A method and system for automatically laying out monitoring points coupled with hierarchical pipe network and catchment zoning

Abstract

The present application relates to the technical field of municipal drainage engineering, and provides a kind of coupling pipe network grading and catchment zoning delineation monitoring point automatic layout method and system, by obtaining and preprocessing multi-source drainage data, the dynamic knowledge graph that is updated with pipe network facilities and monitoring point change is constructed;Based on the dynamic knowledge graph, a pipe network topology graph is constructed, a graph convolution network is used to determine pipe level labels, and the level labels are used as constraints for dividing catchment zones by a deep reinforcement learning network;Using a graph attention network to calculate the importance of the catchment zones, the monitoring equipment is configured and the monitoring points are determined according to the differences;Access real-time monitoring data, incrementally adjust the above network, and update the pipe grading, catchment zoning and monitoring point layout results in linkage. The present application can match the catchment zone boundary with the catchment range of different level pipes, and improve the matching degree of monitoring resource configuration and waterlogging risk.

Description

Technical Field

[0001] This invention relates to the field of municipal drainage engineering technology, and in particular to an automatic deployment method and system for monitoring points that couples pipe network classification and catchment zone delineation. Background Technology

[0002] Urban stormwater drainage networks are responsible for collecting, transferring, and discharging rainfall runoff. Network classification, catchment area delineation, and monitoring point deployment are fundamental tasks in urban drainage and flood control planning and network operation and maintenance. Network classification identifies main, secondary, and branch drainage channels; catchment area delineation determines the spatial allocation of surface runoff and network confluence; and monitoring point deployment collects operational data such as rainfall, water level, and flow rate.

[0003] In existing technologies, pipeline classification typically employs depth-first search or multi-index hierarchical decision-making methods, catchment area delineation typically employs methods based on Thiessen polygons, pipeline topology iteration, or DEM basin analysis, and monitoring point placement typically employs methods based on graph theory indices or comprehensive evaluation ranking.

[0004] However, pipeline classification, catchment area delineation, and monitoring point deployment are usually treated as independent technical steps. Pipeline level labels are not involved in catchment area delineation, resulting in the use of the same zoning rules for areas corresponding to different pipeline levels. The zoning boundaries are difficult to reflect the actual water conveyance and catchment functions of the pipeline network. The scope of the catchment area, the composition of the pipeline level, and the historical flood risk are not used as a unified basis for the configuration of monitoring equipment and the selection of monitoring points, which can easily lead to insufficient monitoring in key areas and duplicate deployment in general areas.

[0005] Furthermore, when new or modified pipelines are added, or when real-time monitoring data reflects new risks of urban flooding, existing methods usually require independent recalculation, making it difficult to update the pipeline classification, catchment area, and monitoring point layout results simultaneously.

[0006] Therefore, it is necessary to propose an automatic deployment method and system for monitoring points that couples pipeline classification and catchment area delineation to solve or at least alleviate the above-mentioned defects. Summary of the Invention

[0007] The main objective of this invention is to provide an automatic monitoring point deployment method and system that couples pipeline network classification and catchment area delineation, in order to solve the technical problems of how to match the catchment area boundary with the catchment range of different pipeline levels, how to use the importance of the catchment area for differentiated configuration of monitoring equipment and determination of monitoring point locations, and how to automatically update the pipeline network classification, catchment area delineation, and monitoring point deployment as pipeline facilities and monitoring data change.

[0008] To achieve the above objectives, this invention provides an automatic monitoring point deployment method that couples pipeline network classification with catchment area delineation, comprising the following steps: S1. Acquire multi-source drainage data of the target area and preprocess it to obtain a standardized dataset; S2, construct a dynamic knowledge graph based on the standardized dataset, and update the dynamic knowledge graph incrementally as new pipeline facilities or monitoring points are added; S3. Construct a pipeline network topology map based on the dynamic knowledge graph, process the pipeline network topology map using a graph convolutional network, and output the hierarchical labels of each pipeline. The hierarchical labels include at least the trunk level, the secondary trunk level, and the branch level. S4, using the hierarchical labels as state constraints of the deep reinforcement learning network, the deep reinforcement learning network is used to divide the catchment area, wherein the zoning constraint for the main pipeline level is stronger than that for the secondary pipeline level, and the secondary pipeline level is stronger than that for the branch pipeline level; output the catchment area zoning result. S5, taking the water catchment area as a node, calculate the importance of each water catchment area using a graph attention network based on the area attribute information, and configure monitoring equipment and determine monitoring points according to the importance difference; S6, access real-time monitoring data to update the dynamic knowledge graph. When the preset triggering conditions are met, incremental adjustments are made to the graph convolutional network, deep reinforcement learning network and graph attention network to update the pipeline's hierarchical labels, water catchment zoning results and monitoring point layout scheme.

[0009] Preferably, the multi-source drainage data includes stormwater pipe network data, water system data, topographic elevation data, and historical hydrological monitoring data; the preprocessing includes the following steps: The spatial data in the rainwater pipe network data, the water system data, the topographic elevation data, and the historical hydrological monitoring data are uniformly converted to the same plane rectangular projection coordinate system; Correct at least one type of abnormal and invalid data, such as pipeline breakpoints, reverse slopes, elevation jumps, and overlapping polygons, using GIS topology inspection tools; A unified field specification is established, and the multi-source drainage data with unified coordinates and anomaly correction is stored in a spatial database to obtain the standardized dataset.

[0010] Preferably, step S2 includes the following steps: S21, Based on the standardized dataset, construct the graph entities and semantic relationships between them; wherein, the graph entities include pipes, manholes, storm drains, storm drain outlets, water storage facilities, river systems, terrain grid units, and monitoring points; the semantic relationships include the upstream and downstream hydraulic connectivity of pipes, the receiving relationship between storm drain outlets and river systems, the spatial dependency relationship between drainage facilities and terrain grid units, and the monitoring correspondence between monitoring points and manholes or pipes; predefine the subordinate relationship type between catchment areas and pipes, and establish the corresponding subordinate relationship after the catchment areas are divided; S22, the rainfall data is associated with the corresponding rainfall monitoring points, and the rainfall data is associated with the corresponding terrain grid units according to the spatial correspondence between the rainfall monitoring points and the terrain grid units; the water level data and flow rate data are associated with the manholes or pipes where the monitoring points are installed; the water accumulation data is associated with the terrain grid units or manholes according to the water accumulation location; the collection time of each data is written into the temporal attribute of the graph entity associated with each data, thus obtaining the dynamic knowledge graph; S23. When adding or modifying pipeline facilities, update the corresponding graph entities and affected semantic relationships based on the identifier, spatial location, and pipeline connection relationship of the added or modified pipeline facilities. When adding monitoring points, update the corresponding graph entities and affected semantic relationships based on the identifier, installation location, and monitoring object of the added monitoring points. Retain the unchanged graph entities and semantic relationships to obtain the updated dynamic knowledge graph.

[0011] Preferably, step S3 includes the following steps: S31, map the manholes and rainwater outlets in the dynamic knowledge graph as nodes in the pipeline topology graph, and map the pipelines as directed edges connecting the nodes along the water flow direction; S32, determine the water conveyance capacity characteristics of each pipeline based on its nominal pipe diameter and laying slope, determine the water system modulation characteristics based on the cumulative path length of the pipeline along the water flow direction to its final downstream rainwater outlet, and normalize and weight the water conveyance capacity characteristics and the water system modulation characteristics to obtain the edge weight of each directed edge. S33, calculate the betweenness centrality, downstream cumulative pipe length and upstream total catchment area of ​​each node according to the pipeline network topology diagram to form node feature vectors; input the node feature vectors and the directed edge relationships of the pipeline network topology diagram into a graph convolutional network to obtain the feature representation of each node; S34, combine the feature representations of the upstream and downstream nodes of each pipeline to form pipeline features, and determine the hierarchical label of each pipeline based on the pipeline features; the hierarchical label includes trunk level, secondary trunk level and branch level.

