Pipeline network space topology relationship analysis method and system based on graph neural network

Abstract

The application relates to the technical field of pipe network data processing, and discloses a pipeline network spatial topological relation analysis method and system based on a graph neural network, which comprises the following steps: collecting pipe section objects, pipeline nodes and constraint objects. A spatial topological index structure is constructed, which comprises a spatial block index, a topological adjacency index and a standard constraint index. A constraint template is selected according to a pipeline category and a topological relation type to determine candidate object pairs and calculate topological indexes, and finally, a preliminary topological relation classification result is obtained. The relation classification input feature is input into a relation classification model, feature transmission and aggregation are carried out on a graph structure formed by the pipe section objects, the pipeline nodes and the constraint objects, and a topological relation classification result is output. Version identifiers are set for spatial blocks and candidate object pairs, and a spatial topological relation analysis result of the pipeline network is determined according to pipeline network data update conditions. The application improves the accuracy of topological rule violation identification and reduces the consumption of computing resources.

Description

Technical Field

[0001] This invention relates to the field of pipeline network data processing technology, and more specifically, to a system and method for analyzing the spatial topology of pipeline networks based on graph neural networks. Background Technology

[0002] With the advancement of urban underground space development and integrated utility tunnel construction, various pipelines such as water supply pipelines, drainage pipelines, gas pipelines, power cables, and communication cables are densely intertwined underground in cities. The safety and standardization of pipeline spatial layout directly affect operational safety and maintenance costs. In engineering practice, geographic information systems and building information modeling platforms are typically used to centrally manage pipeline spatial data. 3D modeling and visualization techniques are used to display pipeline location relationships, and pipeline collision checks, clearance verification, and topological connectivity analysis are conducted during the planning and design, completion acceptance, and operation and maintenance management phases.

[0003] In existing technologies, common practices often rely on full network traversal and simple rule engines to perform distance calculations and specification comparisons on the 3D pipeline model one by one. Even if some solutions introduce rule bases and incremental update mechanisms, they still mainly focus on geometric collisions and local quality checks, lacking multi-layer spatial topology indexes for large-scale pipeline networks and constraint template management that combines topology relationship types and pipeline categories. For example, patent CN115795768B provides a method for 3D pipeline modeling and data updating that takes into account the physical form of the pipeline. It checks the topology relationships of pipeline points and pipelines through a quality check rule engine and supports range-level incremental updates. However, it does not make specific designs for the collaborative organization of spatial block indexes, topology adjacency indexes, and specification constraint indexes, or for a local recalculation mechanism based on version identifiers. It also does not perform consistency scoring on the topology relationships of candidate object pairs, making it difficult to balance the computational efficiency of large-scale pipeline networks and the recognition accuracy of complex topology relationships.

[0004] Therefore, it is necessary to design a method and system for analyzing the spatial topology of pipeline networks based on graph neural networks. Summary of the Invention

[0005] In view of this, the present invention proposes a pipeline network spatial topology relationship analysis method and system based on graph neural networks, aiming to solve the problem that existing pipeline network spatial topology relationship analysis relies on full network traversal and simple rule comparison in scenarios with multi-source data, complex topology and frequent incremental updates, resulting in low computational efficiency and difficulty in timely and accurate identification of non-compliant relationships.

[0006] In one aspect, this invention proposes a method for analyzing the spatial topology of pipeline networks based on graph neural networks, comprising: Collect pipeline network data and extract pipe segment objects, pipeline nodes, and constraint objects; A spatial topology index structure is constructed, which includes a spatial block index, a topology adjacency index, and a specification constraint index. The spatial block index records the pipe segment objects and constraint objects within each spatial block. The topology adjacency index records the connectivity relationship and topology relationship type of adjacent pipe segment objects based on the pipeline nodes. The specification constraint index stores constraint templates containing pipeline category, topology relationship type, and specification parameters. Based on the pipeline categories and topology relationship types recorded in the spatial topology index structure, a constraint template is selected from the specification constraint index, candidate object pairs are determined within the target spatial block, topology indices are calculated for the candidate object pairs, and preliminary topology relationship classification results are obtained based on the constraint template. The topology indices include horizontal clearance and vertical clearance. The candidate object pairs, topological indicators, and preliminary topological relationship classification results are used as input features for relationship classification. These are then input into the relationship classification model, which employs a graph neural network model to perform feature transfer and aggregation on the graph structure composed of the pipe segment objects, pipeline nodes, and constraint objects. The model outputs the topological relationship classification results and the topological consistency score. Version identifiers are set for each spatial block and candidate object pair in the spatial block index, and the spatial topology analysis results of the pipeline network are determined based on the pipeline network data update.

[0007] Furthermore, when collecting pipeline network data and extracting pipe segment objects, pipeline nodes, and constraint objects, the following steps are included: The system collects design drawings, as-built drawings, surveying data, building information modeling (BIM) data, and operation and maintenance log data. It extracts spatial and attribute data representing pipeline centerlines, pipeline connection components, and constraint objects. Through coordinate transformation and elevation benchmark unification, the various types of data are transformed into a unified spatial coordinate system. In the unified spatial coordinate system, the continuous pipeline centerline is divided into several pipe segment objects at the turning points, intersections, and equipment connection points of the pipeline centerline according to preset division rules. Corresponding pipeline nodes are generated at each division point. Duplicate representations are eliminated through matching rules of spatial position overlap and attribute consistency. Building boundaries, road boundaries, and protected area boundaries are extracted as constraint objects.

[0008] Furthermore, when constructing the spatial topology index structure, the following is included: The area where the pipeline network is located is divided into multiple spatial blocks according to a preset spatial resolution and height range. A block record is established for each spatial block in the spatial block index. The pipe segment object identifier, constraint object identifier and corresponding spatial boundary range information located in or intersecting with the spatial block are registered in each block record. In the topology adjacency index, a node record is established for each pipeline node. In each node record, the identifier of the pipe segment object connected to the pipeline node is registered. The corresponding topology relationship type is determined according to the connection direction, horizontal distance, vertical distance and overlap along the pipeline segment object and written into the node record.

[0009] Furthermore, the construction of the spatial topology index structure also includes: Based on the design specification text and pipeline laying standards, the specification clauses involving distance constraints between pipeline categories and between pipeline categories and constraint objects are extracted. Each specification clause is parsed into a constraint template. Each constraint template includes at least a first pipeline category, a second pipeline category or constraint object category, an applicable topology relationship type, and a specification parameter group. The constraint template is stored in the specification constraint index using the first pipeline category, the second pipeline category or the constraint object category, and the topology relationship type as search keys. The specification parameter group includes the horizontal clearance range, the vertical clearance range, the burial depth difference range, the upper limit of the overlap length along the pipeline, or the intersection angle range.

[0010] Furthermore, when selecting constraint templates from the specification constraint index based on the pipeline categories and topology relationship types recorded in the spatial topology index structure, and determining candidate object pairs within the target spatial block, the process includes: Within the target spatial block, the identifiers of pipe segment objects and constraint objects located within or intersecting with the target spatial block are read based on the spatial block index. Combined with the connectivity relationships and topological relationship types of adjacent pipe segment objects corresponding to the pipeline nodes recorded in the topology adjacency index, combinations of pipe segment objects and pipe segment objects or combinations of pipe segment objects and constraint objects that satisfy the constraint template constraint conditions are selected according to the first pipeline category, second pipeline category, or constraint object category in the constraint template and the applicable topological relationship type. The selected combinations are taken as candidate object pairs. For each candidate object pair, the horizontal and vertical clearances are calculated based on the endpoint positions of the pipe segment objects, the burial depth of the pipe segment objects, and the boundary positions of the constraint objects recorded in the spatial block index and the topology adjacency index. If necessary, the spatial nearest distance and the length of overlap along the path of the candidate object pair are further calculated. The horizontal clearance, the vertical clearance, and the optional spatial nearest distance and length of overlap along the path are used as the topological indicators of the candidate object pair.

[0011] Furthermore, preliminary topological relationship classification results are obtained based on the constraint template, including: For each candidate object pair, the topology index is compared item by item with the set of standard parameters specified in the constraint template. When the horizontal clearance and the vertical clearance are both within the safe range of the corresponding set of standard parameters and the selectable closest spatial distance and the length of overlap along the path meet the requirements of the set of standard parameters, the preliminary topology relationship classification result of the candidate object pair is determined as a compliant relationship. When at least one of the topology indicators is located within the boundary between the safety range and the preset warning range of the corresponding specification parameter group, the preliminary topology relationship classification result of the candidate object pair is determined as a general warning relationship; When any of the topology indicators exceeds the allowable range of the corresponding specification parameter group or when the pipe segment object intrudes into the boundary of the protected area, the preliminary topology relationship classification result of the candidate object pair is determined as a serious violation relationship.

[0012] Furthermore, when inputting the candidate object pairs, the topological indices, and the preliminary topological relationship classification results as relationship classification input features into the relationship classification model, the following steps are included: In the relationship classification model, the pipe segment object, the pipeline node, and the constraint object are respectively used as the pipe segment node, pipeline node, and constraint node in the graph structure. Node features such as node type, spatial location, burial depth, and pipeline category are written for each node. Connection edges representing physical connectivity are established between the pipe segment node and the pipeline node based on the topological adjacency index. Neighboring edges representing spatial proximity are established between pipe segment nodes and between pipe segment nodes and constraint nodes based on the spatial block index and the standard constraint index. The corresponding topological relationship type, horizontal clearance, and vertical clearance are written into each connection edge and neighboring edge. The topological index corresponding to each candidate object pair and the preliminary topological relationship classification result are used as the relationship features of that candidate object pair, along with the node features and the edge features, and input into the graph structure.

