Lightweight schedule simulation method for large-scale, multi-scale, multi-entity projects
By unifying multi-source 3D data and building a microservice architecture simulation engine, the problems of large data volume and high rendering load in large-scale, multi-entity engineering projects are solved, and the consistency of progress mapping and continuous verification of simulation results under multi-entity collaboration are achieved.
Patent Information
- Application Number
- CN202611612866.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-19
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies struggle to address the challenges of large-scale, multi-scale, and multi-entity engineering projects, such as the large volume of multi-source heterogeneous 3D data, high computational load for loading and rendering, and inconsistent granularity of progress mapping under multi-entity collaboration. This makes it difficult for engineering progress management methods to achieve lightweight 3D scene construction, multi-scale progress simulation, and multi-entity collaborative applications.
By collecting BIM model data, GIS terrain data, and oblique photography data, coordinate unification and format normalization are performed to build a unified data base. A simulation engine is built based on a microservice architecture. Combined with the construction schedule and process model, the spatiotemporal evolution rules of construction are defined to drive the dynamic updating of 3D model components, output 3D dynamic simulation results, and perform consistency verification and parameter correction through closed-loop training.
It achieves lightweight processing of multi-source heterogeneous 3D data, reduces the computational load of terminal rendering, solves the problem of inconsistent progress mapping granularity under multi-subject collaboration, and realizes continuous verification and consistency check of simulation results.
Smart Images

Figure CN123116382A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of information technology and computer simulation in engineering construction, specifically a lightweight progress simulation method for large-scale, multi-scale, multi-entity projects. Background Technology
[0002] Existing large-scale, multi-scale, multi-entity engineering project schedule management usually requires the integration of BIM model data, geographic information terrain data, and on-site 3D real-scene data to uniformly manage and display the construction plan, actual progress, and component status.
[0003] In related technologies, in order to visualize the progress of a project, the construction process can be displayed based on a 3D model; or, the construction plan can be compared with the data reported on site to determine whether there are any deviations in the project schedule.
[0004] However, in scenarios where projects involve multiple types of work sites and collaboration among multiple participating entities, the former approach is easily affected by factors such as the large scale of multi-source heterogeneous 3D data, high computational load of terminal rendering, and inconsistent component mapping granularity. The latter approach, on the other hand, struggles to simultaneously accommodate the access of progress data from multiple entities, 3D dynamic simulation display, and continuous verification and optimization of simulation results. Therefore, the project progress management methods in related technologies are difficult to simultaneously achieve lightweight 3D scene construction, multi-scale progress simulation, and multi-entity collaborative applications. Summary of the Invention
[0005] The purpose of this invention is to provide a lightweight progress simulation method for large-scale, multi-scale, multi-agent projects, addressing the technical problems of large volumes of multi-source heterogeneous 3D data, high computational loads for loading and rendering, and inconsistent progress mapping granularity under multi-agent collaboration. To solve the above technical problems, the technical solution of this invention includes:
[0006] A lightweight schedule simulation method for large-scale, multi-scale, multi-agent projects includes the following steps:
[0007] Sp1: Collect multi-source heterogeneous 3D data including BIM model data, GIS terrain data and oblique photogrammetry data, unify coordinates and normalize formats to form a unified data base, and generate an oblique photogrammetry real-scene 3D model based on the oblique photogrammetry data.
[0008] Sp2: Lightweight processing of the unified data base plate, registration and fusion of each three-dimensional data in the unified data base plate, and construction of an engineering three-dimensional digital base plate containing three-dimensional model components;
[0009] SP3: A simulation engine built on a microservice architecture, incorporating microservices for model management, rendering, computation scheduling, and data interfaces.
[0010] Sp4: Obtain the construction schedule, extract the three-dimensional model components from the three-dimensional digital base plate of the project, establish mapping relationships and construct a multi-scale simulation view, and combine the construction process model corresponding to the large-scale multi-scale multi-subject project to define the construction spatiotemporal evolution rules including efficiency and resource allocation parameters.
[0011] Sp5: Acquire construction progress data, drive the dynamic update of the three-dimensional model components according to the construction spatiotemporal evolution rules, compare and analyze the deviation between the construction progress data and the construction progress plan, and output the three-dimensional dynamic simulation results.
[0012] Sp6: Perform closed-loop training on the simulation model, verify the consistency between the three-dimensional dynamic simulation results and the construction progress data, and correct the efficiency and resource allocation parameters in the construction spatiotemporal evolution rules based on the deviation analysis results.
[0013] Preferably, in step Sp1, the BIM model data includes BIM model files in open standard format and BIM model files in proprietary format, and extracts component geometric information, component attribute information and spatial relationship information, and establishes a BIM model database according to work points.
[0014] Preferably, in step Sp1, the GIS terrain data includes DEM, DOM, and TIN data, and covers the surface area within the engineering route and its construction impact boundary.
[0015] Preferably, in step Sp2, the lightweighting process includes geometric simplification, which employs a progressive mesh simplification algorithm based on edge folding and combines a quadratic error metric and attribute error to construct a folding cost function; for regular geometric components, parametric surface fitting is used to replace discrete triangular meshes.
[0016] Preferably, in step Sp2, the lightweighting process includes LOD multi-level detail model generation, which includes generating multiple precision levels for each three-dimensional model component in the engineering three-dimensional digital base plate, and automatically switching between precision levels according to the viewpoint distance.
[0017] Preferably, in step Sp2, the lightweight processing of the oblique photogrammetry real-world 3D model includes: performing tile-based cutting on the oblique photogrammetry real-world 3D model and extracting local textures and progressive loading, and combining texture atlas technology to perform atlas merging on the local textures after tile-based cutting.
[0018] Preferably, in step Sp3, the rendering microservice is deployed on a GPU server cluster and the rendering results are transmitted to the client browser as a video stream through a server-side remote distributed rendering method; the computation scheduling microservice allocates progress simulation computation tasks based on a task queue architecture.