[0012] Preferably, step S4 includes the following steps: S41, the target area is discretized into multiple terrain grid units. Based on the surface flow direction and pipeline spatial location of each terrain grid unit, the target pipeline receiving the surface runoff of that terrain grid unit is determined, and the hierarchical label of the target pipeline is used as the grid hierarchical label of that terrain grid unit. The state characteristics of each terrain grid unit are determined, including terrain elevation, surface slope, distance to the nearest main pipeline, distance to the nearest secondary pipeline, grid hierarchical label, current catchment area number, and historical waterlogging risk score. S42, using a deep reinforcement learning network as the decision model, the deep reinforcement learning network takes the state characteristics of each terrain grid unit as input and outputs the division action corresponding to each terrain grid unit. The division action is to divide the terrain grid unit into an existing spatially adjacent catchment area or to create a new catchment area with the terrain grid unit as a seed. S43, During training, when a terrain grid cell is assigned to a catchment area that does not contain its target pipeline, a hierarchical matching penalty is applied; wherein, when the hierarchical labels of the target pipeline are main trunk level, secondary trunk level, and branch level, the hierarchical matching penalty corresponds to a first negative reward, a second negative reward, and a third negative reward, respectively, and the absolute value of the first negative reward is greater than the absolute value of the second negative reward, and the absolute value of the second negative reward is greater than the absolute value of the third negative reward; S44, the reward function of the deep reinforcement learning network is obtained by weighted summation of the hierarchical matching penalty, hydraulic connectivity reward, terrain consistency reward, area balance reward and historical waterlogging risk gain according to their respective weights; S45. After training, the water catchment zoning scheme is obtained according to the division action of each terrain grid unit output by the deep reinforcement learning network. Water catchment zoning with an area smaller than the preset minimum area threshold is merged into water catchment zoning that is spatially adjacent and hydraulically connected to it. The boundary of the merged zoning is smoothed and the boundary of the zoning is aligned with the boundary of the road or river to obtain the water catchment zoning result.

[0013] Preferably, step S5 includes the following steps: S51, using the water catchment area as a node and the hydraulic connectivity between water catchment areas from the upstream water catchment area to the downstream water catchment area according to the direction of water flow in the pipeline network as directed edges, construct a zoning topology graph; S52, determine the node characteristics of each catchment area, the node characteristics include the catchment area, the number of main pipelines in the catchment area, the number of secondary pipelines in the catchment area, the upstream, midstream or downstream location level of the catchment area in the pipeline network and the historical waterlogging risk score. S53, input the node features of each watershed and the directed edge relationships of the partition topology graph into the graph attention network to obtain the feature representation of each watershed, and calculate the importance score of each watershed based on the feature representation of each watershed. S54. Sort the catchment areas from high to low according to the importance score, configure online monitoring equipment for the top 30% of the catchment areas, configure liquid level monitoring equipment for the catchment areas after the top 30% but not exceeding the top 70%, and configure liquid level alarm devices for the manholes in the catchment areas after the top 70% that are located at historical water accumulation points or pipeline overflow risk points. S55. Within each catchment area, the intermediate centrality of each manhole in the catchment area is calculated based on the pipeline topology map. The manholes with the highest intermediate centrality are selected as monitoring candidate points. The monitoring candidate points are then weighted and sorted according to their distances from historical water accumulation points and pipeline overflow risk points to determine the monitoring points within the catchment area.

[0014] Preferably, step S6 includes the following steps: The real-time monitoring data of the target area and its collection time are associated with the corresponding graph entities in the dynamic knowledge graph, and the dynamic knowledge graph is updated. Incremental training will begin when any of the following triggering conditions are met: The first triggering condition is: the cumulative rainfall in any of the durations of 1 hour, 2 hours, 3 hours or 24 hours reaches the standard of once in ten years or above for that duration, and the same location experiences flooding twice in a row. The second triggering condition is: the total length of newly added or modified pipelines entering the warehouse exceeds 5% of the total length of the original pipelines in the target area; The incremental training is as follows: based on the original model parameters of the graph convolutional network, the deep reinforcement learning network, and the graph attention network, the updated data in the dynamic knowledge graph is used as the incremental training data, the learning rate is set to 1 / 10 of the initial learning rate, and the graph convolutional network, the deep reinforcement learning network, and the graph attention network are incrementally trained in sequence, and the updated pipeline hierarchical labels, water catchment zoning results, and monitoring point layout scheme are output.

[0015] Preferably, the main pipeline stage has a nominal pipe diameter greater than or equal to 1000 mm or an upstream total catchment area greater than or equal to 3 km². 2 The pipeline is defined as follows: the secondary trunk level is a pipeline with a nominal diameter greater than or equal to 600 mm and less than 1000 mm, and the branch level is a pipeline with a nominal diameter less than 600 mm.

[0016] Preferably, S3 further includes building a digital twin base; wherein the digital twin base is used to uniformly manage the dynamic knowledge graph, the pipeline topology map, the hierarchical labels of the pipelines, the catchment area results, and the monitoring point layout scheme, and to access the real-time monitoring data; based on the real-time monitoring data, the dynamic knowledge graph, the pipeline topology map, the hierarchical labels of the pipelines, and the catchment area results, rainstorm simulation is performed, and graded urban flooding early warning results are output.

[0017] The present invention also provides an automatic monitoring point deployment system that couples pipeline network classification and catchment area delineation, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the automatic monitoring point deployment method that couples pipeline network classification and catchment area delineation as described above.

[0018] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention transmits the pipe hierarchy labels output by the graph convolutional network to the catchment area partitioning process, and sets different strengths of partition constraints according to the main, secondary, and branch pipes, so that the catchment area corresponding to the main pipe remains intact, the catchment area corresponding to the secondary pipe maintains hydraulic connection, and the area corresponding to the branch pipe can be adjusted according to the local terrain. Compared with the existing methods that process the pipe network hierarchy and catchment area separately, this invention allows the pipe hierarchy information to actually participate in the catchment area decision, which not only improves the matching degree between the partition boundary and the pipe catchment area, but also avoids the main pipe being cut by the partition boundary or the branch area being over-constrained due to the same constraints applied to pipes of different levels.

[0019] (2) This invention constructs the catchment area as nodes in a topological graph with hydraulic connectivity, and uses a graph attention network to calculate the importance of the catchment area based on the area, pipeline hierarchy, upstream and downstream locations, and historical flood risk. Then, monitoring equipment is configured and monitoring points are determined according to the importance. By combining the configuration of equipment at the catchment level with the selection of monitoring points within the catchment area, the equipment type, deployment level, and location of the monitoring points are adapted to the drainage function and flood risk of the catchment area, forming a coordinated deployment effect from catchment evaluation to monitoring point determination.

[0020] (3) This invention associates real-time monitoring data, changes in pipeline facilities, and changes in monitoring points with a dynamic knowledge graph, and makes incremental adjustments to the graph convolutional network, deep reinforcement learning network, and graph attention network, so that the updated pipeline hierarchical labels are transmitted to the catchment area division process, and the updated catchment area division results are further transmitted to the monitoring point deployment process. It can maintain the correspondence between the three types of results after changes in pipeline structure or drainage status, and avoid inconsistencies in the hierarchical, zoning, and deployment results caused by separate updates.

[0021] (4) This invention transforms the originally independent pipeline hierarchy, catchment area and monitoring point layout into a sequential process by constraining the catchment area by pipeline hierarchy, driving the monitoring layout by the importance of the catchment area, and triggering the linkage update by real-time data. This can simultaneously improve the problems of catchment area boundary matching, monitoring resource allocation and dynamic updating of results. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0023] Figure 1 This is a schematic flowchart of one embodiment of the present invention.

[0024] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0025] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0027] Furthermore, the technical solutions of the various embodiments can be combined with each other, but only if they are feasible for those skilled in the art. If the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0028] Please refer to Figure 1 The present invention provides an automatic deployment method for monitoring points that couples pipe network classification and catchment area delineation, comprising the following steps: S1. Acquire multi-source drainage data of the target area and preprocess it to obtain a standardized dataset; S2, construct a dynamic knowledge graph based on the standardized dataset, and update the dynamic knowledge graph incrementally as new pipeline facilities or monitoring points are added; S3. Construct a pipeline network topology map based on the dynamic knowledge graph, process the pipeline network topology map using a graph convolutional network, and output the hierarchical labels of each pipeline. The hierarchical labels include at least the trunk level, the secondary trunk level, and the branch level. S4, using the hierarchical labels as state constraints of the deep reinforcement learning network, the deep reinforcement learning network is used to divide the catchment area, wherein the zoning constraint of the main pipeline is stronger than that of the secondary pipeline, and the secondary pipeline is stronger than that of the branch pipeline; output the catchment area zoning result. S5, taking the water catchment area as a node, calculate the importance of each water catchment area using a graph attention network based on the area attribute information, and configure monitoring equipment and determine monitoring points according to the importance difference; S6, access real-time monitoring data to update the dynamic knowledge graph. When the preset triggering conditions are met, incremental adjustments are made to the graph convolutional network, deep reinforcement learning network and graph attention network to update the pipeline's hierarchical labels, water catchment zoning results and monitoring point layout scheme.

