Graded quantitative evaluation method and system for underground pipe gallery pipe network risk

Abstract

The invention relates to the technical field of quantitative evaluation, in particular to a graded quantitative evaluation method and system for underground pipe gallery pipe network risks. The method comprises the following steps: planning an open network based on pipeline topology and front-end sensing topology; deploying a first compression node by using a low-rank matrix, deploying a second decoupling node by using an explicit feature-implicit feature, and determining preprocessors and deploying the preprocessors at opening positions; according to the open network, executing dynamic sampling and open return, triggering a pre-processor to execute low-rank compression and feature decoupling processing of distributed data, synthesizing a feature coding sequence and returning the feature coding sequence to a rear-end processor, activating a risk antibody module to execute targeted alignment of a feature coding sequence-defense coding structure, and determining risk quantitative data; the technical problem that in the prior art, underground pipe gallery pipe network risk assessment lacks a refined grade division and quantitative assessment mechanism is solved, and the technical effects of underground pipe gallery pipe network risk type identification and magnitude quantitative output are achieved.

Description

Technical Field

[0001] This invention relates to the field of quantitative assessment technology, and in particular to a graded quantitative assessment method and system for the risk of underground utility tunnels and networks. Background Technology

[0002] As a crucial component of urban infrastructure, underground utility tunnels directly impact the safety and stability of urban pipe networks. Current risk assessments for underground utility tunnels largely rely on manual inspections, single-point sensor data collection, or static evaluation models, failing to cover a wide range of multi-dimensional risk factors. Due to the complex structure of underground utility tunnels and the highly dynamic and spatially heterogeneous operating environment, traditional assessment methods often struggle to fully reflect the spatial distribution and evolution trends of risks. Furthermore, assessment results are primarily qualitative, lacking the ability to finely classify and quantify risk levels, thus hindering the provision of clear decision-making support for risk response and emergency management. Summary of the Invention

[0003] Therefore, the technical problem to be solved by the present invention is to overcome the lack of refined classification and quantitative assessment mechanism for risk assessment of underground utility tunnels and networks in the prior art.

[0004] To address the aforementioned technical problems, this invention provides a method for the graded and quantitative assessment of risks in underground utility tunnel networks, comprising: Construct an open network for underground utility tunnel areas based on pipeline topology and front-end sensor topology, where the open locations are data sensor aggregation points; A preprocessor is deployed at each open location, the preprocessor comprising a first compressed node deployed in a low-rank matrix and a second decoupled node deployed in an explicit feature-implicit feature configuration. Dynamic sampling and open backhaul are performed based on the open network, triggering the preprocessor to perform low-rank compression and feature decoupling processing of distributed data, and synthesizing feature encoding sequences; The feature-encoded sequence is sent back to the back-end processor, and the risk antibody module with built-in risk defense structure is activated to perform targeted alignment of the feature-encoded sequence and the defense-encoded structure to determine the risk quantification data.

[0005] Preferably, the construction of the underground utility tunnel area open network based on pipeline topology and front-end sensor topology includes: Set a preset utility tunnel range, divide the underground utility tunnel area into regions based on the preset utility tunnel range, and determine the utility tunnel region data; The locations of the openings are determined based on the pipe gallery segmentation data, wherein the pipe gallery segmentation data and the opening locations correspond one-to-one. The open network is determined based on the pipeline topology, the front-end sensing topology, and the open location.

[0006] Preferably, a first encoding mode is introduced and built into the output port of the preprocessor.

[0007] Preferably, the step of performing dynamic sampling and open-loop feedback based on the open network includes: Using preset sampling rules, the front-end sensing topology is driven to perform multi-source sensing acquisition in the pipe gallery and determine multi-source sensing data; Based on the correspondence between the pipe gallery domain data and the opening location, the multi-source sensor data is transmitted back to the opening location.

[0008] Preferably, the step of triggering the preprocessor to perform low-rank compression and feature decoupling processing of distributed data, and synthesizing the feature encoding sequence includes: The first compression node of the preprocessor is triggered to perform low-rank compression processing on the multi-source sensor data and determine the first compressed data. The first compressed data stream is transferred to the second decoupling node, where a binary classification process of explicit features and implicit features is performed and an identifier is added to determine the second decoupling data. At the output port, the second decoupled data is encoded and converted according to the first encoding mode to determine the feature encoding sequence.

[0009] Preferably, a second encoding mode is introduced to encode the risk defense structure as the defense encoding structure, wherein the risk defense structure and the defense encoding structure are associated with a mapping relationship. The risk antibody module is trained under supervision using the defense coding structure and the coding target trigger based on the first coding mode.

[0010] Preferably, the construction of the risk defense structure includes: Obtain records of risk events in the utility tunnel, perform clustering processing under the same scenario, and determine N groups of record clusters; Traverse the N sets of record clusters to discover risk defense patterns, wherein the risk defense patterns are universal patterns within the clusters; The risk defense model is structurally integrated as the risk defense structure.