[0019] Preferably, the multi-scale simulation view establishes construction process models for full-section hard rock tunnel boring machine tunnels, shield tunnels, pipe jacking, buried pipes, and pumping stations, and associates each process with the state change rules of the corresponding three-dimensional model components.
[0020] Preferably, in step Sp5, the construction progress data includes the actual start date, completed work volume, current construction status, and construction resource input, and corresponding early warning prompt data instructions are generated based on time deviation, work volume deviation, and critical path deviation.
[0021] Preferably, in step Sp6, the closed-loop training includes: consistency verification between the simulation results and the construction progress data;
[0022] Construction dimension deviation matrix ,in The number of evaluation units included in the construction progress data, where each evaluation unit corresponds to a work point, process, or section, and the column dimensions strictly correspond to the dimensionless time deviation, spatial location deviation, and quantity deviation, respectively, and a multi-dimensional accuracy evaluation is performed based on the deviation matrix; and parameter correction based on deviation feedback, wherein the parameter correction adopts an adaptive learning rate gradient descent optimization method.
[0023] Compared with the prior art, the present invention has the following beneficial effects:
[0024] This invention solves the problem of large data volume from multiple sources by collecting heterogeneous 3D data including BIM model data, GIS terrain data, and oblique photogrammetry data, unifying coordinates and normalizing formats, and then lightweight processing, registration, and fusion to construct a 3D digital base plate for engineering. At the same time, it builds a simulation engine based on a microservice architecture, which includes microservices for model management, rendering, computation scheduling, and data interfaces, effectively reducing the terminal rendering computation load.
[0025] This invention solves the problem of inconsistent progress mapping granularity under multi-subject collaboration by extracting three-dimensional model components to establish mapping relationships and constructing multi-scale simulation views. By defining construction spatiotemporal evolution rules containing efficiency and resource allocation parameters, it drives the dynamic updating of components and outputs three-dimensional dynamic simulation results. Combined with closed-loop training of the simulation model, consistency verification and parameter correction are performed based on deviation analysis results, realizing continuous verification of simulation results. Attached Figure Description
[0026] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. The drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0027] Figure 1 This is a flowchart of a lightweight progress simulation method for large-scale, multi-scale, multi-entity projects provided in this embodiment of the invention. Detailed Implementation
[0028] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.
[0029] Please see the appendix Figure 1 A lightweight schedule simulation method for large-scale, multi-scale, multi-entity projects includes the following steps:
[0030] Sp1: Collect multi-source heterogeneous 3D data including BIM model data, GIS terrain data and oblique photogrammetry data, unify coordinates and normalize formats to form a unified data base, and output oblique photogrammetry real-scene 3D model based on oblique photogrammetry data.
[0031] Sp2: Lightweight processing of the unified data base, registering and fusing the various 3D data within the unified data base to construct an engineering 3D digital base containing 3D model components;
[0032] SP3: A simulation engine built on a microservice architecture, incorporating microservices for model management, rendering, computation scheduling, and data interfaces.
[0033] SP4: Obtain the construction schedule, extract the 3D model components from the 3D digital base plate of the project, establish mapping relationships and construct multi-scale simulation views, and combine the construction process model corresponding to large-scale multi-scale and multi-subject projects to define the spatiotemporal evolution rules of construction with efficiency and resource allocation parameters.
[0034] Sp5: Acquire construction progress data, drive dynamic updates of 3D model components according to construction spatiotemporal evolution rules, compare and analyze deviations between construction progress data and construction progress plan, and output 3D dynamic simulation results.
[0035] SP6: Perform closed-loop training on the simulation model, verify the consistency between the three-dimensional dynamic simulation results and the construction progress data, and correct the efficiency and resource allocation parameters in the construction spatiotemporal evolution rules based on the deviation analysis results.
[0036] In an exemplary embodiment, in a real-world application scenario: the construction progress management of large-scale water conservancy projects needs to simultaneously address the data organization issues arising from differences in the scale of projects along the route, the number of participating units, and the types of work sites;
[0037] This implementation method is mainly used to illustrate the processing path of this type of project in the process of unifying the three-dimensional base plate, progress simulation and data verification. Its main implementation steps are to unify the BIM model, GIS terrain and oblique photogrammetry model to the same coordinate reference, and then perform lightweight and fusion processing so that the subsequent simulation and display are all based on the same engineering three-dimensional digital base plate.
[0038] Among them, BIM model data usually comes from component models formed during the design and refinement phases, GIS topographic data outputs the surface environment within the engineering route and construction impact range, and oblique photography data is used to supplement the actual appearance of features on site.
[0039] Because the coordinate systems and file formats of various types of data are inconsistent, the system first completes coordinate transformation and format sorting, and then writes the model files, terrain data and real scene models into a unified data base. The processing of the data base is not a simple data compression, but rather geometric simplification, detail level output, texture compression, instantiation rendering and redundant geometry cleanup according to components, terrain and real scene models respectively.
[0040] For the real-world 3D model formed by oblique photography, tile cutting and progressive loading will be performed to reduce the memory and rendering resource consumption of the client's computing device. After these processes are completed, the BIM model, terrain data and real-world model will be spatially registered to align them under the same engineering conditions, forming a 3D digital base plate for engineering that can be used for subsequent construction progress simulation.
[0041] To ensure that the flow of heterogeneous data within a unified data platform has executable control logic, the system assigns five processing states to each batch of imported objects in sequence: pending parsing, normalized, pending registration, merged, and published. A separate pending review state is set up as an abnormal suspended state. This state is not included in the above five normal processing states and can only return to the pending registration state after correction is completed and the coordinate range and spatial integrity checks are passed again.
[0042] The pending parsing status only allows entry into the format recognition and coordinate check process, the normalized status can enter the lightweight process, the pending registration status is used to wait for the BIM model, terrain data and real scene model under the same work point to be completed before spatial alignment is performed, the merged status indicates that a unified work condition index has been output, and the published status indicates that the batch of data can be called by the simulation task.
