A power infrastructure construction process warning method and system based on a high-precision three-dimensional model

Abstract

The present invention discloses a power infrastructure construction process warning method and system based on a high-precision three-dimensional model, which relates to the technical field of power infrastructure construction. It includes collecting multi-source data at the power infrastructure construction site and constructing a high-precision three-dimensional model containing equipment spatial coordinates, construction time sequence tags, and environmental parameters; inputting the spatio-temporal graph into the spatio-temporal graph convolutional network to output a risk propagation probability matrix in the three-dimensional space and the construction time sequence dimension, and identifying high-risk nodes and associated construction stages; dynamically marking the root cause coordinates and the risk diffusion range in the high-precision three-dimensional model, generating a warning instruction containing a visualization path, a root cause report, and a disposal priority, and pushing it to the terminal in real time through the edge node. Based on the hierarchical spatio-temporal attention-physical constraint graph convolutional network, the present invention dynamically allocates the cross-stage association weights between nodes, combines the physical connection strength to optimize the convolutional kernel parameters, and significantly improves the spatio-temporal correlation of high-risk node identification.

Description

Technical Field

[0001] The present invention relates to the technical field of power infrastructure construction, and in particular, to a power infrastructure construction process warning method and system based on a high-precision three-dimensional model. Background Art

[0002] In recent years, the risk warning technology for power infrastructure projects has gradually developed in the direction of the integration of three-dimensional modeling and spatio-temporal data analysis. In the prior art, a monitoring system based on BIM (Building Information Modeling) has achieved three-dimensional visualization of the construction scene, but mostly focuses on the monitoring of static equipment status, and has insufficient modeling of the dynamic correlation of construction time sequences; at the same time, the application of spatio-temporal graph neural networks (ST-GNN) in fields such as traffic flow prediction has verified its spatio-temporal correlation modeling ability, but it has not been deeply coupled with construction physical constraints (such as equipment connection strength, process connection logic). In addition, Bayesian networks are used for risk causal reasoning, but traditional methods rely on expert experience to construct topologies and are difficult to adapt to the dynamic evolution characteristics of high-dimensional spatio-temporal data.

[0003] The prior art mainly has the following deficiencies: First, traditional high-precision three-dimensional models do not dynamically bind the spatial coordinates of equipment with construction time sequence tags and environmental parameters, resulting in limited modeling accuracy of risk propagation paths. For example, existing methods judge equipment anomalies through fixed thresholds, but ignore the influence of construction stage dependence and physical connection strength on risk conduction, and are prone to false alarms or missed alarms. Second, the risk prediction model based on supervised learning lacks a counterfactual reasoning mechanism and cannot effectively distinguish the contribution of equipment anomalies and normal state fluctuations to the overall risk, resulting in root cause positioning deviation. For example, existing Bayesian networks are difficult to decouple the causal chain in multi-process coupling scenarios because they do not integrate spatio-temporal graph convolution features, affecting the pertinence of disposal strategies. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a power infrastructure construction process warning method based on a high-precision three-dimensional model to solve the problems of inaccurate modeling of dynamic risk propagation, large deviation in root cause equipment positioning, and lag in warning instruction generation during the power infrastructure construction process.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a power infrastructure construction process warning method based on a high-precision three-dimensional model, which includes collecting multi-source data at the power infrastructure construction site and constructing a high-precision three-dimensional model including equipment spatial coordinates, construction time sequence tags, and environmental parameters;

[0008] Map it to a spatio-temporal graph according to the grid topology. The node attributes encode the construction timestamps and the sequential changes in the equipment status, and the edge weights are associated with the construction connection relationship and the physical connection strength between adjacent nodes;

[0009] Input the spatio-temporal graph into the spatio-temporal graph convolutional network, and output the risk propagation probability matrix in the three-dimensional space and the construction time sequence dimension to identify high-risk nodes and associated construction stages;

[0010] Construct a causal Bayesian network, calculate the difference in risk evolution after eliminating equipment anomalies through counterfactual simulation, and locate the root cause equipment and processes;

[0011] Dynamically label the root cause coordinates and the risk diffusion range in the high-precision three-dimensional model, generate a warning instruction including a visualization path, a root cause report, and a disposal priority, and push it to the terminal in real time through the edge node.

[0012] As a preferred solution of the power infrastructure construction process warning method based on the high-precision three-dimensional model described in the present invention, wherein: the multi-source data includes equipment space coordinate data, construction time sequence label data, environmental parameter data, equipment status time sequence data, and construction connection relationship data;

[0013] The construction of the high-precision three-dimensional model including equipment space coordinates, construction time sequence labels, and environmental parameters is as follows.

[0014] Perform spatio-temporal alignment and cleaning on the collected equipment space coordinate data, construction time sequence label data, and environmental parameter data, and output a standardized spatio-temporal data set;

[0015] Extract the equipment space coordinates from the standardized spatio-temporal data set, combine with the point cloud data collected by LiDAR, perform denoising and registration, and generate the cleaned point cloud data;

[0016] Convert the cleaned point cloud data into a three-dimensional grid model, optimize the number of patches, and map the equipment space coordinates to generate a high-precision basic model;

[0017] Bind the construction time sequence label and the environmental parameter to the equipment space node to form a high-precision three-dimensional model with time sequence and environmental attributes.

[0018] As a preferred solution of the power infrastructure construction process warning method based on the high-precision three-dimensional model described in the present invention, wherein: the mapping to the spatio-temporal graph according to the grid topology, the node attributes encode the construction timestamps and the sequential changes in the equipment status, and the edge weights are associated with the construction connection relationship and the physical connection strength between adjacent nodes is as follows.

[0019] Divide the high-precision three-dimensional model into grid cells according to a fixed resolution, and each grid cell serves as a node of the spatio-temporal graph;

[0020] Extract the construction timestamps and the time series data of equipment status from the standardized spatio-temporal dataset, bind them to the corresponding grid nodes, and form node attributes.

[0021] According to the data on the connection relationship between the equipment spatial coordinates and the construction, determine the connection relationships of adjacent grid nodes, and calculate the physical connection strength as the edge weight.

[0022] Integrate the node attributes and the edge weights to generate a spatio-temporal graph, and use a graph optimization algorithm to optimize the graph structure.

[0023] As a preferred solution of the power infrastructure construction process warning method based on a high-precision three-dimensional model according to the present invention, wherein: input the spatio-temporal graph into a spatio-temporal graph convolutional network, output a risk propagation probability matrix in the three-dimensional space and the construction time series dimension, and identify high-risk nodes and associated construction stages. The specific steps are as follows.