[0011] Preferably, the step of sending the feature-encoded sequence back to the back-end processor and activating the risk antibody module with a built-in risk defense structure to perform targeted alignment of the feature-encoded sequence and the defense-encoded structure to determine the risk quantification data includes: According to the network communication protocol, the feature coding sequence is sent back to the back-end processor, the risk antibody module built into the back-end processor is activated, the targeted alignment of the feature coding sequence-defense coding structure is performed, and the targeted defense coding node is determined. Based on the second encoding mode, the risk defense mode is determined and the risk type is deduced by interpreting the targeted defense encoding node, and the risk level is determined based on the risk defense level. The risk type and risk level are associated and distributed integration based on spatiotemporal codes is performed to determine risk quantification data, wherein the spatiotemporal codes include time codes and spatial codes.

[0012] Preferably, after determining the risk defense mode, the method further includes: The aforementioned risk defense models are integrated and stored in a temporary contingency plan database; By triggering the aforementioned temporary contingency plan database, risk emergency management of the underground utility tunnel area is implemented.

[0013] This invention also provides a graded and quantitative assessment system for the risks of underground utility tunnel networks, comprising: The open network construction module constructs an open network for underground utility tunnel areas based on pipeline topology and front-end sensor topology, where the open locations are data sensor aggregation points; A deployment module is used to deploy a preprocessor at each open location, the preprocessor including a first compressed node deployed in a low-rank matrix and a second decoupled node deployed in an explicit feature-implicit feature configuration. The feature encoding module is used to perform dynamic sampling and open backhaul based on the open network, trigger the preprocessor to perform low-rank compression and feature decoupling processing of distributed data, and synthesize a feature encoding sequence. The risk quantification module is used to send the feature-encoded sequence back to the back-end processor and activate the risk antibody module with built-in risk defense structure to perform targeted alignment of the feature-encoded sequence and the defense encoding structure to determine the risk quantification data.

[0014] The technical solution of the present invention has the following advantages over the prior art: The present invention describes a method for hierarchical and quantitative risk assessment of underground utility tunnel networks. For underground utility tunnel areas, an open network based on pipeline topology and front-end sensor topology is planned, where the open locations serve as data sensing convergence points. A first compression node is deployed using a low-rank matrix, and a second decoupling node is deployed using explicit and implicit features. A preprocessor is determined and deployed at each open location. Dynamic sampling and open-end backhaul are performed based on the open network, triggering the preprocessor to perform low-rank compression and feature decoupling processing of distributed data, synthesizing a feature encoding sequence and transmitting it back to the back-end processor. The risk antibody module is activated to perform targeted alignment of the feature encoding sequence and defense encoding structure to determine the risk quantification data. The targeted defense encoding nodes of the targeted alignment determine the risk type and risk level, and the risk antibody module has a built-in risk defense structure. This invention solves the technical problem of the lack of refined hierarchical classification and quantitative assessment mechanisms in existing underground utility tunnel network risk assessments. By constructing an open network and performing feature encoding and targeted alignment processing, it achieves the technical effect of identifying the type and quantifying the level of risk in underground utility tunnel networks. Attached Figure Description

[0015] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings, wherein: Figure 1 This is a flowchart illustrating the implementation of a risk classification and quantitative assessment method for underground utility tunnel networks provided by this invention. Figure 2 This is a structural block diagram of a risk classification and quantitative assessment device for underground utility tunnel networks provided in an embodiment of the present invention. Detailed Implementation

[0016] The core of this invention is to provide a hierarchical and quantitative assessment method and system for underground utility tunnel network risks. By constructing an open network and performing feature encoding and targeted alignment processing, the technical effect of identifying the type and quantifying the magnitude of underground utility tunnel network risks is achieved.

[0017] To enable those skilled in the art to better understand the present invention, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please refer to Figure 1. Figure 1 The flowchart illustrates the implementation of a risk classification and quantitative assessment method for underground utility tunnel networks provided by this invention; the specific operation steps are as follows: S101: Construct an open network of underground utility tunnel areas based on pipeline topology and front-end sensor topology, where the open locations are data sensor aggregation points; S102: Deploy a preprocessor at each open location, the preprocessor comprising a first compressed node deployed in a low-rank matrix and a second decoupled node deployed in an explicit feature-latent feature configuration; S103: Perform dynamic sampling and open backhaul according to the open network, trigger the preprocessor to perform low-rank compression and feature decoupling processing of distributed data, and synthesize feature coding sequences; S104: The feature coding sequence is sent back to the back-end processor, and the risk antibody module with built-in risk defense structure is activated to perform targeted alignment of feature coding sequence and defense coding structure to determine risk quantification data.