[0043] If a data object still exceeds the boundary of the engineering line after coordinate unification, or lacks a minimum identifiable spatial range, it will be marked as pending review and will not enter the subsequent fusion stage for the time being; the object pending review will not participate in the simulation calculation of the current scheduling cycle, but the original import record will be retained so that it can be re-entered into the queue after subsequent correction;
[0044] For large-scale model data that is continuously accessed, the system receives data in segments by work point and uses sequential submission to control concurrency: only one fusion write process is allowed to be in the submission state at the same work point at the same time, and different work points can be processed in parallel, thereby avoiding the repeated overwriting of component indexes in the same spatial range within a submission cycle.
[0045] The simulation engine adopts a microservice approach, which is divided into several independent units: model management, rendering, computation scheduling, and data interface. The model management microservice is responsible for storing the lightweight model and its index, the rendering service is responsible for converting model changes into video streams that can be received by the browser, the computation scheduling microservice is responsible for receiving the schedule plan and organizing simulation tasks, and the data interface microservice is responsible for accessing the progress data reported by each participating entity.
[0046] Therefore, model management, rendering, and computation are not directly coupled, and updating one part does not require modifying the entire process. Furthermore, the model management microservice only publishes the data version that has been published, and the computation scheduling microservice reads only the published version at the beginning of each scheduling cycle and keeps the version fixed within the cycle. If a new version of the base data is generated during the execution of the cycle, the new version will be postponed to the next scheduling cycle, thereby avoiding inconsistencies in the planned mapping results caused by switching the model base data within the same cycle.
[0047] In the progress simulation part, the system first retrieves the construction progress plan from the computation scheduling microservice, and then establishes a correspondence between the work processes in the plan and the 3D model components. After the correspondence is established, the system defines the construction spatiotemporal evolution rules based on the efficiency parameters and resource configuration parameters, and organizes the entire project into different simulation views according to three levels: overall project, work area and work point, and work process components.
[0048] Therefore, the same set of progress data can be used to view the overall progress or to locate a single work point or a single process component for viewing. To avoid inconsistencies in granularity between planned tasks and model components, the mapping relationship adopts a three-level association method of task identifier, work point identifier, and component identifier: a planned task can correspond to one or more components, and a component belongs to only one process task in the execution state at the same time.
[0049] When the same component involves consecutive processes, its assigned task is switched according to the planned start and end time sequence, instead of repeatedly counting the completion of two tasks at the same calculation time; for linear work points that are continuously advanced by section, the system uses the section number as the intermediate association unit, first collects the planned work quantity to the section, and then the section is allocated to the component, so as to ensure that the cumulative work quantity is consistent with the component status update.
[0050] If the task level in the construction schedule is deeper than the component level, the lower-level tasks are first merged into executable segments according to the same process, the same section, and consecutive planned time periods, and then a mapping is established with the components. If the task level is shallower than the component level, the task workload is split according to the section to which the component belongs and the component weight. After splitting, the sum of the cumulative workloads allocated to each component remains consistent with the original task workload to avoid the situation where the total planned amount and the total component amount are not closed in the subsequent deviation analysis.
[0051] After each participating entity reports real-time progress data through the data interface microservice, the system compares the measured progress with the planned progress, drives the state changes of the 3D model components according to the construction time and space evolution rules, and outputs dynamic simulation images in the rendering service.
[0052] If there is a deviation between the reported data and the plan, the time deviation, workload deviation and critical path deviation will be recorded in the deviation analysis, and the deviation information will be sent back to the simulation model for subsequent display and judgment. The real-time reporting here is used to indicate continuous data access and does not require all participating entities to complete synchronous submission at the same physical moment. The system outputs progress snapshots with a unified scheduling cycle. Within the same scheduling cycle, the progress data received in that cycle is frozen first, and then a plan comparison, status update and screen rendering are performed.
[0053] If a certain progress data arrives later than the deadline of the current scheduling cycle, it will be incorporated into the next scheduling cycle for processing, thereby decoupling the data access time from the simulation calculation time and avoiding the repeated rewriting of the same component state within a calculation cycle due to communication delay.
[0054] During the simulation result verification and calibration phase, the system compares the consistency of the three-dimensional dynamic simulation results with the measured construction data. If it is found that the simulation results of certain work points deviate continuously from the measured progress, the efficiency parameters and resource allocation parameters are corrected, and the corrected parameters are written back to the construction spatiotemporal evolution rules. If the measured progress data is missing in a certain period, the stable record of the previous effective period is used first, and the record is marked as a substitute value to ensure that the progress simulation will not stop due to the interruption of data communication of local computing nodes.
[0055] A lightweight progress simulation method for large-scale, multi-scale, multi-entity projects, in step Sp1, BIM model data includes BIM model files in open standard format and BIM model files in special format, and extracts component geometric information, component attribute information and spatial relationship information, and establishes a BIM model database according to work points;
[0056] In the design and construction coordination of large-scale water conservancy projects, BIM models usually exist in both open and proprietary formats. If they are simply stored as files, inconsistencies in data granularity can easily occur during component retrieval, schedule mapping, and work point association. Therefore, this implementation method first distinguishes between public and proprietary BIM model formats after collecting the BIM model, then extracts component geometric information, component attribute information, and spatial relationship information, and establishes a BIM model database by work point.
[0057] As a result, components no longer exist as simple file objects, but are organized into searchable and associative database records; for components under the same work site, the system can directly retrieve their geometric contours, attribute fields and upstream and downstream relationships, providing a basis for subsequent mapping output between construction schedule plans and components;
[0058] If a certain format file has missing local attributes, the identifiable geometric information is retained and the missing field is kept as a null record to avoid affecting the entire work site model from being entered into the database due to the incompleteness of a single attribute item. To avoid duplicate entry of the same component into the database between open format and proprietary format, the system first outputs candidate corresponding groups based on the component's spatial location, component type, and adjacency relationship, and then merges and judges them in the order of geometry first and then attributes: when the enclosing range of two candidate components overlaps and their main axes are consistent, they are first regarded as the same candidate object.