[0024] Based on the node attributes and edge weights of the spatio-temporal graph, construct a hierarchical spatio-temporal attention-physical constraint graph convolutional network, and integrate a causal spatio-temporal attention unit, a physical constraint convolutional kernel, and a counterfactual reasoning layer as the network architecture.

[0025] Based on the causal spatio-temporal attention unit in the network architecture, divide the temporal dependence relationship by combining the construction time series labels, and use the node timestamp difference and the physical connection strength to adjust the spatial attention distribution.

[0026] Dynamically allocate cross-stage node association weights through the Softmax function, strengthen the spatio-temporal correlation of high-risk nodes within the same process, and output the dynamically calculated attention weights.

[0027] Concatenate the physical parameters in the edge weights with the equipment status of the nodes, input them into a multi-layer perceptron to generate a physical perception convolutional kernel, and dynamically adjust the kernel weights according to the environmental parameters.

[0028] Fuse the attention weights and the physical constraint convolutional kernel, perform multi-head graph convolution on the node features, respectively capture the local physical connection, the global process dependence, the short-term state fluctuation, and the long-term environmental trend, and output spatio-temporal hidden features.

[0029] Perform counterfactual intervention on the equipment status of each node, recalculate the graph convolution output, and generate a risk difference matrix by comparing the Euclidean distance between the actual and counterfactual features.

[0030] Weightedly fuse the spatio-temporal hidden features and the risk difference matrix, map them through the Sigmoid function to a risk propagation probability matrix in the three-dimensional space-time dimension, and label the risk values of each grid node in each construction stage.

[0031] Calculate the dynamic threshold of risk probability according to the construction stage, set the warning threshold, screen the nodes exceeding the warning threshold as high-risk targets, trace back the construction stage corresponding to the risk peak, and output the high-risk nodes and associated construction stages.

[0032] As a preferred solution of the power infrastructure construction process warning method based on a high-precision three-dimensional model according to the present invention, wherein: performing counterfactual intervention on the state of each node device, recalculating the graph convolution output, and generating a risk difference matrix by comparing the Euclidean distances of the actual and counterfactual feature vectors. The specific steps are as follows.

[0033] Based on the spatio-temporal hidden feature matrix, extract the actual feature vector of each node as the benchmark for counterfactual comparison.

[0034] Perform counterfactual intervention on the state of each node device, and generate counterfactual state data by setting the device state as the reference value.

[0035] Based on the counterfactual state data, re-perform the graph convolution operation and output the counterfactual feature vector.

[0036] Compare the actual feature vector and the counterfactual feature vector, calculate the Euclidean distance, and generate the node-level risk difference value.

[0037] Map the risk difference values of all nodes according to the grid topology, generate a risk difference matrix, and perform normalization processing to output the standardized risk difference matrix.

[0038] As a preferred solution of the power infrastructure construction process warning method based on a high-precision three-dimensional model according to the present invention, wherein: constructing a causal Bayesian network, calculating the risk evolution difference after eliminating equipment anomalies through counterfactual simulation, and locating the root cause equipment and processes. The specific steps are as follows.

[0039] Based on the high-risk nodes and associated construction stages, construct a Bayesian network of the causal dependence relationship between equipment states and construction processes.

[0040] Generate a counterfactual intervention scenario through the Bayesian network, simulate the state change after eliminating equipment anomalies, and output the counterfactual state data.

[0041] Input the spatio-temporal graph convolution network and calculate the risk propagation probability matrix in the counterfactual scenario.

[0042] Compare the risk propagation probability matrices of the actual scenario and the counterfactual scenario, and generate a risk evolution difference matrix.

[0043] According to the risk evolution difference matrix, identify the root cause equipment with the highest contribution to the overall risk.

[0044] Trace back the influence path of the abnormal state of the root cause equipment on adjacent nodes and construction processes, and locate the key processes.

[0045] As a preferred solution of the power infrastructure construction process warning method based on a high-precision three-dimensional model according to the present invention, wherein: dynamically mark the root cause coordinates and the risk diffusion range in the high-precision three-dimensional model, generate a warning instruction including a visualization path, a root cause report and a disposal priority, and push it to the terminal in real time through an edge node. The specific steps are as follows:

[0046] Based on the root cause device located, extract its spatial coordinate data, and dynamically mark the root cause coordinates in the high-precision three-dimensional model;

[0047] Taking the root cause coordinates as the starting point, according to the risk evolution difference matrix, calculate the risk diffusion range of the root cause device, and visually mark the diffusion boundary in the high-precision three-dimensional model;

[0048] Combined with the risk propagation probability matrix, extract the risk propagation path of high-risk nodes, and draw a visualization path in the high-precision three-dimensional model;

[0049] Integrate the root cause device and related process information to generate a detailed root cause report including device status, abnormal reasons, influence scope and related processes;

[0050] According to the risk value and diffusion range of the nodes in the risk evolution difference matrix, calculate the disposal priority of each root cause device;

[0051] Integrate the visualization path, the root cause report and the disposal priority to generate a structured warning instruction;

[0052] Push the warning instruction to the terminal device in real time through the edge node and take corresponding measures.

[0053] In a second aspect, the present invention provides a power infrastructure construction process warning system based on a high-precision three-dimensional model, including a data acquisition module, a graph structure mapping module, a risk identification module, a root cause location module and a warning push module;

[0054] The data acquisition module is used to collect multi-source data of the power infrastructure construction site and construct a high-precision three-dimensional model including device spatial coordinates, construction time sequence tags and environmental parameters;

[0055] The graph structure mapping module is used to map it into a spatio-temporal graph according to the grid topology, encode the construction time stamp and the time sequence change of the device state for the node attributes, and associate the edge weights with the construction connection relationship and the physical connection strength of adjacent nodes;

[0056] The risk identification module is used to input the spatio-temporal graph into a spatio-temporal graph convolutional network, output a risk propagation probability matrix in the three-dimensional space and the construction time sequence dimension, and identify high-risk nodes and related construction stages;

[0057] The root cause location module is used to construct a causal Bayesian network, calculate the risk evolution difference after eliminating equipment anomalies through counterfactual simulation, and locate the root cause equipment and processes.

[0058] The early warning push module is used to dynamically mark the root cause coordinates and the risk diffusion range in the high-precision three-dimensional model, generate an early warning instruction including a visualization path, a root cause report, and a disposal priority, and push it to the terminal in real time through the edge node.

