Knowledge graph construction method and system based on electrical fault disaster

Abstract

The invention discloses a knowledge graph construction method and system based on electrical fault disaster, and belongs to the technical field of electrical fire early warning and intelligent security and protection. The method comprises the following steps: constructing an electrical fault disaster-causing multi-source data acquisition system, dynamically adjusting an acquisition strategy based on a safety threshold, and establishing an electrical fire disaster-causing link relationship library; identifying a specific entity relation mode of the electrical fault through a corpus special for the electrical field and a field word embedding technology; establishing a knowledge graph with dynamic evolution capability, and solving entity attribute conflicts based on credibility weighted fusion by adopting an incremental updating mechanism and partition life cycle management; and constructing an electrical fire situation perception model, carrying out early warning threshold adaptive optimization, and generating an analysis report containing technical responsibility and management responsibility through a multi-order disaster-inducing factor backtracking mechanism. Full-chain closed-loop management of electrical fire risks from perception and cognition to decision making is achieved, and the accuracy and tracing depth of fire early warning in a complex environment are remarkably improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of electrical fire early warning and intelligent security and protection, and particularly relates to a knowledge graph construction method and system based on electrical fault disaster. BACKGROUND

[0002] With the continuous improvement of electrification level, the power consumption scene is increasingly complex, the electrical fire has high frequency, strong concealment and great destructiveness, and has become one of the main threats to industrial and urban safety. Therefore, it is crucial to realize accurate and active electrical fire risk early warning. However, the main challenge in electrical fire analysis is the coupling and complexity of multiple factors. The traditional electrical fire early warning method is based on statistical analysis or simple causal inference, which is often difficult to capture the complex mechanism of electrical fire occurrence, development and spread. These methods usually rely on post-data analysis, and lack effective prediction and early warning ability for potential risks before electrical fire occurs. In addition, the traditional method is difficult to effectively integrate device parameters, environmental information, personnel behavior and other multi-source heterogeneous data, resulting in one-sidedness and limitation of the analysis results. SUMMARY

[0003] To solve the above problems existing in the prior art, the application provides a knowledge graph construction method and system based on electrical fault disaster; The purpose of the application can be achieved by the following technical solutions: S1: constructing an electrical fault disaster multi-source data collection list, dynamically adjusting the collection interval based on the electrical equipment safety threshold, extracting the disaster link relationship in the historical fire case through the interface of the fire file management system, and constructing an electrical fire disaster link relationship database; S2: based on the multi-source data characteristics in the field of electrical fire, establishing an electrical field special corpus, processing device full life cycle text, fault disposal text and phenomenon risk association text through field word embedding technology, and identifying the entity relationship mode of electrical fault; S3: establishing an electrical fire knowledge graph with dynamic evolution ability, responding to edge node data changes in real time through an incremental update trigger mechanism, optimizing graph storage through partition life cycle management, and solving entity attribute conflicts based on electrical fault disaster multi-source data credibility weighted fusion, and constructing an electrical fault disaster chain model supporting time and space backtracking; S4: based on the knowledge graph, establishing an electrical fire situation perception model, dynamically coupling real-time monitoring data and historical disaster modes, adaptively optimizing the early warning threshold, and generating an analysis report containing direct technical responsibility and indirect management responsibility based on a multi-order disaster factor backtracking mechanism.

[0004] Specifically, the construction of the electrical fault disaster multi-source data collection list has the following specific process: The dynamic acquisition rule is set based on the electrical equipment safety threshold value, and the acquisition interval is dynamically adjusted according to the deviation of the real-time monitoring data and the safety threshold value through an adaptive sampling frequency algorithm; An interface protocol of a fire-fighting archive system is established, equipment entities, fault entities and disaster-causing factor entities are extracted from historical fire disaster cases through an electrical fire entity precise identification algorithm, and a disaster-causing link relationship mapping table is constructed. A time synchronization service is deployed at an electrical fire disaster monitoring edge node, time stamps of multi-source acquisition data are iteratively corrected through a transmission delay correction model, and a standard data frame consistent in time and space is generated.

[0005] Specifically, the fire-fighting archive management system comprises: An unstructured database for storing historical fire disaster case reports, a preprocessing module supporting entity identification algorithm operation, a cause-effect relationship pattern library constructed based on electrical fire field knowledge, and a structured database for storing standardized disaster-causing link relationships, and a system interface for data interaction and entity relationship extraction with a knowledge graph construction system.

[0006] Specifically, the electrical field corpus comprises: Real-time acquisition of equipment operation state texts is performed through an industrial Internet of Things platform interface to construct an equipment operation state corpus layer; historical fire disaster case reports are extracted from the fire-fighting archive management system to construct a fire disaster case corpus layer; time series monitoring data are converted into natural language descriptions through a sensor data conversion module to construct a phenomenon association corpus layer; a corpus quality evaluation mechanism is established, domain relevance verification is performed on the collected corpus based on an electrical field professional dictionary, and the corpus that does not meet the requirements is filtered; a corpus version management mechanism is set, and the corpus is dynamically updated and maintained based on electrical equipment updates and standard revisions.

[0007] Specifically, the field word embedding technology comprises: A multi-source corpus hierarchical system specific to the electrical fire field is constructed, and the model parameters in the equipment full life cycle text, the cause-effect description in the fault disposal text, and the natural language description converted from the sensor time series data are fused to form an electrical fault semantic enhanced corpus; Based on the electrical professional dictionary and the fault disaster-causing chain knowledge, a field semantic enhancement model is constructed, electrical equipment association constraints and fault cause-effect relationship constraints are introduced in the word vector training process, and the terms with cause-effect relationship are kept in close semantic association in the vector space; An electrical field semantic verification mechanism is designed, a test set containing typical fault modes, disaster-causing paths and cause-effect relationships is constructed, and the accuracy of the word vector in electrical fault semantic reasoning is verified; A dynamic semantic evolution model is established, based on newly collected fault cases and device operation state data, to continuously adjust and optimize the word vector space, and adapt the semantic representation to the semantic evolution caused by the aging of electrical equipment and environmental changes.

[0008] Specifically, the entity relationship mode of the electrical fault includes: A device-fault-phenomenon-environment four-element relationship model is established, in which the device entity is connected to the fault entity through the occurrence relationship, the fault entity is connected to the phenomenon entity through the performance relationship, and the environment entity is connected to the fault entity through the induction relationship. An electrical disaster-causing causal relationship chain mode is constructed, based on historical fire case analysis, to establish a multi-level causal transmission relationship of insulation aging-causing-short circuit-triggering-overheating-generating-fire, and form a complete disaster-causing path description; and a time sequence correlation mode is designed to establish a "time correlation" relationship between device state data and fault occurrence time, and model the time sequence of the fault development process. A responsibility correlation mode is established, which associates the fault entity with the responsibility subjects such as personnel and management system through two types of relationship, management responsibility and technical responsibility, supports subsequent responsibility tracing analysis, and based on a multi-source data credibility weighting mechanism, evaluates the confidence of entity relationships from different sources.