[0059] Based on this, if the component types are consistent and there is no conflict in the adjacent relationship, they are merged into a unified component record, and the set of attributes with higher source priority is retained as the current valid attributes, while the remaining source information is retained as the basis for tracing. If the geometry is similar but the attributes conflict, the system does not directly overwrite, but marks the component as conflict pending and suspends its automatic mapping until subsequent manual review or other related components at the same work site are completed before it is lifted.
[0060] A lightweight progress simulation method for large-scale, multi-scale, multi-entity projects. In step Sp1, the GIS terrain data includes DEM, DOM, and TIN data, and covers the surface area within the engineering route and its construction impact boundary.
[0061] When the engineering route is long and the terrain conditions are significantly different, using only a single terrain data type is insufficient to support the subsequent spatial registration and construction display. Therefore, this implementation method uses DEM, DOM and TIN data simultaneously during GIS data collection, and makes them cover the surface area within the engineering route and construction impact boundary. DEM is used to express elevation changes, DOM is used to express surface images, and TIN is used to express the three-dimensional surface morphology of the terrain.
[0062] After the three types of terrain data are entered into the unified data base, the system can form a continuous surface background and output spatial reference for subsequent BIM models, reality models and progress components. If the TIN data density of some local areas is insufficient, the combined results of DEM and DOM will be used first to supplement it, ensuring that the surface area remains continuously available within the scope of the project's influence, without requiring all areas to maintain the same level of accuracy.
[0063] For areas where the coverage of DEM, DOM, and TIN is not completely consistent, the system first forms a unified clipping range based on the engineering line and construction impact boundary, and then supplements it in the order of priority of elevation continuity, image continuity, and local surface details.
[0064] If the difference in elevation values from different sources within the same region exceeds the preset tolerance, the data with the latest collection timestamp will be used as the currently valid form, and the other source will be marked as pending verification and will not participate in the registration calculation in the current cycle, so as to reduce the repeated jumps of the spatial reference surface in adjacent scheduling cycles.
[0065] A lightweight progress simulation method for large-scale, multi-scale, multi-entity projects. In step Sp2, the lightweight processing includes geometric simplification. Geometric simplification adopts a progressive mesh simplification algorithm based on edge folding, and constructs a folding cost function by combining quadratic error metric and attribute error. For regular geometric components, parametric surface fitting is used to replace discrete triangular mesh.
[0066] A lightweight progress simulation method for large-scale, multi-scale, multi-entity projects. In step Sp2, lightweight processing includes LOD multi-level detail model output. LOD multi-level detail model output includes outputting multiple precision levels for each three-dimensional model component in the engineering three-dimensional digital base plate, and automatically switching between precision levels according to the viewpoint distance.
[0067] The lightweight progress simulation method for large-scale, multi-scale, multi-entity projects includes the following steps in Sp2: lightweight processing of the oblique photogrammetry real-scene 3D model: tile-cutting the oblique photogrammetry real-scene 3D model and extracting local textures and progressive loading, and combining texture atlas technology to perform atlas merging of the local textures after tile-cutting.
[0068] In the 3D working conditions of large-scale water conservancy projects, the number of model faces and the number of texture switching times will directly affect the loading and rendering efficiency; if only the original triangular mesh is retained, the model data will be too large, which is not conducive to the subsequent real-time simulation display; if it is compressed too much, the shape of the components will be difficult to identify.
[0069] Therefore, this implementation adopts a progressive mesh simplification algorithm with edge folding for geometric simplification, and uses a quadratic error metric and attribute error to jointly constitute the folding cost function; the quadratic error metric is used to represent the change in distance from the vertex of the candidate edge to the plane containing the adjacent triangle before and after folding, and the attribute error is used to represent the changes in the normal direction, texture coordinates and component boundary identifiers;
[0070] Each attribute error is normalized according to the maximum change range of the corresponding attribute within the current component. When the maximum change range is zero, the attribute error is recorded as zero. The system performs candidate edge folding in ascending order of the folding cost function value and prohibits folding the outer contour edge of the component, the boundary edge of different materials, and the edge with construction status boundary marker, thereby avoiding changes to the component identity or progress display boundary after lightweighting.
[0071] The weights of quadratic error and attribute error remain unchanged within the same batch. Ordinary internal meshes are dominated by quadratic error, while the weight of attribute error is increased in areas with concentrated component boundaries and texture features. To prevent triangular face flipping, non-manifold connections, or local gaps after folding, the system performs a local validity check before each candidate edge is folded: the fold is only accepted if the normals of adjacent triangular faces have not reversed after folding, the local adjacency relationship remains simply connected, and no duplicate edges are introduced; candidate edges that fail the validity check are skipped directly and do not participate in the current round of simplification.
[0072] For regular geometric components such as pipes and elbows, discrete triangular meshes are no longer maintained. Instead, parametric surface fitting is used to express the component as a surface. Regular geometric components are identified based on component type, cross-sectional dimension consistency, and axial continuity.
[0073] After fitting, the maximum spatial deviation between the fitted surface and the original mesh is compared with the allowable modeling error of the component. If the deviation does not exceed the allowable modeling error, the fitted surface is used; otherwise, the simplified triangular mesh is retained to prevent irregular local structures from being incorrectly replaced.
[0074] After geometric simplification, the system outputs multiple precision levels for each 3D model and switches between different levels based on the viewpoint distance. This implementation sets at least three precision levels (high, medium, and low) for each model, using the model's bounding sphere radius as the distance criterion: let the viewpoint distance be... The radius of the enclosing sphere is ;
[0075] when When, call the high-precision level; when When, call the precision level; when When viewing at close range, a lower precision level is called; when viewing at a distance, a higher precision level is called; and when viewing at a distance, a lower precision level is called to reduce unnecessary detail loading. To avoid frequent switching when the viewpoint moves near the level switching boundary, the system sets a hold interval on both sides of the current switching boundary, which is equivalent to 10% of the distance of the boundary.
