Pipe network model automatic construction and correction method based on graph structure state information transmission

Abstract

An automatic pipeline network model construction and correction method based on graph structure state information transmission is proposed. This method first automatically parses engineering drawings or databases, intelligently identifies and matches pipeline and node parameters, and constructs an initial pipeline network model. Then, the model is abstracted into a graph structure, and through a training-free state information transmission algorithm, iterative message passing and node state updates are performed on the topology network based on engineering constraint violations to generate globally coordinated correction priors. Finally, the corrected model is output through constraint fusion solution. This invention realizes full-process automation of pipeline network modeling and correction, overcomes the shortcomings of traditional methods such as difficult annotation matching, low efficiency, and inconsistency in local corrections, and can significantly improve the efficiency of model construction and the overall quality and rationality under complex topology conditions.

Description

Technical Field

[0001] This invention belongs to the technical field of urban drainage engineering and flood numerical model construction, specifically involving an automatic construction and correction method for pipeline network models based on graph structure state information transmission. Background Technology

[0002] With the continuous expansion of urban stormwater and sewage drainage systems, numerical models of drainage pipe networks are widely used in planning, operation analysis, risk assessment, urban flood control, and urban renewal. Current pipe network model building processes typically involve manually retrieving information such as pipeline locations, node locations, pipe diameter markings, and node elevation markings from vector electronic engineering drawings or detailed databases, and then manually converting this information into the structured data required for the pipe network model. However, existing technologies have the following problems:

[0003] 1) The modeling process relies on manual operation, which limits efficiency and consistency.

[0004] Because engineering drawings contain a large number of pipelines and nodes, with a wide distribution range and complex pipeline topology, relevant attribute parameters (such as pipe diameter and node elevation) are usually marked in a scattered manner, which leads to a large amount of manual identification, matching and input during the model building process, making it difficult to meet the requirements of speed and standardization.

[0005] 2) The relationship between annotation information and object entities is complex, which can easily lead to mismatches or omissions.

[0006] In engineering drawings, key attributes such as pipe diameter and node elevation are often expressed in text form and may correspond to target entities through leader lines, associated line segments, or local layout rules. Due to variations in scale, overlapping annotations, dense leader lines, or non-standard labeling in drawings, matching methods based on distance thresholds or simple proximity rules are prone to mismatches, affecting the accuracy of parameter extraction and the reliability of modeling.

[0007] 3) Initial models are prone to engineering consistency issues, increasing the workload of subsequent verification and correction.

[0008] The initial model generated from the error of drawing analysis and parameter matching may have phenomena such as reverse slope, local elevation changes, discontinuity of upstream and downstream parameters, and dead nodes, which will affect the structural consistency and usability of the model, and require additional verification and correction work.

[0009] 4) Existing correction methods mostly use preset rules for local adjustments, making it difficult to guarantee overall consistency under complex topology conditions.

[0010] Existing model correction techniques typically use preset rules to adjust local parameters. However, under complex topology conditions such as multi-branch convergence and long-link transmission, local correction may cause upstream and downstream cascading effects, resulting in insufficient overall consistency and difficulty in stably meeting the requirements of engineering rationality.

[0011] Therefore, there is an urgent need for a closed-loop technical solution that can automatically construct municipal stormwater and sewage pipe network models from engineering drawings and automatically diagnose and correct model anomalies, so as to improve modeling efficiency and model quality. Summary of the Invention

[0012] To overcome the shortcomings of the prior art, the purpose of this invention is to provide an automatic construction and correction method for pipeline network models based on graph structure state information transmission. This method solves the problems in the process of generating pipeline network models from engineering drawings or databases, such as difficulty in annotation matching, insufficient accuracy of parameter extraction, low modeling efficiency, and many anomalies in the initial model, as well as the reliance on manual correction. It features automatic construction of pipeline network models, improved modeling quality and accuracy, and enhanced model structure consistency and engineering rationality.

[0013] To achieve the above objectives, the technical solution adopted by this invention is: an automatic construction and correction method for pipeline network models based on graph structure state information transmission, comprising the following steps:

[0014] Step 1: Obtain the original data source of the municipal stormwater and sewage pipe network, parse the original data source to obtain the pipeline entity set, node entity set, text annotation set, and lead line or associated line segment set; when the original data source is a pipe network database, read the pipe segment table, node table and their attribute fields and establish the connection relationship between pipe segments and nodes; combine the parsed pipeline entity set, node entity set, text annotation set, and lead line or associated line segment set into a unified set of original pipe network objects;

[0015] The original data source includes either vector engineering electronic drawing files or a pipeline database.

[0016] Step 2: Construct a spatial index structure based on the original object set obtained in Step 1 to support nearest neighbor retrieval and candidate set generation, and preprocess the parsing results of the original data source;

[0017] Step 3: Generate a set of candidate annotations for cross-sectional parameters for each pipe segment; score and sort the candidate annotations to determine the cross-sectional parameters; and output the source markers and confidence scores of the cross-sectional parameters; perform unit conversion and format standardization on the cross-sectional parameters.

[0018] Step 4: Generate a set of candidate elevation annotations for each node, parse the format of the candidate elevation annotation set to obtain the ground elevation, node bottom elevation and burial depth fields, and assign values ​​according to the score; when there are competing annotation matches or conflicting elevation values, perform conflict resolution, output the elevation source mark and elevation assignment confidence; fill in missing fields according to the consistency rules.