[0009] Specifically, the incremental update trigger mechanism is used to respond to changes in edge node data in real time, and the specific process is as follows: A real-time data listener is deployed on the edge node to perform frame format verification and timestamp normalization on the collected monitoring data, and to perform message deduplication and batch identification; Based on real-time comparison of electrical parameters and safety thresholds, incremental update tasks are generated and added to the update queue according to priority, and the scheduler calls the corresponding entity recognizer and relationship extractor to perform knowledge extraction according to priority; The extraction results are mapped to fields and converted to units through a standardized interface, and change description metadata is generated for each result, and the real-time incremental update of the knowledge graph is completed by combining graph indexing and credibility weighting.

[0010] Specifically, the partition life cycle management includes: Based on risk level and time dimension, the data is divided into three storage levels: hot zone, warm zone and cold zone, and the real-time monitoring data, historical case data and archived data are stored in the hot zone, warm zone and cold zone respectively. Different storage strategies are implemented for different partitions, the hot zone data is stored in full memory to support real-time query, the warm zone data is stored in compressed form to support periodic analysis, and the cold zone data is archived in compressed form to provide on-demand backtracking loading. A partition pruning and on-demand rollback mechanism is established, based on a space-time index and query conditions, corresponding partitions are automatically selected, cold region archived data is provided with a decompression recovery interface, and a fast rollback analysis of historical disaster-causing cases is performed.

[0011] Specifically, the electrical fault disaster-causing multi-source data credibility weighted fusion solves entity attribute conflicts, and the specific process is as follows: An electrical fire data credibility grading system is established, the current and temperature data collected by the electrical fire monitoring terminal in real time are set as the highest credibility level, the device parameters and maintenance data recorded by the device management system are set as the medium credibility level, and the manual inspection record data are set as the basic credibility level; An attribute fusion rule based on credibility weight is constructed, different fusion weights are assigned to attribute values of different credibility levels, an electrical fault feature priority conflict resolution mechanism is adopted, based on the conflict of attribute values, the data with the highest matching degree with the electrical fault disaster-causing chain feature is preferentially adopted, and the latest effective data is selected in combination with the time stamp information; An attribute evolution tracking and version management is established, the resolution process and decision basis of each attribute conflict are recorded, a complete attribute change trajectory is formed, and the rollback analysis of the electrical fault disaster-causing process is supported.

[0012] Specifically, the electrical fire situation perception model comprises: A multi-dimensional coupling analysis model of electrical equipment operating state and environmental parameters is constructed, load current, cable temperature, environmental temperature and humidity, and device aging coefficient are collected in real time, and a dynamic correlation between electrical fault disaster-causing risk and multi-source monitoring parameters is established; A disaster-causing mode matching engine based on historical electrical fire cases is constructed, the abnormal data monitored in real time are compared with the features of typical electrical fault development modes, and the early features of the three main disaster-causing paths of overload heating, insulation deterioration and poor contact are identified; A dynamic optimization mechanism of warning threshold is established, according to the device operation history data, environmental conditions and real-time load rate, in combination with the fault evolution law in the knowledge graph, the warning threshold of key parameters such as temperature and current is automatically adjusted, and a warning strategy adaptive to specific operation scenarios is formed.

[0013] Specifically, the dynamic coupling of real-time monitoring data and historical disaster-causing modes comprises: An association analysis mechanism of real-time monitoring data and historical disaster-causing cases is established, the load current and cable temperature data collected at present are matched with the typical electrical fault cases stored in the knowledge graph in multiple dimensions; The key electrical parameter change law in the historical disaster-causing mode is extracted, including the duration of overload current, the temperature rise rate threshold and the insulation resistance decline trend, and a feature template library of electrical fault development is constructed; By similarity calculation of real-time data stream and feature template, the matching degree of the current running state and the historical disaster-causing mode is recognized, and when the similarity exceeds the set threshold, a warning is triggered, and a disposal suggestion based on historical experience is generated.

[0014] Specifically, the multi-stage disaster-causing factor backtracking mechanism comprises: Based on the direct cause tracing of the graph query, the edges in the knowledge graph are queried by the graph query statement based on the warning event node, the entities, attributes and relationships directly connected to the warning event node are extracted, and the direct cause entity causing the anomaly is quickly located by the graph pattern matching algorithm, and a direct disaster-causing path graph is generated to find the directly responsible equipment and environmental factors; Through deep root cause mining by electrical fault causal contribution analysis, the direct cause entity is set as the starting point of analysis, the current and temperature historical data sequence within the time window is extracted, the electrical fault evolution model is established, the contribution score of each disaster-causing factor to the current warning event is quantitatively calculated, the key disaster-causing path is identified and its dynamic evolution law is analyzed; Based on the counterfactual responsibility identification of electrical fault characteristics, a benchmark fact scenario is constructed, the key electrical parameters of the actually occurring disaster-causing event sequence are extracted as the basis for analysis, the counterfactual scenario is constructed by adjusting the overload current value, the environmental temperature value or the equipment aging degree parameter, the evolution path of each disaster-causing factor in the normal threshold range is simulated; by comparing the loss degree difference between the benchmark scenario and the counterfactual scenario, the influence weight of each disaster-causing factor is calculated, and an analysis report containing direct technical responsibility and indirect management responsibility is generated by combining the device management relationship in the knowledge graph, to clearly define the technical root cause and management responsibility of the accident.

[0015] A knowledge graph construction system based on electrical fault disaster-causing, comprising: A multi-source heterogeneous data collaborative collection module configured to construct an electrical fault disaster-causing multi-source data collection list, dynamically adjust the collection interval based on the electrical equipment safety threshold, extract the disaster-causing link relationship in the historical fire case from the fire archives management system, and construct an electrical fire disaster-causing link relationship database; A knowledge layer extraction and semantic understanding module configured to establish an electrical field special corpus based on the multi-source data characteristics of the electrical fire field, process the device full life cycle text, fault disposal text and phenomenon risk association text through field word embedding technology, and identify the entity relationship mode of electrical fault; A knowledge graph construction and risk modeling module configured to establish an electrical fire knowledge graph with dynamic evolution capability, respond to edge node data changes in real time through an incremental update trigger mechanism, optimize graph storage through partition life cycle management, and solve entity attribute conflicts based on electrical fault disaster-causing multi-source data credibility weighted fusion, and construct an electrical fault disaster-causing chain model supporting time and space backtracking; The intelligent reasoning and fire warning module is configured to establish an electrical fire situation perception model based on the knowledge graph, to perform adaptive optimization of a warning threshold through dynamic coupling of real-time monitoring data and historical disaster-causing modes, and to generate an analysis report containing direct technical responsibility and indirect management responsibility based on a multi-stage disaster-causing factor backtracking mechanism.

