Underground pipeline three-dimensional modeling method and system and storage medium

Abstract

The invention relates to the technical field of pipeline three-dimensional modeling, and discloses an underground pipeline three-dimensional modeling method and system and a storage medium, and the method comprises the steps: obtaining point cloud data, BIM model data and GIS spatial data, and dynamically distributing weights based on data quality, timeliness and application scenes to form fusion pipeline basic data; on the basis of the pipeline trend, clustering parameters are adjusted according to the local density of the point cloud to carry out self-adaptive clustering, and a pipeline geometric model is generated; constructing a three-dimensional space-time diagram based on the fusion data, and predicting the abnormality of a pipe network through a neural network model containing space feature extraction, time feature extraction and an attention mechanism space-time fusion layer; and updating the visual state of the three-dimensional model based on a prediction result, marking high-risk and normal pipe sections, displaying an abnormal propagation path according to a pipe network topological structure and a fluid dynamics principle, and triggering updating when a preset condition is met. According to the method and the device, the target of meeting the actual demand of pipe network full-life-cycle fine management can be achieved.

Description

Technical Field

[0001] This application relates to the technical field of three-dimensional pipeline modeling, and in particular to a method, system and storage medium for three-dimensional modeling of underground pipelines. Background Technology

[0002] Underground pipelines are a core component of urban infrastructure, and their 3D modeling is a crucial technical support for pipeline network planning and design, daily operation and maintenance, and emergency response. Currently, the industry widely uses technologies such as 3D laser scanning, BIM modeling, and GIS spatial analysis to acquire pipeline-related data. By integrating multi-source data, 3D models are constructed, providing fundamental data support for the utilization of underground space resources and the safety management of pipeline networks.

[0003] Existing 3D modeling methods for underground pipelines lack differentiated adaptation to the characteristics of multi-source data during the data processing stage, have insufficient flexibility in adjusting geometric parameters during the modeling process, and struggle to balance accuracy and intuitiveness in pipeline anomaly prediction and visualization, resulting in limited support for operation and maintenance decisions in practical applications.

[0004] As can be seen from the above, existing 3D modeling methods for underground pipelines have problems such as poor accuracy of multi-source data fusion, insufficient adaptability of geometric modeling, weak targeting of anomaly prediction, and lack of visualization and early warning effects. How to meet the actual needs of refined management of the entire life cycle of pipeline networks remains to be solved. Summary of the Invention

[0005] To meet the practical needs of refined management of the entire life cycle of pipeline networks, this application provides a three-dimensional modeling method, system, and storage medium for underground pipelines.

[0006] Firstly, this application provides a three-dimensional modeling method for underground pipelines, employing the following technical solution: A method for 3D modeling of underground pipelines, comprising: The process involves acquiring point cloud data, BIM model data, and GIS spatial data. Point cloud data is obtained through 3D laser scanning. BIM model data includes parametric attribute information of pipelines, and GIS spatial data includes topological relationships and spatial distribution information of pipelines. Weights are dynamically allocated from multiple data sources based on data quality, timeliness, and application scenario to form fused pipeline baseline data. Data quality assessment is based on point cloud accuracy, BIM parameter completeness, and GIS topological rationality; timeliness assessment is based on data acquisition time; application scenario assessment is based on pipeline type and importance to urban areas; point cloud accuracy assessment coefficient is calculated based on point cloud density gradient; BIM parameter completeness assessment coefficient is calculated based on parameter missing rate; and GIS topological rationality assessment coefficient is calculated based on topology error rate. Based on the pipeline orientation, a cylindrical neighborhood analysis method is used to dynamically adjust the clustering parameters according to the local density of the point cloud, and adaptively cluster the pipeline point cloud to generate a pipeline geometric model. The cylindrical neighborhood is based on the pipeline orientation as the axis, with the horizontal radius proportional to the pipe diameter and the vertical height related to the change in burial depth. A 3D spatiotemporal graph is constructed based on the fused pipeline baseline data, where nodes represent pipeline points and edges represent pipeline segments. Pipeline anomalies are predicted using a 3D spatiotemporal graph neural network model, and the corresponding prediction results are obtained. The 3D spatiotemporal graph neural network model includes a spatial feature extraction layer, a temporal feature extraction layer, and a spatiotemporal information fusion layer. The spatial feature extraction layer uses a graph convolutional network to capture the spatial topological relationships of the pipeline, the temporal feature extraction layer uses an LSTM network to capture the temporal evolution features of the pipeline, and the spatiotemporal information fusion layer dynamically fuses spatial and temporal features through an attention mechanism. The visualization status of the 3D pipeline model is dynamically updated based on the prediction results. High-risk pipeline sections are marked in red, and normal pipeline sections are displayed in green. The abnormal propagation path is displayed in the 3D model. The calculation of the abnormal propagation path is based on the pipeline network topology and fluid dynamics principles. By simulating the flow direction and velocity of fluid in the pipeline network, the path and scope of abnormal propagation are determined. The dynamic update is triggered by the following conditions: the predicted abnormal probability exceeds a threshold, the abnormal duration exceeds a preset time, and the abnormal impact range exceeds a preset area.

[0007] Optionally, the process of dynamically allocating multi-source data weights includes: The underground environmental impact factors are introduced, including soil type, groundwater depth, and intensity of surrounding construction activities in the pipeline area. These underground environmental impact factors are quantified into correction coefficients for data reliability. The correction coefficients are lower for areas with poor soil stability, shallow groundwater depth, and frequent construction. The quality assessment weights are adjusted based on the accuracy level of the data acquisition equipment. The accuracy level of 3D laser scanning equipment is divided into three levels according to industry standards, with higher levels corresponding to higher point cloud data quality weights. The maturity level of BIM modeling tools corresponds to the adjustment of parameter integrity weights. An initial weight allocation model is established based on historical modeling data. The inputs include the quality assessment results of the current data, timeliness information, application scenario requirements, and correction coefficients. The output is the initial weight allocation scheme. The modeling accuracy after initial weight fusion is evaluated by cross-validation. If the accuracy does not meet the preset target, the weight ratio is iteratively optimized until the modeling accuracy meets the requirements. In the iterative optimization of the weight ratio, quality weight is given priority, environmental correction is secondary, and timeliness is dynamically fine-tuned.

[0008] Optionally, the error correction and multi-scale adaptation process after generating the pipeline geometry model includes: Construct a geometric model error detection index system, which includes geometric deviation, topological integrity, and parameter matching degree, and set the qualified threshold for each index; The accuracy of the geometric model is verified by reverse verification of point cloud data. The average distance deviation between the surface of the geometric model and the corresponding point cloud is calculated. If it exceeds the threshold, cylindrical neighborhood clustering is re-executed for the deviation area, and the neighborhood parameters are adjusted. The model's burial depth parameters are corrected by combining ground settlement data. The historical ground settlement of the pipeline area is extracted from GIS spatial data, and the cumulative settlement value is calculated by weighting according to the time series. The burial depth coordinates of the geometric model are then dynamically corrected. Generate multi-scale geometric models and set accuracy levels according to application scenario requirements. Accuracy levels include high-precision operation and maintenance level, medium-precision planning level, and lightweight display level. High-precision models retain pipeline detail features, while lightweight models simplify non-critical geometric information and support on-demand access.

[0009] Optionally, in the process of dynamically fusing spatial and temporal features through an attention mechanism in the spatiotemporal information fusion layer, the following steps are included: The spatial features that need to be focused on are selected. These include pipe sections in the core urban area, pipe sections with frequent historical failures, pipe sections connecting critical infrastructure, and large-diameter trunk pipe sections. Higher spatial feature weights are assigned to the pipe sections corresponding to the spatial features that need to be focused on. Time characteristics are classified into timeliness levels: pipeline operation data within the last three months are marked as high-timeliness data, data from three months to one year are marked as medium-timeliness data, and data from more than one year are marked as low-timeliness data. Different time weights are assigned to different timeliness levels. By calculating the correlation between spatial and temporal features through an attention mechanism, we focus on analyzing the spatiotemporal coupling relationship of key pipe sections corresponding to high-timeliness data, thereby improving the feature fusion weight of key pipe sections corresponding to high-timeliness data. By dynamically adjusting the fusion strategy based on the pipeline operation status, when a real-time abnormal signal corresponding to pressure fluctuation or flow abnormality occurs in a pipeline segment, the corresponding abnormal pipeline is identified, and the time feature weight of the abnormal pipeline segment is temporarily increased. The fused spatiotemporal features are normalized to eliminate the dimensional differences between features of different dimensions, generating a fused feature vector of a unified dimension, which is then input into the anomaly prediction module.

