Multi-graphics engine scheduling and rendering method and system for digital twins

Abstract

This invention discloses a multi-graphics engine scheduling and rendering method and system for digital twins, relating to the field of data processing technology. The method includes: constructing a digital twin 3D scene and forming a multi-source heterogeneous model element library; obtaining rendering performance analysis of multiple graphics rendering engines to acquire performance characteristics of each engine; decomposing the digital twin rendering requirements into multiple rendering task units and constructing multiple task description vectors; updating the scheduling parameters of the multiple rendering task units; optimizing the rendering engine hybrid scheduling strategy for the multiple rendering task units to obtain a combination of multiple target rendering engines and a global hybrid scheduling strategy; and executing the rendering tasks. This invention solves the technical problem in existing digital twin scenes where multiple graphics rendering engines are difficult to efficiently coordinate and schedule according to rendering requirements, resulting in low overall rendering efficiency. It achieves the technical effect of improving the parallel collaborative efficiency of multiple graphics rendering engines and enhancing the overall rendering performance of the digital twin scene.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and more specifically to a multi-graphics engine scheduling and rendering method and system for digital twins. Background Technology

[0002] In digital twin applications, it is often necessary to process heterogeneous data from multiple sources, such as oblique photogrammetry models, BIM models, and laser point cloud models, simultaneously. Different types of rendering tasks have significantly different requirements for the computing power, real-time performance, and stability of graphics rendering engines. Due to the inconsistencies in the architecture and performance characteristics of various graphics rendering engines, a fixed or single engine allocation method is often used during the rendering process. It is difficult to dynamically coordinate and schedule engines according to specific rendering needs and operating status, resulting in some engines being overloaded while other resources are idle, affecting the overall rendering efficiency and the real-time presentation effect of the digital twin scene. Summary of the Invention

[0003] This application provides a multi-graphics engine scheduling and rendering method and system for digital twins, which is used to address the technical problem that multiple graphics rendering engines in existing digital twin scenarios are difficult to coordinate and schedule efficiently according to rendering needs, resulting in low overall rendering efficiency.

[0004] In view of the above problems, this application provides a multi-graphics engine scheduling and rendering method and system for digital twins.

[0005] A first aspect of this application provides a multi-graphics engine scheduling and rendering method for digital twins, the method comprising: A digital twin 3D scene is constructed, comprising an oblique photogrammetry model, a BIM model, a laser point cloud model, an underground pipeline model, and a terrain model. A multi-source heterogeneous model element library is formed through unified spatial coordinate calibration and semantic annotation. Multiple graphics rendering engines of the digital twin platform are acquired, and their rendering performance is analyzed to obtain their performance characteristics. The digital twin rendering requirements are decomposed into multiple rendering task units, and multiple task description vectors are constructed for each unit. The scheduling parameters of the multiple rendering task units are dynamically updated based on real-time situational data, and historical rendering experience is incorporated into the rendering scheduling process. The rendering engine hybrid scheduling strategy for the multiple rendering task units is optimized to obtain a combination of multiple target rendering engines and a global hybrid scheduling strategy. Based on the global hybrid scheduling strategy and the combination of multiple target rendering engines, the multiple graphics rendering engines are driven to execute rendering tasks in parallel.

[0006] A second aspect of this application provides a multi-graphics engine scheduling and rendering system for digital twins, the system comprising: The module generates a digital twin 3D scene, including oblique photogrammetry, BIM, laser point cloud, underground pipeline, and terrain models, and forms a multi-source heterogeneous model element library through unified spatial coordinate calibration and semantic annotation. The performance analysis module acquires multiple graphics rendering engines from the digital twin platform, performs rendering performance analysis, and obtains the performance characteristics of each engine. The decomposition module decomposes the digital twin rendering requirements into multiple rendering task units and constructs multiple task description vectors for each unit. The strategy optimization module dynamically updates the scheduling parameters of the multiple rendering task units based on real-time situational data, incorporates historical rendering experience during the rendering scheduling process, optimizes the hybrid scheduling strategy for multiple rendering task units, and obtains a combination of multiple target rendering engines and a global hybrid scheduling strategy. The driving module drives multiple graphics rendering engines to execute rendering tasks in parallel according to the global hybrid scheduling strategy and the combination of multiple target rendering engines.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: This application constructs a digital twin 3D scene including an oblique photogrammetry model, a BIM model, a laser point cloud model, an underground pipeline model, and a terrain model, and forms a multi-source heterogeneous model element library through unified spatial coordinate calibration and semantic annotation; it acquires multiple graphics rendering engines from the digital twin platform, performs rendering performance analysis, and obtains the performance characteristics of multiple graphics rendering engines; it decomposes the digital twin rendering requirements into multiple rendering task units, and constructs multiple task description vectors for multiple rendering task units; it dynamically updates the scheduling parameters of the multiple rendering task units based on real-time situational data, and introduces historical rendering experience memory into the rendering scheduling process, optimizes the rendering engine hybrid scheduling strategy for multiple rendering task units, and obtains multiple target rendering engine combinations and a global hybrid scheduling strategy; according to the global hybrid scheduling strategy and the multiple target rendering engine combinations, it drives multiple graphics rendering engines to execute rendering tasks in parallel. This invention addresses the technical problem of low overall rendering efficiency in existing digital twin scenarios where multiple graphics rendering engines struggle to coordinate efficiently according to rendering requirements. By introducing a multi-graphics engine hybrid scheduling mechanism based on rendering task unit decomposition and combining real-time situational data with historical rendering experience, this invention improves the parallel coordination efficiency of multiple graphics rendering engines and enhances the overall rendering performance of digital twin scenarios. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 A schematic diagram of the multi-graphics engine scheduling and rendering method for digital twins provided in this application embodiment; Figure 2 This is a schematic diagram of the structure of a multi-graphics engine scheduling and rendering system for digital twins provided in an embodiment of this application.

[0010] Figure labeling: Feature library generation module 11, performance analysis module 12, disassembly module 13, strategy optimization module 14, driving module 15. Detailed Implementation

[0011] This application provides a multi-graphics engine scheduling and rendering method and system for digital twins. It addresses the technical problem of low overall rendering efficiency in existing digital twin scenarios where multiple graphics rendering engines are difficult to coordinate and schedule efficiently according to rendering needs. By introducing a multi-graphics engine hybrid scheduling mechanism based on rendering task unit decomposition and combining real-time situational data and historical rendering experience, the technical effect of improving the parallel collaboration efficiency of multiple graphics rendering engines and enhancing the overall rendering performance of digital twin scenarios is achieved.

[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0013] It should be noted that any variation of the terms "comprising" and "having" is intended to cover non-exclusive inclusion, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to such processes, methods, products, or devices.

[0014] Example 1, as Figure 1 As shown, this application provides a multi-graphics engine scheduling and rendering method for digital twins, the method comprising: Step S100: Construct a digital twin 3D scene including oblique photogrammetry model, BIM model, laser point cloud model, underground pipeline model and terrain model, and form a multi-source heterogeneous model element library through unified spatial coordinate calibration and semantic annotation.