[0016] The present application has the following advantages: (1) By constructing a dynamic knowledge graph with "electrical fault disaster chain" as the core, the causal relationship and unified representation of multi-dimensional risk factors such as equipment, environment and historical cases are realized, overcoming the limitations of data islandization and static analysis of traditional warning methods, and significantly improving the overall cognition and warning accuracy of complex coupled disaster-causing paths.

[0017] (2) By establishing an "electrical fire situation perception model" and a "warning threshold dynamic optimization mechanism", real-time and adaptive evaluation of electrical operation risks is realized, and combined with a "multi-stage disaster-causing factor backtracking mechanism", a full-chain closed-loop management from real-time monitoring, intelligent diagnosis to root cause and responsibility tracing is formed, enhancing the proactive prevention and control capabilities of the system in pre-warning and post-tracing.

[0018] (3) Relying on the causal reasoning ability of the knowledge graph, combined with "electrical fault causal contribution analysis" and "responsibility identification based on counterfactual inference", not only high-credibility warning results can be output, but also quantitative and interpretable "technical and management responsibility analysis reports" can be generated, providing transparent and reliable basis for fire safety decision-making and responsibility definition, and promoting the intelligent upgrading of the electrical fire prevention and control system. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to facilitate the understanding of those skilled in the art, the present application will be further described below with reference to the accompanying drawings.

[0020] Figure 1 A flowchart of a knowledge graph construction method based on electrical fault disaster in the present application; Figure 2 An entity relationship diagram of a knowledge graph construction method based on electrical fault disaster in the present application. DETAILED DESCRIPTION

[0021] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined invention purpose, the specific embodiments, structures, features and effects according to the present application are described in detail as follows in combination with the preferred embodiments and the accompanying drawings.

[0022] Please refer to Figure 1 A knowledge graph construction method based on electrical fault disaster, comprising: S1: Construct a multi-source data collection list of electrical fault-causing disasters, dynamically adjust the collection interval based on the safety threshold of electrical equipment, and extract the disaster-causing link relationship from historical fire cases by connecting to the fire protection file management system to construct an electrical fire disaster-causing link relationship database; S2: Based on the multi-source data characteristics of the electrical fire field, establish a dedicated corpus for the electrical field, and process the full life cycle text of equipment, fault handling text and phenomenon risk association text through domain word embedding technology to identify the entity relationship patterns of electrical faults; S3: Establish an electrical fire knowledge graph with dynamic evolution capabilities, respond to changes in edge node data in real time through an incremental update trigger mechanism, optimize graph storage by adopting partition lifecycle management, and resolve entity attribute conflicts based on the weighted fusion of multi-source data on electrical faults to resolve conflicts, and construct an electrical fault disaster chain model that supports spatiotemporal backtracking. S4: Based on the knowledge graph, establish an electrical fire situation perception model, and adaptively optimize the early warning threshold by dynamically coupling real-time monitoring data with historical disaster-causing patterns. Based on a multi-level disaster-causing factor backtracking mechanism, generate an analysis report that includes direct technical responsibility and indirect management responsibility.

[0023] Specifically, the process of constructing the multi-source data collection list of electrical fault-causing disasters is as follows: Dynamic acquisition rules are set based on electrical equipment safety thresholds, and the acquisition interval is dynamically adjusted according to the deviation between real-time monitoring data and safety thresholds through an adaptive sampling frequency algorithm. Establish an interface protocol for the fire protection archive system, and extract equipment entities, fault entities, and disaster-causing factor entities from historical fire cases through an electrical fire entity precision identification algorithm, and construct a disaster-causing link relationship mapping table. Deploy time synchronization services at the edge nodes of electrical fire monitoring, and use a transmission delay correction model to iteratively correct the timestamps of multi-source collected data to generate spatiotemporally consistent standard data frames.

[0024] In this embodiment, a commercial complex has a total building area of ​​200,000 square meters, and its power distribution system includes 8 dry-type transformers, 56 low-voltage distribution cabinets, and more than 300 main power supply circuits. To achieve accurate early warning of electrical fires, a complete multi-source data acquisition system for electrical fault-related disasters needs to be constructed.

[0025] In the specific implementation process, dynamic acquisition rules are first set based on the safety thresholds of electrical equipment. Different safety threshold parameters are set for different electrical equipment: the current threshold for primary load distribution circuits is set to 105% of the rated value, and for secondary loads it is set to 110%; the temperature threshold for cable joints is set to vary from 70℃ to 90℃ depending on the cable type; the temperature threshold for transformer windings is uniformly set to 130℃. The system adjusts the acquisition interval in real time through an adaptive sampling frequency algorithm. When the current of phase A of the HXD-1000 distribution box reaches 198A (threshold is 200A), the acquisition interval is automatically adjusted from the usual 5 minutes to 30 seconds; when the ambient temperature of a certain floor distribution room reaches 45℃ (threshold is 50℃), the temperature sensor acquisition frequency is increased from once every 10 minutes to once every 1 minute.

[0026] Simultaneously, an interface protocol for the fire protection record system was established, connecting to the municipal fire brigade's fire investigation record management system via a standard API interface to obtain complete case studies of electrical fires in commercial buildings within the past 5 years. The system utilizes an electrical fire entity precision identification algorithm to process these case reports, accurately identifying "aging cable joints" as an equipment entity, "excessive contact resistance" as a fault entity, and "dust accumulation and moisture" as causative factors from the "XX Shopping Mall 2021 Fire Report," and establishing a complete causative chain: "dust accumulation and moisture → excessive contact resistance → localized overheating → ignition of insulation layer." Based on this, a causative chain relationship mapping table contains 136 valid records, each with detailed annotations of 12 key attribute fields, including causative factor type, development duration, and impact range.

[0027] Regarding the deployment of edge nodes for electrical fire monitoring, dedicated data acquisition nodes were deployed in 36 electrical rooms within the shopping mall. Each node is equipped with a GPS timing module and maintains time synchronization with the central server via the NTP protocol, ensuring that the time deviation is controlled within ±10 milliseconds. A Kalman filter model was trained based on historical transmission delay data. By analyzing 12,800 data transmission records from the past 30 days, a transmission delay prediction model was established, achieving a prediction accuracy of 92%. In actual operation, when a 180-millisecond transmission delay was detected between edge node 3 and the central server, the system automatically reverse-corrected the timestamps of the collected current and temperature data to ensure complete consistency with the central system's time reference. The final generated standard data frame strictly adheres to the ISO / IEC 14543-4 standard, containing complete fields such as device ID, parameter type, acquired values, correction timestamp, and data quality identifier.