[0076] Once the viewpoint enters the holding interval, it continues to maintain the current level until it crosses the outer boundary of the holding interval before switching is performed. For models with a bounding sphere radius of zero or where a valid bounding sphere cannot be obtained, the system determines the equivalent radius based on the longest side of the bounding box of its components. If the longest side of the bounding box is still zero, the high-precision level is used directly and does not participate in distance switching.
[0077] The judgment of layer switching, model replacement and visible status submission are all based on the same rendering schedule result. Only one effective precision layer is allowed for the same model within a certain scheduling time. If the target layer resource has not been fully loaded, the current layer will be maintained until the target layer is fully loaded before the whole layer is replaced, thereby avoiding the mixed display of high and low layers of the same model due to asynchronous loading in a single scheduling.
[0078] For oblique photogrammetry real-world 3D models, due to their large coverage area and abundant textures, this implementation method cuts them into spatial tiles and displays them step by step using a progressive loading method; the tiles are organized according to a spatial quadtree, with parent tiles storing low-precision models of the covered area and child tiles storing high-precision models of the corresponding local areas.
[0079] The client first requests the parent tile that intersects with the current view area, and then requests the child tiles according to the view distance and screen projection range. Low-precision tiles are given priority in the application, and high-precision tiles are added later. The texture part is merged into an atlas to reduce the number of texture bindings. Within the same texture atlas, tile textures with the same material type and spatial adjacency are merged first, and the texture coordinate index of each tile is preserved to avoid texture misalignment after merging.
[0080] If some tiles fail to load in time under the current network conditions, low-precision tiles will be displayed first, and then replaced with high-precision tiles after the high-precision tiles have been loaded and passed the integrity check, so as to ensure that the working conditions are always visible.
[0081] In step Sp3 of the lightweight progress simulation method for large-scale, multi-scale, multi-entity projects, the rendering microservice is deployed on a GPU server cluster and the rendering results are transmitted to the client browser as a video stream through server-side remote distributed rendering; the computation scheduling microservice allocates progress simulation computation tasks based on a task queue architecture.
[0082] A lightweight progress simulation method for large-scale, multi-scale, and multi-entity projects is proposed. The multi-scale simulation view establishes construction process models for full-section hard rock tunnel boring machine tunnels, shield tunnels, pipe jacking, buried pipes, and pumping stations, and associates each process with the state change rules of the corresponding three-dimensional model components.
[0083] The lightweight progress simulation method for large-scale, multi-scale, multi-entity projects includes, in step Sp5, the construction progress data including the measured start date, completed work volume, current construction status, and construction resource input. Based on the time deviation, work volume deviation, and critical path deviation, corresponding early warning data instructions are output.
[0084] When multiple entities collaboratively view the project progress, if local direct rendering is still used, loading and display pressure will easily occur as the model size increases. Therefore, this implementation deploys the rendering service on a GPU server cluster and transmits the results to the client browser as a video stream through cloud rendering. The computing scheduling microservice puts the progress simulation task into the task queue and distributes it to different computing nodes for execution according to the current load.
[0085] Therefore, rendering and computation run independently. When a node is busy, it only affects the tasks undertaken by that node and will not interrupt the entire simulation chain. To avoid the same simulation task being executed repeatedly on multiple computing nodes, each task in the task queue is bound to a unique task identifier and its scheduling cycle. The task is locked after it enters the execution state and released after it is completed.
[0086] If a node does not return a result within the specified time, the scheduling service marks the task as timed out and re-enqueues it, instead of directly reusing the intermediate state of the incomplete node. Before re-enqueueing, the scheduling service queries the result storage area based on the unique task identifier. If a complete result already exists, the task is marked as completed and is not recalculated. If a complete result is not formed, the corresponding temporary state is cleared and the task is re-executed to ensure that only one valid simulation result is used in the same scheduling cycle.
[0087] Furthermore, the interaction between the client and the rendering service runs in four stages: session establishment, bitstream stabilization, interaction update, and session release. In the session establishment stage, view permission confirmation and initial working condition binding are completed. In the bitstream stabilization stage, the current working condition video stream is continuously output. In the interaction update stage, only the control information required for viewpoint changes and layer switching is uploaded without returning the complete model. In the session release stage, the corresponding GPU rendering context is reclaimed.
[0088] If the client does not return a receipt confirmation within two consecutive sending cycles, the rendering service will maintain the previous stable bitrate and continue sending; if no confirmation is received after the time limit is exceeded, the high-precision level requests for the session will be frozen and only the basic bitstream will be retained. High-precision rendering will be gradually restored after confirmation is received to prevent the server from continuously outputting high-load images that cannot be consumed when there is network jitter.
[0089] When constructing the simulation model, the system establishes construction process models for TBM tunnels, shield tunnels, pipe jacking, buried pipes, and pumping stations, and associates each process with the state change rules of the corresponding components. As a result, when a certain process is scheduled to be executed, the system can update the state of the relevant components according to the process sequence, instead of just displaying or hiding the components as a whole.
[0090] If a certain type of work point does not appear in the current project segment, the corresponding process model remains inactive and does not participate in the scheduling of the current cycle. The state update granularity of each process model is consistent, and the completion ratio of the process at the end of the scheduling cycle is used as the basis for component state switching, rather than the video frames on the browser side as the calculation basis, so as to avoid logical conflicts when the rendering refresh frequency is inconsistent with the construction progress calculation frequency.
