An underground pipe network partition visualization and safety warning method and system

Abstract

The present application relates to the technical field of underground pipe network monitoring and early warning, and particularly relates to an underground pipe network partition visualization and safety early warning method and system. The method comprises: obtaining a map service source, an underground pipeline service, GIS data, a BIM model, equipment data, historical monitoring data, historical maintenance records, inspection logs and accident reports, performing data alignment and data standardization processing to obtain pipe network topology structure information; performing key node, valve, connection relationship and monitoring coverage density correlation processing to obtain partition objects; performing connection processing to obtain multi-source heterogeneous monitoring data, and performing time sequence feature coding, spatial feature coding, semantic text data feature coding and cross-modal fusion processing to obtain fusion feature vectors; and then performing confidence interval, intelligent alarm instruction, mutation alarm instruction, maximum clustering distance and connection relationship joint processing to obtain correlation alarm state information. The present application can effectively improve the accuracy and timeliness of safety early warning.

Description

Technical Field

[0001] This invention relates to the field of underground pipeline network monitoring and early warning technology, and in particular to a method and system for visualizing and providing safety early warning of underground pipeline network zones. Background Technology

[0002] In the field of underground pipeline network monitoring and early warning technology, existing solutions typically establish visualization layers based on map service sources, underground pipeline services, geographic information system data, building information models, equipment data, historical monitoring data, historical maintenance records, inspection logs, and accident reports. These layers are combined with marked points, intelligent alarm commands, sudden change alarm commands, and the scope of impact to carry out monitoring and early warning. However, these solutions have limitations such as the disconnect between pipeline network topology information and subsequent processing, the separation of marked points and associated alarm status information, and the separation of connection relationships and maximum clustering distance.

[0003] Existing methods often rely on equipment data and marked locations to first generate abnormal records, and then combine historical monitoring data or preset pressure data thresholds for judgment, followed by drawing layers on the abnormal records.

[0004] In scenarios involving pipeline topology information, fused feature vectors, partition boundaries, and connectivity, existing solutions often struggle with the pre-formation of partition objects and the continuous organization of associated alarm status information, failing to meet the requirement for stable implementation around partition objects, connectivity, and associated alarm status information. Regarding the joint processing of pipeline topology information and fused feature vectors, existing technologies generally lack a unified link between historical monitoring data, historical maintenance records, inspection logs, and accident reports, and lack unified joint processing between confidence intervals, intelligent alarm commands, abrupt alarm commands, and maximum clustering distance. This makes it difficult to establish a consistent workflow for acquisition, data alignment, data standardization, correlation processing, linking processing, joint processing, and rendering in underground pipeline partition visualization and safety early warning applications, resulting in an unstable correspondence between associated alarm status information and safety early warning results. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a method for visualizing and providing safety early warning of underground pipe network zones, comprising:

[0006] S100. Acquire multi-source pipeline data, perform spatial alignment, time synchronization and object merging processing to obtain pipeline topology information including pipe segments, key nodes, valves and their connection relationships; wherein, the multi-source pipeline data includes map service source, underground pipeline service, GIS data, BIM model, equipment data and historical operation and maintenance data;

[0007] S200. Based on the pipeline topology information, trace continuous paths along the connection relationships, identify and verify the key nodes and valves to delineate candidate areas, and then correct the boundaries of the candidate areas according to the monitoring coverage density of the equipment data to generate partition objects.

[0008] S300. Based on the partition boundaries of the partition object and the key nodes and valves it contains, real-time and historical monitoring data, maintenance records, inspection logs and accident reports are linked to the corresponding partition object to form multi-source heterogeneous monitoring data.

[0009] S400. For the multi-source heterogeneous monitoring data, temporal encoding is performed using a temporal convolutional network, spatial encoding is performed using a graph convolutional network, and semantic text encoding is performed using an attention mechanism network. The encoding results are then fused across modalities to obtain a fused feature vector.

[0010] S500. Based on the fused feature vector and its real-time monitoring data, generate a confidence interval and determine whether an intelligent alarm or a sudden alarm is triggered; perform clustering distance calculation on the adjacent marked points that trigger the alarm, and generate associated alarm status information in combination with the connection relationship;

[0011] S600. Based on the associated alarm status information, draw the partition boundary, the marked points that triggered the alarm, and the affected valves, pipe sections, users, and pressure change points in the visualization layer to generate a safety warning result.

[0012] Furthermore, the process of acquiring multi-source pipeline network data, performing spatial alignment, time synchronization, and object merging to obtain pipeline network topology information including pipe segments, key nodes, valves, and their connection relationships includes:

[0013] The service maps pipe segments in underground pipeline services to spatial locations in GIS data, maps building information model components in BIM models to road and building locations in map service sources, and merges records of key nodes, valves, and manhole covers falling within the same location range into the same object.

[0014] Synchronize timestamps according to monitoring time, and merge pressure data, flow data, temperature data, hazardous gas data and location data within the same monitoring time window into the same object;

[0015] Arrange the records in chronological order according to maintenance time, inspection time, and accident time, and then match them to objects according to pipe section, key node, and valve, so that maintenance records, inspection records, and accident records are linked to spatial objects one by one.

[0016] The pipeline network topology information also includes monitoring coverage density, reference locations of marked points, correspondence of historical maintenance records, correspondence of inspection logs, and correspondence of accident reports.

[0017] Furthermore, the process of tracing continuous paths along the aforementioned connections, identifying and verifying the key nodes and valves to delineate candidate regions includes:

[0018] The process of identifying and verifying the key nodes and valves to delineate candidate regions includes a first round of association:

[0019] Tracing adjacent pipe segments along the connection relationship, identifying key nodes on the same continuous path, reading the front and rear positions of the valves in the continuous path, and grouping pipe segments and key nodes that are between adjacent valves and have a continuous connection relationship into the same candidate area;

[0020] When there are key nodes and two valves in the candidate area, the historical maintenance record correspondence, the inspection log correspondence, and the accident report correspondence are called for secondary verification. The secondary verification prioritizes checking the historical maintenance record, then the inspection log, and then the accident report. Candidate areas that are consistent with the maintenance, inspection, and accident directions are retained, and conflicting content is written into the boundary verification mark.

[0021] Furthermore, the process of correcting the boundaries of the candidate regions based on the monitoring coverage density of the device data to generate partition objects includes:

[0022] The step of correcting the boundary of the candidate region based on the monitoring coverage density includes a second round of association:

[0023] The number of equipment data sources, the frequency of historical monitoring data records, and the distribution status of the reference positions of the marked points are read segment by segment along the candidate region. For areas where the equipment data source is interrupted, the historical monitoring data records are sparse, or the reference positions of the marked points are obviously separated, new boundary candidates are generated at the corresponding key nodes or valves.

[0024] The partition object includes partition number, partition boundary, key node set, valve set, connection relationship segment, monitoring coverage density classification result, and boundary verification mark.

[0025] Furthermore, the process of linking real-time and historical monitoring data, maintenance records, inspection logs, and incident reports to the corresponding partition objects to form multi-source heterogeneous monitoring data includes:

[0026] The attachment process includes object filtering, hierarchical attachment, association verification, and supplementary recording of marked points;

[0027] Object filtering includes: reading location data from real-time monitoring data and checking whether it falls within the partition boundary; reading historical locations and historical objects from historical monitoring data and checking whether they are consistent with the set of key nodes, the set of valves, and the connection relationship fragments; and reading maintenance pipe sections, maintenance valves, inspection objects, accident objects, and accident locations from historical maintenance records, inspection logs, and accident reports and checking whether they match the key nodes, valves, and connection relationships within the partition object.

[0028] The hierarchical connection includes monitoring layer connection, maintenance layer connection, inspection layer connection and accident layer connection. Among them, the monitoring layer connection merges real-time monitoring data and historical monitoring data according to the zone number and arranges them in order of monitoring time. The maintenance layer connection merges historical maintenance records according to maintenance time and maintenance object. The inspection layer connection merges inspection logs according to inspection location and inspection object. The accident layer connection merges accident reports according to accident location and accident object.

[0029] The correlation verification includes: checking the temporal continuity between real-time monitoring data and historical monitoring data, checking the consistency of historical maintenance records, inspection logs and accident reports with the objects of key node sets, valve sets and connection relationship fragments, and writing monitoring supplementary records, maintenance and inspection conflict records and cross-regional accident records into the connection anomaly record;

[0030] The annotation point supplementation includes: generating annotation points around the key node set, valve set and accident location along the connection relationship segment, and attaching real-time monitoring data, historical monitoring data, historical maintenance records, inspection logs and accident reports to the corresponding annotation points respectively;

[0031] The multi-source heterogeneous monitoring data includes partition numbers, marked points, real-time monitoring data sequences, historical monitoring data sequences, historical maintenance record sequences, inspection log sequences, accident report sequences, and connection anomaly records.

[0032] Furthermore, the process of using a temporal convolutional network for temporal encoding, a graph convolutional network for spatial encoding, and an attention mechanism network for semantic text encoding for the multi-source heterogeneous monitoring data, and then fusing the encoding results across modalities to obtain a fused feature vector, includes:

[0033] The temporal coding includes: receiving real-time monitoring data sequences and historical monitoring data sequences under the same partition number from the input segment of the temporal convolutional network; performing continuous convolution on pressure data, flow data, temperature data and harmful gas data in chronological order from the convolutional segment; and outputting a temporal feature vector.

[0034] The spatial encoding includes: receiving the labeled points, key node sets, and valve sets under the partition number by the node input segment of the graph convolutional network; connecting the object correspondence, spatial order relationship, and connection relationship fragments received by the input segment; and reading the nodes and connections in the order of connection inside the partition boundary by the aggregation segment and performing aggregation processing on adjacent labeled points, adjacent key nodes, and adjacent valves, and outputting a spatial topology feature vector.

[0035] The semantic text encoding includes: receiving a sequence of historical maintenance records, a sequence of inspection logs, a sequence of accident reports, and a record of attached abnormalities by a neural network model based on an attention mechanism; reading the above text objects according to the partition number; assigning the text objects to the corresponding annotation points according to the object correspondence; performing attention allocation according to the time order; and outputting a semantic text feature vector.

[0036] The cross-modal fusion includes concatenation operation, multi-head self-attention network, and refinement and dimensionality reduction. The concatenation operation reads the temporal feature vector, spatial feature vector, and semantic text feature vector corresponding to the same partition number and concatenates them in the order of the labeled points. The multi-head self-attention network performs multiple rounds of attention allocation on the concatenation result. After refining and dimensionality reduction, duplicate content is compressed and the fused feature vector is output.

[0037] Furthermore, the process of generating confidence intervals and determining whether to trigger intelligent alarms or sudden change alarms includes:

[0038] The process of generating confidence intervals includes: reading historical monitoring data sequences by partition number, dividing historical windows according to the relationship between the marked points and time sequence, performing interval merging on the historical windows of the same marked point to form the confidence interval of that marked point, and then performing partition merging on multiple marked points within the same partition number to form the confidence interval of that partition number.