[0013] Furthermore, the relationship classification model employs at least two layers of graph neural network to sequentially perform feature transfer and aggregation on the graph structure. In each layer, the feature weights transferred from adjacent nodes to the target node are controlled based on the edge features. After completing multi-layer feature transfer and aggregation, aggregated features of pipe segment nodes, pipeline nodes, and constraint nodes related to the candidate object pair are extracted for each candidate object pair. The aggregated features and the relationship features of the candidate object pair are input together into the classification output unit, which generates the topology relationship classification result and the topology consistency score.

[0014] Furthermore, setting version identifiers for each spatial block and candidate object pair in the spatial block index, and determining the spatial topology analysis results of the pipeline network based on the pipeline network data update status, includes: When pipeline network data is updated, topology index calculations and relationship classifications are performed on the spatial blocks whose version identifiers have changed and the candidate object pairs within them to obtain the spatial topology relationship analysis results of the pipeline network. For the candidate object pairs whose version identifiers have not changed, the topology relationship classification results and topology consistency scores are reused and cached to obtain the spatial topology relationship analysis results of the pipeline network.

[0015] Compared with existing technologies, the advantages of this invention are as follows: By extracting pipe segment objects, pipeline nodes, and constraint objects, and constructing a spatial topology index structure, pipeline categories, topology relationship types, and specification parameters are structurally bound at the index layer. Combined with precise selection of candidate object pairs and calculation of topology indicators based on constraint templates within the target spatial blocks, this eliminates the reliance on traversal-style geometric collision calculations across the entire pipeline network, reducing the retrieval scope and computational complexity in large-scale pipeline networks. Furthermore, candidate object pairs, topology indicators, and preliminary topology relationship classification results are uniformly mapped to a graph structure, utilizing a graph neural network. The model performs feature transfer and aggregation, outputting topological relationship classification results and topological consistency scores for each candidate object pair. It can comprehensively utilize local clearance information and the entire network topology pattern, improving the accuracy of violation identification under complex spatial relationships and boundary conditions. The topological consistency score provides quantifiable risk references for planning review, completion acceptance, and operation and maintenance decisions. The model sets version identifiers for each spatial block in the spatial block index and candidate object pairs, enabling rapid dynamic analysis of pipeline network spatial topology relationships in scenarios with frequent incremental updates, taking into account the comprehensive requirements of computational efficiency, data scale, and topology identification accuracy.

[0016] On the other hand, this application also provides a pipeline network spatial topology analysis system based on graph neural networks, used to apply the above-mentioned pipeline network spatial topology analysis method based on graph neural networks, including: The acquisition unit is configured to acquire pipeline network data and extract pipe segment objects, pipeline nodes, and constraint objects. The first analysis unit is configured to construct a spatial topology index structure, which includes a spatial block index, a topology adjacency index, and a specification constraint index. The spatial block index records pipe segment objects and constraint objects within each spatial block. The topology adjacency index records the connectivity relationships and topology relationship types of adjacent pipe segment objects based on the pipeline nodes. The specification constraint index stores constraint templates containing pipeline categories, topology relationship types, and specification parameters. The second analysis unit is configured to select a constraint template from the specification constraint index based on the pipeline category and topology relationship type recorded in the spatial topology index structure, determine candidate object pairs within the target spatial block, calculate topology indices for the candidate object pairs, and obtain preliminary topology relationship classification results based on the constraint template. The topology indices include horizontal clearance and vertical clearance. The correction unit is configured to take the candidate object pairs, topological indicators and preliminary topological relationship classification results as relationship classification input features and input them into the relationship classification model. The relationship classification model uses a graph neural network model to perform feature transfer and aggregation on the graph structure composed of the pipe segment objects, pipeline nodes and constraint objects, and outputs the topological relationship classification results and topological consistency score. The update unit is used to set version identifiers for each spatial block and candidate object pair in the spatial block index, and to determine the spatial topology analysis results of the pipeline network based on the pipeline network data update status.

[0017] It is understandable that the above-mentioned pipeline network spatial topology analysis method and system based on graph neural networks have the same beneficial effects, and will not be elaborated further here. Attached Figure Description

[0018] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A flowchart illustrating a pipeline network spatial topology analysis method based on graph neural networks provided in an embodiment of the present invention; Figure 2 This is a functional block diagram of a pipeline network spatial topology analysis system based on graph neural networks, provided in an embodiment of the present invention. Detailed Implementation

[0019] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, embodiments and features in the embodiments of the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0020] In pipeline network spatial topology analysis, existing technologies primarily rely on full network traversal and simple rule engines to perform distance calculations and specification comparisons. They lack multi-layered spatial topology indexes for large-scale pipeline networks and constraint template management mechanisms that combine topology relationship types and pipeline categories. The reduced computational efficiency stems from the increased computational complexity caused by full network traversal. Insufficient accuracy in identifying complex topology relationships arises because local rule engines cannot integrate the global dependencies of spatial proximity and physical connectivity. This problem directly impacts the timeliness and reliability of pipeline network analysis, thus hindering decision-making during the planning, design, completion, acceptance, and operation and maintenance management phases.

[0021] For example, during pipeline collision checks in urban underground utility tunnel construction projects, various pipelines such as water supply pipelines, drainage pipelines, gas pipelines, power cables, and communication cables are densely intertwined in a limited space. Furthermore, when performing topological relationship analysis on the intersection areas of gas pipelines and power cables, existing methods require pairwise distance calculations for all pipeline objects in the entire network, consuming significant computational resources on redundant operations on unrelated object pairs. Simultaneously, for complex topological scenarios such as multi-layered intersections or non-standard connections, rule engines rely solely on geometric distances for local judgments, failing to accurately identify topological relationship types and leading to the omission of potential collision risks.

[0022] If the above problems are not addressed, the inefficiency of pipeline network spatial topology analysis will prolong the decision-making cycle in planning, design, and operation and maintenance management. Insufficient identification accuracy may prevent the timely detection of safety hazards, thereby causing pipeline operation accidents and threatening the safe operation of urban infrastructure.

[0023] For this, please refer to Figure 1 As shown, this application proposes a method for analyzing the spatial topology of pipeline networks based on graph neural networks, including: S100: Collect pipeline network data and extract pipe segment objects, pipeline nodes, and constraint objects.

[0024] S200: Construct a spatial topology index structure, which includes a spatial block index, a topology adjacency index, and a specification constraint index. The spatial block index records the pipe segment objects and constraint objects within each spatial block. The topology adjacency index records the connectivity relationships and topology relationship types of adjacent pipe segment objects based on pipeline nodes. The specification constraint index stores constraint templates containing pipeline categories, topology relationship types, and specification parameters.

[0025] S300: Based on the pipeline categories and topology relationship types recorded in the spatial topology index structure, select constraint templates from the specification constraint index, determine candidate object pairs within the target spatial block, calculate topology indices for the candidate object pairs, and obtain preliminary topology relationship classification results based on the constraint templates. The topology indices include horizontal clearance and vertical clearance.

[0026] S400: The candidate object pairs, topological indicators, and preliminary topological relationship classification results are used as input features for relationship classification. The relationship classification model is then input into the relationship classification model. The relationship classification model uses a graph neural network model to perform feature transfer and aggregation on the graph structure composed of pipe segment objects, pipeline nodes, and constraint objects, and outputs the topological relationship classification results and topological consistency score.

[0027] S500: Set version identifiers for each spatial block and candidate object pair in the spatial block index, and determine the spatial topology analysis results of the pipeline network based on the pipeline network data update status.

[0028] Specifically, in the field of pipeline network spatial topology analysis, existing technologies mainly rely on full network traversal and simple rule engines for distance calculation and specification comparison, making it difficult to balance computational efficiency for large-scale pipeline networks with the accuracy of identifying complex topological relationships. Spatial topology index structure refers to an index system for organizing and managing pipeline network data. It can be implemented using a combination of spatial block indexes, topological adjacency indexes, and specification constraint indexes. For example, spatial block indexes can divide regions based on geographic coordinate grids or quadtree structures; topological adjacency indexes can use adjacency lists or graph databases to store node connection relationships; and specification constraint indexes can use hash tables or database indexes with pipeline category and topological relationship type as search keys. Its main purpose is to achieve rapid location of pipeline objects and efficient matching of constraint conditions. Furthermore, constraint templates refer to rule units that store pipeline categories, topological relationship types, and specification parameters. They can be implemented using structured data templates generated from parsing design specification text or manually defined parameterized rule sets. For example, templates can be formed by extracting distance constraint clauses from pipeline laying standards. Their main purpose is to achieve categorized storage and precise retrieval of specification parameters. Specifically, topological indicators include horizontal and vertical clearances, which can be achieved by calculating Euclidean distance using coordinates or by measuring spatial distances using geographic information system tools. For example, the minimum horizontal distance can be calculated using the coordinate difference between the endpoints of a pipe segment. This is primarily to quantify the spatial relationship characteristics between pipeline objects. As a preferred implementation, the relationship classification model uses a graph neural network model to perform feature transfer and aggregation on the graph structure composed of pipe segment objects, pipeline nodes, and constraint objects. Feature aggregation can be achieved using graph convolutional networks or graph attention network architectures. For example, node features can be iteratively updated through multi-layer neural networks. This is mainly to integrate spatial proximity and physical connectivity relationships to improve classification accuracy. In practical applications, setting version identifiers for each spatial block and candidate object pair in the spatial block index can be achieved using timestamps, hash values, or incremental sequence numbers. For example, a unique identifier can be generated based on the data update time. This is mainly to support local recalculation and result reuse under the incremental update mechanism. Therefore, this application achieves computational range compression through the spatial topology index structure, avoiding the efficiency bottleneck caused by full network traversal. Simultaneously, the relationship classification model utilizes graph neural networks to perform feature transfer and aggregation on the graph structure, capturing global dependencies in complex topologies. Furthermore, the version identification mechanism allows for processing only the changed portions when pipeline network data is updated, improving the efficiency of dynamic analysis. The synergistic effect of these technical features overcomes the limitations of traditional methods in terms of scalability and accuracy improvement, providing an efficient and accurate solution for spatial topology analysis of large-scale pipeline networks.