[0019] Step 5: Construct the pipeline network topology based on the connection relationship between nodes and pipe segments, and generate an initial pipeline network model containing a set of nodes, a set of pipe segments, and their attribute parameters;

[0020] Step 6: Perform consistency verification on the initial pipeline model, identify anomaly types, and perform rule correction based on preset engineering constraints to obtain the rule-corrected initial pipeline model; the set of engineering constraints established during the rule correction process includes at least one or more of the following: minimum slope constraint, continuity constraint, and connectivity constraint.

[0021] Step 7: Abstract the initial pipeline network model after rule correction into graph structure data G=(V,E), construct a node state vector containing node elevation state, confidence level, and source label, and also construct an edge attribute vector containing pipe segment attributes and direction confidence level; perform training-free state information transfer iterative inference on the graph structure data to obtain the node bottom elevation correction prior and pipe segment anomaly confidence level.

[0022] Step 8: Combine the corrected prior from Step 7 with the set of engineering constraints from Step 6 to solve the problem. Obtain the final corrected result that satisfies the engineering constraints through constraint projection or constraint optimization, and output the corrected pipeline network model.

[0023] In step 2, the spatial index structure is an R-tree index, which includes at least an index for the geometric position of the outer frame of text annotations and an index for the geometric objects of pipelines and nodes.

[0024] In step 3, the candidate annotation set for cross-section parameters includes at least: strongly associated candidate annotations obtained by pointing to the leader or associated line segment, weakly associated candidate annotations obtained by spatial proximity retrieval, and regular candidate annotations that satisfy the cross-section format rules.

[0025] In step 3, the scoring is based on at least two of the following: distance cost, lead connectivity consistency, text format matching degree, and layer consistency.

[0026] The scoring described in step 4 is the same as the scoring described in step 3.

[0027] In step 4, the conflict resolution includes: when the difference in candidate annotation scores is greater than a preset threshold, the highest-scoring candidate is adopted; when the difference in scores is less than a preset threshold, it is marked as uncertain and the alternative value is retained, while the confidence field is output.

[0028] In step 4, the consistency rule completion includes: the consistency completion rules include: when the ground elevation and burial depth are known but the node bottom elevation is missing, completion is calculated based on the relationship that the node bottom elevation equals the ground elevation minus the burial depth; when the node bottom elevation and burial depth are known but the ground elevation is missing, completion is calculated based on the relationship that the ground elevation equals the node bottom elevation plus the burial depth; when both the ground elevation and the node bottom elevation are known but the burial depth is missing, completion is calculated based on the relationship that the burial depth equals the ground elevation minus the node bottom elevation; the completion result is written to the corresponding field, and the source marker of the field is set to completion, while the confidence level of the field is set to be lower than the confidence level obtained by directly parsing the text annotation.

[0029] In step 5, determining the direction of the pipe segment includes: determining the direction based on the flow direction identifier on the drawing or the upstream and downstream fields in the database; determining the direction from high to low based on the bottom elevation of the node when the flow direction identifier is missing; and outputting the direction confidence field to characterize the reliability of the direction determination.

[0030] In step 6, the anomaly types include at least two of the following: reverse slope anomaly, node elevation change anomaly, dead-end node or isolated subgraph anomaly, discontinuity of upstream and downstream cross-sectional parameters anomaly, and end node discharge condition anomaly.

[0031] In step 7, the state information transmission iterative inference includes: constructing messages based on engineering constraint violations and performing neighborhood aggregation, and sequentially updating the node bottom elevation state; wherein the update weight is jointly determined by the node elevation confidence and the pipe segment direction confidence, so as to achieve the effect of prioritizing the adjustment of low confidence objects in the iteration; at the same time, one of the state change amount being less than a threshold and reaching the maximum number of iterations is used as the stopping condition.

[0032] In step 8, the fusion solution aims to minimize the deviation between the final correction amount and the correction prior, and performs constraint projection or constraint optimization under the engineering constraint set, which includes at least two of the following: slope constraint, continuity constraint, and connectivity constraint.

[0033] The beneficial effects of this invention are:

[0034] This invention achieves anomaly identification and elevation consistency correction through a closed-loop process of "constraint violation construction - message passing - neighborhood aggregation - state iteration update - convergence criterion - constraint fusion projection" without training data, and outputs a pipeline model that meets engineering constraints. The key features of this invention are its focus on pipeline network model construction and quality improvement. It is compatible with multi-source data input, including electronic vector engineering drawings and pipeline network databases, and can automatically parse and form structured data of nodes and pipe segments to construct a pipeline network topology model. Based on this, the pipeline network is abstracted into a graph structure with nodes as vertices and pipe segments as edges. Node state vectors and edge attribute vectors are constructed, and training-free state information transmission and inference are performed on the graph structure. Prior information for anomaly identification and parameter correction is generated through an iterative process of "constraint violation construction—message passing—neighborhood aggregation—state iteration update—convergence criterion." Furthermore, this prior information is fused with engineering consistency rules for constraint processing. Constraint projection or constraint optimization is used to obtain the final correction result that satisfies the engineering constraints, forming a closed-loop process of "automatic modeling—state inference—fusion correction—result output." This improves the overall consistency and engineering rationality of the pipeline network model under complex topology conditions and reduces the workload of manual verification and correction. The specific advantages of this invention are as follows:

[0035] 1) This invention achieves a high degree of automation in the modeling process, significantly improving modeling efficiency and consistency: By automatically parsing vector engineering electronic drawing files and pipeline databases, it intelligently identifies objects such as pipelines, nodes, and text annotations. Utilizing spatial indexing and multi-factor scoring matching mechanisms, it automatically and accurately assigns values ​​to pipe section cross-sectional parameters and node elevation parameters. This automated process fundamentally changes the traditional modeling method that relies on manual identification, matching, and input, avoiding the inefficiencies and poor consistency caused by cumbersome manual operations and inconsistent standards. It enables the rapid and standardized generation of initial pipeline models.

[0036] 2) Improved accuracy and robustness of parameter matching in complex scenarios: Addressing real-world challenges such as dense annotations, complex leader lines, and non-standard annotations in drawings, this invention constructs a candidate annotation set including strong correlations, weak correlations, and rule-based candidates. It then comprehensively considers multiple dimensions such as distance cost, leader line connectivity consistency, text format matching degree, and layer consistency for scoring and conflict resolution, effectively reducing the risk of parameter mismatch and omission. Simultaneously, the output parameter source label and assignment confidence provide a reliable basis for subsequent processing, significantly improving the accuracy of parameter extraction and its adaptability to complex drawings.

[0037] 3) This invention achieves global and collaborative correction of model anomalies, ensuring overall engineering rationality: The core innovation lies in abstracting the rule-corrected pipeline network model into a graph structure and performing training-free state information transmission iterative inference. Based on engineering constraint violations, this method performs message passing and neighborhood aggregation on the pipeline network topology, enabling it to perceive the cascading impact of local anomalies on the entire pipeline system. Through a strategy of "prioritizing the adjustment of low-confidence objects," node states are iteratively updated, ultimately generating a corrected prior that comprehensively considers global consistency. This graph-based global inference method overcomes the limitations of existing technologies that can only perform local rule corrections, enabling collaborative optimization under complex topology conditions (such as multi-branch merging and long-link transmission), thereby significantly improving the overall consistency and engineering rationality of the corrected pipeline network model.

[0038] 4) A complete "automatic construction-diagnosis-correction" closed loop is formed, greatly reducing the workload of manual verification: This invention integrates the entire process from data parsing, parameter assignment, topology construction, anomaly diagnosis to intelligent correction, forming a complete automated closed-loop solution. Through this method, a high-quality pipeline network model that meets basic engineering constraints such as slope, continuity, and connectivity, along with detailed correction logs, can be directly output. This not only greatly reduces the reliance on the manual experience of professional engineers during model construction but also frees technical personnel from the heavy and error-prone work of later model verification and correction, effectively reducing labor and time costs. Attached Figure Description

[0039] Figure 1 This is a flowchart of a method for automatically constructing and correcting a pipeline network model based on graph structure state information transmission, according to the present invention.

[0040] Figure 2 This is a vector engineering plan of a typical drainage area pipe network in an embodiment of the present invention.

[0041] Figure 3 This is a schematic diagram of the R-Tree spatial index in an embodiment of the present invention.

[0042] Figure 4 This is the effect of the method construction and modification in the embodiments of the present invention. Detailed Implementation

[0043] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0044] See Figure 1An automatic construction and correction method for pipeline network models based on graph structure state information transmission is proposed. Building upon automatic modeling and rule correction, an iterative inference module based on graph structure state information transmission is introduced to perform global inference of the pipeline network topology. It outputs corrected priors for nodes and pipe segments, and combines these with engineering constraints to form a closed-loop correction result. The method includes the following steps:

[0045] Step 1: Parsing and object recognition of vector engineering electronic drawings (or databases)

[0046] Obtain the original data source of the municipal stormwater and sewage pipe network, parse the original data source to obtain the set of pipeline entities, the set of node entities, the set of text annotations, and the set of lead lines or associated line segments; when the original data source is a pipe network database, read the pipe segment table, the node table and their attribute fields and establish the connection relationship between pipe segments and nodes; and uniformly construct the parsed set of pipeline entities, the set of node entities, the set of text annotations, and the set of lead lines or associated line segments into a set of original pipe network objects.

[0047] The original data source includes either vector engineering electronic drawing files or a pipeline database.

[0048] The set of text annotations is used to characterize pipe diameter, node elevation, burial depth, or other attributes;

[0049] The parsing process includes reading the pipe segment table, node table and their attribute fields from the database, and completing the construction of the pipe segment-node connection relationship to form structured input data for subsequent topology modeling and automatic correction.

[0050] Step 2: Multi-source attribute association and spatial index construction

[0051] A spatial index structure is constructed based on the original set of objects from the official website obtained in step 1 to support nearest neighbor retrieval and candidate set generation, and the parsing results of the original data source are preprocessed.

[0052] The spatial index structure is an R-tree index, which includes at least an index for the geometric position of the bounding box of text annotations, and an index for the geometric objects of pipelines and nodes.

[0053] The parsing results of the original data source are preprocessed, including at least: merging similar entities (merging multiple symbols of the same node), cleaning up duplicate line segments, removing zero-length objects, and mapping to a unified object type by layer or symbol library.