[0029] (1) This invention transmits the pipe hierarchy labels output by the graph convolutional network to the catchment area partitioning process, and sets different strengths of partition constraints according to the main, secondary, and branch pipes, so that the catchment area corresponding to the main pipe remains intact, the catchment area corresponding to the secondary pipe maintains hydraulic connection, and the area corresponding to the branch pipe can be adjusted according to the local terrain. Compared with the existing methods that process the pipe network hierarchy and catchment area separately, this invention allows the pipe hierarchy information to actually participate in the catchment area decision, which not only improves the matching degree between the partition boundary and the pipe catchment area, but also avoids the main pipe being cut by the partition boundary or the branch area being over-constrained due to the same constraints applied to pipes of different levels.

[0030] (2) This invention constructs the catchment area as nodes in a topological graph with hydraulic connectivity, and uses a graph attention network to calculate the importance of the catchment area based on the area, pipeline hierarchy, upstream and downstream locations, and historical flood risk. Then, monitoring equipment is configured and monitoring points are determined according to the importance. By combining the configuration of equipment at the catchment level with the selection of monitoring points within the catchment area, the equipment type, deployment level, and location of the monitoring points are adapted to the drainage function and flood risk of the catchment area, forming a coordinated deployment effect from catchment evaluation to monitoring point determination.

[0031] (3) This invention associates real-time monitoring data, changes in pipeline facilities, and changes in monitoring points with a dynamic knowledge graph, and makes incremental adjustments to the graph convolutional network, deep reinforcement learning network, and graph attention network, so that the updated pipeline hierarchical labels are transmitted to the catchment area division process, and the updated catchment area division results are further transmitted to the monitoring point deployment process. It can maintain the correspondence between the three types of results after changes in pipeline structure or drainage status, and avoid inconsistencies in the hierarchical, zoning, and deployment results caused by separate updates.

[0032] (4) This invention transforms the originally independent pipeline hierarchy, catchment area and monitoring point layout into a sequential process by constraining the catchment area by pipeline hierarchy, driving the monitoring layout by the importance of the catchment area, and triggering the linkage update by real-time data. This can simultaneously improve the problems of catchment area boundary matching, monitoring resource allocation and dynamic updating of results.

[0033] Preferably, the multi-source drainage data includes stormwater pipe network data, water system data, topographic elevation data, and historical hydrological monitoring data; the preprocessing includes the following steps: The spatial data in the rainwater pipe network data, the water system data, the topographic elevation data, and the historical hydrological monitoring data are uniformly converted to the same plane rectangular projection coordinate system; The stormwater pipe network data includes the pipe start point, end point, pipe diameter, pipe length, laying slope, and upstream and downstream connection relationships. When collecting the stormwater pipe network data, micro-terminal branch pipes with a nominal pipe diameter of less than 200mm and isolated pipe sections without confluence access are manually screened out in advance to avoid such pipe sections interfering with the pipe network topology construction and hierarchical calculation.

[0034] The water system data includes the spatial boundaries of rivers and lakes, as well as the locations of stormwater outfalls. The topographic elevation data includes a digital elevation model (DEM), which comprises DEM raster grids with a resolution of at least 5 meters. Global topographic elevation, surface slope, and topographic watershed ridges are extracted from the DEM raster using hydrological analysis tools.

[0035] The historical hydrological monitoring data includes cumulative rainfall and peak rainfall intensity for each rainfall event over the years, pipe network water level records, pipe network flow records, historical water accumulation point plane coordinates, maximum water accumulation depth, and inundation range vectors. The multi-source drainage data also includes real-time IoT monitoring data, which includes continuous time-series records of water level, flow rate, and rainfall collected from monitoring points at 5-minute intervals.

[0036] Spatial data from the aforementioned stormwater drainage network data, water system data, topographic elevation data, and historical hydrological monitoring data are uniformly converted to the same Cartesian projection coordinate system. This embodiment uses the CGCS2000 Cartesian projection coordinate system. For time-series data such as rainfall, water level, and flow rate, the acquisition time is retained, and a connection is established with the corresponding manholes, pipelines, or topographic grid units through monitoring point identifiers or spatial locations. By unifying spatial coordinates, pipeline, water system, topographic elevation data, and historical water accumulation data can be uniformly overlaid, distance calculated, and location matched.

[0037] Correct at least one type of abnormal and invalid data, such as pipeline breakpoints, reverse slopes, elevation jumps, and overlapping polygons, using GIS topology inspection tools; Specifically, the rainwater pipe network data and water system data, after coordinate transformation, are imported into a GIS topology inspection tool. The topology inspection function is used to identify pipe network breakpoints and overlapping polygons, while the pipe network elevation and slope attributes are used to identify reverse slopes and elevation jumps. For identified anomalies, corrections are made based on the original pipe network connections, pipe endpoint locations, manhole locations, and original elevation data. Specifically, pipe network breakpoints are corrected to restore the connection between pipes and manholes or adjacent pipes; reverse slopes and elevation jumps are corrected to ensure that the water flow direction in the pipes aligns with the slope direction and that the elevation difference between connected pipes at manholes is within a preset reasonable range; and overlapping polygons are corrected to eliminate duplicate or invalid surface features. Spatial relationships confirmed to be part of the actual engineering structure are preserved.

[0038] A unified field specification is established, and the multi-source drainage data, after coordinate unification and anomaly correction, is stored in a spatial database to obtain the standardized dataset. The unified field specification is established to organize the fields of the data after coordinate transformation and anomaly correction. Fields include a unique identifier for the data object, spatial location, and data category; for data with time attributes, the field also includes the collection time. Specifically, stormwater pipe network data also includes pipe diameter, pipe length, slope, and upstream / downstream connection relationships; historical hydrological monitoring data also includes the monitoring object, monitoring value, and collection time.

[0039] The stormwater drainage network data, water system data, topographic elevation data, and historical hydrological monitoring data, after field processing, are stored in a spatial database to form the standardized dataset. This results in all types of data having unified coordinate references, object identifiers, field structures, and time information, thus providing a unified data input for subsequent dynamic knowledge graph construction and pipeline topology map generation.

[0040] In a preferred embodiment, S2 includes the following steps: S21, Based on the standardized dataset, construct the graph entities and semantic relationships between them; wherein, the graph entities include pipes, manholes, storm drains, storm drain outlets, water storage facilities, river systems, terrain grid units, and monitoring points; the semantic relationships include the upstream and downstream hydraulic connectivity of pipes, the receiving relationship between storm drain outlets and river systems, the spatial dependency relationship between drainage facilities and terrain grid units, and the monitoring correspondence between monitoring points and manholes or pipes; predefine the subordinate relationship type between catchment areas and pipes, and establish the corresponding subordinate relationship after the catchment areas are divided; The dynamic knowledge graph is built upon the data and interfaces provided by the digital twin platform, employing ontology modeling tools, a graph database engine, GIS preprocessing tools, data flow tools, and automated interaction scripts. Specifically, the ontology modeling tool constructs the OWL domain ontology for the pipeline network; the graph database engine stores the relationships between graph nodes and hydraulic systems; the GIS preprocessing tool performs vector data cleaning and standardized export; the data flow tool is a distributed message queue middleware built into the digital twin platform, used to push real-time monitoring time-series data; and the automated interaction scripts call the graph database interface library, enabling bidirectional data synchronization between the relational spatial database and the graph database of the digital twin platform.

[0041] The map entities include pipes, manholes, storm drains, storm outlets, water storage facilities, river systems, and topographic grid units. The data attributes of the map entities include the nominal pipe diameter, pipe section length, and laying slope of the pipes; the planar coordinates of the manholes and storm drains; and the monitored water level and historical water depth.

[0042] Based on the connection relationship of the pipeline endpoints and the direction of water flow in the pipeline, establish upstream and downstream hydraulic connectivity between adjacent pipeline entities; based on the spatial location and actual access relationship between the rainwater outlet and the river system, establish receiving relationship between the rainwater outlet entity and the corresponding river system entity; based on the spatial superposition relationship between the drainage facility entity and the terrain grid unit entity, establish spatial dependency relationship between the drainage facility entity and the terrain grid unit entity; based on the installation location of the monitoring point and the monitoring object, establish monitoring correspondence between the monitoring point entity and the corresponding manhole entity or pipeline entity.

[0043] Predefine the subordination type between the catchment area and the pipeline; after obtaining the catchment area result, construct the catchment area entity according to the correspondence between the catchment area boundary and the pipeline spatial location, and establish the subordination relationship between the catchment area entity and the pipeline entity.