[0028] Through the aforementioned implementation scheme, the system achieved accurate multi-source data collection and spatiotemporal synchronization in the complex power distribution environment of a commercial complex, providing a high-quality data foundation for subsequent knowledge graph construction. In actual operation, the system successfully issued early warnings for three potential electrical faults, with an average warning lead time of 42 minutes, fully validating the effectiveness and practical value of this method in complex building environments. Specifically, the fire protection record management system includes: An unstructured database for storing historical fire case reports, a preprocessing module to support entity recognition algorithms, a causal relationship pattern library built based on knowledge of electrical fires, and a structured database for storing standardized disaster-causing link relationships are all included. The system interacts with the knowledge graph through system interfaces to build the system's data interaction and entity relationship extraction.

[0029] Specifically, the electrical field-specific corpus includes: The system collects real-time equipment operation status text through an industrial IoT platform interface to construct an equipment operation status corpus layer; extracts historical fire case reports from a fire protection file management system to construct a fire case corpus layer; converts time-series monitoring data into natural language descriptions through a sensor data conversion module to construct a phenomenon association corpus layer; establishes a corpus quality assessment mechanism, performs domain relevance verification on the collected corpus based on an electrical field professional dictionary, and filters out corpus that does not meet the requirements; and sets up a corpus version management mechanism to dynamically update and maintain the corpus based on electrical equipment updates and standard revisions.

[0030] Specifically, the domain term embedding technology includes: We construct a multi-source hierarchical corpus system unique to the field of electrical fires, which integrates model parameters in the full life cycle text of equipment, causal descriptions in fault handling text, and natural language descriptions converted from sensor time-series data to form an electrical fault semantic enhancement corpus. Based on electrical engineering lexicon and fault causal chain knowledge, a domain semantic enhancement model is constructed. Electrical equipment association constraints and fault causal relationship constraints are introduced during word vector training to maintain close semantic association between terms with causal relationships in the vector space. We designed a semantic verification mechanism for the electrical domain and verified the accuracy of word vectors in semantic reasoning of electrical faults by constructing a test set containing typical fault modes, disaster-causing paths and causal relationships. A dynamic semantic evolution model is established. Based on newly collected fault cases and equipment operation status data, the word vector space is continuously adjusted and optimized to adapt the semantic representation to the semantic evolution brought about by the aging of electrical equipment and dynamic factors of environmental changes.

[0031] In this embodiment, based on the monitoring scenario of a power distribution system of a large data center, the data center includes four 2000kVA transformers, 128 precision distribution cabinets, and more than 2000 power supply circuits, requiring the construction of an accurate semantic understanding model for electrical faults.

[0032] In constructing the multi-source corpus hierarchical system, the system first establishes a three-layer corpus architecture: the equipment lifecycle text layer, which connects to equipment manufacturer databases to obtain structured data such as Schneider PM8000 series instrument parameters and Siemens SIVACON low-voltage cabinet configuration documents, and establishes a three-level index according to equipment type-model-parameter; the fault handling text layer, which collects fault descriptions recorded by maintenance personnel in real time through the industrial IoT platform interface, such as "Phase B cable joint overheated to 95℃, and after inspection, it was found that poor crimping caused excessive contact resistance," and other complete handling records; the phenomenon-risk association text layer innovatively converts sensor time-series data into natural language descriptions, such as converting the current sequence [185A, 192A, 198A, 205A] into the semantic description "Phase A current continuously increased from 185A to 205A within 30 minutes, exceeding the rated value by 5%." The three types of corpora are integrated through a semantic fusion module based on an electrical professional dictionary to form an electrical fault semantic enhancement corpus containing 120,000 labeled samples.

[0033] The domain semantic enhancement model is trained using an innovative method based on the characteristics of electrical fault propagation. During word vector training, not only are word co-occurrence relationships considered, but more importantly, electrical equipment association constraints and fault causal relationship constraints are introduced. Specifically, the system constructs an electrical fault propagation graph, setting "insulation aging" as the parent node and "partial discharge" and "dielectric loss" as child nodes, forcibly maintaining this causal topology during vector space optimization. Simultaneously, equipment association constraints are established based on electrical connection relationships; for example, the electrical connection path "transformer-low-voltage cabinet-distribution box-terminal equipment" is represented as a continuous semantic manifold in the vector space. Through this dual constraint, the cosine similarity between "insulation aging" and "breakdown discharge" increases from 0.32 in the traditional method to 0.78, while the similarity with the irrelevant word "ambient humidity" decreases from 0.45 to 0.12.

[0034] The implementation of the semantic verification mechanism involved a dedicated electrical fault semantic reasoning test set. This test set included 36 typical fault scenarios, such as verifying the complete disaster path of "overload-heating-insulation degradation-short circuit". Each scenario included multiple sets of semantic reasoning questions, such as "If the A-phase current is continuously overloaded by 15%, what is the most likely subsequent phenomenon?" The system evaluated the model's performance by calculating the semantic reasoning accuracy of word vectors on this test set, requiring an accuracy of at least 85% before deployment. During the verification process, it was found that the accuracy of traditional word vectors in the positive feedback loop reasoning of "poor contact-heating-oxidation-increased contact resistance" was only 62%, while it improved to 89% after domain-enhanced training.

[0035] The innovation of the dynamic semantic evolution model lies in the establishment of a semantic drift tracking mechanism. The system continuously monitors newly collected fault cases. When it detects that a new fault mode, "mechanical jamming leading to delayed tripping," has appeared in a circuit breaker of a certain brand after three years of operation, the system automatically adds the case to the training corpus and readjusts the vector representations of related words. Simultaneously, a semantic decay model is established based on equipment aging data. For example, for equipment that has been in operation for more than five years, the semantic association weight between "insulation resistance" and "aging cracking" is automatically increased by 30%. Changes in environmental factors are also incorporated into the semantic evolution considerations. For instance, during the rainy season, the semantic similarity between "ambient humidity" and "surface creepage" is dynamically adjusted based on real-time monitoring data, with the adjustment magnitude calculated based on the correlation strength between the two in historical data.

[0036] The implementation results show that the word vectors trained with domain enhancement achieve an entity recognition accuracy of 94.2% and a relation extraction F1 score of 88.7% in the electrical fault text understanding task, representing improvements of 15.6% and 22.3% respectively compared to general word vectors. In practical applications, the system successfully identified the semantic features of multiple complex faults. For example, when analyzing a cascading fault of "harmonic current - accelerated temperature rise - insulation aging," it accurately captured the semantic relationships between various phenomena, providing reliable semantic support for early warning decisions. Through continuous dynamic evolution, the system maintained a semantic understanding accuracy of over 92% after one year of operation, demonstrating good adaptability. Specifically, the entity relationship model of the electrical fault includes: Establish a four-element relationship model of equipment-fault-phenomenon-environment, where the equipment entity and the fault entity are connected through the occurrence relationship, the fault entity and the phenomenon entity are connected through the manifestation relationship, and the environment entity and the fault entity are connected through the induction relationship. A causal relationship chain model for electrical disasters is constructed. Based on the analysis of historical fire cases, a multi-level causal transmission relationship is established, which includes insulation aging, short circuit, overheating, and fire, forming a complete description of the disaster path. A time-series correlation model is also designed to establish a "time correlation" relationship between equipment status data and the time of failure occurrence, and to model the time series of the failure development process. A responsibility association model is established, which associates faulty entities with responsible parties such as personnel and management systems through two types of relationship: management responsibility and technical responsibility. This supports subsequent responsibility tracing and analysis, and the confidence level of entity relationships from different sources is assessed based on a multi-source data credibility weighting mechanism.