[0015] In this embodiment, the oblique photogrammetry model, BIM model, laser point cloud model, underground pipeline model, and terrain model are first integrated into the same three-dimensional spatial environment to form a digital twin three-dimensional scene containing multiple source models. Specifically, the oblique photogrammetry model is generated by acquiring multi-angle images using drones or aerial platforms and employing 3D reconstruction methods; the BIM model is created or imported using building information modeling software; the laser point cloud model is formed by acquiring spatial point cloud data using laser scanning equipment and performing basic filtering; the underground pipeline model is obtained by modeling from pipeline design data or detection data; and the terrain model is generated from digital elevation data.

[0016] Subsequently, a unified spatial reference system processing was performed on all the aforementioned models. Spatial coordinate calibration ensured that models from different sources and coordinate systems were positioned consistently within the same spatial reference system. After the unified spatial reference system processing was completed, semantic annotation was applied to each model, adding semantic attribute information such as model category, functional attributes, and spatial hierarchy to model elements. Finally, the oblique photogrammetry models, BIM models, laser point cloud models, underground pipeline models, and terrain models that have undergone spatial coordinate calibration and semantic annotation are centrally stored, forming a multi-source heterogeneous model element library.

[0017] Furthermore, the method provided in the application embodiments constructs a digital twin 3D scene including an oblique photogrammetry model, a BIM model, a laser point cloud model, an underground pipeline model, and a terrain model, and forms a multi-source heterogeneous model element library through unified spatial coordinate calibration and semantic annotation, and also includes: The oblique photogrammetry model, BIM model, laser point cloud model, underground pipeline model, and terrain model are uniformly processed using a spatial reference system. After uniform processing, the models are semantically annotated to obtain additional semantic attribute information for each model. The additional semantic attribute information includes model category, functional attributes, and spatial hierarchy. The oblique photogrammetry model, BIM model, laser point cloud model, underground pipeline model, and terrain model that have completed coordinate calibration and semantic annotation are stored in a multi-source heterogeneous model element library.

[0018] In this embodiment, when performing unified spatial reference system processing on the oblique photogrammetry model, BIM model, laser point cloud model, underground pipeline model, and terrain model, the data files of each model are first read, and parameters such as the spatial reference system identifier, coordinate unit, coordinate axis direction, and origin definition are extracted. Then, the unified spatial reference system used in the digital twin 3D scene is determined, and the coordinate values ​​of each model are converted to the unified spatial reference system according to a predetermined coordinate transformation process, and the coordinate units are unified to the same unit of measurement. After completing the coordinate transformation and unit unification, the common reference point in the digital twin 3D scene is used as the alignment basis, and coordinate calibration processing is performed on each model. By performing translation correction, rotation correction, and scale correction on the entire model, the oblique photogrammetry model, BIM model, laser point cloud model, underground pipeline model, and terrain model achieve consistent position, direction, and scale in the same 3D space, thereby completing the unified spatial reference system processing.

[0019] After unified processing by the spatial reference system, semantic annotation of the models begins by extracting identifiable objects from the oblique photogrammetry model, BIM model, laser point cloud model, underground pipeline model, and terrain model as model elements, and establishing a semantic attribute record for each model element. Subsequently, additional semantic attribute information is populated into each model element. This includes assigning a model category field to identify the model element's category (oblique photogrammetry model, BIM model, laser point cloud model, underground pipeline model, or terrain model), assigning a function attribute field to identify the model element's purpose and meaning in the digital twin business, and assigning a spatial hierarchy field to describe the model element's hierarchical affiliation and subordinate relationship within the 3D scene structure. After the additional semantic attribute information is written, the semantic attribute record is bound to the corresponding model element, ensuring that the additional semantic attribute information for each model fully encompasses the model category, function attribute, and spatial hierarchy, thus completing the semantic annotation and obtaining the additional semantic attribute information for each model.

[0020] Finally, when storing the oblique photogrammetry model, BIM model, laser point cloud model, underground pipeline model, and terrain model, which have undergone coordinate calibration and semantic annotation, into the multi-source heterogeneous model feature library, the geometric data of each model, the spatial coordinate information processed uniformly by the spatial reference system, and the additional semantic attribute information obtained from semantic annotation are first linked and organized to form storage entries indexed by model features. These storage entries are then written into the multi-source heterogeneous model feature library, and a retrieval index is established with model category, functional attributes, and spatial hierarchy as key fields. This enables the multi-source heterogeneous model feature library to achieve unified storage, unified retrieval, and unified access for the oblique photogrammetry model, BIM model, laser point cloud model, underground pipeline model, and terrain model.

[0021] Step S200: Obtain multiple graphics rendering engines from the digital twin platform, perform rendering performance analysis, and obtain the performance characteristics of multiple graphics rendering engines.

[0022] In this embodiment, within the digital twin platform's operating environment, the deployed and callable graphics rendering engines are first acquired. By reading system configuration and engine registration information, multiple graphics rendering engines participating in digital twin rendering are identified. After acquiring these engines, each engine's historical rendering records are sequentially traversed to form a corresponding historical rendering record set. Then, based on this historical rendering record set, different rendering scenarios involved in the rendering process are aggregated in a scene-based manner. Scene-based performance analysis is then performed on the aggregated rendering scenarios to obtain a set of scene-based performance characteristics for each graphics rendering engine under different rendering scenarios. Subsequently, the set of scene-based performance characteristics is traversed and scene-based identified, ultimately yielding graphics rendering engine performance characteristics reflecting the performance of multiple graphics rendering engines under different rendering scenarios.

[0023] Furthermore, the method provided in the application embodiments, which involves obtaining multiple graphics rendering engines of the digital twin platform, performing rendering performance analysis, and obtaining the performance characteristics of the multiple graphics rendering engines, also includes: The process involves iterating through the multiple graphics rendering engines to extract historical rendering records, resulting in multiple sets of historical rendering records; then, based on these sets, the process involves scene-based aggregation and performance analysis of rendering scenarios, resulting in multiple sets of scene-based performance features for the graphics rendering engines; finally, the process involves iterating through these sets of scene-based performance features for the graphics rendering engines to identify scenarios, resulting in multiple performance features for the graphics rendering engines.

[0024] In this embodiment of the application, when extracting historical rendering records from multiple graphics rendering engines, the multiple graphics rendering engines already acquired in the digital twin platform are first traversed one by one. By accessing the rendering log data automatically generated and saved by each graphics rendering engine during operation, historical rendering record information related to the execution of rendering tasks is extracted. The historical rendering record information includes the model type, rendering frame rate, rendering time, resource usage status, and running result status corresponding to the rendering task. The extracted historical rendering records are then organized and collected according to the graphics rendering engine dimension, thereby forming multiple historical rendering record sets corresponding to multiple graphics rendering engines.