[0010] Optionally, the iterative optimization process of the three-dimensional spatiotemporal graph neural network model includes: Collect deviation data between the prediction results of the three-dimensional spatiotemporal neural network model and the actual pipeline network anomalies, and establish deviation evaluation indicators, including the anomaly type identification accuracy, anomaly occurrence time prediction error, and impact range prediction deviation. Filter the pipe segment types that are frequently misjudged in the deviation data, extract the spatial topological features, historical runtime sequence data and environmental correlation data of the frequently misjudged pipe segment types to form a special optimization dataset; adjust the model parameters based on the special optimization dataset, increase the extraction weight of key features, optimize the graph convolution kernel size of the spatial feature extraction layer, and adjust the number of LSTM units in the temporal feature extraction layer. A transfer learning mechanism is introduced to transfer the pipeline network model parameters of mature areas to new modeling areas, and fine-tune them by combining a small amount of measured data from the new modeling areas. The performance of the iterated 3D spatiotemporal neural network model is tested regularly. If all evaluation indicators meet the preset standards, the optimization is completed; otherwise, iterative optimization continues.

[0011] Optionally, the early warning and linkage process for high-risk pipeline sections includes: Based on the anomaly prediction results, the coordinates and burial depth information of high-risk pipe sections are extracted and associated with the distribution of surrounding pipelines in GIS spatial data; and early warning information is generated to clarify the location of high-risk pipe sections and the associated risks of adjacent pipe sections; the early warning information is linked with the visualization status of the 3D model, and clicking on the early warning information directly locates the corresponding pipe section.

[0012] Secondly, this application provides a three-dimensional modeling system for underground pipelines, which adopts the following technical solution: A 3D modeling system for underground pipelines, comprising: The multi-source data acquisition and weighted fusion module acquires point cloud data, BIM model data, and GIS spatial data. The point cloud data is acquired through 3D laser scanning, the BIM model data includes parametric attribute information of pipelines, and the GIS spatial data includes topological relationships and spatial distribution information of pipelines. The module dynamically assigns weights to the multi-source data based on data quality, timeliness, and application scenario to form fused basic pipeline data. Data quality assessment is based on point cloud accuracy, BIM parameter completeness, and GIS topological rationality; timeliness assessment is based on data acquisition time; application scenario assessment is based on pipeline type and importance to the urban area; point cloud accuracy assessment coefficient is calculated based on point cloud density gradient; BIM parameter completeness assessment coefficient is calculated based on parameter missing rate; and GIS topological rationality assessment coefficient is calculated based on topology error rate. The clustering analysis and geometric modeling module uses a cylindrical neighborhood analysis method based on the pipeline direction to dynamically adjust the clustering parameters according to the local density of the point cloud, and performs adaptive clustering on the pipeline point cloud to generate a pipeline geometric model. The cylindrical neighborhood is based on the pipeline direction as the axis, with the horizontal radius proportional to the pipe diameter and the vertical height related to the change in burial depth. The spatiotemporal graph construction and anomaly prediction module constructs a 3D spatiotemporal graph based on the fused pipeline baseline data, where nodes represent pipeline points and edges represent pipeline segments. It then predicts pipeline anomalies using a 3D spatiotemporal graph neural network model, obtaining the corresponding prediction results. This model includes a spatial feature extraction layer, a temporal feature extraction layer, and a spatiotemporal information fusion layer. The spatial feature extraction layer uses a graph convolutional network to capture the spatial topological relationships of the pipelines, the temporal feature extraction layer uses an LSTM network to capture the temporal evolution characteristics of the pipelines, and the spatiotemporal information fusion layer dynamically fuses spatial and temporal features through an attention mechanism. The visualization update and path display module dynamically updates the visualization status of the 3D pipeline model based on the prediction results, marking high-risk pipe sections in red and displaying normal pipe sections in green, and displaying the abnormal propagation path in the 3D model. The calculation of the abnormal propagation path is based on the pipeline network topology and fluid dynamics principles. By simulating the flow direction and velocity of fluid in the pipeline network, the path and impact range of abnormal propagation are determined. The dynamic update is triggered by the following conditions: the predicted abnormal probability exceeds a threshold, the abnormal duration exceeds a preset time, and the abnormal impact range exceeds a preset area.

[0013] Thirdly, this application provides a three-dimensional modeling system for underground pipelines, which adopts the following technical solution: A three-dimensional modeling system for underground pipelines includes a processor, wherein the processor runs a program for the three-dimensional modeling method for underground pipelines described in any one of the above-mentioned methods.

[0014] Fourthly, this application provides a storage medium, which adopts the following technical solution: A storage medium storing a program for the three-dimensional modeling method of underground pipelines as described in any one of the above.

[0015] In summary, this application includes at least one of the following beneficial technical effects: By using a multi-source data dynamic weight fusion mechanism, and by adjusting data weights based on the differences in underground environmental impact factors and equipment accuracy levels, coupled with point cloud reverse verification, ground settlement data correction, and multi-scale adaptation strategies, a high-precision and highly adaptable three-dimensional geometric model of the pipeline is constructed. This ensures the accuracy of data during the planning and design phase, while also meeting the diverse needs for detailed precision and lightweight display during the operation and maintenance phase, laying a solid data and model foundation for the full life cycle management of the pipeline network.

[0016] Meanwhile, relying on the precise spatiotemporal feature fusion and iterative optimization capabilities of the 3D spatiotemporal graph neural network model, combined with the high-risk pipeline section early warning linkage mechanism, it can achieve accurate prediction, dynamic tracking and intuitive presentation of pipeline anomalies. It can identify potential risks in advance and clarify the scope of impact, providing targeted support for operation and maintenance decisions, effectively connecting daily maintenance, emergency response and long-term planning, and fully meeting the core requirements of refined management of the entire pipeline life cycle for data fusion accuracy, model adaptability, anomaly prediction accuracy and intuitive early warning. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a three-dimensional modeling method for underground pipelines according to an exemplary embodiment.

[0018] Figure 2 This is a structural block diagram of a three-dimensional modeling system for underground pipelines, according to an exemplary embodiment. Detailed Implementation

[0019] The embodiments of this application are described in detail below, and examples of the embodiments are shown in the accompanying drawings.

[0020] In the description of this specification, the references to "certain embodiments," "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples" refer to specific features, structures, materials, or characteristics described in connection with the described embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0021] This application discloses a method for three-dimensional modeling of underground pipelines, referring to... Figure 1 ,include; S100 acquires point cloud data, BIM model data, and GIS spatial data. The point cloud data is acquired through 3D laser scanning, the BIM model data includes parametric attribute information of pipelines, and the GIS spatial data includes the topological relationships and spatial distribution information of pipelines. The weights of the multi-source data are dynamically allocated based on data quality, timeliness, and application scenario to form fused pipeline foundation data. Data quality assessment is based on point cloud accuracy, BIM parameter completeness, and GIS topological rationality; timeliness assessment is based on data acquisition time; application scenario assessment is based on pipeline type and importance of urban areas; point cloud accuracy assessment coefficient is calculated based on point cloud density gradient; BIM parameter completeness assessment coefficient is calculated based on parameter missing rate; and GIS topological rationality assessment coefficient is calculated based on topology error rate.

[0022] The S100 execution process specifically includes: Step 1: Conduct targeted collection of multi-source data. A comprehensive 3D laser scanning system is used to scan the area containing underground pipelines. The scan covers the entire pipeline's starting point, ending point, and key nodes along its route. The scanning precision is adjusted according to the pipeline type, with millimeter-level precision used for critical pipelines such as gas and water supply. Point cloud data containing information such as the pipeline's 3D surface coordinates and reflection intensity is acquired, comprehensively recording the pipeline's physical morphology and spatial location. Corresponding BIM model data is extracted from the pipeline design or maintenance database, focusing on collecting structured and parametric attribute information such as pipe diameter, material, wall thickness, design pressure, installation time, manufacturer, and maintenance cycle, ensuring coverage of core parameters required for the entire pipeline lifecycle management. GIS data of underground pipelines in the target area is retrieved through the city's geographic information system, extracting topological relationships such as the connection methods between pipe segments and points, the spatial association between pipelines and surrounding underground structures, and pipeline branch relationships, as well as spatial distribution information such as the pipeline's plane coordinates, burial depth data, and geographical identifiers of the roads or areas where it is located, establishing a mapping between the pipeline and the geographic space.