[0039] The determination of whether to trigger an intelligent alarm includes: reading the fused feature vector and the real-time monitoring data sequence of the current operating cycle according to the partition number, reading the confidence interval field corresponding to each marked point one by one, and if the real-time monitoring data sequence exceeds the confidence interval field and the fused feature vector corresponding to the marked point is consistent with the object correspondence of the historical monitoring data sequence, then writing an intelligent alarm record at the marked point.

[0040] Furthermore, the process of calculating the clustering distance between adjacent marked points that trigger the alarm, and generating associated alarm status information based on the connection relationship, includes:

[0041] Expand along the connection relationship segment segment by segment, calculate the line distance between adjacent marked points, perform cluster distance filtering on multiple line distances within the current partition number to obtain the maximum cluster distance field. If two marked points are spatially close but their connection relationship segments are not continuous, they are not written into the same maximum cluster distance field. Simultaneously read in the intelligent alarm record, the sudden alarm record, the maximum cluster distance field, and the connection relationship segment, and classify the marked points that are continuously distributed on the same connection relationship segment and whose distance does not exceed the maximum cluster distance field into the same associated alarm record.

[0042] The associated alarm status information includes partition number, associated alarm command, associated alarm record, scope of influence, correspondence of marked points, correspondence of connection relationship segments, correspondence of affected valves, correspondence of affected pipe segments, correspondence of affected users, and correspondence of pressure change points. The scope of influence is generated by reading the key node set, valve set, and marked points forward and backward along the connection relationship segment corresponding to the associated alarm record.

[0043] Furthermore, the process of drawing the partition boundaries, alarm trigger points, and affected valves, pipe sections, users, and pressure change points in the visualization layer to generate safety warning results includes:

[0044] Read the partition boundary corresponding to the partition number, read the correspondence between the influence range and the connection relationship fragment, map the influence range fragment falling inside the current partition boundary to the visualization layer, and if the influence range crosses the current partition boundary, then cut off the range falling inside this partition number according to the current partition boundary.

[0045] Read all the marked points within the current partition number according to the correspondence of the marked points, adjust the display order of the marked points according to the time order and spatial order in the associated alarm records, write the associated alarm records first and then write the corresponding marked points for marked points that are associated with both smart alarms and sudden alarms, and write pressure change markers at the original marked point positions for marked points that are only associated with sudden alarms.

[0046] Read the correspondence of affected valves, the correspondence of affected pipe segments, and the correspondence of affected users. The drawing position of the affected valves is determined by the original position of the valves and the current partition boundary. The drawing range of the affected pipe segments is determined by the correspondence of the connection fragments and the affected range. The drawing position of the affected users is determined by the correspondence of the affected users and the current partition number.

[0047] Read the corresponding relationship of the pressure change points and compare it point by point with the drawn marked points. If the pressure change point is not written in the marked point field, then write it into the current partition boundary according to the connection relationship fragment and spatial order relationship.

[0048] The safety warning results include the zone number, zone boundary, marked location, affected valves, affected pipe sections, affected users, pressure change locations, and associated alarm commands.

[0049] Furthermore, an underground pipeline network zoning visualization and safety early warning system, applied to any of the methods described above, includes:

[0050] The first processing module is used to acquire map service sources, underground pipeline services, GIS data, BIM models, equipment data, historical monitoring data, historical maintenance records, inspection logs and accident reports, and perform data alignment and data standardization processing to obtain pipeline network topology information;

[0051] The second processing module is used to perform correlation processing of key nodes, valves, connection relationships and monitoring coverage density based on the pipeline network topology information to obtain partition objects;

[0052] The third processing module is used to perform real-time monitoring data, historical monitoring data, historical maintenance records, inspection logs and accident reports on the partition object to obtain multi-source heterogeneous monitoring data.

[0053] The heterogeneous feature encoding and fusion module is used to perform temporal feature encoding, spatial feature encoding, semantic text data feature encoding and cross-modal fusion processing based on the multi-source heterogeneous monitoring data to obtain a fused feature vector;

[0054] The fourth processing module is used to perform joint processing of confidence interval, intelligent alarm command, mutation alarm command, maximum clustering distance and connection relationship based on the fused feature vector to obtain associated alarm status information;

[0055] The drawing module is used to draw the partition boundaries, marked points, affected valves, affected pipe sections, affected users, and pressure change points based on the associated alarm status information, and to obtain the safety warning results.

[0056] The key innovations of this invention include:

[0057] (1) Based on the pipeline topology information, association processing is performed around key nodes, valves, connection relationships and monitoring coverage density, and partition objects are generated by combining historical maintenance records, inspection logs and accident reports, so that the partition objects are organized before data hooking, feature encoding, joint processing and drawing processing.

[0058] (2) Based on the partition object, real-time monitoring data, historical monitoring data, historical maintenance records, inspection logs and accident reports are linked and processed to form multi-source heterogeneous monitoring data within the same partition object. Then, the multi-source heterogeneous monitoring data is processed by time-series feature encoding, spatial feature encoding, semantic text data feature encoding and cross-modal fusion to obtain a fused feature vector.

[0059] (3) Based on the fusion feature vector, the confidence interval, intelligent alarm command, sudden alarm command, maximum clustering distance and connection relationship are incorporated into the unified joint processing link to generate associated alarm status information including the influence range, the correspondence of the marked point, the correspondence of the affected valve, the correspondence of the affected pipe section, the correspondence of the affected user and the correspondence of the pressure change point, and drive the partition boundary and marked point drawing processing.

[0060] The following are its main beneficial effects:

[0061] (1) In view of the problem that the pipeline topology information is disconnected from the subsequent processing in the existing scheme, the present invention generates partition objects by associating key nodes, valves, connection relationships and monitoring coverage density, so that subsequent real-time monitoring data, historical monitoring data, historical maintenance records, inspection logs and accident reports enter a unified processing link around the partition objects, and a stable correspondence is formed between the partition boundaries, marked points and associated alarm status information.

[0062] (2) In view of the problem that historical maintenance records, inspection logs and accident reports are separated from the monitoring link in the existing scheme, the present invention puts the historical maintenance records, the inspection logs and the accident reports into the partition object generation and attachment process, so that the multi-source heterogeneous monitoring data retains both the monitoring data sequence and the text record sequence, and the fused feature vector and the partition object maintain a continuous correspondence.

[0063] (3) In view of the problem that the intelligent alarm command, the mutation alarm command and the maximum clustering distance are processed separately in the existing scheme, the present invention puts the confidence interval, intelligent alarm command, mutation alarm command, maximum clustering distance and connection relationship into the same joint processing process, so that the associated alarm status information no longer stays at a single marked point, but forms associated alarm records and influence ranges around the partition object.

[0064] (4) In view of the problem of unstable connection between the associated alarm status information and the drawing process in the existing scheme, the present invention directly organizes the corresponding relationship of the affected valve, the corresponding relationship of the affected pipe section, the corresponding relationship of the affected user and the corresponding relationship of the pressure change point in the associated alarm status information, so that the partition boundary, the marked point, the affected valve, the affected pipe section, the affected user and the pressure change point are written in the same drawing process, and the object link of the safety warning result remains consistent.

[0065] (5) In view of the problem that the safety warning results are separated from the subsequent update link in the existing scheme, the present invention places the safety warning results, disposal results, knowledge base, historical monitoring data, historical maintenance records, inspection logs, accident reports and zoning boundaries in the same update link, so that the objects output by the preceding association processing, hook processing, joint processing and drawing processing continue to enter the subsequent update process, and the method link of underground pipe network zoning visualization and safety warning remains continuous. Attached Figure Description

[0066] Figure 1 A flowchart illustrating a method for visualizing and providing safety early warning of underground pipe network zones, provided as an embodiment of this application;

[0067] Figure 2 This is a structural block diagram of an underground pipeline network zoning visualization and safety early warning system provided in an embodiment of this application. Detailed Implementation

[0068] Example 1: Refer to Figure 1 This is a flowchart illustrating a method for visualizing and providing safety early warning of underground pipe network zones according to an embodiment of the present invention. The process may include at least steps S100-S600:

[0069] S100. Acquire multi-source pipeline data, perform spatial alignment, time synchronization and object merging processing to obtain pipeline topology information including pipe segments, key nodes, valves and their connection relationships; wherein, the multi-source pipeline data includes map service source, underground pipeline service, GIS data, BIM model, equipment data and historical operation and maintenance data;

[0070] S200. Based on the pipeline topology information, trace the continuous path along the connection relationship, identify and verify the key nodes and valves to delineate candidate areas, and then correct the boundary of the candidate areas according to the monitoring coverage density of the equipment data to generate partition objects.

[0071] S300. Based on the partition boundaries of the partition object and the key nodes and valves it contains, real-time and historical monitoring data, maintenance records, inspection logs and accident reports are linked to the corresponding partition object to form multi-source heterogeneous monitoring data.

[0072] S400. For the multi-source heterogeneous monitoring data, temporal encoding is performed using a temporal convolutional network, spatial encoding is performed using a graph convolutional network, and semantic text encoding is performed using an attention mechanism network. The encoding results are then fused across modalities to obtain a fused feature vector.

[0073] S500. Based on the fused feature vector and its real-time monitoring data, generate a confidence interval and determine whether an intelligent alarm or a sudden alarm is triggered; perform clustering distance calculation on the adjacent marked points that trigger the alarm, and generate associated alarm status information in combination with the connection relationship;

[0074] S600. Based on the associated alarm status information, draw the partition boundary, the marked points that triggered the alarm, and the affected valves, pipe sections, users, and pressure change points in the visualization layer to generate a safety warning result.

[0075] S100. Acquire multi-source pipeline data, perform spatial alignment, time synchronization and object merging processing to obtain pipeline topology information including pipe segments, key nodes, valves and their connection relationships; wherein, the multi-source pipeline data includes map service source, underground pipeline service, GIS data, BIM model, equipment data and historical operation and maintenance data;

[0076] Specifically, after the underground pipeline network zoning visualization and safety early warning system is activated, the first processing module receives map service sources, underground pipeline services, GIS data, BIM models, equipment data, historical monitoring data, historical maintenance records, inspection logs, and accident reports. The map service sources record the locations of roads, manhole covers, buildings, and the ground surface. The underground pipeline services record the service content of pipe sections, key nodes, valves, and marked points. The GIS data refers to Geographic Information System, which records the spatial location of the underground pipeline network, road layers, and GIS layers. The BIM model refers to Building Information Modeling, which records pipe sections, key nodes, valves, manhole covers, and building information model components. The equipment data records pressure data, flow data, temperature data, hazardous gas data, and location data. The historical monitoring data records monitoring data generated within the previous operating cycle of the same underground pipeline network. The historical maintenance records maintenance time, maintenance pipe section, maintenance valve, and maintenance results. The inspection log records the inspection time, inspection location, inspection object, and inspection description uploaded by the inspection terminal. The accident report records the accident time, accident location, accident object, and handling result. The above data is triggered by the first processing module according to a preset collection cycle. When new equipment data, historical maintenance records, inspection logs, or accident reports are received, a re-retrieval process is triggered simultaneously. When new map service sources, underground pipeline services, GIS data, or BIM models are received, a full reconstruction process is triggered.

