Underground space facility fault diagnosis method, system and equipment

Abstract

The present invention provides a method, system and equipment for diagnosing underground space facility faults, which include: obtaining composite fault characteristics based on multi-source sensor data in the underground space; determining a target fault type based on the composite fault characteristics; locating a set of key nodes related to the target fault type in an underground space topology map; matching potential fault entities corresponding to the target fault type with an empirical knowledge graph to search for a target fault event entity; mapping nodes in the key node set to facility and equipment entities in the empirical knowledge graph to obtain attribute characteristics of the facility and equipment entities; searching for a fault cause entity associated with the target fault event entity based on the empirical knowledge graph, combining the attribute characteristics corresponding to each fault cause entity and each facility and equipment entity to obtain an optimized entity of the current fault cause, calculating a causal relationship score of the optimized entity through a relationship association with the target fault event entity, and selecting the most credible fault cause as the final fault cause based on the relationship score.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent operation and maintenance of underground space facilities, and in particular to a method, system and equipment for diagnosing faults of underground space facilities. Background Art

[0002] Currently, the operation and maintenance of underground facilities generally relies on manual inspections and discrete data analysis for troubleshooting. Existing technical solutions have the following major flaws: Traditional methods often focus on a single data source (such as sensor monitoring or BIM models), lack the ability to integrate multi-source heterogeneous data, and fail to achieve deep integration and dynamic reasoning of multi-source heterogeneous data. This leads to fragmented and unsystematic fault diagnosis, and is unable to achieve dynamic correlation and collaborative analysis of cross-modal information, thus failing to meet the growing demand for operation and maintenance brought about by the large-scale development of urban underground space. Furthermore, existing systems lack the ability to reason automatically based on spatial topological constraints. Fault detection relies on threshold alarms while ignoring the linkage effects between facilities. This leads to inaccurate assessments of the scope of fault impact and difficulty in locating hidden correlations. Root cause location relies on manual experience, resulting in delayed responses, low root cause tracing efficiency, high misjudgment rates, and the omission of potential risks. Summary of the Invention

[0003] In view of the above problems, the present invention is proposed to provide a method, system and equipment for diagnosing underground space facility faults that overcome the above problems or at least partially solve the above problems.

[0004] One aspect of the present invention provides a method for diagnosing underground space facility faults, the method comprising:

[0005] Acquiring monitoring data from various sensors in the underground space to obtain multi-source sensor data, and performing feature fusion on the multi-source sensor data to extract composite fault features for characterizing the confidence of different fault types;

[0006] determining a target fault type according to the composite fault characteristics;

[0007] A graph attention network is used to locate a set of key nodes related to the target fault type in a preset underground space topology map. The key node set includes nodes in the underground space topology map that are within the fault influence range of the target fault type.

[0008] Determine a potential fault entity corresponding to a target fault type, match the potential fault entity with a preset empirical knowledge graph to search for a target fault event entity in the empirical knowledge graph, where the target fault event entity is a fault event entity similar to the potential fault entity. The empirical knowledge graph includes facility and equipment entities, fault event entities, fault cause entities, and relationships between different entities.

[0009] Matching each key node in the key node set with the experience knowledge graph to map each key node to a facility and equipment entity in the experience knowledge graph, and obtaining attribute features associated with each facility and equipment entity;

[0010] According to the empirical knowledge graph, the fault cause entity associated with the target fault event entity is searched, and each fault cause entity is combined with each facility and equipment entity and its attribute characteristics to obtain the optimized entity of the current fault cause. The causal relationship score of the optimized entity of each fault cause is calculated by associating the target fault event entity through the relationship to obtain the relationship score of each fault cause. According to the relationship score of each fault cause, the most credible fault cause is selected as the final fault cause.

[0011] Another aspect of the present invention further provides an underground space facility fault diagnosis system, the system comprising:

[0012] a data fusion module, configured to acquire monitoring data from various sensors in the underground space to obtain multi-source sensor data, and perform feature fusion on the multi-source sensor data to extract composite fault features for characterizing the confidence of different fault types;

[0013] A fault type determination module is used to determine a target fault type according to the composite fault characteristics;

[0014] A spatial topology constraint analysis module is used to locate a set of key nodes related to the target fault type in a preset underground spatial topology map using a graph attention network. The key node set includes nodes in the underground spatial topology map that are within the fault influence range of the target fault type.

[0015] A first knowledge graph association module is configured to determine a potential fault entity corresponding to a target fault type, match the potential fault entity with a preset empirical knowledge graph, and search for a target fault event entity in the empirical knowledge graph. The target fault event entity is a fault event entity similar to the potential fault entity. The empirical knowledge graph includes facility and equipment entities, fault event entities, fault cause entities, and relationships between different entities.

[0016] a second knowledge graph association module, configured to match each key node in the key node set with the experience knowledge graph, so as to map each key node to a facility and equipment entity in the experience knowledge graph, and obtain attribute features associated with each facility and equipment entity;

[0017] The fault diagnosis module is used to search for the fault cause entity associated with the target fault event entity based on the empirical knowledge graph, combine each fault cause entity with each facility and equipment entity and its attribute characteristics to obtain the optimized entity of the current fault cause, calculate the causal relationship score of each fault cause optimized entity through the relationship association target fault event entity, obtain the relationship score of each fault cause, and select the most credible fault cause as the final fault cause according to the relationship score of each fault cause.