[0025] Next, rendering scene-based aggregation and performance analysis are performed based on multiple historical rendering record sets. In this process, the first historical rendering record set is extracted and subjected to rendering scene-based aggregation, forming multiple first aggregated historical rendering record sets. Then, performance feature extraction is performed on these multiple first aggregated historical rendering record sets, resulting in multiple first aggregated performance feature sets. Subsequently, the mean drift of performance features within each of these first aggregated performance feature sets is filtered to eliminate abnormal performance features and retain stable performance features, resulting in multiple filtered aggregated performance features. Finally, these multiple filtered aggregated performance features are used as the first graphics rendering engine scene-based performance feature set and added to the multiple graphics rendering engine scene-based performance feature sets. By repeating the aforementioned process, multiple graphics rendering engine scene-based performance feature sets are obtained.

[0026] Finally, the scene-specific performance feature sets of multiple graphics rendering engines are traversed one by one. During the traversal, the scene-specific performance feature data corresponding to each graphics rendering engine is read sequentially. This scene-specific performance feature data includes performance indicators such as rendering frame rate, rendering time, resource usage, and stability under a specific rendering scenario. Subsequently, based on the rendering scenario from which the scene-specific performance feature originates, scene-specific identification processing is performed on each performance feature, that is, binding the performance feature with its corresponding rendering scenario type, so that each performance feature has a clear rendering scenario identifier. After completing the scene-specific identification, the scene-specific performance features generated by the same graphics rendering engine under different rendering scenarios are summarized and organized to form a set of graphics rendering engine performance features that can reflect the performance of the graphics rendering engine under multiple rendering scenario conditions, thereby obtaining multiple graphics rendering engine performance features.

[0027] Furthermore, in the method provided in the application embodiments, the process of performing rendering scene-based aggregation and rendering scene-based performance analysis based on the multiple historical rendering record sets to obtain multiple graphics rendering engine scene-based performance feature sets also includes: Extract a first historical rendering record set from multiple historical rendering record sets and aggregate it for rendering scenarios to obtain multiple first aggregated historical rendering record sets; traverse the multiple first aggregated historical rendering record sets to extract performance features to obtain multiple first aggregated performance feature sets; perform mean shift filtering on the performance features within each of the multiple first aggregated performance feature sets to obtain multiple filtered aggregated performance features; add the multiple filtered aggregated performance features as a first graphics rendering engine scenario-based performance feature set to the multiple graphics rendering engine scenario-based performance feature sets.

[0028] In this embodiment, when extracting a first historical rendering record set from multiple historical rendering record sets for rendering scene aggregation, the graphics rendering engine identifiers corresponding to the multiple historical rendering record sets are first determined in the digital twin platform, and a first historical rendering record set is selected from them, where "first" refers to any historical rendering record set corresponding to any graphics rendering engine. Then, the historical rendering records in the first historical rendering record set are read one by one according to the generation order of the historical rendering records, and deterministic field information describing the rendering scene is extracted from each historical rendering record. This field information includes the combination of model types involved in rendering, the number of models involved in rendering, the geometric scale parameters of the models involved in rendering, the rendering perspective parameters, and a list of identifiers of the models actually involved in rendering. After extracting the field information, the model type combination is used as the primary aggregation condition. Historical rendering records with completely identical model type combinations are grouped into the same group. Under the premise of identical model type combinations, the model quantity field is compared numerically, and historical rendering records with identical model quantity values ​​are grouped into the same group. Based on identical model quantity values, the model geometry parameters are compared item by item, and historical rendering records with identical model geometry parameters are further grouped into the same group. Next, the rendering viewpoint parameters are compared, and historical rendering records with identical rendering viewpoint direction and viewpoint range parameters are grouped into the same group. Finally, the model identifier list is compared, and historical rendering records with completely identical model identifier lists participating in the actual rendering are grouped into the same group. Thus, through layer-by-layer consistency determination of model type combination, model quantity, model geometry parameters, rendering viewpoint parameters, and model identifier list, the rendering scene aggregation is completed, forming multiple first-aggregated historical rendering record sets. Each first-aggregated historical rendering record set corresponds to a specific and reproducible rendering scene.

[0029] Next, the multiple first-aggregate historical rendering record sets are traversed. One first-aggregate historical rendering record set is selected sequentially, and each historical rendering record in that set is read. Then, performance data fields directly related to rendering performance are extracted from each historical rendering record. These performance data fields include rendering frame rate values, single-frame rendering time values, resource usage status records during rendering, and rendering task completion status records. After the performance data fields are extracted, the rendering frame rate values ​​extracted from the same first-aggregate historical rendering record set are aggregated to form a rendering frame rate dataset; the rendering time values ​​are aggregated to form a rendering time dataset; the resource usage status records are aggregated to form a resource usage status dataset; and the rendering task completion status records are aggregated to form a runtime status dataset, thus constructing a first-aggregate performance feature set corresponding to this rendering scene. By repeating the above performance data extraction and aggregation steps for multiple first-aggregate historical rendering record sets, multiple first-aggregate performance feature sets are obtained.

[0030] Subsequently, when performing mean drift filtering on multiple first aggregated performance feature sets, for any given first aggregated performance feature set, the mean value within the set is first calculated for the rendering frame rate dataset, rendering time dataset, and resource usage status dataset to obtain the corresponding performance mean benchmark. Then, each performance feature record in the first aggregated performance feature set is read one by one, and the rendering frame rate value, rendering time value, and resource usage status value in that record are compared item by item with the corresponding performance mean benchmark. When the deviation between any performance value and the performance mean benchmark exceeds a preset deviation threshold, the performance feature record is determined as an abnormal performance feature and removed from the first aggregated performance feature set. After the abnormal performance feature removal is completed, the remaining performance feature records are retained as stable performance features, thus forming the filtered aggregated performance features. By repeatedly performing the performance mean calculation, performance deviation comparison, and abnormal performance feature removal process on multiple first aggregated performance feature sets, multiple filtered aggregated performance features are obtained.

[0031] After obtaining multiple aggregated performance features, these features are organized according to their corresponding first aggregated historical rendering record sets and stored as the first graphics rendering engine scene-based performance feature set. Subsequently, the first graphics rendering engine scene-based performance feature set is added to multiple graphics rendering engine scene-based performance feature sets. By repeating the above rendering scene-based aggregation, performance feature extraction, and performance feature filtering processes on different historical rendering record sets, multiple graphics rendering engine scene-based performance feature sets covering multiple graphics rendering engines and multiple rendering scenes are gradually constructed.

[0032] Step S300: Decompose the digital twin rendering requirements into multiple rendering task units and construct multiple task description vectors for multiple rendering task units.

[0033] Furthermore, the method provided in the application embodiments also includes: Each of the multiple task description vectors includes multi-model combined features, real-time features, and stability constraints.

