Intelligent engineering whole-cycle digital management system

Abstract

The application discloses an intelligent engineering full-cycle digital management system, and relates to the technical field of digital management, comprising: a real scene generation module, which is used for generating a three-dimensional real scene model according to collected engineering image data, and performing design specification identification in combination with a preset engineering design behavior specification constraint; a construction evolution module, which is used for executing construction evolution simulation, and performing construction compliance review based on a preset construction specification constraint; and a decision optimization module, which is used for constructing a knowledge graph based on a pre-acquired engineering knowledge base, performing engineering decision correlation analysis, and developing decision collaborative optimization. The application constructs a driving management system covering the whole life cycle of design, construction and operation and maintenance, realizes engineering full-process digital management and control, breaks through the fragmentation and experience dependence problems of traditional engineering management, forms a closed loop of data collection, intelligent analysis, knowledge sharing and dynamic optimization, and promotes the transformation of engineering management to full-factor digitalization and full-process intelligentization.

Description

Technical Field

[0001] This invention relates to the field of digital management technology, and more specifically, to a digital management system for the entire lifecycle of intelligent engineering projects. Background Technology

[0002] Intelligent engineering is a product of the deep integration of next-generation information technology and traditional engineering construction. It aims to achieve perception, analysis, decision-making, and optimization throughout the entire lifecycle of a project through advanced technologies such as big data, BIM (Building Information Modeling), and computer vision. Its core objective is to shift from the traditional model relying on human experience, static planning, and post-hoc correction to an intelligent paradigm driven by data, providing real-time feedback, proactive control, and adaptive optimization. However, to truly realize intelligent engineering, the key lies in building an integrated management platform that spans the entire lifecycle of design, construction, and operation and maintenance. Currently, many digital tools are still limited to single stages or isolated functions, leading to information fragmentation, delayed decision-making, and slow risk response. Therefore, the construction of a full-lifecycle digital management system for intelligent engineering is crucial.

[0003] However, existing intelligent engineering full-cycle digital management systems generally suffer from shortcomings such as weak knowledge reuse capabilities, rigid decision-making mechanisms, and insufficient depth of intelligent analysis. Engineering experience is stored in an unstructured form, making it difficult to securely share and transfer knowledge across projects; schedules and resource plans lack the ability to adaptively adjust to market fluctuations and unexpected situations; and system functions are mostly limited to visualization, lacking in-depth modeling of the physical evolution of construction processes and proactive risk warnings. These problems mean that existing systems remain limited to digital recording, failing to achieve true intelligent driving and struggling to break through the experience-dependent and fragmented predicament of traditional engineering management.

[0004] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0005] In response to the problems in related technologies, this invention proposes an intelligent engineering full-cycle digital management system to overcome the aforementioned technical problems existing in the existing related technologies.

[0006] Therefore, the specific technical solution adopted by the present invention is as follows: This invention provides a digital management system for the entire lifecycle of intelligent engineering projects, comprising: The reality generation module is used to generate a 3D reality model based on the collected engineering image data, and to identify the design specifications of the 3D reality model in combination with the preset engineering design behavior specifications to obtain the design specification identification result. The construction evolution module is used to perform construction evolution simulation on the 3D reality model, and conduct construction compliance review based on the construction evolution simulation results and preset construction specification constraints to obtain the construction compliance review results. The decision optimization module is used to construct a knowledge graph based on a pre-acquired engineering knowledge base, perform engineering decision correlation analysis on the design specification identification results and construction compliance review results based on the knowledge graph, and carry out collaborative decision optimization based on the results of the engineering decision correlation analysis.

[0007] Furthermore, the reality generation module includes: The dense point cloud generation module is used to input the collected engineering image data into the motion reconstruction structure algorithm and generate a dense point cloud, and then construct a 3D reality model based on the dense point cloud. The difference region annotation module is used to align the 3D reality model with the building information model based on the improved iterative nearest point algorithm to obtain the aligned 3D reality model, and to annotate the difference regions of the aligned 3D reality model to obtain the 3D reality difference annotation results. The design specification recognition module is used to perform quantitative analysis of design indicators based on the 3D real-world difference annotation results using a visual model, and then combine the quantitative analysis results of design indicators with the preset engineering design behavior specification constraints to perform design specification recognition and obtain the design specification recognition result. The real-scene generation module, construction evolution module, and decision optimization module are connected sequentially; the dense point cloud generation module, difference area annotation module, and design specification identification module are connected sequentially.

[0008] Furthermore, the design specification identification module includes: The semantic boundary completion module is used to construct a causal graph model based on the acquired engineering design data, and to perform causal inference and semantic boundary completion on the 3D real scene difference annotation results using the causal graph model to obtain the completed difference annotation results. The self-attention mechanism capture module is used to input the completed difference annotation results into the Transformer's visual model and use the self-attention mechanism to capture the relationship between geometric structure and global context, so as to obtain a high-dimensional feature embedding representation of each difference region. The design index regression prediction module is used to decode the high-dimensional feature embedding representation using a geometric perception decoding network, and to perform end-to-end design index regression prediction based on the decoding results, so as to obtain the design index quantitative analysis results. The logical matching module is used to logically match the quantitative analysis results of design indicators with the preset engineering design behavior specifications constraints, and generate design specification identification results based on the logical matching results; The semantic boundary completion module, the self-attention mechanism capture module, the design index regression prediction module, and the logical matching module are connected sequentially.

[0009] Furthermore, the semantic boundary completion module includes: The engineering logic construction module is used to extract design goals from the acquired engineering design data, and to implement engineering logic deduction and construction based on the design goals using rule reasoning to obtain a hierarchical design logic chain. The causal graph model building module is used to take the design goal as the top-level node and associate it with the hierarchical design logic chain. It performs geometric and spatial representation mapping and graph structure form expression on the association results to obtain a structured causal graph model. The causal inference module is used to input the 3D real scene difference annotation results into the causal graph model, and use the counterfactual reasoning mechanism to perform context matching causal path query on the missing regions in the 3D real scene difference annotation results. Based on the query results, the missing regions are filled in with geometric boundaries and semantic attributes to obtain the completed difference annotation results. The engineering logic construction module, the cause-effect graph model construction module, and the cause-effect inference module are connected sequentially.

[0010] Furthermore, the construction evolution module includes: The construction specification embedding module is used to embed construction specification clauses into the building information model based on a pre-established industrial basic standard extension framework, thereby obtaining a standardized building information model. The engineering twin construction module is used to integrate standardized building information models, 3D reality models, and pre-acquired engineering environment data to construct a dynamically updated engineering twin. The dynamic simulation module is used to dynamically simulate the construction status at each stage based on the dynamically updated engineering twin and the preset construction control element set, and obtain the construction evolution simulation results. The construction compliance review module is used to extract quantitative features of construction status from the construction evolution simulation results, compare and analyze the quantitative features of construction status with the preset construction specification constraints, judge the construction compliance based on the comparison and analysis results, and obtain the construction compliance review results. The construction specification embedding module, the engineering twin construction module, the dynamic simulation module, and the construction compliance review module are connected sequentially.