[0029] Understandably, the implementation of the pipeline network spatial topology relationship analysis method first involves the collection and extraction of pipeline network data. This process transforms the original design drawings, as-built drawings, and survey data into structured pipe segment objects, pipeline nodes, and constraint objects, thereby avoiding redundant processing of non-critical data. The construction of the spatial topology index structure is a core component. It achieves localized organization of the pipeline network area through spatial block indexing. For example, this spatial block index can specifically adopt a spatial partitioning method based on regular grids, with each spatial block size set at 100 meters × 100 meters to record the spatial boundary range information of pipe segment objects and constraint objects within each block, thus narrowing the query scope. The topology adjacency index, based on pipeline nodes, registers the connectivity and topology relationship types of adjacent pipe segment objects, freeing the retrieval of physical connection relationships from the dependence on full network traversal. The normative constraint index stores constraint templates that combine pipeline categories and topology relationship types. By using pipeline categories and topology relationship types as search keys, normative parameters can be quickly retrieved, avoiding blind scanning of the rule base. Furthermore, based on the pipeline categories and topological relationship types recorded in the spatial topology index structure, constraint templates are precisely selected from the normative constraint index. Candidate object pairs meeting the conditions are screened within the target spatial block, and topological indicators such as horizontal and vertical clearances are calculated only for these objects, reducing unnecessary computation. Subsequently, the topological indicators are compared in real-time according to the normative parameter set of the constraint template, generating preliminary topological relationship classification results and providing a reliable starting point for correction. Based on this, the candidate object pairs, topological indicators, and preliminary classification results are used as input features for relationship classification, which are then input into the relationship classification model. This model can specifically employ a two-layer graph convolutional network architecture to perform feature transfer and aggregation on the graph structure composed of pipeline segments, pipeline nodes, and constraint objects. Edge features control the feature transfer weights between adjacent nodes, allowing node features to fully integrate spatial proximity and physical connectivity relationships, thereby outputting topological relationship classification results and topological consistency scores, solving the accuracy bottleneck of local rules failing to capture global dependencies in complex topologies. Furthermore, version identifiers are set for each spatial block and candidate object pair in the spatial block index. When pipeline network data is updated, only the parts with changed version identifiers are locally recalculated, while the unchanged parts reuse cached results, avoiding redundant analysis of the entire network. Thus, this method addresses the problems of low computational efficiency and insufficient accuracy in identifying complex topological relationships in large-scale pipeline network analysis by compressing the computational scope through a multi-layer index structure, improving the depth of relationship recognition through graph neural networks, and ensuring update efficiency through a version mechanism. This achieves high efficiency and accuracy in spatial topological relationship analysis. Specifically, the collaborative design of the spatial block index and the topological adjacency index significantly reduces the complexity of data retrieval, the fast matching mechanism of the normative constraint index improves the efficiency of constraint application, and the graph neural network's ability to aggregate graph structure features enhances the robustness of topological relationship classification. Ultimately, this ensures the timeliness and reliability of analysis results in the scenario of dynamically updated pipeline network data.

[0030] Specifically, in some of the embodiments described above in this application, steps are proposed to collect pipeline network data and extract pipe segment objects, pipeline nodes, and constraint objects. However, in the implementation process, problems such as inconsistent coordinate systems of multi-source heterogeneous data leading to spatial position deviation, failure to focus on key topological points in the division of pipeline centerlines causing object representation distortion, failure to eliminate duplicate representations causing data redundancy, and non-standard extraction of constraint objects affecting net distance calculation are made difficult to meet the requirements of large-scale pipeline networks in terms of accuracy and efficiency of spatial topology index construction and relationship analysis.

[0031] In response, this application further proposes that when collecting pipeline network data and extracting pipe segment objects, pipeline nodes, and constraint objects, the following should be included: The system collects design drawings, as-built drawings, surveying data, building information modeling (BIM) data, and operation and maintenance records. It extracts spatial and attribute data representing pipeline centerlines, pipeline connection components, and constraint objects. Through coordinate transformation and elevation benchmark unification, the various types of data are converted to a unified spatial coordinate system. In the unified spatial coordinate system, the continuous pipeline centerlines are divided into several pipe segment objects at the turning points, intersections, and equipment connection points of the pipeline centerlines according to preset division rules. Corresponding pipeline nodes are generated at each division point. Duplicate representations are eliminated through matching rules of spatial position overlap and attribute consistency. Building boundaries, road boundaries, and protected area boundaries are extracted as constraint objects.

[0032] Design drawing data refers to pipeline layout drawings generated during the engineering design phase. These can be implemented using computer-aided design (CAD) file formats and aim to provide initial pipeline design parameters. As-built drawing data refers to drawings recording the actual pipeline layout after project completion. These can be implemented using Geographic Information System (GIS) formats and aim to reflect the actual installation status of the pipelines. Surveying data refers to spatial location data of pipelines obtained through on-site measurements. This data can be acquired using Global Navigation Satellite Systems (GNSS) and aims to supplement high-precision spatial information. Building Information Modeling (BIM) data refers to digital model data containing the three-dimensional geometry and attribute information of the pipelines. This data can be implemented using industrial-grade standard formats and aims to provide integrated information on the pipelines and their building environment. Operation and maintenance log data refers to attribute data recorded during pipeline operation and maintenance. This data can be implemented using structured database tables and aims to provide pipeline category and operation and maintenance attribute information.

[0033] In practical applications, the pipeline centerline refers to the spatial curve data representing the pipeline axis, which can be represented by a polyline or parametric curve, and its purpose is to describe the spatial direction of the pipeline. Pipeline connection components refer to the data of the connecting parts between pipelines, which can be represented by node models or joint geometry, and their purpose is to identify the physical connection points of the pipelines. Constraint objects refer to the external boundary data that affects the pipeline layout, which can be represented by polygonal or surface models, and their purpose is to define the restricted areas for pipeline laying.

[0034] Specifically, coordinate transformation refers to the process of converting spatial data from different coordinate systems to a unified reference frame. This can be achieved using affine transformations or polynomial fitting methods, with the aim of eliminating coordinate system differences. Elevation datum unification refers to the process of converting data from different vertical reference systems to a unified elevation datum. This can be achieved using a geoid calibration model, with the aim of ensuring the consistency of elevation data.

[0035] Among them, the preset division rules refer to the criteria for determining the dividing points of the pipeline centerline, which can be achieved based on curvature change detection or direction angle abrupt change threshold, with the aim of accurately capturing key topological points. Turning points refer to points where the pipeline direction changes, which can be identified using spatial derivative analysis. Intersection points refer to points where multiple pipelines intersect or merge, which can be identified using spatial intersection calculation. Equipment connection points refer to points where pipelines connect to equipment such as valves and pump stations, which can be identified using attribute association or spatial proximity matching.

[0036] In practical applications, spatial overlap refers to objects coinciding or being close in spatial coordinates, which can be determined by calculating a spatial distance less than a threshold. Attribute consistency refers to objects being identical in attributes such as category and material, which can be determined by comparing attribute fields. Matching rules refer to the mechanism for eliminating duplicate representations, which can be implemented using a weighted scoring system of spatial distance and attribute similarity to ensure object uniqueness. Building boundaries refer to the spatial boundaries of a building's outer contour, which can be achieved by extracting the building's outline. Road boundaries refer to the spatial boundaries of road curbs or centerlines, which can be achieved by offsetting the road centerline. Protected area boundaries refer to the restricted areas for the safe laying of pipelines, which can be achieved using buffer zone analysis to provide external constraints for regulatory inspections.