[0044] S22, the rainfall data is associated with the corresponding rainfall monitoring points, and the rainfall data is associated with the corresponding terrain grid units according to the spatial correspondence between the rainfall monitoring points and the terrain grid units; the water level data and flow rate data are associated with the manholes or pipes where the monitoring points are installed; the water accumulation data is associated with the terrain grid units or manholes according to the water accumulation location; the collection time of each data is written into the temporal attribute of the graph entity associated with each data, thus obtaining the dynamic knowledge graph; Perform batch import of static entities: The automated interactive script reads the standardized pipeline, terrain and water system vector data from the relational spatial database of the digital twin base, writes them in batches into the graph database to generate graph entities, and automatically matches the hydraulic association relationships defined in the ontology.

[0045] According to the data type of the monitoring, the rainfall data is associated with the corresponding rainfall monitoring point entity, and according to the spatial correspondence between the rainfall monitoring point and the terrain grid unit, the rainfall data is associated with the corresponding terrain grid unit entity.

[0046] The monitoring points are used as the data source for water level and flow rate data. The manhole or pipeline entity corresponding to the monitoring point is used as the monitored object. The association between the water level and flow rate data and the monitored object is established. The water accumulation data is associated with the corresponding terrain grid unit entity or manhole entity according to the water accumulation location.

[0047] The distributed message queue middleware pushes water level, flow rate, and rainfall data collected from monitoring points at 5-minute intervals to the graph database interface script, automatically updating the dynamic temporal attributes of manholes and pipeline entities within the graph database. The data values, data types, and collection times of various monitoring data are saved as temporal attributes of their associated graph entities. When the same graph entity corresponds to monitoring data from multiple collection times, the corresponding data values ​​are saved according to the collection time, resulting in the dynamic knowledge graph. Furthermore, data self-checking can be performed. The dynamic knowledge graph automatically traverses the topology, identifies and marks pipeline breakpoints and reverse slopes.

[0048] S23. When new or modified pipeline facilities or new monitoring points are added, the corresponding graph entities and affected semantic relationships are updated according to the identifiers, spatial locations and pipeline connection relationships of the new or modified pipeline facilities or new monitoring points, while the unchanged graph entities and semantic relationships are retained, resulting in an updated dynamic knowledge graph.

[0049] When new pipeline facilities are added, corresponding graph entities are added to the existing dynamic knowledge graph based on the unique identifier, spatial location, endpoint connection relationship and water flow direction of the new pipeline facilities, and the corresponding association relationship between them and adjacent pipelines, inspection wells, rainwater outlets, river systems or terrain grid units is established.

[0050] When upgrading pipeline facilities, update the attributes of the corresponding pipeline entities and their upstream and downstream hydraulic connectivity based on the upgraded pipeline location, endpoint connection relationship, pipe diameter, laying slope and water flow direction; update the original association relationship that no longer exists due to pipeline upgrade, and save the entity attributes and association relationship before the upgrade as historical state.

[0051] When a new monitoring point is added, a corresponding monitoring point entity is added to the existing dynamic knowledge graph based on the unique identifier, installation location, and monitoring object of the monitoring point, and a monitoring correspondence relationship is established between the monitoring point entity and the corresponding manhole entity or pipeline entity.

[0052] The incremental update is triggered by data change events on the digital twin platform, adding or updating only the corresponding entities and relationships without reconstructing the entire graph, thus achieving lightweight expansion. Unchanged graph entities and semantic relationships are preserved, resulting in an updated dynamic knowledge graph.

[0053] In a preferred embodiment, S3 includes the following steps: S31, map the manholes and rainwater outlets in the dynamic knowledge graph as nodes in the pipeline topology graph, and map the pipelines as directed edges connecting the nodes along the water flow direction; Based on the hydraulic connectivity between manholes, storm drains, and pipes in the dynamic knowledge graph, manholes and storm drains are mapped as nodes in a graph network, and pipes are mapped as directed edges connecting the nodes along the water flow direction. The direction of the directed edges is determined according to the actual water flow direction of the pipes.

[0054] S32, determine the water conveyance capacity characteristics of each pipeline based on its nominal pipe diameter and laying slope, determine the water system modulation characteristics based on the cumulative path length of the pipeline along the water flow direction to its final downstream rainwater outlet, and normalize and weight the water conveyance capacity characteristics and the water system modulation characteristics to obtain the edge weight of each directed edge. For each pipeline, the pipe diameter characteristic weight is determined based on the pipeline's nominal diameter and laying slope. Preferably, the pipe diameter characteristic weight is calculated using the following formula: w pipe =0.7· ln(D) / ln(D max )+ 0.3· (1-tanh(S)), D For pipe diameter, D max Where S is the maximum pipe diameter across the entire area, and S is the pipe slope. The larger the pipe diameter and the gentler the laying slope, the higher the corresponding pipe diameter weight value. The water system modulation factor is determined based on the straight-line distance from the pipeline along the water flow direction to the corresponding water system access node at the final downstream stormwater outlet. Preferably, the water system modulation factor is calculated using the following formula: w water =exp(- λ·d water) ), d water The cumulative path length is the sum of the lengths of each pipe segment traced downstream from the pipeline along the water flow direction to the final downstream storm drain outlet. l For the attenuation coefficient (as an example, take...), l =0.01m -1The closer the pipeline is to the water system access node, the higher the water system modulation weight, so as to prioritize the identification of the backbone drainage corridors along the water system. After normalizing the pipe diameter feature weight and the water system modulation factor, they are weighted according to preset weights to obtain the edge weight of the corresponding directed connection edge.

[0055] In this embodiment, the weighting coefficients for pipe diameter feature weight and water system modulation factor are 0.6 and 0.4, respectively. In other embodiments, the weighting coefficients are optimized on the validation set using a grid search method. The search range for the weighting coefficient of pipe diameter feature weight is 0.3 to 0.8, and the search range for the weighting coefficient of water system modulation factor is 0.2 to 0.7. The search step size is 0.1, and the accuracy of pipe network hierarchical classification is used as the optimization objective.

[0056] S33, calculate the betweenness centrality, downstream cumulative pipe length and upstream total catchment area of ​​each node according to the pipeline network topology diagram to form node feature vectors; input the node feature vectors and the directed edge relationships of the pipeline network topology diagram into a graph convolutional network to obtain the feature representation of each node; The network topology centrality is represented by the normalized value of the node's betweenness centrality, which reflects the node's transit role in the hydraulic path of the pipeline network; the downstream cumulative catchment capacity is measured by the total length of the downstream pipeline along the flow direction of the current node, which characterizes the catchment scale of the downstream pipeline network of the current node; the upstream total catchment area is determined according to the topographic catchment relationship and the pipeline connection relationship, which characterizes the upstream service range that can flow into the current node.

[0057] In this embodiment, the graph convolutional network adopts a three-layer stacked graph convolutional structure, with the hidden layer dimension of each graph convolutional layer set to 64. The input of the graph convolutional network is the node feature vector of each node. The graph convolutional network aggregates the features of adjacent nodes according to the weighted directed adjacency relationship, and performs node self-connection, adjacency relationship normalization, feature transformation and nonlinear activation processing in each graph convolutional layer to obtain the feature representation of each node layer by layer.

[0058] The graph convolutional network is trained under supervision using manually labeled and field-verified pipeline network-level samples. Since pipelines are represented as directed edges in the pipeline network topology graph, and pipeline level labels correspond to pipelines, for each pipeline, the feature representations of the upstream and downstream nodes of the pipeline are combined with the weights of the directed edges corresponding to the pipeline to form the pipeline feature.

[0059] S34, combine the feature representations of the upstream and downstream nodes of each pipeline to form pipeline features, and determine the hierarchical label of each pipeline based on the pipeline features; the hierarchical label includes trunk level, secondary trunk level and branch level.

[0060] The pipeline hierarchy samples are labeled according to the following standards: For main pipelines, the label value is 2: the nominal pipe diameter is greater than or equal to DN1000, or the downstream service catchment area of ​​the pipeline is greater than or equal to 3 km². 2 It is a key regional rainwater transfer corridor; Secondary dry level, with a label value of 1: The nominal pipe diameter is greater than or equal to DN600 and less than DN1000, and it undertakes the function of transferring and transporting water in the area. Branch line level, with a label value of 0: The nominal pipe diameter is less than DN600, and it is responsible for rainwater collection at the end of the plot.

