A pipeline corridor risk identification and positioning method and system under multi-protocol driving

Abstract

This application discloses a method and system for risk identification and location in utility tunnels under a multi-protocol parallel driving approach, relating to the field of utility tunnel risk management technology. The method includes: acquiring a utility tunnel structure diagram; deploying multi-source sensors to construct a sensor topology network; continuously monitoring the utility tunnel; identifying characteristic risks, determining characteristic risk nodes and characteristic risk feature vectors, performing association link retrieval, and determining a target association link set; extracting backtracking data to determine a backtracking data sequence set; identifying serialized risk trends to determine a set of hidden risk nodes, and using the characteristic risk nodes and the set of hidden risk nodes as the risk identification and location results. This application solves the technical problem in existing utility tunnel risk monitoring technologies where single sensor deployment makes it difficult to accurately locate hidden risk nodes, resulting in low risk monitoring efficiency. It achieves efficient identification and location of utility tunnel risks, effectively improving the technical effect of utility tunnel risk monitoring and control capabilities.

Description

Technical Field

[0001] This application relates to the field of utility tunnel risk management technology, specifically to a method and system for identifying and locating utility tunnel risks under a multi-protocol approach. Background Technology

[0002] With the continuous expansion of urban underground utility tunnel construction, the complex environment and diverse facilities within these tunnels place higher demands on risk monitoring and control. Existing utility tunnel risk monitoring technologies generally suffer from problems such as single sensor deployment, poor data transmission protocol compatibility, and reliance on human experience for risk identification, resulting in the inability to obtain real-time and comprehensive data on the operational status of the utility tunnels. At the same time, traditional methods can only identify surface-level risks and are insufficient to locate hidden risk nodes through historical data backtracking and risk trend analysis.

[0003] Existing technologies for monitoring risks in utility tunnels suffer from problems such as the limited deployment of sensors, making it difficult to accurately locate hidden risk nodes and resulting in low monitoring efficiency. Summary of the Invention

[0004] This application provides a method and system for risk identification and location of utility tunnels under multi-protocol parallel driving, which is used to address the technical problem that the existing technology of utility tunnel risk monitoring has low efficiency due to the single deployment of sensors, difficulty in accurately locating hidden risk nodes.

[0005] In view of the above problems, this application provides a method and system for identifying and locating risks in utility tunnels under multi-protocol parallel operation.

[0006] A first aspect of this application provides a method for risk identification and location of utility tunnels under multi-protocol parallel operation, the method comprising:

[0007] A structural diagram of the utility tunnel is obtained. Based on this diagram, multi-source sensors are deployed within the tunnel to construct a sensor topology network. This network is then used for continuous monitoring of the tunnel, and monitoring data is transmitted to a transmission center via various communication protocols to generate a real-time sensor monitoring data topology network. This transmission center includes an intermediate flow and backtracking database. Risk identification is performed based on this real-time sensor monitoring data topology network to identify risk-characterizing nodes and feature vectors. Related link retrieval is conducted within the sensor topology network to determine a target set of related links. For each related topology node in the target set, backtracking data is extracted from the intermediate flow and backtracking database to determine a set of backtracking data sequences for the related topology nodes. Serialized risk trend identification is performed on the backtracking data sequence set to determine a set of hidden risk nodes. The identified risk nodes and the set of hidden risk nodes are used as the risk identification and location results.

[0008] A second aspect of this application provides a multi-protocol parallel-driven utility tunnel risk identification and positioning system, the system comprising:

[0009] The system includes a sensor topology network construction module for acquiring a pipe gallery structure diagram, deploying multi-source sensors within the pipe gallery based on the diagram, and constructing a sensor topology network. A topology network generation module is used for continuous monitoring of the pipe gallery using the sensor topology network and transmitting monitoring data to the transmission center of the sensor topology network via various communication protocols to generate a real-time sensor monitoring data topology network. The transmission center includes an intermediate flow backtracking database. An associated link set determination module is used for risk identification based on the real-time sensor monitoring data topology network, identifying risk-characterizing nodes and risk-characterizing feature vectors, and performing associated link retrieval within the sensor topology network to determine a target associated link set. A backtracking data extraction module is used for extracting backtracking data from each associated topology node in the target associated link set from the intermediate flow backtracking database to determine a set of backtracking data sequences for the associated topology nodes. A hidden risk node set determination module is used for serialized risk trend identification of the set of backtracking data sequences for the associated topology nodes to determine a set of hidden risk nodes. The identified risk nodes and the set of hidden risk nodes are used as the risk identification and location results.

[0010] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0011] The process involves acquiring a structural diagram of the utility tunnel, deploying multi-source sensors, and constructing a sensor topology network. This network is then used for continuous monitoring of the tunnel, transmitting monitoring data to the network's transmission center via various communication protocols to generate a real-time sensor monitoring data topology network. Based on this network, risk identification is performed to determine risk-characterizing nodes and feature vectors. Related link retrieval is then conducted to identify a target set of related links. For each related topology node in the target set, backtracking data is extracted from the intermediate backtracking database to determine a set of backtracking data sequences for the related topology nodes. Serialized risk trend identification is then performed to determine a set of hidden risk nodes. The identified risk nodes and the set of hidden risk nodes are used as the risk identification and location results. This approach achieves efficient identification and location of risks in the utility tunnel, effectively improving the risk monitoring and control capabilities. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 A schematic flowchart of a method for risk identification and location of utility tunnels under multi-protocol parallel driving provided in this application embodiment;

[0014] Figure 2 A schematic diagram of another method for identifying and locating risks in utility tunnels under multi-protocol parallel driving provided in this application embodiment;

[0015] Figure 3 A schematic diagram of another method for identifying and locating risks in utility tunnels under multi-protocol parallel driving provided in this application embodiment;

[0016] Figure 4 A schematic diagram of another method for identifying and locating risks in utility tunnels under multi-protocol parallel driving provided in this application embodiment;

[0017] Figure 5 A schematic diagram of another method for identifying and locating risks in utility tunnels under multi-protocol parallel driving provided in this application embodiment;

[0018] Figure 6 A schematic diagram of another method for identifying and locating risks in utility tunnels under multi-protocol parallel driving provided in this application embodiment;

[0019] Figure 7 A schematic diagram of another method for identifying and locating risks in utility tunnels under multi-protocol parallel driving provided in this application embodiment;

[0020] Figure 8 This is a schematic diagram of a multi-protocol parallel-driven pipe gallery risk identification and positioning system provided in an embodiment of this application.

[0021] Figure labeling: Sensor topology network construction module 10, topology network generation module 20, associated link set determination module 30, backtracking data extraction module 40, hidden risk node set determination module 50. Detailed Implementation

[0022] This application provides a method and system for identifying and locating risks in utility tunnels under a multi-protocol parallel approach, which addresses the technical problem of low risk monitoring efficiency in existing technologies, where the deployment of sensors is limited and it is difficult to accurately locate hidden risk nodes.

[0023] The technical solutions of the embodiments of this application 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 this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0024] like Figure 1 As shown, this application provides a method for risk identification and location of utility tunnels under multi-protocol parallel operation, the method comprising:

[0025] Step S100: Obtain the pipe gallery structure diagram, and deploy multi-source sensors in the pipe gallery based on the pipe gallery structure diagram to construct a sensor topology network.

[0026] Specifically, the process begins by acquiring a structural map of the utility tunnel, including its spatial layout, route direction, and node distribution. Based on these structural features, a set of tunnel routes and the locations of route intersections are extracted. Then, various types of sensors are deployed within the tunnel according to different monitoring needs: acoustic sensors are strategically placed in leak-prone areas such as pipe joints and valves (to capture abnormal sound waves generated by gas leaks); optical sensors are deployed in structurally weak sections (such as bends and material transition zones) (to acquire real-time image information of cracks and deformation); and gas sensors are densely deployed along pipelines transporting flammable and explosive gases (to monitor changes in the concentration of methane, carbon monoxide, etc.). Simultaneously, composite sensor nodes of all types are added at route intersections, forming an initial sensor set covering key risk points. Finally, using each sensor as a topological node, a sensor topology network with spatial coordinates is constructed according to the physical connection relationships of the tunnel routes. The links between nodes reflect the spatial adjacency of the monitoring area, laying the foundation for subsequent spatial correlation analysis of multi-source data.