[0037] Specifically, this application first collects multi-source heterogeneous data covering the entire lifecycle of pipelines to ensure the comprehensiveness and complementarity of data sources. Then, it extracts spatial and attribute data of pipeline centerlines, pipeline connection components, and constraint objects, providing both geometric location and category attribute information for analysis. Next, it transforms all types of data into a unified spatial coordinate system through coordinate transformation and elevation datum unification, eliminating spatial reference differences between different data sources and ensuring spatial consistency of all objects within a unified framework. Based on this, it precisely divides the pipeline according to the turning points, intersections, and equipment connection locations of the pipeline centerlines, forming pipe segment objects that conform to the actual topology. Simultaneously, it generates pipeline nodes at each division location as logical anchor points for topological connections. Furthermore, it employs matching rules based on spatial location overlap and attribute consistency to eliminate duplicate representations in multi-source data, ensuring the uniqueness and accuracy of pipe segment objects and constraint objects. Finally, it extracts building boundaries, road boundaries, and protected area boundaries as constraint objects, precisely defining external environmental constraints. These steps, executed sequentially, form a complete pipeline network basic data construction process, solving problems such as multi-source data integration, object representation distortion, and data redundancy, laying a solid foundation for efficient and accurate spatial topology relationship analysis.

[0038] As a preferred embodiment, the specific implementation of this application is as follows: In an underground integrated pipeline analysis project in a certain city, design drawings, as-built measurement data, field survey point cloud data, building information model data, and operation and maintenance record data of water supply, drainage, gas, and other pipelines are collected. The spatial curves of the centerlines of each pipeline, the location information and attribute data of connecting components such as valves and inspection wells, as well as the spatial boundary data of building outlines and road centerlines are extracted. Through standard coordinate transformation methods and elevation correction techniques, all data are integrated into a unified spatial coordinate reference frame. In the unified coordinate system, based on the direction change points of the pipeline centerlines, the spatial intersection points of multiple pipelines, and the connection points with pumping station equipment, the continuous pipeline centerlines are rationally divided to form pipe segment objects, and pipeline nodes are set at each division location. Using spatial proximity and attribute matching rules, duplicate object representations are identified and merged. Urban building outline polygons, road curb lines, and pipeline safety protection zone boundaries are defined as constraint objects.

[0039] Through the above solution, this application solves the problem of spatial position deviation caused by the inconsistency of coordinate systems of multi-source heterogeneous data, avoids topological relationship errors caused by the distortion of pipeline centerline division, eliminates data redundancy and improves processing efficiency, and standardizes the extraction of constraint objects to ensure the accuracy of net distance calculation, thereby improving the accuracy and analysis efficiency of pipeline network spatial topology index construction.

[0040] In practical applications, some embodiments of this application propose constructing a spatial topology index structure to organize pipeline data. However, during its implementation, the lack of preset rules for spatial block division leads to uneven block distribution, affecting query efficiency. Furthermore, the determination of topological relationships does not consider multi-dimensional geometric parameters, resulting in inaccurate identification and making it difficult to support efficient analysis and accurate topology relationship judgment for large-scale pipeline networks.

[0041] In this regard, this application further proposes: The area where the pipeline network is located is divided into multiple spatial blocks according to the preset spatial resolution and height range. A block record is created for each spatial block in the spatial block index. In each block record, the pipe segment object identifier, constraint object identifier, and corresponding spatial boundary range information located within or intersecting with the spatial block are registered.

[0042] In the topology adjacency index, a node record is created for each pipeline node. The identifier of the pipe segment object connected to the pipeline node is registered in each node record. The corresponding topology relationship type is determined and written into the node record based on the connection direction, horizontal distance, vertical distance and overlap along the pipeline segment object.

[0043] Specifically, the preset spatial resolution refers to the granularity parameter for dividing space horizontally. This can be achieved using uniform grid partitioning or a dynamic grid strategy based on pipeline density distribution. The goal is to ensure that spatial blocks are evenly distributed horizontally, avoiding excessively large blocks leading to overly broad query ranges or excessively small blocks causing index fragmentation. The height range refers to the range parameter for dividing space vertically. This can be achieved using fixed height intervals or adaptive interval partitioning based on pipeline burial depth distribution characteristics. The goal is to adapt to the three-dimensional spatial characteristics of the pipeline network, ensuring that blocks reasonably cover the pipeline distribution vertically. In practical applications, spatial block records refer to the data structure storing object information within a spatial block. This can be implemented using R-tree or quadtree index structures, aiming to efficiently manage the spatial location information of pipe segment objects and constraint objects within the block. Specifically, the topology relationship type refers to the category describing the connection relationship between pipe segment objects. This can include types such as intersection, parallel, or overlap, determined by comprehensively analyzing connection direction, horizontal distance, vertical distance, and overlap along the pipeline. The goal is to accurately characterize the spatial topology characteristics between pipelines.

[0044] Specifically, this application's solution controls the spatial partitioning process by pre-setting spatial resolution and height range, ensuring a uniform distribution of blocks in both horizontal and vertical dimensions, thus resolving query efficiency fluctuations caused by uneven partitioning. In the spatial partitioning index, a partition record is established for each spatial block, registering pipe segment object identifiers, constraint object identifiers, and spatial boundary range information, constructing a structured spatial index framework that supports rapid data location and local processing for target spatial blocks. In the topology adjacency index, node records are established with pipeline nodes as the core, registering connected pipe segment object identifiers, and comprehensively determining the topology relationship type based on connection direction, horizontal distance, vertical distance, and overlap along the pipeline, achieving accurate topology relationship modeling based on the physical connection essence of pipelines. The collaborative organization of the spatial partitioning index and the topology adjacency index enables pipeline data to possess dual indexing characteristics of spatial location and topology connectivity, providing a high-precision, low-overhead data foundation for candidate pair screening and topology index calculation.

[0045] As a specific implementation method, the solution of this application is implemented as follows: The area where the pipeline network is located can be divided into grid-like spatial blocks based on a preset spatial resolution. The spatial block index can use a spatial database index structure to store block records. Each block record contains a list of pipe segment object identifiers, a list of constraint object identifiers, and the coordinates of the block boundary range. The topological adjacency index can use a graph database to store node records. Each node record contains the identifiers of connected pipe segment objects and the topological relationship type. The topological relationship type determines the connection direction by calculating the coordinate vectors of the endpoints of the connected pipe segment objects, measuring the horizontal projection distance as the horizontal distance, calculating the elevation difference as the vertical distance, and detecting the length of the overlapping part of the pipe segments as the overlap along the pipeline.

[0046] Through the above scheme, the spatial partitioning of this application achieves uniform distribution based on preset rules, avoiding fluctuations in query efficiency. The determination of topology relationship types integrates multi-dimensional geometric parameters, improving the accuracy of topology relationship identification, thereby supporting efficient analysis and accurate topology relationship judgment of large-scale pipeline networks.

[0047] Traditional methods for analyzing the spatial topology of pipeline networks primarily rely on spatial block indexes and topological adjacency indexes for data organization when constructing spatial topology index structures. However, in their implementation, they lack a systematic organization and efficient retrieval mechanism for constraint clauses in design specifications and pipeline laying standards. This results in the inability to quickly match relevant distance constraint specifications based on pipeline categories and topology relationship types. Consequently, the selection of candidate pairs and the calculation of topology indicators lack standardized basis, affecting the accuracy of topology relationship classification and the computational efficiency of large-scale pipeline network analysis.

[0048] In response, this application further proposes that when constructing a spatial topology index structure, it also includes: extracting specification clauses involving distance constraints between pipeline categories and between pipeline categories and constraint objects based on design specification texts and pipeline laying standards; parsing each specification clause into a constraint template; each constraint template including at least a first pipeline category, a second pipeline category or constraint object category, an applicable topology relationship type, and a specification parameter group; storing constraint templates in the specification constraint index using the first pipeline category, the second pipeline category or constraint object category, and the topology relationship type as search keys; and the specification parameter group including the horizontal clearance range, the vertical clearance range, the burial depth difference range, the upper limit of the overlap length along the route, or the intersection angle range.

[0049] The extraction of specification clauses refers to the process of identifying and obtaining distance constraint-related clauses from design specification texts and pipeline laying standards. This can be achieved using text mining algorithms based on keyword matching or natural language processing techniques, aiming to ensure the authority and completeness of constraint data and avoid human input errors. Constraint template parsing can be understood as the operation of converting unstructured specification clauses into structured data formats. This can be implemented using template filling engines or semantic parsing rule bases, aiming to enable computers to automatically identify and apply constraint rules. A constraint template is a standardized data unit containing pipeline category, topology relationship type, and specification parameters. It can be designed as a database record or a set of key-value pairs, aiming to accurately match specific pipeline combinations (such as gas pipelines and power pipelines) and topology scenarios (such as parallel or intersecting). Specification constraint index storage refers to a mechanism that organizes constraint templates using pipeline category and topology relationship type as composite search keys. This can be implemented using hash tables or inverted index structures, aiming to establish a multi-dimensional, fast query channel. The standard parameter set can be understood as a set of numerical ranges that define compliance boundaries. Specifically, it may include parameter categories such as horizontal net distance range. Its purpose is to provide a quantitative basis for the calculation of topology indicators.

[0050] Specifically, the proposed solution systematically extracts and parses unstructured constraint clauses from design specification texts and pipeline laying standards into structured constraint templates. A specification constraint index is constructed using pipeline category, constraint object category, and topology relationship type as composite search keys, thus achieving machine-processable constraint data. When candidate object pairs need to be selected, key-value matching is performed directly in the specification constraint index based on the pipeline category and topology relationship type recorded in the spatial topology index structure to quickly locate the applicable constraint template, avoiding a traversal query of all specification clauses across the entire network. This mechanism is organically integrated with the spatial block index and the topology adjacency index: the spatial block index provides spatial distribution information of pipe segment objects and constraint objects; the topology adjacency index provides pipeline node connectivity and topology relationship types; and the specification constraint index provides specification parameter basis based on the pipeline category and topology relationship type output by the former two. These three elements work together to support accurate selection of candidate object pairs and calculation of topology indicators, thereby constructing a multi-layered index system for large-scale pipeline networks.