[0061] When the nominal pipe diameter and the service catchment area correspond to different levels, the higher level is used as the labeling level for the pipe. Among them, DN1000 is classified as the main trunk level, and DN600 is classified as the secondary trunk level, so that the pipe diameter ranges of the main trunk level, secondary trunk level, and branch level do not overlap.

[0062] After training, the feature representations of the upstream and downstream nodes of the pipeline to be classified, along with the corresponding directed connection edge weights, are input into the pipeline hierarchical classification process. The pipeline is then output as a main trunk, secondary trunk, or branch level label, and the pipeline hierarchical label is transmitted to the subsequent catchment area division step.

[0063] In a preferred embodiment, S4 includes the following steps: S41, the target area is discretized into multiple terrain grid units. Based on the surface flow direction and pipeline spatial location of each terrain grid unit, the target pipeline receiving the surface runoff of that terrain grid unit is determined, and the hierarchical label of the target pipeline is used as the grid hierarchical label of that terrain grid unit. The state characteristics of each terrain grid unit are determined, including terrain elevation, surface slope, distance to the nearest main-level pipeline, distance to the nearest secondary-level pipeline, grid hierarchical label, current catchment area number, and historical waterlogging risk score. The terrain elevation is in meters, the surface slope is in percentage, the distances to the nearest main-level pipeline and the nearest secondary-level pipeline are in meters, the grid hierarchical label is assigned a value of 2 for main-level, 1 for secondary-level, and 0 for branch-level, and the historical waterlogging risk score is a normalized value between 0 and 1. The historical waterlogging risk score is determined using a conventional multi-index comprehensive evaluation method based on the historical water depth and water frequency in the historical hydrological monitoring data, with a value range of 0 to 1.

[0064] The target area is discretized into multiple terrain grid cells. For urban built-up areas, the area of ​​a single terrain grid cell is controlled within 0.25 ha; for open suburban areas, the area of ​​a single terrain grid cell is controlled within 1 ha. The centerlines of river channels and main pipelines are used as grid boundaries to avoid the river channels and main pipelines being cut off by unrelated grid boundaries.

[0065] Based on topographic elevation data, the surface runoff direction of each topographic grid unit is determined. Then, considering the spatial relationship between the topographic grid unit and the pipeline, the target pipeline receiving the surface runoff of that topographic grid unit is identified. The target pipeline refers to the pipeline into which the surface runoff of the topographic grid unit flows according to its surface runoff direction.

[0066] S42, using a deep reinforcement learning network as the decision model, the deep reinforcement learning network takes the state characteristics of each terrain grid unit as input and outputs the division action corresponding to each terrain grid unit. The division action is to divide the terrain grid unit into an existing spatially adjacent catchment area or to create a new catchment area with the terrain grid unit as a seed. Using a deep reinforcement learning network as the decision model for water catchment partitioning, the state features of the current terrain grid unit are input into the deep reinforcement learning network according to the preset terrain grid unit processing order (e.g., from upstream to downstream) to obtain the action value corresponding to each executable partitioning action in the current state.

[0067] The executable partitioning actions include: (1) partitioning the current terrain grid cell into an existing catchment area that is spatially adjacent to it; and (2) creating a new catchment area using the current terrain grid cell as a seed.

[0068] The action with the highest action value is selected from all executable actions and used as the action for the current terrain grid cell. The action value represents the expected cumulative reward after executing the action under the current state characteristics. After executing the action, the current catchment area number of the current terrain grid cell is updated, and the subsequent decision state is updated according to the updated catchment area number.

[0069] S43, During training, when a terrain grid cell is assigned to a catchment area that does not contain its target pipeline, a hierarchical matching penalty is applied; wherein, when the hierarchical labels of the target pipeline are main trunk level, secondary trunk level, and branch level, the hierarchical matching penalty corresponds to a first negative reward, a second negative reward, and a third negative reward, respectively, and the absolute value of the first negative reward is greater than the absolute value of the second negative reward, and the absolute value of the second negative reward is greater than the absolute value of the third negative reward; As a preferred example, the first negative reward, the second negative reward, and the third negative reward are -100, -30, and -5, respectively, and the absolute value of the first negative reward is greater than the absolute value of the second negative reward, and the absolute value of the second negative reward is greater than the absolute value of the third negative reward.

[0070] The hierarchical matching penalty is used to reduce the action value of assigning terrain grid units to catchment areas that do not contain their target pipelines, making the catchment affiliation of terrain grid units closer to the service area of ​​their target pipelines. Specifically, trunk pipelines typically serve larger catchment areas, and by applying the largest absolute first negative reward, it is ensured that trunk pipelines are located at the outlet of their respective catchment areas, thus matching the catchment area of ​​the catchment area with the upstream catchment topology of the pipeline. Secondary trunk pipelines undertake regional catchment and transfer functions, and are given a medium negative reward. Branch pipelines mainly undertake local end-point collection functions, allowing their corresponding terrain grid units to be appropriately adjusted based on local terrain and feature boundaries, and therefore are given a smaller negative reward.

[0071] S44, the reward function of the deep reinforcement learning network is obtained by weighted summation of the hierarchical matching penalty, hydraulic connectivity reward, terrain consistency reward, area balance reward and historical waterlogging risk gain according to their respective weights; The hydraulic connectivity bonus is determined based on the consistency between the surface runoff direction of the terrain grid unit and the water flow direction of the target pipeline. The hydraulic connectivity bonus is increased when the surface runoff of the terrain grid unit can flow into the target pipeline along the surface runoff direction, and when the surface runoff of each terrain grid unit within the same catchment area can flow into the same drainage channel within the catchment area along the pipeline water flow direction. The hydraulic connectivity bonus is decreased when the surface runoff direction is inconsistent with the target pipeline water flow direction, or when the terrain grid units within the same catchment area lack continuous drainage paths.

[0072] The terrain consistency bonus is determined based on the spatial relationship between the catchment area boundary and the topographic watershed extracted by the digital elevation model. The bonus is increased when the catchment area boundary extends along the topographic watershed; and decreased when the catchment area boundary crosses the topographic watershed. This ensures that the catchment area boundary is as consistent as possible with the catchment boundary formed by the natural terrain.

[0073] The area balancing reward is determined based on the deviation between the area of ​​each catchment zone and the average catchment zone area. When the deviation between the catchment zone area and the overall average catchment zone area is within ±50%, the area balancing reward is increased; when the deviation exceeds ±50%, the area balancing reward is decreased. The penalty intensity of the hierarchical matching penalty is positively correlated with the degree of hierarchical difference; that is, the penalty intensity when trunk-level pipelines and branch-level pipelines are mixed and assigned to the same zone is greater than the penalty intensity when trunk-level pipelines and secondary-level pipelines are mixed and assigned to the same zone.

[0074] The historical flood risk reward is determined based on the historical flood risk score of the terrain grid units. When the partitioning action results in relatively independent or finer catchment area assignments for terrain grid units with higher historical flood risk scores, the historical flood risk reward is increased. Conversely, when the partitioning action merges terrain grid units with higher historical flood risk scores with large areas of low-risk terrain grid units, leading to a decrease in the accuracy of high-risk area identification, the historical flood risk reward is decreased. The degree of adjustment of the historical flood risk reward increases with the increase of the historical flood risk score determined by historical water depth and water frequency.

[0075] As a preferred example, the weights for hydraulic connectivity reward, terrain consistency reward, hierarchical matching penalty, area balance reward, and historical flooding risk reward are set to 0.30, 0.25, 0.15, 0.10, and 0.20, respectively. Each reward is calculated based on the evaluation result corresponding to the current partitioning action, and the weighted sum is used as the comprehensive reward for the current partitioning action.

[0076] In this embodiment, the deep reinforcement learning network includes three fully connected hidden layers with 128, 64, and 32 neurons respectively. During training, an experience replay mechanism is employed, saving the state features of the terrain grid units, the partitioning action, the comprehensive reward, and the next state after performing the partitioning action to the experience replay pool. The network parameters are then trained using samples from the experience replay pool. Simultaneously, a periodic hard update method for the target network is used to improve the stability of the training process. When the network training loss reaches convergence, training stops, and the trained deep reinforcement learning network is obtained.

[0077] S45. After training, the water catchment zoning scheme is obtained according to the division action of each terrain grid unit output by the deep reinforcement learning network. Water catchment zoning with an area smaller than the preset minimum area threshold is merged into water catchment zoning that is spatially adjacent and hydraulically connected to it. The boundary of the merged zoning is smoothed and the boundary of the zoning is aligned with the boundary of the road or river to obtain the water catchment zoning result.