[0023] Step 2: After data acquisition, perform multi-source data preprocessing: The coordinate systems of BIM model data and GIS spatial data are converted to the 3D laser scanning coordinate system of point cloud data. Matching and calibration are performed using corresponding feature points such as pipeline inflection points and interfaces. The conversion error is calculated to ensure that the coordinate deviation of the three types of data is controlled within a preset threshold. Raw data of different formats are uniformly converted to a preset standardized data format. Point cloud data is converted from LAS format, BIM model data from IFC format, and GIS spatial data from Shapefile format. Simultaneously, the data is cleaned to remove duplicate and invalid noise data, and missing key fields are added to ensure data integrity and consistency.

[0024] Step 3, followed by multi-dimensional data evaluation: In data quality assessment, the point cloud density gradient of different sections along the pipeline axis is calculated. A gentler density gradient indicates a more uniform point cloud distribution, corresponding to a higher point cloud accuracy evaluation coefficient. The number of missing core parameters in the BIM model is counted, and the BIM parameter completeness evaluation coefficient is obtained using the missing rate formula. A lower missing rate results in a higher completeness coefficient. The number of topological connection errors and contradictory branch relationships in the GIS spatial data is checked, and the topological error rate is calculated. A lower error rate corresponds to a higher GIS topological rationality evaluation coefficient. In timeliness assessment, the data collection timestamp is used as the core basis. Data collected within 3 months is marked as high-timeliness data, 3 months to 1 year as medium-timeliness data, and more than 1 year as low-timeliness data. Different timeliness levels correspond to different timeliness weights. In application scenario assessment, priority is assigned based on pipeline type, with critical pipelines for people's livelihoods such as gas, electricity, and water supply having higher priority than ordinary communication pipelines. The importance of urban areas is also considered, with core business districts, transportation hubs, and areas surrounding hospitals and schools having higher priority than ordinary residential areas and suburbs. Priority and regional level are directly related to the application scenario weight allocation.

[0025] Step 4: Dynamically assign weights based on the multi-dimensional evaluation results: Based on data quality assessment coefficients, timeliness levels, and application scenario priorities, a weight allocation model is established to dynamically adjust the weight ratios of the three types of data for different modeling needs. For example, in the modeling of gas pipelines in core areas, the weight ratio of GIS topology rationality and BIM parameter completeness is increased; if point cloud data was collected recently and has high accuracy, the weight of point cloud data is appropriately increased. The specific weight values ​​of the three types of data are calculated through the weight allocation model to ensure that the weight allocation can adapt to the data characteristics and application scenario requirements, avoiding the impact of the limitations of single data on the fusion effect.

[0026] Step 5, finally perform multi-source data weighted fusion: A pre-defined weighted fusion algorithm is used to integrate pre-processed and weighted point cloud data, BIM model data, and GIS spatial data. During the fusion process, format differences, coordinate deviations, and attribute conflicts between data are eliminated. The geometric morphology information, parametric attribute information, topological relationship information, and spatial distribution information of pipelines are organically combined to generate pipeline basic data with unified benchmarks, complete attributes, clear spatial relationships, and reliable data quality, providing core data support for subsequent modeling processes.

[0027] By collecting multi-source data covering pipeline geometry, attributes, and spatial relationships, and eliminating data heterogeneity through coordinate unification and format standardization preprocessing, and then accurately adapting to data characteristics and application scenarios through multi-dimensional evaluation and dynamic weight allocation, the efficient fusion of multi-source data is finally achieved. This not only solves the problems of incomplete information and insufficient accuracy of single data, but also ensures the relevance and reliability of the fused data through dynamic weight adjustment.

[0028] S200 uses a cylindrical neighborhood analysis method based on the pipeline orientation to dynamically adjust clustering parameters according to the local density of the point cloud, and performs adaptive clustering on the pipeline point cloud to generate a pipeline geometric model. The cylindrical neighborhood is based on the pipeline orientation as the axis, with the horizontal radius proportional to the pipe diameter and the vertical height related to the change in burial depth.

[0029] The S200 execution process specifically includes: Step 1: Extract key information from the pipeline basic data generated by S100, combine the pipeline design centerline coordinates in the BIM model data with the pipeline spatial distribution trajectory in the GIS spatial data, and use point cloud data for auxiliary calibration. By fitting the center trajectory of the dense point cloud area, local abnormal point interference is eliminated, and finally the smooth and continuous actual pipeline route is determined, providing an accurate axis benchmark for cylindrical neighborhood analysis.

[0030] Step 2: Using the determined pipeline route as the axis, extract the pipe diameter value of the corresponding pipe segment from the BIM model data, and set the horizontal radius of the cylindrical neighborhood according to the preset scale coefficient to ensure that the radius can completely cover the pipeline point cloud without including too many redundant background points; obtain the burial depth variation range of the pipe segment from the GIS spatial data, and combine it with the terrain slope data to set the initial value of the vertical height so that the vertical direction can adapt to the burial depth fluctuation and avoid pipeline point cloud omissions due to burial depth changes.

[0031] Step 3: Divide the pipeline into several continuous analysis sections according to its route. The section length is dynamically adjusted according to the pipe diameter; the larger the pipe diameter, the longer the section. Count the number of point clouds in each analysis section, calculate the average distribution density of the point clouds within the section, and record the minimum and maximum spacing between point clouds. The difference between average density and spacing comprehensively characterizes the local density of the point clouds, providing data support for adjusting clustering parameters.

[0032] Step 4: If the local density of the point cloud in a certain analysis section is higher than the preset threshold, it indicates that the point cloud data in that area is dense. The horizontal radius and vertical height of the cylindrical neighborhood should be appropriately reduced to decrease the false inclusion of non-pipeline background points (such as soil impurities and underground obstacle point clouds). If the local density is lower than the preset threshold, it indicates that the point cloud distribution is sparse. The horizontal radius and vertical height should be increased accordingly to ensure that all effective pipeline point clouds can be captured. Simultaneously, the horizontal radius ratio coefficient is optimized based on the pipe material characteristics. The ratio coefficient is reduced for metallic pipelines to improve contour accuracy, while the basic ratio coefficient remains unchanged for non-metallic pipelines. For sections with drastic changes in burial depth, the vertical height is further increased to adapt to the point cloud distribution shift caused by burial depth fluctuations.

[0033] Step 6: Based on the adjusted cylindrical neighborhood parameters, perform segment-by-segment clustering analysis on the pipeline point cloud. Point clouds within the neighborhood are classified as belonging to the same pipeline unit, thus separating the effective pipeline point cloud from the background point cloud. During clustering, the spatial distribution pattern of the point cloud within each neighborhood is monitored in real time. By comparing features such as reflection intensity and coordinate deviation, abnormal point clouds caused by soil impurities or equipment errors are identified and removed, ensuring that the clustered data contains only pure pipeline point clouds. For curved pipeline sections, the number of clustering iterations is increased according to the curvature; the greater the curvature, the more iterations are needed. Simultaneously, the neighborhood adjustment step size is reduced to ensure the clustering accuracy of point clouds in curved areas.

[0034] Step 6: Based on the clustered clean pipeline point cloud, a curve fitting algorithm is used to fit the pipeline's central axis. Combined with the pipe diameter parameters in the BIM model, the pipeline cross-sectional dimensions are determined, constructing a preliminary pipeline geometric model. The model accuracy is verified using point cloud data. The average distance deviation between the model surface and the corresponding point cloud is calculated. If the deviation exceeds a preset threshold, the clustering parameters are readjusted for the deviation area, and clustering is repeated until the model accuracy meets the requirements. Finally, a pipeline geometric model is generated that highly matches the actual pipeline shape, has accurate geometric parameters, and a smooth, continuous surface.

[0035] By accurately determining the pipeline route, dynamically configuring neighborhood parameters, and using adaptive clustering analysis and model optimization, the modeling challenges caused by background noise interference, uneven distribution, and complex pipeline morphology in point cloud data are effectively solved.