[0011] Furthermore, the dynamic simulation module includes: The structural response evolution module is used to drive a multiphysics simulation engine based on dynamically updated engineering twins and preset construction control element sets to perform state simulation of the entire construction process at time steps, and output the structural response evolution sequence of each stage. The response field dynamic evolution module is used to extract the displacement field and strain field time series of key structural regions from the structural response evolution sequence, and to construct a dynamic evolution dataset of continuous response field using spatial grid cells as the basic analysis unit. The information entropy flow analysis module is used to perform structural information entropy flow analysis on the dynamic evolution dataset, and to perform dynamic tracking and abnormal pattern recognition based on the structural information entropy flow analysis results to obtain construction evolution simulation results; The structural response evolution module, the response field dynamic evolution module, and the information entropy flow analysis module are connected sequentially.

[0012] Furthermore, the information entropy flow analysis module includes: The spatial difference analysis module is used to construct the time series of physical quantities for each spatial discrete unit of the dynamic evolution dataset, and calculate the spatial difference of the strain gradient field between adjacent units based on the time series, so as to obtain the strain gradient non-uniformity index of each unit at different times. The entropy yield calculation module is used to treat the response fluctuation of each unit as a local information evolution process. It uses the kernel density estimation method to perform probability density modeling on the time window sequence of the non-uniform strain gradient index, and combines the differential entropy formula to calculate the local entropy yield. The synchronization mapping module is used to synchronize the local entropy yield of all units in the spatial and temporal dimensions, and construct a spatiotemporal entropy yield distribution map. The dynamic tracking module is used to perform trend analysis on the entropy value sequence of each spatial location in the spatiotemporal entropy yield distribution map using a sliding time window mechanism, and to perform anomaly pattern recognition based on the trend analysis results. The anomaly pattern recognition results are then correlated with the pre-acquired construction load data to obtain the construction evolution simulation results.

[0013] Furthermore, the expression for calculating the local entropy productivity is as follows: ; In the formula, h i ( t n ) represents a unit i exist t n The estimated value of the differential entropy at time t; N n Indicates time step n The total number of samples; j Indicates the index of the currently evaluated sample; k Indicates the reference sample index used for kernel density estimation; σ n This represents the Gaussian kernel bandwidth at the corresponding time step; Indicates time step t n Next j One observation sample; Indicates time step t n Next kOne observation sample; h i ( t n-1 ) represents a unit i exist t n-1 The estimated value of the differential entropy at time Δ t Indicates the time step; Representation unit i exist t n The local entropy production rate at time t.

[0014] Furthermore, the decision optimization module includes: The common feature extraction module is used to extract common features of engineering schemes from a pre-acquired engineering knowledge base using contrastive learning, and to construct a knowledge graph for engineering decision-making based on the extracted common features. The association matching module is used to extract engineering anomaly features based on the design specification identification results and construction compliance review results, and to perform engineering decision association matching on the engineering anomaly features based on the knowledge graph to obtain engineering decision association analysis results; The collaborative decision optimization module is used to simulate the construction path based on the results of engineering decision correlation analysis, obtain an optimized engineering schedule plan, and carry out multi-objective collaborative decision optimization in combination with pre-acquired market prices to obtain an optimized engineering decision. The common feature extraction module, the association matching module, and the collaborative decision optimization module are connected sequentially.

[0015] Furthermore, the collaborative decision-making optimization module includes: The project schedule determination module is used to perform construction path simulation based on the results of the correlation analysis of project decisions using generative adversarial networks, and dynamically optimize the resource allocation and timing of adversarial training based on the results of the construction path simulation to obtain the optimized project schedule. The procurement plan determination module is used to input the optimized project schedule plan into the deep reinforcement learning model, and combine it with the pre-acquired market prices to construct a mapping relationship between material demand and procurement time window. Based on the mapping relationship, the module makes short-term predictions on building material price trends, determines the material procurement strategy based on the short-term prediction results, and obtains the optimized procurement plan. The engineering decision module is used to integrate the optimized schedule plan and the optimized procurement plan into a multi-objective process, and to perform collaborative optimization of engineering decisions based on the results of the multi-objective integration using Pareto front analysis, so as to obtain the optimized engineering decision. The project schedule determination module, the procurement plan determination module, and the project decision-making module are connected sequentially.

[0016] The beneficial effects of this invention are as follows: 1. This invention constructs a driving management system covering the entire lifecycle of design, construction, and operation and maintenance through the coordinated setup of a real-scene generation module, a construction evolution module, and a decision optimization module. It achieves full-process digital control of engineering projects through 3D modeling using UAV imagery, deep BIM integration, and intelligent algorithms. High-precision 3D real-scene models are generated based on UAV oblique photography and AI image processing, and automatic compliance reviews are conducted in conjunction with BIM design data. Intelligent recognition algorithms are developed to detect construction violations, and cross-project experience is integrated through a shared knowledge base to optimize decision-making. This allows the system to overcome the fragmentation and experience-dependent problems of traditional engineering management, forming a closed loop of data collection, intelligent analysis, knowledge sharing, and dynamic optimization, thus promoting the transformation of engineering management towards full-element digitalization and full-process intelligentization.

[0017] 2. This invention effectively solves the problems of low efficiency, strong subjectivity, and difficulty in quantification of traditional manual inspections through a semantic boundary completion module. In complex irregular structures or dense electromechanical pipeline areas, this module can accurately mark the difference areas and complete the semantic boundaries, making the hidden works visible and the design intent calculable, providing a high-fidelity digital foundation for subsequent construction and significantly improving the ability to control the quality of the project in advance.

[0018] 3. This invention, through its information entropy flow analysis module, overcomes the limitations of traditional construction simulation relying solely on static BIM models. It achieves high-fidelity simulation of the physical state evolution during construction and intelligent early warning of damage precursors. By extracting the spatiotemporal distribution of local entropy productivity from massive response data, it can identify abnormal energy dissipation zones without relying on material constitutive models. This anomaly pattern recognition mechanism based on information entropy can provide early warnings of potential structural instability, support failure, or overload risks, far earlier than traditional displacement or stress threshold alarms. Simultaneously, combined with automatic compliance review based on construction specifications, it can assess in real-time whether the execution of procedures meets safety and technological requirements, realizing a shift from post-event correction to in-process intervention and even pre-event control, greatly enhancing the safety resilience and process controllability of the construction process. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the 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.