[0077] Specifically, the data alignment process first addresses spatial correspondences, then temporal correspondences, and finally object correspondences. For map service sources, underground pipeline services, GIS data, and BIM models, the first processing module reads the locations of pipe segments, key nodes, valves, and manhole covers within the same underground pipeline network, performs location overlap checks and road layer checks, and matches the pipe segments in the underground pipeline service with their spatial locations in the GIS data. It also matches the building information model components in the BIM model with the road and building locations in the map service source, merging records where key nodes, valves, and manhole covers fall within the same location range into the same object. For equipment data and historical monitoring data, the first processing module timestamps the data according to the monitoring time, merging pressure data, flow data, temperature data, hazardous gas data, and location data within the same monitoring time window into the same object. For historical maintenance records, inspection logs, and accident reports, the first processing module arranges them chronologically by maintenance time, inspection time, and accident time, and then maps them to objects by pipe section, key node, and valve, creating a one-to-one connection between maintenance records, inspection records, and accident records and the aforementioned spatial objects. In the event of location conflicts, time conflicts, or object conflicts, the first processing module retains the original records and simultaneously invokes data verification and validation processes to compare the conflicting records again, retaining the comparison results in the current processing chain.

[0078] Specifically, the data standardization process includes timestamp synchronization, missing value handling, outlier filtering, data normalization, data verification, and data validation. Timestamp synchronization unifies the time representation of equipment data, historical monitoring data, historical maintenance records, inspection logs, and accident reports. Missing value handling completes missing monitoring records in equipment data and historical monitoring data, preserving the correspondence between original and completed records. Outlier filtering removes pressure, flow, temperature, and hazardous gas data that exceed the current monitoring range. Data normalization unifies the recording scale of different equipment data, ensuring that equipment data and historical monitoring data within the same underground pipeline network fall under the same recording scope. Data verification checks the consistency of pipe segments, key nodes, valves, and manhole covers between map service sources, underground pipeline services, GIS data, and BIM models. Data validation checks the consistency of the object correspondence between historical maintenance records, inspection logs, and accident reports and the equipment data and historical monitoring data. For records that pass both verification and validation, the first processing module writes the current batch processing result; for records that fail verification or validation, the first processing module retains the original record and continues to participate in subsequent manual inspection and verification, and does not delete the record in the current batch.

[0079] Specifically, in a real-world scenario, when the system accesses map service sources, underground pipeline services, GIS data, and BIM models of the water supply network, the first processing module first reads the locations of pipe segments, key nodes, valves, and manhole covers beneath the road. Then, it reads pressure, flow, and location data for the same location. Historical monitoring data, maintenance records, inspection logs, and accident reports from the past thirty days are then sequentially processed into the same batch. Subsequently, the first processing module performs location mapping on pipe segment records on both sides of the same valve, timestamps and synchronizes equipment data around the same key node, and performs object mapping on inspection logs and accident reports corresponding to the same accident location. After completing the above processing, the roads, pipe segments, key nodes, valves, manhole covers, monitoring data, maintenance records, inspection records, and accident records of the water supply network are organized into a single object link, which can be directly accessed for subsequent key node, valve, connection relationship, and monitoring coverage density association processing.

[0080] Specifically, after the data alignment and standardization processes, the first processing module outputs pipeline network topology information. This information includes pipe segments, key nodes, valves, connection relationships, monitoring coverage density, reference locations of marked points, historical maintenance record correspondences, inspection log correspondences, and accident report correspondences. Key nodes, valves, connection relationships, and monitoring coverage density are entered as output field names in step S200, which involves "based on the pipeline network topology information, performing association processing on key nodes, valves, connection relationships, and monitoring coverage density to obtain partition objects." The reference locations of marked points and various correspondences are also retained in the pipeline network topology information for use in the attachment processing of S300 and the drawing processing of S600.

[0081] This step's technical effects can be summarized as follows: This step integrates map service sources, underground pipeline services, GIS data, BIM models, equipment data, historical monitoring data, historical maintenance records, inspection logs, and accident reports into a single processing chain, forming a unified input for key nodes, valves, and connection relationships. The pipeline topology information output by this step directly inherits from S200, with clear step connections. This step pre-writes the reference locations and corresponding relationships of marked points into the pipeline topology information, ensuring that subsequent partition object processing, connection processing, and drawing processing have the same data foundation.

[0082] S200. Based on the pipeline topology information, trace the continuous path along the connection relationship, identify and verify the key nodes and valves to delineate candidate areas, and then correct the boundary of the candidate areas according to the monitoring coverage density of the equipment data to generate partition objects.

[0083] Specifically, this step is executed by the second processing module, with the pipeline topology information output by S100 as the input source. This pipeline topology information already includes pipe segments, key nodes, valves, connection relationships, monitoring coverage density, reference locations of marked points, historical maintenance record correspondences, inspection log correspondences, and accident report correspondences. The second processing module initiates association processing when the pipeline topology information is entered into the database, and synchronously initiates recalculation processing when updates occur in the map service source, underground pipeline service, GIS data, BIM model, equipment data, historical monitoring data, historical maintenance records, inspection logs, and accident reports. In the GIS data, GIS stands for Geographic Information System. In the BIM model, BIM stands for Building Information Modeling. Key nodes refer to locations where the connection status of a pipe segment changes. Valves refer to control objects located on the pipe segment. Connection relationships refer to the preceding and following correspondences between pipe segments and key nodes, between key nodes and valves, and between valves and pipe segments. The monitoring coverage density refers to the distribution of the equipment data, the historical monitoring data, and the reference locations of the marked points around the pipe section and the key nodes.

[0084] Specifically, the second processing module first extracts key nodes, valves, and connection relationships from the pipeline network topology information, performing a first round of association. This first round of association unfolds according to the pipe segment direction and the location of key nodes. The second processing module traces adjacent pipe segments along the connection relationships, identifies key nodes on the same continuous path, and then reads the preceding and following positions of the valves on that continuous path. Pipe segments and key nodes located between adjacent valves and with continuous connection relationships are grouped into the same candidate region. If a key node corresponds to two valves simultaneously within a candidate region, the second processing module calls the historical maintenance record correspondence, the inspection log correspondence, and the accident report correspondence for secondary verification. During secondary verification, the maintenance pipe segments and valves in the historical maintenance records are checked first, followed by the inspection locations and objects in the inspection logs, and then the accident locations and objects in the accident reports. After secondary verification, candidate regions consistent with the maintenance, inspection, and accident directions are retained, and conflicting content is written to boundary verification markers.

[0085] Furthermore, after completing the first round of association, the second processing module performs a second round of association, which revolves around the monitoring coverage density. The second processing module reads the number of equipment data sources, the frequency of historical monitoring data records, and the distribution of reference locations of marked points segment by segment along the candidate region. For regions with continuous equipment data sources, continuous historical monitoring data records, and continuous reference locations of marked points, the second processing module maintains the current candidate region boundary. For regions with interrupted equipment data sources, sparse historical monitoring data records, or significantly separated reference locations of marked points, the second processing module generates new boundary candidates at the corresponding key nodes or valves. The input to this processing link remains the pipeline topology information, and the processing actions still focus on the four aspects of key nodes, valves, connection relationships, and monitoring coverage density. The processing result manifests as the shrinkage, separation, or merging of candidate regions. When two boundary candidates simultaneously meet the association conditions, the second processing module again calls the historical maintenance records, the inspection logs, and the accident reports for comparison, and writes the boundary candidates retained after comparison into the partition boundary.

[0086] Specifically, in one implementation, the second processing module divides the system by valves and then refines it by key nodes and connection relationships. This method is suitable for water supply networks where valves are clearly distributed and connection relationships are stable. The second processing module first reads the valve positions on a continuous pipe section under a road, defines the continuous pipe section between two valves as the initial area, and then checks the number of key nodes and changes in connection relationships within the initial area. If the key nodes are concentrated and the connection relationships are continuous, the initial area is retained; if the key nodes are dense and the connection relationships are branched, the area boundaries are redefined at the branching points. Subsequently, the second processing module reads the number of equipment data sources, the frequency of historical monitoring data records, and the distribution status of reference locations of marked points within the initial area, corrects the partition boundaries, and forms partition objects. In this implementation, valves act as outer boundaries, key nodes act as internal partitions, and monitoring coverage density acts as boundary correction.

[0087] Specifically, in another implementation, the second processing module first generates continuous paths based on key nodes and connection relationships, and then reads valves and monitoring coverage density for convergence processing. This method is suitable for underground pipe networks with dense key nodes and large valve spacing. Starting from a key node, the second processing module traces continuous pipe segments along connection relationships until it encounters a valve or a point where the connection relationship is interrupted, forming a set of continuous paths. Subsequently, the second processing module reads the monitoring coverage density of each continuous path. If the equipment data source and historical monitoring data are continuously distributed in the continuous path, they are retained as the same partition object; if a sudden drop in monitoring coverage density occurs near a key node, the continuous path is truncated at that key node, and a new partition boundary is generated. In this implementation, key nodes serve as the initial segmentation point, connection relationships serve as the path tracing point, valves serve as the boundary closure point, and monitoring coverage density serves as the truncation determination point.

[0088] Specifically, in one engineering embodiment, a water supply network is laid beneath a city road. The network topology information output by S100 records six consecutive pipe segments, four key nodes, two valves, corresponding equipment data sources, historical monitoring data, historical maintenance records, inspection logs, and accident reports under the road. The second processing module first reads the six pipe segments between the two valves to form an initial candidate area. Then, it checks the connection relationships corresponding to the four key nodes and finds that one key node corresponds to two branch pipe segments. The second processing module continues to read the historical maintenance records and accident reports near the key node and finds that there are long-term single maintenance records and single accident objects at this location, so a new boundary candidate is generated at this key node. Subsequently, the second processing module reads the number of equipment data sources and the distribution status of the reference positions of the marked points within the road segment and finds that there are fewer equipment data sources and larger intervals in the historical monitoring data records on the branch pipe segments, so the branch pipe segments are separated from the original candidate area to form independent partition objects. After this process is completed, the main road segment forms one partition object and the branch segment forms another partition object. Both share the same map service source and underground pipeline service, but the partition boundaries are different, and the subsequent connection processing paths are also different.