[0036] S300 constructs a 3D spatiotemporal graph based on the fused pipeline baseline data, where nodes represent pipeline points and edges represent pipeline segments. It predicts pipeline anomalies through a 3D spatiotemporal graph neural network model and obtains the corresponding prediction results. The 3D spatiotemporal graph neural network model includes a spatial feature extraction layer, a temporal feature extraction layer, and a spatiotemporal information fusion layer. The spatial feature extraction layer uses a graph convolutional network to capture the spatial topological relationships of the pipeline, the temporal feature extraction layer uses an LSTM network to capture the temporal evolution features of the pipeline, and the spatiotemporal information fusion layer dynamically fuses spatial and temporal features through an attention mechanism.

[0037] The S300 execution process specifically includes: Step 1: First, conduct basic data analysis for constructing the 3D spatiotemporal map: From the integrated pipeline basic data generated by S100, the core information of pipe points and pipe segments is accurately extracted: pipe point information includes attributes such as coordinates, burial depth, pipe diameter, material, installation time, and maintenance records; pipe segment information includes connection relationships, length, pressure bearing threshold, and historical fault data. At the same time, the topological relationships in GIS spatial data and the parametric attributes in BIM model are integrated to form the basic dataset for constructing a three-dimensional spatiotemporal map, ensuring that the data covers all dimensions of space, attributes, and time.

[0038] Step 2, then proceed with the construction of the three-dimensional spatiotemporal graph topology: Using pipe points as nodes and pipe segments as edges, a mapping between nodes and edges is established based on the pipeline topology in GIS spatial data: each node corresponds to a unique pipe point, recording its complete attribute information; each edge corresponds to a unique pipe segment, associating the node identifiers at both ends, and binding the physical parameters and operational data of the pipe segment. Using graph structure modeling tools, nodes and edges are organized according to the actual pipeline network connection logic, forming a three-dimensional graph structure containing spatial topological relationships, providing a carrier for spatiotemporal feature extraction.

[0039] Step 3: Next, we will complete the fusion of time series data and graph structure: Time-series data during pipeline operation is collected, including sensor monitoring data such as pressure, flow, and temperature at different time points, as well as time-series information such as historical fault occurrence times and maintenance records. The time-series data is then normalized according to preset time intervals to form a standardized time series. This time series is then associated and bound to corresponding nodes and edges in a 3D graph structure, so that each node and edge contains both spatial attributes and dynamic changes in the time dimension, ultimately constructing an integrated 3D spatiotemporal graph that combines spatial topology and temporal dynamics.

[0040] Subsequently, the 3D spatiotemporal graph neural network model was initialized and configured. For the spatial feature extraction layer, the kernel size, number of layers, and activation function of the graph convolutional network were set, and the network parameters were adjusted according to the complexity of the pipeline network topology to ensure effective capture of the spatial correlation strength and topological dependency between pipeline segments. For the temporal feature extraction layer, the number of hidden layer units, time step size, and dropout ratio of the LSTM network were configured to adapt to the length and fluctuation characteristics of the time series data, ensuring accurate capture of the temporal evolution pattern. For the spatiotemporal information fusion layer, the weight calculation rules of the attention mechanism were set, and the initial value of the fusion ratio of spatial and temporal features was clarified, laying the foundation for dynamic fusion.

[0041] The constructed 3D spatiotemporal map is input into the spatial feature extraction layer. A graph convolutional network is used to perform layer-by-layer convolution operations on the spatial association information between nodes and edges. Centered on nodes, the layer aggregates the attribute information of their neighboring nodes and the connection features of edges, quantifying relationships such as spatial distance between pipe points and the tightness of pipe segment connections. This transforms the spatial topology information into standardized spatial feature vectors, comprehensively representing the spatial distribution and association characteristics of the pipeline network. Bound temporal data is extracted from the nodes and edges of the 3D spatiotemporal map and input into the LSTM network of the temporal feature extraction layer. Through the synergistic effect of the input gate, forget gate, and output gate, effective temporal information is filtered, redundant noise data is forgotten, and the focus is on capturing temporal evolution features such as pressure fluctuation trends, flow change cycles, and fault occurrence patterns. This transforms the temporal data into temporally correlated temporal feature vectors, reflecting the dynamic changes in the pipeline network's operating status.

[0042] Step 4: Achieve dynamic fusion of spatiotemporal features through an attention mechanism: Spatial and temporal feature vectors are input into the spatiotemporal information fusion layer. The attention mechanism first calculates the correlation between the two types of features: key nodes and edges, such as pipe sections in the core urban area and pipe sections with frequent historical failures, are assigned higher spatial feature weights; recent high-time-sensitivity time-series data and data from periods with high failure rates are assigned higher temporal feature weights. Based on the correlation and weight allocation rules, a weighted summation algorithm is used to deeply fuse spatial and temporal features, generating a unified spatiotemporal fusion feature vector that takes into account both spatial topological dependencies and temporal dynamic changes, thus eliminating the limitations of single features.

[0043] Step 5, finally execute the pipeline anomaly prediction and output the results: The spatiotemporal fusion feature vector is input into the prediction layer of the 3D spatiotemporal graph neural network model. Through a fully connected network and classification algorithm, the operational status of the pipeline network is determined: the anomaly type (such as leakage, corrosion, blockage, etc.), the node / edge identifier of the anomaly, and the severity of the anomaly are identified. Simultaneously, the predicted probability value of the anomaly is output. The prediction results are then standardized to form a standardized prediction result containing core anomaly information, providing a direct basis for subsequent visualization and operation and maintenance decisions.

[0044] By constructing a three-dimensional spatiotemporal graph that integrates "space and time," the separation between spatial information and temporal data is broken. Relying on graph convolutional networks and LSTM networks, the spatial topological features and temporal evolution features of the pipeline network are accurately extracted. Then, the attention mechanism is used to achieve dynamic fusion of the two types of features, which effectively improves the comprehensiveness and relevance of feature representation.

[0045] S400 dynamically updates the visualization status of the 3D pipeline model based on the prediction results, marking high-risk pipe sections in red and normal pipe sections in green, and displaying the abnormal propagation path in the 3D model. The calculation of the abnormal propagation path is based on the pipeline network topology and fluid dynamics principles. By simulating the flow direction and speed of fluid in the pipeline network, the path and scope of abnormal propagation are determined. The dynamic update trigger conditions include the predicted abnormal probability exceeding a threshold, the abnormal duration exceeding a preset time, and the abnormal impact range exceeding a preset area.

[0046] The S400 execution process specifically includes: Step 1: First, analyze the prediction results and dynamically update the trigger judgment: Extract the pipeline anomaly prediction results output by S300, and clarify the core information such as the anomaly segment identifier, anomaly type (e.g., leakage, corrosion, blockage), predicted anomaly probability, estimated duration, and initial impact range. Verify each of the preset dynamic update trigger conditions: if the predicted anomaly probability exceeds a set threshold, or the estimated anomaly duration reaches a preset time standard, or the initial impact range covers a preset area, the update trigger requirements are met, and the 3D pipeline model visualization status update process is initiated; if none of the trigger conditions are met, the current model visualization status is maintained, and only the prediction results are recorded for subsequent periodic checks.

[0047] Step 2, followed by precise calculation of the abnormal propagation path: The complete pipeline topology is extracted from the S100 fused pipeline basic data, clarifying key spatial information such as the connection relationships between pipe segments, the specific locations and real-time on / off status of valves, and pipeline interface types. Based on fluid dynamics principles and combined with physical parameters such as pipe segment diameter, length, and inner wall roughness, the fluid flow resistance and basic flow velocity of each pipe segment are calculated to establish a fluid transmission correlation model between pipe segments. The actual flow direction of fluid in the pipeline network is simulated, starting from the anomaly initiation pipe segment and deducing the propagation trajectory along the pipeline topology connection relationships. If a closed valve is encountered, its pipe segment is identified as a propagation obstruction point, and a new detour propagation path is planned. Based on the actual operating pressure and flow rate data of the pipe segment, the anomaly propagation speed is dynamically adjusted; the higher the pressure and the greater the flow rate, the higher the propagation speed. At the same time, combined with the pipe segment spacing and fluid diffusion law, the specific range of the anomaly's impact is accurately calculated, forming complete anomaly propagation path data.