[0018] Another aspect of the present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor; when the computer program is executed by the processor, the steps of the underground space facility fault diagnosis method described above are implemented.

[0019] Another aspect of the present invention provides a computer program product having a computer program stored thereon, which implements the steps of the above-mentioned underground space facility fault diagnosis method when executed by a processor.

[0020] The underground space facility fault diagnosis method, system and equipment provided by the embodiments of the present invention solve the problems of insufficient multi-source data fusion, lack of spatial topology correlation analysis and delayed intelligent decision-making in the diagnosis and troubleshooting of underground space facility faults. The present invention realizes accurate and efficient autonomous diagnosis and troubleshooting of underground space facility faults by fusing multimodal sensor data, dynamic topology constraint modeling and knowledge graph-driven reasoning.

[0021] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are specifically listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present invention. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0023] Figure 1 A flow chart of a method for diagnosing underground space facility faults provided by an embodiment of the present invention;

[0024] Figure 2 A flowchart of a method for diagnosing underground space facility faults provided by another embodiment of the present invention;

[0025] Figure 3A structural block diagram of an underground space facility fault diagnosis system proposed in an embodiment of the present invention. DETAILED DESCRIPTION

[0026] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0027] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a," "an," "said," and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the stated features, integers, steps, operations, elements, and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0028] Those skilled in the art will understand that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by those skilled in the art in the art to which this invention belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art and, unless specifically defined, should not be interpreted in an idealized or overly formal sense.

[0029] Figure 1 The flowchart of the underground space facility fault diagnosis method according to one embodiment of the present invention is schematically shown. Figure 1 The underground space facility fault diagnosis method according to the embodiment of the present invention specifically includes the following steps:

[0030] S11. Acquire monitoring data from various sensors in the underground space to obtain multi-source sensor data, and perform feature fusion on the multi-source sensor data to extract composite fault features for characterizing the confidence of different fault types.

[0031] In this embodiment, after obtaining multi-source sensor data, sensor data preprocessing, basic probability assignment (BPA), Dempster combination rule calculation, and feature vector generation are performed in sequence to integrate multi-sensor data, eliminate noise, and extract composite fault features.

[0032] S12. Determine a target fault type according to the composite fault feature.

[0033] S13. Use a graph attention network to locate a set of key nodes related to the target fault type in a preset underground space topology map, where the key node set includes nodes in the underground space topology map that are within a fault influence range of the target fault type.

[0034] In this embodiment, a graph attention network (GAT) is used to locate the fault impact range and determine the key nodes that require priority treatment. Specifically, by inputting the underground space topology map and the target fault type, a set of key nodes is output.

[0035] S14. Determine the potential fault entity corresponding to the target fault type, match the potential fault entity with the preset experience knowledge graph to find the target fault event entity in the experience knowledge graph, the target fault event entity is a fault event entity similar to the potential fault entity, and the experience knowledge graph includes facility and equipment entities, fault event entities, fault cause entities and the relationship between different entities.

[0036] In the embodiment of the present invention, a time-series enhanced TransE model is used to construct an experience knowledge graph, associate historical cases with regulatory standards, and ultimately output the probability of the cause of the fault.

[0037] In the embodiment of the present invention, the construction of the experience knowledge graph specifically includes the following steps:

[0038] 1. Obtain historical case text data related to underground facility failures and annotate this historical case text data with entities and relationships. The entities include facility and equipment entities, failure event entities, failure cause entities, failure time entities, failure phenomenon entities, and failure response measures entities. Relationships represent the associations between different entities. Specifically, the present invention can collect historical case text data related to underground facility failures from channels such as urban management departments, meteorological departments, and news media. This data includes information such as the time, location, cause, scope of impact, response measures taken, and final results of the underground facility failure. The collected data is then cleaned to remove duplicate, erroneous, and incomplete data records. The data is carefully annotated to identify the entities involved (e.g., specific facility type, such as underground pipelines, ventilation equipment, power systems, etc.; the failure event itself; and the location of the failure, etc.) and relationships (e.g., "underground pipeline aging caused the failure," "ventilation equipment failure was associated with power system anomalies," etc.).

[0039] 2. Extract entities and relationships from the historical case text data to form structured entity-relationship-entity triples. This structured triple data is semantically stored via a knowledge graph to generate an empirical knowledge graph. Nodes in the empirical knowledge graph represent entities, and edges represent relationships between different entities. This method constructs the historical case knowledge graph by sequentially extracting entities, extracting relationships, and storing them in the knowledge graph. Specifically, entity types include: facilities and equipment (e.g., "cast iron pipe," "valve V001," "cable E12"), fault events (e.g., "gas leak"), physical phenomena (e.g., "abnormal flow rate"), and treatment measures (e.g., "close main valve"). Relationship types include "triggered," "manifested as," "needs to be treated as," and "affected by." Furthermore, facility and equipment entities possess attribute characteristics, including "material," "service life," "location coordinates," "sensor reading threshold," and "treatment duration." Ultimately, this is constructed as a structured triple (entity-relationship-entity) and stored in a graph database (such as Neo4j), forming a multidimensional knowledge association network centered on "fault-cause-action." Furthermore, physical phenomenon entities can be included to form a multidimensional knowledge association network centered on "fault-cause-phenomenon-action."