[0054] Step 3, Obtaining and Standardizing Pipe Section Cross-Section Parameters

[0055] For each pipe segment, a candidate annotation set for cross-sectional parameters is generated. The candidate annotation set for cross-sectional parameters is scored and sorted to determine the cross-sectional parameters, and the source markers of the cross-sectional parameters and the confidence level of the cross-sectional assignment are output. The units of the cross-sectional parameters are converted and the format is standardized.

[0056] Strongly related candidates: Text annotations that point from the endpoint of a leader or related line segment to a pipe segment or fall within the pipe segment's buffer zone;

[0057] Weakly related candidates: Textual annotations retrieved by spatial indexing within a preset radius around representative sampling points in the pipe segment;

[0058] Rule candidates: Textual annotations that meet the cross-section format rules (e.g., format patterns, unit and symbol rules for pipe diameter / cross-section dimensions).

[0059] Candidate annotations are scored, with the score obtained by a linear or nonlinear combination of at least the following factors: distance cost, lead line connectivity consistency, text format matching degree, text orientation consistency with pipe segment direction, and layer consistency; the highest score is selected as the source of cross-sectional parameters, and the source of cross-sectional parameters and the assignment confidence are recorded; unit conversion and format standardization are performed on the cross-sectional parameters.

[0060] Step 4: Obtaining and Consistently Organizing Node Elevation Parameters

[0061] For each node, a set of candidate elevation annotations is generated. The format of the candidate elevation annotation set is parsed to obtain the ground elevation, node bottom elevation, and burial depth fields, and values ​​are assigned according to a scoring system. When there are competing matching or conflicting elevation values, conflict resolution is performed, and the elevation source marker and elevation assignment confidence level are output. Missing fields are completed according to consistency rules, including: when the ground elevation and burial depth are known but the node bottom elevation is missing, completion is calculated based on the relationship "node bottom elevation equals ground elevation minus burial depth"; when the node bottom elevation and burial depth are known but the ground elevation is missing, completion is calculated based on the relationship "ground elevation equals node bottom elevation plus burial depth"; when both the ground elevation and node bottom elevation are known but the burial depth is missing, completion is calculated based on the relationship "burial depth equals ground elevation minus node bottom elevation". The completed result is written to the corresponding field, and the source marker of that field is set to complete. Simultaneously, the confidence level of that field is set to be lower than the confidence level obtained by directly parsing the text annotation.

[0062] When the score difference is greater than the threshold, the highest score candidate is selected.

[0063] When the score difference is less than the threshold, it is marked as uncertain, the confidence field is output and alternative values ​​are retained;

[0064] Step 5: Construct the pipeline network topology and generate the initial model

[0065] The pipeline network topology is constructed based on the connection relationship between nodes and pipe segments, and an initial pipeline network model containing a set of nodes, a set of pipe segments, and their attribute parameters is generated.

[0066] The starting and ending nodes of each pipe segment are determined by node number association or spatial connection relationship, and a topology map data structure is generated for subsequent verification and correction.

[0067] Step 6: Initial model anomaly diagnosis and rule correction

[0068] Perform a consistency check on the initial pipeline model generated in step 5, identify the anomaly types in the initial pipeline model, and perform rule corrections based on preset engineering constraints to obtain the rule-corrected initial pipeline model.

[0069] The anomaly types include at least the following categories: pipe section reverse slope anomaly, node elevation change anomaly, starting node anomaly, end node discharge condition anomaly, upstream and downstream cross-sectional parameter discontinuity anomaly, dead end node anomaly, redundant node anomaly, and abnormal branch pipe structure anomaly.

[0070] The anomalies are automatically corrected based on preset engineering constraints to obtain the initial pipeline network model after rule correction. The set of engineering constraints established during the rule correction process includes at least one or more of the following: minimum slope constraint, continuity constraint, and connectivity constraint.

[0071] Step 7: Iterative inference and correction of prior output based on graph structure state information transmission

[0072] The initial pipeline network model obtained in step 6 after rule correction is abstracted into graph structure data G=(V,E), where V is the set of nodes, E is the set of pipe segments, nodes are graph vertices, and pipe segments are graph edges. For each node... Construct a state vector, which includes the node bottom elevation. Ground elevation burial depth and its source tag and confidence level; for each edge Construct an attribute vector that includes at least the pipe segment length. Section parameters, current slope and direction confidence;

[0073] Construct a node state vector containing node elevation status and confidence source label, and an edge attribute vector containing pipe segment attributes and direction confidence; perform training-free state information transfer iterative inference on graph structure data to obtain one or more of the following: corrected prior values ​​of node bottom elevation and confidence of pipe segment anomalies.

[0074] Performing iterative information passing on graph-structured data includes the following sub-steps:

[0075] 1) Information calculation: For each edge Calculate information from neighboring nodes to the target node. The information is determined by the states of adjacent nodes, edge attributes, and engineering constraint functions.

[0076] 2) State Update: Aggregate information from the neighborhood according to a preset aggregation function and update the node state. The update weight can be determined by the confidence of nodes and edges, so that the low-confidence elevation field is adjusted first during the update.