[0048] Step 3, Next, configure the visual status update: A clear color-mapping rule for pipeline segment risk levels is established: pipeline segments with a predicted anomaly probability exceeding the high-risk threshold are marked in red, while normally operating segments without anomaly risk are displayed in green. If medium- or low-risk pipeline segments exist (unless explicitly required, the core rule applies, focusing on the two core states of high risk and normal), corresponding color identifiers can be added according to extended rules, but the clarity of the core color distinction must be prioritized. For anomaly propagation paths, visualization parameters are set: the path trajectory is presented as dynamic lines, with line thickness positively correlated with anomaly severity (thicker lines for higher severity) and color depth corresponding to propagation speed (darker colors for faster speed), ensuring the path characteristics are intuitively identifiable. Simultaneously, additional information tags are configured for high-risk pipeline segments, including key information such as anomaly type, predicted anomaly occurrence time, and impact range boundaries, providing convenience for subsequent review.

[0049] Step 4, then perform 3D pipeline model visualization rendering: The 3D modeling and rendering engine is invoked, and the 3D pipeline model is updated in real time based on the configured color mapping rules and path display parameters: the model color of high-risk pipe sections is switched to red, while normal pipe sections remain green. Dynamic lines of the anomaly propagation path are simultaneously drawn, ensuring that the lines accurately match the pipeline network topology without misalignment or deviation. During rendering, terrain data and underground obstacle distribution information are combined to optimize and adjust the anomaly propagation path lines, automatically avoiding underground structures, other pipelines, and other obstacles, ensuring that the path display is clear and conforms to the actual geographical environment. At the same time, model rendering efficiency is optimized to avoid screen stuttering due to state updates, ensuring the smoothness and stability of the visualization effect.

[0050] Step 5: Finally, optimize the visualization results and adapt the interaction: Establish a visual result verification mechanism to check the accuracy of pipe segment color markings, the consistency of anomaly propagation paths, and the completeness of additional labels. Correct any deviations promptly. Support user interaction: when a user clicks on a red high-risk pipe segment, a details window pops up displaying information such as anomaly type, prediction basis, impact range, and historical fault records. Provide model zooming, panning, and rotation functions to allow users to view anomalies from different perspectives. Establish a visual update log to record the time, triggering conditions, update content, and operator of each update, providing complete data support for subsequent source tracing and auditing.

[0051] By transforming abstract anomaly predictions into concrete and easily understandable 3D model visualizations, the red high-risk pipe section markers and dynamic anomaly propagation paths enable maintenance personnel to quickly locate core risk areas and grasp anomaly spread trends, providing direct visual guidance for emergency response and maintenance planning. Simultaneously, strict adherence to trigger conditions for dynamic updates ensures the timeliness and accuracy of the visualized information, avoiding resource waste from ineffective updates and guaranteeing timely risk warnings. This effectively connects anomaly prediction with actual maintenance decisions, providing intuitive and efficient visualization support for refined management of the entire pipeline lifecycle, significantly improving the response speed and scientific rigor of pipeline risk management.

[0052] Based on the above solution, taking the gas pipeline network in the core business district of a provincial capital city as an example, this area has dense pipelines and is surrounded by high-rise buildings and transportation hubs, requiring extremely high safety and stability of the pipeline network. It necessitates meticulous management throughout the entire lifecycle to mitigate risks such as leakage and corrosion. During the planning and design phase, the S100 multi-source data fusion mechanism of this solution is adopted to collect directional 3D laser scanning point cloud data, gas pipeline network BIM models (including parameters such as pipe diameter, material, and design pressure), and GIS spatial data (including pipeline topology and burial depth distribution). After preprocessing with unified coordinates and standardized formats, and considering the application scenario priority and data timeliness of the core business district, the weighting of BIM parameter integrity and GIS topology rationality is dynamically increased, resulting in the generation of high-precision basic data. By using S200 cylindrical neighborhood adaptive clustering modeling, with the actual pipeline route as the axis, neighborhood parameters are adjusted according to the local density of the point cloud, and background point clouds such as soil impurities are removed to generate a geometric model that highly matches the actual pipeline shape. This accurately identifies potential conflict points between the planned route and power pipelines, providing data support for pipeline layout optimization and reducing the risk of later operation and maintenance from the source.

[0053] During routine maintenance, relying on the S300 3D spatiotemporal graph construction and anomaly prediction mechanism, pipeline points and sections are transformed into graph structure nodes and edges, bound to real-time sensor time-series data such as pressure and flow, as well as historical maintenance records, forming an integrated "space + time" 3D spatiotemporal graph. A graph convolutional network captures the spatial relationships between core pipeline sections and surrounding pipelines, an LSTM network analyzes the pressure fluctuation trends of high-timeliness data from the past three months, and an attention mechanism focuses on strengthening the feature fusion weights of key pipeline sections around hospitals and shopping malls, enabling accurate prediction of anomalies such as corrosion and leakage. In one quarter, the model warned of an abnormal probability exceeding the standard for a certain main pipeline section. Maintenance personnel viewed the high-risk pipeline section marked in red through the S400 visualization interface, combined with the dynamically displayed anomaly propagation path (based on fluid dynamics simulation, considering valve opening and closing status correction), quickly located the source of risk, and carried out anti-corrosion treatment in advance, avoiding a large-scale gas outage accident, demonstrating the foresight and targeted nature of refined maintenance.

[0054] During the emergency response and long-term optimization phases, when a branch section of the pipeline leaks due to an accidental construction, the S400 system immediately triggers a visual update. The leaking section and affected area are marked in red, and dynamic lines clearly show the trajectory and speed of the anomaly's propagation to the surrounding commercial area. The emergency response team can use the model's interactive function to view the historical maintenance records and alternative detour sections for that section, quickly formulating valve shut-off plans and personnel evacuation routes, significantly shortening emergency response time. Simultaneously, the multi-scale model generated by the plan supports viewing a lightweight version on mobile devices at the emergency site, and also allows the command center to retrieve a high-precision model for root cause analysis. Data after each response is automatically synchronized to a 3D spatiotemporal map, and through iterative optimization of the S300 model, the feature extraction weights for that type of pipeline section are updated, improving the accuracy of subsequent anomaly predictions. The entire process forms a closed-loop management system of "planning-operation-emergency-optimization," fully meeting the core requirements of refined management of the entire lifecycle of the core area's gas pipeline network for data accuracy, risk prediction, and efficient response.

[0055] In this embodiment of the application, the process of dynamically allocating weights for multi-source data includes: Step 1, Identification and Quantification of Underground Environmental Impact Factors: Comprehensive underground environmental data of the pipeline area is collected, specifying soil type (such as sandy soil, clay soil, silty soil, etc.), specific groundwater depth (divided into shallow and deep grades according to preset intervals), and intensity of surrounding construction activities (quantified by construction frequency and scope). Based on industry-preset quantitative standards, the three types of environmental factors are converted into data reliability correction coefficients. Among them, soils with poor stability such as sandy soil, groundwater with a depth less than the preset threshold, and areas with weekly construction frequency exceeding the preset number are all assigned lower correction coefficients, which directly reflect the impact of the environment on data reliability.

[0056] Step 2, Weighting of Equipment Accuracy and Tool Maturity: The system queries the industry accuracy level certification of 3D laser scanning equipment (Level 1, Level 2, and Level 3), and assigns point cloud data quality weights according to the corresponding rules for each level, with Level 1 equipment having the highest weight and Level 3 the lowest. At the same time, it verifies the industry maturity rating of BIM modeling tools (such as assessments based on tool parameter compatibility, industry application breadth, and update frequency). The higher the maturity level, the higher the corresponding weight ratio for BIM parameter integrity, ensuring that the performance of data acquisition tools is directly linked to weight allocation.

[0057] Step 3, Initial weight allocation scheme construction: Historical modeling data of the same type of pipeline and the same area are retrieved to establish an initial weight allocation model. The model includes four input dimensions: data quality, timeliness, application scenario, and environmental correction. The point cloud accuracy evaluation coefficient, BIM parameter completeness evaluation coefficient, GIS topology rationality evaluation coefficient (quality evaluation result), high / medium / low timeliness level corresponding to the data collection time (timeliness information), pipeline type priority and regional importance level (application scenario requirements), and environmental correction coefficient quantified in step 1 are input into the initial model one by one. The initial weight ratio scheme of the three types of data (point cloud, BIM, GIS) is calculated and output through the built-in algorithm of the model.