[0089] Furthermore, after the partition object is generated, the second processing module writes the partition boundary, partition number, key node set, valve set, connection relationship fragment, monitoring coverage density grading result, and boundary verification mark into the partition object. The partition number is used to distinguish different partition objects; the partition boundary is used for the drawing process in S600; the key node set, valve set, connection relationship fragment, and monitoring coverage density grading result are used for the attachment process in S300; and the boundary verification mark is used for knowledge base writing and subsequent update processing. Understandably, the output field names in this step are partition number, partition boundary, key node set, valve set, connection relationship fragment, monitoring coverage density grading result, and boundary verification mark. These output field names together constitute the partition object and serve as the direct input in S300 for "attaching real-time monitoring data, historical monitoring data, historical maintenance records, inspection logs, and accident reports based on the partition object."

[0090] In summary, this step organizes key nodes, valves, connections, and monitoring coverage density in the pipeline topology information into stable partitioned objects, eliminating the need for subsequent processing along individual marked points. The partitioned objects generated in this step simultaneously carry partition boundaries and boundary verification markers, ensuring that subsequent connection, joint processing, and drawing processes operate around the same object link. This step also incorporates historical maintenance records, inspection logs, and accident reports into the boundary verification process in advance, making the correspondence between partitioned objects and subsequent safety warning results clearer.

[0091] S300. Based on the partition boundaries of the partition object and the key nodes and valves it contains, real-time and historical monitoring data, maintenance records, inspection logs and accident reports are linked to the corresponding partition object to form multi-source heterogeneous monitoring data.

[0092] Specifically, this step is executed by the third processing module, with the input source being the partition object output by S200. The partition object already contains the partition number, partition boundary, key node set, valve set, connection relationship fragments, monitoring coverage density grading results, and boundary verification markers. After the partition object is generated, the third processing module initiates the attachment process, synchronously triggering incremental attachment processing when equipment data arrives, historical monitoring data is updated, historical maintenance records are written, inspection logs are uploaded, and accident reports are archived. Reattachment processing is triggered when the partition boundary is updated. The real-time monitoring data consists of pressure data, flow data, temperature data, hazardous gas data, and location data generated from equipment data within the current operating cycle. The historical monitoring data is the monitoring data corresponding to the partition object within previous operating cycles. The historical maintenance records include maintenance time, maintenance pipe section, maintenance valve, and maintenance result. The inspection log records the inspection time, inspection location, inspection object, and inspection description uploaded by the inspection terminal. The accident report records the accident time, accident location, accident object, and handling result. The connection process involves incorporating the aforementioned data into the partition object and establishing a correspondence between the partition object and the monitoring data, maintenance records, inspection records, and accident records.

[0093] Specifically, the third processing module first performs object filtering. Object filtering revolves around partition number, partition boundary, key node set, valve set, and connection relationship fragment. The third processing module reads location data from real-time monitoring data and checks whether it falls within the partition boundary; it reads historical locations and historical objects from historical monitoring data and checks whether they are consistent with the key node set, valve set, and connection relationship fragment; it reads maintenance pipe sections, maintenance valves, inspection objects, accident objects, and accident locations from historical maintenance records, inspection logs, and accident reports and checks whether they match the key nodes, valves, and connection relationships within the partition object. Records that meet both boundary and object conditions are added to the current partition object; records that do not meet either boundary or object conditions are transferred to adjacent partition objects for review; records that are still inconsistent after review are written as attachment exception records, and the original records are retained for the next boundary update call. Through this processing chain, the third processing module first completes the determination of "who belongs to this partition object" before proceeding to subsequent classification processing.

[0094] Furthermore, the third processing module performs layered attachment. Layered attachment is divided into monitoring, maintenance, inspection, and incident layers. When attaching to the monitoring layer, the third processing module merges real-time and historical monitoring data according to the partition number, arranges them in chronological order, and then writes them to the corresponding key node and valve sets along the connection relationship segments. When attaching to the maintenance layer, the third processing module merges historical maintenance records according to maintenance time and maintenance object, mapping the same maintenance pipe section and the same maintenance valve to the current partition object. When attaching to the inspection layer, the third processing module merges inspection logs according to inspection location and inspection object, attaching the inspection description to the key node or valve set of the current partition object. When attaching to the incident layer, the third processing module merges incident reports according to incident location and incident object, attaching the handling results to the corresponding connection relationship segment of the current partition object. After completing the monitoring, maintenance, inspection, and incident layers, the third processing module performs unified numbering on the four layers of records within the same partition object and establishes chronological, object-oriented, and spatial order relationships. After this processing, the partition object is no longer just a boundary object, but an operational object with monitoring chain, maintenance chain, inspection chain and accident chain.

[0095] Specifically, the third processing module performs association verification after layered attachment. The association verification first checks the temporal continuity between real-time monitoring data and historical monitoring data, and then checks the consistency of historical maintenance records, inspection logs, and accident reports with the objects of the key node set, valve set, and connection relationship fragment. If real-time monitoring data appears for the same key node in the current operating cycle, but there is no corresponding record in adjacent historical monitoring data, the third processing module marks this record as a supplementary monitoring record and retains its location data and monitoring time. If the same valve has been written with maintenance results in the historical maintenance record, but there is still an abnormal inspection description of the same object in the inspection log, the third processing module marks this group of records as maintenance-inspection conflict records and writes them into the boundary verification mark association area. If the accident location in the accident report spans two partition objects, the third processing module calls the connection relationship fragment and the partition boundary to re-examine the accident location, attaches the accident report to both partition objects simultaneously, and then marks it as a cross-regional accident record. Attached abnormal records, supplementary monitoring records, maintenance-inspection conflict records, and cross-regional accident records all participate in the generation of this multi-source heterogeneous monitoring data and are not removed in this step.

[0096] Further, the third processing module performs annotation point supplementation. The annotation points are derived from the reference positions of the annotation points in S100 and the correspondence between the currently attached objects. The third processing module generates annotation points around the key node set, valve set, and accident location along the connection relationship segment, and attaches real-time monitoring data, historical monitoring data, historical maintenance records, inspection logs, and accident reports to the corresponding annotation points. If an annotation point is associated with both the monitoring layer and the accident layer, then the annotation point records the monitoring time, accident time, and accident object; if an annotation point is associated with both the maintenance layer and the inspection layer, then the annotation point records the maintenance time, maintenance result, inspection time, and inspection description. After the annotation point supplementation is completed, the third processing module writes the annotation point and the partition number together into the attachment result, so that a corresponding structure of "partition boundary - annotation point - recorded object" is formed within the same partition object. This corresponding structure is directly called in the drawing process of S600.

[0097] In one engineering embodiment, a main pipeline section beneath a road corresponds to one partition object, and a branch pipeline section corresponds to another partition object. Two valves and three key nodes are deployed within the main pipeline section partition object, continuously receiving pressure, flow, and location data. During its early morning cycle, the third processing module receives a set of real-time monitoring data. It detects that two location data points fall within the main pipeline section partition boundary, and that each location corresponds to a key node and a valve. Therefore, this set of real-time monitoring data is linked to the main pipeline section partition object. Subsequently, the third processing module reads the historical monitoring data of this partition object from the past seven days and merges it into the same record sequence according to monitoring time. Next, the third processing module discovers that the valve had a historical maintenance record from the previous day, and that an inspection log was uploaded to the same location on the same day, with the inspection description pointing to the same valve. Therefore, both the historical maintenance record and the inspection log are simultaneously linked to the corresponding marked point of the valve. If an accident report is filed that evening, and the accident location falls near the key node, the accident report continues to be linked to the same partition object and enters the accident layer. Through this operation, a continuous monitoring chain, maintenance chain, inspection chain, and accident chain are formed within the main pipeline section zoning object, while the branch pipeline section zoning object only retains data records within its own boundary, and the two are not confused.

[0098] Specifically, after the third processing module completes the connection process, it outputs multi-source heterogeneous monitoring data. This multi-source heterogeneous monitoring data includes partition numbers, marked points, real-time monitoring data sequences, historical monitoring data sequences, historical maintenance record sequences, inspection log sequences, accident report sequences, object correspondences, temporal order relationships, spatial order relationships, and connection anomaly records. Understandably, the aforementioned partition numbers, real-time monitoring data sequences, historical monitoring data sequences, historical maintenance record sequences, inspection log sequences, and accident report sequences constitute the direct input for S400, which performs temporal feature encoding, spatial feature encoding, semantic text data feature encoding, and cross-modal fusion processing. The aforementioned marked points and object correspondences are further invoked in S600, which performs the drawing of partition boundaries, marked points, affected valves, affected pipe sections, affected users, and pressure change points. The aforementioned connection anomaly records are written into the knowledge base update link, awaiting update processing after the subsequent safety warning results are generated.

[0099] In summary, this step integrates the partitioned objects with real-time monitoring data, historical monitoring data, historical maintenance records, inspection logs, and accident reports into a single operational chain, forming multi-source heterogeneous monitoring data with temporal, spatial, and object correspondence relationships. This step writes the marked points into the partitioned objects during the attachment phase, and subsequent feature encoding and drawing processes revolve around the same record structure. This step retains cross-regional accident records and attachment anomaly records within the multi-source heterogeneous monitoring data, ensuring continuous input for subsequent joint processing and knowledge base updates.

[0100] S400. For the multi-source heterogeneous monitoring data, temporal encoding is performed using a temporal convolutional network, spatial encoding is performed using a graph convolutional network, and semantic text encoding is performed using an attention mechanism network. The encoding results are then fused across modalities to obtain a fused feature vector.

[0101] Specifically, this step is executed by the heterogeneous feature encoding and fusion module, with the input source being the multi-source heterogeneous monitoring data output by S300. The multi-source heterogeneous monitoring data includes partition numbers, marked points, real-time monitoring data sequences, historical monitoring data sequences, historical maintenance record sequences, inspection log sequences, accident report sequences, object correspondences, temporal order relationships, spatial order relationships, and connection anomaly records. The temporal feature encoding corresponds to the real-time monitoring data sequences and historical monitoring data sequences. The spatial feature encoding corresponds to partition numbers, marked points, object correspondences, spatial order relationships, and connection relationship fragments. The semantic text data feature encoding corresponds to the historical maintenance record sequences, inspection log sequences, accident report sequences, and connection anomaly records. The cross-modal fusion processing corresponds to the aforementioned three types of encoding results. The heterogeneous feature encoding and fusion module is started after the multi-source heterogeneous monitoring data is written, and re-encoding processing is triggered when connection anomaly records are added, partition numbers are changed, or marked points are updated.