[0077] 3) Stopping condition: Stop when the change in node state is less than the threshold or the maximum number of iterations is reached, and obtain the node bottom elevation correction and the confidence degree of the slope aspect anomaly of the pipe segment as the correction prior.

[0078] The message and update delivery rules based on slope feasibility are as follows:

[0079] The revised pipeline network model is represented as graph-structured data G=(V, E), where the set of nodes is V and the set of pipe segments is E. For any pipe segment... Based on the bottom elevation of the node With pipe section length L e Construct minimum slope engineering constraints, requiring the bottom elevation difference between the upstream and downstream nodes to satisfy the minimum slope i. min (Formula 1):

[0080]

[0081] in: Let i be the elevation of the node base. min For the minimum allowable slope, L e This refers to the length of the pipe section.

[0082] When the current model does not satisfy equation (1), the constraint violation amount of the pipe segment in the sense of minimum slope is defined. (Equation 2)

[0083]

[0084] Furthermore, based on the violation quantity, the message transmitted from the neighboring node to the target node is constructed. The weight This is used to reflect the confidence level of the pipe segment direction and the reliability of the source of the cross-sectional elevation information, so that objects with low confidence or uncertainty may receive greater correction drive in subsequent iterations (Equation 3):

[0085]

[0086] In the formula, Adjacent nodes To the target node The message being delivered Information weight.

[0087] For the target node to its neighborhood All messages are aggregated using a summation aggregation function (Equation 4).

[0088]

[0089] Then, the node bottom elevation status is updated based on the aggregation results. To reflect the strategy of "prioritizing low confidence adjustments," a node update coefficient is introduced. It is determined by the confidence level of the node elevation field; when the node elevation confidence level is low, a larger value is taken. This allows for the priority adjustment of uncertain nodes to meet engineering constraints. The update of the node bottom elevation from the kth iteration to the (k+1)th iteration can be expressed as equation (5):

[0090]

[0091] The above information transmission and status update process is executed iteratively. The process continues until the iterative change in the bottom elevation of all nodes is less than a preset threshold. The iteration is considered to have reached convergence and stopped (Equation 6), or it is forcibly stopped when the maximum number of iterations is reached.

[0092]

[0093] After the iteration stops, it can be... Obtain the prior for the correction of the node bottom elevation. This is then used as a priori input for subsequent engineering constraint fusion solutions. This is to further ensure that the final result simultaneously satisfies the set of engineering constraints such as slope, continuity, and connectivity. In the fusion phase, the objective is to minimize the deviation between the prior correction and the final correction, and this is achieved within the constraint set. Perform constrained projection or constrained optimization to obtain the final correction amount.

[0094]

[0095] in, To correct the prior values ​​of the node bottom elevation output by iterative inference, To integrate the final correction obtained from the solution, the first term of the objective function is used to constrain the final correction to be as close as possible to the prior correction. This represents constraints that include slope, continuity, and connectivity, specifically for any pipe segment. satisfy:

[0096]

[0097] In the formula, This represents the two endpoints of pipe segment (edge) e. As an upstream node, For downstream nodes; This indicates that the constraint holds for all pipe segments in the network. E is the set of pipe segments.

[0098] Step 8: Combine the corrected prior from Step 7 with the set of engineering constraints from Step 6 to solve the problem. Obtain the final corrected result that satisfies the engineering constraints through constraint projection or constraint optimization, and output the corrected pipeline network model.

[0099] In step 2, the spatial index structure is an R-tree index, which includes at least an index for the geometric position of the outer frame of text annotations and an index for the geometric objects of pipelines and nodes.

[0100] In step 3, the candidate annotation set for cross-section parameters includes at least: strongly associated candidate annotations obtained by pointing to the leader or associated line segment, weakly associated candidate annotations obtained by spatial proximity retrieval, and regular candidate annotations that satisfy the cross-section format rules.

[0101] In step 3, the scoring is based on at least two of the following: distance cost, lead connectivity consistency, text format matching degree, and layer consistency.

[0102] The scoring described in step 4 is the same as the scoring described in step 3.

[0103] In step 4, the conflict resolution includes: when the difference in candidate annotation scores is greater than a preset threshold, the highest-scoring candidate is adopted; when the difference in scores is less than a preset threshold, it is marked as uncertain and the alternative value is retained, while the confidence field is output.

[0104] In step 4, the consistency rule completion includes: calculating the bottom elevation based on the ground elevation and burial depth, or calculating the ground elevation based on the bottom elevation and burial depth, and setting the source mark of the calculated elevation value as "calculation".

[0105] In step 5, determining the direction of the pipe segment includes: determining the direction based on the flow direction identifier on the drawing or the upstream and downstream fields in the database; determining the direction from high to low based on the bottom elevation of the node when the flow direction identifier is missing; and outputting the direction confidence field to characterize the reliability of the direction determination.

[0106] In step 6, the anomaly types include at least two of the following: reverse slope anomaly, node elevation change anomaly, dead-end node or isolated subgraph anomaly, discontinuity of upstream and downstream cross-sectional parameters anomaly, and end node discharge condition anomaly.