[0027] Step S200: Continuously monitor the pipe gallery using the sensor topology network, and transmit the monitoring data to the transmission center of the sensor topology network through multiple communication protocols to generate a real-time sensor monitoring data topology network, wherein the transmission center includes an intermediate flow and backtracking database.

[0028] Specifically, the sensor topology network uses multiple types of sensors (acoustic, optical, and gas) to continuously monitor the utility tunnel at high frequency, acquiring multimodal data such as sound waveforms, structural images, and gas concentrations in real time. After being encoded using adapted communication protocols (such as ZigBee for image transmission, LoRa for gas concentration transmission, and Wi-Fi for acoustic signal transmission), the data is aggregated from edge nodes to a transmission center equipped with an intermediate backtracking database. This database serves as a unified data hub, storing raw sensor data and topological association information according to preset time windows (such as the most recent 7 days / 30 days), forming a structured data pool. The transmission center performs timestamp synchronization, spatial coordinate calibration, and format normalization on the data, generating a real-time sensor monitoring data topology network based on the connection relationships between network nodes. This network dynamically maps the physical space of the utility tunnel, providing both real-time data display and rapid historical data retrieval capabilities. Compared to traditional distributed storage models, the intermediate flow backtracking database optimizes data extraction through pre-aggregation and indexing, eliminating the need to retrieve data from each sensor during subsequent risk analysis. Instead, it allows for the batch extraction of backtracking data sequences from target nodes (such as the acoustic-optical-gas data chain of a pipeline section over the past 24 hours) based on topology network links. This reduces data extraction time from minutes to seconds, significantly improving the efficiency of multi-source data fusion analysis.

[0029] Step S300: Based on the real-time sensor monitoring data topology network, identify the risk characteristics, determine the risk-characterizing nodes and risk-characterizing feature vectors, and perform a link retrieval in the sensor topology network to determine the target link set.

[0030] Specifically, based on the real-time sensor monitoring data topology network, each topology node is automatically scanned through preset multi-source risk identification rules (acoustic rule: sound frequency / amplitude exceeds the threshold; optical rule: image feature vector detects cracks or deformations through convolution analysis; gas rule: gas type / concentration exceeds the standard). Nodes that trigger anomalies are marked as risk nodes, and corresponding feature vectors are extracted (such as acoustic feature vectors containing abnormal frequency components, optical feature vectors containing crack pixel ratios, and gas feature vectors containing excessive concentration values). Subsequently, taking the risk-characterizing node as the center, the associated backtracking radius is determined based on the risk-characterizing feature vector matching backtracking radius matching table (a feature-distance mapping table generated by clustering historical risk data). For example, a gas leak risk corresponds to a 50-meter backtracking radius, and a structural crack corresponds to a 30-meter backtracking radius. Based on this radius, all pipe gallery routes and nodes connected to the risk-characterizing node and whose distance is less than or equal to the backtracking radius are retrieved in the sensor topology network, forming a set of target associated links containing potential risk propagation paths. For example, after a node with excessive gas concentration triggers the retrieval, all sensor nodes and connecting routes within 50 meters upstream and downstream of the pipe where it is located are included in the associated links, providing an analysis scope for subsequent tracing of the risk source.

[0031] Step S400: For each associated topology node in the target associated link set, extract backtracking data from the intermediate flow backtracking database to determine the associated topology node backtracking data sequence set.

[0032] Specifically, for each associated topology node within the target associated link set (such as acoustic, optical, and gas sensor nodes in pipelines upstream and downstream of a risk node), historical monitoring data is extracted in batches by node ID and time range (e.g., 7 days prior to the current time) through the efficient indexing mechanism of the intermediate backtracking database. The extracted data includes time-series sequences from multiple sensors (e.g., sound frequency / amplitude curves from acoustic sensors, crack development image sequences from optical sensors, and concentration change trends from gas sensors), equipment operating status parameters (e.g., valve switching frequency, pipeline pressure fluctuations), and topology association metadata (e.g., node spatial coordinates, link distance). The data extraction process is based on a predefined backtracking data structure template (including timestamps, sensor types, monitoring values, and equipment status fields), ensuring that multi-source data are aligned in the spatiotemporal dimensions to form a structured set of associated topology node backtracking data sequences. For example, in an associated link triggered by a gas leak risk node, high-frequency monitoring data chains from all sensors within the pipeline section where that node is located over the past 48 hours can be quickly retrieved, providing continuous and complete dataset support for subsequent analysis of risk evolution trends and avoiding time-consuming data retrieval and association analysis errors caused by scattered storage.

[0033] Step S500: Perform serialized risk trend identification on the backtracking data sequence set of the associated topological nodes, determine the set of hidden risk nodes, and use the representative risk nodes and the set of hidden risk nodes as the risk identification and positioning results.

[0034] Specifically, implicit risks are identified by serializing and analyzing the backtracking data sequences of associated topological nodes. Time-series analysis algorithms (such as sliding window calculation of fluctuation coefficients and dynamic time warping to match historical anomaly patterns) are used to perform multi-dimensional evaluations of each node's historical data (such as sound frequency fluctuation curves, gas concentration change trends, and structural image feature sequences). Nodes with normal real-time data but significant fluctuations (e.g., standard deviation exceeding 20% ​​of the mean), historical intermittent anomaly records (e.g., short-term gas concentration exceeding limits), or predicted trends reaching risk thresholds (e.g., accelerated crack propagation rate) are identified as implicit risk nodes. Finally, the directly triggering anomaly-characteristic risk nodes are merged with the implicit risk nodes identified through trend analysis to form a risk identification and location result encompassing both immediate risks and potential hazards, achieving precise control from single-point anomaly alarms to full-link risk prediction.

[0035] like Figure 2 As shown, in one possible implementation, step S100 further includes:

[0036] Step S110: Extract the set of pipe gallery routes and the set of pipe gallery route features based on the pipe gallery structure diagram.

[0037] Step S120: Deploy multi-source sensors on the pipeline route set according to the pipeline route feature set to obtain an initial sensor deployment set.

[0038] Step S130: Extract the route intersections from the set of pipe gallery routes, and deploy multi-source sensors at each route intersection to obtain a set of intersecting sensors.

[0039] Step S140: Perform a union operation on the initial set of deployed sensors and the cross-deployed set of sensors to obtain the deployed sensor set.

[0040] Step S150: Treat each deployed sensor in the set of deployed sensors as a topology node, and connect each topology node according to the route of the utility tunnel route set to generate the sensor topology network.

[0041] Specifically, the process begins by acquiring electronic or paper structural diagrams of the utility tunnel, including its plan layout, spatial orientation, and structural parameters. Using image recognition technology (such as Convolutional Neural Networks) or manual annotation, spatial geometric information (e.g., start and end coordinates, route length, curvature angle, etc.) for all tunnel routes is extracted from the structural diagrams, forming a set of utility tunnel routes. Simultaneously, attribute features are extracted for each route, including but not limited to: the type of transported medium (e.g., gas, tap water, heat), pipe material (e.g., steel pipe, concrete pipe), service life, design pressure / flow parameters, and historical maintenance records, forming a feature set of utility tunnel routes that corresponds one-to-one with the route set.

[0042] A differentiated sensor deployment strategy is implemented based on the characteristic set of the pipeline corridor routes: For pipelines transporting flammable and explosive media, gas and acoustic sensors are deployed at preset intervals at key nodes such as valves and interfaces, as well as along the route, to form a leak monitoring network; for sections prone to structural deformation, optical and stress sensors are deployed along the route to capture crack propagation and stress changes in real time; in high-temperature and high-pressure pipeline areas, a monitoring chain composed of temperature and pressure sensors is configured. By matching and mapping route characteristics (such as the type of transported medium, material properties, and risk level) with sensor functions, multiple sensor combinations are customized for each route, ultimately forming an initial sensor deployment set covering all pipeline corridor routes, enabling targeted perception and data collection for different risk scenarios.