[0040] 3. A unified standard terminology entity is added to fault event entities with the same or similar fault phenomena in the empirical knowledge graph through semantic relationships. National or industry specification entries matching each standard terminology entity are then associated as rule entities with the corresponding standard terminology entity. This invention further introduces a mapping mechanism for term standardization and specification terms into the empirical knowledge graph. First, the system normalizes natural language fault descriptions in historical case texts (e.g., "gas leak" and "natural gas escape") into unified standard terms (e.g., "GasLeak"). By establishing semantic relationships such as "synonymous with" and "standard term is," the fault event entities in the graph are consistently expressed. Subsequently, based on the standard terminology, matching national or industry specification information is introduced as a new entity in the graph, namely, a standard terminology entity (e.g., "GB50028-2016" and "Section 5.4"). Relationships such as "reference specification," "applicable clause description," and "included clause" are established. This enables graph-level linkage between fault events and national or industry specification entries, providing data support for subsequent specification matching and work order generation.

[0041] S15. Match each key node in the key node set with the experience knowledge graph to map each key node to a facility and equipment entity in the experience knowledge graph, and obtain attribute features associated with each facility and equipment entity.

[0042] S16. Search for fault cause entities associated with the target fault event entity based on the empirical knowledge graph, combine each fault cause entity with each facility and equipment entity and its attribute characteristics to obtain the optimized entity of the current fault cause, calculate the causal relationship score of each fault cause optimized entity through the relationship association target fault event entity, obtain the relationship score of each fault cause, and select the most credible fault cause as the final fault cause based on the relationship score of each fault cause.

[0043] The underground space facility fault diagnosis method provided by the embodiment of the present invention solves the problems of insufficient multi-source data fusion, lack of spatial topology correlation analysis and delayed intelligent decision-making in the diagnosis and troubleshooting of underground space facility faults. The present invention integrates multimodal sensor data, dynamic topology constraint modeling and knowledge graph-driven reasoning, which not only realizes accurate and efficient autonomous diagnosis and troubleshooting of underground space facility faults, but also realizes closed-loop management of rapid positioning, root cause tracing and decision optimization, thereby improving the efficiency and safety of underground space operation and maintenance.

[0044] In the embodiment of the present invention, step S11 of performing feature fusion on the multi-source sensor data to extract composite fault features for characterizing the confidence of different fault types specifically includes the following steps not shown in the drawings:

[0045] S111. Normalize the monitoring data of each sensor in the underground space into standardized data with a unified dimension.

[0046] In this embodiment, one or more types of sensors are used in the underground space to obtain one or more types of raw data, and then the heterogeneous sensor data are standardized into a unified dimension to eliminate magnitude differences.

[0047] First, dynamic range setting is obtained based on historical data statistics.

[0048]

[0049] in, and are the statistical mean and standard deviation of historical normal state data respectively; and is the sensor range.

[0050] Then, normalize the data: ,

[0051] in, The raw data of the sensor (such as pressure value, temperature value, etc.); and is the preset sensor range; This is the normalized data, ranging from [0.1], which is used for subsequent BPA allocation.

[0052] S112. Map the standardized data corresponding to each sensor to the support probabilities of different fault types according to a preset fault mapping relationship to obtain the basic probability distribution of each sensor for different fault types; the fault mapping relationship includes the correspondence between the sensor and the support probabilities of different fault types.

[0053] In this embodiment, mapping rules from sensor types to fault types can be defined in advance based on expert knowledge or historical data to construct a fault mapping relationship to map normalized data to the confidence probability of each fault hypothesis, as follows:

[0054] ,

[0055] in, is the fault type of the i-th sensor pair The support probability of is a pre-set confidence value in the range of [0,1], where 0 indicates no support at all and 1 indicates complete confidence (for example, when the gas concentration sensor detects that the methane concentration exceeds the threshold, the probability of gas leakage may be considered to be 90% according to the fault mapping relationship (i.e. p =0.9)); is the normalized threshold for triggering a fault (e.g. 0.7). Finally, the BPA distribution of each sensor is output. , for subsequent fusion.

[0056] S113 , calculating the comprehensive confidence of each fault type according to the basic probability distribution of each sensor for different fault types, generating a fused feature vector according to the comprehensive confidence of each fault type, and using the fused feature vector as the composite fault feature.

[0057] In this embodiment, the Dempster-Shafer evidence theory (DST) method is used to fuse the BPA distribution of multiple sensors and quantify the joint support and conflict of each fault type, as follows:

[0058] ,

[0059] in, is the set of all possible fault types; Fault type for sensor 1 i +1 basic probability allocation; Fault type for sensor 2 i +2 basic probability distribution, and are mutually exclusive hypotheses; represents the intersection of the fault types supported by the two sensors; is the quantification of the degree of contradiction in the sensor data (range 0~1); is the normalized fault type i The comprehensive confidence level (range 0~1) of all fault types is: , used to construct the feature vector.

[0060] After obtaining the fused probability of each fault type, the fused probability is organized into a standardized vector to obtain the composite fault feature , used for subsequent dynamic fault judgment.

[0061] In the embodiment of the present invention, determining the target fault type according to the composite fault feature in step S12 specifically includes: calculating the normalized probability value of each fault type according to the composite fault feature, and selecting the fault type with the largest normalized probability value as the target fault type.

[0062] In this embodiment, hidden layer feature extraction, output layer linear combination, Softmax probability normalization, and highest probability fault type determination are performed in sequence to classify the fused data and determine the possible fault type. The specific implementation is as follows:

[0063] First, perform hidden layer feature extraction: used to convert low-dimensional features F Mapping to high-dimensional space can capture complex patterns of compound fault features (such as the joint features of "gas leakage + pressure drop").