[0037] Specifically, the process of responding to edge node data changes in real time through the incremental update triggering mechanism is as follows: Deploy real-time data listeners at edge nodes to perform frame format verification and timestamp normalization on the collected monitoring data, and perform message deduplication and batch identification. Based on the real-time comparison of electrical parameters and safety thresholds, incremental update tasks are generated and added to the update queue according to priority. The scheduler calls the corresponding entity recognizer and relation extractor to perform knowledge extraction according to priority. The extracted results are mapped to fields and converted to units using a standardized interface. Metadata describing the changes is generated for each result. Real-time incremental updates of the knowledge graph are completed by combining graph index and credibility weighting.

[0038] Specifically, the partition lifecycle management includes: Data is partitioned based on risk level and time dimension, and real-time monitoring data, historical case data and archived data are stored in three storage levels: hot zone, warm zone and cold zone, respectively. Differentiated storage strategies are implemented for different partitions: hot zone data is stored in full memory to support real-time querying, warm zone data is stored in compressed form to support periodic analysis, and cold zone data is compressed and archived to provide on-demand backtracking loading. Establish a partition pruning and on-demand backtracking mechanism, automatically select the corresponding partition based on spatiotemporal index and query conditions, provide a decompression and recovery interface for cold zone archived data, and conduct rapid backtracking analysis of historical disaster cases.

[0039] Specifically, the process of resolving entity attribute conflicts based on the weighted fusion of multi-source data on electrical fault-induced disasters is as follows: Establish a data credibility grading system for electrical fires, setting the current and temperature data collected in real time by electrical fire monitoring terminals as the highest credibility level, the equipment parameters and maintenance data recorded by the equipment management system as the medium credibility level, and the data recorded by manual inspections as the basic credibility level. An attribute fusion rule based on credibility weight is constructed, and differentiated fusion weights are assigned to attribute values ​​with different credibility levels. An electrical fault feature-priority conflict resolution mechanism is adopted. Based on the conflict of attribute values, the data with the highest matching degree with the electrical fault disaster chain features is selected first, and the latest valid data is selected in combination with timestamp information. Establish attribute evolution tracking and version management to record the resolution process and decision-making basis of each attribute conflict, forming a complete attribute change trajectory, and supporting retrospective analysis of the disaster process caused by electrical faults.

[0040] Specifically, the electrical fire situation awareness model includes: Construct a multidimensional coupled analysis model of electrical equipment operating status and environmental parameters, collect load current, cable temperature, ambient temperature and humidity and equipment aging coefficient in real time, and establish a dynamic correlation between electrical fault disaster risk and multi-source monitoring parameters; A disaster pattern matching engine based on historical electrical fire cases is constructed. The abnormal data monitored in real time is compared with the typical electrical fault development patterns to identify the early characteristics of three main disaster paths: overload heating, insulation deterioration, and poor contact. Establish a dynamic optimization mechanism for early warning thresholds. Based on historical equipment operation data, environmental conditions, and real-time load rate, combined with the fault evolution patterns in the knowledge graph, automatically adjust the early warning thresholds of key parameters such as temperature and current to form an early warning strategy that is adaptive to specific operating scenarios.

[0041] In this embodiment, based on the high-voltage power distribution system of a large data center, the following describes the actual monitoring scenario of the high-voltage power distribution room of a large data center. This power distribution room is responsible for 60% of the power supply to the data center and includes four 2000kVA dry-type transformers, 32 high-voltage switchgear cabinets, and supporting cable systems.

[0042] In the in-depth implementation of the multidimensional coupling analysis model, the system innovatively introduced a dynamic coupling factor matrix calculation method. Taking the monitoring of the No. 3 transformer outgoing switch as an example, the system not only collects traditional load current (385A) and cable surface temperature (68℃), but also simultaneously acquires parameters in 16 dimensions, including current harmonic distortion rate (7.8%), temperature change gradient (2.3℃ / min), ambient dew point temperature (23℃), and equipment aging coefficient (0.72) calculated based on equipment commissioning time and load history. By constructing a deep neural network coupling model, the system discovered implicit correlations that are difficult to identify using traditional methods: when the 5th harmonic content exceeds 6%, even if the current is within the safe range, the temperature rise rate of the cable joint will increase significantly by about 40%. This discovery enabled the system to issue an early warning 47 minutes in advance of a joint overheating risk caused by harmonic current during monitoring on a Wednesday morning, while all parameters of the traditional monitoring system showed normal values.

[0043] The disaster pattern matching engine's effectiveness lies in its ability to extract microscopic features of the fault development process. Based on 286 electrical fire cases collected over the past five years from high-voltage power distribution rooms in large data centers, the system has constructed a refined feature template library encompassing three main fault types: overload-induced heating, insulation degradation, and poor contact. Taking poor contact fault identification as an example, the system not only monitors the absolute temperature value of the connection point but, more importantly, analyzes its thermal dynamic response characteristics: under normal conditions, there is an inertial delay of 3-5 minutes in temperature response when the load current changes, while this delay shortens to 1-2 minutes in cases of poor contact, and the temperature fluctuation amplitude increases significantly. In practical applications, the system calculates the similarity between the real-time monitoring curve and the fault template using a dynamic time warping algorithm. When the temperature response delay of a cable joint suddenly shortens from the normal 4.2 minutes to 1.5 minutes when the current changes, the system immediately identifies it as an early characteristic of poor contact, issuing a warning 32 minutes earlier than the traditional method based on a fixed temperature threshold. Furthermore, the system has established a fault path probability graph model, capable of predicting the likelihood of the current abnormal state developing into a complete disaster-causing link, providing a basis for tiered early warning.

[0044] The core of the dynamic optimization mechanism for early warning thresholds lies in establishing an adaptive adjustment strategy based on the health status of equipment throughout its entire lifecycle. The system creates an independent early warning threshold evolution file for each critical piece of equipment, comprehensively considering multiple factors such as equipment commissioning time, operating environment, load history, and maintenance records. Taking a dry-type transformer that has been in operation for 5 years as an example, based on its historical operating data (annual average load rate of 75%, highest ambient temperature of 38℃, and cumulative start-stop cycles of 1520), combined with an insulation material thermal aging model, the system dynamically adjusts the winding temperature early warning threshold from the factory-set 145℃ to 128℃. The specific optimization process includes: first, calculating the equipment health index (based on parameters such as insulation resistance change rate and partial discharge intensity), and then establishing a threshold adjustment function based on the real-time load rate. When the transformer's load rate is detected to continuously exceed 80%, the system automatically activates an enhanced monitoring mode, further reducing the temperature early warning threshold to 120℃, while simultaneously increasing the monitoring frequency of current harmonic content from 5 minutes / time to 30 seconds / time. This refined dynamic adjustment significantly improved the early warning accuracy and significantly reduced the false alarm rate compared to the traditional fixed threshold method during the three-month trial operation.