[0102] Specifically, the heterogeneous feature encoding and fusion module first performs temporal feature encoding. This temporal feature encoding employs a temporal convolutional network (TCNN). The TCNN consists of an input segment, a convolutional segment, and an output segment. The input segment receives real-time monitoring data sequences and historical monitoring data sequences under the same partition number. The convolutional segment performs continuous convolution on pressure data, flow data, temperature data, and hazardous gas data according to their temporal order. This processing chain progresses according to monitoring time. First, it reads the real-time monitoring data sequence of the current operating cycle, then reads the historical monitoring data sequence of adjacent operating cycles, and then merges the records of the same object before and after each labeled point. If a labeled point has a supplementary monitoring record, the heterogeneous feature encoding and fusion module places this record at the end of the current time window and retains the supplementary mark. If a labeled point only has a historical monitoring data sequence and no real-time monitoring data sequence, then this labeled point is transferred to the supplementary record list and does not enter the current convolutional segment. The convolutional segment outputs the temporal feature vector within the same partition number and records this result as a temporal feature vector field.

[0103] Furthermore, the heterogeneous feature encoding and fusion module performs spatial feature encoding. This spatial feature encoding employs a graph convolutional network (Graph Convolutional Network). The Graph Convolutional Network consists of a node input segment, a connection input segment, and an aggregation segment. The node input segment receives the labeled points, key node sets, and valve sets under the partition number. The connection input segment receives object correspondences, spatial order relationships, and connection fragments. The aggregation segment reads nodes and connections in the order of preceding and following connections within the partition boundary and performs aggregation processing on adjacent labeled points, adjacent key nodes, and adjacent valves. If there are cross-regional accident records in the accident report sequence, the heterogeneous feature encoding and fusion module only extracts the labeled points and object correspondences falling within the current partition number; the portion crossing the current partition boundary is retained in the adjacent partition number. If the boundary verification marker indicates that the current partition boundary is under review, the Graph Convolutional Network pauses cross-boundary aggregation and only performs aggregation on nodes within the boundary. The aggregation segment outputs a spatial topological feature vector and records this result as a spatial feature vector field.

[0104] Specifically, after completing temporal and spatial feature encoding, the heterogeneous feature encoding and fusion module performs semantic text data feature encoding. This semantic text data feature encoding employs an attention-based neural network model. This neural network model receives historical maintenance record sequences, inspection log sequences, accident report sequences, and connection anomaly records. In the historical maintenance record sequence, maintenance time, maintenance pipe section, maintenance valve, and maintenance result are grouped as a set of text objects. In the inspection log sequence, inspection time, inspection location, inspection object, and inspection description are grouped as a set of text objects. In the accident report sequence, accident time, accident location, accident object, and handling result are grouped as a set of text objects. In the connection anomaly records, monitoring supplementary records, maintenance inspection conflict records, and cross-regional accident records are grouped as anomaly text objects. The attention-based neural network model first reads the above text objects according to the partition number, then assigns the text objects to corresponding annotation points according to object correspondence, and then performs attention allocation according to the temporal order, assigning higher weights to text objects more strongly associated with the current partition number, while retaining the original weights of cross-regional text objects and adding cross-regional labels. The processing link outputs a semantic text feature vector, and the result is recorded as a semantic text feature vector field.

[0105] Furthermore, the heterogeneous feature encoding and fusion module performs cross-modal fusion processing. This cross-modal fusion processing includes concatenation, a multi-head self-attention network (MSA), and refinement and dimensionality reduction. The MSA is a Multi-HeadSelf-Attention Network. The concatenation operation first reads the temporal feature vector field, spatial feature vector field, and semantic text feature vector field corresponding to the same partition number, and concatenates them according to the order of the labeled points. The MSA then performs multiple rounds of attention allocation on the concatenated result, ensuring that the monitoring data, connectivity relationships, and text records within the same partition number form a unified reading order. Refinement and dimensionality reduction then compress duplicate content, retaining the main content related to the partition number, labeled points, object correspondence, temporal order, and spatial order. If a partition number has an abnormal connection record, the abnormality marker is retained in the cross-modal fusion processing, and the corresponding features are not deleted. After processing, the heterogeneous feature encoding and fusion module obtains the fused feature vector and records the result as the fused feature vector field.

[0106] In one engineering embodiment, a main pipeline section beneath a road corresponds to a zone number. This zone number contains three marked points, two valves, and multiple pressure and flow data. The heterogeneous feature encoding and fusion module first feeds the pressure and flow data of the current operating cycle along with the historical monitoring data sequence of the past seven days into a temporal convolutional network to obtain a temporal feature vector field. Subsequently, the three marked points, two valves, corresponding object relationships, and spatial order relationships are fed into a graph convolutional network to obtain a spatial feature vector field. Next, the historical maintenance record sequence, inspection log sequence, and accident report sequence under this zone number are fed into an attention-based neural network model to obtain a semantic text feature vector field. Finally, the heterogeneous feature encoding and fusion module feeds the three types of feature vector fields into a concatenated operation, a multi-head self-attention network, and refinement and dimensionality reduction processing to generate a fused feature vector field corresponding to this zone number. This fused feature vector field is then used by the S500 for "joint processing of confidence intervals, intelligent alarm commands, sudden alarm commands, maximum clustering distance, and connectivity relationships."

[0107] Understandably, after this step is completed, the output includes a temporal feature vector field, a spatial feature vector field, a semantic text feature vector field, and a fused feature vector field. The temporal feature vector field, spatial feature vector field, and semantic text feature vector field are retained within the heterogeneous feature encoding and fusion module. The fused feature vector field serves as the main output of this step and enters S500, participating in subsequent processing in conjunction with the aforementioned connection relationships. The aforementioned output also retains the correspondence between partition numbers and labeled points, for review and processing in S600 for drawing partition boundaries, labeled points, and pressure change points.

[0108] In summary, this step integrates monitoring data, connectivity relationships, and text records within the same partition number into a unified fusion feature vector. The aforementioned processing chain unfolds along the partition number and marked points, with clear connections between steps. Abnormal records are retained in this step and entered into the fusion feature vector field, ensuring continuity in subsequent alarm status information processing.

[0109] S500. Based on the fused feature vector and its real-time monitoring data, generate a confidence interval and determine whether an intelligent alarm or a sudden alarm is triggered; perform clustering distance calculation on the adjacent marked points that trigger the alarm, and generate associated alarm status information in combination with the connection relationship;

[0110] Specifically, this step is executed by the fourth processing module, with the fused feature vector output by S400 as the input source. It simultaneously calls upon real-time monitoring data sequences, historical monitoring data sequences, labeled points, partition numbers, object correspondences, temporal order relationships, spatial order relationships, and connection relationship fragments from the multi-source heterogeneous monitoring data. The confidence interval is an interval record of the historical monitoring data sequence within the same partition number. The intelligent alarm command is an alarm record generated by the fourth processing module based on the confidence interval, the fused feature vector, and the real-time monitoring data sequence. The sudden change alarm command is an alarm record generated by the fourth processing module based on the real-time monitoring data sequence, a preset pressure data threshold, and the fused feature vector. The maximum clustering distance is the distance record used when adjacent labeled points within the same partition number are expanded along the connection relationship fragment. The joint processing of connection relationships is a process of placing the intelligent alarm command, the sudden change alarm command, the maximum clustering distance, and the connection relationship fragment into the same processing link to generate associated alarm status information. The fourth processing module starts this step after the fused feature vector is written. It triggers recalculation processing when new data sequences are added in real-time monitoring, historical data sequences are updated, labeled points are updated, or connection relationship fragments are updated.

[0111] Specifically, the fourth processing module first generates confidence intervals. It reads historical monitoring data sequences by partition number and then divides them into historical windows based on the location of the markers and the time sequence. Each historical window retains the original arrangement of pressure, flow, temperature, and hazardous gas data, without disrupting the correspondence between the objects. The fourth processing module performs interval merging on historical windows with the same marker, forming a confidence interval for that marker; it then performs partition merging on multiple markers within the same partition number, forming a confidence interval for that partition number. If supplementary monitoring records exist in the historical monitoring data sequence, these records are placed in adjacent historical windows according to the supplementary marker and still participate in interval merging. If monitoring data corresponding to cross-regional accident records exists in the historical monitoring data sequence, the fourth processing module only retains records falling within the current partition boundary. After the confidence intervals are generated, the fourth processing module writes the upper edge, lower edge, time range, and marker location of the interval into the current partition number and records this information as the confidence interval field.

[0112] Further, the fourth processing module generates intelligent alarm commands. The fourth processing module reads the fused feature vector and the real-time monitoring data sequence of the current operating cycle according to the partition number, and then reads the confidence interval field corresponding to each marked point. If the real-time monitoring data sequence corresponding to a certain marked point exceeds the confidence interval field, and the fused feature vector corresponding to that marked point maintains the same object correspondence with the historical monitoring data sequence, then the fourth processing module writes an intelligent alarm record for that marked point. If multiple consecutive marked points within the same partition number have intelligent alarm records, the fourth processing module retains their spatial and temporal order relationships and does not merge them in this stage. If there are abnormal connection records for the same marked point, the fourth processing module adds an abnormal marker to the intelligent alarm record and continues to send this marker to subsequent connection relationship joint processing. After the intelligent alarm command is generated, the fourth processing module writes the partition number, marked point, intelligent alarm record, and abnormal marker into the intelligent alarm command field.

[0113] Specifically, the fourth processing module generates a sudden change alarm command after the intelligent alarm command. The fourth processing module reads the pressure data of the current operating cycle from the real-time monitoring data sequence, and then reads the pressure data of the adjacent operating cycles and a preset pressure data threshold. The preset pressure data threshold is the pressure data discrimination record of the current underground pipeline network under this zone number. The fourth processing module compares the pressure data changes of the two operating cycles in chronological order, and then combines this with the feature changes of the current marked point in the fused feature vector to determine whether the change constitutes a sudden change record. If the pressure data change crosses the preset pressure data threshold, and the fused feature vector does not switch the connection relationship segment displayed at the same marked point, the fourth processing module writes a sudden change alarm record at that marked point. If multiple marked points within the same zone number simultaneously write sudden change alarm records, the fourth processing module retains their sequential positions according to spatial order. After the sudden change alarm command is generated, the fourth processing module writes the zone number, marked point, sudden change alarm record, and corresponding pressure data into the sudden change alarm command field.

[0114] Further, the fourth processing module generates the maximum clustering distance and performs joint processing of connection relationships. The maximum clustering distance is read from the marked points within the same partition number that have already been written into intelligent alarm records or mutation alarm records. The fourth processing module expands segment by segment along the connection relationship, first calculating the distance along the line between adjacent marked points, and then performing clustering distance filtering on multiple distances along the line within the current partition number to obtain the maximum clustering distance field for the current partition number. If two marked points are spatially close but their connection relationship segments are not continuous, they are not written into the same maximum clustering distance field. During the joint processing of connection relationships, the fourth processing module reads the intelligent alarm instruction field, mutation alarm instruction field, maximum clustering distance field, and connection relationship segment simultaneously. If multiple marked points are continuously distributed on the same connection relationship segment and the distance between them does not exceed the maximum clustering distance field, the fourth processing module assigns the aforementioned marked points to the same associated alarm record. If multiple marked points are spatially close but belong to different connection relationship segments, the fourth processing module splits them into different associated alarm records. If the marked point corresponding to the cross-regional accident record is connected to the boundary of the current partition number, the fourth processing module retains the cross-regional mark and writes the boundary mark into the associated alarm record.