[0064] ,

[0065] in, is the weight parameter from the input layer to the hidden layer, indicating the effect of the i-th input feature on the The contribution weight of hidden nodes; For the hidden layer The bias term of each node is used to adjust the activation threshold; For the i The activation value of the hidden node generates nonlinear features through the ReLU function;

[0066] It should be noted that the weight parameters and bias terms can be randomly initialized, or more preferably, optimized through training. In this embodiment of the present invention, the feature vector F, obtained by multi-source sensor fusion, is trained to cover all possible fault scenarios (such as gas leaks, pipe blockages, cable overheating, etc.) so that the actual probabilities after training are significantly biased towards the actual fault type.

[0067] Then the output layer linear combination is performed: the hidden layer features are linearly combined into a score for each fault type, quantifying the possibility of each fault type.

[0068] ,

[0069] in, is the weight parameter from the hidden layer to the output layer, indicating the The contribution weight of hidden nodes to the i-th fault type; is the bias term of the i-th fault type in the output layer, which is used to adjust the classification boundary; is the unnormalized score of the i-th fault type.

[0070] Then perform Softmax probability normalization: the score Convert to probability value , satisfying the probability distribution constraints.

[0071] ,

[0072] in, Score for the pair Perform exponential operations to ensure non-negativity and amplify differences; is the normalized probability of the i-th fault type, satisfying ,and , get the probability distribution of all fault types , k is the total number of fault types.

[0073] Finally, from the probability distribution T The fault type with the largest normalized probability value is selected as the current diagnosis result.

[0074] ,

[0075] in, is the probability distribution T The fault type corresponding to the largest probability value is the target fault type.

[0076] In the embodiment of the present invention, the step S13 of using the graph attention network to locate the key node set related to the target fault type in the preset underground space topology map specifically includes the following steps (not shown in the figure):

[0077] S131. Extract characteristic information related to the target fault type for each node in the underground space topology map.

[0078] In this embodiment, the input of the graph attention network is the target fault type and underground space topology. Used to specify the current fault type (such as gas leakage, cable overheating), driving GAT to focus on the topological pattern related to the fault. Underground pipe network topology map middle, It is the underground pipe network topology map; For nodes (static attributes: type (valve / pipeline / monitoring point), coordinates, material, service life. Dynamic status: pressure, flow, open and closed status (real-time update)); Edge (connection relationship: physical connection of pipes or cables. Physical parameters: length, diameter, resistance coefficient.).

[0079] The calculation process of the Graph Attention Network (GAT) is as follows: linear transformation of node features, attention weight calculation, and feature aggregation and update are performed in sequence. By modeling the dynamic correlation between pipeline network nodes, combining the physical topology structure and real-time data, the fault impact range and key control nodes can be accurately located.

[0080] This application extracts feature information related to the target fault type for each node in the underground space topology map through linear transformation of node features. Specifically, each node's original feature vector (including type, location, pressure, open / closed state, etc.) is mapped to a new latent space along with the fault type embedding vector through a preset trainable linear transformation. This transforms the node features into a latent layer representation after the linear transformation. This extracts deep pattern information related to the current fault type and obtains feature information related to the target fault type for each node. For example, it identifies features such as "valve nodes are more sensitive to pressure changes in gas leak scenarios."

[0081] S132. Perform attention scoring on all adjacent node pairs in the underground space topology map according to the characteristic information of each node, and calculate the total attention score of each node in the underground space topology map by other nodes according to the scoring results.

[0082] Specifically, by Perform attention scoring and calculate nodes i For Node j The degree of attention, and then normalized by softmax to obtain the normalized attention weight , , i.e. adjacent nodes j At the node i This step dynamically learns "key relationships" within the structure, such as the close linkage between a leak point and its downstream valve. Neighborhood features are then aggregated and activated based on attention weights to generate updated node representations. This allows the identification of key nodes with high influence in the fault propagation chain, such as pressure anomaly monitoring points and emergency stop valves within the leak-affected area.

[0083] Specifically, the total attention score calculation formula is:

[0084] ,

[0085] in, Representation node i The total attention of other nodes. The higher the score, the stronger its core role in the fault propagation chain. is a node set, used to traverse all nodes of interest or nodes in the graph; is the set of neighbor nodes of node i; and For nodes i and nodes j The original eigenvector of ; and is the hidden layer representation after linear transformation of node features, is the transpose of the trainable attention vector.

[0086] S133 . Select a preset number of nodes in descending order of total attention scores to form a set of key nodes related to the target fault type.

[0087] In this embodiment, the total attention score is sorted in descending order, and the first m nodes (such as the leakage point with the highest score) are selected to output the key node set. S， .

[0088] In this embodiment of the present invention, determining the potential fault entities corresponding to the target fault type in step S14 specifically includes the following steps: analyzing the fault meaning of the target fault type in the current specific scenario based on the contextual information obtained when constructing the empirical knowledge graph and a preset knowledge base file; and obtaining potential fault entities related to the target fault type from the contextual information based on the fault meaning. When constructing the empirical knowledge graph based on historical case text data, corresponding contextual information is obtained while extracting entities and relationships and stored as attribute information.

