Regional pipe network safety assessment and early warning system based on big data and edge computing

Abstract

The application discloses a regional pipe network safety evaluation and early warning system based on big data and edge computing, comprising: a four-layer perception synchronization module for building a four-layer architecture and generating synchronous bimodal time series data; a reachable interval generation module for generating pipe network event reachable intervals; a segmentation detection module for performing pipe network constraint wild binary segmentation on the synchronous bimodal time series data to obtain a bimodal variable point feature set; a homologous matching segmentation module for synthesizing a pipe network abnormal event segmentation set; a risk label upload module for generating abnormal segmentation upload data packets; a cloud calibration evaluation module for hydraulic mechanism verification and data-driven identification to generate a pipe network safety evaluation result; and an early warning work order feedback module for generating segmentation strategy correction information and issuing the information to the edge layer to update the next round of segmentation strategy. The application realizes segmented identification and cloud calibration evaluation of pipe network safety risks.

Description

Technical Field

[0001] This invention relates to the field of pipeline safety monitoring technology, and in particular to a regional pipeline safety assessment and early warning system based on big data and edge computing. Background Technology

[0002] With the continuous expansion of urban water supply, drainage, and industrial pipeline networks, the operational status of regional pipeline networks directly affects water supply safety, drainage safety, the stability of urban infrastructure, and the safety of on-site maintenance personnel. Existing pipeline monitoring technologies typically collect pipeline operation data through pressure sensors, flow meters, level gauges, or water quality testing equipment, and upload the collected data to a monitoring platform. The platform then assesses risks such as pipeline leaks, pipe bursts, water hammer, corrosion, and environmental anomalies based on preset thresholds, historical experience, or single anomaly identification models, thereby achieving a certain level of pipeline operation monitoring and fault alarm.

[0003] However, most existing technologies focus on the collection and analysis of single-type data, such as issuing alarms based solely on pressure changes, flow fluctuations, or water quality anomalies. They lack high-frequency synchronous acquisition and joint analysis of pressure time-series data and acoustic time-series data, making it difficult to accurately identify the correlation between leakage jet acoustic waves, negative pressure waves, water hammer disturbances, and transient pressure changes during pipe rupture. While some systems can collect multi-source data, the lack of unified time base alignment and pipeline topology constraints among these data sources makes it difficult to effectively utilize the propagation time relationships of abnormal signals between different monitoring nodes, easily leading to false alarms, missed alarms, or inaccurate anomaly location.

[0004] Meanwhile, existing anomaly detection methods are typically based on fixed thresholds, global statistical analysis, or centralized cloud processing. They lack constraints that consider pipe segment connections, pipe length, pipe diameter, pipe material, valve status, and the reachability range of pressure waves and sound waves. This makes it difficult to quickly screen and segment effective anomaly fragments at the edge. Uploading large amounts of raw data directly to the cloud also increases communication bandwidth and cloud computing burden, affecting the real-time performance of early warnings.

[0005] Therefore, how to provide a regional pipeline network safety assessment and early warning system based on big data and edge computing is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a regional pipeline network safety assessment and early warning system based on big data and edge computing. This invention achieves multi-source data acquisition, real-time edge segmentation and detection, and cloud-based collaborative calibration and assessment through a four-layer architecture. This can improve the real-time performance, accuracy, and reliability of pipeline network abnormal event identification, reduce invalid data uploads and false alarms and missed alarms, and update edge detection parameters through early warning work orders to achieve closed-loop early warning and handling of regional pipeline network safety risks.

[0007] According to an embodiment of the present invention, a regional pipeline network safety assessment and early warning system based on big data and edge computing includes: The four-layer perception synchronization module is used to construct a four-layer architecture including a perception layer, an edge layer, a cloud big data layer, and an application layer. The perception layer collects pressure time-series data and acoustic time-series data, and performs time synchronization through the edge layer to generate synchronous dual-modal time-series data. The reachability interval generation module is used to generate reachability intervals for pipeline events based on pipeline propagation constraint parameters. The segmentation and detection module is used to generate random detection intervals within the reachable interval of pipeline events, and to perform pipeline-constrained wild binary segmentation on synchronous dual-modal time-series data to obtain a dual-modal change point feature set; The homogeneous matching segmentation module is used to perform asynchronous propagation time matching on the dual-modal variable point feature set according to the pipeline propagation constraint parameters, to obtain pressure-acoustic homogeneous variable point pairs, and to synthesize pipeline abnormal event segmentation sets based on the pressure-acoustic homogeneous variable point pairs; The risk labeling and uploading module is used to extract the risk characteristics of abnormal segments based on the segment set of abnormal events in the pipeline network, generate early warning value, perform segmented handling and labeling, and generate abnormal segment uploading data packets. The cloud calibration and evaluation module is used to perform hydraulic mechanism verification and data-driven identification based on the data packets uploaded in abnormal segments and the cloud verification and evaluation parameters. It also performs collaborative calibration on the mechanism verification results and data identification results to generate pipeline safety assessment results. The early warning work order feedback module is used to generate early warning information and operation and maintenance work orders based on the pipeline safety assessment results. The cloud big data layer generates segmentation strategy correction information based on the early warning information and operation and maintenance work orders, and sends it to the edge layer to update the next round of segmentation strategy.

[0008] Optionally, the four-layer sensing synchronization module includes: Sensing layer acquisition devices are deployed at regional pipeline network monitoring nodes, and a four-layer architecture consisting of sensing layer, edge layer, cloud big data layer and application layer is established, while data transmission channels between layers are also established. The sensing layer collects pressure timing data and acoustic timing data through sensing layer acquisition devices, and also collects pipeline network operating environment data, and transmits them together to the edge layer. After receiving force timing data and acoustic timing data at the edge layer, the pressure timing data and acoustic timing data are processed by unified time base marking, sampling time correction and timing alignment according to satellite timing information to generate synchronous dual-mode timing data.

