Electric power engineering construction management and control method and system based on digital twinning

Abstract

The invention relates to the technical field of engineering management and control, in particular to an electric power engineering construction management and control method and system based on digital twinning. The method comprises the following steps: acquiring heterogeneous sensing data of a construction area, and carrying out dynamic modeling on a structural member to obtain a member-level digital twin grid; performing environmental factor-material performance response coupling based on the component-level digital twin grid to obtain a material environmental response distribution diagram; component environment sensitivity dynamic injection is carried out according to the component-level digital twin grids, and a construction process dynamic map is obtained; performing multi-mode construction state field deduction based on the construction process dynamic map to obtain a construction state evolution distribution field; and performing reversible process scheduling optimization and simulation according to the construction state evolution distribution map to obtain a reversible scheduling optimization map. The efficiency of electric power engineering construction and the resource utilization rate can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of engineering management and control, and particularly to a method and system for power engineering construction management and control based on digital twin. Background Art

[0002] During the construction process of the power system, there are many construction links involved, the construction period is long, the construction environment is complex and changeable, and there is a highly coupled relationship among multi-dimensional indicators such as project quality, safety, and progress. At present, the traditional power engineering construction management methods generally rely on means such as static two-dimensional drawings, manual on-site supervision, and offline communication and coordination, lacking the ability of dynamic modeling and real-time perception of the entire project life cycle. On the one hand, it is difficult for traditional methods to achieve data penetration in the design, construction, and operation and maintenance stages, resulting in serious problems of information fragmentation and data islands, and it is impossible to timely grasp the actual progress and construction quality status of the project; on the other hand, the project management decision-making process highly depends on the experience of managers, lacking scientific support based on multi-source data fusion analysis, and prone to problems such as project node delays, unreasonable allocation of construction resources, and lagging discovery of quality hazards, directly affecting the project construction efficiency and operation reliability.

[0003] In addition, in the face of the complex and changeable construction site environment, traditional methods lack effective technical means in engineering simulation, process optimization, construction simulation, etc., and it is difficult to make quick responses and dynamic adjustments to emergencies during the project process, reducing the project's risk response ability and system resilience. Especially in large-scale engineering projects across regions and specialties, there are many participating units, scattered data sources, and complex management systems, and it is even more difficult for traditional methods to achieve collaborative perception and intelligent management and control of the entire project system. Summary of the Invention

[0004] Based on this, it is necessary for the present invention to provide a method and system for power engineering construction management and control based on digital twin to solve at least one of the above technical problems.

[0005] To achieve the above object, a method for power engineering construction management and control based on digital twin includes the following steps:

[0006] Step S1: Obtain heterogeneous sensing data of the construction area, and perform dynamic modeling of structural components based on the heterogeneous sensing data of the construction area to obtain a component-level digital twin grid;

[0007] Step S2: Analyze the state of building components and the state of the construction area environment based on the component-level digital twin grid, and perform environmental factor-material property response coupling on the state of building components and the state of the construction area environment to obtain a material environment response distribution map;

[0008] Step S3: Analyze the engineering construction process dependencies based on the component-level digital twin grid to obtain a multi-dimensional construction process dependency graph; dynamically inject the component environmental sensitivity into the multi-dimensional construction process dependency graph using the material environment response distribution map to obtain a dynamic construction process map;

[0009] Step S4: Deduce the multi-modal construction state field based on the dynamic construction process map to obtain a construction state evolution distribution field;

[0010] Step S5: Conduct state anomaly multi-hop causal backtracking based on the construction state evolution distribution map to obtain an abnormal causal conduction chain diagram, and perform reversible process scheduling optimization and simulation based on the abnormal causal conduction chain diagram to obtain a reversible scheduling optimization map.

[0011] The present invention realizes comprehensive information integration and dynamic monitoring in the process of power engineering construction by constructing a multi-dimensional construction management system based on digital twins. First, by acquiring heterogeneous sensing data in the construction area and performing dynamic modeling of structural components, a component-level digital twin grid is obtained, realizing real-time perception and monitoring of construction site data, improving data timeliness and accuracy, and timely discovering potential problems. The coupling of environmental factors and material performance responses generates a material environment response distribution map, revealing the impact of construction conditions on materials, optimizing material selection and use, and improving construction efficiency. By analyzing the construction process dependencies and dynamically injecting them to generate a dynamic construction process map, real-time monitoring and adjustment of the construction process are realized. Using this dynamic map for environmental adaptability optimization between processes reduces construction delays caused by environmental changes. At the same time, multi-modal construction state field deduction is carried out to help identify potential risks, and the abnormal causal backtracking mechanism can trace construction anomalies, optimize decisions and reduce risks. Through reversible process scheduling optimization and simulation, flexible adjustment during the construction process is achieved, improving project resilience and response capabilities. In summary, the present invention can accurately simulate and control key variables in the construction process, effectively enhancing the perception, response and prediction capabilities of the entire project management system, thus significantly improving the scientificity, timeliness and accuracy of project management. This construction management solution based on multi-source data fusion not only breaks through the problems of data fragmentation and information islands in the traditional construction management mode, but also provides strong data support and intelligent prediction for project decision-making, significantly improving the efficiency, quality and safety of project construction.

[0012] Optionally, Step S1 is specifically:

[0013] Step S11: Acquire heterogeneous sensing data in the construction area and perform data preprocessing to obtain standardized heterogeneous sensing data, including lidar scan data, building structure camera image sets, component material coding sets, GNSS positioning data, and environmental sensing data;

[0014] Step S12: Perform semantic segmentation and component recognition on the standardized heterogeneous sensing data, extract the semantic features of component boundaries, structural categories, and material properties, and obtain the initial semantic label set of components;

[0015] Step S13: Based on the initial semantic label set of components and the lidar scan data, perform three-dimensional contour modeling of components, and perform topological relationship reasoning on the three-dimensional contours of components to generate a three-dimensional component topological structure set;

[0016] Step S14: Integrate the three-dimensional component topological structure set and the GNSS positioning data to establish a spatial mapping relationship, and perform coordinate registration and error correction to generate a component spatial position mapping model;

[0017] Step S15: Combine the initial semantic label set of components, the three-dimensional component topological structure set, and the component spatial position mapping model to construct a component-level digital twin unit, and generate a set of component twins;

[0018] Step S16: Organize the set of component twins in a grid, establish a multi-level digital structure mapping relationship, and output a component-level digital twin grid.

[0019] The present invention obtains and standardizes the heterogeneous sensing data in the construction area, integrates data from different sources, eliminates the problem of inconsistent formats, provides an accurate data basis for subsequent analysis, ensures seamless fusion of data in the same coordinate system, and improves the reliability of analysis. During semantic segmentation and component recognition, the component boundaries, structural categories, and material property are accurately extracted, laying a foundation for component modeling and quality monitoring. The three-dimensional contour modeling and topological reasoning of components, combined with lidar and semantic labels, accurately describe the spatial relationships of components, improving the accuracy of spatial positioning and risk assessment. Integrating the three-dimensional component topology with GNSS positioning data ensures the accurate position of components at the construction site and improves the positioning accuracy. The constructed set of digital twins provides a real-time virtual model for each component, facilitating real-time monitoring and risk prediction. The grid organization and multi-level structure mapping relationship ensure the consistency and accuracy of components, improving the intelligence and efficiency of construction. This solution greatly improves the visualization, data integration, and management capabilities at the construction site, solves the problems of information fragmentation and data silos in traditional methods, and enhances the management efficiency and safety of construction projects.

[0020] Optionally, step S15 is specifically as follows:

[0021] Step S151: Extract the boundary coordinates, structural categories, and material property parameters of each component in the initial semantic label set of components, and perform component ID-level matching with the three-dimensional component topological structure set to generate a semantic-structure association index table;

[0022] Step S152: Perform component three-dimensional geometry-material semantic attribute binding based on the semantic-structure association index table, establish a component geometric semantic fusion unit, and generate a component semantic structure fusion body;

[0023] Step S153: Perform global coordinate pose mapping transformation on the component coordinate data in the component spatial position mapping model and the position anchor points in the component semantic structure fusion body to obtain a component instantiation unit;

[0024] Step S154: Perform attribute structuring processing on the component instantiation unit to generate a component attribute data set;

[0025] Step S155: Establish multi-instance component twin mapping rules based on the constructed attribute data set, and use the multi-instance component twin mapping rules to perform structure instance merging and index rearrangement to obtain a set of component twins.

[0026] By extracting the boundary coordinates, structure categories, and material properties in the initial semantic tags of components and performing ID-level matching with the three-dimensional component topology structure set, the present invention can ensure the accurate integration of component information, thereby providing an efficient semantic-structure association for subsequent modeling and analysis. The generation of this index table improves the query and mapping efficiency of component information, enabling the rapid acquisition of detailed component attributes and structural relationships in a complex construction environment. During the component geometry-material semantic attribute binding process, the geometric shape of the component can be accurately combined with the material properties, enhancing the ability to comprehensively describe component characteristics and providing a reliable basis for subsequent construction quality control and condition monitoring. The global coordinate pose mapping transformation further improves the spatial accuracy and positioning accuracy of components by matching the component spatial position with the position anchor points in the semantic structure fusion body, facilitating real-time tracking and management in a dynamic construction environment. The structuring processing of the component attribute data set further standardizes data storage and management, enabling the attribute data to support the integration and analysis of large-scale construction data in a more efficient manner. Finally, by establishing component twin mapping rules and instance merging, the relationships between multi-instance components can be effectively processed, improving the scalability and adaptability of the system, ensuring the consistency and accuracy of component information throughout the project life cycle, and significantly enhancing the intelligent and efficient management capabilities of the construction process.

[0027] The present invention extracts the boundary coordinates, structural categories, and material properties from the initial semantic tags of components, and performs ID-level matching with the three-dimensional component topological structure to ensure the accurate integration of component information, providing an efficient semantic-structure association for subsequent modeling and analysis. The generation of this index table improves the information query and mapping efficiency, helping to quickly obtain the detailed properties and structural relationships of components in a complex construction environment. During the binding process of component geometric-material semantic attributes, the geometric shape of the component is precisely combined with the material properties, enhancing the ability to comprehensively describe component characteristics and providing a reliable basis for construction quality control and condition monitoring. The global coordinate pose mapping transformation further improves the spatial accuracy and positioning accuracy of components, facilitating real-time tracking and management in a dynamic construction environment. The structured processing of the component attribute dataset standardizes data storage and management, supporting the integration and analysis of large-scale construction data. Finally, through component twin mapping rules and instance merging, the relationships of multi-instance components are processed, improving the system scalability, ensuring the consistency and accuracy of component information throughout the project lifecycle, and enhancing the intelligent and efficient management capabilities of the construction process.

[0028] Optionally, step S16 is specifically as follows:

[0029] Step S161: Set the minimum partition granularity to 1.0 m 3 to perform spatial pose partitioning and classification on the component twin set, obtaining a component spatial partition table;

[0030] Step S162: Based on the component spatial partition table, construct the hierarchical relationship of the spatial partition structure, and set the maximum hierarchical depth to 5 for recursive parsing of the structure hierarchy, obtaining a multi-level component partition tree;

[0031] Step S163: Perform hierarchical reorganization and position optimization on the multi-level component partition tree, calculate the spatial aggregation degree, and perform aggregation adjustment on the structural nodes of the multi-level component partition tree according to the spatial aggregation degree to generate an optimized structural hierarchy map;

[0032] Step S164: Based on the optimized structural hierarchy map, and set the minimum side length of the three-dimensional grid cell to 0.5 m to perform spatial grid organization on the component twin set, thereby constructing a component grid index set;

[0033] Step S165: Combine the component grid index set and the multi-level component partition tree for grid cell classification and encapsulation to generate a set of grid-component cluster units;

[0034] Step S166: Perform unified data structure encapsulation on the set of grid-component cluster units and embed a 10 s dynamic update mechanism to obtain a component-level digital twin grid.