[0020] Figure 1 This is a schematic diagram of the intelligent engineering full-cycle digital management system according to an embodiment of the present invention; Figure 2 This is an application flowchart of the intelligent engineering full-cycle digital management system according to an embodiment of the present invention.

[0021] In the picture: 1. Real-world scene generation module; 2. Construction evolution module; 3. Decision optimization module. Detailed Implementation

[0022] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention.

[0023] According to an embodiment of the present invention, a digital management system for the entire lifecycle of intelligent engineering is provided.

[0024] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figures 1-2 As shown, the intelligent engineering full-cycle digital management system according to an embodiment of the present invention includes: The real-scene generation module 1 is used to generate a three-dimensional real-scene model based on the collected engineering image data, and to perform design specification recognition on the three-dimensional real-scene model in combination with the preset engineering design behavior specification constraints to obtain the design specification recognition result.

[0025] Specifically, the reality generation module 1 includes: The dense point cloud generation module is used to input the collected engineering image data into the motion reconstruction structure algorithm and generate a dense point cloud, and then construct a 3D reality model based on the dense point cloud. The difference region annotation module is used to align the 3D reality model with the building information model based on the improved iterative nearest point algorithm, obtain the aligned 3D reality model, and annotate the difference regions of the aligned 3D reality model to obtain the 3D reality difference annotation results.

[0026] The design specification recognition module is used to perform quantitative analysis of design indicators based on the 3D reality difference annotation results using a visual model, and then combine the quantitative analysis results of design indicators with the preset engineering design behavior specification constraints to perform design specification recognition and obtain the design specification recognition result.

[0027] The real-scene generation module 1, construction evolution module 2, and decision optimization module 3 are connected sequentially; the dense point cloud generation module, difference area annotation module, and design specification identification module are connected sequentially.

[0028] Specifically, the design specification identification module includes: The semantic boundary completion module is used to construct a causal graph model based on the acquired engineering design data, and to perform causal inference and semantic boundary completion on the 3D real scene difference annotation results using the causal graph model to obtain the completed difference annotation results.

[0029] Specifically, the semantic boundary completion module includes: The engineering logic construction module is used to extract design goals from the acquired engineering design data, and to implement engineering logic deduction and construction based on the design goals using rule reasoning to obtain a hierarchical design logic chain. The causal graph model building module is used to take the design goal as the top-level node and associate it with the hierarchical design logic chain. It performs geometric and spatial representation mapping and graph structure form expression on the association results to obtain a structured causal graph model. The causal inference module is used to input the 3D real scene difference annotation results into the causal graph model, and use the counterfactual reasoning mechanism to perform context matching causal path query on the missing regions in the 3D real scene difference annotation results. Based on the query results, the missing regions are filled in with geometric boundaries and semantic attributes to obtain the completed difference annotation results. The engineering logic construction module, the cause-effect graph model construction module, and the cause-effect inference module are connected sequentially.

[0030] The self-attention mechanism capture module is used to input the completed difference annotation results into the Transformer's visual model and use the self-attention mechanism to capture the relationship between geometric structure and global context, so as to obtain a high-dimensional feature embedding representation of each difference region.

[0031] The design index regression prediction module is used to decode the high-dimensional feature embedding representation using a geometric perception decoding network, and to perform end-to-end design index regression prediction based on the decoding results, so as to obtain the design index quantitative analysis results.

[0032] The logical matching module is used to logically match the quantitative analysis results of design indicators with the preset engineering design behavior norms and generate design norm identification results based on the logical matching results.

[0033] The semantic boundary completion module, the self-attention mechanism capture module, the design index regression prediction module, and the logical matching module are connected sequentially.

[0034] Specifically, the real-scene generation module 1 includes intelligent acquisition and modeling of 3D engineering image data. The engineering image data is acquired through drones, specifically multi-rotor drones executing a gridded flight path at an altitude of 150 meters and an overlap rate of 80%, acquiring 0.02m resolution oblique photogrammetric images. Dense point clouds with a density >1000 points / m are generated using the Structure of Motion (SfM) algorithm. 2Simultaneously, image processing is performed using a point cloud-BIM automatic registration algorithm. This involves employing an improved ICP (Iterative Closest Point) algorithm to align the real-world point cloud with the BIM (Building Information Model), achieving a registration error of <3mm and automatically marking discrepancies, such as elevation deviations >10cm. The real-world generation module 1 enables intelligent recognition, analyzing aerial images based on a Transformer visual model to detect three typical violations: first, construction boundary violations (comparing the real-world view with the BIM geofence to identify illegal material stacking within the river's blue line, with a detection rate of 98%); second, safety violations (detecting behaviors such as not wearing safety helmets and lack of edge protection, with an accuracy rate of 96%); and third, process violations (analyzing issues such as excessive concrete pouring intervals and insufficient curing through time-series image analysis).

[0035] Specifically, the real-scene generation module 1, through deep integration of computer vision, causal reasoning, Transformer representation learning, and normative knowledge engineering, achieves an end-to-end automated process from raw engineering images to automatic identification and risk warning of design compliance. In a super high-rise commercial complex project, a drone equipped with a camera was deployed to collect a total of 5327 RGB images, with a single image resolution of 4000×3000 pixels and a ground sampling distance of GSD=6.2mm, covering the core tube, standard floors, and podium area, obtaining high-precision multi-source image data of the engineering site, i.e., engineering image data. These images are first fed into the dense point cloud generation module, which uses the COLMAP open-source framework to perform structure of motion reconstruction (SfM) to obtain the initial pose containing sparse feature points; then, OpenMVS is called to perform multi-view stereo matching (MVS) to generate a dense point cloud with a high density, with the average point spacing controlled within 8mm. After statistical outlier removal (SOR) filtering, 98.7% of the valid points are retained in the point cloud, and a textured triangular mesh model is generated through Poisson surface reconstruction to form a high-fidelity 3D real-scene model.

[0036] The real-world model then enters the difference area annotation module for spatial alignment with the project's BIM model. Traditional ICP algorithms are susceptible to local geometric noise, so an improved strategy is introduced. First, a kernel-point convolutional network is used to perform semantic segmentation on the point cloud, identifying 12 types of components, including walls, beams, columns, and floor slabs. Then, during the ICP iteration process, differentiated weights are assigned to different component categories; for example, load-bearing walls are weighted at 1.5, and non-load-bearing partitions at 0.8. A flatness constraint term is also added, requiring the fitted plane residual to be <5mm. After 12 iterations, the overall registration error RMS is reduced to 11.3mm, meeting the unified standard requirements for construction quality acceptance in building engineering. Based on this, the Hausdorff distance field between the real-world model and the BIM surface is calculated, with a threshold of ±20mm, automatically annotating 327 difference areas, including 112 missing components (e.g., uncast floor slabs); 143 positional offsets (e.g., column center offset >15mm); and 72 geometric deformations (e.g., uneven beam bottoms).