[0043] Spatial coordinate analysis is used to identify all route intersections (such as pipeline branch points, crossroads, and junctions of pipelines carrying different media) from the pipeline corridor route set. These intersections are prone to becoming high-risk areas due to their complex structures, stress concentrations, or media interactions. For each intersection, a composite monitoring unit containing multiple types of sensors, including acoustic, optical, and gas sensors, is deployed: acoustic sensors capture abnormal sound waves caused by fluid diversion or pressure changes; optical sensors monitor structural deformation and crack initiation at the intersection; and gas sensors monitor changes in gas concentration in the intersection area in real time (such as the spread of gas leaks). By deploying multi-source sensors at intersections, a cross-deployed sensor array is formed. For example, a composite sensor deployed at a crossroads node of a pipeline corridor can simultaneously collect vibration, image, and gas data from pipelines in four directions, achieving three-dimensional monitoring of complex nodes, compensating for the monitoring blind spots of single-type sensors, and improving the comprehensiveness and accuracy of risk perception.

[0044] The initial set of sensors covering the main body of the utility tunnel route is merged with the set of sensors deployed at intersections. By removing duplicate nodes (such as sensors that coincide with route endpoints and intersections) and retaining all types of sensors, a complete set of deployed sensors is generated. This set integrates monitoring nodes for the main route and monitoring nodes for key intersections, ensuring continuous coverage of the linear path of the utility tunnel while enhancing three-dimensional monitoring of high-risk nodes.

[0045] Each sensor in the deployed sensor set is defined as a topology node and assigned a unique identifier (such as node ID and spatial coordinates). Based on the spatial orientation and connection relationships of the utility tunnel route set, adjacent nodes are connected via virtual links according to the start-end logic: for straight routes, nodes are sequentially connected to form a chain topology; for branch routes, a star-shaped branch structure is constructed with the intersection node as the center; for loop routes, a closed loop topology is generated. During the connection process, attributes such as physical distance and medium flow direction between nodes are recorded simultaneously, ultimately generating a sensor topology network that completely maps to the physical structure of the utility tunnel.

[0046] like Figure 3 As shown, in one possible implementation, step S300 further includes:

[0047] Step S310: Obtain the risk identification rules.

[0048] Step S320: Use the risk identification rules to identify risks in the real-time sensor monitoring data topology network and obtain risk-representing nodes and risk-representing feature vectors.

[0049] Specifically, risk identification rules are obtained by integrating the anomaly detection logic of multiple sensors. These rules include: for acoustic sensors, setting normal threshold ranges for sound frequency and amplitude; exceeding these thresholds triggers anomaly detection; for optical sensors, defining feature vector matching rules for structural anomalies using a pre-defined crack image feature convolution model; and for gas sensors, setting concentration safety limits (e.g., 10% of the lower explosive limit of methane) based on gas type as anomaly detection conditions. After standardization, these rules form a set of rules encompassing acoustic, optical, and gas-related risk identification logics, providing a unified basis for risk screening of real-time data.

[0050] Based on the acquired risk identification rules, the data of each node in the real-time sensor monitoring data topology is matched type by type: For acoustic sensor nodes, the frequency and amplitude values ​​of the monitored sound are extracted and compared with the threshold range in the rules. If they exceed the threshold, the node is marked as an acoustic risk node, and the frequency and amplitude values ​​are encapsulated into a feature vector; For optical sensor nodes, the real-time image is convolutionally analyzed using a crack image feature recognizer to generate an image feature vector. If the crack feature parameters in the vector exceed the preset value of the rules, it is marked as an optical risk node, and the feature vector carries the convolution analysis results; For gas sensor nodes, the gas type and concentration are monitored in real time. If the concentration reaches the safety limit set by the rules (e.g., methane concentration ≥ 10% of the lower explosive limit), it is marked as a gas risk node, and the feature vector contains the gas type and concentration value. By executing multiple rules in parallel, nodes that trigger anomalies and their multi-dimensional feature vectors are captured in real time to form initial risk identification results. For example, a node is marked as a gas-type risk node because it detects a methane concentration of 0.8%LEL (below the lower explosive limit but exceeding the warning value), and its feature vector is recorded as [methane, 0.8%LEL], providing a clear risk anchor point for subsequent correlation link analysis.

[0051] like Figure 4 As shown, in one possible implementation, step S310 further includes:

[0052] Step S311: Obtain acoustic risk identification rules, wherein the acoustic risk identification rules are as follows: when the sensor type is a sound sensor, extract the sound frequency and sound amplitude monitored by the sound sensor for anomaly identification. If an anomaly is identified, the topology node where the sound sensor is located is used as a risk characterization node, and the sound frequency and sound amplitude are embedded into an initially empty vector to construct a risk characterization feature vector.

[0053] Step S312: Obtain optical risk identification rules, wherein the optical risk identification rules are as follows: when the sensor type is an optical sensor, extract the image information monitored by the optical sensor, use a crack image feature recognizer to perform image feature convolution analysis on the image information to obtain an image feature vector, perform anomaly identification based on the image feature vector, and if an anomaly is identified, use the topological node where the optical sensor is located as a risk characterization node and use the image feature vector as a risk characterization feature vector.

[0054] Step S313: Obtain gas risk identification rules, wherein the gas risk identification rules are as follows: when the sensor is a gas sensor, extract the gas type and gas concentration detected by the gas sensor for anomaly identification. If an anomaly is identified, the topology node where the gas sensor is located is used as a risk characterization node, and the gas type and gas concentration are embedded into an initially empty vector to construct a risk characterization feature vector.

[0055] Step S314: Summarize the acoustic risk identification rules, optical risk identification rules and gas risk identification rules to generate a risk identification rule characterizing the risk.

[0056] Specifically, for risk identification using sound sensors, acoustic risk identification rules are defined: After the sound sensor collects sound data in the pipe gallery in real time, it automatically extracts the sound frequency and amplitude parameters, and determines anomalies based on preset normal operating threshold ranges (e.g., frequency 20-2000Hz, amplitude ≤80dB). If the monitored value exceeds the threshold (e.g., detecting a high-frequency abnormal noise at 3000Hz or a sudden vibration at 95dB), the anomaly identification logic is triggered. The topological node where the sensor is located is marked as an acoustic risk node, and the actual monitored sound frequency and amplitude values ​​are embedded into an initially empty feature vector (e.g., the vector is represented as [frequency value, amplitude value]), forming a risk characteristic vector carrying specific anomaly parameters, providing acoustic data support for subsequent risk type classification and correlation analysis.

[0057] For risk identification using optical sensors, an optical risk identification rule is constructed: After the optical sensor acquires real-time images of the pipe gallery structure, the image information is automatically extracted and input into a pre-trained crack image feature recognizer (built based on a convolutional neural network algorithm). The recognizer traverses the image through multiple convolutional kernels, extracting parameters such as edge contours, pixel distribution, and texture features, generating an image feature vector containing indicators such as crack length, width, and pixel percentage (e.g., vector format: [crack length, width, pixel percentage]). A preset anomaly judgment threshold is set (e.g., crack width ≥ 0.2 mm or pixel percentage ≥ 5%). If the parameters in the image feature vector exceed the threshold, a structural anomaly is determined, the corresponding topological node where the optical sensor is located is marked as an optical risk node, and the image feature vector is directly used as the risk characterization feature vector, providing a visual and quantitative analysis basis for structural deformation for subsequent risk localization.

[0058] To address the risk identification of gas sensor types, gas risk identification rules are established: When a gas sensor monitors gas data within the pipe gallery in real time, it automatically analyzes the gas type (e.g., methane, carbon monoxide, oxygen, etc.) and its corresponding concentration value, comparing it with preset thresholds in industry safety standards (e.g., the lower explosive limit of methane is 5% LEL, and the warning threshold is set at 1% LEL). If the detected target gas concentration exceeds the warning threshold (e.g., methane concentration reaches 1.2% LEL) or an abnormal gas type (e.g., flammable gas not transported by a pipeline) is detected, it is determined to be a gas leak or pollution risk. The topology node where the sensor is located is marked as a gas-type risk characterization node, and the actual monitored gas type and concentration value are embedded into an initially empty feature vector (e.g., the vector is represented as [gas type, concentration value]), forming a risk characterization feature vector carrying specific risk parameters, providing key gas-dimensional data support for subsequent risk tracing and diffusion analysis.