[0045] Specifically, the dynamic coupling of real-time monitoring data with historical disaster-causing patterns includes: Establish a correlation analysis mechanism between real-time monitoring data and historical disaster cases, and perform multi-dimensional feature matching between the currently collected load current and cable temperature data and typical electrical fault cases stored in the knowledge graph; Extract the variation patterns of key electrical parameters from historical disaster-causing patterns, including the duration of overload current, the threshold of temperature rise rate, and the decreasing trend of insulation resistance, and construct a feature template library for the development of electrical faults. By calculating the similarity between real-time data streams and feature templates, the system identifies the degree of matching between the current operating status and historical disaster-causing patterns. When the similarity exceeds a set threshold, an early warning is triggered, and disposal suggestions based on historical experience are generated.

[0046] Specifically, the multi-level disaster-causing factor backtracking mechanism includes: Direct cause tracing based on graph query: Based on the early warning event node, the graph query statement performs graph traversal query on the edges in the knowledge graph to extract the entities, attributes and relationships directly connected to the early warning event node. At the same time, the graph pattern matching algorithm is used to quickly locate the direct cause entity that caused the anomaly and generate a direct disaster path graph to find the directly responsible equipment and environmental factors. By analyzing the causal contribution of electrical faults, we can deeply explore the root causes. We take the direct cause entity as the starting point of the analysis, extract the historical data sequence of current and temperature within the time window, establish an electrical fault evolution model, quantify the contribution score of each disaster-causing factor to the current warning event, identify key disaster-causing paths, and analyze their dynamic evolution law. Counterfactual liability determination based on electrical fault characteristics constructs a baseline factual scenario. Key electrical parameters are extracted from the actual disaster-causing event sequence as the basis for analysis. By adjusting parameters such as overload current, ambient temperature, or equipment aging, a counterfactual scenario is constructed to simulate the evolution path of each disaster-causing factor within its normal threshold range. By comparing the difference in the degree of loss between the baseline scenario and the counterfactual scenario, the influence weight of each disaster-causing factor is calculated. Combined with equipment management relationships in the knowledge graph, an analysis report containing direct technical responsibility and indirect management responsibility is generated to clarify the technical root cause and management responsibility of the accident.

[0047] In this embodiment, during the analysis of a short-circuit fire accident in the power distribution system of a chemical plant, the system initiated a multi-stage disaster-causing factor backtracking mechanism for in-depth analysis. The fire in the chemical plant's 10kV power distribution system was caused by a cable fault, resulting in significant losses to the production line. The system then initiated the backtracking mechanism for comprehensive analysis.

[0048] In the direct cause tracing process based on electrical feature graphs, the system employs a graph query method enhanced with electrical fault features. Starting with the "cable fire" early warning event node, the system performs a deep graph traversal query based on electrical characteristics within the electrical fire knowledge graph. Unlike traditional graph queries, the system first identifies entity networks with electrical connectivity and then performs a bidirectional traversal along current and heat conduction paths. The query results show a strong correlation between the "10kV cable joint A37" equipment entity and the "insulation breakdown" fault entity, with a correlation strength of 0.87. Simultaneously, through electrical environment coupling analysis, the system identifies a synergistic influence coefficient of 0.68 for the "high temperature environment" environmental entity. During path construction, the system introduces electrical fault propagation timing constraints, accurately reconstructing the complete accident chain of "insulation aging - leading to - partial discharge - triggering - short circuit - generation - fire," and calculating the time interval and energy transfer efficiency of each link. In particular, the system discovers hidden paths that are difficult to identify using traditional methods: through abnormal fluctuations in the cable shielding potential, it traces the harmonic pollution impact of upstream equipment.

[0049] In the in-depth quantitative analysis phase of the causal contribution of electrical faults, the system adopted a dynamic evolution modeling method based on electrophysical processes. Using "insulation breakdown" as the analysis anchor point, the system extracted multi-dimensional historical data from the 30 days prior to the accident, including the harmonic spectrum of three-phase currents, the temperature gradient distribution of cable joints, the coordinated changing trends of ambient temperature and humidity, and the frequency characteristic changes of insulation resistance. The system innovatively constructed an electrical fault coupled dynamics model, which comprehensively considers the interaction of electro-thermal-chemical multi-physics fields. In the contribution calculation, the system not only analyzes the absolute values ​​of parameters but also focuses on their rate of change and the strength of their interactions. The quantitative results show that the contribution of the equipment aging coefficient is 0.78, of which insulation aging based on tanδ measurement accounts for 0.42, and connection performance degradation based on mechanical strength degradation accounts for 0.36. Of the 0.65 contribution of ambient temperature, the direct temperature rise effect accounts for 0.38, and the accelerated aging effect accounts for 0.27. This detailed analysis reveals the key disaster mechanism: "synergistic effect of equipment aging and high-temperature environment → decrease in glass transition temperature of insulating material → decrease in partial discharge initiation voltage → insulation breakdown".

[0050] In the responsibility determination process based on counterfactual deduction of electrical parameters, the system established a simulation platform for the chain reaction of electrical faults. By constructing a baseline factual scenario, the system extracted key electrical state parameters from the accident sequence, including core data such as the waveform characteristics of operating current, the spatiotemporal distribution of equipment temperature, and changes in the dielectric properties of insulating materials. In the counterfactual deduction, the system adopted a step-by-step substitution method for electrical parameters: first, it simulated a scenario where cable joints were replaced according to a prescribed cycle, calculating the dielectric strength retention rate of the insulating material; then, it simulated the ambient temperature distribution after thermal isolation measures were implemented, analyzing its effect on suppressing the rate of insulation aging; finally, it simulated the current spectrum characteristics after optimizing the operating load, evaluating its degree of improvement on the thermal stability of the equipment. By comparing the differences in system energy distribution between the baseline scenario and the counterfactual scenario, the system accurately calculated the impact weight of each disaster-causing factor, finding that equipment exceeding its service life increased the probability of insulation material carbonization by 62%, ambient temperature exceeding the standard increased the frequency of partial discharge activity by 45%, and excessive operating load accelerated the electrothermal aging process by 28%. By combining the equipment management relationship network and electrical safety responsibility matrix in the knowledge graph, the system generates a quantitative responsibility analysis report, which clearly delineates the specific responsibilities of the equipment maintenance department for lack of preventive maintenance, the production management department for dereliction of duty in environmental control, and the operation scheduling department for improper load management.