[0035] The present invention sets the minimum partition granularity to 1.0 m 3And carry out component spatial pose zoning and classification, which can effectively refine the division of component information according to spatial positions, improve the efficiency and accuracy of spatial data management, and contribute to subsequent spatial analysis and construction decision-making. When constructing the hierarchical relationship of the spatial zoning structure, the maximum hierarchical depth is set to 5, ensuring a moderate depth of hierarchical parsing, which can not only reflect the fine-grained relationships between components but also avoid overly complex computational overhead, facilitating management and optimization. The hierarchical reorganization and position optimization of the multi-level component zoning tree effectively improve the spatial aggregation degree. Through the aggregation adjustment of structural nodes, the spatial layout of components is optimized, and the resource utilization rate and construction efficiency at the construction site are improved. When carrying out grid organization, the side length of the minimum unit of the three-dimensional grid is set to 0.5 m, which can improve the computational efficiency of grid division while ensuring accuracy, making data processing in a large-scale construction environment more efficient. The combination of the component grid index set and the multi-level zoning tree further optimizes the classification and encapsulation of grid cells, improving the flexibility and adaptability of grid management. Through unified data structure encapsulation and a dynamic update mechanism (updated every 10 seconds), it is ensured that the component-level digital twin grid can reflect on-site changes in real time during the construction process, improving the real-time performance and accuracy of construction management.

[0036] The present invention refines the division of the component space by setting the minimum zoning granularity to 1.0 m 3 Refines the division of the component space, improves the efficiency and accuracy of spatial data management, and provides support for subsequent spatial analysis and construction decision-making. When constructing the hierarchical structure of the spatial zoning, the maximum hierarchical depth is 5, ensuring a moderate depth of hierarchical parsing, which can not only reflect the fine-grained relationships between components but also avoid overly complex computational overhead. The reorganization and optimization of the multi-level component zoning tree improve the spatial aggregation degree, optimize the spatial layout of components, and improve the resource utilization rate and construction efficiency. The side length of the minimum unit of the three-dimensional grid is set to 0.5 m, which improves the computational efficiency of grid division while ensuring accuracy and is suitable for a large-scale construction environment. The combination of the component grid index set and the zoning tree optimizes the flexibility and adaptability of grid management. Through unified data structure encapsulation and a dynamic update mechanism every 10 seconds, it is ensured that the component-level digital twin grid can reflect on-site changes in real time, improving the real-time performance and accuracy of construction management.

[0037] Optionally, the coupling of environmental factors - material property responses in step S2 is specifically as follows:

[0038] Based on the component-level digital twin grid, and combined with GNSS positioning data and the building structure camera image set, identify the actual construction component environmental exposure status to obtain a set of component environmental exposure parameters;

[0039] Perform spatial interpolation and time alignment on the set of component environmental exposure parameters and environmental sensing data to generate a spatio-temporal distribution matrix of component-level environmental factors;

[0040] Obtain the material property parameter library, and based on the material property parameter library, conduct material type - environmental factor association to obtain the component material response mapping matrix;

[0041] Fuse the component material response mapping matrix and the component - level environmental factor spatio - temporal distribution matrix, and conduct tensor coupling modeling to establish the material response tensor field;

[0042] Conduct multivariate sensitivity analysis and response weight normalization on the material response tensor field, evaluate the component performance degradation risk degree, and generate the component performance risk sensitivity matrix;

[0043] Combine the component spatial position mapping model with the component performance risk sensitivity matrix, conduct spatial mapping visualization, and obtain the material - environment response distribution map.

[0044] The present invention combines component - level digital twin grids and GNSS positioning data to accurately identify the environmental exposure status of construction components, generate an environmental exposure parameter set, and achieve real - time monitoring. By performing spatial interpolation and time alignment on component environmental exposure parameters and environmental sensing data, a component - level environmental factor spatio - temporal distribution matrix is generated to ensure data consistency and provide a reliable basis for subsequent analysis. Using the material property parameter library to associate material types with environmental factors, a material response mapping matrix is constructed to reflect the impact of environmental factors on material properties. Through tensor coupling modeling, the component material response mapping matrix and the environmental factor spatio - temporal distribution matrix are fused to establish a material response tensor field, providing a multi - dimensional analysis perspective. Conduct sensitivity analysis and weight normalization on the material response tensor field, evaluate the component performance degradation risk, generate a risk sensitivity matrix, and optimize construction decisions. Finally, combine the component spatial position mapping model with the risk matrix for spatial mapping visualization, intuitively display the material - environment response distribution, and improve the risk prediction and response capabilities.

[0045] Optionally, constructing the response tensor field specifically includes:

[0046] Extract the material types, response model parameters, and coupling factor weight information in the component material response mapping matrix, identify the material response function families corresponding to various components, and perform material property mapping indexing according to the component IDs in the component - level digital twin grid to generate a component response function index table;

[0047] Extract the time - series environmental variables of the components from the component - level environmental factor spatio - temporal distribution matrix, construct a component - environmental factor - time three - order input tensor, and perform environmental variable response channel matching according to the component response function index table to generate a component environmental response input tensor;

[0048] According to the component material coding set, use the material property parameter library to extract the rotational degrees of freedom of the contact joints, the coating paint thickness and aging grade, and the electrochemical sensitivity coefficients of the electrode components;

[0049] Perform structural behavior response analysis and quantification of adjustment factors for the rotational degrees of freedom of the contact joint, the thickness of the painted surface layer, the aging level, and the electrochemical sensitivity coefficient of the electrode component, respectively, to obtain the joint flexibility response adjustment factor, the paint layer shielding attenuation factor, and the electrode corrosion response factor;

[0050] Normalize the joint flexibility response adjustment factor, the paint layer shielding attenuation factor, and the electrode corrosion response factor, establish a structural sensitivity factor table, and expand the structural sensitivity factor table into a structural response adjustment factor matrix;

[0051] Perform tensor-level coupling and fusion of the component environmental response input tensor and the structural response adjustment factor matrix to generate an initial material response tensor field;

[0052] Perform tensor normalization and dimension compression on the initial material response tensor field, and reconstruct the compressed tensor field to obtain the material response tensor field.

[0053] The present invention can efficiently generate a response function index table by accurately extracting the material types and coupling factor weight information in the component material response mapping matrix, and performing attribute mapping in combination with the component ID, which improves the organization and query efficiency of component material information. The constructed third-order input tensor matches the response channels, ensuring the accurate mapping of environmental variables to material responses, and providing a stable data basis for subsequent material response analysis. By extracting the structural characteristics of the contact joint and the electrochemical sensitivity of the electrode, it is possible to quantify and optimize the effects of joint flexibility, paint layer attenuation, and electrode corrosion on material properties, enhancing the prediction ability of component performance changes. The normalized structural sensitivity factors further improve the comparability between different factors, and the constructed response adjustment factor matrix provides a reliable basis for dynamic adjustment and optimization. Coupling and fusing the environmental response input tensor and the adjustment factor matrix improves the accuracy and application range of the material response tensor field. Finally, through tensor normalization and compression, it ensures the efficient calculation and storage of the response model, and optimizes the scientific nature of component performance evaluation and construction decision-making.

[0054] Optionally, step S3 is specifically as follows:

[0055] Step S31: According to the spatial topological structure and component material label information in the component-level digital twin grid, identify the installation, nesting, support, and occlusion relationships between components, and construct a component-level process dependency graph;

[0056] Step S32: Based on the CNSS positioning data and the component spatial position mapping model, perform spatial position projection and adjacency analysis on the component-level process dependency graph to generate a multi-dimensional construction process dependency graph;

[0057] Step S33: Invoke the material environment response distribution map, and embed the component environment response intensity value in the material environment response distribution map into each component node of the multi-dimensional construction process dependency graph to form a node environment response injection graph;

[0058] Step S34: Based on the node environment response injection graph, extract the environmental response gradient changes of the front and rear component nodes connected by each process edge, and calculate the environmental adaptation weight of the process dependency path in the component-level process dependency graph to generate a process edge response sensitivity matrix;

[0059] Step S35: Integrate the node environment response injection graph and the process edge response sensitivity matrix, construct a weighted directed graph, and perform node and path dynamic feasible region correction to obtain a construction process dynamic graph.

[0060] Through the combination of component-level digital twin grids and material label information, the present invention can efficiently identify the installation, support, and occlusion relationships between components, construct a process dependency graph, and provide a basis for the visualization and optimization of the construction process. When performing spatial position projection and adjacency analysis, by combining CNSS positioning data and the component spatial position mapping model, the accuracy of the component positions in the process dependency graph is ensured, and the scientificity and accuracy of the construction path planning are improved. By embedding the material environment response intensity value into the component nodes, the influence of different environments on each component can be accurately reflected, enhancing the dynamic adaptability during the construction process. When calculating the environmental adaptation weight of the process dependency path, through the analysis of the environmental response gradient, it is ensured that the construction decision-making can be optimized according to the real-time environmental changes. Integrating the node environment response injection graph and the process edge response sensitivity matrix, the constructed weighted directed graph realizes the dynamic adjustment and optimization of the construction process, thereby improving the coordination and efficiency among various processes during the construction process, and ensuring the high efficiency, flexibility, and controllability of the construction process.

[0061] Optionally, step S4 is specifically as follows:

[0062] Step S41: Based on the node environmental response characteristics and path weight parameters in the construction process dynamic graph, construct an input feature vector to generate a construction process state input tensor set;

[0063] Step S42: Perform dynamic time expansion and multi-channel nested encoding on the construction process state input tensor set to establish a process state evolution and propagation mechanism, thereby obtaining a construction process state propagation model;

[0064] Step S43: Through the construction process state propagation model, perform simulation deduction on the construction process dynamic graph to obtain a state path simulation sequence set;

[0065] Step S44: Perform space-time domain fusion analysis on the state path simulation sequence set, extract the activity intensity, state transition frequency and resource consumption change rate of the process nodes in different time periods, and obtain the construction state aggregation index map;

[0066] Step S45: spatially fuse the construction status aggregation index map with the component space mapping model, construct a visualized spatial field of the construction status evolution in each period, and output the construction status evolution distribution field.

[0067] The present invention constructs an input feature vector based on the node environmental response characteristics and path weight parameters of the dynamic graph of the construction process, providing an accurate data basis for the analysis of the subsequent construction process status. When performing dynamic time expansion and multi-channel nested coding, the evolution and propagation mechanism of the process state can be effectively simulated, so that various state changes in the construction process can be more accurately captured and predicted. By simulating and deducing the construction process state propagation model, the changes in the process state can be predicted, providing decision support for construction optimization. The space-time domain fusion analysis of the state path simulation sequence set extracts the active intensity, state transition frequency and resource consumption change rate of the process node, helping to identify the resource consumption trend and potential bottlenecks in the construction process. Finally, the construction state aggregation index graph and the component space mapping model are spatially fused, which can vividly display the evolution process of the construction state and ensure real-time monitoring and efficient management of the construction process.