[0051] As a preferred embodiment, the solution of this application is implemented as follows: In the scenario of gas pipelines and water supply pipelines crossing, a constraint template is queried from the specification constraint index using "gas pipeline - water supply pipeline - crossing" as a composite search key. This constraint template can be specifically stored in the index table of a relational database, where the specification parameter group can include parameter categories such as horizontal clearance range and vertical clearance range. When processing pipeline network data, the type of cross-topological relationship between gas pipelines and water supply pipelines at pipeline nodes is determined according to the topological adjacency index. Then, the corresponding constraint template is quickly obtained through the specification constraint index, thereby guiding the screening of candidate object pairs and the calculation process of topological indicators.

[0052] Through the above scheme, this application realizes the structured storage and efficient retrieval of constraint templates, which provides a clear standard for the selection process of candidate object pairs, avoids the topological relationship classification error caused by inaccurate constraint matching, reduces the computational overhead in large-scale pipeline network analysis, and improves the accuracy and execution efficiency of spatial topological relationship analysis.

[0053] In practical applications, some of the embodiments described above in this application propose a standard constraint index to store constraint templates. However, in its implementation, how to efficiently use the constraint templates to filter out the candidate object pairs that need to be checked and calculate the relevant topology indicators in order to avoid the computational overhead caused by a full traversal of the entire pipeline network, while ensuring that the filtering process can accurately match the pipeline category, constraint object category and topology relationship type, has become a key bottleneck in improving the efficiency of large-scale pipeline network analysis.

[0054] In response, this application further proposes a method for selecting constraint templates from the specification constraint index based on the pipeline categories and topology relationship types recorded in the spatial topology index structure. When determining candidate object pairs within a target spatial block, the method includes: reading the pipe segment object identifiers and constraint object identifiers located within or intersecting with the target spatial block based on the spatial block index; combining the connectivity relationships and topology relationship types of adjacent pipe segment objects corresponding to the pipeline nodes recorded in the topology adjacency index; and selecting the constraint object pairs according to the first pipeline category, second pipeline category, or constraint object category in the constraint template, as well as applicable... The topology relationship type is filtered to find pipe segment objects and combinations of pipe segment objects or pipe segment objects and constraint objects that meet the constraint template conditions. The filtered combinations are used as candidate object pairs. For each candidate object pair, the horizontal and vertical clearances are calculated based on the endpoint positions of the pipe segment objects, the burial depth of the pipe segment objects, and the boundary positions of the constraint objects recorded in the spatial block index and the topology adjacency index. If necessary, the spatial nearest distance and the length of overlap along the path of the candidate object pair are further calculated. The horizontal clearance, vertical clearance, and optional spatial nearest distance and length of overlap along the path are used as the topology indicators of the candidate object pair.

[0055] The spatial block index refers to a data structure that organizes the pipeline network area according to its spatial location. It can be implemented using grid partitioning or a quadtree structure, aiming to limit the analysis scope to a specific spatial region and avoid global data scanning. The topological adjacency index can be understood as an index mechanism that records the connection relationships between pipeline nodes and pipe segments. It can be stored in the form of an adjacency list or a graph database, aiming to pre-store physical connectivity information and reduce the computational burden of real-time topology analysis. The constraint template specifically refers to a rule template containing pipeline categories and specification parameters. It can be generated based on the parsing of design specification text, aiming to provide a standardized basis for specification checks. Candidate object pairs can be understood as combinations of objects requiring topological relationship verification. They can be combinations of pipe segments or pipe segments and constraint objects filtered through the index, aiming to narrow the inspection scope and focus on potential non-compliant objects. Topological indicators specifically refer to quantitative parameters used to evaluate pipeline spatial relationships. They can include horizontal and vertical distance measurements, aiming to provide an objective basis for judging specification compliance.

[0056] Specifically, the proposed solution strictly limits the analysis scope to the target spatial block through spatial block indexing, reading only the pipe segment objects and constraint object identifiers within or intersecting that block, thus avoiding traversal of the entire network data. Based on this, and combined with the pre-stored pipeline node connectivity relationships in the topology adjacency index, it quickly obtains the combination information and topology relationship type of adjacent pipe segment objects. Subsequently, according to the first pipeline category, second pipeline category, or constraint object category specified in the constraint template, and the applicable topology relationship type, it precisely matches the selected object combinations, retaining only candidate object pairs that meet the constraint conditions. For the selected candidate object pairs, it directly uses the pipe segment endpoint positions, burial depths, and constraint object boundary data stored in the spatial block index and topology adjacency index to calculate basic indicators such as horizontal and vertical clearances. Supplementary calculations of the spatial nearest distance and along-the-line overlap length are triggered only when a specific topology relationship type (such as parallel laying) is detected. Finally, these indicators are integrated into the topology indicator input analysis process. The above steps form a complete logical chain from spatial scope limitation and topological relationship screening to index calculation. Each link is closely connected and together realizes an efficient screening mechanism based on index-driven and template constraints.

[0057] As a specific implementation method, this application is implemented as follows: In the scenario of urban underground pipeline network analysis, firstly, the identifiers of water supply pipeline segments and road boundary constraint objects within a target spatial block are read based on the spatial block index. Then, combined with the pipeline node information recorded in the topological adjacency index, the combinations of power pipelines adjacent to the water supply pipeline and their parallel laying topological relationship types are identified. Candidate object pairs to be checked are selected according to the "water supply-power" category and the "parallel laying" topological relationship type in the constraint template. Based on the pipeline endpoint coordinates and burial depth data stored in the index, the horizontal and vertical clearances between the water supply pipeline and the power pipeline are calculated. For parallel-laid pipeline segments, the overlap length along the route is further calculated. Finally, these indicators are used as topological indicators input into the relationship classification model for analysis.

[0058] Through the above technical solution, this application can reduce the computational complexity of pipeline network topology analysis and avoid the computational overhead caused by a full traversal of the entire pipeline network. At the same time, it ensures that the selection process strictly follows design specifications, accurately matching pipeline categories, constraint object categories, and topology relationship types. While guaranteeing the relevance and compliance of topology index generation, it significantly improves the computational efficiency of spatial topology analysis for large-scale pipeline networks.

[0059] In some of the embodiments described above in this application, a preliminary topological relationship classification result is obtained based on a constraint template to quickly filter candidate object pairs. However, in its implementation, classification is based only on simple rules, which cannot handle complex situations such as indicators being in the boundary interval or intruding into the boundary, resulting in inaccurate preliminary classification results and affecting the input quality of the relationship classification model.

[0060] In this regard, this application further proposes obtaining preliminary topological relationship classification results based on constraint templates, including: For each candidate object pair, the topology indicators are compared with the standard parameter set specified in the constraint template item by item. When the horizontal and vertical clearances are both within the safe range of the corresponding standard parameter set and the selectable nearest spatial distance and the length of overlap along the path meet the limits of the standard parameter set, the preliminary topology relationship classification result of the candidate object pair is determined as a compliant relationship.

[0061] When at least one of the topology indicators is located within the boundary between the safe range and the preset warning range of the corresponding specification parameter group, the preliminary topology relationship classification result of the candidate object pair is determined as a general warning relationship.

[0062] When any of the topology indicators exceeds the allowable range of the corresponding specification parameter group or when the characterizing pipe segment object intrudes into the boundary of the protected area, the preliminary topology relationship classification result of the candidate object pair is determined as a serious violation relationship.

[0063] The specification parameter set refers to the collection of parameters stored in the constraint template, including horizontal and vertical clearance ranges. It can be implemented using structured data tables or rule engine configuration files, aiming to provide differentiated distance constraint benchmarks for different pipeline categories. The safe range and preset warning range can be understood as hierarchical threshold intervals defined in the specification parameter set. These can be implemented through numerical interval division or fuzzy logic rules, aiming to distinguish the boundary states of complete compliance, potential risk, and serious violation. The boundary interval specifically refers to the transition area between the upper limit of the safe range and the lower limit of the warning range. It can be based on engineering experience to set a buffer zone or dynamically calculate boundary values, aiming to capture complex situations where indicators are in a critical state. The boundary of the intrusion into the protected area refers to the spatial location of a pipeline segment exceeding the physical boundary defined by the constraint object. It can be determined through spatial geometric operations or rasterization analysis, aiming to identify direct spatial conflicts between pipelines and buildings or protected areas.