[0059] The rules for identifying risks in acoustic, optical, and gas sensors are integrated and summarized to form a unified risk identification rule. Specifically, for acoustic sensors, risks are determined based on anomaly thresholds of sound frequency and amplitude, and feature vectors are constructed. For optical sensors, risks are identified through image feature convolution analysis and crack parameter thresholds, and feature vectors are output. For gas sensors, risk nodes are marked and feature vectors are generated based on whether the gas type and concentration exceed safety limits. The fusion of these three types of rules forms a standardized risk identification logic covering multiple sensor sources. This enables parallel detection of risks such as acoustic anomalies, structural deformations, and gas leaks in the real-time monitoring data topology network, uniformly generating risk identification results that include risk type, node location, and multi-dimensional feature parameters. This provides structured rule support for subsequent risk correlation analysis and precise location.

[0060] Continue as Figure 3 As shown, in one possible implementation, step S300 further includes:

[0061] Step S330: Based on the location of the risk-representing node, extract the associated utility tunnel routes from the sensor topology network to obtain a set of associated utility tunnel routes.

[0062] Step S340: Perform backtracking radius matching based on the risk characteristic vector to obtain the associated backtracking radius.

[0063] Step S350: Based on the risk-characterizing nodes and the associated backtracking radius, extract the associated links from the associated utility tunnel route set to determine the target associated link set.

[0064] Specifically, based on whether the risk-representing node is located at a route intersection, differentiated associated utility tunnel routes are extracted from the sensor topology network: If the risk-representing node is located at a non-route intersection (i.e., the middle section or end point of a single utility tunnel route), the unique utility tunnel route containing that node is directly extracted as the associated route. For example, if an optical risk-representing node is located in the middle of gas pipeline L3, the associated utility tunnel route set only includes L3. If the risk-representing node is located at a route intersection (such as the junction of two or more utility tunnel routes), all utility tunnel routes passing through that intersection are extracted through topological network connectivity analysis. For example, if an acoustic risk-representing node is located at the intersection of routes L1, L2, and L3, the associated utility tunnel route set will include all routes of L1, L2, L3, and their extension directions. Through this differentiated extraction strategy, the physical path range of the risk node is accurately identified.

[0065] A pre-built backtracking radius matching mechanism is used to analyze risk feature vectors to obtain corresponding associated backtracking radii. First, a matching table containing historical risk feature vector prototypes and preset backtracking radii is called (this table is generated based on feature vectors and risk propagation distances from historical utility tunnel risk logs; for example, the historical maximum associated distance for a certain type of gas leak feature is 30 meters). The current risk feature vector is compared with the aggregated historical risk feature vector prototypes in the table using similarity calculations (e.g., cosine similarity), and the prototype with the highest similarity is selected as the matching result. Then, the prototype backtracking radius corresponding to this prototype is extracted from the matching table as the associated backtracking radius for this analysis. For example, if the current feature vector is [methane, 1.5% LEL], and its similarity to the prototype vector marked as a gas leak in the matching table reaches 92%, its corresponding prototype backtracking radius is 40 meters, then the associated backtracking radius is determined to be 40 meters. This process, through historical data-driven feature matching, dynamically adapts the backtracking range to different risk types, providing a scientific spatial retrieval threshold for subsequent associated link extraction.

[0066] Centered on the risk-characterizing node, and combining the associated backtracking radius and the set of associated utility tunnel routes, spatially limited associated link extraction is performed: First, the spatial coordinates of the risk-characterizing node are used as a reference point. Within the extracted set of associated utility tunnel routes (such as a path network including main routes and intersection routes), the extraction extends upstream and downstream along the route direction, defining the physical distance retrieval range according to the associated backtracking radius (such as 30 meters or 50 meters). Then, all topological nodes and connecting links within this range are traversed, and links directly or indirectly connected to the risk-characterizing node are extracted, forming a target associated link set containing attributes such as node connection relationships and media flow direction. For example, if the risk-characterizing node is located 100 meters from gas pipeline L2, and the associated backtracking radius is 40 meters, then in L2 and its intersection routes L1 and L3, all sensor nodes and pipeline connections within the interval from 60 meters to 140 meters of L2 are extracted, including links at the L1 / L2 intersection and the L2 / L3 intersection. This set not only covers the directly affected areas around the risk nodes, but also extends to neighboring paths that may be affected by risk propagation through route correlation, providing a structured link network model for subsequent retrospective data extraction and risk trend analysis, ensuring that potential paths of risk propagation can be fully captured.

[0067] like Figure 5 As shown, in one possible implementation, step S340 further includes:

[0068] Step S341: Pre-build the backtracking radius matching table.

[0069] Step S342: Perform similarity matching between the risk representation feature vector and the aggregated historical risk representation feature vector prototype in the backtracking radius matching table, and take the aggregated historical risk representation feature vector prototype corresponding to the maximum similarity as the matching result.

[0070] Step S343: Extract the prototype backtrack radius of the matching result from the backtrack radius matching table as the associated backtrack radius.

[0071] Specifically, a backtracking radius matching table is pre-constructed using a historical data-driven approach. The process is as follows: First, historical utility tunnel risk logs are collected, and historical risk characteristic vectors (covering parameters such as acoustic frequency / amplitude, optical image features, gas type / concentration, etc.) and the historical association distances between corresponding historical risk characteristic nodes and hidden risk nodes are extracted. Then, a clustering algorithm is used to aggregate the historical risk characteristic vectors into K aggregated feature vector sets, and the historical association distances corresponding to each set are aggregated into K distance sets (e.g., taking the maximum value or average value). Next, a prototype is constructed for each aggregated feature vector set by iteratively calculating the spatial density in M-dimensional space to determine the feature prototype (e.g., by randomly selecting spatial points, constructing subspaces, and verifying the iteration direction). Finally, each feature prototype and its corresponding maximum historical association distance (i.e., the prototype backtracking radius) are stored in the table, forming a mapping relationship between the aggregated historical risk characteristic vector prototype and the prototype backtracking radius. For example, the high-frequency vibration characteristic prototype is 50 meters, and the methane leakage characteristic prototype is 40 meters, providing historical data support for subsequent matching of backtracking radii based on real-time risk characteristic vectors.

[0072] The similarity calculation is performed between the currently acquired risk-representing feature vector (e.g., [sound frequency 2500Hz, amplitude 90dB, methane concentration 1.2%LEL]) and each aggregated historical risk-representing feature vector prototype in the backtracking radius matching table. First, the feature vectors are standardized (e.g., z-score standardization) to eliminate the influence of dimensions. Then, the cosine similarity algorithm is used to calculate the directional similarity between vectors, while Euclidean distance is combined to assess spatial distance. A weighted fusion is then used to form a comprehensive similarity index (e.g., cosine similarity weight 0.6, Euclidean distance weight 0.4). For high-dimensional feature vectors (e.g., vectors containing image features), principal component analysis (PCA) is used to reduce the dimensionality to within 30 dimensions before calculation. All prototypes in the matching table are traversed, and the maximum similarity value and corresponding prototype ID are recorded. If the maximum similarity exceeds a preset threshold (e.g., 0.8), the prototype is used as the matching result; otherwise, a fuzzy matching mechanism is activated to select the closest prototype and trigger a manual review process. For example, if the overall similarity between the current vector and the prototype of the pipeline vibration feature is 0.92, which exceeds the threshold, then the prototype is determined to be the matching result, and its corresponding prototype backtracking radius (such as 50 meters) is directly mapped as the associated backtracking radius to ensure that the intelligent prediction of the risk propagation range is consistent with historical patterns.

[0073] Based on the matching results, the corresponding prototype backtracking radius is retrieved from the pre-constructed backtracking radius matching table as the associated backtracking radius for the current risk analysis. Each aggregated historical risk feature vector prototype in the matching table is associated with a prototype backtracking radius based on historical data statistics (e.g., 50 meters for a high-frequency abnormal vibration prototype, 40 meters for a methane concentration exceeding the standard prototype). When the prototype with the highest similarity to the current risk feature vector is determined (e.g., a gas leak prototype is matched), the prototype backtracking radius corresponding to that prototype (e.g., 40 meters) is directly extracted as the spatial range threshold for subsequent association link extraction. This process transforms abstract feature vectors into specific physical distance parameters through the mapping relationship of historical risk data, providing a quantitative basis for accurately delineating the risk impact range, ensuring that the association link analysis covers potential risk propagation paths while avoiding meaningless range expansion, thus improving the efficiency and accuracy of risk positioning.