[0115] Specifically, the fourth processing module summarizes the associated alarm status information after generating associated alarm records. This associated alarm status information includes the partition number, associated alarm command, associated alarm record, scope of influence, correspondence of marked points, correspondence of connection segments, correspondence of affected valves, correspondence of affected pipe segments, correspondence of affected users, and correspondence of pressure change points. The process of generating the scope of influence is as follows: the fourth processing module reads the key node set, valve set, and marked points forward and backward along the connection segments corresponding to the associated alarm record, and then writes the key nodes, valves, pipe segments, and users belonging to the same associated alarm record into the same scope of influence. If boundary markers exist, the scope of influence ends at the current partition boundary, and the portion crossing partitions is retained in the adjacent partition number. After completing this process, the fourth processing module records the aforementioned content uniformly as the associated alarm status information field.

[0116] In one engineering embodiment, a main pipeline section beneath a road corresponds to a zone number. This zone number contains three marked points and two valves. The fourth processing module first reads the historical monitoring data sequence of the past seven days, generating confidence interval fields for each of the three marked points. Then, it reads the real-time monitoring data sequence and fused feature vector of the current operating cycle, finding that the pressure data of the first and second marked points both exceed their respective confidence interval fields, thus writing intelligent alarm records for these two marked points. Next, the fourth processing module compares the pressure data of the current operating cycle with that of adjacent operating cycles, finding that the pressure data change at the second marked point crosses a preset pressure data threshold, thus writing a sudden change alarm record for the second marked point. Then, the fourth processing module calculates the distance along the connection relationship segment between the first and second marked points. This distance falls into the maximum clustering distance field of the current zone number, thus grouping these two marked points into the same associated alarm record. Finally, the fourth processing module reads the valve set and pipeline record along the connection relationship segment corresponding to this associated alarm record, forming the scope of influence, and writes the correspondence between affected valves, affected pipelines, and pressure change points. The associated alarm status information is then sent to the S600 for retrieval.

[0117] Understandably, the output of this step is an associated alarm status information field. This field includes the partition number, associated alarm command, associated alarm record, affected area, correspondence of marked points, correspondence of connection segments, correspondence of affected valves, correspondence of affected pipe segments, correspondence of affected users, and correspondence of pressure change points. These fields directly serve as input to S600's "drawing process for partition boundaries, marked points, affected valves, affected pipe segments, affected users, and pressure change points based on the associated alarm status information." Meanwhile, the partition number and associated alarm command are retained within the fourth processing module for subsequent knowledge base update processing.

[0118] Summary of the technical effects of this step: This step integrates confidence intervals, intelligent alarm commands, mutation alarm commands, maximum clustering distance, and connection relationship fragments into the same processing chain, forming associated alarm status information centered around the partition number. The output of this step already includes the scope of influence and the corresponding relationships of affected objects; subsequent rendering processes unfold along the same object chain. Cross-regional records and anomaly markers are retained in this step, ensuring continuity in subsequent security warning results and knowledge base update processing.

[0119] S600. Based on the associated alarm status information, draw the partition boundary, the marked points that triggered the alarm, and the affected valves, pipe sections, users, and pressure change points in the visualization layer to generate a safety warning result.

[0120] Specifically, this step is executed by the drawing module, with the input source being the associated alarm status information output by S500. The associated alarm status information already includes the partition number, associated alarm command, associated alarm record, affected area, correspondence of marked points, correspondence of connection segments, correspondence of affected valves, correspondence of affected pipe segments, correspondence of affected users, and correspondence of pressure change points. The partition boundary is the boundary content written to the partition object by S200. The marked points are the point content corresponding to the partition number after the S300 connection processing. The affected valves, affected pipe segments, and affected users are the object content merged by S500 along the connection segments and affected area. The pressure change points are the point content retained by S500 based on the sudden alarm command and associated alarm record. The drawing module starts this step after the associated alarm status information is written, triggering redrawing processing when the partition boundary is updated, the marked points are updated, the affected area changes, or the associated alarm command changes.

[0121] Specifically, the drawing module first performs partition boundary drawing processing. The drawing module reads the partition boundary corresponding to the partition number, then reads the correspondence between the influence range and the connection relationship fragment, and maps the influence range fragment falling within the current partition boundary to the visualization layer. The visualization layer first writes the partition number, then the partition boundary, and then the influence range boundary. If the influence range crosses the current partition boundary, the drawing module truncates the range falling within this partition number according to the current partition boundary, and retains the portion that crosses the boundary in the waiting queue of the adjacent partition number. If there is a boundary marker in the associated alarm status information, the drawing module retains the original partition boundary and overlays the influence range boundary, without rewriting the original partition boundary record. After completing this processing, the drawing module writes the partition boundary corresponding to the current partition number into the layer record and records this content as the partition boundary field.

[0122] Further, the drawing module performs annotation point drawing processing after writing the partition boundaries. The drawing module reads all annotation points within the current partition number according to the annotation point correspondence, and then adjusts the display order of the annotation points according to the temporal and spatial order relationships in the associated alarm records. For annotation points simultaneously associated with both intelligent alarm commands and sudden alarm commands, the drawing module first writes the associated alarm records, and then writes the corresponding annotation points. For annotation points only associated with intelligent alarm commands, the drawing module writes them point by point along the connection relationship segment in sequential order. For annotation points only associated with sudden alarm commands, the drawing module writes a pressure change marker at the original annotation point location. If the same annotation point corresponds to maintenance records, inspection logs, and accident reports simultaneously, the drawing module retains the same point, does not split it into multiple locations, and attaches the corresponding content to the same layer object. After completing this processing, the drawing module writes the annotation point corresponding to the current partition number into the layer record and records this content as the annotation point field.

[0123] Specifically, after the marked points are written, the drawing module performs drawing processing for affected valves, affected pipe segments, and affected users. The drawing module first reads the correspondence of affected valves, then the correspondence of affected pipe segments, and finally the correspondence of affected users. The drawing position of the affected valve is determined by the valve's original position and the current partition boundary. The drawing range of the affected pipe segment is determined by the correspondence of the connection relationship segments and the affected range. The drawing position of the affected user is determined by the correspondence of the affected user and the current partition number. If the same affected valve is located at the intersection of two adjacent partition boundaries, the drawing module prioritizes retaining the valve drawing content within the current partition boundary and retains the corresponding content from the adjacent partition numbers in the drawing queue. If the same affected pipe segment spans multiple key nodes, the drawing module writes it segment by segment along the connection relationship segments, without directly merging segments. If there is no corresponding relationship for affected users in the current partition number, the drawing module retains the original record in the associated alarm status information and does not add new user positions. After completing this processing, the drawing module writes the affected valves, affected pipe segments, and affected users into the layer record, and records this content as the affected valve field, affected pipe segment field, and affected user field, respectively.

[0124] Further, the drawing module performs pressure change point drawing processing. The drawing module reads the correspondence between the pressure change points and compares them point-by-point with the labeled point field. If a pressure change point is already written in the labeled point field, the drawing module overlays the pressure change record at the original labeled point location without generating a new point. If a pressure change point is not written in the labeled point field, the drawing module adds it to the current partition boundary according to the connection relationship fragments and the spatial order relationship. If a pressure change point corresponds to a cross-zone accident record, the drawing module only writes the portion falling within the current partition boundary, leaving the portion outside the boundary in the adjacent partition number. After completing this processing, the drawing module writes the pressure change points to the layer record and records this content as the pressure change point field.

[0125] In one engineering embodiment, a main pipeline segment under a road corresponds to a zone number. The associated alarm status information generated by S500 shows that within this zone number there are two marked points, one affected valve, two affected pipe segments, and one pressure change point, with the influence extending to the downstream valve location. The drawing module first writes the zone boundary of this zone number into the visualization layer, then overlays the influence range onto the area inside the zone boundary. Subsequently, the drawing module writes two marked points inside the zone boundary, and overlays one of the points simultaneously associated with a sudden change alarm command as a pressure change point. Next, the drawing module writes the affected valve and two affected pipe segments along the connection relationship segments, and writes the affected users according to the positional relationship of the current zone number. If there is a cross-zone pipe segment outside the downstream valve, this segment is not drawn in this zone number, but is reserved in the drawing queue of the adjacent zone number. After completing the above processing, a layer object is formed in the visualization layer, consisting of the zone boundary, marked points, affected valve, affected pipe segment, affected users, and pressure change point.

[0126] Specifically, the drawing module generates a safety warning result after completing all drawing processes. This safety warning result includes a partition number, partition boundary field, marked point field, affected valve field, affected pipe segment field, affected user field, pressure change point field, and associated alarm command. Understandably, the aforementioned elements collectively constitute the fields of the safety warning result. The partition number, partition boundary field, marked point field, affected valve field, affected pipe segment field, affected user field, and pressure change point field are written into the safety warning result as output field names in this step and are subsequently called in the knowledge base update process, corresponding to historical monitoring data, historical maintenance records, inspection logs, accident reports, and partition boundary update links. The associated alarm command is retained in the safety warning result for subsequent handling results to be written and called.

[0127] In summary, this step transforms associated alarm status information into safety warning results dominated by zone boundaries, maintaining the same processing order for preceding zone objects and subsequent layer objects. Zone boundaries, marked locations, affected valves, affected pipe sections, affected users, and pressure change points are output within the same processing chain, with knowledge base updates directly invoked. Cross-zone content retains the separation relationship between inside and outside the boundary in this step, ensuring continuity in subsequent zone boundary update chains.

[0128] Example 2: Figure 2 This diagram illustrates a structural block diagram of an underground pipe network zoning visualization and safety early warning system according to an embodiment of the present invention. Figure 2 As shown, the structure may include:

[0129] The first processing module 01 is used to acquire map service sources, underground pipeline services, geographic information system data, building information model (BIM) data, equipment data, historical monitoring data, historical maintenance records, inspection logs, and accident reports, and perform data alignment and standardization processing to obtain pipeline network topology information. Specifically, the first processing module receives road locations, manhole cover locations, and building locations from the map service source; pipe segments, key nodes, valves, and reference locations of marked points from the underground pipeline service; spatial locations and road layers from the geographic information system data; pipe segment models, valve models, and manhole cover models from the BIM; pressure data, flow data, temperature data, hazardous gas data, and location data from the equipment data; and time and object records from historical monitoring data, historical maintenance records, inspection logs, and accident reports. The first processing module first performs location correspondence processing, then performs time sequence merging processing, and then performs object consistency verification processing. The location correspondence processing groups road locations, spatial locations, pipe segment models, valve models, and manhole cover models into the same underground pipeline network object. The time-sequence merging process arranges equipment data, historical monitoring data, historical maintenance records, inspection logs, and accident reports into a unified timeline based on monitoring time, maintenance time, inspection time, and accident time. The object consistency verification process maps maintenance objects, inspection objects, and accident objects to pipe segments, key nodes, and valves. For location conflict records and object conflict records, the first processing module writes an anomaly flag and retains the original record. The first processing module outputs pipeline topology information, which includes pipe segments, key nodes, valves, connection relationships, monitoring coverage density, reference locations of marked points, and their corresponding relationships. This pipeline topology information is then passed to the second processing module as input for the association processing of key nodes, valves, connection relationships, and monitoring coverage density.