[0091] The completion rate of a process is limited to between 0% and 100%. When the reported value is below zero, it is treated as zero; when it is above 100%, it is treated as 100%. The status of a component is divided into three categories: not started, under construction, and completed. When the completion rate is zero, it is recorded as not started; when it is greater than zero and less than 100%, it is recorded as under construction; and when it reaches 100%, it is recorded as completed. If a preceding process is not completed, even if the subsequent process receives completion data, it will not drive the component status switch. Instead, the data will be retained until the first scheduling cycle after the preceding process is completed.
[0092] When the same component receives multiple reports from participating entities within the same scheduling cycle, the system first merges them according to the work point identifier, process identifier, and component identifier. Then, it uses the last valid report before the end of the scheduling cycle as the value of the status field and the cumulative amount of work completed within the cycle as the value of the quantity field.
[0093] If the status field does not correspond to the cumulative completed work volume, the completion ratio calculated from the cumulative completed work volume will be used to redetermine the component status, thereby avoiding conflicting markers of the same component being both completed and not started within a scheduling cycle; for processes with parallel branches that then merge in the process sequence, the system will only allow the merged process to enter the execution state when all parallel preceding branches have met the completion conditions.
[0094] If only a portion of the branches are completed, the merging process remains in a waiting state, and the data it receives is only stored in the temporary storage area without triggering a component state switch. For linear components constructed in a continuous advancement manner, the system also records the section where the construction front is located and requires that the direction of advancement of the front be consistent with the previous effective cycle within the same scheduling cycle. If the reported position shows a non-continuous change in the reverse spatial coordinates and there is no rework mark, the position record is only used for early warning and is not directly rewritten to the front position.
[0095] During the progress data access phase, the system receives the measured start date, completed work volume, current construction status, and construction resource input, and compares this data with the planned progress. The time deviation is determined by the difference in calendar days between the measured node date and the corresponding planned node date. A positive value indicates that the measured progress is lagging behind, and a negative value indicates that the measured progress is ahead of schedule.
[0096] The deviation of the quantity of work is determined by the ratio of the difference between the planned cumulative completed quantity and the actual cumulative completed quantity to the planned cumulative completed quantity. When the planned cumulative completed quantity is zero, this ratio is not calculated. Instead, the time deviation is judged based on whether the process started as planned. The deviation of the critical path is determined by the number of calendar days that the expected completion date of the critical path task exceeds the planned completion date.
[0097] The warning threshold should preferentially adopt the threshold approved by the project schedule management rules and written into the project baseline; if the project baseline is not configured separately, the time deviation is set to [value missing]. The proportion of delayed work volume is The critical path deviation is When satisfied , or If any of the conditions in the above conditions are met, output a level 1 warning;
[0098] When satisfied , or If any of the conditions in the above conditions are met, the warning level will be raised by one level; the warning will be lifted after the deviation recovers to within the corresponding trigger threshold and remains within two consecutive scheduling cycles, so as to avoid repeated triggering and lifting near the critical value; negative time deviations generated by early completion will not trigger delayed warnings, but will be retained in the deviation record for subsequent evaluation.
[0099] If a resource input data item is not reported, the most recent valid record will be used as the reference value, while retaining the missing data marker to avoid interruption of deviation judgment due to the absence of a single field. The reference value formed by the most recent valid record is only used to maintain the simulation of the current cycle and is not used as a direct basis for whether the resource input has met the plan. If three consecutive scheduling cycles are missing, a data quality prompt will be output. For the completed work volume, the system will convert it according to the measurement standard consistent with the plan before participating in the deviation judgment.
[0100] If the on-site report specifies the section length, number of excavation rings, or equipment shifts, the standard completion quantity for the corresponding process should first be converted based on the unit length engineering quantity, single ring standard engineering quantity, or single shift standard output recorded in the baseline of that process. Then, it should be written into the current scheduling cycle to avoid direct comparison between different work sites due to different units of measurement. The above conversion, status merging, and deviation judgment are all completed on the data snapshot frozen in the same scheduling cycle.
[0101] After a snapshot is taken and before the start of the next scheduling cycle, newly arriving data is only stored in the database without rewriting the status of components already submitted in this cycle, so as to ensure that a deviation analysis, a status update and a rendering output use the same input data; in order to ensure that the data reported by multiple construction entities can be compared on the same time axis, the system uses the scheduling service clock as a unified reference time.
[0102] After the data reported by each entity arrives, the receiving time is recorded first, and then the service time carried by it is compared with the unified reference time. If the offset is within the allowable range, it is assigned to the corresponding scheduling cycle according to its service time. If the offset exceeds the allowable range, it is assigned to the most recent cycle according to the receiving time and a clock offset mark is added.
[0103] When multiple consecutive cycles show offsets in the same direction, the system applies a fixed compensation amount to the reporting source, so that its subsequent data is prioritized for freezing according to the compensated service time. Once the offset direction is reversed or exceeds the applicable range of compensation, the original compensation is canceled and processing is restored according to the receiving time, thereby reducing the continuous impact of heterogeneous terminal clock drift on cycle division.
[0104] A lightweight progress simulation method for large-scale, multi-scale, multi-entity projects, step Sp6, closed-loop training includes: consistency verification between simulation results and construction progress data;
[0105] Construction dimension deviation matrix ,in The evaluation unit is the number of evaluation units contained in the construction progress data. Each evaluation unit corresponds to a work point, process, or section. The column dimensions strictly correspond to the dimensionless time deviation, spatial location deviation, and quantity deviation, respectively. Multi-dimensional accuracy evaluation is performed based on the deviation matrix. Parameter correction is based on deviation feedback. The parameter correction adopts an adaptive learning rate gradient descent optimization method.
[0106] During the long-term operation of the construction progress simulation, if the initial efficiency parameters and resource allocation parameters are used continuously, the model output will gradually deviate from the actual construction process. Therefore, this implementation method adds three processing steps after the simulation is run: verification, evaluation and parameter correction.