[0037] Due to construction obstructions or equipment blind spots, the initial annotations contained 98 semantically incomplete areas, such as stair platforms that were only partially visible. To address this, the semantic boundary completion module initiated a causal reasoning mechanism. The engineering logic construction module extracted 42 high-rise design objectives from BIM attribute parameters, construction drawings, and engineering specifications, such as ensuring the clear width of evacuation staircases is ≥1.1m and that fire compartment boundaries require solid walls with a fire resistance rating ≥2.0h. Based on these objectives, a hierarchical logic chain of "functional requirements, component layout rules, and geometric parameter constraints" was derived layer by layer using a rule reasoning engine based on Drools 8.0, generating a total of 215 executable rules. The causal graph model construction module used the design objectives as top-level nodes, constructing a causal graph in the Neo4j graph database containing 1248 nodes, covering four categories: objectives, functions, components, and geometric representations, and 2876 directed edges. Each edge was accompanied by a causal strength of zero to one and a specification reference ID. The causal inference module inputs the initial difference annotations into the graph model and initiates counterfactual inference for the missing areas. For example, when a 3.2m long wall is missing at the boundary of a fire compartment, the system queries the causal path "fire compartment integrity, continuous enclosure structure required, wall should exist and thickness ≥200mm", and combines it with the orientation of adjacent existing walls to complete its geometric boundary, extending to the shear walls on both sides and semantic attributes, namely, the material is aerated concrete block; the fire resistance limit is 2.5h, and outputs the completed difference annotation results, improving the integrity to 99.1%.

[0038] The completed result is presented in voxelized form with a resolution of 2 cm. The input is captured by a self-attention mechanism module, employing a Point Transformer architecture with 12 encoder layers, 512 embedding dimensions, and 8 heads. Each difference region is represented as a subset of 1024 points. The model dynamically calculates the geometric and semantic association weights between any two points through self-attention. For example, for the stair tread region, the model automatically enhances the feature interactions with adjacent platforms and railings. For instance, during training, the model is pre-trained on ScanNet v2 and a self-built construction site dataset. During fine-tuning, only the last two layers are updated, achieving an average accuracy of 89.4% on the validation set after convergence. The output is a 512-dimensional high-dimensional feature embedding for each difference region, encoding local shape details and global design context.

[0039] The design metric regression prediction module utilizes a lightweight geometry-aware MLP decoder to perform end-to-end regression on the embeddings, outputting 18 categories of key design metrics. For example, for evacuation corridor areas, the system quantifies the measured net width as 1.08m, while the specification requires ≥1.10m, resulting in a deviation. For core cylinder columns, the verticality deviation is 1 / 480, while the standard limit is 1 / 500. For floor slab flatness, the maximum gap of a 2m straightedge is 4.2mm, while the standard limit is ≤4mm. All indicators are accompanied by confidence scores, with an average of 0.93. The quantification error was verified by laser scanning, with a mean absolute error (MAE) of 2.7mm. The design specification identification module matches the quantification results with a structured specification library containing 867 machine-readable rules using OWL ontology modeling. The rule engine performs Boolean judgments and automatically generates a compliance report, identifying 23 non-compliant items, including 3 high-risk items (e.g., a measured fire separation distance of 3.8m < the standard 4.0m, triggering a red alert), 12 medium-risk items (e.g., a stair tread height difference >10mm), and 8 low-risk items. These are then directly pushed to the project management platform.

[0040] Construction Evolution Module 2 is used to perform construction evolution simulation on the 3D real-scene model, and conduct construction compliance review based on the construction evolution simulation results and preset construction specification constraints to obtain the construction compliance review results.

[0041] Specifically, construction evolution module 2 includes: The construction specification embedding module is used to embed construction specification clauses into the building information model based on a pre-established industrial basic standard extension framework, thereby obtaining a standardized building information model. The engineering twin construction module is used to integrate standardized building information models, 3D reality models, and pre-acquired engineering environment data to construct a dynamically updated engineering twin. The dynamic simulation module is used to dynamically simulate the construction status at each stage based on the dynamically updated engineering twin and the preset construction control element set, and obtain the construction evolution simulation results.

[0042] Specifically, the dynamic simulation module includes: The structural response evolution module is used to drive a multiphysics simulation engine based on dynamically updated engineering twins and preset construction control element sets to perform state simulation of the entire construction process at time steps, and output the structural response evolution sequence of each stage. The response field dynamic evolution module is used to extract the displacement field and strain field time series of key structural regions from the structural response evolution sequence, and to construct a dynamic evolution dataset of continuous response field using spatial grid cells as the basic analysis unit. The information entropy flow analysis module is used to perform structural information entropy flow analysis on the dynamic evolution dataset, and to perform dynamic tracking and abnormal pattern recognition based on the results of the structural information entropy flow analysis to obtain the construction evolution simulation results.

[0043] Specifically, the information entropy flow analysis module includes: The spatial difference analysis module is used to construct the time series of physical quantities for each spatial discrete unit of the dynamic evolution dataset, and calculate the spatial difference of the strain gradient field between adjacent units based on the time series, so as to obtain the strain gradient non-uniformity index of each unit at different times. The entropy yield calculation module treats the response fluctuation of each unit as a local information evolution process, uses the kernel density estimation method to model the probability density of the time window sequence of the non-uniform strain gradient index, and combines the differential entropy formula to calculate the local entropy yield.

[0044] Specifically, the expression for calculating the local entropy productivity is as follows: ; In the formula, h i ( t n ) represents a unit i exist t n The estimated value of the differential entropy at time t; N n Indicates time step n The total number of samples; j Indicates the index of the currently evaluated sample; k Indicates the reference sample index used for kernel density estimation; σ n This represents the Gaussian kernel bandwidth at the corresponding time step; Indicates time step t n Next j One observation sample; Indicates time step t n Next k One observation sample; h i ( t n-1 ) represents a unit i exist t n-1 The estimated value of the differential entropy at time Δ t Indicates the time step; Representation unit i exist t n The local entropy production rate at time t.

[0045] The synchronization mapping module is used to synchronize the local entropy yield of all units in the spatial and temporal dimensions, and construct a spatiotemporal entropy yield distribution map.

[0046] The dynamic tracking module is used to perform trend analysis on the entropy value sequence of each spatial location in the spatiotemporal entropy yield distribution map using a sliding time window mechanism, and to perform anomaly pattern recognition based on the trend analysis results. The anomaly pattern recognition results are then correlated with the pre-acquired construction load data to obtain the construction evolution simulation results.

[0047] The spatial difference analysis module, entropy yield calculation module, synchronization mapping module, and dynamic tracking module are connected sequentially.