[0059] In a third aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and: when the computer program is executed by the processor, any step of the power infrastructure construction process early warning method based on a high-precision three-dimensional model as described in the first aspect of the present invention is implemented.

[0060] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and: when the computer program is executed by the processor, any step of the power infrastructure construction process early warning method based on a high-precision three-dimensional model as described in the first aspect of the present invention is implemented.

[0061] The beneficial effects of the present invention are as follows: By constructing a high-precision three-dimensional model that integrates equipment spatial coordinates, construction time sequence tags, and environmental parameters, and mapping it into a spatio-temporal graph structure, multi-dimensional feature encoding of the construction process is realized. Based on the hierarchical spatio-temporal attention-physical constraint graph convolutional network, cross-stage association weights between nodes are dynamically allocated, and convolutional kernel parameters are optimized in combination with physical connection strength, significantly improving the spatio-temporal correlation of high-risk node recognition. A risk difference matrix is generated through the counterfactual intervention mechanism to quantify the contribution of equipment anomalies to the overall risk, and the abnormal propagation path is traced back in combination with the causal Bayesian network, making the root cause equipment location more accurate. In addition, the root cause coordinates and the risk diffusion range are dynamically marked, and a visualization early warning instruction including a disposal priority is generated and pushed in real time through the edge node, greatly shortening the risk response time. This method breaks through the limitations of traditional static monitoring and experience-driven causal reasoning, realizes accurate modeling of risk propagation in the three-dimensional space-time dimension, and provides a full-link closed-loop solution for power infrastructure projects from risk identification, root cause tracing to intelligent decision-making. Description of the Drawings

[0062] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for description in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings without creative efforts based on these drawings.

[0063] Figure 1 It is a flowchart of the power infrastructure construction process early warning method based on a high-precision three-dimensional model in Embodiment 1.

[0064] Figure 2 It is the flowchart for generating the risk difference matrix in Embodiment 1. Detailed implementation manners

[0065] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following describes the detailed implementation manners of the present invention with reference to the drawings of the specification.

[0066] Many specific details are set forth in the following description to facilitate a thorough understanding of the present invention. However, the present invention may be implemented in other ways different from those described herein. Persons skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0067] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation manner of the present invention. The appearances of "in one embodiment" in different places in this specification do not all refer to the same embodiment, nor are they separate or alternative embodiments that exclude each other with other embodiments.

[0068] Embodiment 1, referring to Figure 1 and Figure 2 is the first embodiment of the present invention. This embodiment provides a power infrastructure construction process warning method based on a high-precision three-dimensional model, including the following steps:

[0069] S1. Collect multi-source data of the power infrastructure construction site, and construct a high-precision three-dimensional model including equipment spatial coordinates, construction time sequence tags, and environmental parameters.

[0070] Specifically, it includes the following steps:

[0071] The multi-source data includes equipment spatial coordinate data, construction time sequence tag data, environmental parameter data, equipment status time sequence data, and construction connection relationship data;

[0072] Specifically, use unmanned aerial vehicle oblique photography, laser scanning (LiDAR), and ground three-dimensional laser scanning technology to collect multi-perspective and multi-resolution data of the power infrastructure project area, ensure millimeter-level accuracy of elements such as terrain, landform, buildings, and roads, obtain multi-source data, and provide a reliable data basis for subsequent construction of a high-precision three-dimensional model.

[0073] To construct a high-precision three-dimensional model including equipment spatial coordinates, construction time sequence tags, and environmental parameters, the specific steps are as follows.

[0074] Perform spatio-temporal alignment and cleaning on the collected equipment spatial coordinates, construction time sequence tags, and environmental parameter data, and output a standardized spatio-temporal data set.

[0075] Among them, the spatio-temporal alignment method includes timestamp alignment and spatial coordinate alignment; the data cleaning method includes outlier detection, missing value handling, and redundant data elimination.

[0076] Extract the device spatial coordinates from the standardized spatio-temporal dataset, combine with the point cloud data collected by LiDAR, perform denoising and registration, and generate the cleaned point cloud data;

[0077] Specifically, use the RANSAC algorithm to initially separate the device-related point cloud from the point cloud data, set the sampling times and inlier threshold; then use statistical filtering to remove low-density outliers, and further remove noise by combining adaptive radius filtering; subsequently use the FPFH feature descriptor for rough registration to optimize the initial pose; in the fine registration stage, introduce the ICP algorithm with physical constraints, and accelerate the nearest neighbor search through KD-Tree; finally calculate the distance between the registered point cloud and the device coordinates, ensure that the error is within the allowable range, and verify the surface continuity, and output the cleaned point cloud data that meets the quality requirements.

[0078] Convert the cleaned point cloud data into a three-dimensional mesh model, optimize the number of patches and map the device spatial coordinates to generate a high-precision basic model;

[0079] Specifically, first use point cloud processing software (such as CloudCompare or MeshLab) to perform triangulation on the cleaned point cloud data to generate a preliminary three-dimensional mesh model, ensuring that the three-dimensional mesh model can accurately represent the spatial structure of the point cloud; then reduce the number of patches of the model through mesh optimization algorithms (such as edge collapse or vertex merging) to reduce the computational complexity while retaining key geometric features; subsequently perform spatial alignment on the device spatial coordinate data and the optimized mesh model, and use the coordinate transformation matrix to accurately map the device position information to the corresponding position of the mesh model to ensure the consistency between the three-dimensional mesh model and the actual space; finally, attach environmental parameters or device status information to the surface of the three-dimensional mesh model through texture mapping or color coding technology to generate a high-precision basic model, providing an intuitive and accurate three-dimensional visualization basis for subsequent construction simulation and data analysis.

[0080] Bind the construction time sequence tags and environmental parameters to the device spatial nodes to form a high-precision three-dimensional model with time sequence and environmental attributes;

[0081] Among them, the device spatial node refers to an abstract node representing the device position in the three-dimensional mesh model or spatio-temporal graph. It not only contains the device spatial coordinates, but may also be bound with other attributes (such as construction time sequence tags, environmental parameters, device status, etc.).

[0082] For example: Attach attribute fields {construction stage: foundation pouring, ambient humidity: 65%, red line distance: 2.3m} to each device node;

[0083] Specifically, during the process of constructing a high-precision 3D model, multi-source data collected needs to be fused through 3D modeling software (such as ContextCapture, Bentley, etc.) to generate a high-precision 3D model containing a geographic coordinate system, supporting refined expressions of terrain elevation, underground pipelines, and above-ground facilities.