[0058] Step 4, Cross-validation and Weight Iterative Optimization: Cross-validation is employed to randomly divide the multi-source data after the initial weighting scheme is fused into training and validation sets. Modeling accuracy (such as geometric deviation and parameter matching degree) is calculated by modeling on the training set and verifying on the validation set. If the accuracy does not meet the preset standard, the weight ratios of the three types of data are adjusted in a gradient manner according to the principle of "prioritizing quality weight adjustment, supplementing with environmental correction coefficient, and dynamically fine-tuning timeliness weight" (e.g., the adjustment range of quality weight does not exceed 10% each time, and the adjustment range of environmental correction coefficient does not exceed 5%). The cross-validation and weight adjustment process is repeated until the modeling accuracy meets the preset target, and the final weight allocation scheme is determined.

[0059] The process of adapting equipment and tools to performance, building initial models, and iterating cross-validation enables precise and dynamic allocation of weights for multi-source data, further improving the adaptability and reliability of multi-source data fusion.

[0060] In this embodiment of the application, the error correction and multi-scale adaptation process after generating the pipeline geometry model includes: Step 1: Construct an error detection index system and set acceptable thresholds: Define the specific definitions of geometric deviation, topological integrity, and parameter matching degree: geometric deviation refers to the difference in spatial distance between the surface of the pipeline geometric model and the original point cloud data; topological integrity refers to the correct proportion of the connection relationship between pipe segments and pipe points in the model; and parameter matching degree refers to the degree of consistency between the model parameters and the BIM design parameters. Based on industry modeling accuracy standards and actual application needs, set qualified thresholds for each indicator (such as geometric deviation ≤ 5 mm, topological integrity ≥ 99%, and parameter matching degree ≥ 98%) to form a standardized error detection basis.

[0061] Step 2: Verify the accuracy of the point cloud data and adjust the clustering parameters: The Euclidean distance method is used to calculate the average distance deviation between the surface of the geometric model and the corresponding original point cloud. The regions with deviations exceeding the threshold are located segment by segment. For these regions, the cylindrical neighborhood clustering process is re-executed, and the neighborhood parameters such as the horizontal radius and vertical height are adjusted by a gradient of 5% to 10% to reduce the clustering range of the deviation region or expand the effective point cloud capture range. After re-clustering, the average distance deviation is calculated again until the deviation meets the qualified threshold requirements.

[0062] Step 3: Correct the burial depth parameters based on ground settlement data: Historical ground settlement data for the past 5-10 years in the area where the pipeline is located were extracted from GIS spatial data. Weighting rules were set according to the time series (0.6 for data in the past 3 years and 0.4 for data in the past 3-10 years), and the cumulative settlement value was calculated by weighted summation. Based on the correction formula of "original burial depth coordinates + cumulative settlement value", the burial depth coordinates of the corresponding pipe section in the geometric model were dynamically corrected point by point to ensure that the burial depth of the model is consistent with the current actual underground pipeline location.

[0063] Step 4: Generate multi-scale models and support on-demand calling: Based on application scenarios, the core requirements for three accuracy levels are defined: High-precision operation and maintenance level models retain detailed features such as pipeline wall thickness, interface threads, and weld locations, adapting to daily maintenance and fault location scenarios; medium-precision planning level models simplify non-critical details (such as deleting decorative interface structures) while retaining core parameters such as pipe diameter, length, and connection relationships, adapting to pipeline network expansion and route optimization planning scenarios; lightweight display level models retain only the pipeline centerline, pipe diameter, and core topological relationships, compressing model data volume through polygon simplification algorithms, adapting to reporting and display scenarios and mobile terminal quick viewing scenarios; a model call trigger mechanism is established to automatically match the model of the corresponding accuracy level according to user operation commands or logged-in roles, realizing on-demand loading.

[0064] Through multi-dimensional error detection, dynamic correction of settlement data, and multi-scale model adaptation, the spatial accuracy and parameter accuracy of the pipeline geometric model are ensured, and flexible application in different scenarios is realized. This provides accurate and efficient model support for the diverse needs of operation, maintenance, planning, and display in the entire life cycle management of pipeline networks.

[0065] In this embodiment of the application, the process of dynamically fusing spatial and temporal features through an attention mechanism in the spatiotemporal information fusion layer includes: Step 1: Filter spatial features, focus on key aspects, and assign weights: Pipe segment attributes and spatial information are extracted from the three-dimensional spatiotemporal map to accurately identify pipe segments in the core urban area, pipe segments with frequent historical failures, pipe segments connecting critical infrastructure (such as hospitals and water plants), and large-diameter main pipe segments. These pipe segments are marked as key spatial features. According to preset rules, higher spatial feature weights are assigned to key pipe segments (e.g., the basic weight of non-key pipe segments is 1, and the weight of key pipe segments is increased to 1.5-2.0) to strengthen the representation of key spatial features.

[0066] Step 2, Time-sensitivity classification and weight assignment of time features: Pipeline operation sequence data are collected and classified according to the data collection time: data within the last three months is marked as high-time data and given the highest time weight; data from three months to one year is marked as medium-time data and given a medium weight; data older than one year is marked as low-time data and given the lowest weight. The weight difference highlights the impact of recent data on feature fusion.

[0067] Step 3: Calculate the spatiotemporal feature correlation and strengthen key coupling relationships: The attention mechanism calculates the correlation between spatial and temporal features through a built-in algorithm, focusing on the spatiotemporal coupling relationship between high-time data and key spatial segments (such as pressure change data of core area segments in the past three months), and automatically increases the feature fusion weight of such "high-time data + key spatial data" combinations to ensure that key spatiotemporal information is given priority representation.

[0068] Step 4: Dynamically adjust the fusion strategy based on the operational status: The system monitors the pipeline's operating status in real time. When the sensor detects real-time abnormal signals such as pressure fluctuations exceeding the preset range or abnormal flow changes in a certain pipeline section, it quickly locates the abnormal pipeline section. It temporarily increases the weight of the time feature of the pipeline section (e.g., increases the weight by 0.3-0.5) to strengthen the contribution of the time feature of the abnormal period to the fusion result and improve the sensitivity of anomaly identification.

[0069] Step 5, Feature normalization processing: The spatiotemporal fusion features after weight adjustment are normalized. A standardization method is used to map the feature values ​​of different dimensions (such as spatial distance, pressure value, and time span) to the same numerical range to eliminate the interference caused by the difference in dimensions. A spatiotemporal fusion feature vector with unified dimension and numerical standardization is generated and directly input into the pipeline network anomaly prediction module for subsequent calculation.

[0070] By accurately selecting key spatial features, assigning time weights hierarchically, strengthening key spatiotemporal coupling relationships, dynamically adapting to operational status, and normalizing the process, targeted fusion of spatiotemporal features was achieved, enhancing the representational ability and effectiveness of feature vectors.

[0071] In this embodiment of the application, the iterative optimization process of the three-dimensional spatiotemporal graph neural network model includes: Step 1: Collect deviation data and establish evaluation indicators: The model outputs anomaly prediction results (including anomaly type, occurrence time, and impact range) and the actual anomaly situation in the pipeline network are recorded synchronously, and the deviation data between the two is calculated. Three types of deviation evaluation indicators are defined: anomaly type identification accuracy (the proportion of correctly identified anomalies to the total number of anomalies), anomaly occurrence time prediction error (the difference between the predicted time and the actual occurrence time), and impact range estimation deviation (the proportion of the difference between the predicted number of affected pipe sections and the actual number of affected sections), forming a standardized model performance evaluation system.

[0072] Step 2: Screen for frequently misjudged pipe sections and optimize model parameters: Statistical analysis of the deviation data was performed to screen out high-frequency misjudged pipe segment types (such as small-diameter branch pipe segments in old residential areas) with high anomaly misjudgment rates, large time errors, or excessive range prediction deviations. Spatial topological features (connection method with the main pipe), historical operating sequence data (pressure / flow data for the past 1-2 years), and environmental correlation data (soil type, surrounding construction conditions) of such pipe segments were extracted to construct a special optimization dataset. Based on this dataset, the model parameters were adjusted: the extraction weights of key features such as pipe segment material and operating years were increased, the size of the graph convolution kernel of the spatial feature extraction layer was optimized (e.g., from 3×3 to 5×5 to improve the ability to capture local features), and the number of LSTM units in the temporal feature extraction layer was adjusted (e.g., from 64 to 128 to strengthen the learning of temporal patterns).

[0073] Step 3: Introduce a transfer learning mechanism to adapt to the new modeling region: For newly added pipeline network modeling areas (without sufficient historical data), retrieve the parameters of mature regional pipeline network models of the same type and environment as initial parameters; collect a small amount of measured data (such as 3-6 months of operating data and 1-2 abnormal records) in the new modeling area, and make local fine-tuning of the migration parameters to adapt to the pipeline network characteristics of the new modeling area (such as pipeline layout and differences in operating pressure) and shorten the model training cycle.