[0107] In step 7, the state information transmission iterative inference includes: constructing a message based on the engineering constraint violation and performing neighborhood aggregation to update the node bottom elevation state; wherein the update weight is determined by one of the node elevation confidence and the pipe segment direction confidence, so as to achieve the effect of prioritizing the adjustment of low confidence objects in the iteration; at the same time, one or more of the following are used as the stopping conditions for the iteration: the state change amount is less than the threshold, the maximum number of iterations is reached.

[0108] In step 8, the fusion solution aims to minimize the deviation between the final correction amount and the correction prior, and performs constraint projection or constraint optimization under the engineering constraint set, which includes at least two of the following: slope constraint, continuity constraint, and connectivity constraint.

[0109] Example 1, see Figure 1 An automatic construction and correction method for pipeline network models based on graph structure state information transmission includes the following steps:

[0110] Input data and preset configuration

[0111] The inputs in this implementation include at least one of the following two types of data sources:

[0112] 1) Vector engineering electronic drawing files: including pipeline elements, node elements, text annotations (pipe diameter / section dimensions, node elevation, burial depth, etc.) and leader lines or related line segments;

[0113] 2) Pipeline database: It should include at least a node table and a pipe segment table. The node table contains fields related to node coordinates and elevation, and the pipe segment table contains fields such as connection relationship, cross section and length.

[0114] To improve the adaptability of automatic parsing, this embodiment can optionally read the model's preset configuration file, which is used to define layer-object type mapping, field mapping relationship, annotation format rules, unit conversion rules, and engineering constraint parameters.

[0115] Step 1: Parsing and object recognition of vector engineering electronic drawings (or databases)

[0116] When the data source is a vector engineering drawing file, the drawing elements are parsed and identified to include at least:

[0117] Pipeline entity collection Line segments, polylines, etc.;

[0118] Node entity set Point and block references, etc.;

[0119] Text annotation set T;

[0120] A set of leader or related line segments L, or a broken line with an arrowhead;

[0121] Simultaneously, extract the geometric information (coordinates, bounding box, orientation angle, layer name, line type, etc.) of each object, and extract the text content, font size, rotation angle, and bounding box of the text annotations.

[0122] When the data source is a pipeline database, the node table, pipe segment table and their attribute fields are read from the database, and the pipe segment-node connection relationship is established to form a structured initial input.

[0123] Step 2: Multi-source attribute association and spatial index construction

[0124] A spatial index is built based on the object set from step S1 to support fast nearest neighbor retrieval and candidate set generation. In this embodiment:

[0125] Create an R-tree index for the center point of the bounding box of the text annotation. ;

[0126] Create an R-tree index for pipeline geometry objects ;

[0127] Build an R-tree index for node geometry objects .

[0128] To reduce noisy data, this implementation method preprocesses the parsing results, including:

[0129] Merge similar entities (merge multiple symbols at the same node, merge duplicate pipe segments);

[0130] Clean up duplicate segments and remove zero-length objects;

[0131] Map to a uniform object type by layer / symbol library.

[0132] Step 3, Obtaining and Standardizing Pipe Section Cross-Section Parameters

[0133] Generate a candidate annotation set Ce for cross-sectional parameters for each pipe segment. Candidate sources must include at least:

[0134] Strongly related candidates: Text annotations that point from the endpoint of the lead wire to the pipe segment or fall within the buffer zone of the pipe segment;

[0135] Weakly correlated candidates: Text annotations retrieved by index within a preset radius based on the midpoint of the pipe segment or equidistant sampling points;

[0136] Rule candidates: Text annotations that meet the cross-section format rules (such as DN, Φ, rectangular cross-section “width × height” patterns, etc.).

[0137] The candidate annotations are scored and ranked, with scoring factors including at least: distance cost, leader connectivity consistency, format matching, layer consistency, and text orientation consistency with pipe segment direction. The highest-scoring annotation is selected as the source of cross-sectional parameters and output.

[0138] Section parameter field (pipe diameter / section size);

[0139] Cross-section source marking (Drawing annotations / database fields / calculated values);

[0140] Assign confidence level ;

[0141] Unit conversion and standardization are performed on the cross-sectional parameters to meet the requirements of subsequent model construction.

[0142] Step 4: Obtaining and Consistently Organizing Node Elevation Parameters

[0143] Generate a set of candidate elevation annotations for each node. Candidate sources include strong correlations with lead lines and weak correlations with spatial proximity; ground elevation is obtained by parsing candidate texts. Node bottom elevation Burial depth D field;

[0144] When annotation conflicts occur, perform conflict resolution:

[0145] When the difference between candidate scores is greater than a threshold, the highest-scoring candidate is selected.

[0146] When the score difference is less than the threshold, it is marked as uncertain, alternative values ​​are retained, and the elevation confidence level is output. With source tag ;

[0147] Perform consistent completion on missing fields: for example, when With D known and When missing, according to Estimation; or when Zb and D are known, Zg can be estimated. The estimated field is marked as "estimated" in the source tag, and the confidence level is reduced accordingly.

[0148] Step 5, Pipeline topology construction and initial model generation

[0149] The pipeline topology is constructed based on the connection relationships between nodes and pipe segments, forming an initial pipeline model containing a set of nodes, a set of pipe segments, and their attribute parameters. When the drawings lack explicit connection fields, endpoint snapping or spatial snapping can be used to establish the association between pipe segment endpoints and nodes; when the database has start and end node fields, connections are established directly.