[0074] like Figure 6 As shown, in one possible implementation, step S341 further includes:

[0075] Step S3411: Obtain the historical utility tunnel risk log set.

[0076] Step S3412: Extract the historical association distance between historically represented risk nodes and historically implied risk nodes in the historical utility tunnel risk log set, as well as the historically represented risk feature vector, to obtain the historically represented risk feature vector set and the historical association distance set.

[0077] Step S3413: Aggregate the set of historical risk feature vectors of the same type to obtain K aggregated sets of historical risk feature vectors of the same type, where K is a positive integer.

[0078] Step S3414: Map and aggregate the historical association distance set according to the K aggregated historical characterization risk feature vector sets to obtain K aggregated historical association distance sets.

[0079] Step S3415: Prototype construction is performed on the set of K aggregated historical risk feature vectors to generate K aggregated historical risk feature vector prototypes.

[0080] Step S3416: Take the maximum value in the set of K aggregated historical association distances as the K prototype backtracking radii.

[0081] Step S3417: Construct a backtracking radius matching table based on the K aggregated historical risk feature vector prototypes and the K prototype backtracking radii.

[0082] Specifically, by connecting to the utility tunnel operation and maintenance database or historical monitoring system, a batch of historical utility tunnel risk logs are obtained. These logs cover various risk event data recorded during the operation of the utility tunnel, including but not limited to the occurrence time, sensor node location, monitoring parameters, and subsequent risk development of risk types such as acoustic anomalies, structural cracks, and gas leaks.

[0083] The historical utility tunnel risk log set is structured and analyzed: For each risk log, the spatial coordinates of historically representative risk nodes (i.e., sensor nodes where anomalies were first detected, such as sound sensor nodes corresponding to acoustic anomalies) and historically implied risk nodes (i.e., nodes subsequently confirmed as risk sources or related influences, such as optical sensor nodes corresponding to pipeline damage points) are first extracted. The Manhattan or Euclidean distance between the two points is calculated to obtain the historical association distance (e.g., 35 meters), and all distance values ​​are summarized into a historical association distance set. Simultaneously, the monitoring data corresponding to the historically representative risk nodes are extracted from the logs, and historically representative risk feature vectors are constructed according to sensor type: For sound sensors, sound frequency and amplitude values ​​are extracted (e.g., [2500Hz, 90dB]); for optical sensors, vector parameters after convolutional analysis of crack image features are extracted (e.g., [crack length 5cm, width 0.3mm]); for gas sensors, gas type and concentration values ​​are extracted (e.g., [methane, 1.8%LEL]). Finally, a set of historically representative risk feature vectors containing multi-dimensional parameters is formed. The above extraction process provides a crucial quantitative data foundation for subsequent risk feature clustering and backtracking radius calculation.

[0084] Clustering algorithms (such as K-means and DBSCAN) are used for unsupervised learning of the historical risk feature vector set to group vectors with similar feature parameters into the same category, thus aggregating risk features of the same type. The specific process is as follows: First, the multidimensional feature vectors are normalized (e.g., the range of parameters in each dimension is standardized) to eliminate dimensional differences. Then, by iteratively calculating the Euclidean distance or cosine similarity between vectors, vectors that are close in distance in the feature space are aggregated into K clusters (the value of K can be determined by the elbow rule or silhouette coefficient). Each cluster corresponds to an aggregated set of historical risk feature vectors. For example, samples with frequencies higher than 2000Hz and amplitudes greater than 85dB in all acoustic feature vectors are clustered into a high-frequency strong vibration cluster, and samples with methane concentrations exceeding 1% LEL in gas feature vectors are clustered into a gas leak cluster. Through this aggregation, massive historical risk feature data is abstracted into K representative categories, providing structured data support for subsequent prototype construction and backtracking radius calculation.

[0085] The generated set of K aggregated historical risk feature vectors is used as a mapping index to classify and aggregate the historical correlation distance set: Each historical correlation distance is traversed, and the distance value is assigned to the corresponding category based on the aggregation category to which its corresponding risk feature vector belongs. For example, if the feature vector corresponding to a historical correlation distance of 35 meters belongs to the high-frequency strong vibration cluster, then 35 meters is assigned to the distance set corresponding to that cluster. Through this mapping relationship, the original historical correlation distance set is divided into K independent aggregated historical correlation distance sets, each set corresponding to a specific type of risk feature (e.g., the high-frequency strong vibration cluster corresponds to the distance set {25 meters, 35 meters, 40 meters}, and the gas leak cluster corresponds to {30 meters, 45 meters, 50 meters}). This aggregation method realizes the correlation modeling between risk features and propagation distance, providing a classification and statistical basis for subsequently determining the tracing radius of different risk types, ensuring that the delineation of the tracing range matches the propagation characteristics of the risk type.

[0086] For each aggregated historical risk feature vector set, a prototype is constructed to generate a typical vector representing the characteristics of that category. For each set, an initial point is randomly selected in its feature space, and a subspace is constructed with a preset radius. The mean vector of all vectors in the subspace is calculated as the new center point. This process is iterated, and the final center point is determined by verifying the convergence of the iteration direction (e.g., the distance between the center points of two consecutive iterations is less than a threshold). This point is the feature prototype of the set. For example, for the feature vector set of a gas leak cluster, the prototype vector obtained through iterative calculation is [methane, 1.5% LEL, sound frequency 2500Hz, amplitude 90dB]. For high-dimensional feature vectors (such as image features), principal component analysis (PCA) is used to reduce the dimensionality to within 30 dimensions before prototype calculation to ensure computational efficiency. Finally, each aggregated set generates a unique feature prototype, forming K aggregated historical risk feature vector prototypes, providing a standardized reference template for subsequent risk matching.

[0087] For each aggregated historical correlation distance set, an extreme value extraction operation is performed: Iterate through K sets, and for each set (e.g., the distance set corresponding to a gas leak cluster {30 meters, 45 meters, 50 meters}), directly select the maximum value (e.g., 50 meters) as the prototype backtracking radius for that set. This operation is based on the principle of conservative estimation of risk propagation range, ensuring that the selected radius can cover the maximum impact range exhibited by this type of risk in historical cases. For example, if the historical correlation distance set for a structural vibration cluster is {25 meters, 35 meters, 40 meters}, then its prototype backtracking radius is set to 40 meters. In this way, K aggregated historical correlation distance sets are transformed into K prototype backtracking radii, providing a quantitative basis for matching backtracking radii based on real-time risk characteristics, ensuring that the spatial retrieval range during risk analysis is sufficient to cover potential associated risk nodes.

[0088] A structured backtracking radius matching table is constructed by associating K aggregated historical risk feature vector prototypes with their corresponding K prototype backtracking radii. Key-value pairs are established using the feature prototype as the key and the prototype backtracking radius as the value, forming a mapping relationship table between feature prototypes and backtracking radii. For example, a gas leak prototype vector [methane, 1.5% LEL, 2500Hz, 90dB] is associated with 50 meters. This table uses a multi-dimensional index structure to optimize query efficiency, setting composite indexes for feature parameters of different dimensions such as acoustics, optics, and gases, ensuring that real-time risk feature vectors can quickly match the corresponding prototypes. The final matching table is stored in an in-memory database (such as Redis) as the core retrieval basis for similarity matching, supporting risk backtracking radius queries with millisecond-level response times, providing precise spatial range definition for the extraction of associated links.

[0089] like Figure 7 As shown, in one possible implementation, step S3415 further includes:

[0090] Step S34151: Extract the first aggregated historical representation risk feature vector set from the K aggregated historical representation risk feature vector sets.

[0091] Step S34152: Map the first aggregated historical risk feature vector set to M-dimensional space to obtain a set of spatial points, where M is a positive integer and the coordinates of each spatial point correspond to a first aggregated historical risk feature vector.