[0089] Furthermore, when multiple potential fault entities are obtained, a candidate entity list is generated based on the obtained potential fault entities, and each potential fault entity in the candidate entity list is matched with the empirical knowledge graph to search for fault event entities similar to each candidate entity in the empirical knowledge graph; fault event entities similar to each candidate entity are filtered according to the fault description information in the context information, and fault event entities that match the fault description information are preferentially selected as the target fault event entity that is finally mapped. If there are multiple potential candidate entities in the candidate list, and if multiple candidate entities are matched to multiple knowledge graph entities, the multiple fault entities matched in the knowledge graph are filtered through context information (such as fault description, occurrence time, location, etc.). For example, if the context mentions "Gas leak on Haitai Road in May 2025", the entity that matches the time and location is preferentially selected as the fault event entity that is finally mapped.

[0090] In the embodiment of the present invention, the target fault type is clarified by combining the context information and the relevant knowledge base files. " in a specific scenario, analyze the context information when it appears to obtain related clues, and then use named entity recognition technology to identify the relevant entities from the context. "The potential entities related to the target fault type are obtained. Then the potential fault entities are compared with the fault event entities in the knowledge graph, and matching entities are found by calculating the semantic similarity. If the direct match fails, the similarity matching algorithm is used to determine the most similar entity, and the matching result is verified in combination with the context. Finally, the obtained The fault event entity mapped to the knowledge graph is used as the target fault event entity. The relevant knowledge base documents refer to unstructured or semi-structured materials used to assist in understanding the meaning of the entity, such as historical operation and maintenance records, industry standards, operation manuals, or entity definitions in existing cases. This information helps improve the accuracy of entity recognition and matching.

[0091] In the embodiment of the present invention, the specific implementation process of step S15 is to pass in the key node set S , match and map each key node in the key node set with the entities in the experience knowledge graph, and map each node in S (such as valve, leakage point) to the facility and equipment entity in the knowledge graph (such as "valve V001", "cast iron pipe P002"); then perform attribute association to obtain the attribute characteristics associated with each facility and equipment entity (such as material, service life) as evidence for causal reasoning.

[0092] In the embodiment of the present invention, the specific implementation process of calculating the fault probability using the TransE model in step S16 is as follows: searching for the fault cause entity associated with the target fault event entity based on the empirical knowledge graph, combining each fault cause entity with each facility and equipment entity and its attribute characteristics to obtain the optimized entity of the current fault cause, mapping the optimized entity of the fault cause, the target fault event entity, and the relationship between the two into a low-dimensional vector representation through a pre-trained entity embedding table, and obtaining 、 and , calculate the causal relationship score of the optimized entity of each fault cause through the relationship association target fault event entity, and obtain the relationship score of each fault cause. The lower the score, the better. pass association The higher the probability, the more credible the fault cause is selected as the final fault cause according to the relationship score of each fault cause. The relationship score function is as follows:

[0093] .

[0094] In another embodiment of the present invention, the experience knowledge graph also includes a fault time entity and the relationship between the fault time entity and the fault event entity.

[0095] Furthermore, the most credible fault cause is selected as the final fault cause based on the relationship score of each fault cause, including:

[0096] The fault time entity associated with the target fault event entity is searched based on the empirical knowledge graph. The weight of the relationship score corresponding to the fault cause of the same knowledge graph path is determined based on the obtained fault time entity. This allows for time-weighted correlation of historical cases (e.g., recent cases have a higher weight). The formula is as follows:

[0097] ,

[0098] in, is the interval between the current time and the time information represented by the fault time entity, is the preset attenuation coefficient;

[0099] Traverse all knowledge graph paths related to the current fault in the experience knowledge graph, perform weighted summation of the relationship scores of each fault cause under all associated paths in the experience knowledge graph, and calculate the probability value of the current fault cause. , the formula is as follows:

[0100] ,

[0101] in, is the probability value of the i-th fault cause, is the vector representation of the optimized entity corresponding to the fault cause entity in the experience knowledge graph, is the vector representation of the target fault event entity in the experience knowledge graph, is the relationship vector connecting the two, Score the relationship of the corresponding failure cause; To traverse all knowledge graph paths related to the target fault event in the experience knowledge graph KG and perform weighted summation; is the Sigmoid function, which maps the weighted score to the probability interval of [0,1];

[0102] The fault cause with the largest probability value is selected as the final fault cause. Specifically, the fault cause probability distribution C can be obtained through the above calculation: ; and get the highest probability reason:

[0103] .

[0104] In another embodiment of the present invention, the empirical knowledge graph further includes a fault handling measure entity, a standard term entity, an industry specification item entity, and the relationships between the fault handling measure entity and the fault cause entity, between the standard term entity and the fault event entity, and between the industry specification item entity and the standard term entity.

[0105] Furthermore, as Figure 2 shown, the underground space facility fault diagnosis method proposed by the present invention can also integrate all information (the standardized processing result of the fault type, the extraction result of the key node topology information, the generation result of the fault root cause and the disposal plan) into an executable structured maintenance work order, specifically including:

[0106] S17. Generate the first structured information for fault description and specification reference according to the target fault event entity and the industry specification item entity associated with the target fault event entity. Specifically, the present invention obtains the highest probability fault type , such as "gas leakage", and calls the fault type entity mapping module in the knowledge graph to convert it into a unified standard term entity (such as GasLeak). Subsequently, the system retrieves the national or industry specification clauses semantically associated with it in the knowledge graph according to this standard term entity, and connects the standard term entity with the corresponding industry specification item (such as Section 5.4 of 《GB50028-2016》) through the predefined relationship types (such as "refer to specification", "apply standard", "constraint clause") in the graph. These association relationships are established as structured triples (such as <GasLeak, refer to specification, GB50028-2016 Section 5.4>) through manual annotation and standard document parsing in the graph construction stage, enabling the system to automatically extract the normative information matching the current fault and generate the first structured information for fault description and specification reference in the work order, obtaining the "fault type description" and "reference standard" fields, ensuring that the maintenance process complies with regulatory requirements and has standard support.