[0150] For each pipe segment, the start and end nodes are determined, and a graph structure G=(V, E) is generated. For the flow direction, this implementation uses priority determination:

[0151] 1) If the drawing contains flow arrows or the database contains upstream and downstream fields, determine the direction based on these arrows and set the direction confidence level. For high;

[0152] 2) If there is no clear flow direction, then tentatively determine the direction from high to low based on the node's bottom elevation, and set... For the middle;

[0153] 3) If the elevation is also uncertain, then leave it as uncertain in direction and set... The result is low and will be corrected in subsequent iterations.

[0154] Step 6: Initial model anomaly diagnosis and rule correction

[0155] Perform a consistency check on the initial model and identify anomaly types, including:

[0156] Abnormal slope or reverse slope of the pipeline section;

[0157] Abnormal abrupt changes in node elevation;

[0158] Broken nodes, isolated subgraphs, or redundant nodes;

[0159] The parameters of the upstream and downstream cross sections are discontinuous and abnormal.

[0160] Abnormal start / end boundary conditions;

[0161] The anomalies are corrected according to the preset engineering constraints, resulting in a rule-corrected pipeline model. An anomaly marker field is generated for weight control and priority adjustment in the subsequent state inference stage.

[0162] Step 7: Iterative inference and correction of prior output based on graph structure state information transmission

[0163] In this embodiment, information calculation and state update can be constructed according to the minimum slope constraint. The specific message function, aggregation function, state update function, and convergence criterion are shown in equations (1) to (7), respectively; where, the weights and The prior is determined by the confidence levels of direction and cross-section / elevation source, prioritizing adjustments based on lower confidence levels. A revised prior is obtained after the iteration terminates. .

[0164] Step 8: Correct the fusion solution of prior and engineering constraint sets.

[0165] The corrected prior obtained in step 7 is used as the prior input for the fusion solution. The fusion solution aims to minimize the deviation between the final corrected amount and the corrected prior, and performs constraint projection or constraint optimization under the engineering constraint set to obtain the final corrected amount. The engineering constraint set includes at least two of the following: slope constraint, continuity constraint, and connectivity constraint. Specifically, the slope constraint may include at least a minimum slope constraint, meaning that for any pipe segment, the difference between the corrected upstream node bottom elevation and the downstream node bottom elevation meets the minimum allowable slope requirement. The node bottom elevations and related pipe segment attributes are updated based on the final corrected amount, and the corrected pipe network model is output.

[0166] See Figure 2 An electronic vector diagram of the drainage network of a typical drainage zone (data storage format: .dxf) is used for this implementation example.

[0167] See Figure 3 R-trees establish a hierarchical index for spatial objects: the root node R1 covers the global scope, branch nodes R2 and R3 cover local sub-regions, and leaf nodes L1, L2, and L3 store object entries within their respective regions, enclosed by a minimum bounding rectangle. During retrieval, a query range is constructed for the target segment or node, and only nodes intersecting the query range are traversed from top to bottom. Finally, the set of candidate objects is returned from the matched leaf nodes, thus avoiding full graph traversal and improving the efficiency of nearest neighbor retrieval and annotation matching.

[0168] See Figure 4 The effect of constructing and correcting the pipeline network model using this method can correct erroneous topological relationships such as misconnections, abnormal reverse slopes, complex branch pipes, and large pipes connecting to small pipes.

[0169] This invention parses the original data source to identify pipeline entities, node entities, text annotations, and lead lines or associated line segments; constructs a spatial index to generate a candidate set of annotations, and scores and resolves conflicts based on distance cost, lead line connectivity consistency, and format matching, achieving automatic assignment of pipeline section parameters and node elevation parameters, while outputting source tags and confidence levels; on this basis, it constructs a pipeline network topology and performs consistency checks, and performs rule corrections based on preset engineering constraints; further, it abstracts the rule-corrected pipeline network into a graph structure, constructs node state vectors and edge attribute vectors, and generates node bottom elevation correction priors and anomaly confidence levels through iterative inference using training-free state information transmission, and performs constraint projection or constraint optimization under engineering constraint sets such as slope, continuity, and connectivity to obtain the final correction result; it outputs a pipeline network model that meets engineering constraints and a correction log, thereby improving the overall consistency and engineering rationality of the pipeline network model under complex topology conditions and reducing the workload of manual verification.