[0084] Furthermore, seamlessly connect the high-precision 3D model with the GIS system, integrate vector data such as land red lines, planning scopes, and construction boundaries through a spatial database (such as ArcGIS GeoDatabase) to form a spatial-attribute integrated model. Based on the spatial analysis function of GIS, automatically calculate the spatial relationship between the construction area and the land red line, and mark potential conflict areas (such as the buffer zone for red line expansion).

[0085] S2. Map it to a spatio-temporal graph according to the grid topology, encode the construction timestamp and the temporal sequence change of the device status for the node attributes, and associate the edge weights with the construction connection relationship and physical connection strength between adjacent nodes.

[0086] Specifically, it includes the following steps:

[0087] Divide the high-precision 3D model into grid cells at a fixed resolution, and each grid cell serves as a node of the spatio-temporal graph;

[0088] Extract the construction timestamp and the temporal sequence data of the device status from the standardized spatio-temporal dataset, bind them to the corresponding grid nodes to form node attributes;

[0089] Specifically, first, use a data processing tool (such as the Pandas library in Python) to filter out the construction timestamp and the temporal sequence data of the device status from the standardized spatio-temporal dataset to ensure the integrity and non-missing of the time series of the construction timestamp and the temporal sequence data of the device status; then perform spatio-temporal matching on the extracted timestamp and device status data with the nodes of the 3D grid model, and use a spatial indexing algorithm (such as KD-Tree or R-Tree) to quickly locate the grid node corresponding to each timestamp and device status data; then use the construction timestamp and the temporal sequence data of the device status as attribute information and bind them to the matching grid nodes, and store the mapping relationship between the nodes and the attributes through a data structure (such as a dictionary or an attribute table); finally, verify the bound node attributes to ensure that the timestamp and device status data are consistent with the spatio-temporal positions of the grid nodes, forming complete node attribute information, providing dynamic and accurate data support for subsequent construction process simulation and status analysis.

[0090] Determine the connection relationship between adjacent grid nodes according to the device spatial coordinates and construction connection relationship data, and calculate the physical connection strength as the edge weight;

[0091] Among them, the device spatial coordinates refer to the specific position of the device in three-dimensional space, usually represented in the form of a three-dimensional coordinate system (such as X, Y, Z coordinates). It describes the precise spatial position of the device at the construction site.

[0092] It should be noted that the position of the device in the grid is located through the device spatial coordinates, and adjacent nodes are identified by combining construction processes and interaction data; the physical connection strength is dynamically calculated based on the actual contact frequency of the devices, the construction time sequence interval, and the spatial distance, quantifying the cooperation tightness between devices; the generated weighted connection relationship graph can clearly display the critical path and high-frequency interaction nodes in the construction network, providing a topological structure basis for subsequent resource allocation optimization, construction conflict detection, and progress simulation, and ensuring that the scheduling strategy accurately matches the device collaboration requirements in actual construction.

[0093] Furthermore, embed an automatic collision detection algorithm (such as the Navisworks collision detection module based on BIM), analyze the spatial conflicts between construction equipment, temporary facilities and the red line, underground pipelines, and existing buildings in real time, generate a conflict heat map and associate it with the corresponding space-time nodes. Encode the collision detection results as additional attributes of the nodes (such as conflict type, severity level) for subsequent risk propagation probability calculation.

[0094] Integrate the node attributes and edge weights to generate a space-time graph, and use a graph optimization algorithm to optimize the graph structure.

[0095] S3. Input the space-time graph into the space-time graph convolutional network, output the risk propagation probability matrix in the three-dimensional space and construction time sequence dimensions, and identify high-risk nodes and associated construction stages.

[0096] Specifically, it includes the following steps:

[0097] Based on the node attributes and edge weights of the space-time graph, construct a hierarchical space-time attention - physical constraint graph convolutional network, integrating causal space-time attention units, physical constraint convolutional kernels, and counterfactual reasoning layers as the network architecture;

[0098] Preferably, based on the node attributes and edge weights of the space-time graph, construct a hierarchical space-time attention - physical constraint graph convolutional network, integrating causal space-time attention units, physical constraint convolutional kernels, and counterfactual reasoning layers as the network architecture, which can effectively improve the accuracy and intelligent level of power infrastructure construction process modeling.

[0099] Specifically, by combining node attributes (such as construction timestamps and device status time-series data) with edge weights (such as construction connection relationships), the hierarchical spatio-temporal attention-physical constraint graph convolutional network can capture the dynamic spatio-temporal dependencies during the construction process; the causal spatio-temporal attention unit enhances the ability to identify key construction events by analyzing the causal relationships between nodes; the physical constraint convolutional kernel combines construction physical rules (such as device movement constraints and environmental limitations) to ensure that the model output conforms to the actual construction conditions; the counterfactual reasoning layer provides a scientific basis for optimizing construction decisions by simulating the results under different construction conditions. This network architecture can not only achieve high-precision construction process simulation but also provide intelligent support for improving construction efficiency, risk prediction, and resource optimization, significantly enhancing the scientific nature and efficiency of power infrastructure construction management.

[0100] Based on the causal spatio-temporal attention unit in the network architecture, the temporal dependency relationship is divided by combining construction time-series labels, and the spatial attention distribution is adjusted using the node timestamp difference and physical connection strength.

[0101] Specifically, in the causal spatio-temporal attention unit, the causal relationships and spatio-temporal dependencies in the time-series data are captured through the attention mechanism.

[0102] The expression based on the attention mechanism is:

[0103] ;

[0104] ;

[0105] Among them, represents the attention weight of the current node to the neighbor node , is the attention score between the current node and the neighbor node , represents the learnable parameter vector, is the variable used to traverse all neighbor nodes of the current node , represents the current node 's neighbor set, represents the learnable weight matrix, and are the feature vectors of the previous node to the neighbor node respectively, represents the vector concatenation operation, represents the activation function;

[0106] It should be noted that the attention mechanism formula is based on the framework of the Graph Attention Network (GAT). Specifically, by introducing the attention mechanism, it can dynamically allocate the association weights between nodes, thereby capturing the temporal dependence relationship and physical connection strength between nodes.

[0107] Dynamically allocate the cross-stage node association weights through the Softmax function, strengthen the spatio-temporal correlation of high-risk nodes within the same process, and output the dynamically calculated attention weights;

[0108] Preferably, by dynamically allocating the cross-stage node association weights through the Softmax function, strengthening the spatio-temporal correlation of high-risk nodes, generating dynamic attention weights, it can accurately capture the dependence relationship between construction key nodes, provide accurate input for the subsequent causal spatio-temporal attention unit and counterfactual reasoning layer, optimize risk prediction and resource scheduling, and improve the construction modeling accuracy and intelligent decision-making ability.