[0009] Optionally, the reachability interval generation module includes: The edge layer reads the monitoring node identifier, sampling time range, and data acquisition status corresponding to the synchronous dual-modal time series data, and calls the pipeline propagation constraint parameters corresponding to the monitoring node identifier; The edge layer filters the pipeline connectivity between monitoring nodes based on the pipe segment connection relationship and valve status in the pipeline propagation constraint parameters. Pipe segments that are in a disconnected state, closed state, or do not meet the propagation path conditions are removed, while associated monitoring nodes that meet the propagation conditions of abnormal events are retained. The effective propagation path is determined based on the length, diameter, and material of the pipe segments traversed between the associated monitoring nodes. The edge layer determines the reachable time range of pressure events and the reachable time range of acoustic events between associated monitoring nodes based on the effective propagation path, pipeline propagation constraint parameters, and data acquisition status, and merges them to generate the reachable interval of pipeline events.

[0010] Optionally, the segmentation detection module includes: The edge layer generates a random mother interval within the reachable interval of pipeline events, and maps the random mother interval to the reachable sub-intervals of pressure events and acoustic events respectively, and combines them to obtain the accompanying random detection interval; The accompanying random detection interval is mapped to synchronous dual-modal time series data. Pressure detection sequences are extracted from pressure event reachable sub-intervals, and acoustic detection sequences are extracted from acoustic event reachable sub-intervals. A set of candidate segmentation times from the same source is constructed based on pipeline propagation constraint parameters and synchronization state parameters. Pipeline-constrained wild binary segmentation is performed on the pressure detection sequence and acoustic detection sequence. Within each associated random detection interval, the pressure detection sequence and acoustic detection sequence are divided into front and back segments according to the set of candidate segmentation times from the same source. Pressure mutation features and acoustic mutation features are calculated respectively. Based on the pressure mutation features, acoustic mutation features and the corresponding conditions of dual-modal reachable time, a dual-modal associated segmentation score is generated. Candidate time combinations that satisfy the change point retention conditions in the dual-modal associated segmentation score are selected as associated candidate change points. Using the candidate change point as the dividing boundary, the random detection interval is recursively divided into binary segments to generate the next round of random detection sub-intervals. The pipeline-constrained wild binary segmentation process is repeated in the next round of random detection sub-intervals until the recursive stopping condition is met. The candidate variable points obtained from each round of recursive partitioning are aggregated and screened to remove candidate variable points that do not meet the conditions of bimodal reachability time correspondence, random interval repeated verification, and bimodal accompanying change, thus obtaining the bimodal variable point feature set.

[0011] Optionally, the same-source matching segmentation module includes: Call the pipeline propagation constraint parameters corresponding to the dual-modal variable point feature set, perform negative pressure wave time difference verification based on pressure variable point characteristics and pressure wave propagation path, and obtain pressure side time difference matching results; Acoustic spectrum analysis is performed based on the acoustic variable point features and sound wave propagation path in the dual-modal variable point feature set to obtain acoustic side feature matching results; By combining the pressure-side time difference matching results, the acoustic-side feature matching results, and the dual-modal reachability time correspondence, asynchronous propagation time matching is performed to obtain pressure-acoustic homogeneous variable point pairs. Dual-modal fusion positioning preprocessing is performed using pressure-acoustic homogeneous variable point pairs, and the abnormal event segment boundaries are determined based on the fusion positioning preprocessing results to synthesize a pipeline abnormal event segment set.

[0012] Optionally, the risk marker uploading module includes: The edge layer reads the pipeline abnormal event segment set, locates the segment start and end time corresponding to each abnormal event segment according to the abnormal event segment identifier, and cuts back the pressure segment data and acoustic segment data from the synchronous dual-modal time series data according to the segment start and end time to form an abnormal event segment record; The pipeline network operation environment data is aligned to the abnormal event segment records according to the associated monitoring nodes, effective propagation paths and segment start and end times, and data that does not fall within the segment start and end times are removed to obtain segment-related operation data; Based on the pressure segment data and acoustic segment data, the abnormal changes within the segment are compared, and the pressure-acoustic co-source change point is combined to verify whether the pressure abnormality and acoustic abnormality correspond to the same pipeline network abnormal event. At the same time, the verification results are correlated with the segment-related operating environment data to obtain the risk characteristics of the abnormal segment. The risk of each abnormal event segment is ranked according to its risk characteristics, and a warning value is assigned based on the risk ranking results. Based on the early warning value, the abnormal events in the pipeline network are segmented and marked for segmented handling. The segmented handling and marking process includes removing markers, retaining markers, uploading markers, and cloud verification markers. Abnormal events with upload and cloud review tags are segmented and encapsulated into abnormal segment upload data packets, while abnormal event segments with removal and retention tags are retained in the local records of the edge layer.

[0013] Optionally, the cloud-based calibration and evaluation module includes: The cloud-based big data layer receives and parses abnormal segmented data packets, and then cleans, deduplicates, standardizes, and spatiotemporally aligns the parsed data before writing it into the pipeline holographic database. The cloud-based verification and evaluation parameters are invoked to map the abnormal segment association information to the pipeline network topology, and the hydraulic mechanism verification is performed on the target pipe segment, associated monitoring node and effective propagation path corresponding to the abnormal event segment to obtain the mechanism verification results. Initiate data-driven identification by inputting the abnormal segmented data packets into the cloud AI algorithm library to identify abnormal events such as leaks, pipe bursts, water hammer, and abnormal operating conditions, and obtain data identification results. The consistency of the mechanism verification results and the data identification results is compared. When they are consistent, the abnormal event is confirmed. When they are inconsistent, the hydraulic mechanism verification results are used to constrain the data identification results. The data identification results are then used to reversely correct the local parameters in the hydraulic mechanism verification, and a collaborative calibration result is generated. Based on the collaborative calibration results, the target assessment area and the target abnormal pipe section are determined. The target assessment area is divided into assessment units according to pipe section, node and equipment. Structural safety assessment, operational safety assessment, environmental safety assessment and equipment health assessment are performed on each assessment unit. The assessment results are weighted and summarized and then input into the fuzzy comprehensive evaluation process to generate the pipeline safety assessment results.

[0014] Optionally, the early warning work order feedback module includes: The application layer receives the pipeline security assessment results and calls the hierarchical early warning engine to generate early warning information; The warning information is pushed to the BI dashboard, the warning handling module, and the mobile terminal APP, and maintenance work orders are generated based on the warning information; The cloud-based big data layer receives early warning information and maintenance work orders, associates them with the corresponding network anomaly event segments, generates segmentation strategy correction information, and sends the segmentation strategy correction information to the edge layer to update the next round of segmentation strategy.