[0048] The structural response evolution module, the response field dynamic evolution module, and the information entropy flow analysis module are connected sequentially.

[0049] The construction compliance review module is used to extract quantitative features of construction status from the construction evolution simulation results, compare and analyze the quantitative features of construction status with the preset construction specification constraints, judge the construction compliance based on the comparison and analysis results, and obtain the construction compliance review results.

[0050] The construction specification embedding module, the engineering twin construction module, the dynamic simulation module, and the construction compliance review module are connected sequentially.

[0051] Specifically, Construction Evolution Module 2 includes deep BIM integration and intelligent review. Dynamic BIM model construction includes: establishing an extended framework of the IFC (Industrial Foundation Class) standard, embedding construction specification clauses (such as the acceptance specification for water conservancy and hydropower projects) into the BIM model, and supporting computable attributes; developing a lightweight BIM engine to support second-level loading and interaction of component models, such as WebGL optimization. Simultaneously, automatic compliance review includes: geometric compliance (real-time comparison of the real-world site cloud and the BIM model to detect structural dimensional deviations, with a threshold of ±5cm); process compliance (linking construction logs and BIM progress plans to provide early warnings of delayed processes, such as formwork removal earlier than the specified deadline); and safety compliance (integrating a safety specification knowledge graph to automatically check the completeness of life-saving equipment configuration for water-related operations). Furthermore, it implements engineering construction evolution through a digital twin sand table, integrating the BIM model, real-world site cloud, and sensor data (such as concrete temperature and humidity) to construct a dynamically updated engineering twin; and supports rainstorm condition simulation (coupling a CFD engine to calculate flow diversion capabilities and provide early warnings of cofferdam overflow risks). Construction Evolution Module 2 constructs a dynamic deduction system that integrates standard knowledge, multi-source perception, and physical simulation, realizing a closed loop from static model to simulation of the entire construction process's state evolution and intelligent compliance review. Its specific implementation begins with the construction standard embedding module. Based on the Industry Foundation Classes (IFC) extended framework, the system converts 867 quantifiable clauses from 12 core construction standards, such as the Building Construction Safety Inspection Standard and the Concrete Structure Engineering Construction Standard, into machine-readable semantic rules. For example, rules like "spacing of formwork support uprights ≤ 1.2m" and "layered support required for foundation pit excavation depth > 5m" are transformed into machine-readable semantic rules. These rules are then embedded into the project BIM model through the IFC Property Set, forming a standardized Building Information Model (Normative BIM). Each component carries a compliance constraint label, such as the maximum deflection limit of IfcBeam = L / 250. The engineering twin construction module integrates the standard BIM with the high-precision 3D reality model and engineering environment data output by the reality generation module 1, including real-time wind speed or temperature from weather stations, foundation bearing capacity distribution detected by ground-penetrating radar, and load time series monitored by tower cranes. It performs multimodal fusion, uses a spatiotemporal alignment algorithm to unify the coordinate system and clock reference, uses graph neural networks (GNN) to establish the correspondence between BIM components and real-world cloud clusters, and introduces uncertainty quantification mechanisms, such as Monte Carlo Dropout to assess the confidence of data fusion. This constructs a dynamically updated engineering twin with four-dimensional attributes: geometric, semantic, physical, and environmental, and supports automatic status updates according to the construction progress.

[0052] Based on this, the dynamic simulation module drives the state evolution simulation of the entire construction process. Its core sub-module, the structural response evolution module, uses the engineering twin as the initial field and combines it with a preset set of construction control elements, including 214 parameters such as process logic, material strength development curve, temporary support removal sequence, and construction load application path. It calls a multiphysics simulation engine, such as one based on the OpenSees and CFD coupled framework, to perform forward simulation with a time step of 6 hours. For example, during the simulation of the 18th floor slab pouring stage, the engine sequentially loads the steel reinforcement self-weight of 25 kN / m³, wet concrete load of 24 kN / m³, and construction live load of 3 kN / m², considering the time-varying effect of the early-age elastic modulus of concrete. It outputs a structural response evolution sequence containing displacement, strain, and stress, covering all 32 key construction stages from foundation excavation to main structure completion. The response field dynamic evolution module extracts the core stress areas from this sequence, such as the displacement and principal strain fields of transfer beams, cantilever slabs, and deep foundation pit support piles. Using finite element mesh elements as the basic analysis units (0.5m × 0.5m × 0.3m), it constructs a continuous spatiotemporal response field dataset. Each unit stores its physical quantity time series under 32 stages × 4 samplings per stage, forming a tensor with dimensions [480000 × 128]. This dataset is input into the information entropy flow analysis module to perform constitutive-independent damage precursor identification. That is, the spatial difference analysis module analyzes each unit... i A strain gradient nonuniformity index is constructed by calculating the L2 norm difference of the principal strain gradient between adjacent units. A sliding window with a window length of N=10 is used to obtain the strain gradient nonuniformity index of each unit at different times. The entropy yield calculation module treats the strain gradient nonuniformity index sequence as a local information evolution process and uses Gaussian kernel density estimation (KDE) to model its probability distribution. The bandwidth is adaptively set using the Silverman rule, and the local entropy yield is calculated by substituting it into the differential entropy formula. After calculating the entropy yield of each unit at each stage, the synchronous mapping module synchronously maps it in the spatial and temporal dimensions to generate a spatiotemporal entropy yield distribution map with a resolution of 0.5m×0.5m×0.3m×6h. The dynamic tracking module uses a sliding time window with a window length of 5 stages to perform trend analysis on the entropy value sequence at each spatial location. That is, if an entropy yield > 0.15 bit / (m³·h) appears in a certain region for 3 consecutive stages, it is defined as an entropy burst region, reflecting the concentration of plastic dissipation, or the entropy yield < A negative entropy island, which indicates brittle fracture, is marked as an anomaly. The spatiotemporal coordinates of the anomaly are then correlated with construction load logs, such as pump truck operation periods and load locations, to generate construction evolution simulation results that include risk levels, causal inferences, and suggested measures. For example, when simulating the construction of the 22nd floor, the system identified a negative entropy island at the root of the transfer beam with an entropy production rate of -0.12. Correlation revealed that the steel reinforcement was overloaded during this period, with the measured load reaching 8.2 kN / m², which is greater than the design value of 5 kN / m², thus triggering a high-risk warning.

[0053] The construction compliance review module extracts 12 quantitative features of construction status from the construction evolution simulation results, such as maximum displacement rate, the proportion of entropy surge area, and the duration of negative entropy islands. These features are then compared in real-time with a pre-set construction specification constraint library containing 214 dynamic threshold rules, such as requiring work stoppage for displacement rates > 2 mm / d and considering damage for entropy surge area > 10% of component cross-section. The system uses fuzzy logic reasoning to handle uncertainty and outputs three types of review conclusions: green (compliant), yellow (observation), and red (violation). In a real-world project, this module successfully identified three violations, including the aforementioned overload event and seven observation items.