[0078] Identify catchment areas in the initial catchment zoning scheme whose areas are smaller than a preset minimum area threshold. For catchment areas without independent drainage outlets and whose areas are smaller than the preset minimum area threshold, merge them into catchment areas that are spatially adjacent and hydraulically connected. Smooth the boundaries of the merged catchment areas to reduce jagged boundaries formed by terrain grid boundaries. For zone boundary segments adjacent to roads or rivers and suitable as references for zone boundaries, adjust the zone boundary segments along the corresponding road or river boundary lines so that the zone boundary coincides with or substantially coincides with the corresponding road or river boundary; when boundary adjustment would disrupt the original surface flow direction, pipeline hydraulic connectivity, or zone drainage outlet relationships, retain the original zone boundary.

[0079] This implementation method transmits pipeline level labels to corresponding terrain grid cells and sets different levels of level matching penalties according to different pipeline levels. This results in strong constraints on the catchment areas corresponding to trunk pipelines, moderate constraints on the catchment areas corresponding to secondary trunk pipelines, and allows for local adjustment of the catchment areas corresponding to branch pipelines. At the same time, it combines hydraulic connectivity, topographic watersheds, area equilibrium, and historical flood risk to provide comprehensive rewards, thereby ensuring that the obtained catchment zoning results are consistent with pipeline levels, actual catchment relationships, and historical flood risk.

[0080] In a preferred embodiment, S5 includes the following steps: S51, using the water catchment area as a node and the hydraulic connectivity between water catchment areas from the upstream water catchment area to the downstream water catchment area according to the direction of water flow in the pipeline network as directed edges, construct a zoning topology graph; Based on the water catchment zoning results, determine the water catchment zoning to which each manhole node and rainwater outlet node in the pipeline network topology diagram belongs, and traverse the directed edges of the pipelines in the pipeline network topology diagram: When the upstream and downstream nodes of a directed edge of a pipeline belong to different catchment areas, the catchment area to which the upstream node belongs is determined as the upstream catchment area, and the catchment area to which the downstream node belongs is determined as the downstream catchment area. A directed edge is then established between the two catchment areas, pointing from the upstream catchment area to the downstream catchment area. When the upstream and downstream nodes of a directed edge of a pipeline belong to the same catchment area, the directed edge of the pipeline is regarded as the hydraulic connection relationship within the catchment area, and no cross-catchment directed edge is established. When multiple directed edges of pipelines correspond to the same upstream and downstream catchment areas, the cross-area hydraulic connectivity relationships corresponding to the multiple directed edges of pipelines are merged into a single zone directed edge.

[0081] Therefore, a regional topology graph is constructed by using the catchment areas as nodes and the cross-regional hydraulic connectivity relationships from the upstream catchment areas to the downstream catchment areas as directed edges.

[0082] S52, determine the node characteristics of each catchment area, the node characteristics include the catchment area, the number of main pipelines in the catchment area, the number of secondary pipelines in the catchment area, the upstream, midstream or downstream location level of the catchment area in the pipeline network and the historical waterlogging risk score. In this embodiment, the zoning location hierarchy is determined according to the main drainage direction of the target area: Based on the direction of the directed edges in the partition topology graph and the directed reachability between each catchment area and the stormwater outlet, the location level of each catchment area is determined as follows: the catchment area that does not receive water from other catchment areas and transports stormwater to downstream catchment areas through directed edges is determined as the upstream location level and assigned a location level value of 3; the catchment area that receives water from the upstream catchment area and transports stormwater to the downstream catchment area is determined as the midstream location level and assigned a location level value of 2; the catchment area that is directly connected to the stormwater outlet and receives water from other catchment areas is determined as the downstream location level and assigned a location level value of 1.

[0083] The node characteristics reflect the spatial scale of the catchment area, the functional composition of the pipeline network, the upstream and downstream locations, and the historical risk of waterlogging.

[0084] S53, input the node features of each watershed and the directed edge relationships of the partition topology graph into the graph attention network to obtain the feature representation of each watershed, and calculate the importance score of each watershed based on the feature representation of each watershed. The graph attention network uses neighboring watersheds connected to the current watershed via directed edges as neighborhood nodes. It assigns attention weights to the node features of the neighborhood nodes according to the connection direction of the directed edges, and then performs weighted aggregation of the node features of the neighborhood nodes according to the attention weights to obtain the feature representation of each watershed.

[0085] The feature representation includes the node features of the current catchment area itself and the node features of adjacent catchment areas that are hydraulically connected to the current catchment area. The feature representation of each catchment area is input into the importance scoring process, which maps the feature representation of each catchment area to a numerical value to obtain the importance score of each catchment area.

[0086] The importance scoring process is implemented using a fully connected scoring layer. As a preferred implementation, the catchment area, the number of main pipelines within the catchment area, the number of secondary pipelines within the catchment area, the location level value, and the historical flood risk score are all processed using Min-Max normalization and uniformly mapped to the [0,1] interval to obtain the normalized catchment area A and the normalized number of main pipelines. Normalized number of secondary trunk pipelines The system generates a normalized location hierarchy L and a normalized historical flood risk score R. The original location hierarchy values ​​are upstream = 3, midstream = 2, and downstream = 1, which are then normalized to the [0,1] interval. These normalized features are input into the fully connected scoring layer, which calculates the importance score for each catchment area using the following formula: R; in, ~ These are the weighting coefficients for each factor. In this embodiment... ~ All are set to 1. In practical applications, the weights of each factor can be adjusted according to the needs.

[0087] S54. Sort the catchment areas from high to low according to the importance score, configure online monitoring equipment for the top 30% of the catchment areas, configure liquid level monitoring equipment for the catchment areas after the top 30% but not exceeding the top 70%, and configure liquid level alarm devices for the manholes in the catchment areas after the top 70% that are located at historical water accumulation points or pipeline overflow risk points. The top 30% of catchment areas are designated as the first monitoring level, and equipped with online monitoring devices capable of continuously collecting water level, flow rate, and rainfall data. The remaining 30% but not exceeding the top 70% of catchment areas are designated as the second monitoring level, and equipped with level monitoring devices to collect level data, as well as rainfall data as needed. The remaining 70% of catchment areas are designated as the third monitoring level, and level alarm devices are only installed in manholes corresponding to historical water accumulation points or pipeline overflow risk points. The first, second, and third monitoring levels correspond to different types and quantities of monitoring equipment. The first monitoring level employs multi-parameter continuous monitoring, the second monitoring level employs low-power monitoring, and the third monitoring level only alarms for locations with historical water accumulation or pipeline overflow risk.

[0088] By configuring different monitoring devices according to the importance of the catchment area, the catchment areas with higher importance can obtain multiple types of operational data such as water level, flow rate and rainfall, the catchment areas with medium importance can obtain basic liquid level monitoring, and the catchment areas with low importance can only be equipped with alarm devices at known risk locations. This ensures the monitoring capability of key areas while reducing the duplication of equipment deployment in general areas.

[0089] S55. Within each catchment area, the intermediate centrality of each manhole in the catchment area is calculated based on the pipeline topology map. The manholes with the highest intermediate centrality are selected as monitoring candidate points. The monitoring candidate points are then weighted and sorted according to their distances from historical water accumulation points and pipeline overflow risk points to determine the monitoring points within the catchment area.

[0090] The intermediate centrality of each manhole is calculated based on the pipeline topology subgraph, and the top three manholes with intermediate centrality from high to low are selected as monitoring candidate points; when the number of manholes in the corresponding catchment area is less than three, all manholes in the catchment area are selected as monitoring candidate points.

[0091] When a catchment area contains multiple historical water accumulation points or multiple pipeline overflow risk points, the minimum distance between each candidate monitoring point and the corresponding risk point is taken as the distance between the candidate monitoring point and the risk point of that category. This distance is then normalized; the smaller the distance between the candidate monitoring point and the risk point, the higher the spatial correlation of the risk corresponding to that candidate monitoring point.

[0092] The centrality of each monitoring candidate point, its spatial correlation with historical water accumulation points, and its spatial correlation with pipeline overflow risk points are weighted according to preset weights to obtain a comprehensive ranking result for each monitoring candidate point. Based on the number of monitoring devices to be deployed according to the monitoring level of the corresponding catchment area, a corresponding number of manholes are selected as actual monitoring points from the top-ranked monitoring candidate points in the comprehensive ranking result, and the manhole identifier, spatial coordinates, monitoring device type, and deployment quantity of the actual monitoring points are output.