[0015] The beneficial effects of this invention are: This invention constructs a four-layer collaborative architecture consisting of a perception layer, an edge layer, a cloud big data layer, and an application layer. This architecture enables high-frequency synchronous acquisition and unified time-base processing of pressure time-series data and acoustic time-series data. It can fully utilize the correlation between pressure fluctuation information and acoustic information, improve the real-time performance and accuracy of abnormal event identification, and reduce false alarms and missed alarms caused by analysis of a single data source.

[0016] This invention achieves accurate segmentation and identification of abnormal events by constructing reachable intervals for pipeline events and performing pipeline-constrained wild binary segmentation within these intervals. By combining pipeline topology, propagation paths, and bimodal variable point information for source matching, it can effectively identify pressure and acoustic anomalies caused by the same abnormal event, improving the reliability of abnormal event location and providing an accurate data foundation for subsequent leak location, pipe burst analysis, and risk assessment.

[0017] This invention performs risk ranking, early warning value calculation, and segmented handling marking on abnormal events at the edge. Only abnormal events with high analytical value are uploaded to the cloud, thereby reducing data transmission volume and cloud computing pressure, and improving the overall system operating efficiency and early warning response speed.

[0018] This invention combines physical mechanism constraints with intelligent data recognition through a collaborative calibration mechanism of hydraulic mechanism model and data-driven model. This enables mutual verification and dynamic correction of mechanism verification results and data recognition results, improving the credibility and stability of anomaly identification results under complex working conditions and enhancing the accuracy of regional pipeline network safety assessment results.

[0019] This invention combines pipeline safety index, risk level, and weak links to generate graded early warning information and operation and maintenance work orders. It also uses the feedback results of the work orders to continuously correct the edge anomaly segment detection parameters, forming a closed-loop operation mechanism from anomaly detection, risk assessment, early warning and handling to parameter optimization, thereby improving the intelligent level of regional pipeline safety management and predictive operation and maintenance capabilities. Attached Figure Description

[0020] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of the regional pipeline network safety assessment and early warning system based on big data and edge computing proposed in this invention; Figure 2 This is a schematic diagram of the pipeline constraint segmentation and edge-cloud feedback of the regional pipeline safety assessment and early warning system based on big data and edge computing proposed in this invention. Detailed Implementation

[0021] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0022] refer to Figures 1-2 A regional pipeline network safety assessment and early warning system based on big data and edge computing, including: The four-layer perception synchronization module is used to construct a four-layer architecture including a perception layer, an edge layer, a cloud big data layer, and an application layer. The perception layer collects pressure time-series data and acoustic time-series data, and performs time synchronization through the edge layer to generate synchronous dual-modal time-series data. The reachability interval generation module is used to generate reachability intervals for pipeline events based on pipeline propagation constraint parameters. The segmentation and detection module is used to generate random detection intervals within the reachable interval of pipeline events, and to perform pipeline-constrained wild binary segmentation on synchronous dual-modal time-series data to obtain a dual-modal change point feature set; The homogeneous matching segmentation module is used to perform asynchronous propagation time matching on the dual-modal variable point feature set according to the pipeline propagation constraint parameters, to obtain pressure-acoustic homogeneous variable point pairs, and to synthesize pipeline abnormal event segmentation sets based on the pressure-acoustic homogeneous variable point pairs; The risk labeling and uploading module is used to extract the risk characteristics of abnormal segments based on the segment set of abnormal events in the pipeline network, generate early warning value, perform segmented handling and labeling, and generate abnormal segment uploading data packets. The cloud calibration and evaluation module is used to perform hydraulic mechanism verification and data-driven identification based on the data packets uploaded in abnormal segments and the cloud verification and evaluation parameters. It also performs collaborative calibration on the mechanism verification results and data identification results to generate pipeline safety assessment results. The early warning work order feedback module is used to generate early warning information and operation and maintenance work orders based on the pipeline safety assessment results. The cloud big data layer generates segmentation strategy correction information based on the early warning information and operation and maintenance work orders, and sends it to the edge layer to update the next round of segmentation strategy.

[0023] In this embodiment, the four-layer sensing synchronization module includes: Sensing layer acquisition devices are deployed at regional pipeline network monitoring nodes, and a four-layer architecture consisting of sensing layer, edge layer, cloud big data layer and application layer is established. At the same time, data transmission channels between layers are established. Regional pipeline network monitoring nodes include valve wells, pipe section connection points, pressure zone boundaries and historical leakage locations. Sensing layer acquisition devices include high-frequency pressure sensors, high-sensitivity hydrophone sensors and extended sensors. The sensing layer collects pressure and acoustic time-series data through sensing layer acquisition devices, and simultaneously collects pipeline network operating environment data, all of which are transmitted to the edge layer. The pressure time-series data includes pipeline transient pressure fluctuation data, negative pressure wave data, water hammer data, and pipe burst pressure disturbance data. The acoustic time-series data includes leakage jet acoustic wave data, pipe wall vibration data, and acoustic energy change data. The high-frequency pressure sensor is made of stainless steel and has an IP68 protection rating. It collects pipeline transient pressure fluctuation data, negative pressure wave data, water hammer data, and pipe burst pressure disturbance data at a sampling frequency of 1kHz and is used to sense -0.1MPa negative pressure. The high-sensitivity hydrophone sensor is made of stainless steel and has an IP68 protection rating. It collects leakage jet acoustic wave data, pipe wall vibration data, and acoustic energy change data at a sampling frequency of 8kHz and a bandwidth of 10kHz to 20kHz. Pipeline network operating environment data is collected through extended sensors, including flow data, water quality data, liquid level data, and harmful gas data. After receiving pressure and acoustic time-series data at the edge layer, the pressure and acoustic time-series data are processed with unified time base marking, sampling time correction, and time alignment based on satellite timing information to generate synchronous dual-mode time-series data. Satellite timing information includes BeiDou or GPS timing information. Unified time base marking is used to mark the pressure and acoustic time-series data to the same timing reference. Sampling time correction is used to correct the sampling time deviation between the pressure and acoustic time-series data. Time alignment processing is used to correspond the pressure and acoustic time-series data according to the same monitoring node and the same timing reference. Synchronous dual-mode time-series data includes pressure and acoustic time-series data that have been marked with unified time reference.