[0109] Concatenate the physical parameters in the edge weights with the node device status, input them into a multi-layer perceptron to generate a physical perception convolution kernel, and dynamically adjust the kernel weights according to the environmental parameters;

[0110] Specifically, use a data preprocessing tool to concatenate the physical parameters in the edge weights (such as the physical connection strength between devices or the construction connection relationship) with the time-series data of the node device status (such as the operating status or performance indicators of the device) to form a multi-dimensional feature vector; then input the multi-dimensional feature vector into a multi-layer perceptron (MLP). The MLP extracts and transforms the multi-dimensional feature vector layer by layer through a fully connected layer and a non-linear activation function (such as ReLU) to generate preliminary convolution kernel weights; then dynamically adjust the convolution kernel weights in combination with the environmental parameter data (such as temperature, humidity, or wind speed). Through the correlation calculation between the environmental parameters and the convolution kernel weights, ensure that the convolution kernel can adapt to the physical constraints under different environmental conditions; finally, apply the dynamically adjusted convolution kernel weights to the graph convolution operation to capture the physical perception relationship between nodes and provide convolution kernel support that conforms to the actual physical rules for the subsequent spatio-temporal graph convolution network.

[0111] Fuse the attention weights and the physical constraint convolution kernel, perform multi-head graph convolution on the node features, respectively capture the local physical connection, global process dependence, short-term state fluctuations, and long-term environmental trends, and output spatio-temporal hidden features;

[0112] Preferably, fuse the attention weights and the physical constraint convolution kernel, respectively capture the local physical connection, global process dependence, short-term state fluctuations, and long-term environmental trends through multi-head graph convolution, comprehensively extract the multi-dimensional spatio-temporal information of the node features, output spatio-temporal hidden features, and provide multi-dimensional data support for accurately modeling the dynamic construction process, optimizing risk prediction and resource scheduling in the future.

[0113] Perform counterfactual interventions on the status of each node device, recalculate the graph convolution output, and generate a risk difference matrix by comparing the Euclidean distances between the actual and counterfactual features;

[0114] Specifically, based on the spatio-temporal hidden feature matrix, extract the actual feature vector of each node as the benchmark for counterfactual comparison; perform counterfactual interventions on the device status of each node, generate counterfactual status data by setting the device status to the benchmark value; based on the counterfactual status data, re-execute the graph convolution operation to output the counterfactual feature vector; compare the actual feature vector with the counterfactual feature vector, calculate the Euclidean distance, and generate the node-level risk difference value; map the risk difference values of all nodes according to the grid topology to generate a risk difference matrix, and perform normalization processing to output the standardized risk difference matrix.

[0115] Furthermore, based on the risk difference matrix, combine the Monte Carlo method to dynamically simulate the construction behaviors of high-risk nodes (such as the movement paths of equipment and the stacking ranges of materials), and compare them with the land use red lines in the GIS in real time to calculate the conflict probabilities (such as the overstep probability and the intrusion area). At the same time, set multi-level warning thresholds (such as the distance threshold from the red line and the conflict area threshold). When the simulation results show that the construction activities are approaching or exceeding the red line, trigger hierarchical warnings (yellow warning, red warning).

[0116] It should be noted that in the actual construction scenario, the equipment may cause a decrease in construction efficiency or an increase in risks due to abnormal states (such as failures or performance degradation). The actual feature vector is the key data for capturing these real states, providing a basis for subsequent risk analysis and modeling. The counterfactual feature vector provides a benchmark for comparison with the actual feature vector, can quantify the impact of equipment anomalies on the construction process, reveal the risk differences brought by abnormal states, and provide a scientific basis for optimizing construction decisions and risk prevention and control.

[0117] Preferably, the comparative analysis of the actual feature vector and the counterfactual feature vector can accurately quantify the impact of equipment anomalies on the construction process, generate node-level risk difference values, help identify high-risk nodes and their associated construction stages, provide a data-driven scientific basis for construction risk prediction, resource optimization, and decision support, and significantly improve the intelligent level of construction management.

[0118] Weightedly fuse the spatio-temporal hidden features and the risk difference matrix, map them through the Sigmoid function into a risk propagation probability matrix in the three-dimensional space-time dimension, and label the risk values of each grid node at each construction stage;

[0119] Specifically, perform multi-level feature extraction on the spatio-temporal graph through the spatio-temporal graph convolution network to capture local physical connections, global process dependencies, short-term state fluctuations, and long-term environmental trends.

[0120] Spatio-temporal graph convolution, the expression is:

[0121] ;

[0122] Among them, and respectively represent the node feature matrices of the th layer and the th layer, represents the th spatio-temporal adjacency matrix, represents the degree matrix of the th spatio-temporal adjacency matrix, represents the weight matrix of the th layer and the th convolution kernel, represents the ReLU activation function;

[0123] Input the spatio-temporal hidden features of the last layer into the fully connected layer, and generate the risk propagation probability matrix through the function, the expression is:

[0124] ;

[0125] Among them, represents the normalization operation, which is used to convert the output into a probability distribution;

[0126] Fuse the spatio-temporal hidden features with the risk difference matrix to generate the fused feature matrix, the expression is:

[0127] ;

[0128] Among them, represents the fused feature matrix, represents the weight parameter, which is used to balance the contributions of the spatio-temporal hidden features and the risk difference matrix, and the value range is ;

[0129] Input the fused feature matrix into the Sigmoid function, and map it to the risk propagation probability matrix in the three-dimensional space - time series dimension, the expression is:

[0130]

[0131] Among them, and respectively represent the learned weight matrix and bias vector, Denote the activation function, which is used to limit the output within the range.

[0132] Calculate the dynamic threshold of the risk probability according to the construction stage, set the warning threshold, screen the nodes exceeding the warning threshold as high-risk targets, trace back the construction stage corresponding to the risk peak, and output the high-risk nodes and the associated construction stages;

[0133] Specifically, the specific steps for setting the warning threshold are as follows: First, based on the spatio-temporal hidden features, use probability statistical methods (such as kernel density estimation or Gaussian mixture model) to calculate the node risk probability distribution according to the construction stage, and dynamically determine the risk probability threshold for each stage; Then, combine historical construction data and personnel experience, and dynamically adjust the risk probability threshold through a sliding window or an adaptive algorithm to ensure that the risk probability threshold can reflect the actual situation of the current construction stage; Then, set the warning line according to the risk probability distribution and the threshold, and mark the nodes exceeding the warning line as potential high-risk targets to provide a basis for subsequent risk screening.