[0093] Among them, betweenness centrality reflects the transit role of manholes in the hydraulic path of the pipe network within the catchment area, while the distance between the manhole and historical waterlogging points or pipeline overflow risk points reflects the spatial coverage capacity of the manhole for existing waterlogging and overflow risks. By combining these two types of indicators, it is possible to avoid selecting manholes lacking historical risk coverage capacity based solely on pipe network topology centrality, and also to avoid selecting manholes without pipe network transmission function based solely on the distance to risk points.

[0094] In a preferred embodiment, step S6 includes the following steps: The real-time monitoring data of the target area and its collection time are associated with the corresponding graph entities in the dynamic knowledge graph, and the dynamic knowledge graph is updated. Incremental training will begin when any of the following triggering conditions are met: The first triggering condition is: the cumulative rainfall in any of the durations of 1 hour, 2 hours, 3 hours or 24 hours reaches the standard of once in ten years or above for that duration, and the same location experiences flooding twice in a row. The second triggering condition is: the total length of newly added or modified pipelines entering the warehouse exceeds 5% of the total length of the original pipelines in the target area; The incremental training is as follows: based on the original model parameters of the graph convolutional network, the deep reinforcement learning network, and the graph attention network, the updated data in the dynamic knowledge graph is used as the incremental training data, the learning rate is set to 1 / 10 of the initial learning rate, and the graph convolutional network, the deep reinforcement learning network, and the graph attention network are incrementally trained in sequence, and the updated pipeline hierarchical labels, water catchment zoning results, and monitoring point layout scheme are output.

[0095] The incremental training is performed based on the original model parameters of the graph convolutional network, deep reinforcement learning network, and graph attention network without re-initializing the model parameters. The incremental training learning rate for each network is 1 / 10 of its initial learning rate used during the initial training phase. This reduces the adjustment range of model parameters for new data and minimizes the possibility of unnecessary changes to the original pipeline classification, catchment area, and monitoring point layout results.

[0096] Incremental training is performed in the order of the graph convolutional network, deep reinforcement learning network, and graph attention network. The updated dynamic knowledge graph, containing newly added monitoring data or pipeline structure data, is input into the graph convolutional network to update pipeline level labels. Then, the updated pipeline level labels are input into the deep reinforcement learning network to update the state features and partitioning constraints of the terrain grid cells, outputting the updated catchment area partitioning results. Finally, the updated catchment area partitioning results and their partitioning attributes are input into the graph attention network to update the catchment area importance score and monitoring point layout scheme.

[0097] As a preferred example, the main pipeline stage has a nominal pipe diameter greater than or equal to 1000 mm or an upstream total catchment area greater than or equal to 3 km². 2 The pipeline is defined as follows: the secondary trunk level is a pipeline with a nominal diameter greater than or equal to 600 mm and less than 1000 mm, and the branch level is a pipeline with a nominal diameter less than 600 mm.

[0098] As another preferred embodiment, S3 further includes building a digital twin base; wherein, the digital twin base is used to uniformly manage the dynamic knowledge graph, the pipeline topology map, the hierarchical labels of the pipelines, the catchment area results, and the monitoring point layout scheme, and to access the real-time monitoring data; based on the real-time monitoring data, the dynamic knowledge graph, the pipeline topology map, the hierarchical labels of the pipelines, and the catchment area results, rainstorm simulation is performed, and graded urban flooding early warning results are output.

[0099] The digital twin platform includes a perception and transmission layer, a data middleware layer, a geographic visualization platform layer, an algorithm interaction service layer, and a front-end application layer.

[0100] The sensing and transmission layer is used to receive real-time rainfall, water level and flow data collected by external 4G / 5G IoT monitoring terminals, and transmit the real-time monitoring data and its collection time to the data platform layer.

[0101] The data platform layer stores standardized drainage data, real-time monitoring data, dynamic knowledge graphs, pipeline topology diagrams, pipeline hierarchical labels, catchment area results, and monitoring point deployment schemes, and provides a bidirectional API interface. This bidirectional API interface provides data read / write support for the dynamic knowledge graph, the pipeline topology diagram, and graph convolutional networks, deep reinforcement learning networks, and graph attention networks. The data platform layer is responsible for the unified management of standardized drainage data, real-time monitoring data, and computational results. The dynamic knowledge graph, the pipeline topology diagram, the graph convolutional networks, the deep reinforcement learning networks, and the graph attention networks do not have independent databases; all read / write operations are completed by calling the open API of the digital twin platform.

[0102] The geographic visualization base layer is used to load rainwater pipe network, water system, terrain elevation, pipe hierarchy, catchment area and monitoring point layers. Each layer is displayed according to a unified spatial coordinate to show the spatial relationship between pipe hierarchy, catchment area and monitoring point, and supports the import and export of vector data.

[0103] The algorithm interaction service layer is used to schedule the computational tasks and incremental training tasks of the graph convolutional network, the deep reinforcement learning network, and the graph attention network, and to transmit pipeline level labels, water catchment zoning results, and monitoring point deployment schemes between the graph convolutional network, the deep reinforcement learning network, and the graph attention network.

[0104] The front-end application layer is used to visualize the results and export the output.

[0105] During rainstorm simulation, the algorithm interaction service layer reads real-time rainfall, water level, and flow rate data, pipeline hydraulic attributes, pipeline topology, terrain elevation data, pipeline hierarchical labels, and catchment area results from the data platform layer. This data is then input into the rainstorm simulation module to calculate the water accumulation status of each terrain grid unit, the water level or flow rate status of each pipeline, and the flooding status of each catchment area under rainfall conditions. The water accumulation status and pipeline water level or flow rate status are compared with preset flooding judgment thresholds, and a graded flooding warning result is determined based on the corresponding pipeline hierarchical labels and the importance of the catchment area. The graded flooding warning result is displayed through the front-end application layer and sent to the corresponding management terminal via a warning information push interface.

[0106] When the dynamic knowledge graph, the pipeline topology map, the pipeline hierarchical labels, or the catchment area results are updated, the algorithm interaction service layer reads the updated data and calls the rainstorm simulation module to recalculate; when the monitoring point layout scheme is updated, the monitoring point layer in the geographic visualization base layer and the monitoring point layout scheme in the data platform layer are updated synchronously to ensure that the simulation results and early warning results are consistent with the updated pipeline structure and catchment area.

[0107] The present invention also provides an automatic monitoring point deployment system that couples pipeline network classification and catchment area delineation, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the automatic monitoring point deployment method that couples pipeline network classification and catchment area delineation as described above.

[0108] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. An automatic deployment method for monitoring points coupled with pipeline network classification and catchment area delineation, characterized in that, Includes the following steps: S1. Acquire multi-source drainage data of the target area and preprocess it to obtain a standardized dataset; S2, construct a dynamic knowledge graph based on the standardized dataset, and update the dynamic knowledge graph incrementally as new pipeline facilities or monitoring points are added; S3. Construct a pipeline network topology map based on the dynamic knowledge graph, process the pipeline network topology map using a graph convolutional network, and output the hierarchical labels of each pipeline. The hierarchical labels include at least the trunk level, the secondary trunk level, and the branch level. S4, using the hierarchical labels as state constraints of the deep reinforcement learning network, the deep reinforcement learning network is used to divide the catchment area, wherein the zoning constraint for the main pipeline level is stronger than that for the secondary pipeline level, and the secondary pipeline level is stronger than that for the branch pipeline level; output the catchment area zoning result. S5, taking the water catchment area as a node, calculate the importance of each water catchment area using a graph attention network based on the area attribute information, and configure monitoring equipment and determine monitoring points according to the importance difference; S6, access real-time monitoring data to update the dynamic knowledge graph. When the preset triggering conditions are met, incremental adjustments are made to the graph convolutional network, deep reinforcement learning network and graph attention network to update the pipeline's hierarchical labels, water catchment zoning results and monitoring point layout scheme.

2. The automatic deployment method for monitoring points in the coupled pipe network classification and catchment area delineation according to claim 1, characterized in that, The multi-source drainage data includes stormwater pipe network data, water system data, topographic elevation data, and historical hydrological monitoring data; the preprocessing includes the following steps: The spatial data in the rainwater pipe network data, the water system data, the topographic elevation data, and the historical hydrological monitoring data are uniformly converted to the same plane rectangular projection coordinate system; Correct at least one type of abnormal and invalid data, such as pipeline breakpoints, reverse slopes, elevation jumps, and overlapping polygons, using GIS topology inspection tools; A unified field specification is established, and the multi-source drainage data with unified coordinates and anomaly correction is stored in a spatial database to obtain the standardized dataset.