[0024] In this embodiment, the reachable interval generation module includes: The edge layer reads the monitoring node identifier, sampling time range, and data acquisition status corresponding to the synchronous dual-modal time series data, and calls the pipeline propagation constraint parameters corresponding to the monitoring node identifier. The monitoring node identifier includes the monitoring node number, sensor device number, monitoring node installation location, and pipe segment information. The data acquisition status includes the pressure time series data acquisition status, acoustic time series data acquisition status, and satellite time synchronization status. The pipeline propagation constraint parameters include the monitoring node number, monitoring node location, pipe segment connection relationship, pipe segment length, pipe diameter, pipe material, valve status, pressure wave propagation speed range, and acoustic propagation speed range. The edge layer filters the pipeline connectivity between monitoring nodes based on the pipe segment connection relationship and valve status in the pipeline propagation constraint parameters. Pipe segments that are in an open, closed, or do not meet the propagation path conditions are removed, while associated monitoring nodes that meet the propagation conditions of abnormal events are retained. The effective propagation path is determined based on the pipe segment length, pipe diameter, and pipe material between the associated monitoring nodes. Associated monitoring nodes are two or more monitoring nodes that can receive pressure disturbance signals or acoustic disturbance signals of abnormal events in the same pipeline. The effective propagation path is the pipe segment path that the abnormal event can reach between the associated monitoring nodes along the pipeline topology. The edge layer determines the reachable time ranges of pressure events and acoustic events between associated monitoring nodes based on the effective propagation path, pipeline propagation constraint parameters, and data acquisition status. These ranges are then merged to generate the reachable interval of pipeline events. Both the reachable time ranges of pressure events and acoustic events are determined through propagation constraint matching, time base correction, and interval pruning. Propagation constraint matching determines the candidate arrival time periods of the corresponding disturbance signals based on the effective propagation path and pipeline propagation constraint parameters. Time base correction corrects the time deviation of the candidate arrival time periods based on the synchronization status parameters. Interval pruning limits the candidate arrival time periods to the sampling time range. The reachable interval of pipeline events is the candidate time interval of abnormal events that meets the pipeline topology connectivity, pipe segment propagation conditions, pressure wave propagation conditions, and acoustic propagation conditions within the sampling time range of the synchronous dual-modal time series data.

[0025] In this embodiment, the segmentation detection module includes: The edge layer generates a random mother interval within the reachable interval of pipeline events, and maps the random mother interval to the reachable sub-intervals of pressure events and acoustic events respectively, and combines them to obtain the accompanying random detection interval, which includes the reachable sub-intervals of pressure events and acoustic events. The accompanying random detection interval is mapped to synchronous dual-modal time series data. Pressure detection sequences are extracted from pressure event reachable sub-intervals, and acoustic detection sequences are extracted from acoustic event reachable sub-intervals. A set of candidate segmentation times from the same source is constructed based on pipeline propagation constraint parameters and synchronization state parameters. The set of candidate segmentation times from the same source includes candidate time combinations that satisfy the dual-modal reachable time correspondence conditions between pressure candidate segmentation times and acoustic candidate segmentation times. Pipeline-constrained wild binary segmentation is performed on the pressure detection sequence and acoustic detection sequence. Within each accompanying random detection interval, the pressure detection sequence and acoustic detection sequence are divided into front and back segments according to the set of candidate segmentation times from the same source. Pressure mutation features and acoustic mutation features are calculated separately. Based on the pressure mutation features, acoustic mutation features and the dual-modal reachability time correspondence condition, a dual-modal accompanying segmentation score is generated. Candidate time combinations that satisfy the change point retention condition of the dual-modal accompanying segmentation score are selected as accompanying candidate change points. The dual-modal reachability time is the time correspondence between the reachability time range of the pressure event and the reachability time range of the acoustic event. Using the candidate change points as the segmentation boundary, the random detection interval of the companion is recursively divided into binary segments to generate the next round of random detection intervals of the companion. The pipeline-constrained wild binary segmentation process is repeated in the next round of random detection intervals of the companion until the recursive stopping condition is met. The recursive stopping condition includes the length of the random detection interval of the companion being lower than the minimum detection length, the bimodal companion segmentation score being lower than the change point retention condition, or the number of recursive segments reaching the preset number. The candidate variable points obtained from each round of recursive partitioning are aggregated and screened to remove those that do not meet the conditions of bimodal reachability time correspondence, random interval repeated verification, and bimodal accompanying change. This results in a bimodal variable point feature set. The bimodal reachability time correspondence condition means that the pressure candidate variable point is located within the reachable sub-interval of the pressure event, and the acoustic candidate variable point is located within the reachable sub-interval of the acoustic event. The random interval repeated verification condition means that the same candidate variable point is repeatedly detected in multiple accompanying random detection intervals or multiple rounds of recursive partitioning. The bimodal accompanying change condition means that there is a pressure mutation feature at the pressure candidate variable point, and there is an acoustic mutation feature at the acoustic candidate variable point, and the pressure mutation feature and the acoustic mutation feature both point to the same pipeline network abnormal event. The bimodal variable point feature set includes pressure variable point features, acoustic variable point features, accompanying random detection interval identifiers, effective propagation path identifiers, and bimodal reachability time correspondence.

[0026] This invention limits the random detection interval to the reachable interval of pipeline events by segmentation detection module, and performs adjoint mapping between the reachable sub-interval of pressure events and the reachable sub-interval of acoustic events. This makes the wild binary segmentation process subject to both pipeline propagation constraints and dual-modal reachable time relationship constraints, thereby improving the joint detection capability of pressure mutation and acoustic mutation, reducing invalid candidate change points, and enhancing the accuracy of early identification of abnormal events and subsequent homologous matching.