[0134] Furthermore, screen out the nodes with a risk probability exceeding the warning threshold as high-risk targets; Then, analyze the risk change curve of these nodes during the construction process through a backtracking algorithm, and locate the construction stage corresponding to the risk peak; Finally, integrate and output the high-risk nodes and their associated construction stages to form a risk warning report, providing accurate risk positioning and stage association information for construction management, and supporting risk prevention and control and construction optimization decisions.

[0135] S4. Construct a causal Bayesian network, calculate the difference in risk evolution after eliminating equipment anomalies through counterfactual simulation, and locate the root cause equipment and processes.

[0136] Specifically, it includes the following steps:

[0137] Based on the high-risk nodes and the associated construction stages, construct a Bayesian network of the causal dependence relationship between equipment states and construction processes;

[0138] Specifically, extract the equipment state time series data and construction process information from the high-risk node and associated construction stage data output by the spatio-temporal graph convolutional network, and clarify the correlation between high-risk nodes and construction processes; Then, define the potential causal relationship between equipment states and construction processes through historical data analysis, and determine the node and edge structure of the Bayesian network; Then, use the maximum likelihood estimation or Bayesian learning method to train the network parameters, and quantify the conditional probability distribution between equipment states and construction processes; Finally, verify the accuracy of the network structure and the rationality of the causal relationship to ensure that the Bayesian network can accurately reflect the causal dependence relationship between equipment states and construction processes, providing a causal reasoning basis for subsequent counterfactual simulation and risk evolution analysis.

[0139] Generate counterfactual intervention scenarios through a Bayesian network, simulate the state changes after eliminating equipment anomalies, and output counterfactual state data;

[0140] It should be noted that the counterfactual state data is generated by simulating the operating state of the equipment after eliminating anomalies through a Bayesian network. The specific process is to construct a causal Bayesian network based on actual construction data, define the equipment anomaly state and its influencing factors; simulate the operation of the equipment under normal conditions in the Bayesian network through intervention operations (such as "assuming the equipment has not malfunctioned") to generate counterfactual scenarios; then use Bayesian inference to calculate the state changes of the equipment in the counterfactual scenarios and output the counterfactual state data. These data reflect the operating state of the equipment under assumed normal conditions, provide benchmark data for the calculation of the subsequent risk propagation probability matrix, and help quantify the impact of equipment anomalies on the construction process.

[0141] Input into the spatio-temporal graph convolutional network to calculate the risk propagation probability matrix under counterfactual scenarios;

[0142] Compare the risk propagation probability matrices of the actual scenario and the counterfactual scenario to generate a risk evolution difference matrix;

[0143] It should be noted that the actual scenario is generated based on the real data at the construction site, including information such as equipment state time-series data, environmental parameter data, construction timestamps, and node attributes, reflecting the state and risk propagation of nodes under real conditions during the construction process; the counterfactual scenario is generated by simulating the operating state of the equipment after eliminating anomalies through a Bayesian network, reflecting the state and risk propagation of the equipment under assumed normal conditions. By comparing the risk propagation probability matrices of the actual scenario and the counterfactual scenario to generate a risk evolution difference matrix, it is possible to accurately quantify the impact of equipment anomalies on the overall risk, identify the root cause equipment with the highest risk contribution, and trace back the impact path of its abnormal state on adjacent nodes and construction processes.

[0144] Preferably, the counterfactual state data and the comparative analysis of the actual scenario and the counterfactual scenario can accurately locate the root cause equipment and key processes, reveal the risk impact path of equipment anomalies on the construction process, and significantly improve the intelligent level of construction management.

[0145] Identify the root cause equipment with the highest contribution to the overall risk according to the risk evolution difference matrix;

[0146] Specifically, extract the risk contribution values of each equipment node from the risk evolution difference matrix. These risk contribution values reflect the differences in risk changes before and after the elimination of equipment anomalies. Then, use a sorting algorithm (such as Top-K sorting) to screen out the equipment nodes with the highest risk contribution values as potential root cause equipment. Next, combine the equipment status time-series data and construction process information to analyze the abnormal status of these root cause equipment during construction and its influence scope, and verify the rationality of their being root cause equipment.

[0147] For example, if the risk contribution value of a certain crane in multiple construction stages is significantly higher than that of other equipment, and its abnormal status (such as failure or performance degradation) directly leads to construction delays or safety accidents, then this crane can be determined as the root cause equipment. Finally, integrate and output the identified root cause equipment and its risk contribution values to provide accurate target equipment for construction risk prevention and control and resource optimization, and support subsequent decision-making and risk mitigation measures.

[0148] Trace back the influence path of the abnormal status of the root cause equipment on adjacent nodes and construction processes to locate the key processes.

[0149] S5. Dynamically mark the root cause coordinates and the risk diffusion range in the high-precision three-dimensional model, generate a warning instruction containing a visualization path, a root cause report, and a disposal priority, and push it to the terminal in real time through the edge node.

[0150] Specifically, it includes the following steps:

[0151] Based on the located root cause equipment, extract its spatial coordinate data and dynamically mark the root cause coordinates in the high-precision three-dimensional model;

[0152] It should be noted that dynamically marking the root cause coordinates can intuitively display the specific location of the root cause equipment, providing an accurate spatial reference for construction risk positioning, resource scheduling, and decision optimization.

[0153] Starting from the root cause coordinates, according to the risk evolution difference matrix, calculate the risk diffusion range of the root cause equipment and visually mark the diffusion boundary in the high-precision three-dimensional model.

[0154] Preferably, by visually marking the diffusion boundary, the area range affected by the risk can be intuitively displayed, providing an accurate spatial basis for construction risk prevention and control, resource optimization, and decision-making, and significantly improving the scientificity and efficiency of construction management.

[0155] Combine the risk propagation probability matrix, extract the risk propagation paths of high-risk nodes, and draw a visualization path in the high-precision three-dimensional model.

[0156] Integrate the root cause equipment and related process information to generate a detailed root cause report including equipment status, abnormal reasons, influence scope, and related processes.

[0157] Specifically, through data fusion and knowledge graph technology, the status data, abnormal causes, influence scope, and associated process information of the root cause equipment are structurally integrated to generate a detailed root cause report, providing comprehensive and accurate data support for construction risk analysis and decision-making optimization.