[0034] In this embodiment, when decomposing the digital twin rendering requirements into multiple rendering task units, the digital twin rendering requirements are first read item by item to clarify the specific object to be rendered, the rendering area, the rendering content, and the business conditions that trigger the rendering. The overall rendering requirements are then expanded according to time sequence and spatial scope. Subsequently, using the information necessary for a single rendering execution as the decomposition standard, the overall rendering requirements are divided into several independently executable rendering segments. Each rendering segment corresponds to a specific rendering execution process, and this rendering segment is defined as a rendering task unit. During the decomposition process, it is ensured that each rendering task unit corresponds to a defined set of rendering objects, a defined rendering viewpoint state, and defined rendering trigger conditions, thereby forming multiple rendering task units.

[0035] Next, when constructing the task description vector, the types of models actually involved in rendering within the rendering task unit are first checked. The oblique photogrammetry model, BIM model, laser point cloud model, underground pipeline model, and terrain model involved in rendering within the task unit are marked one by one, and the corresponding model participation is written into the multi-model combination feature field of the task description vector to clearly indicate the model combination form involved in the rendering task unit. Then, the refresh rate and allowable response time required by the rendering task unit during operation are read, and the refresh rate and response time requirements are written into the real-time feature field of the task description vector to clarify the rendering speed and update rhythm requirements of the rendering task unit. Next, the requirements for image continuity and stable output during business operation of the rendering task unit are read, and constraints on whether image interruption, image jitter, and continuous output are allowed during rendering are written into the stability constraint field of the task description vector to clarify the stability requirements of the rendering task unit during execution.

[0036] After completing the filling in of the multi-model combination features, real-time features, and stability constraints, the task description vectors corresponding to each rendering task unit are uniformly organized and summarized, and finally multiple task description vectors for multiple rendering task units are constructed.

[0037] Step S400: Dynamically update the scheduling parameters of the multiple rendering task units based on real-time situational data, and introduce historical rendering experience memory into the rendering scheduling process to optimize the rendering engine hybrid scheduling strategy of multiple rendering task units, thereby obtaining a combination of multiple target rendering engines and a global hybrid scheduling strategy.

[0038] In this embodiment, when dynamically updating the scheduling parameters of multiple rendering task units based on real-time situational data and introducing historical rendering experience memory during the rendering scheduling process, the real-time situational data is first parsed to extract situational information related to spatial location, object state, and event type, forming multiple measurement point situational information. The scheduling parameters of multiple rendering task units are then dynamically updated based on the measurement point situational information. Subsequently, from the two dimensions of spatial location and business association, the multiple measurement point situational information are mapped to the corresponding rendering task units to obtain multiple sets of associated rendering task units, and a mapping relationship between measurement point situational information and sets of associated rendering task units is constructed. Under the constraints of the mapping relationship between measurement point situational information and sets of associated rendering task units, historical rendering experience memory is introduced to optimize the rendering engine hybrid scheduling strategy for multiple rendering task units, determine multiple target rendering engine combinations, and form a global hybrid scheduling strategy based on the multiple target rendering engine combinations.

[0039] Furthermore, in the method provided in the application embodiments, the scheduling parameters of the multiple rendering task units are dynamically updated based on real-time situational data, and historical rendering experience memory is introduced during the rendering scheduling process. The method optimizes the rendering engine hybrid scheduling strategy for the multiple rendering task units to obtain a combination of multiple target rendering engines and a global hybrid scheduling strategy. It also includes: The real-time situational data is analyzed to extract situational information related to spatial location, object status, and event type, obtaining situational information from multiple measurement points. From the two dimensions of spatial location and business association, the situational information from multiple measurement points is mapped to the corresponding rendering task units to obtain multiple sets of associated rendering task units, and a mapping relationship between the situational information from measurement points and the sets of associated rendering task units is constructed. Based on the mapping relationship between the situational information from measurement points and the sets of associated rendering task units, the rendering engine hybrid scheduling strategy is optimized to obtain a combination of multiple target rendering engines and a global hybrid scheduling strategy.

[0040] In this embodiment, the real-time situational data is first analyzed. This data originates from the digital twin platform's real-time monitoring and event reporting of real-world objects and business systems. Its content corresponds to the oblique photography model, BIM model, laser point cloud model, underground pipeline model, and terrain model already constructed in the digital twin 3D scene, indicating which spatial location experienced an object state change and what event type triggered it. The analysis process reads the real-time situational data item by item, extracting three types of information from each data item: spatial location, object state, and event type. The spatial location determines the 3D coordinates or spatial area where the situation occurred; the object state describes the current state change of the corresponding object; and the event type describes the type of business event that triggered the change. The spatial location, object state, and event type extracted from the same real-time situational data item are combined to form a single measurement point situational information, thus obtaining multiple measurement point situational information items. For example, if the spatial location corresponds to the area where a building is located, the object state is an abnormal temperature, and the event type is an alarm event, then this measurement point situational information in the digital twin 3D scene points to the relevant elements in the oblique photography model and BIM model containing that building.

[0041] After obtaining situational information from multiple measurement points, when mapping this information to corresponding rendering task units from both spatial location and business relevance dimensions, the process first considers the spatial location dimension. Each measurement point's situational information is then processed to locate its corresponding spatial region within the digital twin 3D scene, and the rendering task unit responsible for rendering that region is determined and included in the associated rendering task unit set. Next, the same measurement point's situational information is processed from the business relevance dimension. Based on its object status and event type, scene elements requiring highlighting, updating, or increased refresh rates are identified. Accordingly, the task unit containing the model elements corresponding to these objects is selected and added to the associated rendering task unit set. These objects may correspond to components and equipment in the BIM model, pipe segments and ancillary facilities in the underground pipeline model, spatial entity changes in the laser point cloud model, surface location elements in the oblique photogrammetry model, and surface areas in the terrain model. For example, if the spatial location falls on a road segment and the event type is a pipeline alarm, the business association points to the corresponding pipe segment and manhole in the underground pipeline model, and simultaneously associates it with the oblique photography model for surface positioning display. Then, the rendering task units involving the underground pipeline model and the oblique photography model are added to the associated rendering task unit set. After completing the spatial location dimension mapping and the business association dimension mapping, the corresponding associated rendering task unit set is recorded for each measurement point's situation information, forming a measurement point situation information-associated rendering task unit set mapping relationship.

[0042] Finally, based on the mapping relationship between measurement point situation information and associated rendering task unit sets, the rendering engine hybrid scheduling strategy is optimized. In this process, based on the mapping relationship between measurement point situation information and associated rendering task unit sets, firstly, using multiple rendering task units as indices, the mapping relationship is retrieved to determine multiple rendering task unit index clues corresponding to the current measurement point situation information. Then, multiple rendering task unit index clues are traversed to extract a set of multiple historical rendering task unit index clues corresponding to each rendering task unit within a preset historical window. Next, historical rendering experience is introduced, and directional local rendering engine hybrid scheduling is performed on multiple rendering task unit index clues through multiple historical rendering task unit index clue sets, obtaining multiple local rendering engine hybrid scheduling strategies and corresponding combinations of multiple target rendering engines. Finally, the multiple local rendering engine hybrid scheduling strategies are globally jointly scheduled to form a global hybrid scheduling strategy covering multiple rendering task units.