[0019] Based on the above embodiments, this embodiment will provide a detailed description of step S101: In some embodiments, constructing an open network for underground utility tunnel areas based on pipeline topology and front-end sensor topology includes: Set a preset utility tunnel range, divide the underground utility tunnel area into regions based on the preset utility tunnel range, and determine the utility tunnel region data; The locations of the openings are determined based on the pipe gallery segmentation data, wherein the pipe gallery segmentation data and the opening locations correspond one-to-one. The open network is determined based on the pipeline topology, the front-end sensing topology, and the open location.

[0020] It should be noted that in these embodiments, for the underground utility tunnel area, a preset tunnel range is first set, and the target area is spatially divided based on this range to generate tunnel domain data. Subsequently, based on the tunnel domain data, corresponding open locations are deployed in each domain. These open locations serve as data sensing aggregation points, used to collect multi-source sensor data and trigger preprocessing mechanisms. After the open location deployment is completed, the pipeline topology inside the underground utility tunnel (describing the physical connection relationships of various pipelines in the underground utility tunnel network) and the deployed front-end sensor topology (describing the spatial layout and communication paths of multimodal sensors) are integrated. The above elements are jointly modeled with the open locations to finally determine the open network for full-area perception and data backhaul.

[0021] Furthermore, in some specific embodiments: First, the GIS spatial modeling method is used to set the preset scope of the utility tunnel, that is, to delineate the target area covered by this risk assessment in the map system. This scope is defined based on the administrative boundaries of the underground utility tunnel, engineering drawings or existing BIM models, and digital boundaries are formed by two-dimensional or three-dimensional coordinate annotation. Through this process, the preset scope of the utility tunnel is determined. Within the designated pre-defined utility tunnel area, a Voronoi diagram-based regional partitioning algorithm is used to segment the underground utility tunnel region. This method uses the pipeline nodes or intersections of the tunnel as the generation center and constructs a distance-based partitioning structure to divide the entire underground utility tunnel into multiple non-overlapping, fully covered regional units, thus obtaining the utility tunnel partitioning data. Each partitioning data includes its corresponding region number, information on the covered pipeline segments, functional attributes (such as power compartment, heating compartment), and subsequent sensor deployment requirements.

[0022] After obtaining the segmented data of the utility tunnel, a shortest path-redundancy optimization point placement method is adopted to calculate the point placement within each segment and select the optimal point as the open location of that segment. This method constructs a set of paths starting from the segment center and connecting to the main communication node (such as a convergence switch or main control PLC), calculates the communication cost, sensor visibility overlap, and power consumption of each candidate point, and finally determines the point with the shortest path, most complete coverage, and best energy efficiency as the data sensing aggregation point. This point placement strategy ensures a one-to-one correspondence between the open location and the segmented data of the utility tunnel, and has stable data acquisition and transmission capabilities. Finally, based on the topology fusion modeling method, the pipeline topology and the front-end sensing topology in the underground utility tunnel are integrated and modeled, and the deployed open locations are connected as core nodes to ultimately construct a complete open network. Specifically, the pipeline topology is converted from CAD drawings into a graph representation, reflecting the spatial relationships and logical dependencies of various pipelines; the front-end sensing topology collects node distribution and communication links of protocols such as LoRa, NB-IoT, or Ethernet to form a sensing network graph; the two are then fused with the open nodes to form a multi-level, heterogeneous open network.

[0023] Based on the above embodiments, this embodiment will provide a detailed description of step S102: It should be noted that, in order to achieve efficient preprocessing of sensor data at open locations, a preprocessor with compression and feature separation capabilities is deployed. This preprocessor consists of a first compression node and a second decoupling node, and its functions are configured based on low-rank matrix processing and explicit feature-latent feature partitioning methods.

[0024] Specifically, the first compression node is deployed as a low-rank matrix, which is used to compress data collected by multiple sensors. The operation process involves constructing a multi-dimensional observation matrix using various raw data collected from open locations. Rows in this matrix represent sampling times, and columns represent various sensor channels (such as temperature, humidity, vibration, gas concentration, etc.). Then, a rank estimation method (such as singular value analysis) is used to determine the approximate rank of this matrix, retaining only the dominant information components and removing redundant noise components, outputting the compressed low-rank data. This processing reduces data dimensionality and communication load, enabling compressed transmission and rapid processing of data at the edge.

[0025] Building upon this foundation, a second decoupling node is deployed using an explicit feature-implicit feature partitioning approach to perform feature classification processing on the compressed data. This node distinguishes between two types of features: explicit features that can be directly identified from the compressed data, such as numerical limits exceeding limits, abrupt changes, and physical threshold breaches; and implicit features that require multi-channel information fusion or time series analysis, such as trend accumulation, periodic changes, and superposition of abnormal patterns. The second decoupling node completes the decoupling and separation of these two types of features and outputs their identification through a pre-defined feature recognition mechanism, resulting in a structured feature set.