[0074] Step 4, Regular performance testing and iterative iteration: The performance of the iterated model is tested at a preset cycle (e.g., quarterly). The test data is input into the model, and three types of deviation evaluation indicators are calculated. If all indicators meet the preset standards (e.g., anomaly type identification accuracy ≥ 95%, time prediction error ≤ 2 hours, range prediction deviation ≤ 10%), the optimization is completed. If any indicator fails to meet the standard, return to step 1 to collect deviation data again, repeat the parameter adjustment and migration fine-tuning process until the model performance meets the requirements.

[0075] By optimizing parameters driven by bias data, adapting to new regions through transfer learning, and periodically verifying performance, we continuously improve the anomaly prediction accuracy and scene adaptability of the 3D spatiotemporal neural network model, ensuring the long-term stability of the model to support the prediction needs of refined management of the entire pipeline network lifecycle.

[0076] In this embodiment of the application, the process of early warning linkage for high-risk pipeline sections includes: Step 1: Extract core information of high-risk pipeline sections and correlate it with surrounding data: From the anomaly prediction results output by S300, key spatial information such as the three-dimensional coordinates and burial depth of high-risk pipe sections are accurately extracted; the GIS spatial data fused by S100 is called to associate the distribution of pipelines around the high-risk pipe section (including the type, diameter, connection relationship and distance of adjacent pipe sections), clarify the degree of correlation between the surrounding pipelines and the high-risk pipe section, and provide data support for subsequent risk assessment.

[0077] Step 2: Generate precise early warning information: Based on the extracted high-risk pipeline information and related surrounding pipeline data, core information such as anomaly type (e.g., leakage, corrosion), predicted anomaly probability, and associated risk level of adjacent pipelines (e.g., direct or indirect association) is integrated to generate standardized early warning information. The information should be concise and clear, including both the specific location description of the high-risk pipeline (e.g., "gas pipeline 3 meters underground at the intersection of XX Road and XX Street") and the range that may be affected by adjacent pipelines, so that maintenance personnel can quickly grasp key information.

[0078] Step 3: Implement the linkage between early warning information and 3D model visualization: The generated early warning information is bound to the S400's 3D pipeline model visualization system, and the early warning list is displayed in a prominent position on the model interface (such as the side or top of the screen). A linkage trigger mechanism is set up so that when maintenance personnel click on an early warning, the 3D model automatically scales and pans to the corresponding high-risk pipeline segment, and highlights the pipeline segment and its associated adjacent segments, intuitively presenting the core risk area and the scope of impact without the need for manual location.

[0079] By accurately extracting information on high-risk pipeline sections and generating targeted early warning prompts by associating them with data from surrounding pipelines, and by achieving visualized linkage and positioning with 3D models, the efficiency and accuracy of risk information acquisition by operation and maintenance personnel have been greatly improved. This provides convenient support for the rapid investigation and early handling of high-risk pipeline sections, and further strengthens the early warning and response capabilities of refined management throughout the entire life cycle of the pipeline network.

[0080] This application discloses a three-dimensional modeling system for underground pipelines, referring to... Figure 2 ,include; The multi-source data acquisition and weighted fusion module acquires point cloud data, BIM model data, and GIS spatial data. The point cloud data is acquired through 3D laser scanning, the BIM model data includes parametric attribute information of pipelines, and the GIS spatial data includes topological relationships and spatial distribution information of pipelines. The module dynamically assigns weights to the multi-source data based on data quality, timeliness, and application scenario to form fused basic pipeline data. Data quality assessment is based on point cloud accuracy, BIM parameter completeness, and GIS topological rationality; timeliness assessment is based on data acquisition time; application scenario assessment is based on pipeline type and importance to the urban area; point cloud accuracy assessment coefficient is calculated based on point cloud density gradient; BIM parameter completeness assessment coefficient is calculated based on parameter missing rate; and GIS topological rationality assessment coefficient is calculated based on topology error rate. The clustering analysis and geometric modeling module uses a cylindrical neighborhood analysis method based on the pipeline direction to dynamically adjust the clustering parameters according to the local density of the point cloud, and performs adaptive clustering on the pipeline point cloud to generate a pipeline geometric model. The cylindrical neighborhood is based on the pipeline direction as the axis, with the horizontal radius proportional to the pipe diameter and the vertical height related to the change in burial depth. The spatiotemporal graph construction and anomaly prediction module constructs a 3D spatiotemporal graph based on the fused pipeline baseline data, where nodes represent pipeline points and edges represent pipeline segments. It then predicts pipeline anomalies using a 3D spatiotemporal graph neural network model, obtaining the corresponding prediction results. This model includes a spatial feature extraction layer, a temporal feature extraction layer, and a spatiotemporal information fusion layer. The spatial feature extraction layer uses a graph convolutional network to capture the spatial topological relationships of the pipelines, the temporal feature extraction layer uses an LSTM network to capture the temporal evolution characteristics of the pipelines, and the spatiotemporal information fusion layer dynamically fuses spatial and temporal features through an attention mechanism. The visualization update and path display module dynamically updates the visualization status of the 3D pipeline model based on the prediction results, marking high-risk pipe sections in red and displaying normal pipe sections in green, and displaying the abnormal propagation path in the 3D model. The calculation of the abnormal propagation path is based on the pipeline network topology and fluid dynamics principles. By simulating the flow direction and velocity of fluid in the pipeline network, the path and impact range of abnormal propagation are determined. The dynamic update is triggered by the following conditions: the predicted abnormal probability exceeds a threshold, the abnormal duration exceeds a preset time, and the abnormal impact range exceeds a preset area.

[0081] This application also discloses a three-dimensional modeling system for underground pipelines, including a processor, wherein the processor runs a program of any one of the above-described three-dimensional modeling methods for underground pipelines.

[0082] This application also discloses a storage medium storing a program for the three-dimensional modeling method of underground pipelines described in any one of the above embodiments.

[0083] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for three-dimensional modeling of underground pipelines, characterized in that, include: The process involves acquiring point cloud data, BIM model data, and GIS spatial data. Point cloud data is obtained through 3D laser scanning. BIM model data includes parametric attribute information of pipelines, and GIS spatial data includes topological relationships and spatial distribution information of pipelines. Weights are dynamically allocated from multiple data sources based on data quality, timeliness, and application scenario to form fused pipeline baseline data. Data quality assessment is based on point cloud accuracy, BIM parameter completeness, and GIS topological rationality; timeliness assessment is based on data acquisition time; application scenario assessment is based on pipeline type and importance to urban areas; point cloud accuracy assessment coefficient is calculated based on point cloud density gradient; BIM parameter completeness assessment coefficient is calculated based on parameter missing rate; and GIS topological rationality assessment coefficient is calculated based on topology error rate. Based on the pipeline orientation, a cylindrical neighborhood analysis method is used to dynamically adjust the clustering parameters according to the local density of the point cloud, and adaptively cluster the pipeline point cloud to generate a pipeline geometric model. The cylindrical neighborhood is based on the pipeline orientation as the axis, with the horizontal radius proportional to the pipe diameter and the vertical height related to the change in burial depth. A 3D spatiotemporal graph is constructed based on the fused pipeline baseline data, where nodes represent pipeline points and edges represent pipeline segments. Pipeline anomalies are predicted using a 3D spatiotemporal graph neural network model, and the corresponding prediction results are obtained. The 3D spatiotemporal graph neural network model includes a spatial feature extraction layer, a temporal feature extraction layer, and a spatiotemporal information fusion layer. The spatial feature extraction layer uses a graph convolutional network to capture the spatial topological relationships of the pipeline, the temporal feature extraction layer uses an LSTM network to capture the temporal evolution features of the pipeline, and the spatiotemporal information fusion layer dynamically fuses spatial and temporal features through an attention mechanism. The visualization status of the 3D pipeline model is dynamically updated based on the prediction results. High-risk pipeline sections are marked in red, and normal pipeline sections are displayed in green. The abnormal propagation path is displayed in the 3D model. The calculation of the abnormal propagation path is based on the pipeline network topology and fluid dynamics principles. By simulating the flow direction and velocity of fluid in the pipeline network, the path and scope of abnormal propagation are determined. The dynamic update is triggered by the following conditions: the predicted abnormal probability exceeds a threshold, the abnormal duration exceeds a preset time, and the abnormal impact range exceeds a preset area.