3. The automatic deployment method for monitoring points in the coupled pipe network classification and catchment area delineation according to claim 1, characterized in that, S2 includes the following steps: S21, Based on the standardized dataset, construct the graph entities and semantic relationships between them; wherein, the graph entities include pipes, manholes, storm drains, storm drain outlets, water storage facilities, river systems, terrain grid units, and monitoring points; the semantic relationships include the upstream and downstream hydraulic connectivity of pipes, the receiving relationship between storm drain outlets and river systems, the spatial dependency relationship between drainage facilities and terrain grid units, and the monitoring correspondence between monitoring points and manholes or pipes; predefine the subordinate relationship type between catchment areas and pipes, and establish the corresponding subordinate relationship after the catchment areas are divided; S22, the rainfall data is associated with the corresponding rainfall monitoring points, and the rainfall data is associated with the corresponding terrain grid units according to the spatial correspondence between the rainfall monitoring points and the terrain grid units; the water level data and flow rate data are associated with the manholes or pipes where the monitoring points are installed; the water accumulation data is associated with the terrain grid units or manholes according to the water accumulation location; the collection time of each data is written into the temporal attribute of the graph entity associated with each data, thus obtaining the dynamic knowledge graph; S23. When adding or modifying pipeline facilities, update the corresponding graph entities and affected semantic relationships based on the identifier, spatial location, and pipeline connection relationship of the added or modified pipeline facilities. When adding monitoring points, update the corresponding graph entities and affected semantic relationships based on the identifier, installation location, and monitoring object of the added monitoring points. Retain the unchanged graph entities and semantic relationships to obtain the updated dynamic knowledge graph.

4. The automatic deployment method for monitoring points in the coupled pipe network classification and catchment area delineation according to claim 3, characterized in that, S3 includes the following steps: S31, map the manholes and rainwater outlets in the dynamic knowledge graph as nodes in the pipeline topology graph, and map the pipelines as directed edges connecting the nodes along the water flow direction; S32, determine the water conveyance capacity characteristics of each pipeline based on its nominal pipe diameter and laying slope, determine the water system modulation characteristics based on the cumulative path length of the pipeline along the water flow direction to its final downstream rainwater outlet, and normalize and weight the water conveyance capacity characteristics and the water system modulation characteristics to obtain the edge weight of each directed edge. S33, calculate the betweenness centrality, downstream cumulative pipe length and upstream total catchment area of ​​each node according to the pipeline network topology diagram to form node feature vectors; input the node feature vectors and the directed edge relationships of the pipeline network topology diagram into a graph convolutional network to obtain the feature representation of each node; S34, combine the feature representations of the upstream and downstream nodes of each pipeline to form pipeline features, and determine the hierarchical label of each pipeline based on the pipeline features; the hierarchical label includes trunk level, secondary trunk level and branch level.

5. The automatic deployment method for monitoring points in the coupled pipe network classification and catchment area delineation according to claim 4, characterized in that, S4 includes the following steps: S41, the target area is discretized into multiple terrain grid units. Based on the surface flow direction and pipeline spatial location of each terrain grid unit, the target pipeline receiving the surface runoff of that terrain grid unit is determined, and the hierarchical label of the target pipeline is used as the grid hierarchical label of that terrain grid unit. The state characteristics of each terrain grid unit are determined, including terrain elevation, surface slope, distance to the nearest main pipeline, distance to the nearest secondary pipeline, grid hierarchical label, current catchment area number, and historical waterlogging risk score. S42, using a deep reinforcement learning network as the decision model, the deep reinforcement learning network takes the state characteristics of each terrain grid unit as input and outputs the division action corresponding to each terrain grid unit. The division action is to divide the terrain grid unit into an existing spatially adjacent catchment area or to create a new catchment area with the terrain grid unit as a seed. S43, During training, when a terrain grid cell is assigned to a catchment area that does not contain its target pipeline, a hierarchical matching penalty is applied; wherein, when the hierarchical labels of the target pipeline are main trunk level, secondary trunk level, and branch level, the hierarchical matching penalty corresponds to a first negative reward, a second negative reward, and a third negative reward, respectively, and the absolute value of the first negative reward is greater than the absolute value of the second negative reward, and the absolute value of the second negative reward is greater than the absolute value of the third negative reward; S44, the reward function of the deep reinforcement learning network is obtained by weighted summation of the hierarchical matching penalty, hydraulic connectivity reward, terrain consistency reward, area balance reward and historical waterlogging risk gain according to their respective weights; S45. After training, the water catchment zoning scheme is obtained according to the division action of each terrain grid unit output by the deep reinforcement learning network. Water catchment zoning with an area smaller than the preset minimum area threshold is merged into water catchment zoning that is spatially adjacent and hydraulically connected to it. The boundary of the merged zoning is smoothed and the boundary of the zoning is aligned with the boundary of the road or river to obtain the water catchment zoning result.

6. The automatic deployment method for monitoring points in the coupled pipe network classification and catchment area delineation according to claim 5, characterized in that, S5 includes the following steps: S51, using the water catchment area as a node and the hydraulic connectivity between water catchment areas from the upstream water catchment area to the downstream water catchment area according to the direction of water flow in the pipeline network as directed edges, construct a zoning topology graph; S52, determine the node characteristics of each catchment area, the node characteristics include the catchment area, the number of main pipelines in the catchment area, the number of secondary pipelines in the catchment area, the upstream, midstream or downstream location level of the catchment area in the pipeline network and the historical waterlogging risk score; S53, input the node features of each watershed and the directed edge relationships of the partition topology graph into the graph attention network to obtain the feature representation of each watershed, and calculate the importance score of each watershed based on the feature representation of each watershed. S54. Sort the catchment areas from high to low according to the importance score, configure online monitoring equipment for the top 30% of the catchment areas, configure liquid level monitoring equipment for the catchment areas after the top 30% but not exceeding the top 70%, and configure liquid level alarm devices for the manholes in the catchment areas after the top 70% that are located at historical water accumulation points or pipeline overflow risk points. S55. Within each catchment area, the intermediate centrality of each manhole in the catchment area is calculated based on the pipeline topology map. The manholes with the highest intermediate centrality are selected as monitoring candidate points. The monitoring candidate points are then weighted and sorted according to their distances from historical water accumulation points and pipeline overflow risk points to determine the monitoring points within the catchment area.

7. The automatic deployment method for monitoring points in the coupled pipe network classification and catchment area delineation according to claim 3, characterized in that, S6 includes the following steps: The real-time monitoring data of the target area and its collection time are associated with the corresponding graph entities in the dynamic knowledge graph, and the dynamic knowledge graph is updated. Incremental training will begin when any of the following triggering conditions are met: The first triggering condition is: the cumulative rainfall in any of the durations of 1 hour, 2 hours, 3 hours or 24 hours reaches the standard of once in ten years or above for that duration, and the same location experiences flooding twice in a row. The second triggering condition is: the total length of newly added or modified pipelines entering the warehouse exceeds 5% of the total length of the original pipelines in the target area; The incremental training is as follows: based on the original model parameters of the graph convolutional network, the deep reinforcement learning network, and the graph attention network, the updated data in the dynamic knowledge graph is used as the incremental training data, the learning rate is set to 1 / 10 of the initial learning rate, and the graph convolutional network, the deep reinforcement learning network, and the graph attention network are incrementally trained in sequence, and the updated pipeline hierarchical labels, water catchment zoning results, and monitoring point layout scheme are output.

8. The automatic deployment method for monitoring points in the coupled pipe network classification and catchment area delineation according to claim 1, characterized in that, The main pipeline is defined as having a nominal pipe diameter greater than or equal to 1000 mm or an upstream total catchment area greater than or equal to 3 km². 2 The pipeline is defined as follows: the secondary trunk level is a pipeline with a nominal diameter greater than or equal to 600 mm and less than 1000 mm, and the branch level is a pipeline with a nominal diameter less than 600 mm.

9. The automatic deployment method for monitoring points in the coupled pipe network classification and catchment area delineation according to claim 1, characterized in that, The S3 further includes building a digital twin base; wherein, the digital twin base is used to uniformly manage the dynamic knowledge graph, the pipeline topology map, the hierarchical labels of the pipelines, the catchment area results, and the monitoring point layout scheme, and to access the real-time monitoring data; based on the real-time monitoring data, the dynamic knowledge graph, the pipeline topology map, the hierarchical labels of the pipelines, and the catchment area results, rainstorm simulation is performed, and graded urban flooding early warning results are output.

10. An automatic monitoring point deployment system that couples pipe network classification and catchment area delineation, characterized in that, The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the automatic deployment method for monitoring points of the coupled pipe network classification and catchment area delineation as described in any one of claims 1 to 9.

Images