[0026] After completing the deployment of the two types of nodes, the first compression node and the second decoupling node are integrated into a single preprocessor unit and installed in various open locations to achieve local compression and feature structure extraction of the original multimodal sensor data, thereby providing a unified and efficient feature encoding input for the backend risk analysis module.

[0027] In some embodiments, a first encoding mode is introduced and built into the output port of the preprocessor.

[0028] It should be noted that, to achieve unified formatting and recognizable transmission of the preprocessor's output data, a first encoding mode is introduced and built into the preprocessor's output port. Specifically, the first encoding mode is an encoding method used to structurally encapsulate the decoupled feature data, possessing field structures such as feature classification identifiers, numerical hierarchy labels, time indexes, and spatial numbers. After the preprocessor completes low-rank compression and explicit-implicit feature decoupling processing of the multi-source sensor data, the system calls the built-in first encoding mode to encode and transform the generated feature results, giving them a unified data structure and format, and outputting a standardized feature encoding sequence. This encoding method ensures consistent expression of decoupled features across different exposed nodes.

[0029] Based on the above embodiments, this embodiment will provide a detailed description of step S103: In some embodiments, performing dynamic sampling and open-loop backhaul based on the open network includes: Using preset sampling rules, the front-end sensing topology is driven to perform multi-source sensing acquisition in the pipe gallery and determine multi-source sensing data; Based on the correspondence between the pipe gallery domain data and the opening location, the multi-source sensor data is transmitted back to the opening location.

[0030] It should be noted that, based on the established open network, the front-end sensor topology deployed in each utility tunnel segment is first driven by preset sampling rules to perform dynamic acquisition tasks of multi-dimensional environmental and structural states in the target area. The acquired data includes temperature, humidity, gas concentration, displacement changes, stress response, etc., generating structured multi-source sensor data. After acquisition, according to the one-to-one correspondence between the utility tunnel segment data and the open location, the data is uploaded to the corresponding open location through the regional communication link to complete the data backhaul process, ensuring that all types of sensing data are spatially logically consistent and structurally clear.

[0031] Furthermore, in some specific embodiments: Firstly, a method combining fixed-period and event threshold settings is employed to establish preset sampling rules, which control the frequency and timing of sensor sampling. Specifically, fixed-period sampling is used to acquire routine operational status data, such as collecting temperature, humidity, or gas concentration data every 5 minutes; event threshold sampling is used to respond to sudden anomalies, such as immediately triggering encrypted sampling when the gas concentration exceeds a preset safety threshold, thereby improving the ability to detect sudden risks. This sampling rule is configured in synchronous or asynchronous mode, depending on the monitoring task requirements, and is used to drive the front-end sensing topology to execute sampling tasks, determining the triggering conditions and sampling time windows for multi-source sensor data.

[0032] Subsequently, the front-end sensing topology is driven via a command-based approach. This topology consists of various types of sensors deployed within each pipe gallery area, including temperature and humidity sensors, gas sensors, vibration sensors, and displacement sensors. These sensors are connected to the local controller via bus or wireless communication. Based on sampling rules, they send acquisition commands to trigger each sensor node to initiate its sampling program, completing the data acquisition of corresponding physical quantities to obtain multiple types of physical parameters and determine the raw measurement results of the multi-source sensing data. After sampling, a local cache extraction method is used to organize and summarize the multi-source sensing data collected from the pipe gallery. Each type of sensor stores the collected data in its local cache, indexed by a timestamp and sensor number. The controller then extracts and generates the multi-source sensing data.

[0033] Finally, a static mapping table method is used to transmit multi-source sensor data back to the corresponding open locations. A one-to-one correspondence between the pipe gallery's domain data and open locations is predefined, with each domain bound to a specific open number. The domain identifier attached to the data is read, and the data is accurately assigned to the corresponding open location according to the mapping table. The data transmission process is then completed via a local communication link (such as LoRa, RS485, or NB-IoT), thus completing the data transmission of multi-source sensor data to the open locations.

[0034] In some embodiments, triggering the preprocessor to perform low-rank compression and feature decoupling processing of distributed data, and synthesizing a feature encoding sequence includes: The first compression node of the preprocessor is triggered to perform low-rank compression processing on the multi-source sensor data and determine the first compressed data. The first compressed data stream is transferred to the second decoupling node, where a binary classification process of explicit features and implicit features is performed and an identifier is added to determine the second decoupling data. At the output port, the second decoupled data is encoded and converted according to the first encoding mode to determine the feature encoding sequence.