[0064] Specifically, the proposed solution compares topological indicators with the specification parameter set item by item, avoiding the generalized processing of multiple indicators found in simple rule engines, and ensuring that each indicator participates independently in classification decisions. Since the specification parameter set fully defines the hierarchical relationship between the safe range, warning range, and permissible range, when both the horizontal and vertical clearances meet the safe range and the selectable indicators comply with the limitations, a compliant relationship is strictly determined, thus eliminating the risk of misjudgment caused by a single indicator meeting the standard. Given that the boundary interval covers the transition zone between the safe and warning ranges, a general warning relationship is triggered when any indicator falls into this interval, thus capturing potential risk states ignored in traditional binary classification. Considering that indicators exceeding the permissible range or intruding into boundaries directly represent spatial conflicts, a serious violation relationship is immediately determined, thereby achieving immediate identification of high-risk scenarios. This classification mechanism works in conjunction with the spatial topology index structure, based on the pipeline node connectivity relationships provided by the topology adjacency index and the constraint templates stored in the specification constraint index, ensuring that the classification process strictly follows the constraints of pipeline category and topology relationship type, forming a complete multi-level classification logic chain.

[0065] In some preferred embodiments, this application is implemented as follows: When a gas pipeline and a water supply pipeline form a candidate object pair within a target spatial block, the corresponding constraint template in the specification constraint index is read. Its specification parameter set specifies a horizontal clearance safety range of 0.5–1.0 meters and a warning range of 0.3–0.5 meters. If the calculated horizontal clearance is 0.45 meters and the vertical clearance is within the safety range, it is determined to be a general warning relationship because the horizontal clearance falls within the boundary interval. If the spatial location of the pipeline object extends into the road boundary constraint object, it is directly determined to be a serious violation relationship. This process quickly locates candidate object pairs using the spatial block index and verifies the connectivity of pipeline nodes using the topological adjacency index, ensuring that the classification results are consistent with the actual spatial topology.

[0066] Through the above technical solutions, this application improves the accuracy of preliminary topological relationship classification, avoids misjudgments caused by improper handling of boundary conditions, and provides reliable input features for the relationship classification model. It can accurately identify pipeline combinations in critical states, thereby ensuring the accuracy of the graph neural network correction process and ultimately enhancing the overall reliability of pipeline spatial topological relationship analysis.

[0067] Specifically, in some of the embodiments described above in this application, candidate object pairs, topological indices, and preliminary topological relationship classification results are proposed to be input into the relationship classification model. However, in its implementation, the preliminary topological relationship classification only relies on the local comparison of the specification parameter group and fails to integrate the global topological dependencies and spatial context information between pipe segment objects, pipe nodes, and constraint objects in the pipeline network. This results in the classification results lacking consideration of the overall consistency of the network, and is prone to misjudgment, especially in complex pipeline networks, and cannot accurately reflect the true state of the topological relationship.

[0068] In response, this application further proposes using candidate object pairs, topological indices, and preliminary topological relationship classification results as input features for relationship classification. When inputting these features into the relationship classification model, the following are included: In the relation classification model, pipe segment objects, pipeline nodes, and constraint objects are used as pipe segment nodes, pipeline nodes, and constraint nodes in the graph structure, respectively. Node features such as node type, spatial location, burial depth, and pipeline category are written for each node. Based on the topological adjacency index, connection edges representing physical connectivity are established between pipe segment nodes and pipeline nodes. Based on the spatial block index and the standard constraint index, neighboring edges representing spatial proximity are established between pipe segment nodes and between pipe segment nodes and constraint nodes. The corresponding topological relation type, horizontal clearance, and vertical clearance are written into each connection edge and neighboring edge. The topological indicators and preliminary topological relation classification results corresponding to each candidate object pair are used as the relation features of that candidate object pair, along with the node features and edge features, and input into the graph structure.

[0069] In practical applications, a pipe segment node refers to a basic unit of a graph structure that abstracts pipe segments in a pipeline network. This can be achieved by assigning a unique node identifier to each pipe segment object. The purpose is to enable the graph neural network to directly process the geometric and attribute features of pipe segment objects, thereby supporting feature transfer and aggregation. A pipeline node can be understood as an abstraction of connection points in a pipeline network as a graph node. This can be achieved by constructing nodes based on pipeline node coordinates and connecting pipe segment information. Its purpose is to accurately represent the topological connectivity at pipeline nodes, providing a structural foundation for modeling physical connectivity. Specifically, a constraint node refers to an abstraction of constraint objects such as building boundaries and road boundaries as graph nodes. This can be achieved by converting the spatial boundaries of constraint objects into node features. Its purpose is to incorporate external constraints into the graph structure, enabling the model to consider the spatial relationships between pipe segment objects and constraint objects. In practical applications, node features refer to a set of data describing the attributes of graph nodes. This can be implemented as a vector form including node type identifiers, spatial coordinates, burial depth values, and pipeline category codes. Its purpose is to provide contextual information for the nodes in the graph neural network, enabling the model to distinguish different categories of objects and their spatial characteristics. Connecting edges can be understood as graph edges representing the physical connectivity between pipe segment nodes and pipeline nodes. They can be constructed using pipeline node connection information recorded in the topological adjacency index, aiming to accurately model the actual connection logic in the pipeline network and ensure the structural accuracy of topological relationship analysis. Specifically, proximity edges represent the spatial proximity relationships between pipe segment nodes or between pipe segment nodes and constraint nodes. They can be constructed using proximity relationship information stored in the spatial block index and the canonical constraint index, aiming to capture spatial context information, such as potential interactions between pipe segment objects and boundary relationships with constraint objects. In practical applications, edge features refer to the dataset describing graph edge attributes. They can be implemented as vectors including topological relationship type encoding, horizontal net distance values, and vertical net distance values, aiming to quantify the strength of the relationships represented by the edges, enabling graph neural networks to perform inference based on precise distance parameters. Among them, relation features can be understood as a set of specific attributes of candidate object pairs. They can be implemented by encoding topological indicators and preliminary topological relation classification results into feature vectors. The purpose is to provide initial classification basis for graph neural networks, so that the model can correct local misjudgments during feature transmission.

[0070] Specifically, the proposed solution first maps pipe segment objects, pipe nodes, and constraint objects in the pipeline network to pipe segment nodes, pipe node nodes, and constraint nodes in a graph structure, respectively. For each node, node features including node type, spatial location, burial depth, and pipeline category are written, thus constructing the basic unit of the graph structure. Next, based on the topological adjacency index, connection edges representing physical connectivity are established between pipe segment nodes and pipe node nodes. Then, based on the spatial block index and the normative constraint index, neighboring edges representing spatial proximity are established between pipe segment nodes and between pipe segment nodes and constraint nodes, forming a complete graph connection relationship. Next, edge features representing the corresponding topological relationship type, horizontal clearance, and vertical clearance are written into each connection edge and neighboring edge to quantify the spatial relationship represented by the edge. Finally, the topological indices and preliminary topological relationship classification results corresponding to candidate object pairs are used as the relationship features of the candidate object pairs. These, along with node features and edge features, are input into the graph structure, enabling the graph neural network to perform feature transfer and aggregation based on global topological information and spatial context, thereby outputting the topological relationship classification result and topological consistency score. By constructing this graph structure and using this feature input method, the scheme integrates the global topological dependencies and spatial context information of the pipeline network, overcoming the limitations of preliminary classification methods that rely solely on local comparisons.

[0071] As a preferred embodiment, the specific implementation of this application is as follows: When analyzing the urban underground pipeline network, gas pipeline segments are mapped to segment nodes, pipeline junctions are mapped to pipeline nodes, and road boundary constraint objects are mapped to constraint nodes. For gas pipeline segment nodes, node type is written as gas pipeline, spatial location coordinates, burial depth value, and pipeline category characteristics. For pipeline nodes, node type is written as junction point, coordinates, and connecting segment information. For constraint nodes, node type is written as road boundary and spatial boundary characteristics. Based on the topological adjacency index, connecting edges are established between gas pipeline segment nodes and junction pipeline nodes. Based on the spatial block index and specification constraint index, adjacent edges are established between gas pipeline segment nodes and power cable pipeline segment nodes, and topological relationship types of parallel, horizontal clearance, and vertical clearance values ​​are written into the edges. The topological indicators and preliminary classification results of candidate gas pipeline segment and power cable pipeline segment pairs are used as relational features input to the graph structure. The graph neural network aggregates global information through feature transmission and outputs the topological relationship classification results.

[0072] Through the above scheme, this application integrates the global topological dependencies and spatial context information of pipeline networks, overcomes the defect that the classification results lack overall network consistency due to the reliance on local comparisons in the preliminary classification, improves the accuracy and reliability of topological relationship identification in complex pipeline networks, and avoids pipeline safety hazards caused by misjudgment.

[0073] In some of the embodiments described above in this application, a relation classification model is proposed to perform feature transfer and aggregation on the graph structure composed of pipe segment objects, pipeline nodes and constraint objects. However, in its implementation, a single-layer graph neural network may not be able to fully capture the topological dependencies of multi-hop neighbor nodes, resulting in insufficient recognition accuracy of complex spatial topological relationships, and lacking a quantitative evaluation mechanism for the reliability of classification results.

[0074] In response, this application further proposes a relation classification model that employs at least two layers of a graph neural network to sequentially perform feature transfer and aggregation on the graph structure, including: In each layer, the feature weights passed from neighboring nodes to the target node are controlled based on edge features.

[0075] After completing the multi-layer feature transfer and aggregation, aggregated features of the pipe segment nodes, pipeline nodes and constraint nodes related to the candidate object pair are extracted for each candidate object pair.

[0076] The aggregated features and the relational features of the candidate object pairs are input together into the classification output unit.