[0027] In this embodiment, the same-source matching segmentation module includes: The pipeline propagation constraint parameters corresponding to the dual-modal variable point feature set are called. The negative pressure wave time difference is checked according to the pressure variable point features and the pressure wave propagation path to obtain the pressure side time difference matching result. The negative pressure wave time difference check is specifically as follows: the inflection point time of the negative pressure wave is located from the pressure variable point features of different associated monitoring nodes. The arrival order and arrival time difference of the negative pressure wave inflection points are compared according to the pressure wave propagation path. Pressure variable points that do not conform to the pressure wave propagation path are eliminated. This is used to determine whether the pressure variable points on different associated monitoring nodes conform to the time difference relationship of the arrival of the negative pressure wave along the pressure wave propagation path. Acoustic spectrum analysis is performed based on the acoustic variable point features and sound wave propagation path in the dual-modal variable point feature set to obtain acoustic side feature matching results. The acoustic spectrum analysis specifically involves: extracting acoustic time-series segments near the acoustic variable points, performing frequency band energy analysis on the acoustic time-series segments, identifying leakage jet sound waves, pipe wall vibration, and acoustic energy peaks, and removing acoustic variable points that do not conform to the sound wave propagation path. This is used to identify leakage jet sound wave features, pipe wall vibration features, and acoustic energy peak features from the acoustic variable point features, and to determine whether the acoustic variable points on different associated monitoring nodes conform to the time difference relationship of the arrival of leakage sound waves along the sound wave propagation path. By combining the pressure-side time difference matching results, the acoustic-side feature matching results, and the dual-mode reachability time correspondence, asynchronous propagation time matching is performed to obtain pressure-acoustic co-originating change point pairs. The asynchronous propagation time matching is specifically as follows: the inflection point time of the negative pressure wave is paired with the acoustic change point time according to the dual-mode reachability time correspondence, and the change point combination that simultaneously satisfies the correspondence of pressure wave propagation path, sound wave propagation path, and abnormal feature is retained. Dual-modal fusion localization preprocessing is performed using pressure-acoustic co-source variable point pairs. Based on the fusion localization preprocessing results, the segment boundaries of abnormal events are determined, and a set of pipeline abnormal event segments is synthesized. The dual-modal fusion localization preprocessing involves associating the pressure-side time difference matching results, acoustic-side feature matching results, pressure wave propagation paths, and sound wave propagation paths with the same pipeline abnormal event to determine the candidate pipeline segments and candidate time periods corresponding to the abnormal event.

[0028] This invention uses a common-source matching segmentation module to jointly analyze pressure change point features, acoustic change point features, and pipeline propagation constraint parameters. It uses negative pressure wave time difference verification and acoustic spectrum analysis to identify pressure anomalies and acoustic anomalies corresponding to the same abnormal event. Furthermore, it uses asynchronous propagation time matching and dual-modal fusion positioning preprocessing to determine candidate pipe segments and candidate time periods, thereby improving the accuracy of common-source identification, segmented positioning, and subsequent risk assessment of abnormal events.

[0029] In this embodiment, the risk marker uploading module includes: The edge layer reads the pipeline abnormal event segment set, locates the segment start and end time corresponding to each abnormal event segment according to the abnormal event segment identifier, and cuts back the pressure segment data and acoustic segment data from the synchronous dual-modal time series data according to the segment start and end time to form an abnormal event segment record; The pipeline network operation environment data is aligned to the abnormal event segment records according to the associated monitoring nodes, effective propagation paths and segment start and end times, and data that does not fall within the segment start and end times are removed to obtain segment-related operation data; Based on the pressure segment data and acoustic segment data, the abnormal changes within the segment are compared, and the pressure-acoustic co-source change point is combined to verify whether the pressure abnormality and acoustic abnormality correspond to the same pipeline network abnormal event. At the same time, the verification results are correlated with the segment-related operating environment data to obtain the risk characteristics of the abnormal segment. Each abnormal event segment is ranked according to its risk characteristics, and a warning value is assigned based on the risk ranking results. The warning value is used to characterize the priority of abnormal event segments entering the cloud big data layer for security assessment and graded warning. Based on the early warning value, the segmented set of abnormal events in the pipeline network is segmented and marked. The segmented marking process includes removal marking, retention marking, upload marking, and cloud verification marking. Removal marking corresponds to abnormal event segments that do not meet the dual-modal accompanying change condition. Retention marking corresponds to abnormal event segments that need to be tracked at the edge layer. Upload marking corresponds to abnormal event segments that need to be uploaded to the cloud big data layer for safety assessment. Cloud verification marking corresponds to abnormal event segments that need to be verified by combining hydraulic mechanism verification and data-driven identification. Abnormal events with upload and cloud review tags are segmented and encapsulated into abnormal segment upload data packets, while abnormal event segments with removal and retention tags are retained in the local records of the edge layer.

[0030] This invention uses a risk-marking upload module to perform dual-modal back-switch, operating environment association, source verification, and risk ranking for pipeline network anomaly events. It can filter out low-value anomaly segments at the edge and prioritize the upload of high-risk segments, reducing invalid data transmission and cloud review pressure, and improving the pertinence, real-time performance, and reliability of anomaly segment handling.