[0068] Optionally, the state abnormal multi-hop causal backtracking in step S5 is specifically as follows:

[0069] Based on the construction state evolution distribution field, the component state mutation detection is performed, the sliding window is set to 5min and the mutation detection threshold is set to 3σ to identify the abnormal state point set and generate the state abnormality annotation set;

[0070] Call the construction process dynamic graph and state anomaly annotation set, set the upper limit of the backtracking hop count to 4, and set the minimum threshold of the path confidence to 0.65 to perform multi-hop dependent path backtracking of abnormal state points. In each hop propagation, calculate the anomaly transmission weight and generate a multi-hop anomaly backtracking path diagram;

[0071] Based on the multi-hop abnormal backtracking path graph, the starting state, path triggering timestamp, environmental response value, material sensitivity factor and structural logic role of each component node on the backtracking path are extracted to generate a component-level causal propagation vector set.

[0072] Perform Bayesian probability graph modeling on the component-level causal propagation vector set, infer and identify the dominant anomaly source factors, evaluate the anomaly attribution weights of each path, and output the anomaly causal attribution map;

[0073] Integrate the abnormal causal attribution map and the construction process dynamic map, set the injection ratio of the influence coefficient to 0.2 to inject the influence coefficient into the process dependence path, and mark the risk sensitivity level of each component node to generate an abnormal causal conduction chain diagram.

[0074] Through component state mutation detection based on the construction state evolution distribution field, the present invention can effectively identify abnormal state point sets, and accurately capture abnormal state changes during the construction process through a sliding window and a mutation detection threshold (3σ), ensuring timely discovery of potential problems. When performing multi-hop dependence path backtracking of abnormal state points, setting the upper limit of the backtracking hop number (4) and the path reliability threshold (0.65) can effectively control the depth and accuracy of the backtracking process, thereby accurately tracing the abnormal propagation path. By extracting the component node information on the backtracking path, a complete causal propagation vector set can be constructed, and the dominant abnormal source factor can be identified through Bayesian probability graph modeling to ensure the accurate positioning of the abnormal source. Combining the abnormal causal attribution map and the construction process dynamic map, injecting the influence coefficient (0.2) to optimize the abnormal response of the process dependence path makes the risk sensitivity marking more accurate, thereby realizing the generation of the abnormal causal conduction chain diagram.

[0075] Optionally, this specification also provides a digital-twin-based construction control system for power engineering, which is used to execute the digital-twin-based construction control method for power engineering as described above. The digital-twin-based construction control system for power engineering includes:

[0076] A component digital twin module, which is used to obtain heterogeneous sensing data in the construction area and perform dynamic modeling of structural components based on the heterogeneous sensing data in the construction area to obtain a component-level digital twin grid;

[0077] An environmental response analysis module, which is used to analyze the state of building components and the environmental state of the construction area based on the component-level digital twin grid, and perform environmental factor-material property response coupling on the state of building components and the environmental state of the construction area to obtain a material-environment response distribution map;

[0078] An environmental sensitivity injection module, which is used to analyze the dependence relationship of the engineering construction process based on the component-level digital twin grid to obtain a multi-dimensional construction process dependence graph; and dynamically inject the component environmental sensitivity into the multi-dimensional construction process dependence graph by using the material-environment response distribution map to obtain a construction process dynamic map;

[0079] A construction state field deduction module, which is used to perform multi-modal construction state field deduction based on the construction process dynamic map to obtain a construction state evolution distribution field;

[0080] The process scheduling optimization module is used to perform multi-hop causal backtracking of state anomalies based on the construction state evolution distribution diagram, obtain the abnormal causal conduction chain diagram, and perform reversible process scheduling optimization and simulation based on the abnormal causal conduction chain diagram to obtain the reversible scheduling optimization diagram.

[0081] The digital twin-based power engineering construction management and control system of the present invention can implement any digital twin-based power engineering construction management and control method of the present invention. It is a medium for coordinating the operations and signal transmissions between various modules to complete the digital twin-based power engineering construction management and control method. The internal modules of the system cooperate with each other, thereby improving the efficiency and resource utilization rate of power engineering construction. Brief Description of the Drawings

[0082] By reading the detailed description of the non-limiting embodiments with reference to the following drawings, other features, objects, and advantages of the present invention will become more apparent:

[0083] Figure 1 It is a schematic diagram of the step flow of the digital twin-based power engineering construction management and control method of the present invention;

[0084] Figure 2 It is a detailed schematic diagram of the step flow of step S1 in the present invention;

[0085] The realization, functional characteristics, and advantages of the objectives of the present invention will be further described with reference to the embodiments and the drawings. Detailed Embodiments

[0086] The technical method of the present invention will be clearly and completely described below with reference to the drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0087] It should be understood that although the terms "first", "second", etc. may be used here to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, the first unit may be called the second unit, and similarly, the second unit may be called the first unit. The term "and / or" used here includes any and all combinations of one or more of the listed related items.

[0088] To achieve the above objectives, please refer to Figures 1 to 2 , the present invention provides a digital twin-based power engineering construction management and control method, and the method includes the following steps:

[0089] Step S1: Obtain heterogeneous sensing data of the construction area, and perform dynamic modeling of structural components based on the heterogeneous sensing data of the construction area to obtain component-level digital twin grids;

[0090] In this embodiment, a construction site of a transmission line in a mountainous area is selected as the implementation scenario, and multiple types of sensing nodes are deployed, including lidar (model Velodyne VLP-32C), high-definition cameras (4K resolution, 15fps), component encoders, GNSS receivers (error less than 5cm), and environmental sensors such as temperature and humidity, wind speed, and ultraviolet rays. The acquisition period is set to 30 seconds. First, the original data is subjected to timestamp alignment and format standardization processing, and invalid data and noise are removed to form a standardized data set in a unified format. Through a deep convolutional semantic segmentation network (such as DeepLabV3+), the boundaries and material properties of different components such as concrete and steel structures in the image are identified, and combined with lidar point cloud data, a voxel grid filtering algorithm and RANSAC plane fitting are used for three-dimensional contour modeling. The structural connection relationship between components is realized by heuristic topology rule reasoning. For example, if the surface normal angle between components is less than 15°, it is determined as a contact support relationship. The spatial error is corrected by the registration of GNSS and point cloud, and the error is controlled within 3cm through the ICP algorithm. Finally, the semantic labels, component topology, and pose data are fused, and the Octree structure is used to divide the components into three-dimensional grids. The generated component-level digital twin grids contain approximately 400,000 nodes and spatial index units.

[0091] Step S2: Analyze the state of building components and the environmental state of the construction area based on the component-level digital twin grids, and perform environmental factor-material property response coupling on the state of building components and the environmental state of the construction area to obtain a material-environment response distribution map;

[0092] In this embodiment, the GNSS positions and image data in different spatial regions of the component digital twin grids are used to extract indicators such as the sunshine exposure time, occlusion situation, and surface color change of each component, and combined with the on-site environmental sensor data to construct component environmental exposure parameters. Taking 30 minutes as the time step, IDW spatial interpolation and Kalman filtering are used to generate the spatio-temporal distribution matrix of component-level environmental factors. Material property parameters (such as thermal expansion coefficient, corrosion rate, etc.) are extracted from common power component materials such as concrete, steel structure, and insulating ceramics, and a material-environment response mapping relationship is established based on empirical models. Subsequently, a tensor coupling model is constructed by tensor coupling the environmental factor matrix and the material response mapping matrix through a third-order tensor structure to construct a response tensor field. For structural features such as rotating joints, paint surfaces, and electrodes, the joint rotation degree of freedom, paint surface aging level, and electrochemical sensitivity coefficient are respectively introduced as response correction factors, and the performance degradation risk degree of each type of component is calculated through principal component analysis and sensitivity regression. Finally, the material-environment response distribution map is visually output.

[0093] Step S3: Analyze the engineering construction process dependencies based on the component-level digital twin grid to obtain a multi-dimensional construction process dependency graph; dynamically inject the component environmental sensitivity into the multi-dimensional construction process dependency graph using the material environment response distribution map to obtain a construction process dynamic map;

[0094] In this embodiment, the structure connection information in the component grid is used to identify the nesting, installation, and occlusion logics of each component, and a component-level process dependency graph is generated according to the construction standards. On this basis, the relative position distance and spatial adjacency of the components are calculated according to the GNSS position projection to construct a multi-dimensional construction process dependency graph. Call the material environment response distribution map, and inject the maximum response value (unit μg / m 3 or °C) of each component into the corresponding node in the graph to form an environment injection graph. Calculate the environmental adaptability of the process path using the response difference between the front and rear nodes, extract the positions where the gradient change value is greater than the set threshold (such as 2.5) as high-sensitivity nodes, generate a process edge response sensitivity matrix through the graph traversal algorithm, and fuse it to generate a construction process dynamic map.

[0095] Step S4: Deduce the multi-modal construction state field based on the construction process dynamic map to obtain a construction state evolution distribution field;

[0096] In this embodiment, call the construction process dynamic map, use the graph neural network (GAT) as the core model, and fuse multi-modal inputs (component position, material properties, environmental response, etc.) to construct a multi-modal state field deduction framework. Taking every 2 hours as the time period, simulate the state evolution of the construction process under different environmental conditions, consider construction delays (such as when the temperature > 35 °C, the delay factor is set to 1.2) and the risk of component failure, and generate a state evolution distribution field. This field is displayed in the form of a three-dimensional heat map, and different colors reflect the construction efficiency and the structural risk level.

[0097] Step S5: Conduct multi-hop causal backtracking for state anomalies based on the construction state evolution distribution map to obtain an abnormal causal conduction chain graph, and perform reversible process scheduling optimization and simulation based on the abnormal causal conduction chain graph to obtain a reversible scheduling optimization map.

[0098] In this embodiment, state mutation points are extracted based on the construction state evolution distribution field, and the mutation threshold is set such that the change rate of the state index is greater than twice the standard deviation of the mean, so as to generate an abnormal state annotation set. The multi-hop abnormal path backtracking algorithm is executed through a dynamic graph, and the maximum number of backtracking hops is set to 4. The calculation method of the propagation reliability for each hop is the weighted average of the edge weight and the response gradient. After constructing the abnormal propagation path graph, the starting state (such as "support not completed"), trigger time, response value, and structural role of the component nodes in each path are extracted to construct a component-level causal propagation vector, and a Bayesian network is used for inference to identify the most likely main cause node. The influence weights of each path are marked in combination with the process graph, the maximum risk path and its conduction range are evaluated, and finally an abnormal causal conduction chain graph is output. Based on this, a reversible path generation mechanism is implemented, mainly by swapping the order of non-critical paths, delaying the construction of sensitive components, etc. for scheduling optimization. The scheduling optimization uses a ternary sorting metric function Pi = α × Ri + β × Ti + γ × Ei, where Ri is the risk level, Ti is the path order value, Ei is the environmental response intensity, and the parameters α, β, and γ are 0.5, 0.3, and 0.2 respectively. Non-critical path nodes are selected for order adjustment according to the sorting results, and the pre-operation procedures of high-risk nodes are optimized through a local topology adjustment algorithm (such as a critical path iterative optimization algorithm) to enable construction in a more stable time window. During the simulation process, three types of scheduling strategies are considered: 1) swapping paths at the same layer, which is applicable to parallel components; 2) merging and translating paths, where the start times of adjacent high-risk paths are uniformly shifted backward; 3) retaining node jumps, where abnormal path nodes are skipped and a re-construction mark is set under certain conditions. Each optimization strategy is simulated 30 times, and the simulation platform screens the optimal construction process rearrangement plan by comparing indicators such as the state mutation frequency and the change amount of path reliability in the original scheduling graph and the optimized graph. If the reliability of the abnormal path drops by more than 15% after optimization and the total construction period is extended by no more than 8 hours, it is determined as an "acceptable reversible scheduling plan", and an updated reversible scheduling optimization graph is output to provide a basis for real-time dynamic scheduling.