[0077] The classification output unit generates topological relationship classification results and topological consistency scores.

[0078] In practical applications, at least two layers in a graph neural network refer to a graph neural network with a depth of two or more layers. This can be implemented using graph convolutional networks, graph attention networks, or graph isomorphic networks. The purpose is to allow features to propagate more widely across the pipeline network graph through multi-layer stacking, thereby capturing long-distance topological dependencies. Edge feature control refers to dynamically adjusting the intensity of information transmission using edge attribute information. This can be implemented using attention mechanisms, weighted averaging, or gating mechanisms. The goal is to ensure that the feature transmission process is closely related to the physical properties of the pipeline, avoiding interference from irrelevant connections. Specifically, focused extraction refers to aggregating features only for nodes directly related to the current analysis object. This can be implemented using max pooling, average pooling, or attention pooling. The goal is to avoid redundancy in full-graph computation and improve the targeting of local topological relationship analysis. The classification output unit is the component used to generate the final classification decision. It can be implemented using fully connected neural network layers, softmax classifiers, or decision trees. The purpose is to fuse deep aggregated features and relational features to provide a reliable classification basis. In practical applications, topological consistency score is an indicator that quantifies the reliability of classification results. It can be implemented using confidence scores, probability distributions, or uncertainty measures. Its purpose is to provide a reliability measure for classification results and enhance adaptability in complex scenarios.

[0079] This application's scheme employs at least two layers of a graph neural network for sequential feature propagation and aggregation, enabling features to spread across multiple nodes in the pipeline network graph and thus capturing long-distance topological dependencies. In each layer, the weights of feature propagation are dynamically adjusted based on information such as the topological relationship type, horizontal clearance, and vertical clearance contained in the edge features, ensuring that high-risk connections receive higher attention. After multi-layer aggregation, for each candidate object pair, only the aggregated features of the relevant nodes are extracted, avoiding redundant computation. Subsequently, the deep aggregated features and the relationship features of the candidate object pairs are jointly input into the classification output unit, fusing explicit rule information and implicit topological patterns. Finally, the classification output unit generates topological relationship classification results and a topological consistency score, providing dual evidence and reliability assessment for classification decisions.

[0080] In some preferred embodiments, this application is implemented as follows: In scenarios where gas pipelines and power cables intersect, the relationship classification model first performs feature transfer on the graph structure through two layers of a graph neural network. In each layer, weights are controlled based on edge features (such as intersection angle and horizontal clearance) to strengthen features near the intersection point. Then, aggregated features of relevant nodes are extracted for the candidate object pair. The aggregated features and preliminary classification results are input into the classification output unit to generate a corrected classification result (such as compliance relationship) and a topological consistency score (such as high confidence). If the score is low, manual review is triggered.

[0081] Through the above technical solutions, this application improves the recognition accuracy of complex spatial topological relationships and provides a quantitative evaluation of the reliability of classification results, solving the problems that single-layer graph neural networks cannot fully capture multi-hop dependencies and lack a reliability mechanism.

[0082] In some of the embodiments described above in this application, a local recalculation mechanism is proposed to set a version identifier to realize pipeline network data updates. However, in its implementation, when the pipeline network data undergoes local changes, it is necessary to recalculate the topological relationship of all spatial blocks and candidate object pairs, resulting in wasted computing resources and low efficiency. Especially for large-scale pipeline networks, the full network traversal recalculation cannot reuse the existing results of the unchanged parts, making it difficult to meet real-time requirements.

[0083] In response, this application further proposes setting version identifiers for each spatial block and candidate object pair in the spatial block index, and determining the spatial topology analysis results of the pipeline network based on the pipeline network data update status.

[0084] In practical applications, version identification refers to the marking information used to identify changes in the status of pipeline network data. It can be implemented using techniques such as timestamps, hash values, or sequence numbers, with the aim of accurately tracking data changes in each spatial block and candidate object pair. Determining the analysis results based on pipeline network data updates can be understood as a mechanism for dynamically deciding the calculation scope based on version identification differences. Specifically, this may include comparing version identifications to distinguish between changed and stable areas, aiming to avoid recalculating the entire network and performing topology analysis only on the changed portions.

[0085] This application's solution sets version identifiers for each spatial block and candidate object pair in the spatial block index, enabling precise identification of the data status of each local area. When pipeline network data is updated, by comparing the differences in version identifiers, the changed areas and stable areas can be quickly distinguished. Therefore, only the spatial blocks with changed version identifiers and their candidate object pairs are used to perform topology index calculations and relationship classifications, avoiding redundant analysis of the entire network. Furthermore, the cached results of unchanged version identifiers, including topology relationship classification results and topology consistency scores, can be intelligently reused, significantly reducing redundant computation.

[0086] As a specific implementation method, the solution of this application is implemented as follows: In the urban underground pipeline network management system, when a gas pipeline undergoes a partial update, the change in the version identifier of the affected spatial block is automatically detected. Only the candidate objects in these changed blocks are recalculated for topological indicators and relationship classification. For other areas, the existing topological relationship classification results and topological consistency scores are directly reused, thereby quickly generating an updated spatial topological relationship analysis report.

[0087] Through the above technical solution, this application solves the problem of computational efficiency when pipeline network data undergoes local changes, reduces redundant computation, and is especially suitable for dynamic update scenarios of large-scale pipeline networks. It can meet real-time requirements while ensuring the continuity and consistency of analysis results.

[0088] In summary, this application extracts pipe segment objects, pipeline nodes, and constraint objects, and constructs a spatial topology index structure. This allows for the structured binding of pipeline categories, topology relationship types, and specification parameters at the index layer. Combined with precise selection of candidate object pairs and calculation of topology indicators based on constraint templates within the target spatial blocks, it eliminates the need for traversal-style geometric collision calculations across the entire pipeline network, reducing the search scope and computational complexity in large-scale pipeline networks. Furthermore, candidate object pairs, topology indicators, and preliminary topology relationship classification results are uniformly mapped to a graph structure. A graph neural network model is used for feature transfer and aggregation, outputting topology relationship classification results and a topology consistency score for each candidate object pair. This approach comprehensively utilizes local clearance information and the overall network topology pattern, improving the accuracy of violation identification under complex spatial relationships and boundary conditions. The topology consistency score provides a quantifiable risk reference for planning review, completion acceptance, and operation and maintenance decisions. By setting version identifiers for each spatial block and candidate object pair in the spatial block index, a rapid and dynamic analysis of the spatial topology of pipeline networks in scenarios with frequent incremental updates is achieved, taking into account the comprehensive requirements of computational efficiency, data scale, and topology identification accuracy.

[0089] Based on another preferred embodiment described above, see [link to preferred embodiment]. Figure 2 As shown, this embodiment provides a pipeline network spatial topology analysis system based on graph neural networks, used to apply the above-described pipeline network spatial topology analysis method based on graph neural networks, including: The acquisition unit is configured to acquire pipeline network data and extract pipe segment objects, pipeline nodes, and constraint objects.

[0090] The first analysis unit is configured to construct a spatial topology index structure, which includes a spatial block index, a topology adjacency index, and a specification constraint index. The spatial block index records the pipe segment objects and constraint objects within each spatial block. The topology adjacency index records the connectivity relationships and topology relationship types of adjacent pipe segment objects based on pipeline nodes. The specification constraint index stores constraint templates containing pipeline categories, topology relationship types, and specification parameters.

[0091] The second analysis unit is configured to select constraint templates from the normative constraint index based on the pipeline categories and topological relationship types recorded in the spatial topology index structure, determine candidate object pairs within the target spatial block, calculate topological indices for the candidate object pairs, and obtain preliminary topological relationship classification results based on the constraint templates. The topological indices include horizontal and vertical clearances.

[0092] The correction unit is configured to take candidate object pairs, topological indicators, and preliminary topological relationship classification results as input features for relationship classification, input them into the relationship classification model, and the relationship classification model uses a graph neural network model to perform feature transfer and aggregation on the graph structure composed of pipe segment objects, pipeline nodes, and constraint objects, and outputs the topological relationship classification results and topological consistency score.

[0093] The update unit is used to set version identifiers for each spatial block and candidate object pair in the spatial block index, and to determine the spatial topology analysis results of the pipeline network based on the update status of the pipeline network data.

[0094] Understandably, the synergy between the multi-layered index structure and the correction mechanism effectively addresses the computational efficiency bottleneck caused by full network traversal in existing technologies and the accuracy deficiency of local rule engines in identifying complex topological relationships. Specifically, the core innovation of this embodiment lies in combining a multi-layered spatial topology index structure with a constraint template management mechanism, introducing the feature transfer and aggregation capabilities of graph neural networks for graph structures, and a local recalculation mechanism based on version identifiers. This reduces the computational scope while enhancing the global dependency of topological relationship identification and supporting efficient incremental updates, thereby improving the efficiency and accuracy of spatial topology relationship analysis in large-scale pipeline networks.