[0107] S18. Generate the second structured information for fault location according to the facility equipment entity corresponding to each key node in the key node set and the attribute characteristics corresponding to each facility equipment entity. After obtaining the key node set S in the present invention, each node (such as a valve, a pipeline) in it is mapped to a facility equipment entity, and the topological attribute characteristics (such as material, service life, location coordinates, caliber, etc.) information of the facility equipment entity is extracted to generate the second structured information for fault location as a work order field.

[0108] S19. Search for the fault handling measure entity associated with the corresponding fault cause entity from the empirical knowledge graph according to the final fault cause, and generate the third structured information for recommending fault handling methods according to the fault handling measure entity. The system obtains the final fault cause After the cause is found (such as "pipe aging"), the standard handling path related to the cause in the knowledge graph is called to generate the third structured information for recommending fault handling methods, and obtain the work order fields including "cause description", "fault handling operation and handling suggestions".

[0109] S20: Generate a structured maintenance work order according to the first structured information, the second structured information, and the third structured information.

[0110] In this embodiment, after obtaining the output fields of the three parts, namely the first structured information, the second structured information and the third structured information, they are spliced ​​according to the work order template to output a structured maintenance work order. The fields include the fault type, reference specifications, involved equipment, location information, fault handling operations and handling suggestions, which are suitable for the scheduling system to automatically send to the maintenance unit.

[0111] ,

[0112] in, It is the first structured information; is the second structured information; It is the third structured information.

[0113] The present invention combines multi-source sensor data fusion with Dempster-Shafer evidence theory to achieve complementary features and noise suppression in heterogeneous data, significantly improving the accuracy of extracting complex fault features. Based on the graph attention network, the spatial topological relationship of underground facilities is dynamically modeled to accurately capture the linkage effect and fault propagation path between nodes, solving the problem of the lack of implicit association analysis in traditional methods. Through the time-series enhanced TransE model to drive the knowledge graph association reasoning, historical cases and industry standards are integrated to achieve automatic tracing and specification matching of the root cause of the fault, breaking through the decision-making lag bottleneck caused by reliance on manual experience. Finally, a closed-loop decision-making system of "data fusion-topology deduction-knowledge-driven" is formed, which greatly improves the accuracy of fault location and the timeliness of handling, while reducing the cost of manual maintenance and the risk of misjudgment.

[0114] For simplicity of description, the method embodiments are described as a series of actions. However, those skilled in the art should be aware that the embodiments of the present invention are not limited by the order of the actions described, because certain steps can be performed in other orders or simultaneously according to the embodiments of the present invention. Secondly, those skilled in the art should also be aware that the embodiments described in this specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present invention.

[0115] Another embodiment of the present invention further provides an underground space facility fault diagnosis system, which includes a functional module for implementing the underground space facility fault diagnosis method as described in any one of the above items. Figure 3 The structure block diagram of the underground space facility fault diagnosis system according to another embodiment of the present invention is schematically shown. Figure 3 The underground space facility fault diagnosis system of this embodiment specifically includes a data fusion module 301, a fault type judgment module 302, a spatial topology constraint analysis module 303, a first knowledge graph association module 304, a second knowledge graph association module 305, and a fault diagnosis module 306, wherein:

[0116] The data fusion module 301 is used to obtain monitoring data of various sensors in the underground space to obtain multi-source sensor data, and perform feature fusion on the multi-source sensor data to extract composite fault features for characterizing the confidence of different fault types;

[0117] A fault type determination module 302 is configured to determine a target fault type based on the composite fault characteristics;

[0118] A spatial topology constraint analysis module 303 is configured to locate a set of key nodes related to a target fault type in a preset underground spatial topology map using a graph attention network, where the key node set includes nodes in the underground spatial topology map that are within a fault impact range of the target fault type.

[0119] A first knowledge graph association module 304 is configured to determine a potential fault entity corresponding to a target fault type, match the potential fault entity with a preset empirical knowledge graph, and search for a target fault event entity in the empirical knowledge graph. The target fault event entity is a fault event entity similar to the potential fault entity. The empirical knowledge graph includes facility and equipment entities, fault event entities, fault cause entities, and relationships between different entities.

[0120] The second knowledge graph association module 305 is configured to match each key node in the key node set with the experience knowledge graph, so as to map each key node to a facility and equipment entity in the experience knowledge graph, and obtain attribute features associated with each facility and equipment entity;

[0121] The fault diagnosis module 306 is used to search for the fault cause entity associated with the target fault event entity based on the empirical knowledge graph, combine each fault cause entity with each facility and equipment entity and its attribute characteristics to obtain the optimized entity of the current fault cause, calculate the causal relationship score of each fault cause optimized entity through the relationship association target fault event entity, obtain the relationship score of each fault cause, and select the most credible fault cause as the final fault cause based on the relationship score of each fault cause.

[0122] In an embodiment of the present invention, the experiential knowledge graph also includes a fault handling measure entity, a standard term entity, an industry specification entry entity, and the relationship between the fault handling measure entity and the fault cause entity, the relationship between the standard term entity and the fault event entity, and the relationship between the industry specification entry entity and the standard term entity.