[0035] It should be noted that the returned multi-source sensor data will trigger the preprocessor at the open position to execute the data compression and feature recognition process. First, the first compression node within the preprocessor is activated, using a low-rank matrix compression method to perform rank estimation and principal component extraction on the multi-source sensor data, preserving key feature information and outputting the first compressed data after dimensionality reduction. Subsequently, this compressed data is passed to the second decoupling node. This node, based on whether it has direct observability, divides the structural information in the data into explicit and implicit features, and adds corresponding feature identifiers to each type of feature, obtaining the second decoupling data with clear classification. At the output port of the preprocessor, according to a preset first encoding mode, the decoupling data undergoes structural compression and encoding encapsulation, outputting a feature encoding sequence in a unified format.

[0036] Furthermore, in some specific embodiments: First, the first compression node of the preprocessor is triggered, employing Singular Value Decomposition (SVD) to perform low-rank compression processing on the multi-source sensor data. Specifically, the time-series data collected by various types of sensors (such as temperature, humidity, gas concentration, displacement, etc.) are constructed into a two-dimensional observation matrix, with rows representing time points and columns representing sensor channels. This matrix is ​​decomposed using SVD to extract the most representative top k singular values ​​and eigenvectors, thereby reconstructing an approximate matrix with a rank much smaller than the original matrix. This process removes redundancy and noise, reduces data dimensionality, and retains dominant information features, outputting sparse and change-sensitive first-compressed data.

[0037] Next, the first compressed data stream is transferred to the second decoupling node, where a binary classification method based on observability—explicit features versus implicit features—is used for feature decoupling. This method uses whether a feature can be directly observed as the criterion. If a feature can be directly identified by a single-variable abrupt change (e.g., temperature > 70℃, vibration > 2 mm / s), it is classified as an explicit feature; if it requires multi-channel joint analysis or trend regression (e.g., gradual accumulation of gas concentration, resonance response of different sensors), it is classified as an implicit feature. In the processing flow, the node classifies the features according to a preset feature type rule base and attaches a corresponding feature identifier to each feature, including a category code (e.g., E / I), channel number, sampling timestamp, and open location code, generating structured second decoupling data.

[0038] Finally, the pre-defined first encoding mode is invoked at the output port of the preprocessor, and the second decoupled data is encoded and converted using a field concatenation and index mapping method. This method is based on a field sequence, concatenating the information in the feature identifier in sequence according to a set format, such as [feature type] + [channel ID] + [location ID] + [time code] + [numerical level], and establishing a fast index relationship for the backend recognition structure through hash mapping, ultimately generating a set of standardized feature encoding sequences.

[0039] Based on the above embodiments, this embodiment will provide a detailed description of step S104: In some embodiments, a second encoding mode is introduced to encode the risk defense structure as the defense encoding structure, wherein the risk defense structure and the defense encoding structure are associated with a mapping relationship. The risk antibody module is trained under supervision using the defense coding structure and the coding target trigger based on the first coding mode.

[0040] It should be noted that, in order to achieve accurate alignment between the front-end feature recognition results and the back-end risk response strategy, a second encoding mode is introduced to perform structured expression and standardization of the existing risk defense structure, generate a defense encoding structure that can be used for target matching, and enhance the recognition and response capabilities of the risk antibody module through encoding mapping and supervised training.

[0041] Specifically, a second coding mode is first introduced, employing a field grouping coding method to encode the risk defense structure. The risk defense structure is a set of emergency response strategies extracted from historical risk events in underground utility tunnels, containing key information such as "risk type," "trigger threshold," "response level," and "response methods." Through preset coding rules, this structured information is mapped to semantically meaningful coded fields. For example, "gas leak" → [RT01], "Level II response" → [RL2], "forced ventilation + audible and visual alarm" → [DP07]. These fields are concatenated sequentially to generate a complete coded string, such as [RT01][RL2][DP07], forming a unified format for the defense coding structure.

[0042] Subsequently, a mapping relationship is established between the risk defense structure and the defense coding structure. A hash index mapping method is used, with the defense coding structure as the key and the risk defense structure it points to as the value, forming a one-to-one correspondence table. This mapping ensures that each set of codes is not only unique but also allows for backtracking to the original defense scheme after matching, achieving interpretability and traceability of risk response, and constructing a well-structured defense strategy index system.

[0043] Next, the risk antibody module is trained under supervised training using the defensive coding structure and the coding target trigger in the first coding mode. The feature coding sequence output by the preprocessor is constructed based on the first coding mode, and includes feature type (explicit / recessive), channel number, acquisition time, spatial location, etc. When making risk judgments, this feature coding sequence is aligned with the defensive coding structure, i.e., the semantic matching degree or field overlap between them is calculated. This process can use a label-pair supervised training method, using the feature coding sequence as input samples and the corresponding defensive coding structure as supervision labels to train the classification and matching capabilities of the risk antibody module.