[0092] Step S34153: Randomly select a first spatial point from the set of spatial points, and construct a first subspace according to a preset spatial radius.

[0093] Step S34154: Randomly select a spatial point at the edge of the first subspace as a direction verification spatial point, and construct a direction verification subspace based on the preset spatial radius.

[0094] Step S34155: Determine whether the spatial density of the first subspace is less than or equal to the spatial density of the directional verification subspace. If so, take the direction from the first spatial point to the directional verification spatial point as the iteration direction, and continue to iterate the directional verification spatial point according to the iteration direction until the preset number of iterations is met. Take the first aggregated historical representation risk feature vector corresponding to the spatial point obtained in the last iteration as the prototype of the first aggregated historical representation risk feature vector.

[0095] Step S34156: Traverse the set of K aggregated historical risk feature vectors to construct prototypes and obtain the prototypes of the K aggregated historical risk feature vectors.

[0096] Specifically, among the K aggregated historical risk feature vector sets generated by the clustering algorithm, the first set is extracted in a preset order (such as the natural order of clustering results, the priority order of risk types, etc.) and used as the target set for the current prototype construction. This set is denoted as the first aggregated historical risk feature vector set.

[0097] Each feature vector in the first aggregated historical risk feature vector set is converted into a coordinate point in M-dimensional space, forming a set of spatial points. The dimension of the feature vector (such as acoustic frequency, amplitude, gas type concentration, etc.) directly corresponds to the dimension M of the space (for example, if the feature vector contains three parameters: sound frequency, amplitude, and gas concentration, then M=3). The value of each parameter is standardized (e.g., normalized to the [0, 1] interval) and then used as the coordinate value of the spatial point in the corresponding dimension. For example, a feature vector of [sound frequency 2000Hz, amplitude 80dB, methane concentration 1.2%LEL] is converted into a point (0.6, 0.7, 0.8) in three-dimensional space after standardization. Through this mapping, the abstract risk feature parameters are transformed into a positional distribution in geometric space, which facilitates the subsequent determination of feature prototypes through spatial density analysis and provides an intuitive geometric model basis for the iterative calculation of prototype construction.

[0098] A point is randomly selected from the set of spatial points (i.e., the set of points mapped from the first aggregated historical risk feature vector set to M-dimensional space) as the first spatial point. The preset spatial radius is a pre-defined parameter used to define the extent of the subspace (e.g., it can be set to 10% of the maximum Euclidean distance after feature vector standardization). Centered on this spatial point and with the preset spatial radius as the range, a first subspace of spherical or hypercube shape is constructed in M-dimensional space. This subspace contains all spatial points whose distance from the first spatial point is less than or equal to the preset radius.

[0099] First, the edge region of the first subspace (i.e., the set of points whose distance from the first spatial point is equal to a preset spatial radius) is determined. Then, a spatial point is randomly selected from this edge region as the direction verification spatial point. Next, a direction verification subspace is constructed centered on this direction verification spatial point, using the same preset spatial radius. This subspace is also a spherical or hypercube structure and is used to evaluate the feature distribution density in the current direction. For example, if the center of the first subspace is (0.6, 0.7, 0.8), the preset radius is 0.2, and the coordinates of edge point A are (0.8, 0.7, 0.8), then the direction verification subspace is centered on point A, with a radius of 0.2, and includes all spatial points whose distance from point A is ≤0.2. This step, by selecting a verification point at the edge of the subspace and constructing a new subspace, provides a benchmark for subsequent comparison of spatial densities in different directions and determination of the iterative direction of the feature prototype, ensuring that the prototype converges to the region with the densest feature distribution.

[0100] The iteration direction of the feature prototype is dynamically adjusted by comparing the spatial density (i.e., the number of spatial points per unit volume) of the first subspace and the directional verification subspace: First, the density of the two subspaces is calculated: Spatial density = Number of points in the subspace / Subspace volume. If the density of the first subspace is less than or equal to the density of the directional verification subspace, it indicates that there is a denser feature distribution along the direction from the first spatial point to the directional verification spatial point. This direction is set as the iteration direction, and the process of constructing the subspace, selecting edge points, and calculating the density is repeated with the directional verification spatial point as the new center until the preset number of iterations (e.g., 10 times) is reached. Each iteration moves towards the region with higher density, ensuring that the prototype gradually approaches the core region of the feature distribution. For example, if the density of the first subspace is 5 points / unit volume and the density of the directional verification subspace is 8 points / unit volume in the first iteration, then the iteration direction is from the first spatial point to the directional verification spatial point, the new center is set as the verification point, the subspace is constructed again, and the judgment is repeated. When the iteration ends, the feature vector corresponding to the spatial point obtained in the last iteration is taken as the prototype of the first aggregated historical representation risk feature vector. If the density of the first subspace is greater than that of the direction verification subspace, the current direction is abandoned, and a direction verification space point is randomly selected again from the edge of the first subspace (retrying a maximum of a preset number of times, such as 5 times). If none of these conditions are met, the vector corresponding to the initial first space point is used directly as the prototype to ensure the robustness of the process. This process, through a density-guided iterative mechanism, enables the prototype to accurately reflect the core distribution of the aggregated feature set, improving the accuracy of subsequent risk matching.

[0101] After completing the prototype construction of the first aggregated historical risk feature vector set (i.e., the first set), the remaining K-1 aggregated sets are selected sequentially according to a preset order (such as the index order of clustering results, risk type priority, etc.), and the prototype construction process of steps S34151 to S34155 is repeated: for each set, its feature vectors are first mapped to an M-dimensional space to form a point set, and then the feature prototype of the set is determined by random point selection, subspace construction, density comparison, and iterative calculation. For example, if K=3, after completing the prototype construction of the gas leak category set, the same operation is performed on the structural vibration category set and the gas concentration anomaly category set in sequence. After traversal, a set containing K prototypes is generated, each prototype corresponding to the core feature vector of a cluster, such as the high-frequency strong vibration prototype, the methane leak prototype, etc. These prototypes constitute the core retrieval key of the backtracking radius matching table, ensuring that subsequent real-time risk feature vectors can be quickly matched to the corresponding prototypes and backtracking radii based on historical clustering rules, providing an efficient feature comparison benchmark for pipeline risk positioning under multi-protocol parallel drive.

[0102] In one possible implementation, step S3415 further includes:

[0103] When the spatial density of the first subspace is greater than the spatial density of the directional verification subspace, the directional verification spatial point is reselected, and the number of reselections is recorded. When the number of reselections is greater than or equal to the preset number of selections, the first aggregated historical representation risk feature vector corresponding to the first spatial point is used as the prototype of the first aggregated historical representation risk feature vector.

[0104] Specifically, if the spatial density of the first subspace is greater than that of the directional verification subspace, it indicates that the directional feature distribution of the currently selected directional verification point is sparser. This direction will be abandoned, and a new directional verification point will be randomly selected from the edge of the first subspace, with the number of reselections recorded. For example, if the preset number of selections is 5, and the initial comparison does not meet the condition, a new point will be selected, and the subspace will be reconstructed for density comparison. If, after 5 consecutive reselections, the density of all directional verification subspaces is still less than that of the first subspace, the first point is considered to be in a region of maximum local density, and its corresponding feature vector is directly used as the prototype of the first aggregated historical risk feature vector. This mechanism ensures that the algorithm can converge even with complex or scattered feature distributions by limiting the number of retries, avoiding infinite loops, while ensuring that the prototype construction results reflect the core features of the data as much as possible, thus improving the reliability of subsequent risk matching.

[0105] Based on the same inventive concept as the multi-protocol parallel-driven pipe gallery risk identification and location method in the foregoing embodiments, such as Figure 8 As shown, this application provides a multi-protocol parallel-driven utility tunnel risk identification and positioning system. The system and method embodiments in this application are based on the same inventive concept. The system includes:

[0106] The sensor topology network construction module 10 is used to acquire the pipe gallery structure diagram, deploy multi-source sensors in the pipe gallery based on the pipe gallery structure diagram, and construct a sensor topology network.