[0054] The decision optimization module 3 is used to construct a knowledge graph based on a pre-acquired engineering knowledge base, perform engineering decision correlation analysis on the design specification identification results and construction compliance review results based on the knowledge graph, and carry out collaborative decision optimization based on the engineering decision correlation analysis results.

[0055] Specifically, the decision optimization module 3 includes: The common feature extraction module is used to extract common features of engineering schemes from a pre-acquired engineering knowledge base using contrastive learning, and to construct a knowledge graph for engineering decision-making based on the extracted common features. The association matching module is used to extract engineering anomaly features based on the design specification identification results and construction compliance review results, and to perform engineering decision association matching on the engineering anomaly features based on the knowledge graph to obtain engineering decision association analysis results; The collaborative decision optimization module is used to simulate construction paths based on the results of engineering decision correlation analysis, obtain an optimized engineering schedule plan, and carry out multi-objective collaborative decision optimization in combination with pre-acquired market prices to obtain an optimized engineering decision.

[0056] Specifically, the collaborative decision-making optimization module includes: The project schedule determination module is used to perform construction path simulation based on the results of the correlation analysis of project decisions using generative adversarial networks, and dynamically optimize the resource allocation and timing of adversarial training based on the results of the construction path simulation to obtain the optimized project schedule. The procurement plan determination module is used to input the optimized project schedule plan into the deep reinforcement learning model, and combine it with the pre-acquired market prices to construct a mapping relationship between material demand and procurement time window. Based on the mapping relationship, the module makes short-term predictions on building material price trends, determines the material procurement strategy based on the short-term prediction results, and obtains the optimized procurement plan. The engineering decision module is used to integrate the optimized schedule plan and the optimized procurement plan into a multi-objective process, and to perform collaborative optimization of engineering decisions based on the results of the multi-objective integration using Pareto front analysis, so as to obtain the optimized engineering decision. The project schedule determination module, the procurement plan determination module, and the project decision-making module are connected sequentially.

[0057] The common feature extraction module, the association matching module, and the collaborative decision optimization module are connected sequentially.

[0058] Specifically, Decision Optimization Module 3, or shared knowledge-driven decision optimization, is built through a cross-project knowledge base. This base centrally stores national engineering data, including design models, construction problem databases, and operation and maintenance records. Common features are extracted through comparative learning. A knowledge graph is constructed to link engineering types, typical problems, and solutions, supporting semantic retrieval and solution recommendations. The intelligent optimization engine includes schedule optimization and cost optimization. Schedule optimization uses a GAN (Generative Adversarial Network) to simulate multiple versions of construction paths, dynamically adjusting resource allocation and shortening the construction period by 15%. Cost optimization uses a reinforcement learning model to train material procurement strategies, combining market price fluctuations to predict the optimal procurement time, reducing costs by 12%. The collaborative management platform develops a dual-engine visualization interface for BIM and GIS, supporting multiple roles, including online collaborative annotation and commenting by designers, construction engineers, and supervisors. Violations automatically trigger rectification work orders and link them to the knowledge base to recommend handling cases, with a response time of less than 5 minutes. Decision Optimization Module 3 constructs an intelligent collaborative decision engine that integrates knowledge graphs, generative AI, and reinforcement learning, achieving a closed loop from multi-source anomaly identification to multi-objective engineering decision optimization. Its implementation begins with the common feature extraction module. The system integrates multimodal data from a historical project database, encompassing 217 building, bridge, and underground engineering cases, industry standards, construction logs, supervision reports, and expert experience databases, forming an engineering knowledge base with structured records. Based on this, a contrastive learning framework is used for common feature extraction. Each engineering project is encoded as a high-dimensional vector, and a BERT-BiLSTM hybrid encoder processes the text description, while GCN processes the BIM topology. The InfoNCE loss function is used to narrow down similar solutions, such as the embedding distance for super high-rise core tube construction and to push away dissimilar solutions. After training, the model achieves a clustering accuracy of 86.4% on the validation set, successfully extracting generalizable common features for engineering decisions. For example, deep foundation pit support requires consideration of groundwater dynamics, and large-span steel structures require segmented hoisting. These features, acting as nodes, are combined with entity relationships such as dependency, constraint, and substitution to construct a knowledge graph covering four dimensions—design, construction, cost, and safety—for engineering decision-making. This knowledge graph is stored in the Neo4j graph database, containing nodes and edges, and supports SPARQL semantic queries. The association matching module receives input from preceding modules, including design specification identification results from the reality generation module 1 (e.g., insufficient clear width of evacuation staircases, missing fire separation distances) and construction compliance review results from the construction evolution module 2 (e.g., negative entropy islands in transfer beams, overloaded formwork supports). The system first uses a RoBERTa-based fine-tuned Named Entity Recognition (NER) model to extract engineering anomaly features, identifying three main categories and 47 subtypes: structural (e.g., missing components, excessive deformation); process (e.g., reversed procedures, insufficient maintenance); and resource (e.g., insufficient equipment, delayed materials).These abnormal features are used as query conditions to perform multi-hop reasoning in the knowledge graph. For example, when the input is that the verticality deviation of the core tube wall is >1 / 400, the system searches along the path of abnormality, affected components, related construction schemes, and recommended corrective measures, and matches historical similar cases with a similarity >0.82 and their corresponding decisions, such as adjusting the hydraulic synchronization accuracy of the climbing formwork and increasing the retesting frequency of the laser plumb bob. The system outputs structured engineering decision association analysis results, which include four elements: the cause of the abnormality, the related process, the recommended measures, and the expected effect.