[0130] The second processing module 02 receives the pipeline topology information and performs correlation processing on key nodes, valves, connection relationships, and monitoring coverage density to obtain partition objects. Specifically, the second processing module receives key nodes, valves, connection relationships, monitoring coverage density, and reference positions of marked points from the pipeline topology information from the first processing module. The second processing module first traces continuous pipe segments according to connection relationships, then identifies branch and merging positions according to key nodes, then identifies boundary positions according to valves, and then reads the monitoring coverage density to verify the boundary positions. For areas with dense key nodes, the second processing module checks whether adjacent connection relationships are continuous; for areas with large valve spacing, the second processing module checks whether the monitoring coverage density is interrupted; for positions marked in historical maintenance records, inspection logs, and accident reports, the second processing module calls the corresponding relationships to perform boundary re-verification. After the boundary re-verification is completed, the second processing module generates a partition number, partition boundary, key node set, valve set, connection relationship fragment, and boundary verification mark. The partition number, partition boundary, key node set, valve set, connection relationship segment, monitoring coverage density classification result and boundary verification mark together constitute the partition object, which is passed to the third processing module as the input object for the attachment processing.

[0131] The third processing module 03 receives the partition object and, based on equipment data, marked points, and connection relationships, performs concatenation processing on real-time monitoring data, historical monitoring data, historical maintenance records, inspection logs, and accident reports to obtain multi-source heterogeneous monitoring data. Specifically, the third processing module receives the partition number, partition boundary, key node set, valve set, connection relationship fragment, and boundary verification mark from the partition object in the second processing module, and simultaneously receives equipment data and marked points. The third processing module first filters real-time monitoring data and historical monitoring data according to the partition boundary, then merges monitoring records at the same location according to the marked points, and then concatenates historical maintenance records, inspection logs, and accident reports according to the connection relationship fragments to the corresponding key nodes, valves, or pipe sections. For records that fall within the current partition boundary and have consistent object correspondence, the third processing module writes the current partition number; for accident reports and inspection logs that cross adjacent partition boundaries, the third processing module retains the boundary mark and forms cross-region records; for monitoring records with missing marked points, the third processing module supplements the marked points based on the location data. After the connection is completed, the third processing module outputs multi-source heterogeneous monitoring data. The multi-source heterogeneous monitoring data includes partition number, marked point location, real-time monitoring data sequence, historical monitoring data sequence, historical maintenance record sequence, inspection log sequence, accident report sequence, object correspondence, time sequence, spatial sequence, and connection anomaly record. The multi-source heterogeneous monitoring data is transmitted to the heterogeneous feature encoding and fusion module as the input object for encoding and fusion processing.

[0132] The heterogeneous feature encoding and fusion module 04 is used to receive the multi-source heterogeneous monitoring data, perform temporal feature encoding, spatial feature encoding, semantic text data feature encoding, and cross-modal fusion processing to obtain a fused feature vector. Specifically, the heterogeneous feature encoding and fusion module receives real-time monitoring data sequences, historical monitoring data sequences, marked points, object correspondences, temporal order relationships, spatial order relationships, historical maintenance record sequences, inspection log sequences, and accident report sequences from the multi-source heterogeneous monitoring data from the third processing module. The heterogeneous feature encoding and fusion module first performs temporal feature encoding on the real-time monitoring data sequences and historical monitoring data sequences to form temporal feature vectors within the same partition number; then it performs spatial feature encoding on the marked points, object correspondences, spatial order relationships, and connection relationship segments to form spatial feature vectors; subsequently, it performs semantic text data feature encoding on the historical maintenance record sequences, inspection log sequences, accident report sequences, and attached abnormal records to form semantic text feature vectors. After the three types of feature vectors are generated, the heterogeneous feature encoding and fusion module performs concatenation, attention allocation, and refinement processing on the aforementioned feature vectors, writing features belonging to the same partition number and the same labeled point into a unified record. The heterogeneous feature encoding and fusion module outputs a fused feature vector, which includes the partition number, labeled point, temporal feature content, spatial feature content, and semantic text feature content. The fused feature vector is passed to the fourth processing module as the input object for joint processing.

[0133] The fourth processing module 05 is used to receive the fused feature vector, generate a confidence interval based on historical monitoring data, generate a sudden change alarm command based on real-time monitoring data and a preset pressure data threshold, and perform joint processing based on the intelligent alarm command, the sudden change alarm command, the maximum clustering distance, and the connection relationship to obtain associated alarm status information. Specifically, the fourth processing module receives the fused feature vector from the heterogeneous feature encoding and fusion module, and simultaneously calls the historical monitoring data sequence, the real-time monitoring data sequence, the marked points, the connection relationship fragments, and the object correspondence. The fourth processing module first merges historical windows from the historical monitoring data sequence according to the partition number and the marked points to form a confidence interval; then it reads the real-time monitoring data sequence of the current operating cycle and writes the records that exceed the confidence interval into the intelligent alarm command; subsequently, it reads the preset pressure data threshold, performs comparison processing on the pressure data difference between the current operating cycle and the adjacent operating cycle, and writes the records that meet the judgment conditions into the sudden change alarm command. After the intelligent alarm command and the sudden change alarm command are generated, the fourth processing module calculates the distance along the line between adjacent marked points along the connection relationship segment to form the maximum clustering distance. Then, the intelligent alarm command, the sudden change alarm command, the maximum clustering distance, and the connection relationship segment are put into the same processing link to merge continuously distributed marked points whose distances fall within the current range, forming associated alarm records. The fourth processing module outputs associated alarm status information, which includes the partition number, associated alarm command, associated alarm record, affected area, correspondence of marked points, correspondence of connection relationship segments, correspondence of affected valves, correspondence of affected pipe segments, correspondence of affected users, and correspondence of pressure change points. The associated alarm status information is transmitted to the drawing module as the input object for drawing processing.

[0134] The drawing module 06 receives the associated alarm status information and, based on the associated alarm status information and the scope of influence, performs drawing processing on the partition boundaries, marked points, affected valves, affected pipe sections, affected users, and pressure change points to obtain a safety warning result. Specifically, the drawing module receives the partition number, scope of influence, correspondence of marked points, correspondence of affected valves, correspondence of affected pipe sections, correspondence of affected users, and correspondence of pressure change points from the associated alarm status information from the fourth processing module, and calls the partition boundaries output by the second processing module. The drawing module first writes the partition number and partition boundary in the visualization layer, then maps the scope of influence to the corresponding partition boundary, and then writes the marked points according to the correspondence of marked points. After the marked points are written, the drawing module sequentially writes the affected valves, affected pipe sections, and affected users along the connection relationship segments, and superimposes the corresponding pressure change points onto the written marked points or independent point positions. For cross-zone content, the drawing module extracts the drawing objects falling within the current partition number according to the current partition boundary, and retains the objects outside the boundary for adjacent partition numbers. After the drawing is completed, the drawing module outputs a safety warning result, which includes a partition number, partition boundary field, marked point field, affected valve field, affected pipe section field, affected user field, pressure change point field, and associated alarm command. The safety warning result is passed to the update module as the input object for update processing.

[0135] The update module 07 receives the safety warning results and handling results, writes the handling results into the knowledge base, and updates historical monitoring data, historical maintenance records, inspection logs, accident reports, and zoning boundaries. Specifically, the update module receives the zoning number, zoning boundary field, marked point field, affected valve field, affected pipe section field, affected user field, pressure change point field, and associated alarm command from the drawing module in the safety warning results, and receives the handling results. The update module first maps the handling results to the zoning number and associated alarm command, and then writes the corresponding results into the knowledge base. After the knowledge base is written, the update module writes back the historical monitoring data, historical maintenance records, inspection logs, and accident reports according to the zoning number, writes the handling time and handling object for records that have been handled, and rewrites the zoning boundary for locations where boundary changes have occurred. If the handling results involve cross-zoning records, the update module synchronously calls the zoning boundary field and object correspondence of adjacent zoning numbers to perform a joint update. After the update is completed, the update module returns the new historical monitoring data, historical maintenance records, inspection logs, accident reports, and partition boundaries to the first and second processing modules for use in the next round of data alignment, data standardization, and correlation processing. The first and second processing modules are connected, the second processing module is connected to the third processing module, the third processing module is connected to the heterogeneous feature encoding and fusion module, the heterogeneous feature encoding and fusion module is connected to the fourth processing module, the fourth processing module is connected to the drawing module, and the drawing module is connected to the update module. Specifically, the above connections are passed sequentially according to data flow and control flow. The output object of the previous module is written into the module interface as the input object of the next module. The update object returned by the update module re-enters the first and second processing modules, forming a closed-loop processing chain.

Claims

1. A method for visualizing and providing safety early warning of underground pipe network zones, characterized in that, include: S100. Acquire multi-source pipeline data, perform spatial alignment, time synchronization and object merging processing to obtain pipeline topology information including pipe segments, key nodes, valves and their connection relationships; wherein, the multi-source pipeline data includes map service source, underground pipeline service, GIS data, BIM model, equipment data and historical operation and maintenance data; S200. Based on the pipeline topology information, trace the continuous path along the connection relationship, identify and verify the key nodes and valves to delineate candidate areas, and then correct the boundary of the candidate areas according to the monitoring coverage density of the equipment data to generate partition objects. S300. Based on the partition boundaries of the partition object and the key nodes and valves it contains, real-time and historical monitoring data, maintenance records, inspection logs and accident reports are linked to the corresponding partition object to form multi-source heterogeneous monitoring data. S400. For the multi-source heterogeneous monitoring data, temporal encoding is performed using a temporal convolutional network, spatial encoding is performed using a graph convolutional network, and semantic text encoding is performed using an attention mechanism network. The encoding results are then fused across modalities to obtain a fused feature vector. S500. Based on the fused feature vector and its real-time monitoring data, generate a confidence interval and determine whether an intelligent alarm or a sudden alarm is triggered; perform clustering distance calculation on the adjacent marked points that trigger the alarm, and generate associated alarm status information in combination with the connection relationship; S600. Based on the associated alarm status information, draw the partition boundary, the marked points that triggered the alarm, and the affected valves, pipe sections, users, and pressure change points in the visualization layer to generate a safety warning result.