[0107] The system first verifies the consistency between the simulation results and the construction progress data. Then, it constructs a deviation matrix based on time deviation, spatial location deviation, and quantity deviation, and performs a multi-dimensional accuracy evaluation accordingly. The time deviation is the difference in calendar days between the simulation node completion date and the actual node completion date. The spatial location deviation is the distance along the path between the simulated construction front and the reported construction front within the same scheduling cycle. The quantity deviation is the difference between the cumulative completed quantity in the simulation and the cumulative completed quantity in the actual measurement. All three types of deviations retain positive and negative directions to distinguish between simulation ahead and simulation lag.
[0108] The deviation matrix is organized with evaluation units as rows and deviation dimensions as columns within the scheduling cycle. Each evaluation unit corresponds to one of the following: a work point, a process, or a section. The same object is not recorded repeatedly in the same matrix. If an evaluation unit can only obtain measured data for two or one dimension in the current cycle, then only the valid value is recorded at the corresponding dimension position of the unit, and the other positions are recorded as missing. Missing values are not directly written as zero.
[0109] For the spatial position deviation of linear work points, the distance along the path is calculated cumulatively according to the mileage direction of the centerline of the work point; if the location reported on site is represented by ring number, station number or section number, it is first converted to the same centerline mileage before participating in the deviation calculation, so as to ensure that the spatial position deviation adopts a unique length caliber within the same cycle.
[0110] To avoid evaluation distortion caused by the direct superposition of time deviation, spatial location deviation and quantity deviation due to different dimensions, the system performs dimensionless processing after constructing the deviation matrix; the time deviation is based on the planned duration of the corresponding process, and if the planned duration is less than 1 day or cannot be obtained, it is treated as 1 day.
[0111] Spatial position deviation is based on the length of the corresponding construction section. If the section length is less than 1m or cannot be obtained, it is treated as 1m. Quantity deviation is based on the planned quantity of the corresponding process. If the planned quantity of the process is less than the minimum unit of measurement or cannot be obtained, it is treated as the minimum unit of measurement. Each original deviation is divided by the corresponding benchmark to obtain the dimensionless deviation, thereby ensuring that the same dimension is used consistently across different work sites and avoiding the normalized benchmark being zero.
[0112] For missing dimensions, zero-value filling is not performed; instead, the evaluation results for the current period are calculated only on the existing dimensions, and the missing markers for that period are retained. For each evaluation unit, the dimensionless time deviation, spatial location deviation, and quantity deviation are first obtained. Then, all valid evaluation units within the current scheduling period are weighted and summarized according to their corresponding planned quantity proportions to form the period-level indicators for the current period. , and ;
[0113] For the current scheduling cycle, dimensionless data of each evaluation unit in the deviation matrix $E$ is extracted. A linear weighted sum is then performed using the proportion of the planned workload of each evaluation unit to the total planned workload of the cycle as the weight, and cycle-level indicators are calculated accordingly. , and The comprehensive evaluation objective can be expressed as:
[0114]
[0115] in, , and These represent the dimensionless periodic time deviation, spatial location deviation, and engineering quantity deviation, respectively. , and These are weighting coefficients, obtained through the analytic hierarchy process or based on pre-defined project characteristics; this is done when all dimensions of data are valid. , and Take values of 0.4, 0.3, and 0.3 respectively;
[0116] When there are missing dimensions, the weights of the missing dimensions are reset to zero and redistributed according to the original weights of the remaining valid dimensions, so that the sum of the valid weights is still 1. Thus, gradient descent corrects the parameters corresponding to the unified evaluation objective, rather than directly mixing and calculating the original deviations with different physical meanings and dimensions.
[0117] When the evaluation results show that the deviation persists, the system filters the efficiency parameters and resource allocation parameters that need to be corrected from the deviation feedback, and then uses the gradient descent method with adaptive learning rate to update the parameters; the efficiency parameters include the standard amount of work completed per unit time, and the resource allocation parameters include the number of personnel, the number of equipment, and the daily effective working hours;
[0118] When the deviation direction of the same dimension is consistent for three consecutive scheduling cycles and the overall evaluation target has not decreased, parameter correction is initiated. Parameter correction is performed item by item, rather than applying perturbations to all parameters to be corrected simultaneously in the same round. For each parameter to be corrected, the remaining parameters are kept unchanged, a perturbation is added to the current value, and the simulation of the same cycle is re-executed. The perturbation is 1% of the current parameter value.
[0119] When the current parameter value is zero, the disturbance amount is 1% of the minimum allowable value of the parameter; the minimum allowable value is the lower limit of the parameter preset in the project baseline or the default constant set by the system; the system approximates the local rate of change of the objective function with respect to the parameter by a single-parameter two-sided disturbance, and determines the direction of descent accordingly; if neither increasing nor decreasing the direction causes the comprehensive evaluation target to decrease, the parameter remains unchanged in this round.
[0120] This yields an approximate gradient of the coordinate direction for the current parameter set. After updating only one parameter at a time, the new value of that parameter is immediately written back before proceeding to the judgment of the next parameter to be corrected. This avoids misjudgment of direction due to mutual coupling when multiple parameters are updated simultaneously.
[0121] The initial learning rate is 0.05. If the comprehensive evaluation target decreases consecutively after two consecutive updates, the learning rate will be increased by 10%, but not exceeding 0.2. If the comprehensive evaluation target increases, the update will be rolled back and the learning rate will be halved, but not lower than 0.005. After each update, the change in the standard completion amount per unit time and the effective daily working time relative to the time before the update shall not exceed 10%. The number of personnel and equipment shall be adjusted in whole numbers and the single increase or decrease shall not exceed one.
[0122] The personnel and equipment numbers are first determined by continuous values, then rounded to the nearest integer, and the comprehensive evaluation target after rounding is used for verification. If the verification result is not better than before the update, the integer adjustment is cancelled. All parameters are simultaneously constrained by the corresponding minimum and maximum values in the project baseline. If they exceed the limit, the corresponding boundary value is used. When the decrease of the comprehensive evaluation target is less than 1% for three consecutive scheduling cycles, or when the preset upper limit of 20 updates is reached, the parameter correction for this round is stopped.