[0043] Furthermore, in the method provided in the application embodiments, based on the mapping relationship between measurement point situation information and associated rendering task unit sets, the rendering engine hybrid scheduling strategy is optimized to obtain a combination of multiple target rendering engines and a global hybrid scheduling strategy, and the method further includes: Using multiple rendering task units as indexes, the mapping relationship between measurement point situation information and associated rendering task unit sets is retrieved to determine multiple rendering task unit index clues; the multiple rendering task units are traversed to extract multiple historical rendering task unit index clue sets within a preset historical window; historical rendering experience memory is introduced, and directional local rendering engine hybrid scheduling is performed on the multiple rendering task unit index clues through the multiple historical rendering task unit index clue sets to obtain multiple local rendering engine hybrid scheduling strategies and multiple target rendering engine combinations; the multiple local rendering engine hybrid scheduling strategies are globally jointly scheduled to obtain a global hybrid scheduling strategy.

[0044] In this embodiment, when retrieving the mapping relationship between measurement point situation information and associated rendering task unit sets using multiple rendering task units as indexes, a list of identifiers for multiple rendering task units is first prepared, and one of the rendering task unit identifiers is retrieved as the current retrieval object. Then, the record containing the rendering task unit identifier is searched in the mapping relationship between measurement point situation information and associated rendering task unit sets. After locating the record, the corresponding measurement point situation information content is read, and the correspondence between the rendering task unit identifier and the measurement point situation information is recorded. By repeatedly performing the search, location, reading, and recording steps for multiple rendering task unit identifiers, multiple rendering task unit index clues are obtained, where each rendering task unit index clue contains a rendering task unit identifier and the measurement point situation information identifier corresponding to that rendering task unit.

[0045] Next, multiple rendering task units are traversed to extract a set of historical rendering task unit index clues within a preset historical window. In this process, the start and end times of the preset historical window are first determined for each rendering task unit, and a search is performed in the historical scheduling records of the digital twin platform according to the time range. Then, using the rendering task unit identifier as the matching condition, historical records within the preset historical window are filtered one by one, retaining those matching the identifier. In each filtered historical record, the corresponding measurement point situation information identifier, the rendering engine combination identifier used at that time, and the recorded rendering effect information are read, and this information, along with the rendering task unit identifier, is organized into a historical rendering task unit index clue. By repeating the filtering and organization steps for all historical records within the preset historical window, a set of historical rendering task unit index clues corresponding to that rendering task unit is formed. By repeatedly performing the steps of determining the time range, searching historical records, filtering and matching records, and organizing index clues for multiple rendering task units, multiple sets of historical rendering task unit index clues are obtained.

[0046] Next, historical rendering experience memory is introduced, and directional local rendering engine hybrid scheduling is performed on multiple rendering task unit index clue sets through multiple historical rendering task unit index clue sets. In this process, based on the introduced historical rendering experience memory, similarity matching is performed on multiple historical rendering task unit index clue sets based on the multiple rendering task unit index clues to obtain multiple matching historical rendering task unit index clue sets. Then, the corresponding historical rendering engine combination sets and historical rendering effect sets are extracted from the multiple matching historical rendering task unit index clue sets. Next, based on the comprehensive performance of the historical rendering engine combination sets and historical rendering effect sets, multiple hybrid scheduling directions are determined. Finally, based on the multiple hybrid scheduling directions and multiple rendering task unit index clues, directional local rendering engine hybrid scheduling is performed to form multiple local rendering engine hybrid scheduling strategies and obtain multiple target rendering engine combinations.

[0047] Finally, a global joint scheduling of multiple local rendering engine hybrid scheduling strategies is performed. This process begins by obtaining multiple importance weights and schedulable resource sets corresponding to multiple rendering task units. Then, based on these importance weights, a global joint scheduling of multiple local rendering engine hybrid scheduling strategies under the schedulable resource sets is performed. This unified coordination and resource allocation of rendering engine combinations for different rendering task units ultimately yields the global hybrid scheduling strategy.

[0048] Furthermore, the method provided in the application embodiment introduces historical rendering experience memory, performs directional local rendering engine hybrid scheduling on the multiple rendering task unit index threads through multiple historical rendering task unit index thread sets, obtains multiple local rendering engine hybrid scheduling strategies and multiple target rendering engine combinations, and further includes: Based on the multiple rendering task unit index clues, experience-memory similarity matching is performed on the multiple historical rendering task unit index clue sets to obtain multiple matching historical rendering task unit index clue sets; multiple historical rendering engine combination sets and multiple historical rendering effect sets are obtained from the multiple matching historical rendering task unit index clue sets; multiple hybrid scheduling directions are determined based on the multiple historical rendering engine combination sets and multiple historical rendering effect sets; directional local rendering engine hybrid scheduling is performed according to the multiple hybrid scheduling directions and multiple rendering task unit index clues to obtain multiple local rendering engine hybrid scheduling strategies and multiple target rendering engine combinations.

[0049] In this embodiment, when performing empirical memory similarity matching on multiple sets of historical rendering task unit index clues based on multiple rendering task unit index clues, the task description vector features corresponding to each rendering task unit index clue are first expanded. The multi-model combination features are then split into participation identifiers for oblique photogrammetry models, BIM models, laser point cloud models, underground pipeline models, and terrain models. The real-time features are split into rendering refresh frequency requirements and response time requirements, and the stability constraints are split into rendering continuity requirements and interruption tolerance requirements. Subsequently, the multiple sets of historical rendering task unit index clues are traversed, and the same feature expansion is performed on each historical rendering task unit index clue. Each clue is then compared with the current rendering task unit index clue item by item. When the three types of features—multi-model combination features, real-time features, and stability constraints—all satisfy the condition that the corresponding field values ​​are completely consistent, the historical rendering task unit index clue is included in the matching results, and multiple sets of matched historical rendering task unit index clues are obtained.

[0050] Then, for each matching historical rendering task unit index clue set, the historical rendering task unit index clue records are read one by one. The corresponding historical rendering engine combination identifier is extracted from each record and collected to form a historical rendering engine combination set. At the same time, the corresponding historical rendering effect records are extracted and collected to form a historical rendering effect set. The historical rendering effect records include rendering frame rate, rendering time, resource usage status and running result status. In this way, multiple historical rendering engine combination sets and multiple historical rendering effect sets are obtained.

[0051] Next, when determining multiple hybrid scheduling directions based on multiple historical rendering engine combinations and multiple historical rendering effect sets, a particle swarm optimization method is adopted. Each historical rendering engine combination is assigned as a candidate particle, and a fitness value is calculated for each candidate particle. The fitness value calculation process takes the rendering frame rate, rendering time, and stability performance from the historical rendering effect set as inputs. First, the rendering frame rate value is normalized to make the rendering frame rates of different records comparable. Then, the rendering time value is normalized to make the rendering time of different records comparable. Finally, the stability performance value is normalized to make the stability performance of different records comparable. After normalization, the normalized rendering frame rate result is used as a positive indicator, the normalized rendering time result as a negative indicator, and the normalized stability performance result as a positive indicator. The three indicators are weighted and summed according to preset weights to obtain the fitness value of the candidate particle. This allows the fitness value to simultaneously reflect the comprehensive effect of the historical rendering engine combination in terms of rendering frame rate, rendering time, and stability performance. During the iteration process of the particle swarm optimization method, candidate particles are updated according to their fitness values, so that candidate particles with higher fitness values ​​have a higher retention probability in the update, and thus multiple hybrid scheduling directions are formed.