[0051] Implementation results demonstrate the unique value of this retrospective mechanism in complex industrial scenarios. Through electrical feature-enhanced analysis, the system uncovered multiple coupled disaster-causing factors that are difficult to identify using traditional methods, including the synergistic effect of harmonic pollution and insulation aging, and the deep-seated mechanisms of the interaction between ambient temperature and equipment load. Compared to traditional fire investigation methods, this system reduces accident analysis time from 5 days to 4 hours, increases the accuracy of liability determination from 68% to 94%, and provides a quantitative chain of evidence based on electrophysical processes. More importantly, the counterfactual analysis report generated by the system provides chemical plants with specific improvement directions, including optimization suggestions for harmonic mitigation measures, ambient temperature control standards, and equipment replacement cycles, effectively preventing the recurrence of similar accidents.

[0052] A knowledge graph construction system based on electrical fault-induced disasters includes: The multi-source heterogeneous data collaborative acquisition module is configured to build a multi-source data acquisition list of electrical faults, dynamically adjust the acquisition interval based on the safety threshold of electrical equipment, and extract the disaster-causing link relationship from historical fire cases by connecting to the fire protection file management system to build an electrical fire disaster-causing link relationship database. The knowledge layer extraction and semantic understanding module is configured to build a dedicated corpus for the electrical field based on the multi-source data characteristics of the electrical fire field. It processes the full life cycle text of equipment, fault handling text and phenomenon risk association text through domain word embedding technology to identify the entity relationship patterns of electrical faults. The knowledge graph construction and risk modeling module is configured to build an electrical fire knowledge graph with dynamic evolution capabilities. It responds to changes in edge node data in real time through an incremental update trigger mechanism, optimizes graph storage by using partitioned lifecycle management, and resolves entity attribute conflicts by weighted fusion of multi-source data on electrical faults. It also builds an electrical fault disaster chain model that supports spatiotemporal backtracking. The intelligent reasoning and fire early warning module is configured to establish an electrical fire situation perception model based on the knowledge graph, adaptively optimize the early warning threshold by dynamically coupling real-time monitoring data and historical disaster-causing patterns, and generate an analysis report containing direct technical responsibility and indirect management responsibility based on a multi-level disaster-causing factor backtracking mechanism.

[0053] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for constructing a knowledge graph based on electrical fault-induced disasters, characterized in that, include: S1: Construct a multi-source data collection list of electrical fault-causing disasters, dynamically adjust the collection interval based on the safety threshold of electrical equipment, and extract the disaster-causing link relationship from historical fire cases by connecting to the fire protection file management system to construct an electrical fire disaster-causing link relationship database; S2: Based on the multi-source data characteristics of the electrical fire field, establish a dedicated corpus for the electrical field, and process the full life cycle text of equipment, fault handling text and phenomenon risk association text through domain word embedding technology to identify the entity relationship patterns of electrical faults; S3: Establish an electrical fire knowledge graph with dynamic evolution capabilities, respond to changes in edge node data in real time through an incremental update trigger mechanism, optimize graph storage by adopting partition lifecycle management, and resolve entity attribute conflicts based on the weighted fusion of multi-source data on electrical faults to resolve conflicts, and construct an electrical fault disaster chain model that supports spatiotemporal backtracking. S4: Based on the knowledge graph, establish an electrical fire situation perception model, and adaptively optimize the early warning threshold by dynamically coupling real-time monitoring data with historical disaster-causing patterns. Based on a multi-level disaster-causing factor backtracking mechanism, generate an analysis report that includes direct technical responsibility and indirect management responsibility.

2. The method according to claim 1, characterized in that, The specific process for constructing the multi-source data collection list of electrical fault-causing disasters is as follows: Dynamic acquisition rules are set based on electrical equipment safety thresholds, and the acquisition interval is dynamically adjusted according to the deviation between real-time monitoring data and safety thresholds through an adaptive sampling frequency algorithm. Establish an interface protocol for the fire protection archive system, and extract equipment entities, fault entities, and disaster-causing factor entities from historical fire cases through an electrical fire entity precision identification algorithm, and construct a disaster-causing link relationship mapping table. Deploy time synchronization services at the edge nodes of electrical fire monitoring, and use a transmission delay correction model to iteratively correct the timestamps of multi-source collected data to generate spatiotemporally consistent standard data frames.

3. The method according to claim 1, characterized in that, The fire protection record management system includes: An unstructured database for storing historical fire case reports, a preprocessing module to support entity recognition algorithms, a causal relationship pattern library built based on knowledge of electrical fires, and a structured database for storing standardized disaster-causing link relationships are all included. The system interacts with the knowledge graph through system interfaces to build the system's data interaction and entity relationship extraction.

4. The method according to claim 1, characterized in that, The electrical field-specific corpus includes: The system collects real-time equipment operation status text through an industrial IoT platform interface to construct an equipment operation status corpus layer; extracts historical fire case reports from a fire protection file management system to construct a fire case corpus layer; converts time-series monitoring data into natural language descriptions through a sensor data conversion module to construct a phenomenon association corpus layer; establishes a corpus quality assessment mechanism, performs domain relevance verification on the collected corpus based on an electrical field professional dictionary, and filters out corpus that does not meet the requirements; and sets up a corpus version management mechanism to dynamically update and maintain the corpus based on electrical equipment updates and standard revisions.

5. The method according to claim 1, characterized in that, The domain term embedding technology includes: We construct a multi-source hierarchical corpus system unique to the field of electrical fires, which integrates model parameters in the full life cycle text of equipment, causal descriptions in fault handling text, and natural language descriptions converted from sensor time-series data to form an electrical fault semantic enhancement corpus. Based on electrical engineering lexicon and fault causal chain knowledge, a domain semantic enhancement model is constructed. Electrical equipment association constraints and fault causal relationship constraints are introduced during word vector training to maintain close semantic association between terms with causal relationships in the vector space. We designed a semantic verification mechanism for the electrical domain and verified the accuracy of word vectors in semantic reasoning of electrical faults by constructing a test set containing typical fault modes, disaster-causing paths and causal relationships. A dynamic semantic evolution model is established. Based on newly collected fault cases and equipment operation status data, the word vector space is continuously adjusted and optimized to adapt the semantic representation to the semantic evolution brought about by the aging of electrical equipment and dynamic factors of environmental changes.

6. The method according to claim 1, characterized in that, The entity relationship model of the electrical fault includes: Establish a four-element relationship model of equipment-fault-phenomenon-environment, where the equipment entity and the fault entity are connected through the occurrence relationship, the fault entity and the phenomenon entity are connected through the manifestation relationship, and the environment entity and the fault entity are connected through the induction relationship. A causal relationship chain model for electrical disasters is constructed. Based on the analysis of historical fire cases, a multi-level causal transmission relationship is established, which includes insulation aging, short circuit, overheating, and fire, forming a complete description of the disaster path. A time-series correlation model is also designed to establish a "time correlation" relationship between equipment status data and the time of failure occurrence, and to model the time series of the failure development process. A responsibility association model is established, which associates faulty entities with responsible parties such as personnel and management systems through two types of relationship: management responsibility and technical responsibility. This supports subsequent responsibility tracing and analysis, and the confidence level of entity relationships from different sources is assessed based on a multi-source data credibility weighting mechanism.