[0123] The updated parameters are written back to the construction spatiotemporal evolution rules as the calculation basis for the next round of simulation. If an evaluation cannot complete the comparison of all dimensions due to missing measured data, only the existing dimensions are calculated, and the records of the uncompared dimensions are retained to maintain the continuity of the parameter correction link. The substitute values formed by the records of the previous effective period are only used to maintain the continuity of the simulation and are not used as measured observation values for gradient direction judgment and parameter correction.
[0124] When there are fewer than two valid dimensions, only consistency checks and warnings are performed in this cycle, and parameter correction is not initiated. To avoid drastic parameter fluctuations caused by a single abnormal report, parameter updates are triggered when the deviation direction is consistent across multiple consecutive scheduling cycles. If an isolated outlier occurs in a certain cycle, the result of that cycle will only participate in the warning and will not directly trigger a significant parameter correction.
[0125] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.
Claims
1. A lightweight schedule simulation method for large-scale, multi-scale, multi-entity projects, characterized in that: Includes the following steps: Sp1: Collect multi-source heterogeneous 3D data including BIM model data, GIS terrain data and oblique photogrammetry data, unify coordinates and normalize formats to form a unified data base, and generate an oblique photogrammetry real-scene 3D model based on the oblique photogrammetry data. Sp2: Lightweight processing of the unified data base plate, registration and fusion of each three-dimensional data in the unified data base plate, and construction of an engineering three-dimensional digital base plate containing three-dimensional model components; SP3: A simulation engine built on a microservice architecture, incorporating microservices for model management, rendering, computation scheduling, and data interfaces. Sp4: Obtain the construction schedule, extract the three-dimensional model components from the three-dimensional digital base plate of the project, establish mapping relationships and construct a multi-scale simulation view, and combine the construction process model corresponding to the large-scale multi-scale multi-subject project to define the construction spatiotemporal evolution rules including efficiency and resource allocation parameters. Sp5: Acquire construction progress data, drive the dynamic update of the three-dimensional model components according to the construction spatiotemporal evolution rules, compare and analyze the deviation between the construction progress data and the construction progress plan, and output the three-dimensional dynamic simulation results. Sp6: Perform closed-loop training on the simulation model, verify the consistency between the three-dimensional dynamic simulation results and the construction progress data, and correct the efficiency and resource allocation parameters in the construction spatiotemporal evolution rules based on the deviation analysis results.
2. The lightweight schedule simulation method for large-scale, multi-scale, multi-entity projects according to claim 1, characterized in that: In step Sp1, the BIM model data includes BIM model files in open standard format and BIM model files in proprietary format, and extracts component geometric information, component attribute information and spatial relationship information, and establishes a BIM model database according to work points.
3. The lightweight schedule simulation method for large-scale, multi-scale, multi-entity projects according to claim 1, characterized in that: In step Sp1, the GIS terrain data includes DEM, DOM, and TIN data, and covers the surface area within the engineering route and its construction impact boundary.
4. The lightweight schedule simulation method for large-scale, multi-scale, multi-entity projects according to claim 1, characterized in that: In step Sp2, the lightweighting process includes geometric simplification, which employs a progressive mesh simplification algorithm based on edge folding and combines a quadratic error metric and attribute error to construct a folding cost function; for regular geometric components, parametric surface fitting is used to replace discrete triangular meshes.
5. The lightweight schedule simulation method for large-scale, multi-scale, multi-entity projects according to claim 1, characterized in that: In step Sp2, the lightweighting process includes LOD multi-level detail model generation, which includes generating multiple precision levels for each 3D model component in the engineering 3D digital base plate, and automatically switching between precision levels according to the viewpoint distance.
6. The lightweight schedule simulation method for large-scale, multi-scale, multi-entity projects according to claim 1, characterized in that: In step Sp2, the lightweight processing of the oblique photogrammetry real-world 3D model includes: tile-slicing the oblique photogrammetry real-world 3D model and extracting local textures and progressively loading them, and combining the texture atlas technology to perform atlas merging of the local textures after tile-slicing.
7. The lightweight schedule simulation method for large-scale, multi-scale, multi-entity projects according to claim 1, characterized in that: In step Sp3, the rendering microservice is deployed on a GPU server cluster, and the rendering results are transmitted to the client browser as a video stream through server-side remote distributed rendering. The computation scheduling microservice allocates progress simulation computation tasks based on a task queue architecture.
8. The lightweight schedule simulation method for large-scale, multi-scale, multi-entity projects according to claim 1, characterized in that: The multi-scale simulation view establishes construction process models for full-section hard rock tunnel boring machine tunnels, shield tunnels, pipe jacking, buried pipes, and pumping stations, and associates each process with the state change rules of the corresponding three-dimensional model components.
9. The lightweight schedule simulation method for large-scale, multi-scale, multi-entity projects according to claim 1, characterized in that: In step Sp5, the construction progress data includes the actual start date, completed work volume, current construction status, and construction resource input. Corresponding early warning prompt data instructions are generated based on time deviation, work volume deviation, and critical path deviation.
10. The lightweight schedule simulation method for large-scale, multi-scale, multi-entity projects according to claim 1, characterized in that: In step Sp6, the closed-loop training includes: consistency verification between the simulation results and the construction progress data; Construction dimension deviation matrix ,in The number of evaluation units included in the construction progress data, where each evaluation unit corresponds to a work point, process, or section, and the column dimensions strictly correspond to the dimensionless time deviation, spatial location deviation, and quantity deviation, respectively, and a multi-dimensional accuracy evaluation is performed based on the deviation matrix; and parameter correction based on deviation feedback, wherein the parameter correction adopts an adaptive learning rate gradient descent optimization method.