[0099] Optionally, step S1 is specifically as follows:

[0100] Step S11: Obtain heterogeneous sensing data in the construction area and perform data preprocessing to obtain standardized heterogeneous sensing data, including lidar scan data, building structure camera image sets, component material coding sets, GNSS positioning data, and environmental sensing data;

[0101] In this embodiment, heterogeneous sensing devices including a Velodyne HDL-32E lidar, a FLIR infrared imaging camera, a GNSS base station, and distributed environmental monitoring nodes (temperature, humidity, wind speed, PM2.5, etc.) are deployed in the construction area. The acquisition frequency is set to 10 Hz, and the data is aggregated to the edge server through the UDP protocol synchronously. The collected data is subjected to unified clock calibration (maximum time offset ≤ 30 ms), and based on the Kalman filtering algorithm, abnormal point clouds and image occlusion areas are cleaned. At the same time, PCA (principal component analysis) is used to standardize the material coding feature vectors, and finally a standardized heterogeneous sensing data set in a unified format is formed. The data format is encapsulated in a JSON structure, including fields: sensing type, spatial coordinates, timestamp, material label, and environmental attribute fields.

[0102] Step S12: Perform semantic segmentation and component recognition on the standardized heterogeneous sensing data, extract semantic features of component boundaries, structural categories, and material properties, and obtain an initial semantic label set of components;

[0103] In this embodiment, the DeepLabV3+ network is used to perform multi-scale semantic segmentation training on the image data in the standardized heterogeneous sensing data set. The training set selects 5000 labeled component images, uses the cross-entropy loss function, and sets the initial learning rate to 0.001. The component boundary contour lines are identified in the segmentation results, and the lidar point cloud data is structurally matched based on the boundary line integral error controlled within 5 pixels. The extraction of material properties combines the image spectral features and the material coding library for fusion verification. Finally, an initial semantic label set of components is output, including component numbers, structural categories (such as beams, columns, slabs, etc.), material types (such as C30 concrete, Q235 steel, etc.), boundary point sets, and relative positions to the image center.

[0104] Step S13: Based on the initial semantic label set of components and the lidar scan data, perform three-dimensional contour modeling of components, and perform topological relationship reasoning on the three-dimensional contours of components to generate a three-dimensional component topological structure set;

[0105] In this embodiment, three-dimensional contour modeling is performed based on the boundary point set in the initial semantic label set of components and the lidar point cloud. The volume modeling method based on the α-shape algorithm is used to generate a 3D envelope, and the overlapping parts are removed by combining Boolean operations. The point cloud density of each component is required to be greater than 1500 points / m 2 ². Subsequently, using the Voronoi diagram partitioning and Delaunay triangulation algorithms, the contact surface relationship and inclusive structure hierarchy relationship between components are constructed, the connection, nesting, and occlusion topological structures of components are inferred, and a three-dimensional component topological structure set is formed. Each topological edge in the structure contains attribute fields such as connection type, contact area ratio, and relative spatial distance.

[0106] Step S14: Integrate the 3D component topological structure set with GNSS positioning data to establish a spatial mapping relationship, perform coordinate registration and error correction, and generate a component spatial position mapping model;

[0107] In this embodiment, the centroid point of each component in the 3D component topological structure set is used as the initial spatial positioning point, and ID matching is performed with the physical component identifier in the GNSS positioning data. The point cloud registration method based on the RANSAC algorithm is used to optimize the rigid transformation of the component spatial position. The number of iterations is set to 1000, and the error tolerance is 0.02m. At the same time, a differential positioning error model is introduced to correct the GNSS data, and the maximum positioning error does not exceed 5cm. After completing the coordinate alignment, a component spatial position mapping model is generated, and the 3D spatial position and rotation vector in the world coordinate system are output with the component ID as the key value.

[0108] Step S15: Combine the initial semantic label set of components, the 3D component topological structure set, and the component spatial position mapping model to construct component-level digital twin units, and generate a set of component twins;

[0109] In this embodiment, the component category and material property fields in the initial semantic label set of components are associated with the spatial geometric information in the 3D component topological structure set, and the GNSS mapping result is indexed by the unique component number. During the process of constructing digital twin units, image labels, topological positions, material library reference IDs, and current real-time status fields are bound to each component instance. The BIM-IFC standard is used for data encapsulation, and the output fields include: component number, material ID, 3D mesh data (OBJ format), spatial coordinates, topological relationships with other components, etc., and finally a set of component twins is formed.

[0110] Step S16: Organize the set of component twins in a grid manner, establish a multi-level digital structure mapping relationship, and output a component-level digital twin grid.

[0111] In this embodiment, according to the 3D dimensions and spatial positions of each component in the set of component twins, the grid granularity is set to 0.5m 3 , and the construction area is divided into several grid units. Components are assigned to the corresponding grid units according to their centroid positions and the cross-sectional area of the outer envelope. Subsequently, a spatial hierarchical mapping relationship is established, with the top layer being the regional grid index and the secondary layer being the component clustering unit. Through clustering analysis (setting K = 10), adjacent components are spatially aggregated to form a grid-based component cluster. Each component cluster is encapsulated as a digital twin index unit, bound with an update period (refreshing the status every 10 minutes), and embedded with a data access interface API to construct a component-level digital twin grid that supports real-time query and status synchronization.

[0112] Particularly importantly, Step S13 is specifically:

[0113] Invoke the initial semantic tag set of components and perform semantic point cloud clustering on the lidar scan data to generate a set of component point cloud subsets;

[0114] In this embodiment, the structure category and material property tags in the initial semantic tag set of components are invoked, and semantic point cloud clustering is performed in combination with the lidar scan data. The DBSCAN clustering algorithm is used to perform density clustering on the lidar point cloud. The clustering radius parameter is set to 0.15m, and the minimum number of points is 100. The semantic tags of components are used to perform semantic screening and labeling on the clustering results. Each point cloud clustering subset is classified into the corresponding component category, and a set of component point cloud subsets is formed. The clustering accuracy is evaluated by the IoU index, and the clustering accuracy is required to reach above 0.85 to ensure the accuracy of component recognition.

[0115] Perform three-dimensional geometric fitting modeling on the set of component point cloud subsets, extract the contour boundaries of each component, and generate a set of three-dimensional contour models of components;

[0116] In this embodiment, three-dimensional geometric fitting modeling is performed on the set of component point cloud subsets. The plane and cylinder fitting method based on the RANSAC algorithm is used to perform parametric modeling on structures such as beams, columns, and walls. The model fitting error is controlled within ±0.01m. The α-shape boundary envelope processing is performed on each component point cloud, its three-dimensional boundary contour is extracted, and a set of three-dimensional contour models of components is output. The contour model of each component is saved in the OBJ format, including vertex coordinates, face patch information, and normal vector attributes.

[0117] Calculate the relative contact surfaces and boundary distances between components based on the set of three-dimensional contour models of components, and set the minimum contact area threshold to 0.02m 2 and the maximum boundary spacing to 0.1m to screen the associated component pairs;

[0118] In this embodiment, relative spatial analysis is performed on the set of three-dimensional contour models of components to calculate the contact area and boundary spacing between components. The AABB (axis-aligned bounding box) is used to quickly screen the possible contact component pairs, and the contact area of the triangular mesh is accurately calculated. The minimum contact area threshold is set to 0.02m 2 ; at the same time, the KD-Tree nearest neighbor search is performed on the contour boundary point set, and the component pairs with a boundary spacing less than 0.1m are determined as potential associated objects. The screening results are output as the component pair index and the corresponding spatial parameters, and a set of associated component pairs is generated.

[0119] Perform contact type recognition and directional analysis on the associated component pairs, label the intersecting, adjacent, nested, and covering component relationships, and generate a set of semantic topological edges of components;

[0120] In this embodiment, the set of associated component pairs is further subjected to contact type identification and directional analysis. The spatial relative relationship between components is determined by using the included angle between the main direction vectors of the two components and the position of the centroid projection. If there is a large overlapping contact surface and the height difference is <0.05 m, it is determined as "intersecting" or "supporting"; if they are parallel and the distance is between 0.02 - 0.1 m, it is defined as "adjacent"; if one component encloses more than 80% of the envelope of another, it is marked as "nested"; if the upper component blocks ≥50% of the projected area of the lower component, it is identified as "covering". Finally, a set of semantic topological edges of the components is generated, and each edge contains fields such as component pair ID, relationship type, direction vector, and geometric parameters.

[0121] Taking the set of three-dimensional contour models of components as graph nodes and the set of semantic topological edges of components as connecting edges, a topological graph structure is constructed, and connectivity checking and topological closed-loop correction are performed to output a set of three-dimensional component topological structures.

[0122] In this embodiment, each component instance in the set of three-dimensional contour models of components is used as a graph node, and the set of semantic topological edges of components is used as the directed graph connecting edges to construct a three-dimensional component topological graph structure. Connectivity checking is performed through a graph traversal algorithm to ensure that each component node is reachable, and topological closed-loop correction is performed on areas with isolated nodes or missing closed loops. The correction strategies include adding auxiliary edges or adjusting edge weights to make the final component topological graph have complete structural information and geometric consistency.

[0123] Optionally, step S15 is specifically as follows:

[0124] Step S151: Extract the boundary coordinates, structural categories, and material property parameters of each component in the initial semantic label set of components, and perform component ID-level matching with the set of three-dimensional component topological structures to generate a semantic-structure association index table;

[0125] In this embodiment, each component data in the initial semantic label set of components is structurally extracted, and its two-dimensional or three-dimensional boundary coordinates (such as the eight-point coordinates of the bounding box), structural categories (such as beams, columns, walls, slabs), and material property parameters (such as concrete strength grade C30, steel model HRB400, coating thickness, etc.) are extracted. Subsequently, one-to-one ID-level matching is performed with the set of three-dimensional component topological structures through the unique identifier of the component (such as component number or BIM ID). The matching adopts a hash index plus multi-level ID verification mechanism to ensure a stable mapping relationship is established between the component information in the semantic label and the three-dimensional geometric topological structure. Finally, a semantic-structure association index table is generated, and each row records the component ID, semantic category, material property, and the corresponding topological structure ID.

[0126] Step S152: Based on the semantic-structure association index table, perform component three-dimensional geometry-material semantic attribute binding, establish a component geometric semantic fusion unit, and generate a component semantic structure fusion body;

[0127] In this embodiment, the three-dimensional geometric data of each component is bound to its semantic label and material attributes according to the semantic-structure association index table. Specifically, a component fusion unit is defined in JSON format, and the fields include component ID, geometric model path, structure category, material type, surface treatment parameters, component size, orientation angle, etc. The fusion data is encapsulated through a nested structure. During the fusion process, unit conversion is performed on the material parameters, such as converting "mm" to "m", and the strength parameters are standardized and encoded. Finally, a set of component semantic structure fusion bodies is formed.

[0128] Step S153: Perform global coordinate pose mapping conversion on the component coordinate data in the component spatial position mapping model and the position anchor points in the component semantic structure fusion body to obtain a component instantiation unit;

[0129] In this embodiment, the global positioning coordinate data (such as GNSS coordinates or local control point coordinate system) in the component spatial position mapping model is used to perform spatial mapping on the anchor point positions of the geometric models in the component fusion body. During the spatial pose conversion process, a combination of quaternion rotation and Euclidean displacement matrix is used to calculate the actual position that the component should be in the three-dimensional space. To ensure accuracy, the ICP algorithm (Iterative Closest Point) is used to correct the coordinate alignment error, and the error threshold is set to 0.005m. Finally, a component instantiation unit with absolute spatial pose is output, and each instance contains attributes such as position (X, Y, Z), attitude (Pitch, Roll, Yaw), and structure category.