[0044] In some embodiments, the risk antibody module can be constructed based on encoding similarity comparison methods, such as cosine similarity, hash comparison, or rule-based segmented matching, to achieve rapid matching and accurate classification of feature sequences to defense strategies. Through continuous training, the module gradually masters the semantic mapping relationship between different encoding patterns and has the ability to automatically match the optimal defense response strategy to new feature combinations.

[0045] In some embodiments, the construction of a risk defense structure includes: Obtain records of risk events in the utility tunnel, perform clustering processing under the same scenario, and determine N groups of record clusters; Traverse the N sets of record clusters to discover risk defense patterns, wherein the risk defense patterns are universal patterns within the clusters; The risk defense model is structurally integrated as the risk defense structure.

[0046] Furthermore, in some specific embodiments: First, a multi-source data fusion method was employed to obtain the risk event records for the utility tunnel. By retrieving heterogeneous data sources from the underground utility tunnel operation system, including sensor alarm logs, manual inspection records, video event tags, and dispatch instruction records, the data was standardized and formatted to transform it into a structured event dataset. The utility tunnel risk event records include fields such as timestamp, event location, risk type, trigger indicator value, response measures, and result feedback.

[0047] Next, a feature vector clustering method based on K-Means is used to perform clustering processing under the same source scenario. In this process, for each record in the utility tunnel risk event record, its key fields are extracted to construct a feature vector, including risk type code, spatial location code, trigger parameter value and response time, etc. The K-Means clustering algorithm is used to perform aggregation processing under the same source scenario (such as the same type of risk or the same partition). By setting the intra-cluster distance threshold, N record clusters are finally determined.

[0048] Subsequently, a high-frequency strategy path extraction method was employed to traverse N record clusters and uncover risk defense patterns. Specifically, statistical analysis was performed on event samples within each record cluster to identify the processing flow with the highest repetition rate in the response measure field as the representative strategy. Risk defense patterns are response path combinations that are widely adopted and effective within the cluster. For example, if the combination of "Level II response + ventilation activation + audible and visual alarm" occurs more than 80% of the time in most events, it can be identified as a universally applicable pattern within that cluster.

[0049] Finally, a structural template mapping method is used to parse the fields and populate the templates of the risk defense model, thus structurally integrating the risk defense model as a risk defense structure. Specifically, the extracted response path is broken down into standard fields, including "risk type", "trigger condition", "response level", "handling method", and "control unit action", etc., and integrated according to a unified template format to generate a structured data object. This structure is the standardized risk defense structure.

[0050] In some embodiments, the feature-encoded sequence is fed back to the back-end processor, and a risk antibody module with a built-in risk defense structure is activated to perform targeted alignment of the feature-encoded sequence and the defense-encoded structure, determining risk quantification data including: According to the network communication protocol, the feature coding sequence is sent back to the back-end processor, the risk antibody module built into the back-end processor is activated, the targeted alignment of the feature coding sequence-defense coding structure is performed, and the targeted defense coding node is determined. Based on the second encoding mode, the risk defense mode is determined and the risk type is deduced by interpreting the targeted defense encoding node, and the risk level is determined based on the risk defense level. The risk type and risk level are associated and distributed integration based on spatiotemporal codes is performed to determine risk quantification data, wherein the spatiotemporal codes include time codes and spatial codes.

[0051] It should be noted that the feature-encoded sequence is transmitted back to the backend processor via a communication link, activating its built-in risk antibody module. This module first performs a targeted alignment operation between the feature-encoded sequence and the defense coding structure, matching the targeted defense coding node that is closest to the current data. This node originates from a pre-stored risk defense structure, which is constructed by first acquiring historical records of utility tunnel risk events, dividing them into N record clusters through clustering under the same scenario, then performing pattern mining on each record cluster to extract risk defense patterns, and finally completing structured integration to generate a standardized defense coding structure library. The risk antibody module uses this structure library to align and map the feature-encoded sequence.

[0052] After target alignment is completed, the identified target defense coding nodes are interpreted based on the second coding mode to determine their corresponding risk defense modes. This further allows for the derivation of the current risk type (e.g., gas leak, structural anomaly, electrical anomaly) and risk level (e.g., Level I alert, Level II warning, Level III severe). Finally, the risk type and risk level are integrated according to their region and time label, combined with the embedded spatiotemporal code (including time and space codes), to output quantitative risk data with location index, time label, and risk level description.

[0053] Furthermore, in some specific embodiments: First, the feature encoding sequence is transmitted from the front-end processor to the back-end processor according to a preset network communication protocol (such as TCP / IP or a custom low-latency industrial communication protocol). This sequence is constructed based on the first encoding mode and contains multi-dimensional feature information collected by the front end (such as feature type, time label, collection number, spatial coordinates, etc.), and maintains temporal consistency and semantic integrity during the transmission process.