2. The method according to claim 1, characterized in that, The process of acquiring multi-source pipeline network data, performing spatial alignment, time synchronization, and object merging to obtain pipeline network topology information including pipe segments, key nodes, valves, and their connection relationships includes: The service maps pipe segments in underground pipeline services to spatial locations in GIS data, maps building information model components in BIM models to road and building locations in map service sources, and merges records of key nodes, valves, and manhole covers falling into the same location range into the same object. Synchronize timestamps according to monitoring time, and merge pressure data, flow data, temperature data, hazardous gas data and location data within the same monitoring time window into the same object; Arrange the records in chronological order according to maintenance time, inspection time, and accident time, and then match them to objects according to pipe section, key node, and valve, so that maintenance records, inspection records, and accident records are linked to spatial objects one by one. The pipeline network topology information also includes monitoring coverage density, reference locations of marked points, correspondence of historical maintenance records, correspondence of inspection logs, and correspondence of accident reports.

3. The method according to claim 2, characterized in that, The process of tracing continuous paths along the aforementioned connections, identifying and verifying the key nodes and valves to delineate candidate regions includes: The process of identifying and verifying the key nodes and valves to delineate candidate regions includes a first round of association: Tracing adjacent pipe segments along the connection relationship, identifying key nodes on the same continuous path, reading the position of the valve before and after in the continuous path, and grouping pipe segments and key nodes that are between adjacent valves and have a continuous connection relationship into the same candidate area; when there is a situation in the candidate area where a key node corresponds to two valves at the same time, calling the historical maintenance record correspondence, the inspection log correspondence, and the accident report correspondence for secondary verification. The secondary verification prioritizes checking the historical maintenance record, then checks the inspection log, and then checks the accident report, retaining the candidate area that is consistent with the maintenance, inspection, and accident directions, and writing the conflict content into the boundary verification mark.

4. The method according to claim 3, characterized in that, The process of correcting the boundaries of the candidate regions and generating partition objects based on the monitoring coverage density of the device data includes: The step of correcting the boundary of the candidate region based on the monitoring coverage density includes a second round of association: The number of equipment data sources, the frequency of historical monitoring data records, and the distribution status of the reference positions of the marked points are read segment by segment along the candidate region. For areas where the equipment data source is interrupted, the historical monitoring data records are sparse, or the reference positions of the marked points are obviously separated, new boundary candidates are generated at the corresponding key nodes or valves. The partition object includes partition number, partition boundary, key node set, valve set, connection relationship segment, monitoring coverage density classification result, and boundary verification mark.

5. The method according to claim 4, characterized in that, The process of linking real-time and historical monitoring data, maintenance records, inspection logs, and incident reports to the corresponding partition objects to form multi-source heterogeneous monitoring data includes: The attachment process includes object filtering, hierarchical attachment, association verification, and supplementary recording of marked points; Object filtering includes: reading location data from real-time monitoring data and checking whether it falls within the partition boundary; reading historical locations and historical objects from historical monitoring data and checking whether they are consistent with the set of key nodes, the set of valves, and the connection relationship fragments; and reading maintenance pipe sections, maintenance valves, inspection objects, accident objects, and accident locations from historical maintenance records, inspection logs, and accident reports and checking whether they match the key nodes, valves, and connection relationships within the partition object. The hierarchical connection includes monitoring layer connection, maintenance layer connection, inspection layer connection and accident layer connection. Among them, the monitoring layer connection merges real-time monitoring data and historical monitoring data according to the zone number and arranges them in order of monitoring time. The maintenance layer connection merges historical maintenance records according to maintenance time and maintenance object. The inspection layer connection merges inspection logs according to inspection location and inspection object. The accident layer connection merges accident reports according to accident location and accident object. The correlation verification includes: checking the temporal continuity between real-time monitoring data and historical monitoring data, checking the consistency of historical maintenance records, inspection logs and accident reports with the objects of key node sets, valve sets and connection relationship fragments, and writing monitoring supplementary records, maintenance and inspection conflict records and cross-regional accident records into the connection anomaly record; The annotation point supplementation includes: generating annotation points around the key node set, valve set and accident location along the connection relationship segment, and attaching real-time monitoring data, historical monitoring data, historical maintenance records, inspection logs and accident reports to the corresponding annotation points respectively; The multi-source heterogeneous monitoring data includes partition numbers, marked points, real-time monitoring data sequences, historical monitoring data sequences, historical maintenance record sequences, inspection log sequences, accident report sequences, and connection anomaly records.

6. The method according to claim 5, characterized in that, The process of using a temporal convolutional network for temporal encoding, a graph convolutional network for spatial encoding, and an attention mechanism network for semantic text encoding to obtain the multi-source heterogeneous monitoring data, and then fusing the encoding results across modalities to obtain the fused feature vector, includes: The temporal coding includes: receiving real-time monitoring data sequences and historical monitoring data sequences under the same partition number from the input segment of the temporal convolutional network; performing continuous convolution on pressure data, flow data, temperature data and harmful gas data in chronological order from the convolutional segment; and outputting a temporal feature vector. The spatial encoding includes: receiving the labeled points, key node sets, and valve sets under the partition number by the node input segment of the graph convolutional network; connecting the object correspondence, spatial order relationship, and connection relationship fragments received by the input segment; and reading the nodes and connections in the order of connection inside the partition boundary by the aggregation segment and performing aggregation processing on adjacent labeled points, adjacent key nodes, and adjacent valves, and outputting a spatial topology feature vector. The semantic text encoding includes: receiving a sequence of historical maintenance records, a sequence of inspection logs, a sequence of accident reports, and a record of attached abnormalities by a neural network model based on an attention mechanism; reading the above text objects according to the partition number; assigning the text objects to the corresponding annotation points according to the object correspondence; performing attention allocation according to the time order; and outputting a semantic text feature vector. The cross-modal fusion includes concatenation operation, multi-head self-attention network, and refinement and dimensionality reduction. The concatenation operation reads the temporal feature vector, spatial feature vector, and semantic text feature vector corresponding to the same partition number and concatenates them in the order of the labeled points. The multi-head self-attention network performs multiple rounds of attention allocation on the concatenation result. After refining and dimensionality reduction, duplicate content is compressed and the fused feature vector is output.

7. The method according to claim 6, characterized in that, The process of generating confidence intervals and determining whether to trigger smart alarms or sudden alarms includes: The process of generating confidence intervals includes: reading historical monitoring data sequences by partition number, dividing historical windows according to the relationship between the marked points and time sequence, performing interval merging on the historical windows of the same marked point to form the confidence interval of that marked point, and then performing partition merging on multiple marked points within the same partition number to form the confidence interval of that partition number. The determination of whether to trigger an intelligent alarm includes: reading the fused feature vector and the real-time monitoring data sequence of the current operating cycle according to the partition number, reading the confidence interval field corresponding to each marked point one by one, and if the real-time monitoring data sequence exceeds the confidence interval field and the fused feature vector corresponding to the marked point is consistent with the object correspondence of the historical monitoring data sequence, then writing an intelligent alarm record at the marked point.

8. The method according to claim 7, characterized in that, The process of calculating the clustering distance between adjacent marked points that trigger the alarm, and generating associated alarm status information based on the connection relationship, includes: Expand along the connection relationship segment segment by segment, calculate the line distance between adjacent marked points, perform cluster distance filtering on multiple line distances within the current partition number to obtain the maximum cluster distance field. If two marked points are spatially close but their connection relationship segments are not continuous, they are not written into the same maximum cluster distance field. Simultaneously read in the intelligent alarm record, the sudden alarm record, the maximum cluster distance field, and the connection relationship segment, and classify the marked points that are continuously distributed on the same connection relationship segment and whose distance does not exceed the maximum cluster distance field into the same associated alarm record. The associated alarm status information includes partition number, associated alarm command, associated alarm record, scope of influence, correspondence of marked points, correspondence of connection relationship segments, correspondence of affected valves, correspondence of affected pipe segments, correspondence of affected users, and correspondence of pressure change points. The scope of influence is generated by reading the key node set, valve set, and marked points forward and backward along the connection relationship segment corresponding to the associated alarm record.

9. The method according to claim 8, characterized in that, The process of drawing the partition boundaries, alarm trigger points, and affected valves, pipe sections, users, and pressure change points in the visualization layer to generate safety warning results includes: Read the partition boundary corresponding to the partition number, read the correspondence between the influence range and the connection relationship fragment, map the influence range fragment falling inside the current partition boundary to the visualization layer, and if the influence range crosses the current partition boundary, then cut off the range falling inside this partition number according to the current partition boundary. Read all the marked points within the current partition number according to the correspondence of the marked points, adjust the display order of the marked points according to the time order and spatial order in the associated alarm records, write the associated alarm records first and then write the corresponding marked points for marked points that are associated with both smart alarms and sudden alarms, and write pressure change markers at the original marked point positions for marked points that are only associated with sudden alarms. Read the correspondence of affected valves, the correspondence of affected pipe segments, and the correspondence of affected users. The drawing position of the affected valves is determined by the original position of the valves and the current partition boundary. The drawing range of the affected pipe segments is determined by the correspondence of the connection fragments and the affected range. The drawing position of the affected users is determined by the correspondence of the affected users and the current partition number. Read the corresponding relationship of the pressure change points and compare it point by point with the drawn marked points. If the pressure change point is not written in the marked point field, then write it into the current partition boundary according to the connection relationship fragment and spatial order relationship. The safety warning results include the zone number, zone boundary, marked location, affected valves, affected pipe sections, affected users, pressure change locations, and associated alarm commands.

10. An underground pipeline network zoning visualization and safety early warning system, applied to the method of any one of claims 1-9, characterized in that, include: The first processing module is used to acquire map service sources, underground pipeline services, GIS data, BIM models, equipment data, historical monitoring data, historical maintenance records, inspection logs and accident reports, and perform data alignment and data standardization processing to obtain pipeline network topology information; The second processing module is used to perform correlation processing of key nodes, valves, connection relationships and monitoring coverage density based on the pipeline network topology information to obtain partition objects; The third processing module is used to perform real-time monitoring data, historical monitoring data, historical maintenance records, inspection logs and accident reports on the partition object to obtain multi-source heterogeneous monitoring data. The heterogeneous feature encoding and fusion module is used to perform temporal feature encoding, spatial feature encoding, semantic text data feature encoding and cross-modal fusion processing based on the multi-source heterogeneous monitoring data to obtain a fused feature vector; The fourth processing module is used to perform joint processing of confidence interval, intelligent alarm command, mutation alarm command, maximum clustering distance and connection relationship based on the fused feature vector to obtain associated alarm status information; The drawing module is used to draw the partition boundaries, marked points, affected valves, affected pipe sections, affected users, and pressure change points based on the associated alarm status information, and to obtain the safety warning results.

Images