[0130] Step S154: Perform attribute structuring on the component instantiation unit to generate a component attribute data set;

[0131] In this embodiment, further attribute structuring is performed on the component instantiation unit to convert it into a data structure oriented to calculation. Python dictionary nested structure is used for data modeling, and the fields include basic geometric attributes (such as the number of boundary points, volume, surface area), physical parameters (such as weight, density), semantic attributes (such as functional partition, construction stage), and topological association information (such as upstream and downstream component IDs). At the same time, to improve the system operation efficiency, the structured attribute data set is serialized into a binary file or stored in a database for unified management to form a highly accessible component attribute data set.

[0132] Step S155: Establish multi-instance component twin mapping rules based on the constructed attribute data set, and use the multi-instance component twin mapping rules to perform structure instance merging and index rearrangement to obtain a set of component twins.

[0133] In this embodiment, a multi-instance component twin mapping rule is established based on the content of the component attribute dataset. The mapping rule is merged based on the relative relationship between the component ID, component category, and spatial position. For example, if the distance between components of the same type is less than 0.5 m, they can be regarded as replicas of the same template instance. The mapping rule is used to merge and cluster all component instances, perform semantic consistency checks, and reorder the indexes. The index reordering adopts a two-dimensional KD-Tree index structure, which is convenient for fast spatial queries and scheduling operations, and finally outputs a set of component twins.

[0134] Optionally, step S16 is specifically as follows:

[0135] Step S161: Set the minimum partition granularity to 1.0 m 3 to perform spatial pose partitioning and classification on the set of component twins, and obtain a component spatial partition table;

[0136] In this embodiment, the minimum partition granularity is set to 1.0 m 3 , and each component in the set of component twins is partitioned according to its three-dimensional spatial center point coordinates. The specific method is to divide the construction area into equal-volume spatial cubic units, calculate the geometric centroid coordinates of each component, and divide the component into the spatial unit where it is located to construct a spatial partition mapping table. During the division process, if a component spans multiple spatial units, it is classified according to the unit with the largest occupied area to ensure the uniqueness of the partition. Finally, a component spatial partition table is generated, recording each component ID, the corresponding partition number, and the corresponding spatial center point coordinates.

[0137] Step S162: Based on the component spatial partition table, construct a hierarchical relationship of the spatial partition structure, and set the maximum hierarchical depth to 5 for recursive parsing of the structure hierarchy to obtain a multi-level component partition tree;

[0138] In this embodiment, based on the component spatial partition table, a hierarchical tree model of the spatial partition structure is constructed. Using a recursive partitioning method based on the spatial octree (Octree), the maximum hierarchical depth is set to 5 layers, and the partitions are refined and organized layer by layer. In each layer, the partition nodes establish a parent-child hierarchical relationship according to their spatial relative positions. For example, the first layer is a macroscopic division of the building area, and the second to fifth layers correspond to floors, component blocks, component groups, and component units respectively. By recursively parsing the spatial relative subordination relationship of each layer of nodes, a multi-level spatial partition tree is established to ensure the unique positioning of each component in the tree structure and the availability of hierarchical context information.

[0139] Step S163: Perform hierarchical reorganization and position optimization on the multi-level component partition tree, calculate the spatial aggregation degree, and perform aggregation adjustment on the structure nodes of the multi-level component partition tree according to the spatial aggregation degree to generate an optimized graph of the structure hierarchy;

[0140] In this embodiment, after obtaining the multi-level component partition tree, structural level reorganization and position optimization are further carried out. First, calculate the spatial aggregation degree under each structural node, which is defined as the spatial compactness of the component centroids under this node, and the calculation method is the inverse of the average distance between components within the node. If the aggregation degree of a certain node is lower than the set threshold (such as 0.3 / m), then perform the structural node merging operation, and aggregate this node with its adjacent nodes. Combining the aggregation degree information, optimize the positions of the nodes, and adopt the spatial centroid migration strategy to adjust the positions of their parent nodes to better reflect the densely distributed areas of the components. Finally, output the structural level optimization map, recording the merging history of the nodes, the spatial aggregation degree, and the optimized position information.

[0141] Step S164: Based on the structural level optimization map, and setting the minimum side length of the three-dimensional grid unit to 0.5m, organize the component twin set in a spatial grid manner, thereby constructing a component grid index set;

[0142] In this embodiment, based on the structural level optimization map, organize the component twin set in a spatial grid manner, and set the minimum side length of the three-dimensional grid unit to 0.5m. Adopt the three-dimensional grid voxelization method (Voxelization) to divide the entire construction area into uniform voxel cubes, and map each component to the corresponding grid unit according to its geometric boundary. If a component spans multiple grid units, record its cross-grid relationship for subsequent index optimization. Generate a component grid index set, which contains information such as component ID, grid number, partition level, spatial voxel coordinates, and structural category, for rapid positioning and regional scheduling.

[0143] Step S165: Combine the component grid index set and the multi-level component partition tree to classify and encapsulate the grid units, generating a set of grid-based component cluster units;

[0144] In this embodiment, after constructing the grid index, combine the established multi-level component partition tree to classify and encapsulate the grid units. According to the spatial level and aggregation degree information of each grid, encapsulate multiple spatially adjacent and structurally similar grid units into grid-based component cluster units. The components within each cluster unit have a certain geometric connectivity or structural function coupling (such as floor beam-column groups, precast wall panel groups, etc.), and each cluster unit is represented by a unified ID, while retaining its internal component list and structural semantic tags. The generated set of grid-based component cluster units can be used for subsequent structural analysis and parallel simulation.

[0145] Step S166: Perform unified data structure encapsulation on the set of grid-based component cluster units, and embed a 10s dynamic update mechanism to obtain a component-level digital twin grid.

[0146] In this embodiment, a unified data structure encapsulation is performed on all component cluster unit sets. An object-oriented data encapsulation model is adopted to encapsulate each component cluster unit as a structure object, and the fields include cluster ID, component list, spatial grid boundary, structural function category, the construction stage it belongs to, and physical property statistical information, etc. At the same time, to achieve the dynamic response and real-time update of the system, a dynamic update mechanism with a period of 10 s is embedded. This mechanism triggers the automatic resampling and status refresh of component properties and spatial positions through a timer to ensure that the state of the twin grid is kept in real-time synchronization with the on-site construction state. Finally, a component-level digital twin grid is output, providing a fine spatial basis for multi-scale construction state perception and dynamic scheduling.

[0147] Optionally, the coupling of environmental factors - material property response in step S2 is specifically as follows:

[0148] Based on the component-level digital twin grid, combined with GNSS positioning data and the building structure camera image set, the actual construction component environmental exposure state is identified to obtain a component environmental exposure parameter set;

[0149] In this embodiment, first, based on the component-level digital twin grid, by extracting the three-dimensional spatial position, grid cell index, and ID information of each component, its spatial coordinate range in the actual construction site is determined. Combining the actual placement position information of the component in the GNSS positioning data, a position correction algorithm (such as least squares fitting correction) is used to register the theoretical position and the measured position, and the pose of the twin component is dynamically adjusted. The building structure camera image set is synchronously called, and the preset YOLOv8 visual detection model is used to identify and semantically segment the components in the image to determine whether the components are in a state of occlusion, exposure, semi-occlusion, etc. If the proportion of the visible pixel area of the component in the image is greater than 80%, it is judged as "fully exposed"; the proportion between 30% and 80% is "partially exposed"; less than 30% is "occluded". Combining the image recognition results and the spatial pose relationship, a component environmental exposure parameter set is generated, recording information such as the exposure state, visible ratio, and position credibility of each component.

[0150] The component environmental exposure parameter set and the environmental sensing data are spatially interpolated and temporally aligned to generate a component-level environmental factor spatio-temporal distribution matrix;

[0151] In this embodiment, the component environment exposure parameter set is fused with the environmental sensor data deployed inside the construction area. The inverse distance weighting method (IDW) is used to perform spatial interpolation estimation on the data (temperature, humidity, PM2.5, wind speed, etc.) of discrete sensor sampling points within the spatial position range of the component. The interpolation radius is set to 5 meters, and sensors beyond the radius do not participate in the weight calculation. In the time dimension, the sampling synchronization period is set to 60 seconds, and the sensing data is aligned with the component exposure timestamp to ensure double consistency in spatial position and time series. Finally, an environmental parameter record with multiple moments and multiple factors is formed for each component, and a spatio-temporal distribution matrix of component-level environmental factors is constructed, with fields including component ID, moment, temperature, humidity, wind speed, exposure status, etc.

[0152] Obtain the material property parameter library, and based on the material property parameter library, perform material type-environment factor association to obtain the component material response mapping matrix;

[0153] In this embodiment, the material property parameter library is then called, and the component materials are classified according to the material codes specified in the initial semantic labels of the components. Each type of material (such as C30 concrete, HRB400 steel bars, foam insulation boards, etc.) corresponds to a set of environmental factor sensitivity parameters, such as freeze-thaw resistance index, wet expansion coefficient, thermal expansion coefficient, etc. A correlation matrix of material type and environmental factors is constructed, with fields including material category, corresponding factors (temperature, humidity, corrosive gas, etc.), and sensitivity coefficients (between 0 and 1). Use this matrix to map the response intensity of component materials to different environmental factors, and form a component material response mapping matrix to provide a coupling basis between factors for subsequent coupling modeling.

[0154] Fuse the component material response mapping matrix and the spatio-temporal distribution matrix of component-level environmental factors, and perform tensor coupling modeling to establish a material response tensor field;

[0155] In this embodiment, during the stage of constructing the material response tensor field, the above-mentioned component material response mapping matrix and the spatio-temporal distribution matrix of component-level environmental factors are fused by tensor product. A third-order tensor structure is adopted: the dimensions are component ID, time step t, and environmental factor type respectively, and the fusion result is the response value of the component material per unit time. Through tensor slicing, the response trend of any component under multiple factors at any time can be observed, or the action difference of a certain factor on all components can be observed. This tensor structure is stored in the form of a floating-point matrix and is accompanied by response level annotations (such as slight, moderate, severe) for hierarchical management.

[0156] Perform multivariate sensitivity analysis and response weight normalization on the material response tensor field, evaluate the component performance decay risk degree, and generate a component performance risk sensitivity matrix;

[0157] In this embodiment, a sensitivity analysis is further performed on the material response tensor field. The Sobol sensitivity index is used to calculate the main effects and interaction effects of each environmental factor on the material properties of the component, and to evaluate the proportion of the non-linear influence between the factors. A risk sensitivity evaluation threshold is set to 0.7. If the response value of the component under a certain factor combination exceeds this threshold, it is marked as a high-risk item. The sensitivity results of all components at different time points are subjected to standard normalization (min-max normalization to [0,1]), and a component performance risk sensitivity matrix is output. Each row corresponds to a component, and each column corresponds to an environmental risk item. The matrix value represents the relative risk level.

[0158] The component spatial position mapping model is combined with the component performance risk sensitivity matrix for spatial mapping visualization to obtain a material environment response distribution map.

[0159] In this embodiment, the component performance risk sensitivity matrix is integrated with the component spatial position mapping model, and a three-dimensional visualization engine (such as Cesium or Unity3D) is called to construct a spatial heat distribution map. In the digital twin space, the components are rendered as model voxels according to their spatial coordinates, and different color tags are assigned based on the sensitivity values: green represents low risk (x < 0.3), yellow represents medium risk (0.3 < x < 0.6), and red represents high risk (x > 0.6). Interactive query and timeline sliding are supported to view the response change trends of components at different times. Finally, a material environment response distribution map is output, providing spatio-temporal data support for construction status evaluation and maintenance decision-making.