[0052] Finally, when performing directional local rendering engine hybrid scheduling based on multiple hybrid scheduling directions and multiple rendering task unit index threads, each rendering task unit index thread is associated with its corresponding hybrid scheduling direction. Within the candidate particle range defined by the hybrid scheduling direction, target rendering engine combinations are selected from high to low fitness values, and these target rendering engine combinations are written into the scheduling result of that rendering task unit to form a local rendering engine hybrid scheduling strategy. By repeatedly performing the steps of associating hybrid scheduling directions, sorting fitness values, and selecting target rendering engine combinations on multiple rendering task unit index threads, multiple local rendering engine hybrid scheduling strategies and multiple target rendering engine combinations are finally obtained.

[0053] Furthermore, in the method provided in the application embodiments, the global joint scheduling of the multiple local rendering engine hybrid scheduling strategies to obtain a global hybrid scheduling strategy further includes: Obtain multiple importance weights and schedulable resource sets for multiple rendering task units; based on the multiple importance weights, perform global joint scheduling of the schedulable resource sets using multiple local rendering engine hybrid scheduling strategies to obtain a global hybrid scheduling strategy.

[0054] In this embodiment, to obtain multiple importance weights for multiple rendering task units, the multiple rendering task units currently participating in the scheduling are first summarized, and a weight calculation record is established for each rendering task unit. Then, the mapping relationship between the measurement point situation information and the associated rendering task unit set is read to determine whether each rendering task unit is associated with the measurement point situation information. If an association exists, the situation association value is recorded as 1 in the weight calculation record of that rendering task unit; otherwise, it is recorded as 0. Next, the task description vector corresponding to the rendering task unit is read, and real-time features are extracted from the task description vector. The refresh frequency requirement in the real-time features is converted into discrete values ​​according to a preset rule, for example, a high refresh frequency requirement is set to 3, a medium refresh frequency requirement is set to 2, and a low refresh frequency requirement is set to 1. This value is written into the weight calculation record. Simultaneously, the response time requirement is converted into discrete values ​​according to a preset rule, for example, a short response time requirement is set to 3, a medium response time requirement is set to 2, and a long response time requirement is set to 1, and this value is also written into the weight calculation record. Next, stability constraints are read, and rendering continuity requirements are converted according to whether continuous display is mandatory. A value of 2 is assigned for mandatory continuous display, and a value of 1 is assigned for allowing interruptions. This value is then written into the weight calculation record. Next, multi-model combination features are read, and the number of model types involved in the rendering task unit is counted. The number of model types is converted into discrete values ​​based on their order of magnitude; for example, a value of 1 is assigned for involving one model, 2 for involving two models, and 3 for involving three or more models. This value is then written into the weight calculation record. After filling in the weight calculation records for each rendering task unit, the situational correlation value, refresh rate value, response time value, stability constraint value, and number of model types are summed for each rendering task unit to obtain the corresponding raw weight value. Then, the raw weight values ​​of all rendering task units are normalized by dividing the raw weight value of each rendering task unit by the sum of the raw weight values ​​of all rendering task units. This ensures that the sum of the normalized importance weights of all rendering task units is 1, thus obtaining multiple importance weights for multiple rendering task units.

[0055] Meanwhile, in order to obtain the set of schedulable resources, the system reads the list of multiple graphics rendering engines that have been connected to the digital twin platform, queries the current running status and occupancy status of each graphics rendering engine one by one, registers the graphics rendering engines that are running and have not been allocated to other rendering task units as available resources, and records their engine identifier and availability status. Finally, all available graphics rendering engines are aggregated to form a set of schedulable resources.

[0056] Finally, when globally co-scheduling multiple local rendering engine hybrid scheduling strategies on the schedulable resource set based on multiple importance weights, the multiple rendering task units are first sorted according to the normalized importance weights to form a scheduling order. Then, according to this scheduling order, the target rendering engine combinations determined in the local rendering engine hybrid scheduling strategy are allocated to the rendering task units sequentially. Rendering task units with higher importance weights are prioritized when the schedulable resource set allows. When the schedulable resources are insufficient to simultaneously satisfy all rendering task units, the resource allocation is adjusted according to the proportion of importance weights to ensure consistency between the allocation results and the importance weights. Finally, the resource allocation results and execution order of each rendering task unit are unified and organized to form a global hybrid scheduling strategy based on importance weights under the constraints of the schedulable resource set.

[0057] Step S500: Based on the global hybrid scheduling strategy and the combination of the multiple target rendering engines, drive multiple graphics rendering engines to execute rendering tasks in parallel.

[0058] In this embodiment, based on a global hybrid scheduling strategy and a combination of multiple target rendering engines, multiple rendering task units are first associated with their corresponding target rendering engines one by one. A unified rendering execution arrangement is then generated according to the execution order, resource allocation relationships, and parallel constraints determined in the global hybrid scheduling strategy. Subsequently, according to the rendering execution arrangement, the rendering instructions of each rendering task unit are distributed to the respective graphics rendering engines in its corresponding target rendering engine combination, ensuring that each graphics rendering engine only processes the rendering task units assigned to it. During the rendering execution phase, multiple graphics rendering engines synchronously start rendering processing according to the global hybrid scheduling strategy. Under parallel execution conditions, they perform rendering calculations on the rendering task units they are responsible for and output rendering results, thereby achieving parallel rendering execution of multiple graphics rendering engines under the unified control of the global hybrid scheduling strategy.

[0059] In summary, the embodiments of this application have at least the following technical effects: This application constructs a digital twin 3D scene including an oblique photogrammetry model, a BIM model, a laser point cloud model, an underground pipeline model, and a terrain model, and forms a multi-source heterogeneous model element library through unified spatial coordinate calibration and semantic annotation; it acquires multiple graphics rendering engines from the digital twin platform, performs rendering performance analysis, and obtains the performance characteristics of multiple graphics rendering engines; it decomposes the digital twin rendering requirements into multiple rendering task units, and constructs multiple task description vectors for multiple rendering task units; it dynamically updates the scheduling parameters of the multiple rendering task units based on real-time situational data, and introduces historical rendering experience memory into the rendering scheduling process, optimizes the rendering engine hybrid scheduling strategy for multiple rendering task units, and obtains multiple target rendering engine combinations and a global hybrid scheduling strategy; according to the global hybrid scheduling strategy and the multiple target rendering engine combinations, it drives multiple graphics rendering engines to execute rendering tasks in parallel. This invention addresses the technical problem of low overall rendering efficiency in existing digital twin scenarios where multiple graphics rendering engines struggle to coordinate efficiently according to rendering requirements. By introducing a multi-graphics engine hybrid scheduling mechanism based on rendering task unit decomposition and combining real-time situational data with historical rendering experience, this invention improves the parallel coordination efficiency of multiple graphics rendering engines and enhances the overall rendering performance of digital twin scenarios.