[0054] Subsequently, upon receiving the feature-encoded sequence, the backend processor activates the risk antibody module. This module contains multiple pre-trained defense coding structures, which are obtained by processing the risk defense structure through a second coding mode, forming structured labels that can be matched and identified. At this stage, a targeted alignment algorithm (such as field overlap rate matching and semantic label alignment) is used to match the feature-encoded sequence with the defense coding structure, determining the targeted defense coding nodes. Each node represents a callable risk defense response entry point, possessing mapping capabilities and reverse indexing functionality.

[0055] Next, based on the second encoding pattern, the matched targeted defense encoding nodes are decoded to restore their original semantic structure, i.e., the corresponding risk defense pattern. Here, field parsing and label mapping are used to restore the defense content represented by the nodes, such as "risk type," "trigger condition," and "response mechanism." Based on this, combined with the training model of the risk antibody module, the risk type is inversely deduced using a reverse attribution method, and the risk level (e.g., Level I, Level II, Level III, etc.) is determined according to the response level field contained within the node.

[0056] Finally, the identified risk types and risk levels are merged, and a spatiotemporal code is introduced for distributed integration. The spatiotemporal code comprises two dimensions: a time code recording the timestamp or time period of the event, and a spatial code identifying the sensor location, area number, or utility tunnel segment information to which the characteristic event belongs. By using hash-based merging or index aggregation methods, the time code and spatial code are bound to the risk results, generating quantified risk data with spatiotemporal labels.

[0057] In some embodiments, after determining the risk defense mode, the method further includes: The aforementioned risk defense models are integrated and stored in a temporary contingency plan database; By triggering the aforementioned temporary contingency plan database, risk emergency management of the underground utility tunnel area is implemented.

[0058] Furthermore, in some specific embodiments: First, risk defense models are integrated and stored in a structured format in a temporary contingency plan repository. This integration process is based on field mapping and rule classification methods, uniformly modeling each risk defense model according to elements such as "risk type, triggering conditions, response level, and emergency measures," and establishing an index mapping based on spatiotemporal tags, making the defense models searchable and activatable. The stored temporary contingency plan repository adopts a lightweight structure organization method, supporting fast querying and dynamic updates.

[0059] Once a specific risk event is detected and the risk is quantified, a temporary contingency plan database is triggered. Based on risk type and level, a matching search is performed, and the corresponding contingency response strategy is invoked. Subsequently, according to the matched contingency plan, emergency management operations for the underground utility tunnel area are executed, such as activating ventilation equipment, activating audible and visual alarms, and issuing emergency notices. This completes a closed-loop control process from identification to response, thereby improving the timeliness and accuracy of the overall risk response.

[0060] Please refer to Figure 2 , Figure 2 A structural block diagram of a risk classification and quantitative assessment device for underground utility tunnel networks provided in this embodiment of the invention; the specific device may include: The open network construction module 100 constructs an open network for underground utility tunnel areas based on pipeline topology and front-end sensor topology, where the open locations are data sensor aggregation points; Deployment module 200 is used to deploy preprocessors at each open location, the preprocessors including a first compressed node deployed in a low-rank matrix and a second decoupled node deployed in an explicit feature-implicit feature configuration; Feature encoding module 300 is used to perform dynamic sampling and open backhaul according to the open network, trigger the preprocessor to perform low-rank compression and feature decoupling processing of distributed data, and synthesize feature encoding sequence; The risk quantification module 400 is used to send the feature encoding sequence back to the back-end processor and activate the risk antibody module with the built-in risk defense structure to perform targeted alignment of the feature encoding sequence and the defense encoding structure to determine the risk quantification data. The underground utility tunnel network risk classification and quantification assessment device of this embodiment is used to implement the aforementioned underground utility tunnel network risk classification and quantification assessment method. Therefore, the specific implementation of the underground utility tunnel network risk classification and quantification assessment device can be seen in the previous embodiment section of the underground utility tunnel network risk classification and quantification assessment method. For example, the open network construction module 100, deployment module 200, feature encoding module 300, and risk quantification module 400 are respectively used to implement steps S101, S102, S103, and S104 in the above-mentioned underground utility tunnel network risk classification and quantification assessment method. Therefore, its specific implementation can be referred to the description of the corresponding embodiment, which will not be repeated here.

[0061] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for hierarchical and quantitative assessment of risks in underground utility tunnel networks, characterized in that, include: Construct an open network for underground utility tunnel areas based on pipeline topology and front-end sensor topology, where the open locations are data sensor aggregation points; A preprocessor is deployed at each open location, the preprocessor comprising a first compressed node deployed in a low-rank matrix and a second decoupled node deployed in an explicit feature-implicit feature configuration. Dynamic sampling and open backhaul are performed based on the open network, triggering the preprocessor to perform low-rank compression and feature decoupling processing of distributed data, and synthesizing feature encoding sequences; The feature-encoded sequence is sent back to the back-end processor, and the risk antibody module with built-in risk defense structure is activated to perform targeted alignment of the feature-encoded sequence and the defense-encoded structure to determine the risk quantification data.