[0160] Optionally, constructing the response tensor field specifically includes:

[0161] Extract the material type, response model parameters, and coupling factor weight information from the component material response mapping matrix, identify the material response function families corresponding to each type of component, and perform material property mapping indexing according to the component ID in the component-level digital twin grid to generate a component response function index table;

[0162] In this embodiment, a field parsing operation is performed on the component material response mapping matrix to extract the material types used for each type of component (such as C50 high-performance concrete, 6061-T6 aluminum alloy, EPDM rubber seal, etc.), and their corresponding response model parameters, including Young's modulus E, Poisson's ratio μ, thermal conductivity λ, expansion coefficient α, etc. At the same time, the environmental coupling factor weight values are read (for example, the influence weight of temperature on concrete carbonation is 0.42, and the weight of humidity on the metal corrosion rate is 0.36). Subsequently, according to the component ID field in the component-level digital twin grid, index matching of the material property fields is performed to construct a three-tuple data structure of "component ID - material type - response function family number", which is saved as a response function index table for quick retrieval in the subsequent modeling stage.

[0163] Extract the time - series environmental variables of components from the spatio - temporal distribution matrix of component - level environmental factors, construct a three - order input tensor of component - environmental factor - time, and perform environmental variable response channel matching according to the component response function index table to generate a component environmental response input tensor;

[0164] In this embodiment, five environmental variables of temperature, humidity, wind speed, rainfall, and ultraviolet intensity of each component are extracted from the spatio - temporal distribution matrix of component - level environmental factors under a specified time window (for example, within 24 hours, sampling once every 30 minutes), and a three - order input tensor with the dimension of [number of components×number of environmental factors×number of time steps] is constructed; to ensure the adaptability of variable channels and material response functions, a channel alignment operation is performed according to the previously generated response function index table. For example, wind speed and ultraviolet channels are shielded for concrete - type components, while all channels are enabled for metal - type components, so that each type of component only activates the channels related to its material response, generating a component environmental response input tensor to ensure the logical consistency of tensor dimensions.

[0165] According to the component material coding set, use the material property parameter library to extract the rotational degrees of freedom of the contact joints, the thickness of the coated paint surface and the aging grade, and the electrochemical sensitivity coefficients of the electrode components;

[0166] In this embodiment, using the component material coding set as the main index, call the standard material property parameter library to extract the rotational degrees of freedom information (such as ±30°) of mechanical connection components, the thickness of the coated paint surface (such as 120μm) and the aging grade (divided into grades A - D) of various surface components, and extract the electrochemical sensitivity coefficients (such as a temperature coefficient of 2.1e - 5V / ℃) of all components with electrode parts.

[0167] Conduct structural behavior response analysis and adjustment factor quantization on the rotational degrees of freedom of the contact joints, the thickness of the coated paint surface and the aging grade, and the electrochemical sensitivity coefficients of the electrode components respectively to obtain the joint flexibility response adjustment factor, the paint layer shielding attenuation factor, and the electrode corrosion response factor;

[0168] In this embodiment, for these extracted sensitive parameters, use a structural simulation module (such as COMSOL or OpenSees) to perform behavior modeling analysis. For example, simulate the stiffness response of the connector under different rotational degrees of freedom, analyze the shielding efficiency of water vapor diffusion under different paint layer thicknesses, calculate the oxidation attenuation curve of the electrode at different pH values based on the corrosion rate model, and finally generate the joint flexibility response adjustment factor (such as 0.78), the paint layer shielding attenuation factor (such as 0.62), and the electrode corrosion response factor (such as 1.31) respectively.

[0169] Normalize the joint flexibility response adjustment factor, the paint layer shielding attenuation factor, and the electrode corrosion response factor, establish a structural sensitivity factor table, and expand the structural sensitivity factor table into a structural response adjustment factor matrix;

[0170] In this embodiment, the joint flexibility response adjustment factor, the paint layer shielding attenuation factor, and the electrode corrosion response factor are normalized in the range of [0, 1]. At the same time, a structure sensitivity factor table is established according to the component ID, and the fields include the adjustment factor name, the original value, the normalized value, and the unit, etc. On this basis, a structure response adjustment factor matrix with a dimension of [number of components × factor channels] is extended and constructed for the coupling calculation with the environmental response tensor.

[0171] The component environmental response input tensor and the structure response adjustment factor matrix are coupled and fused at the tensor level to generate an initial material response tensor field;

[0172] In this embodiment, the component environmental response input tensor and the structure response adjustment factor matrix are extended in channels and aligned in dimensions, and the tensor product coupling method (Tucker decomposition model) is used to construct the fused initial material response tensor field, and the dimension is increased to the fourth order: [number of components × number of environmental factors × number of time steps × number of adjustment factor channels].

[0173] The initial material response tensor field is tensor-normalized and dimension-compressed, and the compressed tensor field is reconstructed to obtain the material response tensor field.

[0174] In this embodiment, the initial material response tensor field is subjected to Z-score normalization and PCA dimension compression, and the first K characteristic dimensions (usually K = 10 - 15) with a cumulative contribution rate of more than 95% are retained, and the compressed third-order tensor is reconstructed. Finally, the component-level material response tensor field is output, providing basic data input for subsequent performance risk modeling.

[0175] Especially important is that the structural behavior response analysis and the adjustment factor quantification are carried out separately, specifically:

[0176] Based on the rotational degrees of freedom of the contact joints and combined with the recorded relative pose changes of the components in the component space position mapping model, a joint rotation-thermal expansion and contraction response function is constructed to quantify the thermal strain amplification coefficient and generate the joint flexibility response adjustment factor;

[0177] In this embodiment, the rotational degree of freedom parameters of the joint components in the component material coding set are extracted (for example, the hinge joint allows a rotation range of ±25°), and then combined with the relative pose change data of the components recorded in the component space position mapping model, the actual rotation angles of each component node under specific temperature change conditions are compared and analyzed. The linear thermal expansion formula ΔL = α·L·ΔT is used to deduce the axis length change caused by thermal expansion and contraction, and then the rotation angle change function is constructed where is the initial angle, α is the linear expansion coefficient, and L is the length of the component. The error regression fitting is performed on the actually observed angle change and the theoretical calculated value, the temperature sensitivity amplification factor is extracted as the thermal strain amplification coefficient, and the joint flexibility response adjustment factor is calculated jointly with the degree-of-freedom constraint coefficient. For example, the range [0.5, 1.5] is taken for normalization to characterize the actual response intensity of joint flexibility at high or low temperatures.

[0178] Based on the painted surface thickness and the aging grade, and combined with the time-series image features in the building structure camera image set, the fading degree and crack density are identified and extracted, and a protective layer degradation function is constructed to obtain the paint layer shielding attenuation factor;

[0179] In this embodiment, the painted surface thickness (such as 150 μm) and the aging grade (such as grade C) are extracted from the component material properties, and an initial degradation model of thickness-time-aging grade is established. Then, the time-series images in the building structure camera image set are preprocessed (including grayscale conversion, edge enhancement, and illumination normalization), and a convolutional neural network (such as ResNet50) is used for image semantic segmentation to identify the faded area and crack area on the component surface, and the paint fading degree (based on the RGB histogram distribution) and crack density (the sum of the crack pixel lengths per unit area) in each image are extracted. Based on the above extracted features, a protective layer degradation function is constructed. For example, the weighted cumulative form D(t) = w1·F(t) + w2·C(t) is adopted, where F(t) is the fading degree, C(t) is the crack density, and w1, w2 are empirical weight factors. Finally, the loss degree of the overall shielding performance of the paint layer is evaluated by comparing the output value of the degradation function with the initial value of the design thickness, and a numerical paint layer shielding attenuation factor is obtained. Its value range is generally set to [0, 1], indicating the shielding ability from completely effective to completely ineffective.

[0180] Based on the electrochemical sensitivity coefficient of the electrode component, the humidity and pH information in the environmental sensing data are fused to estimate the electrode micro-corrosion activity and generate an electrode corrosion response factor.

[0181] In this embodiment, the electrochemical sensitivity coefficient in the component material properties is called (for example, the electrochemical response coefficient of a copper alloy electrode in a pH = 4 environment is 1.8e-3 A / cm 2 / (pH), and then fuse the environmental sensing data deployed at the construction site, including relative humidity (%RH) and pH value. In each spatial grid unit where the electrode component is located, extract the environmental sequence data within 24 consecutive hours, calculate the average humidity value H_avg and the pH change rate ΔpH, and substitute the two into the micro-corrosion response estimation model: M_corr = k·S·H_avg·|ΔpH|, where k is the material constant and S is the surface area. Then, perform normalization conversion in combination with the electrode sensitivity coefficient to obtain the electrode corrosion response factor representing the corrosion sensitivity level. For example, set the sensitivity upper limit to 1.0 and perform Sigmoid normalization processing on the M_corr value. The finally output response factor is used as the basis for weight assignment in the material attenuation risk modeling.

[0182] Optionally, step S3 is specifically as follows:

[0183] Step S31: According to the spatial topology structure and component material label information in the component-level digital twin grid, identify the installation, nesting, support, and occlusion relationships between components, and construct a component-level process dependency graph;

[0184] In this embodiment, based on the set of components that have completed spatial grid division in the component-level digital twin grid, analyze the geometric contact relationships and position overlap situations between components. Combine the component material label information (such as "reinforced concrete beam", "prefabricated infill wall", etc.) to construct a logical rule library for identifying the installation, nesting, support, and occlusion relationships between components. For example, mark the pair of components with upper and lower center-of-gravity contact and a material hardness difference of less than 10 as a "support" relationship; mark the component whose outer contour encloses more than 80% of the volume of another component as a "nesting" relationship. By traversing the position adjacency relationship matrix between components and the material matching rules, identify all valid structural relationships, and construct a preliminary component-level process dependency graph with component IDs as nodes and structural logical relationships as edges, and perform four-category labeling on the edges (installation, nesting, support, occlusion) to support the dependency judgment of subsequent process reasoning.

[0185] Step S32: Based on the CNSS positioning data and the component spatial position mapping model, perform spatial position projection and adjacency analysis on the component-level process dependency graph to generate a multi-dimensional construction process dependency graph;

[0186] In this embodiment, the real-time position information of the construction nodes recorded in the GNSS positioning data is called, and combined with the global coordinates of the components defined in the component spatial position mapping model, coordinate system unification and attitude registration processing are carried out to ensure the alignment of the positioning data at the construction site and the design model data. On this basis, spatial position projection is performed on the two end components of each edge in the component-level process dependency graph, the Euclidean distances in the X-Y plane and the Z-axis direction are calculated, and potential spatial dependencies are identified using a set adjacent threshold (such as 2.0 m). A dimension nesting parameter is introduced to remap the spatial structure of the original dependency graph, adding height dependency weights (Z-axis), horizontal proximity weights (X-Y), and operation path intersection discrimination parameters to construct a multi-dimensional construction process dependency graph in three-dimensional space, with each edge attached with spatial coordinate differences and operation precedence.