[0031] In this embodiment, the cloud-based calibration and evaluation module includes: The cloud-based big data layer receives and parses abnormal segmented data packets, and performs cleaning, deduplication, standardization, and spatiotemporal alignment on the parsed data before writing it into the pipeline holographic database. The pipeline holographic database is used to store abnormal segmented data, pipeline operating environment data, historical fault data, and equipment operating data uploaded via the edge layer. The cloud-based verification and evaluation parameters are invoked to map the abnormal segment association information to the pipeline network topology. Hydraulic mechanism verification is then performed on the target pipe segment, associated monitoring node, and effective propagation path corresponding to the abnormal event segment, yielding the mechanism verification results. The cloud-based verification and evaluation parameters include pipeline network topology parameters, hydraulic boundary parameters, equipment operating parameters, and historical fault parameters. The hydraulic mechanism verification process utilizes a hydraulic calculation model to simulate pressure, flow rate, and velocity changes within the area corresponding to the abnormal event segment. The simulation results are compared with the pressure changes, acoustic changes, and pipeline network operating environment changes in the uploaded data packet of the abnormal segment. The hydraulic calculation model refers to a pipeline network fluid dynamics calculation model established based on the pipeline network topology, pipe segment length, pipe diameter, pipe material, valve status, pressure boundary, and flow boundary. This model is used to map the abnormal segment association information in the uploaded data packet of the abnormal segment to the pipeline network topology, determine the target verification area corresponding to the abnormal event segment, and simulate pressure, flow rate, and velocity changes under leakage, pipe burst, water hammer, or abnormal operating conditions within the target verification area, providing physical constraints for data-driven identification. Initiate data-driven identification by inputting the abnormal segmented data packets into the cloud AI algorithm library to identify leakage, pipe burst, water hammer, and abnormal operating conditions in the abnormal event segments and obtain data identification results. The cloud AI algorithm library refers to the set of abnormal identification models deployed in the cloud big data layer, which is used to receive abnormal segmented data packets, historical fault data, and pipeline network operation environment data after data governance, and to perform data-driven identification of abnormal event segments. The consistency of the mechanism verification results and the data identification results is compared. When they are consistent, the abnormal event is confirmed. When they are inconsistent, the hydraulic mechanism verification results are used to constrain the data identification results. The data identification results are then used to reversely correct the local parameters in the hydraulic mechanism verification, and a collaborative calibration result is generated. Based on the collaborative calibration results, the target assessment area and target abnormal pipe sections are determined. The target assessment area is then divided into assessment units according to pipe sections, nodes, and equipment. Structural safety assessment, operational safety assessment, environmental safety assessment, and equipment health assessment are performed on each assessment unit. The assessment results are weighted and summarized before being input into the fuzzy comprehensive evaluation process to generate the pipeline network safety assessment results. The weighting and summarization process involves assigning corresponding weights to the structural safety assessment results, operational safety assessment results, environmental safety assessment results, and equipment health assessment results according to the importance of each assessment unit in the target assessment area, the scope of abnormal impact, and historical failure information. The weighted assessment results are then combined into a safety score. The fuzzy comprehensive evaluation process involves matching the safety score with the preset risk level membership intervals to determine the degree of membership of the target assessment area to different risk levels. The risk level with the highest degree of membership is selected as the pipeline network risk level. The pipeline network safety assessment results include the pipeline network safety index, risk level, target abnormal pipe section, scope of abnormal impact, and weak links.

[0032] This invention uses a cloud-based calibration and evaluation module to manage, verify, identify, and collaboratively calibrate abnormal segmented uploaded data. This allows the judgment of pipeline network anomalies to be constrained by both hydraulic physical laws and historical data models, improving the accuracy of identifying leaks, pipe bursts, water hammer, and abnormal operating conditions, reducing false alarms and missed alarms. Furthermore, it generates pipeline network safety indices, risk levels, and weak points through multi-dimensional safety assessments, enhancing the reliability and precision of regional pipeline network safety assessments.

[0033] In this embodiment, the early warning work order feedback module includes: The application layer receives the pipeline safety assessment results and calls the hierarchical early warning engine to generate early warning information. The hierarchical early warning engine divides the early warning level according to the pipeline safety index, the risk characteristics of abnormal segments, and the scope of abnormal impact. The early warning level includes general early warning, relatively severe early warning, serious early warning, and extremely serious early warning. The system pushes early warning information to the BI dashboard, early warning handling module, and mobile terminal APP, and generates operation and maintenance work orders based on the early warning information. The BI dashboard is used to display the pipeline topology, monitoring points, early warning distribution, and risk heat map. The early warning handling module is used to generate dispatch tasks based on the target abnormal pipeline section, abnormal event type, and early warning level. The mobile terminal APP is used to receive early warning information, perform on-site inspections, and provide feedback on handling results. The cloud-based big data layer receives early warning information and maintenance work orders, associates them with the corresponding pipeline network abnormal event segmentation set, generates segmentation strategy correction information, and sends the segmentation strategy correction information to the edge layer to update the next round of segmentation strategy. The next round of segmentation strategy includes pipeline network event reachability interval generation, accompanying random detection interval generation, dual-modal change point screening, pipeline network abnormal event segmentation synthesis, and segmentation handling marking.

[0034] Example 1: To verify the feasibility of this invention in practice, it was applied to a regional water supply network safety monitoring scenario. This network includes main pipe sections, branch pipe sections, valve wells, pipe connection points, and pressure zone boundaries. Problems include pipe aging, significant local pressure fluctuations, frequent historical leaks, and disturbances caused by valve opening and closing. Traditional monitoring methods mainly rely on pressure threshold alarms and manual inspections. When faced with small-flow leaks, early pipe burst disturbances, and water hammer impacts, these methods are prone to problems such as alarm lag, numerous false alarms, excessively large abnormal location ranges, and low on-site investigation efficiency.

[0035] In this scenario, sensing layer acquisition devices are deployed along the target area's pipeline network, including valve wells, pipe connection points, pressure zone boundaries, and historical leakage locations. High-frequency pressure sensors collect pressure time-series data, high-sensitivity hydrophone sensors collect acoustic time-series data, and extended sensors collect flow rate, water quality, liquid level, and hazardous gas data. Pressure time-series data is collected at a sampling frequency of 1 kHz, and acoustic time-series data at an 8 kHz sampling frequency. A unified time base label, sampling time correction, and time-series alignment are performed by an edge layer based on satellite timing information to generate synchronous dual-modal time-series data. A total of 36 monitoring nodes are set up in the test pipeline network, including 14 key monitoring nodes and 22 ordinary monitoring nodes, covering a main pipe section of approximately 18.6 km and a branch pipe section of approximately 31.4 km, forming a 50 km-level regional pipeline network monitoring range.

[0036] The edge layer filters effective propagation paths based on pipeline propagation constraint parameters, determines the reachable time ranges of pressure events and acoustic events, generates pipeline event reachable intervals, and generates random detection intervals only within these reachable intervals. Pipeline-constrained wild binary segmentation is then performed on the synchronous dual-modal time-series data. The system forms accompanying random detection intervals through the reachable sub-intervals of pressure and acoustic events, extracts pressure and acoustic abrupt change features, and eliminates candidate change points that do not meet the dual-modal reachable time correspondence conditions, thus obtaining a dual-modal change point feature set.