Claims

1. A method for automatic construction and correction of pipeline network models based on graph structure state information transmission, characterized in that, Includes the following steps: Step 1: Obtain the original data source of the municipal stormwater and sewage pipe network, parse the original data source to obtain the pipeline entity set, node entity set, text annotation set, and lead line or associated line segment set; when the original data source is a pipe network database, read the pipe segment table, node table and their attribute fields and establish the connection relationship between pipe segments and nodes; combine the parsed pipeline entity set, node entity set, text annotation set, and lead line or associated line segment set into a unified set of original pipe network objects; The original data source includes either vector engineering electronic drawing files or a pipeline database. Step 2: Construct a spatial index structure based on the original object set obtained in Step 1 to support nearest neighbor retrieval and candidate set generation, and preprocess the parsing results of the original data source; Step 3: Generate a set of candidate annotations for cross-sectional parameters for each pipe segment; score and sort the candidate annotations to determine the cross-sectional parameters; and output the source markers and confidence scores of the cross-sectional parameters; perform unit conversion and format standardization on the cross-sectional parameters. Step 4: Generate a set of candidate elevation annotations for each node, parse the format of the candidate elevation annotation set to obtain the ground elevation, node bottom elevation and burial depth fields, and assign values ​​according to the score; when there are competing annotation matches or conflicting elevation values, perform conflict resolution, output the elevation source mark and elevation assignment confidence; fill in missing fields according to the consistency rules. Step 5: Construct the pipeline network topology based on the connection relationship between nodes and pipe segments, and generate an initial pipeline network model containing a set of nodes, a set of pipe segments, and their attribute parameters; Step 6: Perform consistency verification on the initial pipeline model, identify anomaly types, and perform rule correction based on preset engineering constraints to obtain the rule-corrected initial pipeline model; the set of engineering constraints established during the rule correction process includes at least one or more of the following: minimum slope constraint, continuity constraint, and connectivity constraint. Step 7: Abstract the initial pipeline network model after rule correction into graph structure data G=(V,E), construct a node state vector containing node elevation state, confidence level, and source label, and also construct an edge attribute vector containing pipe segment attributes and direction confidence level; perform training-free state information transfer iterative inference on the graph structure data to obtain the node bottom elevation correction prior and pipe segment anomaly confidence level. Step 8: Combine the corrected prior from Step 7 with the set of engineering constraints from Step 6 to solve the problem. Obtain the final corrected result that satisfies the engineering constraints through constraint projection or constraint optimization, and output the corrected pipeline network model.

2. The method for automatic construction and correction of pipeline network model based on graph structure state information transmission according to claim 1, characterized in that, In step 2, the spatial index structure is an R-tree index, which includes at least an index for the geometric position of the outer frame of text annotations and an index for the geometric objects of pipelines and nodes.

3. The method for automatic construction and correction of pipeline network model based on graph structure state information transmission according to claim 1, characterized in that, In step 3, the candidate annotation set for cross-section parameters includes at least: strongly associated candidate annotations obtained by pointing to the leader or associated line segment, weakly associated candidate annotations obtained by spatial proximity retrieval, and regular candidate annotations that satisfy the cross-section format rules.

4. The method for automatic construction and correction of pipeline network model based on graph structure state information transmission according to claim 1, characterized in that, In step 3, the scoring is based on at least two of the following: distance cost, lead connectivity consistency, text format matching degree, and layer consistency. The scoring described in step 4 is the same as the scoring described in step 3.

5. The method for automatic construction and correction of pipeline network model based on graph structure state information transmission according to claim 1, characterized in that, In step 4, the conflict resolution includes: when the difference in candidate annotation scores is greater than a preset threshold, the highest-scoring candidate is adopted; when the difference in scores is less than a preset threshold, it is marked as uncertain and the alternative value is retained, while the confidence field is output.

6. The method for automatic construction and correction of pipeline network model based on graph structure state information transmission according to claim 1, characterized in that, In step 4, the consistency completion rules include: when the ground elevation and burial depth are known but the node bottom elevation is missing, completion is calculated based on the relationship that the node bottom elevation equals the ground elevation minus the burial depth; when the node bottom elevation and burial depth are known but the ground elevation is missing, completion is calculated based on the relationship that the ground elevation equals the node bottom elevation plus the burial depth; when both the ground elevation and the node bottom elevation are known but the burial depth is missing, completion is calculated based on the relationship that the burial depth equals the ground elevation minus the node bottom elevation; the completion result is written to the corresponding field, and the source marker of the field is set to completion, while the confidence level of the field is set to be lower than the confidence level obtained by directly parsing the text annotation.

7. The method for automatic construction and correction of pipeline network model based on graph structure state information transmission according to claim 1, characterized in that, In step 5, determining the direction of the pipe segment includes: determining the direction based on the flow direction identifier on the drawing or the upstream and downstream fields in the database; determining the direction based on the bottom elevation of the node from high to low when the flow direction identifier is missing; and outputting a direction confidence field to characterize the reliability of the direction determination.

8. The method for automatic construction and correction of pipeline network model based on graph structure state information transmission according to claim 1, characterized in that, In step 6, the anomaly types include at least two of the following: reverse slope anomaly, node elevation change anomaly, dead-end node or isolated subgraph anomaly, discontinuity of upstream and downstream cross-sectional parameters anomaly, and end node discharge condition anomaly.

9. The method for automatic construction and correction of pipeline network model based on graph structure state information transmission according to claim 1, characterized in that, In step 7, the state information transmission iterative inference includes: constructing messages based on engineering constraint violations and performing neighborhood aggregation, and sequentially updating the node bottom elevation state; wherein the update weight is jointly determined by the node elevation confidence and the pipe segment direction confidence, so as to achieve the effect of prioritizing the adjustment of low confidence objects in the iteration; at the same time, one of the state change amount being less than a threshold and reaching the maximum number of iterations is used as the stopping condition.

10. The method for automatic construction and correction of pipeline network model based on graph structure state information transmission according to claim 1, characterized in that, In step 8, the fusion solution aims to minimize the deviation between the final correction amount and the correction prior, and performs constraint projection or constraint optimization under the engineering constraint set, which includes at least two of the following: slope constraint, continuity constraint, and connectivity constraint.

Images