[0158] Calculate the disposal priority of each root cause equipment according to the risk value and diffusion scope of the nodes in the risk evolution difference matrix;

[0159] Integrate the visualization path, root cause report, and disposal priority to generate a structured warning instruction;

[0160] Furthermore, data visualization and interaction support will be provided based on a 3D visualization platform using WebGL / Three.js, supporting real-time loading of high-precision models, conflict heat maps, and warning information on browsers or mobile devices.

[0161] The provided interaction functions are as follows: multi-dimensional view switching: overlay / hide the land use red line, planning scope, and construction progress layers; conflict penetration query: click on the conflict area to view the specific equipment number, responsible unit, and disposal time limit; warning disposal tracking: mark the processed area and retain the operation log to form a closed-loop management.

[0162] Push the warning instruction to the terminal device in real time through the edge node and take corresponding measures.

[0163] This embodiment also provides a power infrastructure construction process warning system based on a high-precision 3D model, including: a data acquisition module, a graph structure mapping module, a risk identification module, a root cause location module, and a warning push module; the data acquisition module is used to collect multi-source data of the power infrastructure construction site and construct a high-precision 3D model containing equipment spatial coordinates, construction time sequence tags, and environmental parameters; the graph structure mapping module is used to map it into a spatio-temporal graph according to the grid topology, with node attributes encoding the construction time stamp and the time sequence change of the equipment status, and the edge weight associating the construction connection relationship and physical connection strength between adjacent nodes; the risk identification module is used to input the spatio-temporal graph into the spatio-temporal graph convolutional network, output the risk propagation probability matrix in the three-dimensional space and construction time sequence dimension, and identify high-risk nodes and associated construction stages; the root cause location module is used to construct a causal Bayesian network, calculate the risk evolution difference after eliminating equipment anomalies through counterfactual simulation, and locate the root cause equipment and processes; the warning push module is used to dynamically mark the root cause coordinates and risk diffusion scope in the high-precision 3D model, generate a warning instruction containing a visualization path, a root cause report, and a disposal priority, and push it to the terminal in real time through the edge node.

[0164] This embodiment also provides a computer device, which is applicable to the situation of the power infrastructure construction process warning method based on a high-precision three-dimensional model, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the power infrastructure construction process warning method based on a high-precision three-dimensional model as proposed in the above embodiment.

[0165] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or may be a button, a trackball, or a touchpad provided on the housing of the computer device, or may also be an external keyboard, a touchpad, or a mouse, etc.

[0166] This embodiment also provides a storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the power infrastructure construction process warning method based on a high-precision three-dimensional model as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM for short), Electrically Erasable Programmable Read-Only Memory (EEPROM for short), Erasable Programmable Read-Only Memory (EPROM for short), Programmable Read-Only Memory (PROM for short), Read-Only Memory (ROM for short), magnetic memory, flash memory, a magnetic disk, or an optical disc.

[0167] In summary, the present invention realizes the multi-dimensional feature encoding of the construction process by constructing a high-precision three-dimensional model that integrates device space coordinates, construction time sequence tags, and environmental parameters, and mapping it into a spatio-temporal graph structure. Based on the hierarchical spatio-temporal attention-physical constraint graph convolutional network, the cross-stage correlation weights between nodes are dynamically allocated, and the convolutional kernel parameters are optimized by combining the physical connection strength, significantly improving the spatio-temporal correlation of high-risk node recognition. Through the counterfactual intervention mechanism, a risk difference matrix is generated to quantify the contribution of device anomalies to the overall risk. Combining with the causal Bayesian network to trace back the anomaly propagation path makes the root cause device positioning more accurate. In addition, the root cause coordinates and the risk diffusion range are dynamically marked, and a visual warning instruction with disposal priority is generated and pushed in real time through the edge node, greatly shortening the risk response time. This method breaks through the limitations of traditional static monitoring and experience-driven causal reasoning, realizes the accurate modeling of risk propagation in the three-dimensional space-time dimension, and provides a full-link closed-loop solution for power infrastructure projects from risk identification, root cause tracing to intelligent decision-making.

[0168] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A power infrastructure construction process warning method based on a high-precision three-dimensional model, characterized in that: including Collecting multi-source data of the power infrastructure construction site and constructing a high-precision three-dimensional model including equipment spatial coordinates, construction time series tags, and environmental parameters; Mapping it into a spatio-temporal graph according to the grid topology, encoding the construction timestamp and the temporal change of equipment status in the node attributes, and associating the edge weights with the construction connection relationship and physical connection strength between adjacent nodes; Inputting the spatio-temporal graph into a spatio-temporal graph convolutional network, outputting a risk propagation probability matrix in the three-dimensional space and construction time series dimensions, and identifying high-risk nodes and associated construction stages; Constructing a causal Bayesian network, calculating the difference in risk evolution after eliminating equipment anomalies through counterfactual simulation, and locating the root cause equipment and processes; Dynamically annotating the root cause coordinates and risk diffusion range in the high-precision three-dimensional model, generating a warning instruction including a visualization path, a root cause report, and a disposal priority, and pushing it to the terminal in real time through an edge node; The multi-source data includes equipment spatial coordinate data, construction time series tag data, environmental parameter data, equipment status time series data, and construction connection relationship data; The construction of a high-precision three-dimensional model including equipment spatial coordinates, construction time series tags, and environmental parameters is as follows: Performing spatio-temporal alignment and cleaning on the collected equipment spatial coordinates, construction time series tags, and environmental parameter data, and outputting a standardized spatio-temporal dataset; Extracting equipment spatial coordinates from the standardized spatio-temporal dataset, combining with the point cloud data collected by LiDAR, performing denoising and registration, and generating cleaned point cloud data; Converting the cleaned point cloud data into a three-dimensional mesh model, optimizing the number of patches and mapping the equipment spatial coordinates, and generating a high-precision basic model; Binding the construction time series tags and environmental parameters to the equipment spatial nodes to form a high-precision three-dimensional model with time series and environmental attributes; The mapping into a spatio-temporal graph according to the grid topology, encoding the construction timestamp and the temporal change of equipment status in the node attributes, and associating the edge weights with the construction connection relationship and physical connection strength between adjacent nodes is as follows: Dividing the high-precision three-dimensional model into grid cells according to a fixed resolution, and each grid cell serves as a node of the spatio-temporal graph; Extracting the construction timestamp and equipment status time series data from the standardized spatio-temporal dataset and binding them to the corresponding grid nodes to form node attributes; According to the equipment spatial coordinates and construction connection relationship data, determining the connection relationship between adjacent grid nodes, and calculating the physical connection strength as the edge weight; Integrating the node attributes and edge weights to generate a spatio-temporal graph, and using a graph optimization algorithm to optimize the graph structure; The input of the spatio-temporal graph into a spatio-temporal graph convolutional network, outputting a risk propagation probability matrix in the three-dimensional space and construction time series dimensions, and identifying high-risk nodes and associated construction stages is as follows: Based on the node attributes and edge weights of the spatio-temporal graph, constructing a hierarchical spatio-temporal attention-physical constraint graph convolutional network, integrating a causal spatio-temporal attention unit, a physical constraint convolutional kernel, and a counterfactual reasoning layer as the network architecture; Based on the causal spatio-temporal attention unit in the network architecture, dividing the temporal dependence relationship in combination with the construction time series tags, and using the node timestamp difference and physical connection strength to adjust the spatial attention distribution; Dynamically allocate cross-stage node association weights through the Softmax function, strengthen the spatio-temporal correlation of high-risk nodes within the same process, and output dynamically calculated attention weights; Concatenate the physical parameters in the edge weights with the node device status, input them into a multi-layer perceptron to generate a physically aware convolutional kernel, and dynamically adjust the kernel weights with environmental parameters; Fuse the attention weights and physically constrained convolutional kernels, perform multi-head graph convolution on the node features to capture local physical connections, global process dependencies, short-term state fluctuations, and long-term environmental trends respectively, and output spatio-temporal hidden features; Perform counterfactual intervention on the status of each node device, recalculate the graph convolution output, and generate a risk difference matrix by comparing the Euclidean distances between the actual and counterfactual features; Weightedly fuse the spatio-temporal hidden features and the risk difference matrix, map them through the Sigmoid function to a risk propagation probability matrix in the three-dimensional space-time dimension, and label the risk values of each grid node at each construction stage; Calculate the dynamic threshold of the risk probability according to the construction stage, set the warning threshold, screen the nodes exceeding the warning threshold as high-risk targets, and trace back to the construction stage corresponding to the risk peak value, and output the high-risk nodes and associated construction stages.