[0059] The analysis results drive the collaborative decision-making optimization module to perform multi-objective optimization. Its sub-module, the project schedule determination module, first uses a generative adversarial network (GAN) to simulate construction paths. The generator G takes the results of correlation analysis as input, such as the need to insert corrective actions, and outputs multiple versions of construction paths that satisfy the logical constraints of the actions, represented as a sequence of triples of actions, resources, and time. The discriminator D evaluates the feasibility of the paths based on historical high-quality project data, such as resource conflict rate <5% and critical path float time >0. After training for 120 rounds with a Wasserstein generative adversarial network with gradient penalty (WGAN-GP), the generated paths improved the schedule rationality index by 23%. The system selects representative paths from the Pareto optimal solution set and combines them with resource smoothing algorithms, such as the minimum variance method, to dynamically adjust the allocation of manpower and machinery, outputting an optimized project schedule plan. In a certain actual project, the original planned construction period of 580 days was compressed to 493 days, a reduction of 15%, and the peak resource resource level decreased by 18%. The procurement plan determination module takes the schedule plan as input and combines it with pre-acquired building material market price data, including daily prices of 12 main materials such as steel, concrete, and aluminum formwork over the past three years, to construct a mapping relationship between material demand and procurement time windows. Specifically, based on the bill of materials (BOM) and time sequence of each process in the schedule plan, it generates daily material demand curves for the next 180 days. On this basis, an LSTM network is used to make short-term price trend predictions, with an input window of 30 days, outputting the average price and fluctuation range for the next 7 days. The MAPE on the test set is 4.7%. Subsequently, the prediction results are embedded into a deep reinforcement learning framework, specifically training the agent using the Proximal Policy Optimization (PPO) algorithm. The state space includes current inventory, price prediction, and schedule constraints; the action space includes procurement quantity and procurement timing; and the reward function is defined as a negative total cost, including procurement costs, warehousing costs, and losses due to material shortages and downtime. After multiple interactive training sessions, the agent learned to make bulk purchases during periods of low prices, such as when steel mills are ramping up production at the end of a quarter, and output optimized procurement plans. In pilot projects, steel procurement costs decreased by 12.3%, overall concrete costs decreased by 9.8%, and the risk of material shortages was reduced to below 0.5%. The engineering decision-making module integrates the optimized schedule plan with the procurement plan through multi-objective fusion, constructing a four-dimensional objective function that includes the construction period T, total cost C, resource balance R, and safety risk S, where S is quantified by the entropy flow early warning frequency of the construction evolution module. The NSGA-II multi-objective genetic algorithm is used to solve the Pareto front, with a population size of 200 and 150 generations, ultimately selecting three non-dominated solutions for decision-makers to choose from. The system automatically generates a decision report, including a comparison of indicators for each plan, such as Plan A with T=493d and C=¥284 million; Plan B with T=510d and C=¥276 million, sensitivity analysis (e.g., the impact of ±10% steel price on costs), and an implementation roadmap.In a certain super high-rise project, the final selected solution achieved comprehensive benefits including a 15% reduction in construction period, an 11.7% reduction in cost, and a 22% reduction in high-risk operations.

[0060] In summary, this system achieves intelligent full-cycle digital management of engineering projects through five core technologies: First, it integrates UAVs and BIM for modeling, using oblique photogrammetry to achieve millimeter-level registration with BIM models, supporting dynamic comparison between design and reality; second, it constructs a multi-dimensional compliance engine to automatically review over 300 regulatory clauses related to geometry, process, and safety; third, it conducts spatiotemporal analysis of violations, integrating time-series images and construction logs to deeply mine violation patterns; fourth, it establishes a shared knowledge network architecture to achieve secure, anonymized sharing and knowledge distillation of cross-project data; and fifth, it creates a dynamic digital twin sandbox, coupling BIM, reality, and fluid simulation to provide integrated decision support. To support these capabilities, the system has made breakthroughs in five key technologies: First, it employs a point cloud and BIM high-precision registration algorithm, improving registration efficiency by 5 times through enhanced ICP feature matching strategies, achieving a single project registration time of less than 10 minutes. Second, it develops a computable technology for specification clauses, transforming textual specifications into BIM attribute constraints, increasing the review automation rate from 30% to 95%. Third, it constructs a spatiotemporal violation identification model, combining Transformer temporal analysis and 3D geofencing, achieving a violation detection false negative rate of less than 2%. Fourth, it innovates a knowledge distillation and sharing mechanism, enabling safe transfer of experience across projects through feature extraction and model fine-tuning, with accuracy loss controlled within 3%. Fifth, it develops a generative schedule optimization engine, using conditional GANs to dynamically simulate construction paths and supporting adaptive adjustments for unexpected conditions. For example, in practical applications, construction errors in quality control are reduced by 80%, and the number of acceptance and rectification times decreases by 70%; safety management is significantly strengthened, with a violation detection rate of 98% and an accident rate reduced by 90%; decision-making efficiency is greatly improved, with compliance review time shortened from 3 days to 2 hours and cross-departmental collaboration efficiency increased tenfold; cost control is outstanding, with material waste reduced by 35% and losses due to delays in a single large project reduced by 50 million yuan; the industry empowerment value is prominent, providing a reusable full-cycle digital management paradigm for the infrastructure sector and effectively promoting a 40% increase in the industry's intelligent penetration rate.

[0061] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A digital management system for the entire lifecycle of intelligent engineering projects, characterized in that: include: The reality generation module is used to generate a 3D reality model based on the collected engineering image data, and to identify the design specifications of the 3D reality model in combination with the preset engineering design behavior specifications to obtain the design specification identification result. The construction evolution module is used to perform construction evolution simulation on the 3D reality model, and conduct construction compliance review based on the construction evolution simulation results and preset construction specification constraints to obtain the construction compliance review results. The construction evolution module includes: The construction specification embedding module is used to embed construction specification clauses into the building information model based on a pre-established industrial basic standard extension framework, thereby obtaining a standardized building information model. The engineering twin construction module is used to integrate standardized building information models, 3D reality models, and pre-acquired engineering environment data to construct a dynamically updated engineering twin. The dynamic simulation module is used to dynamically simulate the construction status at each stage based on the dynamically updated engineering twin and the preset construction control element set, and obtain the construction evolution simulation results. The dynamic simulation module includes: The structural response evolution module is used to drive a multiphysics simulation engine based on dynamically updated engineering twins and preset construction control element sets to perform state simulation of the entire construction process at time steps, and output the structural response evolution sequence of each stage. The response field dynamic evolution module is used to extract the displacement field and strain field time series of key structural regions from the structural response evolution sequence, and to construct a dynamic evolution dataset of continuous response field using spatial grid cells as the basic analysis unit. The information entropy flow analysis module is used to perform structural information entropy flow analysis on the dynamic evolution dataset, and to perform dynamic tracking and abnormal pattern recognition based on the structural information entropy flow analysis results to obtain construction evolution simulation results; The information entropy stream analysis module includes: The spatial difference analysis module is used to construct the time series of physical quantities for each spatial discrete unit of the dynamic evolution dataset, and calculate the spatial difference of the strain gradient field between adjacent units based on the time series, so as to obtain the strain gradient non-uniformity index of each unit at different times. The entropy yield calculation module is used to treat the response fluctuation of each unit as a local information evolution process. It uses the kernel density estimation method to perform probability density modeling on the time window sequence of the non-uniform strain gradient index, and combines the differential entropy formula to calculate the local entropy yield. The synchronization mapping module is used to synchronize the local entropy yield of all units in the spatial and temporal dimensions, and construct a spatiotemporal entropy yield distribution map. The dynamic tracking module is used to perform trend analysis on the entropy value sequence of each spatial location in the spatiotemporal entropy yield distribution map using a sliding time window mechanism, and to perform anomaly pattern recognition based on the trend analysis results. The anomaly pattern recognition results are then correlated with the pre-acquired construction load data to obtain the construction evolution simulation results. The decision optimization module is used to construct a knowledge graph based on a pre-acquired engineering knowledge base, perform engineering decision correlation analysis on the design specification identification results and construction compliance review results based on the knowledge graph, and carry out collaborative decision optimization based on the results of the engineering decision correlation analysis.