[0060] Example 2, based on the same inventive concept as the multi-graphics engine scheduling and rendering method for digital twins in the foregoing examples, such as... Figure 2 As shown, this application provides a multi-graphics engine scheduling and rendering system for digital twins. The system and method embodiments in this application are based on the same inventive concept. The system includes: The feature library generation module 11 is used to construct a digital twin 3D scene including oblique photogrammetry model, BIM model, laser point cloud model, underground pipeline model and terrain model, and form a multi-source heterogeneous model feature library through unified spatial coordinate calibration and semantic annotation; the performance analysis module 12 is used to obtain multiple graphics rendering engines of the digital twin platform, perform rendering performance analysis, and obtain the performance characteristics of multiple graphics rendering engines; the decomposition module 13 is used to decompose the digital twin rendering requirements into multiple rendering task units and construct multiple task description vectors for multiple rendering task units; the strategy optimization module 14 is used to dynamically update the scheduling parameters of the multiple rendering task units based on real-time situational data, and introduce historical rendering experience memory in the rendering scheduling process to optimize the rendering engine hybrid scheduling strategy of multiple rendering task units, and obtain multiple target rendering engine combinations and global hybrid scheduling strategy; the driving module 15 is used to drive multiple graphics rendering engines to execute rendering tasks in parallel according to the global hybrid scheduling strategy and the multiple target rendering engine combinations.

[0061] Furthermore, the system is also used to implement the following functions: The oblique photogrammetry model, BIM model, laser point cloud model, underground pipeline model, and terrain model are uniformly processed using a spatial reference system. After uniform processing, the models are semantically annotated to obtain additional semantic attribute information for each model. The additional semantic attribute information includes model category, functional attributes, and spatial hierarchy. The oblique photogrammetry model, BIM model, laser point cloud model, underground pipeline model, and terrain model that have completed coordinate calibration and semantic annotation are stored in a multi-source heterogeneous model element library.

[0062] Furthermore, the system is also used to implement the following functions: The process involves iterating through the multiple graphics rendering engines to extract historical rendering records, resulting in multiple sets of historical rendering records; then, based on these sets, the process involves scene-based aggregation and performance analysis of rendering scenarios, resulting in multiple sets of scene-based performance features for the graphics rendering engines; finally, the process involves iterating through these sets of scene-based performance features for the graphics rendering engines to identify scenarios, resulting in multiple performance features for the graphics rendering engines.

[0063] Furthermore, the system is also used to implement the following functions: Extract a first historical rendering record set from multiple historical rendering record sets and aggregate it for rendering scenarios to obtain multiple first aggregated historical rendering record sets; traverse the multiple first aggregated historical rendering record sets to extract performance features to obtain multiple first aggregated performance feature sets; perform mean shift filtering on the performance features within each of the multiple first aggregated performance feature sets to obtain multiple filtered aggregated performance features; add the multiple filtered aggregated performance features as a first graphics rendering engine scenario-based performance feature set to the multiple graphics rendering engine scenario-based performance feature sets.

[0064] Furthermore, the system is also used to implement the following functions: Each of the multiple task description vectors includes multi-model combined features, real-time features, and stability constraints.

[0065] Furthermore, the system is also used to implement the following functions: The real-time situational data is analyzed to extract situational information related to spatial location, object status, and event type, obtaining situational information from multiple measurement points. From the two dimensions of spatial location and business association, the situational information from multiple measurement points is mapped to the corresponding rendering task units to obtain multiple sets of associated rendering task units, and a mapping relationship between the situational information from measurement points and the sets of associated rendering task units is constructed. Based on the mapping relationship between the situational information from measurement points and the sets of associated rendering task units, the rendering engine hybrid scheduling strategy is optimized to obtain a combination of multiple target rendering engines and a global hybrid scheduling strategy.

[0066] Furthermore, the system is also used to implement the following functions: Using multiple rendering task units as indexes, the mapping relationship between measurement point situation information and associated rendering task unit sets is retrieved to determine multiple rendering task unit index clues; the multiple rendering task units are traversed to extract multiple historical rendering task unit index clue sets within a preset historical window; historical rendering experience memory is introduced, and directional local rendering engine hybrid scheduling is performed on the multiple rendering task unit index clues through the multiple historical rendering task unit index clue sets to obtain multiple local rendering engine hybrid scheduling strategies and multiple target rendering engine combinations; the multiple local rendering engine hybrid scheduling strategies are globally jointly scheduled to obtain a global hybrid scheduling strategy.

[0067] Furthermore, the system is also used to implement the following functions: Based on the multiple rendering task unit index clues, experience-memory similarity matching is performed on the multiple historical rendering task unit index clue sets to obtain multiple matching historical rendering task unit index clue sets; multiple historical rendering engine combination sets and multiple historical rendering effect sets are obtained from the multiple matching historical rendering task unit index clue sets; multiple hybrid scheduling directions are determined based on the multiple historical rendering engine combination sets and multiple historical rendering effect sets; directional local rendering engine hybrid scheduling is performed according to the multiple hybrid scheduling directions and multiple rendering task unit index clues to obtain multiple local rendering engine hybrid scheduling strategies and multiple target rendering engine combinations.

[0068] Furthermore, the system is also used to implement the following functions: Obtain multiple importance weights and schedulable resource sets for multiple rendering task units; based on the multiple importance weights, perform global joint scheduling of the schedulable resource sets using multiple local rendering engine hybrid scheduling strategies to obtain a global hybrid scheduling strategy.

[0069] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0070] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A multi-graphics engine scheduling and rendering method for digital twins, characterized in that, The method includes: A digital twin 3D scene is constructed, which includes oblique photogrammetry model, BIM model, laser point cloud model, underground pipeline model and terrain model, and a multi-source heterogeneous model element library is formed through unified spatial coordinate calibration and semantic annotation; Obtain multiple graphics rendering engines from the digital twin platform, perform rendering performance analysis, and obtain the performance characteristics of multiple graphics rendering engines; The rendering requirements of digital twins are broken down into multiple rendering task units, and multiple task description vectors are constructed for these multiple rendering task units. The scheduling parameters of the multiple rendering task units are dynamically updated based on real-time situational data, and historical rendering experience memory is introduced into the rendering scheduling process. The rendering engine hybrid scheduling strategy of multiple rendering task units is optimized to obtain a combination of multiple target rendering engines and a global hybrid scheduling strategy. Based on the global hybrid scheduling strategy and the combination of multiple target rendering engines, multiple graphics rendering engines are driven to execute rendering tasks in parallel.