[0107] The topology generation module 20 is used to continuously monitor the pipe gallery using the sensor topology network and transmit the monitoring data to the transmission center of the sensor topology network through various communication protocols to generate a real-time sensor monitoring data topology network. The transmission center includes an intermediate flow and backtracking database.

[0108] The associated link set determination module 30 is used to identify risk characteristics based on the real-time sensor monitoring data topology network, determine risk characteristic nodes and risk characteristic vectors, perform associated link retrieval in the sensor topology network, and determine the target associated link set.

[0109] The backtracking data extraction module 40 is used to extract backtracking data from the intermediate flow backtracking database for each associated topology node in the target associated link set, and to determine the associated topology node backtracking data sequence set.

[0110] The hidden risk node set determination module 50 is used to perform serialized risk trend identification on the backtracking data sequence set of the associated topological nodes, determine the hidden risk node set, and use the representative risk node and the hidden risk node set as the risk identification and positioning result.

[0111] Furthermore, the system is also used for the following functions:

[0112] Based on the utility tunnel structure diagram, extract the utility tunnel route set and the utility tunnel route feature set; according to the utility tunnel route feature set, deploy multi-source sensors on each of the utility tunnel route sets to obtain an initial deployment sensor set; extract the route intersections in the utility tunnel route set, and deploy multi-source sensors at each route intersection to obtain an intersection deployment sensor set; perform a union operation on the initial deployment sensor set and the intersection deployment sensor set to obtain a deployment sensor set; treat each deployment sensor in the deployment sensor set as a topology node, and connect each topology node according to the route direction based on the utility tunnel route set to generate the sensor topology network.

[0113] Furthermore, the system is also used for the following functions:

[0114] Obtain risk identification rules; use the risk identification rules to identify risks in the real-time sensor monitoring data topology network, and obtain risk-characterizing nodes and risk-characterizing feature vectors.

[0115] Furthermore, the system is also used for the following functions:

[0116] Acoustic risk identification rules are obtained, wherein when the sensor type is a sound sensor, the sound frequency and amplitude detected by the sound sensor are extracted for anomaly identification. If an anomaly is identified, the topological node where the sound sensor is located is used as a risk characterization node, and the sound frequency and amplitude are embedded into an initially empty vector to construct a risk characterization feature vector. Optical risk identification rules are obtained, wherein when the sensor type is an optical sensor, image information detected by the optical sensor is extracted, and image feature convolution analysis is performed on the image information using a crack image feature recognizer to obtain an image feature vector. Based on the image feature vector... Anomaly identification is performed using feature vectors. If an anomaly is detected, the topological node where the optical sensor is located is designated as a risk characterization node, and the image feature vector is designated as a risk characterization feature vector. Gas risk identification rules are obtained, wherein when the sensor is a gas sensor, the gas type and gas concentration detected by the gas sensor are extracted for anomaly identification. If an anomaly is detected, the topological node where the gas sensor is located is designated as a risk characterization node, and the gas type and gas concentration are embedded into an initially empty vector to construct a risk characterization feature vector. The acoustic risk identification rules, optical risk identification rules, and gas risk identification rules are summarized to generate a risk characterization identification rule.

[0117] Furthermore, the system is also used for the following functions:

[0118] Based on the location of the risk-representing nodes, the associated utility tunnel routes are extracted from the sensor topology network to obtain a set of associated utility tunnel routes; backtracking radius matching is performed based on the risk-representing feature vectors to obtain the associated backtracking radius; and associated links are extracted from the set of associated utility tunnel routes based on the risk-representing nodes and the associated backtracking radius to determine the target set of associated links.

[0119] Furthermore, the system is also used for the following functions:

[0120] A backtracking radius matching table is pre-constructed; the risk characterization feature vector is matched with the aggregated historical risk characterization feature vector prototype in the backtracking radius matching table based on similarity, and the aggregated historical risk characterization feature vector prototype corresponding to the maximum similarity is taken as the matching result; the prototype backtracking radius of the matching result is extracted from the backtracking radius matching table as the associated backtracking radius.

[0121] Furthermore, the system is also used for the following functions:

[0122] Obtain a set of historical utility tunnel risk logs; extract the historical association distances between historically represented risk nodes and historically implied risk nodes in the historical utility tunnel risk log set, as well as the historically represented risk feature vectors, to obtain a set of historically represented risk feature vectors and a set of historical association distances; aggregate the historically represented risk feature vector sets to obtain K aggregated historically represented risk feature vector sets, where K is a positive integer; map and aggregate the historical association distance sets according to the K aggregated historically represented risk feature vector sets to obtain K aggregated historical association distance sets; construct prototypes for the K aggregated historically represented risk feature vector sets to generate K aggregated historically represented risk feature vector prototypes; use the maximum value in the K aggregated historical association distance sets as the K prototype backtracking radii; construct a backtracking radius matching table based on the K aggregated historically represented risk feature vector prototypes and the K prototype backtracking radii.

[0123] Furthermore, the system is also used for the following functions:

[0124] Extract a first aggregated historical representation risk feature vector set from the K aggregated historical representation risk feature vector sets; map the first aggregated historical representation risk feature vector set to an M-dimensional space to obtain a set of spatial points, where M is a positive integer, and the coordinates of each spatial point correspond to a first aggregated historical representation risk feature vector; randomly select a first spatial point from the set of spatial points, and construct a first subspace according to a preset spatial radius; randomly select a spatial point on the edge of the first subspace as a direction verification spatial point, and construct a direction verification subspace based on the preset spatial radius; determine whether the spatial density of the first subspace is less than or equal to the spatial density of the direction verification subspace; if so, take the direction from the first spatial point to the direction verification spatial point as the iteration direction, and continue to iterate the direction verification spatial point according to the iteration direction until the preset number of iterations is met, and take the first aggregated historical representation risk feature vector corresponding to the spatial point obtained in the last iteration as the prototype of the first aggregated historical representation risk feature vector; traverse the K aggregated historical representation risk feature vector sets to construct the prototype, and obtain the K aggregated historical representation risk feature vector prototypes.

[0125] Furthermore, the system is also used for the following functions:

[0126] When the spatial density of the first subspace is greater than the spatial density of the directional verification subspace, the directional verification spatial point is reselected, and the number of reselections is recorded. When the number of reselections is greater than or equal to the preset number of selections, the first aggregated historical representation risk feature vector corresponding to the first spatial point is used as the prototype of the first aggregated historical representation risk feature vector.

[0127] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0128] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0129] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A method for risk identification and location of utility tunnels under multi-protocol parallel driving, characterized in that, The method includes: Obtain the structure diagram of the utility tunnel, and deploy multi-source sensors within the utility tunnel based on the structure diagram to construct a sensor topology network; The sensor topology network is used for continuous monitoring of the pipe gallery, and the monitoring data is transmitted to the transmission center of the sensor topology network through multiple communication protocols to generate a real-time sensor monitoring data topology network. The transmission center includes an intermediate flow and backtracking database. Based on the real-time sensor monitoring data topology network, risk identification is performed to determine risk nodes and risk feature vectors. Then, backtracking radius matching is performed based on the risk feature vectors to obtain the associated backtracking radius. Based on the risk nodes and associated backtracking radii, associated link retrieval is performed in the sensor topology network to determine the target associated link set. For each associated topology node in the target associated link set, backtracking data is extracted from the intermediate flow backtracking database to determine the associated topology node backtracking data sequence set; The sequenced risk trend identification is performed on the backtracking data sequence set of the associated topological nodes to determine the set of hidden risk nodes. The risk-characterizing nodes and the set of hidden risk nodes are used as the risk identification and positioning results. The step of matching the backtracking radius based on the risk characteristic vector to obtain the associated backtracking radius includes: Obtain a collection of historical utility tunnel risk logs; The historical association distance between historically represented risk nodes and historically implied risk nodes in the historical utility tunnel risk log set, as well as the historically represented risk feature vector, are extracted to obtain the historically represented risk feature vector set and the historical association distance set. The historical risk feature vector set is aggregated to obtain K aggregated historical risk feature vector sets, where K is a positive integer; Based on the K aggregated historical risk feature vector sets, the historical association distance sets are mapped and aggregated to obtain K aggregated historical association distance sets; Prototype construction is performed on the set of K aggregated historical risk feature vectors to generate K aggregated historical risk feature vector prototypes; The maximum value among the K aggregated historical association distance sets is used as the K prototype backtracking radii; A backtracking radius matching table is constructed based on the K aggregated historical risk feature vector prototypes and the K prototype backtracking radii. The risk characterization feature vector is matched with the aggregated historical risk characterization feature vector prototype in the backtracking radius matching table, and the aggregated historical risk characterization feature vector prototype corresponding to the maximum similarity is taken as the matching result. The prototype backtrack radius of the matching result is extracted from the backtrack radius matching table and used as the associated backtrack radius.