[0095] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for analyzing the spatial topology of pipeline networks based on graph neural networks, characterized in that, include: Collect pipeline network data and extract pipe segment objects, pipeline nodes, and constraint objects; A spatial topology index structure is constructed, which includes a spatial block index, a topology adjacency index, and a specification constraint index. The spatial block index records the pipe segment objects and constraint objects within each spatial block. The topology adjacency index records the connectivity relationship and topology relationship type of adjacent pipe segment objects based on the pipeline nodes. The specification constraint index stores constraint templates containing pipeline category, topology relationship type, and specification parameters. Based on the pipeline categories and topology relationship types recorded in the spatial topology index structure, a constraint template is selected from the specification constraint index, candidate object pairs are determined within the target spatial block, topology indices are calculated for the candidate object pairs, and preliminary topology relationship classification results are obtained based on the constraint template. The topology indices include horizontal clearance and vertical clearance. The candidate object pairs, topological indicators, and preliminary topological relationship classification results are used as input features for relationship classification. These are then input into the relationship classification model, which employs a graph neural network model to perform feature transfer and aggregation on the graph structure composed of the pipe segment objects, pipeline nodes, and constraint objects. The model outputs the topological relationship classification results and the topological consistency score. Version identifiers are set for each spatial block and candidate object pair in the spatial block index, and the spatial topology analysis results of the pipeline network are determined based on the pipeline network data update.

2. The pipeline network spatial topology analysis method based on graph neural networks according to claim 1, characterized in that, When collecting pipeline network data and extracting pipe segment objects, pipeline nodes, and constraint objects, the following is included: The system collects design drawings, as-built drawings, surveying data, building information modeling (BIM) data, and operation and maintenance log data. It extracts spatial and attribute data representing pipeline centerlines, pipeline connection components, and constraint objects. Through coordinate transformation and elevation benchmark unification, the various types of data are transformed into a unified spatial coordinate system. In the unified spatial coordinate system, the continuous pipeline centerline is divided into several pipe segment objects at the turning points, intersections, and equipment connection points of the pipeline centerline according to preset division rules. Corresponding pipeline nodes are generated at each division point. Duplicate representations are eliminated through matching rules of spatial position overlap and attribute consistency. Building boundaries, road boundaries, and protected area boundaries are extracted as constraint objects.

3. The pipeline network spatial topology analysis method based on graph neural networks according to claim 1, characterized in that, When constructing the spatial topology index structure, the following are included: The area where the pipeline network is located is divided into multiple spatial blocks according to a preset spatial resolution and height range. A block record is established for each spatial block in the spatial block index. The pipe segment object identifier, constraint object identifier and corresponding spatial boundary range information located in or intersecting with the spatial block are registered in each block record. In the topology adjacency index, a node record is established for each pipeline node. In each node record, the identifier of the pipe segment object connected to the pipeline node is registered. The corresponding topology relationship type is determined according to the connection direction, horizontal distance, vertical distance and overlap along the pipeline segment object and written into the node record.

4. The pipeline network spatial topology analysis method based on graph neural networks according to claim 3, characterized in that, When constructing the spatial topology index structure, the following are also included: Based on the design specification text and pipeline laying standards, the specification clauses involving distance constraints between pipeline categories and between pipeline categories and constraint objects are extracted. Each specification clause is parsed into a constraint template. Each constraint template includes at least a first pipeline category, a second pipeline category or constraint object category, an applicable topology relationship type, and a specification parameter group. The constraint template is stored in the specification constraint index using the first pipeline category, the second pipeline category or the constraint object category, and the topology relationship type as search keys. The specification parameter group includes the horizontal clearance range, the vertical clearance range, the burial depth difference range, the upper limit of the overlap length along the pipeline, or the intersection angle range.

5. The pipeline network spatial topology analysis method based on graph neural networks according to claim 4, characterized in that, Based on the pipeline categories and topology relationship types recorded in the spatial topology index structure, when selecting constraint templates from the specification constraint index and determining candidate object pairs within the target spatial block, the process includes: Within the target spatial block, the identifiers of pipe segment objects and constraint objects located within or intersecting with the target spatial block are read based on the spatial block index. Combined with the connectivity relationships and topological relationship types of adjacent pipe segment objects corresponding to the pipeline nodes recorded in the topology adjacency index, combinations of pipe segment objects and pipe segment objects or combinations of pipe segment objects and constraint objects that satisfy the constraint template constraint conditions are selected according to the first pipeline category, second pipeline category, or constraint object category in the constraint template and the applicable topological relationship type. The selected combinations are taken as candidate object pairs. For each candidate object pair, the horizontal and vertical clearances are calculated based on the endpoint positions of the pipe segment objects, the burial depth of the pipe segment objects, and the boundary positions of the constraint objects recorded in the spatial block index and the topology adjacency index. If necessary, the spatial nearest distance and the length of overlap along the path of the candidate object pair are further calculated. The horizontal clearance, the vertical clearance, and the optional spatial nearest distance and length of overlap along the path are used as the topological indicators of the candidate object pair.

6. The pipeline network spatial topology analysis method based on graph neural networks according to claim 5, characterized in that, Preliminary topological relationship classification results are obtained based on the constraint template, including: For each candidate object pair, the topology index is compared item by item with the set of standard parameters specified in the constraint template. When the horizontal clearance and the vertical clearance are both within the safe range of the corresponding set of standard parameters and the selectable closest spatial distance and the length of overlap along the path meet the requirements of the set of standard parameters, the preliminary topology relationship classification result of the candidate object pair is determined as a compliant relationship. When at least one of the topology indicators is located within the boundary between the safety range and the preset warning range of the corresponding specification parameter group, the preliminary topology relationship classification result of the candidate object pair is determined as a general warning relationship; When any of the topology indicators exceeds the allowable range of the corresponding specification parameter group or when the pipe segment object intrudes into the boundary of the protected area, the preliminary topology relationship classification result of the candidate object pair is determined as a serious violation relationship.

7. The pipeline network spatial topology analysis method based on graph neural networks according to claim 6, characterized in that, When the candidate object pairs, the topological indices, and the preliminary topological relationship classification results are used as relationship classification input features and input into the relationship classification model, the following are included: In the relationship classification model, the pipe segment object, the pipeline node, and the constraint object are respectively used as the pipe segment node, pipeline node, and constraint node in the graph structure. Node features such as node type, spatial location, burial depth, and pipeline category are written for each node. Connection edges representing physical connectivity are established between the pipe segment node and the pipeline node based on the topological adjacency index. Neighboring edges representing spatial proximity are established between pipe segment nodes and between pipe segment nodes and constraint nodes based on the spatial block index and the standard constraint index. The corresponding topological relationship type, horizontal clearance, and vertical clearance are written into each connection edge and neighboring edge. The topological index corresponding to each candidate object pair and the preliminary topological relationship classification result are used as the relationship features of that candidate object pair, along with the node features and the edge features, and input into the graph structure.

8. The pipeline network spatial topology analysis method based on graph neural networks according to claim 7, characterized in that, The relationship classification model employs at least two layers of graph neural networks to sequentially perform feature transfer and aggregation on the graph structure. In each layer, the feature weights transferred from adjacent nodes to the target node are controlled based on the edge features. After completing multi-layer feature transfer and aggregation, aggregated features of pipe segment nodes, pipeline nodes, and constraint nodes related to the candidate object pair are extracted for each candidate object pair. The aggregated features and the relationship features of the candidate object pair are input into the classification output unit, which generates the topology relationship classification result and the topology consistency score.

9. The pipeline network spatial topology analysis method based on graph neural networks according to claim 1, characterized in that, When setting version identifiers for each spatial block and candidate object pair in the spatial block index, and determining the spatial topology analysis results of the pipeline network based on the pipeline network data update status, the following steps are included: When pipeline network data is updated, topology index calculations and relationship classifications are performed on the spatial blocks whose version identifiers have changed and the candidate object pairs within them to obtain the spatial topology relationship analysis results of the pipeline network. For the candidate object pairs whose version identifiers have not changed, the topology relationship classification results and topology consistency scores are reused and cached to obtain the spatial topology relationship analysis results of the pipeline network.

10. A pipeline network spatial topology analysis system based on graph neural networks, used to apply the pipeline network spatial topology analysis method based on graph neural networks as described in any one of claims 1-9, characterized in that, include: The acquisition unit is configured to acquire pipeline network data and extract pipe segment objects, pipeline nodes, and constraint objects. The first analysis unit is configured to construct a spatial topology index structure, which includes a spatial block index, a topology adjacency index, and a specification constraint index. The spatial block index records pipe segment objects and constraint objects within each spatial block. The topology adjacency index records the connectivity relationships and topology relationship types of adjacent pipe segment objects based on the pipeline nodes. The specification constraint index stores constraint templates containing pipeline categories, topology relationship types, and specification parameters. The second analysis unit is configured to select a constraint template from the specification constraint index based on the pipeline category and topology relationship type recorded in the spatial topology index structure, determine candidate object pairs within the target spatial block, calculate topology indices for the candidate object pairs, and obtain preliminary topology relationship classification results based on the constraint template. The topology indices include horizontal clearance and vertical clearance. The correction unit is configured to take the candidate object pairs, topological indicators and preliminary topological relationship classification results as relationship classification input features and input them into the relationship classification model. The relationship classification model uses a graph neural network model to perform feature transfer and aggregation on the graph structure composed of the pipe segment objects, pipeline nodes and constraint objects, and outputs the topological relationship classification results and topological consistency score. The update unit is used to set version identifiers for each spatial block and candidate object pair in the spatial block index, and to determine the spatial topology analysis results of the pipeline network based on the pipeline network data update status.

Images