[0037] Subsequently, the homogeneous matching segmentation module performs negative pressure wave time difference verification on pressure change points and acoustic spectrum analysis on acoustic change points. It then performs asynchronous propagation time matching on the pressure-side time difference matching results, acoustic-side feature matching results, and the dual-modal reachability time correspondence to obtain pressure-acoustic homogeneous change point pairs, and synthesizes a set of pipeline abnormal event segments. The risk labeling upload module generates abnormal segment risk characteristics based on pressure changes, acoustic changes, and pipeline operating environment changes in the abnormal event segments. It then performs risk ranking and segmented handling labeling on the abnormal segments, encapsulating only the abnormal segments corresponding to the upload label and cloud verification label into an abnormal segment upload data packet.

[0038] After receiving the abnormal segmented data packets, the cloud-based big data layer cleans, deduplicates, standardizes, and spatiotemporally aligns them, then writes them into the pipeline network holographic database. A hydraulic calculation model is used for hydraulic mechanism verification, and a cloud-based AI algorithm library is used to identify leaks, pipe bursts, water hammer, and abnormal operating conditions. The cloud collaboratively calibrates the mechanism verification results and data identification results to generate a pipeline network safety index, risk level, target abnormal pipe segments, abnormal impact range, and weak points. Based on the assessment results, the application layer generates early warning information and maintenance work orders, and corrects the edge abnormal segment detection parameters through on-site feedback.

[0039] During continuous verification, the system of this invention was compared with a traditional threshold alarm combined with centralized cloud analysis. The traditional method mainly relies on a single pressure threshold and manual review, with all high-frequency data uploaded to the cloud for analysis. Specific comparison data are shown in Table 1. Table 1 Comparison of the Implementation Effects of Regional Pipeline Network Safety Monitoring

[0040] As shown in Table 1, under the same monitoring scenario, this invention reduces the average daily uploaded data volume from 186GB to 38GB, and increases the proportion of effective abnormal segments from 18.7% to 71.3%, indicating that the edge-side risk marking upload mechanism can reduce invalid data transmission and improve the value of uploaded data. The average identification time is shortened from 8.6min to 1.9min, the accuracy rate of small flow leakage identification is increased from 71.4% to 90.5%, and the accuracy rate of water hammer event identification is increased from 76.7% to 93.3%, indicating that pipeline-constrained wild binary segmentation and pressure-acoustic homologous matching improve the efficiency and accuracy of anomaly identification. The overall false alarm rate is reduced from 23.8% to 7.6%, the anomaly candidate location range is narrowed from 180m to 260m to 28m to 55m, and the average on-site investigation time is shortened from 96min to 38min, indicating that this invention can improve the reliability of early warning, narrow the investigation range, and improve the efficiency of on-site handling.

[0041] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A regional pipeline network safety assessment and early warning system based on big data and edge computing, characterized in that, include: The four-layer perception synchronization module is used to construct a four-layer architecture including a perception layer, an edge layer, a cloud big data layer, and an application layer. The perception layer collects pressure time-series data and acoustic time-series data, and performs time synchronization through the edge layer to generate synchronous dual-modal time-series data. The reachability interval generation module is used to generate reachability intervals for pipeline events based on pipeline propagation constraint parameters. The segmentation and detection module is used to generate random detection intervals within the reachable interval of pipeline events, and to perform pipeline-constrained wild binary segmentation on synchronous dual-modal time-series data to obtain a dual-modal change point feature set; The homogeneous matching segmentation module is used to perform asynchronous propagation time matching on the dual-modal variable point feature set according to the pipeline propagation constraint parameters, to obtain pressure-acoustic homogeneous variable point pairs, and to synthesize pipeline abnormal event segmentation sets based on the pressure-acoustic homogeneous variable point pairs; The risk labeling and uploading module is used to extract the risk characteristics of abnormal segments based on the segment set of abnormal events in the pipeline network, generate early warning value, perform segmented handling and labeling, and generate abnormal segment uploading data packets. The cloud calibration and evaluation module is used to perform hydraulic mechanism verification and data-driven identification based on the data packets uploaded in abnormal segments and the cloud verification and evaluation parameters. It also performs collaborative calibration on the mechanism verification results and data identification results to generate pipeline safety assessment results. The early warning work order feedback module is used to generate early warning information and operation and maintenance work orders based on the pipeline safety assessment results. The cloud big data layer generates segmentation strategy correction information based on the early warning information and operation and maintenance work orders, and sends it to the edge layer to update the next round of segmentation strategy.

2. The regional pipeline network safety assessment and early warning system based on big data and edge computing according to claim 1, characterized in that, The four-layer sensing synchronization module includes: Sensing layer acquisition devices are deployed at regional pipeline network monitoring nodes, and a four-layer architecture consisting of sensing layer, edge layer, cloud big data layer and application layer is established, while data transmission channels between layers are also established. The sensing layer collects pressure timing data and acoustic timing data through sensing layer acquisition devices, and also collects pipeline network operating environment data, and transmits them together to the edge layer. After receiving force timing data and acoustic timing data at the edge layer, the pressure timing data and acoustic timing data are processed by unified time base marking, sampling time correction and timing alignment according to satellite timing information to generate synchronous dual-mode timing data.

3. The regional pipeline network safety assessment and early warning system based on big data and edge computing according to claim 1, characterized in that, The reachability interval generation module includes: The edge layer reads the monitoring node identifier, sampling time range, and data acquisition status corresponding to the synchronous dual-modal time series data, and calls the pipeline propagation constraint parameters corresponding to the monitoring node identifier; The edge layer filters the pipeline connectivity between monitoring nodes based on the pipe segment connection relationship and valve status in the pipeline propagation constraint parameters. Pipe segments that are in a disconnected state, closed state, or do not meet the propagation path conditions are removed, while associated monitoring nodes that meet the propagation conditions of abnormal events are retained. The effective propagation path is determined based on the length, diameter, and material of the pipe segments traversed between the associated monitoring nodes. The edge layer determines the reachable time range of pressure events and the reachable time range of acoustic events between associated monitoring nodes based on the effective propagation path, pipeline propagation constraint parameters, and data acquisition status, and merges them to generate the reachable interval of pipeline events.