2. The early warning method for the power infrastructure construction process based on a high-precision three-dimensional model according to claim 1, characterized in that: The above-mentioned counterfactual intervention on the status of each node device, recalculating the graph convolution output, and generating a risk difference matrix by comparing the Euclidean distances between the actual and counterfactual feature vectors, the specific steps are as follows, Based on the spatio-temporal hidden feature matrix, extract the actual feature vector of each node as the benchmark for counterfactual comparison; Perform counterfactual intervention on the status of each node's device, and generate counterfactual status data by setting the device status to the benchmark value; Based on the counterfactual status data, re-perform the graph convolution operation and output the counterfactual feature vector; Compare the actual feature vector and the counterfactual feature vector, calculate the Euclidean distance, and generate a node-level risk difference value; Map the risk difference values of all nodes according to the grid topology, generate a risk difference matrix, and perform normalization processing to output a standardized risk difference matrix.

3. The early warning method for the power infrastructure construction process based on a high-precision three-dimensional model according to claim 1, characterized in that: The above-mentioned construction of a causal Bayesian network, calculating the risk evolution difference after eliminating equipment anomalies through counterfactual simulation, and locating the root cause equipment and processes, the specific steps are as follows, Based on the high-risk nodes and associated construction stages, construct a Bayesian network of the causal dependence relationship between device status and construction processes; Generate a counterfactual intervention scenario through the Bayesian network, simulate the state change after eliminating equipment anomalies, and output counterfactual status data; Input into the spatio-temporal graph convolution network to calculate the risk propagation probability matrix in the counterfactual scenario; Compare the risk propagation probability matrices of the actual scenario and the counterfactual scenario to generate a risk evolution difference matrix; According to the risk evolution difference matrix, identify the root cause equipment with the highest contribution to the overall risk; Trace back the influence path of the abnormal state of the root cause equipment on adjacent nodes and construction processes to locate the key processes.

4. The early warning method for the power infrastructure construction process based on a high-precision three-dimensional model according to claim 1, wherein: The above-mentioned dynamically annotating the root cause coordinates and the risk diffusion range in the high-precision three-dimensional model, generating a warning instruction including a visualization path, a root cause report, and a disposal priority, and pushing it to the terminal in real time through the edge node, the specific steps are as follows, Based on the located root cause equipment, extract its spatial coordinate data and dynamically annotate the root cause coordinates in the high-precision three-dimensional model; Starting from the root cause coordinates, calculate the risk diffusion range of the root cause equipment according to the risk evolution difference matrix, and visually mark the diffusion boundary in the high-precision three-dimensional model; Combined with the risk propagation probability matrix, extract the risk propagation paths of high-risk nodes and draw visual paths in the high-precision three-dimensional model; Integrate the information of the root cause equipment and related processes to generate a detailed root cause report including equipment status, abnormal reasons, influence scope and related processes; Calculate the disposal priority of each root cause equipment according to the risk value and diffusion range of the nodes in the risk evolution difference matrix; Integrate the visual path, root cause report and disposal priority to generate a structured warning instruction; Push the warning instruction to the terminal device in real time through the edge node and take corresponding measures.

5. A power infrastructure construction process warning system based on a high-precision 3D model, based on the power infrastructure construction process warning method based on a high-precision 3D model according to any one of claims 1 to 4, characterized in that: Including a data acquisition module, a graph structure mapping module, a risk identification module, a root cause location module and a warning push module; The data acquisition module is used to collect multi-source data of the power infrastructure construction site and construct a high-precision three-dimensional model including equipment spatial coordinates, construction time sequence tags and environmental parameters; The graph structure mapping module is used to map into a spatio-temporal graph according to grid topology, encode the construction time stamp and the time sequence change of equipment status for node attributes, and associate the construction connection relationship and physical connection strength between adjacent nodes for edge weights; The risk identification module is used to input the spatio-temporal graph into the spatio-temporal graph convolutional network, output the risk propagation probability matrix in the three-dimensional space and construction time sequence dimension, and identify high-risk nodes and related construction stages; The root cause location module is used to construct a causal Bayesian network, calculate the risk evolution difference after eliminating equipment anomalies through counterfactual simulation, and locate the root cause equipment and processes; The warning push module is used to dynamically mark the root cause coordinates and risk diffusion range in the high-precision three-dimensional model, generate a warning instruction including visual path, root cause report and disposal priority, and push it to the terminal in real time through the edge node.

6. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the power infrastructure process warning method based on a high-precision three-dimensional model according to any one of claims 1 to 4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the power infrastructure process warning method based on a high-precision three-dimensional model according to any one of claims 1 to 4.

Images