[0123] Furthermore, the system also includes a decision and feedback generation unit not shown in the accompanying drawings, and the decision and feedback generation unit is used to generate first structured information for fault description and specification reference based on the target fault event entity and the industry specification entry entity associated with the target fault event entity; generate second structured information for fault location based on the facility and equipment entities corresponding to each key node in the key node set and the attribute characteristics corresponding to each facility and equipment entity; search the fault handling measure entity associated with the corresponding fault cause entity from the empirical knowledge graph according to the final fault cause, and generate third structured information for recommending fault handling methods based on the fault handling measure entity; generate a structured maintenance work order based on the first structured information, the second structured information and the third structured information.

[0124] As for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0125] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of this embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0126] In addition, another embodiment of the present invention also provides a computer device, including a memory, a processor, and a computer program stored on the memory and capable of running on the processor; when the computer program is executed by the processor, the steps of the underground space facility fault diagnosis method described above are implemented.

[0127] In addition, another embodiment of the present invention further provides a computer program product, which stores a computer program. When the computer program is executed by a processor, the steps of the underground space facility fault diagnosis method described above are implemented.

[0128] Those skilled in the art will appreciate that although some embodiments herein include certain features included in other embodiments but not other features, the combination of features from different embodiments is intended to be within the scope of the present invention and to form different embodiments. For example, any of the claimed embodiments may be used in any combination.

[0129] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for diagnosing underground space facility faults, characterized in that: The method comprises: Acquiring monitoring data from various sensors in the underground space to obtain multi-source sensor data, and performing feature fusion on the multi-source sensor data to extract composite fault features for characterizing the confidence of different fault types; determining a target fault type according to the composite fault characteristics; A graph attention network is used to locate a set of key nodes related to the target fault type in a preset underground space topology map. The key node set includes nodes in the underground space topology map that are within the fault influence range of the target fault type. Determine a potential fault entity corresponding to a target fault type, match the potential fault entity with a preset empirical knowledge graph to search for a target fault event entity in the empirical knowledge graph, where the target fault event entity is a fault event entity similar to the potential fault entity. The empirical knowledge graph includes facility and equipment entities, fault event entities, fault cause entities, and relationships between different entities. Matching each key node in the key node set with the experience knowledge graph to map each key node to a facility and equipment entity in the experience knowledge graph, and obtaining attribute features associated with each facility and equipment entity; Searching for fault cause entities associated with the target fault event entity based on the empirical knowledge graph, combining each fault cause entity with each facility and equipment entity and its attribute characteristics to obtain an optimized entity for the current fault cause, calculating a causal relationship score between each optimized entity of the fault cause and the target fault event entity through a relationship, and obtaining a relationship score for each fault cause. Based on the relationship score of each fault cause, the most credible fault cause is selected as the final fault cause; The experience knowledge graph also includes a fault time entity and the relationship between the fault time entity and the fault event entity; The most credible fault cause is selected as the final fault cause based on the relationship score of each fault cause, including: The fault time entity associated with the target fault event entity is searched based on the empirical knowledge graph. The weight of the relationship score corresponding to the fault cause of the same knowledge graph path is determined based on the obtained fault time entity. The formula is as follows: ， in, is the interval between the current time and the time information represented by the fault time entity, is the preset attenuation coefficient; Traverse all knowledge graph paths related to the current fault in the experience knowledge graph, perform weighted summation of the relationship scores of each fault cause under all associated paths in the experience knowledge graph, and calculate the probability value of the current fault cause. , the formula is as follows: ， in, is the probability value of the i-th fault cause, is the vector representation of the optimized entity corresponding to the fault cause entity in the experience knowledge graph, is the vector representation of the target fault event entity in the experience knowledge graph, is the relationship vector connecting the two, Score the relationship of the corresponding failure cause; To traverse all knowledge graph paths related to the target fault event in the experience knowledge graph KG and perform weighted summation; is the Sigmoid function, which maps the weighted score to the probability interval of [0,1]; The fault cause with the largest probability value is selected as the final fault cause.

2. The method according to claim 1, characterized in that The experience knowledge graph also includes a fault handling measure entity, a standard term entity, an industry specification item entity, and the relationship between the fault handling measure entity and the fault cause entity, the relationship between the standard term entity and the fault event entity, and the relationship between the industry specification item entity and the standard term entity; The method further comprises: Generate first structured information for fault description and specification reference according to the target fault event entity and the industry specification entry entity associated with the target fault event entity; generating second structured information for fault location based on the facility and equipment entities corresponding to each key node in the key node set and the attribute characteristics corresponding to each facility and equipment entity; Searching the experience knowledge graph for a fault handling measure entity associated with the corresponding fault cause entity according to the final fault cause, and generating third structured information for recommending a fault handling method according to the fault handling measure entity; A structured maintenance work order is generated according to the first structured information, the second structured information, and the third structured information.

3. The method according to claim 1, characterized in that Performing feature fusion on the multi-source sensor data to extract composite fault features for characterizing the confidence of different fault types includes: Normalize the monitoring data of each sensor in the underground space into standardized data with a unified dimension; According to a preset fault mapping relationship, the standardized data corresponding to each sensor is mapped to the support probability of different fault types to obtain the basic probability distribution of each sensor for different fault types; the fault mapping relationship includes the correspondence between the sensor and the support probability of different fault types; The comprehensive confidence of each fault type is calculated according to the basic probability distribution of each sensor to different fault types, and a fused feature vector is generated according to the comprehensive confidence of each fault type, and the fused feature vector is used as the composite fault feature.