2. The method for hierarchical and quantitative assessment of risks in underground utility tunnel networks according to claim 1, characterized in that, The construction of the underground utility tunnel area open network based on pipeline topology and front-end sensor topology includes: Set a preset utility tunnel range, divide the underground utility tunnel area into regions based on the preset utility tunnel range, and determine the utility tunnel region data; The locations of the openings are determined based on the pipe gallery segmentation data, wherein the pipe gallery segmentation data and the opening locations correspond one-to-one. The open network is determined based on the pipeline topology, the front-end sensing topology, and the open location.

3. The method for hierarchical and quantitative assessment of risks in underground utility tunnel networks according to claim 2, characterized in that, A first encoding mode is introduced and built into the output port of the preprocessor.

4. The method for hierarchical and quantitative assessment of risks in underground utility tunnel networks according to claim 3, characterized in that, The step of performing dynamic sampling and open-loop transmission based on the open network includes: Using preset sampling rules, the front-end sensing topology is driven to perform multi-source sensing acquisition in the pipe gallery and determine multi-source sensing data; Based on the correspondence between the pipe gallery domain data and the opening location, the multi-source sensor data is transmitted back to the opening location.

5. The method for hierarchical and quantitative assessment of risks in underground utility tunnel networks according to claim 4, characterized in that, The step of triggering the preprocessor to perform low-rank compression and feature decoupling processing of distributed data, and synthesizing the feature encoding sequence includes: The first compression node of the preprocessor is triggered to perform low-rank compression processing on the multi-source sensor data and determine the first compressed data. The first compressed data stream is transferred to the second decoupling node, where a binary classification process of explicit features and implicit features is performed and an identifier is added to determine the second decoupling data. At the output port, the second decoupled data is encoded and converted according to the first encoding mode to determine the feature encoding sequence.

6. The method for hierarchical and quantitative assessment of risks in underground utility tunnel networks according to claim 5, characterized in that, A second encoding mode is introduced to encode the risk defense structure, which serves as the defense encoding structure, wherein the risk defense structure and the defense encoding structure are associated with a mapping relationship. The risk antibody module is trained under supervision using the defense coding structure and the coding target trigger based on the first coding mode.

7. The method for hierarchical and quantitative assessment of risks in underground utility tunnel networks according to claim 6, characterized in that, The construction of the risk defense structure includes: Obtain records of risk events in the utility tunnel, perform clustering processing under the same scenario, and determine N groups of record clusters; Traverse the N sets of record clusters to discover risk defense patterns, wherein the risk defense patterns are universal patterns within the clusters; The risk defense model is structurally integrated as the risk defense structure.

8. The method for hierarchical and quantitative assessment of risks in underground utility tunnel networks according to claim 6, characterized in that, The step of sending the feature-encoded sequence back to the back-end processor and activating the risk antibody module with a built-in risk defense structure to perform targeted alignment of the feature-encoded sequence and the defense-encoded structure, and determining the risk quantification data, includes: According to the network communication protocol, the feature coding sequence is sent back to the back-end processor, the risk antibody module built into the back-end processor is activated, the targeted alignment of the feature coding sequence-defense coding structure is performed, and the targeted defense coding node is determined. Based on the second encoding mode, the risk defense mode is determined and the risk type is deduced by interpreting the targeted defense encoding node, and the risk level is determined based on the risk defense level. The risk type and risk level are associated and distributed integration based on spatiotemporal codes is performed to determine risk quantification data, wherein the spatiotemporal codes include time codes and spatial codes.

9. The method for hierarchical and quantitative assessment of risks in underground utility tunnel networks according to claim 8, characterized in that, After determining the risk defense mode, the following is also included: Integrate the aforementioned risk defense models and store them in a temporary contingency plan database; By triggering the aforementioned temporary contingency plan database, risk emergency management of the underground utility tunnel area is implemented.

10. A graded and quantitative assessment system for the risks of underground utility tunnel networks, characterized in that, include: The open network construction module constructs an open network for underground utility tunnel areas based on pipeline topology and front-end sensor topology, where the open locations are data sensor aggregation points; A deployment module is used to deploy a preprocessor at each open location, the preprocessor including a first compressed node deployed in a low-rank matrix and a second decoupled node deployed in an explicit feature-implicit feature configuration. The feature encoding module is used to perform dynamic sampling and open backhaul based on the open network, trigger the preprocessor to perform low-rank compression and feature decoupling processing of distributed data, and synthesize a feature encoding sequence. The risk quantification module is used to send the feature-encoded sequence back to the back-end processor and activate the risk antibody module with built-in risk defense structure to perform targeted alignment of the feature-encoded sequence and the defense encoding structure to determine the risk quantification data.

Images