2. The method for risk identification and location of utility tunnels under multi-protocol parallel driving as described in claim 1, characterized in that, Obtain a structural diagram of the utility tunnel, and based on this diagram, deploy multi-source sensors within the tunnel to construct a sensor topology network, including: Based on the aforementioned utility tunnel structure diagram, extract the utility tunnel route set and the utility tunnel route feature set; Based on the feature set of the utility tunnel route, multi-source sensors are deployed for each of the utility tunnel route sets to obtain an initial set of deployed sensors. Extract the route intersections from the set of utility tunnel routes, and deploy multi-source sensors at each route intersection to obtain the set of intersecting sensors. The initial set of deployed sensors and the cross-deployed set of sensors are combined to obtain the set of deployed sensors. Each sensor in the set of deployed sensors is treated as a topology node. Each topology node is connected according to the route of the utility tunnel to generate the sensor topology network.

3. The method for risk identification and location of utility tunnels under multi-protocol parallel driving as described in claim 1, characterized in that, Based on the real-time sensor monitoring data topology network, risk identification is performed to determine risk-representing nodes and risk-representing feature vectors, including: Obtain the rules for identifying risk characteristics; Risk identification is performed on the real-time sensor monitoring data topology network using risk characterization rules to obtain risk characterization nodes and risk characterization feature vectors.

4. The method for risk identification and location of utility tunnels under multi-protocol parallel driving as described in claim 3, characterized in that, Obtain the risk identification rules, including: Acquire acoustic risk identification rules, wherein when the sensor type is a sound sensor, extract the sound frequency and sound amplitude detected by the sound sensor for anomaly identification. If an anomaly is identified, use the topology node where the sound sensor is located as a risk characterization node, embed the sound frequency and sound amplitude into an initially empty vector, and construct a risk characterization feature vector. Obtain optical risk identification rules, wherein when the sensor type is an optical sensor, extract the image information monitored by the optical sensor, use a crack image feature recognizer to perform image feature convolution analysis on the image information to obtain an image feature vector, perform anomaly identification based on the image feature vector, and if an anomaly is identified, use the topological node where the optical sensor is located as a risk characterization node and use the image feature vector as a risk characterization feature vector. Obtain gas risk identification rules, wherein the gas risk identification rules are as follows: when the sensor is a gas sensor, extract the gas type and gas concentration detected by the gas sensor for anomaly identification; if an anomaly is identified, use the topology node where the gas sensor is located as a risk characterization node, embed the gas type and gas concentration into an initially empty vector, and construct a risk characterization feature vector. The acoustic risk identification rules, optical risk identification rules, and gas risk identification rules are summarized to generate risk identification rules that characterize the risk.

5. A method for risk identification and location of utility tunnels under multi-protocol parallel drive as described in any one of claims 1-4, characterized in that, Based on the identified risk nodes and associated backtracking radii, association link retrieval is performed in the sensor topology network to determine the target association link set, including: Based on the location of the risk-representing nodes, the associated utility tunnel routes are extracted from the sensor topology network to obtain a set of associated utility tunnel routes; based on the risk-representing nodes and the associated backtracking radius, the associated links are extracted from the set of associated utility tunnel routes to determine the target set of associated links.

6. A method for risk identification and location of utility tunnels under multi-protocol parallel drive as described in any one of claims 1-5, characterized in that, Prototype construction is performed on the set of K aggregated historical risk feature vectors to generate K aggregated historical risk feature vector prototypes, including: Extract the first aggregated historical representation risk feature vector set from the K aggregated historical representation risk feature vector sets; The first aggregated historical risk feature vector set is mapped to an M-dimensional space to obtain a set of spatial points, where M is a positive integer, and the coordinates of each spatial point correspond to a first aggregated historical risk feature vector. A first spatial point is randomly selected from the set of spatial points, and a first subspace is constructed according to a preset spatial radius; A spatial point is randomly selected from the edge of the first subspace as a direction verification spatial point, and a direction verification subspace is constructed based on the preset spatial radius; Determine whether the spatial density of the first subspace is less than or equal to the spatial density of the directional verification subspace. If so, take the direction from the first spatial point to the directional verification spatial point as the iteration direction, and continue to iterate the directional verification spatial point according to the iteration direction until the preset number of iterations is met. Take the first aggregated historical representation risk feature vector corresponding to the spatial point obtained in the last iteration as the prototype of the first aggregated historical representation risk feature vector. The prototypes of the K aggregated historical risk feature vector sets are constructed by traversing the set of K aggregated historical risk feature vectors.

7. The method for risk identification and location of utility tunnels under multi-protocol parallel driving as described in claim 6, characterized in that, When the spatial density of the first subspace is greater than the spatial density of the directional verification subspace, the directional verification spatial point is reselected, and the number of reselections is recorded. When the number of reselections is greater than or equal to the preset number of selections, the first aggregated historical representation risk feature vector corresponding to the first spatial point is used as the prototype of the first aggregated historical representation risk feature vector.

8. A multi-protocol parallel-driven utility tunnel risk identification and positioning system, characterized in that, The system is used to implement the multi-protocol parallel drive method for identifying and locating risks in utility tunnels as described in any one of claims 1-7, and the system includes: The sensor topology network construction module is used to acquire the pipe gallery structure diagram, deploy multi-source sensors in the pipe gallery based on the pipe gallery structure diagram, and construct a sensor topology network. The topology generation module is used to continuously monitor the pipe gallery using the sensor topology network and transmit the monitoring data to the transmission center of the sensor topology network through multiple communication protocols to generate a real-time sensor monitoring data topology network. The transmission center includes an intermediate flow and backtracking database. The associated link set determination module is used to identify the risk characteristics based on the real-time sensor monitoring data topology network, determine the risk characteristics nodes and risk characteristics feature vectors, perform backtracking radius matching based on the risk characteristics feature vectors to obtain the associated backtracking radius, and perform associated link retrieval in the sensor topology network based on the risk characteristics nodes and associated backtracking radius to determine the target associated link set. The backtracking data extraction module is used to extract backtracking data from the intermediate flow backtracking database for each associated topology node in the target associated link set, and determine the backtracking data sequence set of associated topology nodes; The hidden risk node set determination module is used to perform serialized risk trend identification on the backtracking data sequence set of the associated topological nodes, determine the hidden risk node set, and use the representative risk nodes and the hidden risk node set as the risk identification and positioning results. The associated link set determination module is further configured to: Obtain a collection of historical utility tunnel risk logs; The historical association distance between historically represented risk nodes and historically implied risk nodes in the historical utility tunnel risk log set, as well as the historically represented risk feature vector, are extracted to obtain the historically represented risk feature vector set and the historical association distance set. The historical risk feature vector set is aggregated to obtain K aggregated historical risk feature vector sets, where K is a positive integer; Based on the K aggregated historical risk feature vector sets, the historical association distance sets are mapped and aggregated to obtain K aggregated historical association distance sets; Prototype construction is performed on the set of K aggregated historical risk feature vectors to generate K aggregated historical risk feature vector prototypes; The maximum value among the K aggregated historical association distance sets is used as the K prototype backtracking radii; A backtracking radius matching table is constructed based on the K aggregated historical risk feature vector prototypes and the K prototype backtracking radii. The risk characterization feature vector is matched with the aggregated historical risk characterization feature vector prototype in the backtracking radius matching table, and the aggregated historical risk characterization feature vector prototype corresponding to the maximum similarity is taken as the matching result. The prototype backtrack radius of the matching result is extracted from the backtrack radius matching table and used as the associated backtrack radius.

Images