2. The intelligent engineering full-cycle digital management system according to claim 1, characterized in that, The real-scene generation module includes: The dense point cloud generation module is used to input the collected engineering image data into the motion reconstruction structure algorithm and generate a dense point cloud, and then construct a 3D reality model based on the dense point cloud. The difference region annotation module is used to align the 3D reality model with the building information model based on the improved iterative nearest point algorithm to obtain the aligned 3D reality model, and to annotate the difference regions of the aligned 3D reality model to obtain the 3D reality difference annotation results. The design specification recognition module is used to perform quantitative analysis of design indicators based on the 3D real-world difference annotation results using a visual model, and then combine the quantitative analysis results of design indicators with the preset engineering design behavior specification constraints to perform design specification recognition and obtain the design specification recognition result. The real-scene generation module, the construction evolution module, and the decision optimization module are connected in sequence; the dense point cloud generation module, the difference region annotation module, and the design specification identification module are connected in sequence.

3. The intelligent engineering full-cycle digital management system according to claim 2, characterized in that, The design specification identification module includes: The semantic boundary completion module is used to construct a causal graph model based on the acquired engineering design data, and to perform causal inference and semantic boundary completion on the 3D real scene difference annotation results using the causal graph model to obtain the completed difference annotation results. The self-attention mechanism capture module is used to input the completed difference annotation results into the Transformer's visual model and use the self-attention mechanism to capture the relationship between geometric structure and global context, so as to obtain a high-dimensional feature embedding representation of each difference region. The design index regression prediction module is used to decode the high-dimensional feature embedding representation using a geometric perception decoding network, and to perform end-to-end design index regression prediction based on the decoding results, so as to obtain the design index quantitative analysis results. The logical matching module is used to logically match the quantitative analysis results of design indicators with the preset engineering design behavior norms and generate design norm identification results based on the logical matching results. The semantic boundary completion module, the self-attention mechanism capture module, the design index regression prediction module, and the logical matching module are connected in sequence.

4. The intelligent engineering full-cycle digital management system according to claim 3, characterized in that, The semantic boundary completion module includes: The engineering logic construction module is used to extract design goals from the acquired engineering design data, and to implement engineering logic deduction and construction based on the design goals using rule reasoning to obtain a hierarchical design logic chain. The causal graph model building module is used to take the design goal as the top-level node and associate it with the hierarchical design logic chain. It performs geometric and spatial representation mapping and graph structure form expression on the association results to obtain a structured causal graph model. The causal inference module is used to input the 3D real scene difference annotation results into the causal graph model, and use the counterfactual reasoning mechanism to perform context matching causal path query on the missing regions in the 3D real scene difference annotation results. Based on the query results, the missing regions are filled in with geometric boundaries and semantic attributes to obtain the completed difference annotation results. The engineering logic construction module, the cause-effect graph model construction module, and the cause-effect inference module are connected sequentially.

5. The intelligent engineering full-cycle digital management system according to claim 1, characterized in that, The construction evolution module also includes: The construction compliance review module is used to extract quantitative features of construction status from the construction evolution simulation results, compare and analyze the quantitative features of construction status with the preset construction specification constraints, judge the construction compliance based on the comparison and analysis results, and obtain the construction compliance review results. The construction specification embedding module, the engineering twin construction module, the dynamic simulation module, and the construction compliance review module are sequentially connected.

6. The intelligent engineering full-cycle digital management system according to claim 1, characterized in that, The structural response evolution module, the response field dynamic evolution module, and the information entropy flow analysis module are connected sequentially.

7. The intelligent engineering full-cycle digital management system according to claim 1, characterized in that, The spatial difference analysis module, the entropy productivity calculation module, the synchronization mapping module, and the dynamic tracking module are connected in sequence.

8. The intelligent engineering full-cycle digital management system according to claim 7, characterized in that, The expression for calculating the local entropy yield is as follows: ； In the formula, Representation unit i exist t n The estimated value of the differential entropy at time t; N n Indicates time step n The total number of samples; j Indicates the index of the currently evaluated sample; k Indicates the reference sample index used for kernel density estimation; σ n This represents the Gaussian kernel bandwidth at the corresponding time step; Indicates time step t n Next j One observation sample; Indicates time step t n Next k One observation sample; h i ( t n-1 ) represents a unit i exist t n-1 The estimated value of the differential entropy at time Δ t Indicates the time step; Representation unit i exist t n The local entropy production rate at time t.

9. The intelligent engineering full-cycle digital management system according to claim 1, characterized in that, The decision optimization module includes: The common feature extraction module is used to extract common features of engineering schemes from a pre-acquired engineering knowledge base using contrastive learning, and to construct a knowledge graph for engineering decision-making based on the extracted common features. The association matching module is used to extract engineering anomaly features based on the design specification identification results and construction compliance review results, and to perform engineering decision association matching on the engineering anomaly features based on the knowledge graph to obtain engineering decision association analysis results; The collaborative decision optimization module is used to simulate the construction path based on the results of engineering decision correlation analysis, obtain an optimized engineering schedule plan, and carry out multi-objective collaborative decision optimization in combination with pre-acquired market prices to obtain an optimized engineering decision. The common feature extraction module, the association matching module, and the collaborative decision optimization module are connected in sequence.

10. The intelligent engineering full-cycle digital management system according to claim 9, characterized in that, The collaborative decision-making optimization module includes: The project schedule determination module is used to perform construction path simulation based on the results of the correlation analysis of project decisions using generative adversarial networks, and dynamically optimize the resource allocation and timing of adversarial training based on the results of the construction path simulation to obtain the optimized project schedule. The procurement plan determination module is used to input the optimized project schedule plan into the deep reinforcement learning model, and combine it with the pre-acquired market prices to construct a mapping relationship between material demand and procurement time window. Based on the mapping relationship, the module makes short-term predictions on building material price trends, determines the material procurement strategy based on the short-term prediction results, and obtains the optimized procurement plan. The engineering decision module is used to integrate the optimized schedule plan and the optimized procurement plan into a multi-objective process, and to perform collaborative optimization of engineering decisions based on the results of the multi-objective integration using Pareto front analysis, so as to obtain the optimized engineering decision. The project schedule determination module, the procurement plan determination module, and the project decision-making module are connected sequentially.

Images