminary scene is as follows: Obtain the preset scene template library and preset scene material library; Based on business needs, select a scene template from the preset scene template library; Based on the latitude and longitude range information of the preset scene template, the geospatial data service is indexed to obtain the geospatial data of the scene; The scene base is constructed within the scene template using digital orthophoto maps, digital elevation models, and digital surface models obtained from the geospatial database using standard geospatial data services. AI remote sensing image recognition technology is used to extract features from digital orthophoto maps, and ground features are identified and classified according to the features to obtain classification results. The classification results are then used to match the corresponding ground feature models in a preset scene material library, and the ground feature models in the preset scene material library are then connected to the scene template. The BIM model is obtained by loading and calling the standard 3D model data service in a streaming manner, and then the BIM model is connected to the scene template to complete the initial scene construction.

7. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in claim 6.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method of claim 6.

Images