2. The multi-graphics engine scheduling and rendering method for digital twins as described in claim 1, characterized in that, A digital twin 3D scene was constructed, comprising oblique photogrammetry model, BIM model, laser point cloud model, underground pipeline model, and terrain model. A multi-source heterogeneous model element library was formed through unified spatial coordinate calibration and semantic annotation, including: The oblique photogrammetry model, BIM model, laser point cloud model, underground pipeline model, and terrain model are processed using a unified spatial reference system. After unified processing by the spatial reference system, the models are semantically annotated to obtain additional semantic attribute information for each model. The additional semantic attribute information includes model category, functional attributes, and spatial hierarchy. The oblique photogrammetry model, BIM model, laser point cloud model, underground pipeline model, and terrain model, which have completed coordinate calibration and semantic annotation, will be stored in the multi-source heterogeneous model element library.

3. The multi-graphics engine scheduling and rendering method for digital twins as described in claim 1, characterized in that, The system retrieves multiple graphics rendering engines from the digital twin platform, performs rendering performance analysis, and obtains the performance characteristics of these engines, including: The historical rendering records are extracted by traversing the multiple graphics rendering engines to obtain multiple sets of historical rendering records; Based on the multiple historical rendering record sets, rendering scene aggregation and rendering scene performance analysis are performed to obtain multiple graphics rendering engine scene performance feature sets. The multiple graphics rendering engine scene-specific performance feature sets are traversed and scene-specific identification is performed to obtain the performance features of multiple graphics rendering engines.

4. The multi-graphics engine scheduling and rendering method for digital twins as described in claim 3, characterized in that, Based on the aforementioned sets of historical rendering records, rendering scene aggregation and rendering scene performance analysis are performed to obtain multiple sets of scene-based performance features for graphics rendering engines, including: Extract the first historical rendering record set from multiple historical rendering record sets and aggregate them in a rendering scene to obtain multiple first aggregated historical rendering record sets; The multiple first aggregated historical rendering record sets are traversed to extract performance features, resulting in multiple first aggregated performance feature sets. Each of the multiple first aggregate performance feature sets is filtered by mean shift of performance features within the set to obtain multiple filtered aggregate performance features; The multiple filtered and aggregated performance features are used as the first set of scene-based performance features for graphics rendering engines, and then added to the set of scene-based performance features for multiple graphics rendering engines.

5. The multi-graphics engine scheduling and rendering method for digital twins as described in claim 1, characterized in that, Each of the multiple task description vectors includes multi-model combined features, real-time features, and stability constraints.

6. The multi-graphics engine scheduling and rendering method for digital twins as described in claim 1, characterized in that, The scheduling parameters of the multiple rendering task units are dynamically updated based on real-time situational data. Historical rendering experience is incorporated into the rendering scheduling process, and a mixed scheduling strategy for multiple rendering task units is optimized to obtain a combination of multiple target rendering engines and a global mixed scheduling strategy, including: The real-time situational data is analyzed to extract situational information related to spatial location, object status, and event type, thereby obtaining situational information from multiple measurement points. From the two dimensions of spatial location and business association, the situation information of the multiple measurement points is mapped to the corresponding rendering task units to obtain multiple sets of associated rendering task units, and a mapping relationship between the situation information of measurement points and the sets of associated rendering task units is constructed. Based on the mapping relationship between measurement point situation information and associated rendering task unit set, the rendering engine hybrid scheduling strategy is optimized to obtain multiple target rendering engine combinations and a global hybrid scheduling strategy.

7. The multi-graphics engine scheduling and rendering method for digital twins as described in claim 6, characterized in that, Based on the mapping relationship between measurement point situation information and associated rendering task unit sets, the rendering engine hybrid scheduling strategy is optimized to obtain multiple target rendering engine combinations and a global hybrid scheduling strategy, including: Using multiple rendering task units as indexes, the mapping relationship between measurement point situation information and associated rendering task unit sets is retrieved to determine multiple rendering task unit index clues; Traverse the multiple rendering task units to extract a set of index clues for multiple historical rendering task units within a preset history window; By introducing historical rendering experience memory, and using multiple sets of historical rendering task unit index clues to perform directional local rendering engine hybrid scheduling on the multiple rendering task unit index clues, multiple local rendering engine hybrid scheduling strategies and multiple target rendering engine combinations are obtained. A global hybrid scheduling strategy is obtained by performing global joint scheduling on the hybrid scheduling strategies of the multiple local rendering engines.

8. The multi-graphics engine scheduling and rendering method for digital twins as described in claim 7, characterized in that, By incorporating historical rendering experience memory, and using multiple sets of historical rendering task unit index threads to perform directional local rendering engine hybrid scheduling on these multiple rendering task unit index threads, a hybrid scheduling strategy for multiple local rendering engines and a combination of multiple target rendering engines are obtained, including: Based on the multiple rendering task unit index clues, experience memory similarity matching is performed on the multiple historical rendering task unit index clue sets to obtain multiple matching historical rendering task unit index clue sets. Obtain a set of multiple historical rendering engine combinations and a set of multiple historical rendering effects that match multiple sets of historical rendering task unit index clues; Based on the aforementioned set of multiple historical rendering engine combinations and multiple sets of historical rendering effects, multiple hybrid scheduling directions are determined; Based on the multiple hybrid scheduling directions and multiple rendering task unit index clues, directional local rendering engine hybrid scheduling is performed to obtain multiple local rendering engine hybrid scheduling strategies and multiple target rendering engine combinations.

9. The multi-graphics engine scheduling and rendering method for digital twins as described in claim 7, characterized in that, A global joint scheduling strategy is obtained by performing global joint scheduling on the multiple local rendering engine hybrid scheduling strategies, including: Obtain multiple importance weights and schedulable resource sets for multiple rendering task units; Based on the multiple importance weights, the schedulable resource set is globally jointly scheduled using multiple local rendering engine hybrid scheduling strategies to obtain a global hybrid scheduling strategy.

10. A multi-graphics engine scheduling and rendering system for digital twins, characterized in that, The system is used to execute the multi-graphics engine scheduling and rendering method for digital twins as described in any one of claims 1-9, the system comprising: The feature library generation module is used to construct a digital twin 3D scene containing oblique photogrammetry model, BIM model, laser point cloud model, underground pipeline model and terrain model, and form a multi-source heterogeneous model feature library through unified spatial coordinate calibration and semantic annotation. The performance analysis module is used to obtain multiple graphics rendering engines of the digital twin platform, perform rendering performance analysis, and obtain the performance characteristics of multiple graphics rendering engines. The decomposition module is used to decompose the digital twin rendering requirements into multiple rendering task units and construct multiple task description vectors for multiple rendering task units. The strategy optimization module is used to dynamically update the scheduling parameters of the multiple rendering task units based on real-time situational data, and to introduce historical rendering experience memory into the rendering scheduling process. It optimizes the rendering engine hybrid scheduling strategy for multiple rendering task units to obtain a combination of multiple target rendering engines and a global hybrid scheduling strategy. The driving module is used to drive multiple graphics rendering engines to execute rendering tasks in parallel according to the global hybrid scheduling strategy and the combination of multiple target rendering engines.

Images