4. The regional pipeline network safety assessment and early warning system based on big data and edge computing according to claim 1, characterized in that, The segmentation detection module includes: The edge layer generates a random mother interval within the reachable interval of pipeline events, and maps the random mother interval to the reachable sub-intervals of pressure events and acoustic events respectively, and combines them to obtain the accompanying random detection interval; The accompanying random detection interval is mapped to synchronous dual-modal time series data. Pressure detection sequences are extracted from pressure event reachable sub-intervals, and acoustic detection sequences are extracted from acoustic event reachable sub-intervals. A set of candidate segmentation times from the same source is constructed based on pipeline propagation constraint parameters and synchronization state parameters. Pipeline-constrained wild binary segmentation is performed on the pressure detection sequence and acoustic detection sequence. Within each associated random detection interval, the pressure detection sequence and acoustic detection sequence are divided into front and back segments according to the set of candidate segmentation times from the same source. Pressure mutation features and acoustic mutation features are calculated respectively. Based on the pressure mutation features, acoustic mutation features and the corresponding conditions of dual-modal reachable time, a dual-modal associated segmentation score is generated. Candidate time combinations that satisfy the change point retention conditions in the dual-modal associated segmentation score are selected as associated candidate change points. Using the candidate change point as the dividing boundary, the random detection interval is recursively divided into binary segments to generate the next round of random detection sub-intervals. The pipeline-constrained wild binary segmentation process is repeated in the next round of random detection sub-intervals until the recursive stopping condition is met. The candidate variable points obtained from each round of recursive partitioning are aggregated and screened to remove candidate variable points that do not meet the conditions of bimodal reachability time correspondence, random interval repeated verification, and bimodal accompanying change, thus obtaining the bimodal variable point feature set.

5. The regional pipeline network safety assessment and early warning system based on big data and edge computing according to claim 1, characterized in that, The homogeneous matching segmentation module includes: Call the pipeline propagation constraint parameters corresponding to the dual-modal variable point feature set, perform negative pressure wave time difference verification based on pressure variable point characteristics and pressure wave propagation path, and obtain pressure side time difference matching results; Acoustic spectrum analysis is performed based on the acoustic variable point features and sound wave propagation path in the dual-modal variable point feature set to obtain acoustic side feature matching results; By combining the pressure-side time difference matching results, the acoustic-side feature matching results, and the dual-modal reachability time correspondence, asynchronous propagation time matching is performed to obtain pressure-acoustic homogeneous variable point pairs. Dual-modal fusion positioning preprocessing is performed using pressure-acoustic homogeneous variable point pairs, and the abnormal event segment boundaries are determined based on the fusion positioning preprocessing results to synthesize a pipeline abnormal event segment set.

6. The regional pipeline network safety assessment and early warning system based on big data and edge computing according to claim 1, characterized in that, The risk marker uploading module includes: The edge layer reads the pipeline abnormal event segment set, locates the segment start and end time corresponding to each abnormal event segment according to the abnormal event segment identifier, and cuts back the pressure segment data and acoustic segment data from the synchronous dual-modal time series data according to the segment start and end time to form an abnormal event segment record; The pipeline network operation environment data is aligned to the abnormal event segment records according to the associated monitoring nodes, effective propagation paths and segment start and end times, and data that does not fall within the segment start and end times are removed to obtain segment-related operation data; Based on the pressure segment data and acoustic segment data, the abnormal changes within the segment are compared, and the pressure-acoustic co-source change point is combined to verify whether the pressure abnormality and acoustic abnormality correspond to the same pipeline network abnormal event. At the same time, the verification results are correlated with the segment-related operating environment data to obtain the risk characteristics of the abnormal segment. The risk of each abnormal event segment is ranked according to its risk characteristics, and a warning value is assigned based on the risk ranking results. Based on the early warning value, the abnormal events in the pipeline network are segmented and marked for segmented handling. The segmented handling and marking process includes removing markers, retaining markers, uploading markers, and cloud verification markers. Abnormal events with upload and cloud review tags are segmented and encapsulated into abnormal segment upload data packets, while abnormal event segments with removal and retention tags are retained in the local records of the edge layer.

7. The regional pipeline network safety assessment and early warning system based on big data and edge computing according to claim 1, characterized in that, The cloud-based calibration and evaluation module includes: The cloud-based big data layer receives and parses abnormal segmented data packets, and then cleans, deduplicates, standardizes, and spatiotemporally aligns the parsed data before writing it into the pipeline holographic database. The cloud-based verification and evaluation parameters are invoked to map the abnormal segment association information to the pipeline network topology, and the hydraulic mechanism verification is performed on the target pipe segment, associated monitoring node and effective propagation path corresponding to the abnormal event segment to obtain the mechanism verification results. Initiate data-driven identification by inputting the abnormal segmented data packets into the cloud AI algorithm library to identify abnormal events such as leaks, pipe bursts, water hammer, and abnormal operating conditions, and obtain data identification results. The consistency of the mechanism verification results and the data identification results is compared. When they are consistent, the abnormal event is confirmed. When they are inconsistent, the hydraulic mechanism verification results are used to constrain the data identification results. The data identification results are then used to reversely correct the local parameters in the hydraulic mechanism verification, and a collaborative calibration result is generated. Based on the collaborative calibration results, the target assessment area and the target abnormal pipe section are determined. The target assessment area is divided into assessment units according to pipe section, node and equipment. Structural safety assessment, operational safety assessment, environmental safety assessment and equipment health assessment are performed on each assessment unit. The assessment results are weighted and summarized and then input into the fuzzy comprehensive evaluation process to generate the pipeline safety assessment results.

8. The regional pipeline network safety assessment and early warning system based on big data and edge computing according to claim 1, characterized in that, The early warning work order feedback module includes: The application layer receives the pipeline security assessment results and calls the hierarchical early warning engine to generate early warning information; The warning information is pushed to the BI dashboard, the warning handling module, and the mobile terminal APP, and maintenance work orders are generated based on the warning information; The cloud-based big data layer receives early warning information and maintenance work orders, associates them with the corresponding network anomaly event segments, generates segmentation strategy correction information, and sends the segmentation strategy correction information to the edge layer to update the next round of segmentation strategy.

Images