2. The three-dimensional modeling method for underground pipelines according to claim 1, characterized in that, The process of dynamically allocating weights from multiple data sources includes: The underground environmental impact factors are introduced, including soil type, groundwater depth, and intensity of surrounding construction activities in the pipeline area. These underground environmental impact factors are quantified into correction coefficients for data reliability. The correction coefficients are lower for areas with poor soil stability, shallow groundwater depth, and frequent construction. The quality assessment weights are adjusted based on the accuracy level of the data acquisition equipment. The accuracy level of 3D laser scanning equipment is divided into three levels according to industry standards, with higher levels corresponding to higher point cloud data quality weights. The maturity level of BIM modeling tools corresponds to the adjustment of parameter integrity weights. An initial weight allocation model is established based on historical modeling data. The inputs include the quality assessment results of the current data, timeliness information, application scenario requirements, and correction coefficients. The output is the initial weight allocation scheme. The modeling accuracy after initial weight fusion is evaluated by cross-validation. If the accuracy does not meet the preset target, the weight ratio is iteratively optimized until the modeling accuracy meets the requirements. In the iterative optimization of the weight ratio, quality weight is given priority, environmental correction is secondary, and timeliness is dynamically fine-tuned.

3. The three-dimensional modeling method for underground pipelines according to claim 1, characterized in that, The error correction and multi-scale adaptation process after generating the pipeline geometry model includes: Construct a geometric model error detection index system, which includes geometric deviation, topological integrity, and parameter matching degree, and set the qualified threshold for each index; The accuracy of the geometric model is verified by reverse verification of point cloud data. The average distance deviation between the surface of the geometric model and the corresponding point cloud is calculated. If it exceeds the threshold, cylindrical neighborhood clustering is re-executed for the deviation area, and the neighborhood parameters are adjusted. The model's burial depth parameters are corrected by combining ground settlement data. The historical ground settlement of the pipeline area is extracted from GIS spatial data, and the cumulative settlement value is calculated by weighting according to the time series. The burial depth coordinates of the geometric model are then dynamically corrected. Generate multi-scale geometric models and set accuracy levels according to application scenario requirements. Accuracy levels include high-precision operation and maintenance level, medium-precision planning level, and lightweight display level. High-precision models retain pipeline detail features, while lightweight models simplify non-critical geometric information and support on-demand access.

4. The three-dimensional modeling method for underground pipelines according to claim 1, characterized in that, In the process of dynamically fusing spatial and temporal features through an attention mechanism in the spatiotemporal information fusion layer, the following are included: The spatial features that need to be focused on are selected. These include pipe sections in the core urban area, pipe sections with frequent historical failures, pipe sections connecting critical infrastructure, and large-diameter trunk pipe sections. Higher spatial feature weights are assigned to the pipe sections corresponding to the spatial features that need to be focused on. Time characteristics are classified into timeliness levels: pipeline operation data within the last three months are marked as high-timeliness data, data from three months to one year are marked as medium-timeliness data, and data from more than one year are marked as low-timeliness data. Different time weights are assigned to different timeliness levels. By calculating the correlation between spatial and temporal features through an attention mechanism, we focus on analyzing the spatiotemporal coupling relationship of key pipe sections corresponding to high-timeliness data, thereby improving the feature fusion weight of key pipe sections corresponding to high-timeliness data. By dynamically adjusting the fusion strategy based on the pipeline operation status, when a real-time abnormal signal corresponding to pressure fluctuation or flow abnormality occurs in a pipeline segment, the corresponding abnormal pipeline is identified, and the time feature weight of the abnormal pipeline segment is temporarily increased. The fused spatiotemporal features are normalized to eliminate the dimensional differences between features of different dimensions, generating a fused feature vector of a unified dimension, which is then input into the anomaly prediction module.

5. The three-dimensional modeling method for underground pipelines according to claim 1, characterized in that, The iterative optimization process of the three-dimensional spatiotemporal graph neural network model includes: Collect deviation data between the prediction results of the three-dimensional spatiotemporal neural network model and the actual pipeline network anomalies, and establish deviation evaluation indicators, including the anomaly type identification accuracy, anomaly occurrence time prediction error, and impact range prediction deviation. Filter the pipe segment types that are frequently misjudged in the deviation data, extract the spatial topological features, historical runtime sequence data and environmental correlation data of the frequently misjudged pipe segment types to form a special optimization dataset; adjust the model parameters based on the special optimization dataset, increase the extraction weight of key features, optimize the graph convolution kernel size of the spatial feature extraction layer, and adjust the number of LSTM units in the temporal feature extraction layer. A transfer learning mechanism is introduced to transfer the pipeline network model parameters of mature areas to new modeling areas, and fine-tune them by combining a small amount of measured data from the new modeling areas. The performance of the iterated 3D spatiotemporal neural network model is tested regularly. If all evaluation indicators meet the preset standards, the optimization is completed; otherwise, iterative optimization continues.

6. The three-dimensional modeling method for underground pipelines according to claim 1, characterized in that, The process of early warning and coordination for high-risk pipeline sections includes: Based on the anomaly prediction results, the coordinates and burial depth information of high-risk pipe sections are extracted and associated with the distribution of surrounding pipelines in GIS spatial data; and early warning information is generated to clarify the location of high-risk pipe sections and the associated risks of adjacent pipe sections; the early warning information is linked with the visualization status of the 3D model, and clicking on the early warning information directly locates the corresponding pipe section.

7. A three-dimensional modeling system for underground pipelines, characterized in that, include: The multi-source data acquisition and weighted fusion module acquires point cloud data, BIM model data, and GIS spatial data. The point cloud data is acquired through 3D laser scanning, the BIM model data includes parametric attribute information of pipelines, and the GIS spatial data includes topological relationships and spatial distribution information of pipelines. The module dynamically assigns weights to the multi-source data based on data quality, timeliness, and application scenario to form fused basic pipeline data. Data quality assessment is based on point cloud accuracy, BIM parameter completeness, and GIS topological rationality; timeliness assessment is based on data acquisition time; application scenario assessment is based on pipeline type and importance to the urban area; point cloud accuracy assessment coefficient is calculated based on point cloud density gradient; BIM parameter completeness assessment coefficient is calculated based on parameter missing rate; and GIS topological rationality assessment coefficient is calculated based on topology error rate. The clustering analysis and geometric modeling module uses a cylindrical neighborhood analysis method based on the pipeline direction to dynamically adjust the clustering parameters according to the local density of the point cloud, and performs adaptive clustering on the pipeline point cloud to generate a pipeline geometric model. The cylindrical neighborhood is based on the pipeline direction as the axis, with the horizontal radius proportional to the pipe diameter and the vertical height related to the change in burial depth. The spatiotemporal graph construction and anomaly prediction module constructs a 3D spatiotemporal graph based on the fused pipeline baseline data, where nodes represent pipeline points and edges represent pipeline segments. It then predicts pipeline anomalies using a 3D spatiotemporal graph neural network model, obtaining the corresponding prediction results. This model includes a spatial feature extraction layer, a temporal feature extraction layer, and a spatiotemporal information fusion layer. The spatial feature extraction layer uses a graph convolutional network to capture the spatial topological relationships of the pipelines, the temporal feature extraction layer uses an LSTM network to capture the temporal evolution characteristics of the pipelines, and the spatiotemporal information fusion layer dynamically fuses spatial and temporal features through an attention mechanism. The visualization update and path display module dynamically updates the visualization status of the 3D pipeline model based on the prediction results, marking high-risk pipe sections in red and displaying normal pipe sections in green, and displaying the abnormal propagation path in the 3D model. The calculation of the abnormal propagation path is based on the pipeline network topology and fluid dynamics principles. By simulating the flow direction and velocity of fluid in the pipeline network, the path and impact range of abnormal propagation are determined. The dynamic update is triggered by the following conditions: the predicted abnormal probability exceeds a threshold, the abnormal duration exceeds a preset time, and the abnormal impact range exceeds a preset area.

8. A three-dimensional modeling system for underground pipelines, characterized in that, Includes a processor, wherein the processor runs a program for the three-dimensional modeling method for underground pipelines as described in any one of claims 1-6.

9. A storage medium, characterized in that, A program storing the three-dimensional modeling method for underground pipelines as described in any one of claims 1-6.

Images