[0187] Step S33: Call the material environment response distribution map, and embed the component environment response intensity value in the material environment response distribution map as an environment-sensitive parameter into each component node of the multi-dimensional construction process dependency graph to form a node environment response injection map;

[0188] In this embodiment, the constructed material environment response distribution map is called, and the maximum environment response intensity value R_max, the minimum response value R_min, and the average value R_avg of each component during a given construction period are extracted as environment-sensitive parameters. Then, each component node in the multi-dimensional construction process dependency graph is traversed, and the corresponding triple environment parameters are extracted according to the component ID mapping and written into the node attribute dictionary in an embedded manner to construct a node environment response injection map. During the injection process, a response value normalization parameter α = 0.01 is set to perform normalization mapping on values such as R_max to ensure that the environment response values of all component nodes are unified within the range of [0, 1], for subsequent gradient difference calculation and sensitivity weight fitting.

[0189] Step S34: Based on the node environment response injection map, extract the environmental response gradient changes of the front and rear component nodes connected by each process edge, and calculate the environmental adaptation weight of the process dependency path in the component-level process dependency graph to generate a process edge response sensitivity matrix;

[0190] In this embodiment, based on the node environment response injection map, the normalized environmental response values R_A and R_B of the pre-component node A and the post-component node B connected by each process edge are extracted, the environmental response gradient ΔR = |R_B - R_A| is calculated, and the environmental adaptation weight is constructed according to the empirical formula W_env = e^(γ·ΔR), where γ is the response gradient exponential amplification factor (recommended value γ = 1.5 - 2.0). The response weights of all edges are filled into the edge weight matrix of the component-level process dependency graph in sequence to form a process edge response sensitivity matrix, which is combined with the original structure dependency weight to form a double-weighted structure for characterizing the environmental sensitivity of process execution.

[0191] Step S35: Integrate the node environment response injection graph and the process edge response sensitivity matrix, construct a weighted directed graph, and perform dynamic feasible region correction on nodes and paths to obtain the construction process dynamic atlas.

[0192] In this embodiment, the node environment response injection graph and the process edge response sensitivity matrix are integrated to construct an environmental dynamic weighted directed graph. Each node in the graph carries component position information and environmental response values, and each edge has both structural dependence and environmental weight attributes. Subsequently, a node and path dynamic feasible region correction mechanism is introduced. Based on the setting of the environmental response threshold (for example, when the node response value > 0.85, it is determined as a highly sensitive node), the priority of the corresponding path is lowered or temporarily blocked; and the optimal process path set is extracted through a graph traversal algorithm (such as heuristic A* search), and finally a construction process dynamic atlas that can adapt to environmental changes in real time is formed.

[0193] Optionally, step S4 is specifically as follows:

[0194] Step S41: Construct an input feature vector based on the node environment response characteristics and path weight parameters in the construction process dynamic atlas, and generate a construction process state input tensor set;

[0195] In this embodiment, the environmental response characteristics of each node in the construction process dynamic atlas (such as node environmental response value R i , response gradient ΔR ij ) and the weight parameters of each path (such as path dependence weight W dep , environmental adaptation coefficient W env , structural constraint weight W str ) are extracted to form a feature vector F ij = [R i , R j , ΔR ij , W dep , W env , W str . For each path unit, its corresponding source node, target node, and feature vector are combined to form a state input sample. Traverse all paths in the entire dynamic atlas to construct a third-order tensor where n is the number of paths, m is the time step, and d is the single-path feature dimension (set to 6). At the same time, construction stage labels (such as "hoisting", "welding", "concrete pouring", etc.) are added to the node attributes as additional encodings.

[0196] Step S42: Perform dynamic time expansion and multi-channel nested encoding on the construction process state input tensor set, and establish a process state evolution and propagation mechanism to obtain a construction process state propagation model;

[0197] In this embodiment, the input tensor set T stateDynamic time unfolding processing is performed using a sliding window mechanism (the time window is set to τ = 5) to construct a time series tensor to reflect the local state change sequence. At the same time, a multi-channel nested coding method is introduced to map different process types to independent channels respectively, and a channel expansion tensor is constructed where c is the number of process categories. Subsequently, a gated graph attention propagation network (Gated-GAT) is introduced. By jointly modeling the node state propagation path through a time gate unit and a structural adjacency matrix propagation unit, the contagious state transition between paths is considered during the propagation process, and finally, the construction process state propagation model M propagation :T multi →S pred is output, where represents the predicted state sequence

[0198] Step S43: Simulate and deduce the construction process dynamic graph through the construction process state propagation model to obtain a state path simulation sequence set

[0199] In this embodiment, the construction process state propagation model M propagation is used to perform forward inference step by step on the input tensor T multi to obtain a simulation output sequence of the evolution of the node state path over time. The simulation sequence set represents the state change trajectories of all component nodes within the prediction period [t1, t k . Among them, each represents the state vector of component i at time point t

[0200] Step S44: Perform spatio-temporal domain fusion analysis on the state path simulation sequence set, extract the active intensity, state transition frequency, and resource consumption change rate of process nodes in different time periods to obtain a construction state aggregation index graph

[0201] In this embodiment, sliding average and local gradient enhancement are performed on the time dimension to extract the state active intensity of each node in each time period state transition frequency (statistical by the number of state code changes) and resource consumption change rate (calculated by the change of resource index components). After all the indicators are summarized, an indicator matrix is formed, where each dimension represents the component number, time period number, and three types of indicator values respectively. Combined with the process classification coding and path labels, it can be further extended to a four-order tensor of component-time-process-indicator for constructing a construction state aggregation index graph

[0202] Step S45: Perform spatial fusion on the construction state aggregation index map and the component space mapping model to construct a visual spatial field for the evolution of the construction state at each time period, and output the construction state evolution distribution field.

[0203] In this embodiment, the construction state aggregation index map M ind is mapped to the corresponding three-dimensional coordinate system in the component space position model, and the spatial coordinate point cloud P i =(x i , y i , z i ) of each component is obtained through component ID indexing. The active intensity state frequency and resource change rate within the time period are encoded in the form of color channels, and a three-dimensional heat distribution map of the construction state evolution is constructed through voxel reconstruction. The output construction state evolution distribution field V evo (x, y, z, t) is represented by continuous four-dimensional voxel grid data, supporting multi-time period slice analysis and interactive visualization display.

[0204] Optionally, the state anomaly multi-hop causal backtracking in step S5 is specifically as follows:

[0205] Based on the construction state evolution distribution field, perform component state mutation detection. Set the sliding window to 5 minutes and the mutation detection threshold to 3σ to identify the abnormal state point set, and generate the state anomaly annotation set;

[0206] In this embodiment, the sliding window is set to 5 minutes, and the mutation detection threshold is set to 3σ. This means that within every 5-minute time window, the states of each component on the construction site will be monitored. If the state deviates from the normal value by more than 3 standard deviations (σ), it is determined as an abnormal state. The setting of this threshold can effectively filter out normal fluctuations and only focus on mutation events with high deviations, which helps to identify potential engineering problems in real time, such as equipment failures or construction quality problems. Through the generation of these abnormal state point sets, the abnormal changes in the entire construction process can be accurately marked.

[0207] Call the construction process dynamic map and the state anomaly annotation set, set the upper limit of the backtracking jump number to 4, and set the minimum threshold of the path reliability to 0.65 to perform multi-hop dependent path backtracking of the abnormal state points. In each hop propagation, calculate the abnormal conduction weight and generate a multi-hop abnormal backtracking path map;

[0208] In this embodiment, the dynamic map of construction processes is invoked and multi-hop dependency path backtracking is performed using the set of state anomaly annotations. The upper limit of the backtracking hop count is set to 4, and the minimum threshold of path confidence is set to 0.65. This means that when backtracking the abnormal state points, only up to four steps of backtracking are allowed, and the backtracking confidence of each path needs to be greater than 0.65 to be considered a valid abnormal propagation path. During the backtracking process, the abnormal conduction weight is calculated for each hop of propagation, and the magnitude of the conduction weight determines the contribution degree of the path to the final abnormal state.

[0209] Based on the multi-hop abnormal backtracking path graph, extract the starting state, path trigger timestamp, environmental response value, material sensitivity factor, and structural logic role of each component node on the backtracking path, and generate a component-level causal propagation vector set;

[0210] In this embodiment, based on the multi-hop abnormal backtracking path graph, extract the starting state, path trigger timestamp, environmental response value, material sensitivity factor, and structural logic role of each component node on the backtracking path, and generate a component-level causal propagation vector set. This step can record in detail the specific performance of each component during the abnormal conduction process, including information in multiple dimensions such as time, environmental response, and material sensitivity. These causal propagation vectors will be used to infer the root cause of the anomaly and provide data support for subsequent Bayesian probability graph modeling.

[0211] Perform Bayesian probability graph modeling on the component-level causal propagation vector set, infer and identify the dominant abnormal source factors, evaluate the abnormal attribution weights of each path, and output an abnormal causal attribution graph;

[0212] In this embodiment, modeling is performed based on the component-level causal propagation vector set to infer and identify the dominant abnormal source factors. The Bayesian model uses the dependency relationships and abnormal propagation paths between each component and process to calculate the abnormal attribution weight of each path. The weight of each path represents the contribution degree of the path to the anomaly. Through Bayesian inference, the root cause of the anomaly can be more accurately located, and it can be identified which components, processes, or external factors are most likely to cause the current abnormal state. The output abnormal causal attribution graph will present the roles and impacts of each component and process in the occurrence of the anomaly.

[0213] Fuse the abnormal causal attribution graph and the dynamic map of construction processes, set the influence coefficient injection ratio to 0.2 to inject influence coefficients into the process dependency paths, and label the risk sensitivity levels of each component node to generate an abnormal causal conduction chain graph.

[0214] In this embodiment, the abnormal causal attribution map is integrated with the construction process dynamic map. The injection ratio of the influence coefficient is set to 0.2, which means that on the process dependency path, the influence degree of the abnormal source factor is injected into 20% of the process path to reflect the influence degree of abnormal propagation on the entire construction process. In addition, the risk sensitivity levels of each component node are marked, and an abnormal causal conduction chain diagram is generated according to the weight of abnormal causal propagation and the risk exposure of each component.

[0215] Optionally, this specification also provides a digital-twin-based construction control system for power engineering, which is used to execute the digital-twin-based construction control method for power engineering described above. The digital-twin-based construction control system for power engineering includes:

[0216] A component digital twin module, which is used to obtain heterogeneous sensing data of the construction area and perform dynamic modeling of structural components based on the heterogeneous sensing data of the construction area to obtain a component-level digital twin grid;

[0217] An environmental response analysis module, which is used to analyze the state of building components and the environmental state of the construction area based on the component-level digital twin grid, and perform environmental factor-material property response coupling on the state of building components and the environmental state of the construction area to obtain a material environment response distribution map;

[0218] An environmental sensitivity injection module, which is used to analyze the process dependency relationship of the engineering construction process based on the component-level digital twin grid to obtain a multi-dimensional construction process dependency diagram; and perform dynamic injection of component environmental sensitivity on the multi-dimensional construction process dependency diagram by using the material environment response distribution map to obtain a construction process dynamic map;

[0219] A construction state field deduction module, which is used to perform multi-modal construction state field deduction based on the construction process dynamic map to obtain a construction state evolution distribution field;

[0220] A process scheduling optimization module, which is used to perform state abnormal multi-hop causal backtracking based on the construction state evolution distribution diagram to obtain an abnormal causal conduction chain diagram, and perform reversible process scheduling optimization and simulation based on the abnormal causal conduction chain diagram to obtain a reversible scheduling optimization diagram.