4. The method according to claim 1, wherein Determining a target fault type according to the composite fault feature includes: The normalized probability value of each fault type is calculated according to the composite fault feature, and the fault type with the largest normalized probability value is selected as the target fault type.

5. The method according to claim 1, wherein The potential fault entities corresponding to the target fault type are determined to include: Analyze the fault meaning of the target fault type in the current scenario based on the context information obtained when building the experience knowledge graph and the preset knowledge base file; A potential fault entity related to a target fault type is acquired from the context information according to the fault meaning.

6. The method according to claim 5, characterized in that When multiple potential fault entities are obtained, a candidate entity list is generated based on the obtained potential fault entities, and each potential fault entity in the candidate entity list is matched with the experience knowledge graph to search for fault event entities similar to each candidate entity in the experience knowledge graph; Fault event entities similar to the candidate entities are filtered according to the fault description information in the context information, and the fault event entity matching the fault description information is preferentially selected as the target fault event entity finally mapped.

7. The method according to claim 1, characterized in that The graph attention network is used to locate the key node sets related to the target fault type in the preset underground space topology map, including: Extract characteristic information related to the target fault type from each node in the underground space topology map; According to the characteristic information of each node, the attention score of all adjacent node pairs in the underground space topology map is scored, and the total attention score of each node in the underground space topology map by other nodes is calculated based on the scoring results; A preset number of nodes are selected in descending order of total attention scores to form a set of key nodes related to the target fault type.

8. The method according to claim 1, characterized in that The steps of constructing the experience knowledge graph include: Obtain historical case text data related to underground space facility failures, and annotate the historical case text data with entities and relationships, where the entities include facility and equipment entities, failure event entities, failure cause entities, failure time entities, failure phenomenon entities, and failure handling measures entities, and the relationships represent associations between different entities; Extract entities and relationships from historical case text data to form structured entity-relationship-entity triples. The structured triple data is semantically stored in a knowledge graph to obtain an empirical knowledge graph. The nodes in the empirical knowledge graph represent entities, and the edges in the empirical knowledge graph represent the relationships between different entities. A unified standard term entity is added to the fault event entities with the same fault phenomenon in the experience knowledge graph through semantic relationships, and the national or industry specification entries matching each standard term entity are associated with the corresponding standard term entity as rule entities.

9. An underground space facility fault diagnosis system, characterized in that: The system comprises: a data fusion module, configured to acquire monitoring data from various sensors in the underground space to obtain multi-source sensor data, and perform feature fusion on the multi-source sensor data to extract composite fault features for characterizing the confidence of different fault types; A fault type determination module is used to determine a target fault type according to the composite fault characteristics; A spatial topology constraint analysis module is used to locate a set of key nodes related to the target fault type in a preset underground spatial topology map using a graph attention network. The key node set includes nodes in the underground spatial topology map that are within the fault influence range of the target fault type. A first knowledge graph association module is configured to determine a potential fault entity corresponding to a target fault type, match the potential fault entity with a preset empirical knowledge graph, and search for a target fault event entity in the empirical knowledge graph. The target fault event entity is a fault event entity similar to the potential fault entity. The empirical knowledge graph includes facility and equipment entities, fault event entities, fault cause entities, and relationships between different entities. a second knowledge graph association module, configured to match each key node in the key node set with the experience knowledge graph, so as to map each key node to a facility and equipment entity in the experience knowledge graph, and obtain attribute features associated with each facility and equipment entity; A fault diagnosis module is used to search for fault cause entities associated with the target fault event entity based on the empirical knowledge graph, combine each fault cause entity with each facility and equipment entity and its attribute characteristics to obtain an optimized entity for the current fault cause, calculate a causal relationship score between each optimized entity of the fault cause and the target fault event entity through the relationship, obtain a relationship score for each fault cause, and select the most credible fault cause as the final fault cause based on the relationship score of each fault cause; The experience knowledge graph also includes a fault time entity and the relationship between the fault time entity and the fault event entity; The most credible fault cause is selected as the final fault cause based on the relationship score of each fault cause, including: The fault time entity associated with the target fault event entity is searched based on the empirical knowledge graph. The weight of the relationship score corresponding to the fault cause of the same knowledge graph path is determined based on the obtained fault time entity. The formula is as follows: ， in, is the interval between the current time and the time information represented by the fault time entity, is the preset attenuation coefficient; Traverse all knowledge graph paths related to the current fault in the experience knowledge graph, perform weighted summation of the relationship scores of each fault cause under all associated paths in the experience knowledge graph, and calculate the probability value of the current fault cause. , the formula is as follows: ， in, is the probability value of the i-th fault cause, is the vector representation of the optimized entity corresponding to the fault cause entity in the experience knowledge graph, is the vector representation of the target fault event entity in the experience knowledge graph, is the relationship vector connecting the two, Score the relationship of the corresponding failure cause; To traverse all knowledge graph paths related to the target fault event in the experience knowledge graph KG and perform weighted summation; is the Sigmoid function, which maps the weighted score to the probability interval of [0,1]; The fault cause with the largest probability value is selected as the final fault cause.

10. A computer device, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and executable on the processor; when the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 8 are implemented.

11. A computer program product, characterized in that The computer program product stores a computer program, and when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images