[0221] The above are only specific embodiments of the present invention, which enable those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A digital-twin-based control method for power engineering construction, characterized in that, Including the following steps: Step S1: Obtain heterogeneous sensing data of the construction area, and perform dynamic modeling of structural components based on the heterogeneous sensing data of the construction area to obtain a component-level digital twin grid; Step S2: Analyze the states of building components and the environmental state of the construction area based on the component-level digital twin grid, and perform environmental factor-material property response coupling on the states of building components and the environmental state of the construction area to obtain a material-environment response distribution map; Step S3: Analyze the dependencies of engineering construction processes based on the component-level digital twin grid to obtain a multi-dimensional construction process dependency graph; use the material-environment response distribution map to dynamically inject component environmental sensitivity into the multi-dimensional construction process dependency graph to obtain a construction process dynamic atlas; Step S4: Deduce the multi-modal construction state field based on the construction process dynamic atlas to obtain a construction state evolution distribution field; Step S5: Conduct state anomaly multi-hop causal backtracking based on the construction state evolution distribution map to obtain an abnormal causal conduction chain graph, and perform reversible process scheduling optimization and simulation based on the abnormal causal conduction chain graph to obtain a reversible scheduling optimization atlas.

2. The method for controlling and managing the construction of a power project based on digital twins according to claim 1, wherein Specifically, step S1 is as follows: Step S11: Obtain heterogeneous sensing data of the construction area, and perform data preprocessing to obtain standardized heterogeneous sensing data, including lidar scan data, building structure camera image sets, component material coding sets, GNSS positioning data, and environmental sensing data; Step S12: Perform semantic segmentation and component recognition on the standardized heterogeneous sensing data, and extract semantic features of component boundaries, structural categories, and material properties to obtain an initial component semantic label set; Step S13: Perform three-dimensional contour modeling of components based on the initial component semantic label set and lidar scan data, and perform topological relationship reasoning on the three-dimensional contours of components to generate a three-dimensional component topological structure set; Step S14: Integrate the three-dimensional component topological structure set and GNSS positioning data to establish a spatial mapping relationship, and perform coordinate registration and error correction to generate a component spatial position mapping model; Step S15: Combine the initial component semantic label set, the three-dimensional component topological structure set, and the component spatial position mapping model to construct a component-level digital twin unit, and generate a set of component twins; Step S16: Organize the set of component twins into a grid, establish a multi-level digital structure mapping relationship, and output a component-level digital twin grid.

3. The method for controlling and managing the construction of a power project based on digital twins according to claim 2, wherein, Specifically, step S15 is as follows: Step S151: Extract the boundary coordinates, structural categories, and material property parameters of each component in the initial component semantic label set, and perform component ID-level matching with the three-dimensional component topological structure set to generate a semantic-structure association index table; Step S152: Bind the three-dimensional geometry-material semantic attributes of components according to the semantic-structure association index table, establish a component geometric-semantic fusion unit, and generate a component semantic-structure fusion body; Step S153: Perform global coordinate pose mapping conversion on the component coordinate data in the component spatial position mapping model and the position anchor points in the component semantic-structure fusion body to obtain a component instantiation unit; Step S154: Perform attribute structuring processing on the component instantiation unit to generate a component attribute data set; Step S155: Establish a multi-instance component twin mapping rule based on the constructed attribute dataset, and use the multi-instance component twin mapping rule to perform structure instance merging and index rearrangement to obtain a set of component twins.

4. The method for controlling and managing the construction of a power project based on digital twins according to claim 2, wherein, Step S16 specifically is: Step S161: Set the minimum partition granularity to 1.0 m 3 to perform spatial pose partitioning and classification on the component twin set to obtain a component spatial partition table; Step S162: Construct a hierarchical relationship of spatial partitions based on the component spatial partition table, and set the maximum hierarchical depth to 5 for recursive parsing of the structure hierarchy to obtain a multi-level component partition tree; Step S163: Perform hierarchical reorganization and position optimization on the multi-level component partition tree, calculate the spatial aggregation degree, and perform aggregation adjustment on the structure nodes of the multi-level component partition tree according to the spatial aggregation degree to generate an optimized structure hierarchy map; Step S164: Based on the optimized structure hierarchy map, and set the side length of the minimum unit of the three-dimensional grid to 0.5m to organize the set of component twins in space, thereby constructing a component grid index set; Step S165: Combine the component grid index set and the multi-level component partition tree for grid cell classification and encapsulation to generate a set of grid component cluster cells; Step S166: Perform unified data structure encapsulation on the set of grid component cluster cells and embed a 10s dynamic update mechanism to obtain a component-level digital twin grid.

5. The method for controlling and managing the construction of a power project based on digital twins according to claim 1, characterized in that, The environmental factor-material property response coupling in Step S2 specifically is: Based on the component-level digital twin grid, and combined with GNSS positioning data and the building structure camera image set to identify the actual construction component environmental exposure status, and obtain a set of component environmental exposure parameters; Perform spatial interpolation and time alignment on the set of component environmental exposure parameters and the environmental sensing data to generate a component-level environmental factor spatio-temporal distribution matrix; Obtain the material property parameter library, and perform material type-environmental factor association based on the material property parameter library to obtain a component material response mapping matrix; Fuse the component material response mapping matrix and the component-level environmental factor spatio-temporal distribution matrix, and perform tensor coupling modeling to establish a material response tensor field; Perform multi-variable sensitivity analysis and response weight normalization on the material response tensor field, evaluate the component performance attenuation risk degree, and generate a component performance risk sensitivity matrix; Combine the component spatial position mapping model and the component performance risk sensitivity matrix for spatial mapping visualization to obtain a material environment response distribution map.

6. The method for controlling and managing the construction of a power project based on digital twin according to claim 5, wherein, Constructing the response tensor field specifically is: Extract the material types, response model parameters, and coupling factor weight information in the component material response mapping matrix, identify the material response function families corresponding to various components, and perform material attribute mapping indexing according to the component ID in the component-level digital twin grid to generate a component response function index table; Extract the time series environmental variables of the components from the component-level environmental factor spatio-temporal distribution matrix, construct a component-environmental factor-time three-order input tensor, and perform environmental variable response channel matching according to the component response function index table to generate a component environmental response input tensor; According to the component material coding set, use the material property parameter library to extract the rotational degrees of freedom of the contact joints, the coating paint thickness and aging grade, and the electrochemical sensitivity coefficient of the electrode components; The rotational degrees of freedom of the contact joint, the thickness of the painted surface layer, the aging level, and the electrochemical sensitivity coefficient of the electrode component are respectively analyzed for structural behavior response and the adjustment factors are quantified to obtain the joint flexibility response adjustment factor, the paint layer shielding attenuation factor, and the electrode corrosion response factor; The joint flexibility response adjustment factor, the paint layer shielding attenuation factor, and the electrode corrosion response factor are normalized, a structural sensitivity factor table is established, and the structural sensitivity factor table is extended into a structural response adjustment factor matrix; The component environmental response input tensor and the structural response adjustment factor matrix are coupled and fused at the tensor level to generate an initial material response tensor field; The initial material response tensor field is normalized and dimensionally compressed, and the compressed tensor field is reconstructed to obtain the material response tensor field.

7. The method for controlling and managing the construction of a power project based on digital twins according to claim 1, characterized in that Step S3 is specifically as follows: Step S31: According to the spatial topology structure and component material label information in the component-level digital twin grid, identify the installation, nesting, support, and occlusion relationships between components, and construct a component-level process dependency graph; Step S32: Based on the CNSS positioning data and the component spatial position mapping model, perform spatial position projection and adjacency analysis on the component-level process dependency graph to generate a multi-dimensional construction process dependency graph; Step S33: Call the material environmental response distribution map, and embed the component environmental response intensity value in the material environmental response distribution map as an environmental sensitivity parameter into each component node of the multi-dimensional construction process dependency graph to form a node environmental response injection graph; Step S34: Based on the node environmental response injection graph, extract the environmental response gradient changes of the front and rear component nodes connected by each process edge, and calculate the environmental adaptation weight of the process dependency path in the component-level process dependency graph to generate a process edge response sensitivity matrix; Step S35: Integrate the node environmental response injection graph and the process edge response sensitivity matrix, construct a weighted directed graph, and perform node and path dynamic feasible region correction to obtain a construction process dynamic graph.

8. The method for controlling and managing the construction of a power project based on digital twins according to claim 1, wherein Step S4 is specifically as follows: Step S41: Based on the node environmental response characteristics and path weight parameters in the construction process dynamic graph, construct an input feature vector to generate a construction process state input tensor set; Step S42: Perform dynamic time expansion and multi-channel nested coding on the construction process state input tensor set to establish a process state evolution propagation mechanism, thereby obtaining a construction process state propagation model; Step S43: Simulate and deduce the construction process dynamic graph through the construction process state propagation model to obtain a state path simulation sequence set; Step S44: Perform spatio-temporal domain fusion analysis on the state path simulation sequence set, extract the active intensity, state transition frequency, and resource consumption change rate of the process nodes at different time periods to obtain a construction state aggregation index graph; Step S45: Fuse the construction state aggregation index graph and the component spatial mapping model in space to construct a visual spatial field of the evolution of the construction state at each time period, and output a construction state evolution distribution field.

9. The method for controlling and managing the construction of a power project based on digital twins according to claim 1, wherein The state anomaly multi-hop causal backtracking in Step S5 is specifically as follows: Based on the construction state evolution distribution field, component state mutation detection is carried out. A sliding window of 5 minutes and a mutation detection threshold of 3σ are set to identify abnormal state point sets, and a state anomaly annotation set is generated; The construction process dynamic atlas and the state anomaly annotation set are called. The upper limit of the backtracking jump count is set to 4, and the minimum threshold of the path reliability is set to 0.65 to perform multi-hop dependent path backtracking of abnormal state points. In each hop propagation, the abnormal conduction weight is calculated to generate a multi-hop abnormal backtracking path graph; Based on the multi-hop abnormal backtracking path graph, the starting state, path trigger timestamp, environmental response value, material sensitivity factor, and structural logic role of each component node on the backtracking path are extracted to generate a component-level causal propagation vector set; Bayesian probability graph modeling is performed on the component-level causal propagation vector set to infer and identify the dominant abnormal source factors, and the abnormal attribution weights of each path are evaluated to output an abnormal causal attribution graph; The abnormal causal attribution graph and the construction process dynamic atlas are fused. The influence coefficient injection ratio is set to 0.2 to inject the influence coefficient into the process dependent path, and the risk sensitivity level of each component node is marked to generate an abnormal causal conduction chain graph.

10. A power engineering construction management and control system based on digital twin, characterized in that, For implementing the digital twin-based construction control method for power engineering as described in claim 1, the digital twin-based construction control system for power engineering includes: A component digital twin module for obtaining heterogeneous sensing data in the construction area and performing dynamic modeling of structural components based on the heterogeneous sensing data in the construction area to obtain a component-level digital twin grid; An environmental response analysis module for analyzing the state of building components and the environmental state of the construction area based on the component-level digital twin grid, and performing environmental factor-material property response coupling on the state of building components and the environmental state of the construction area to obtain a material environment response distribution map; An environmental sensitivity injection module for analyzing the dependency relationship of the engineering construction process based on the component-level digital twin grid to obtain a multi-dimensional construction process dependency graph; using the material environment response distribution map to perform dynamic injection of component environmental sensitivity into the multi-dimensional construction process dependency graph to obtain a construction process dynamic atlas; A construction state field deduction module for performing multi-modal construction state field deduction based on the construction process dynamic atlas to obtain a construction state evolution distribution field; A process scheduling optimization module for performing multi-hop causal backtracking of state anomalies based on the construction state evolution distribution map to obtain an abnormal causal conduction chain graph, and performing reversible process scheduling optimization and simulation based on the abnormal causal conduction chain graph to obtain a reversible scheduling optimization graph.

Images