7. The method according to claim 1, characterized in that, The process of responding to edge node data changes in real time through the incremental update triggering mechanism is as follows: Deploy real-time data listeners at edge nodes to perform frame format verification and timestamp normalization on the collected monitoring data, and perform message deduplication and batch identification. Based on the real-time comparison of electrical parameters and safety thresholds, incremental update tasks are generated and added to the update queue according to priority. The scheduler calls the corresponding entity recognizer and relation extractor to perform knowledge extraction according to priority. The extracted results are mapped to fields and converted to units using a standardized interface. Metadata describing the changes is generated for each result. Real-time incremental updates of the knowledge graph are completed by combining graph index and credibility weighting.

8. The method according to claim 1, characterized in that, The partition lifecycle management includes: Data is partitioned based on risk level and time dimension, and real-time monitoring data, historical case data and archived data are stored in three storage levels: hot zone, warm zone and cold zone, respectively. Differentiated storage strategies are implemented for different partitions: hot zone data is stored in full memory to support real-time querying, warm zone data is stored in compressed form to support periodic analysis, and cold zone data is compressed and archived to provide on-demand backtracking loading. Establish a partition pruning and on-demand backtracking mechanism, automatically select the corresponding partition based on spatiotemporal index and query conditions, provide a decompression and recovery interface for cold zone archived data, and conduct rapid backtracking analysis of historical disaster cases.

9. The method according to claim 1, characterized in that, The specific process for resolving entity attribute conflicts based on the weighted fusion of multi-source data on electrical fault-induced disasters is as follows: Establish a data credibility grading system for electrical fires, setting the current and temperature data collected in real time by electrical fire monitoring terminals as the highest credibility level, the equipment parameters and maintenance data recorded by the equipment management system as the medium credibility level, and the data recorded by manual inspections as the basic credibility level. An attribute fusion rule based on credibility weight is constructed, and differentiated fusion weights are assigned to attribute values ​​with different credibility levels. An electrical fault feature-priority conflict resolution mechanism is adopted. Based on the conflict of attribute values, the data with the highest matching degree with the electrical fault disaster chain features is selected first, and the latest valid data is selected in combination with timestamp information. Establish attribute evolution tracking and version management to record the resolution process and decision-making basis of each attribute conflict, forming a complete attribute change trajectory, and supporting retrospective analysis of the disaster process caused by electrical faults.

10. The method according to claim 1, characterized in that, The electrical fire situation awareness model includes: Construct a multidimensional coupled analysis model of electrical equipment operating status and environmental parameters, collect load current, cable temperature, ambient temperature and humidity and equipment aging coefficient in real time, and establish a dynamic correlation between electrical fault disaster risk and multi-source monitoring parameters; A disaster pattern matching engine based on historical electrical fire cases is constructed. The abnormal data monitored in real time is compared with the typical electrical fault development patterns to identify the early characteristics of three main disaster paths: overload heating, insulation deterioration, and poor contact. Establish a dynamic optimization mechanism for early warning thresholds. Based on historical equipment operation data, environmental conditions, and real-time load rate, combined with the fault evolution patterns in the knowledge graph, automatically adjust the early warning thresholds of key parameters such as temperature and current to form an early warning strategy that is adaptive to specific operating scenarios.

11. The method according to claim 1, characterized in that, The dynamically coupled real-time monitoring data and historical disaster-causing patterns include: Establish a correlation analysis mechanism between real-time monitoring data and historical disaster cases, and perform multi-dimensional feature matching between the currently collected load current and cable temperature data and typical electrical fault cases stored in the knowledge graph; Extract the variation patterns of key electrical parameters from historical disaster-causing patterns, including the duration of overload current, the threshold of temperature rise rate, and the decreasing trend of insulation resistance, and construct a feature template library for the development of electrical faults. By calculating the similarity between real-time data streams and feature templates, the system identifies the degree of matching between the current operating status and historical disaster-causing patterns. When the similarity exceeds a set threshold, an early warning is triggered, and disposal suggestions based on historical experience are generated.

12. The method according to claim 1, characterized in that, The multi-level disaster-causing factor backtracking mechanism includes: Direct cause tracing based on graph query: Based on the early warning event node, the graph query statement performs graph traversal query on the edges in the knowledge graph to extract the entities, attributes and relationships directly connected to the early warning event node. At the same time, the graph pattern matching algorithm is used to quickly locate the direct cause entity that caused the anomaly and generate a direct disaster path graph to find the directly responsible equipment and environmental factors. By analyzing the causal contribution of electrical faults, we can deeply explore the root causes. We take the direct cause entity as the starting point of the analysis, extract the historical data sequence of current and temperature within the time window, establish an electrical fault evolution model, quantify the contribution score of each disaster-causing factor to the current warning event, identify key disaster-causing paths, and analyze their dynamic evolution law. Counterfactual liability determination based on electrical fault characteristics constructs a baseline factual scenario. Key electrical parameters are extracted from the actual disaster-causing event sequence as the basis for analysis. By adjusting parameters such as overload current, ambient temperature, or equipment aging, a counterfactual scenario is constructed to simulate the evolution path of each disaster-causing factor within its normal threshold range. By comparing the difference in the degree of loss between the baseline scenario and the counterfactual scenario, the influence weight of each disaster-causing factor is calculated. Combined with equipment management relationships in the knowledge graph, an analysis report containing direct technical responsibility and indirect management responsibility is generated to clarify the technical root cause and management responsibility of the accident.

13. A knowledge graph construction system based on electrical fault-induced disasters, used to execute the method as described in any one of claims 1-12, characterized in that, include: The multi-source heterogeneous data collaborative acquisition module is configured to build a multi-source data acquisition list of electrical faults, dynamically adjust the acquisition interval based on the safety threshold of electrical equipment, and extract the disaster-causing link relationship from historical fire cases by connecting to the fire protection file management system to build an electrical fire disaster-causing link relationship database. The knowledge layer extraction and semantic understanding module is configured to build a dedicated corpus for the electrical field based on the multi-source data characteristics of the electrical fire field. It processes the full life cycle text of equipment, fault handling text and phenomenon risk association text through domain word embedding technology to identify the entity relationship patterns of electrical faults. The knowledge graph construction and risk modeling module is configured to build an electrical fire knowledge graph with dynamic evolution capabilities. It responds to changes in edge node data in real time through an incremental update trigger mechanism, optimizes graph storage by using partitioned lifecycle management, and resolves entity attribute conflicts by weighted fusion of multi-source data on electrical faults. It also builds an electrical fault disaster chain model that supports spatiotemporal backtracking. The intelligent reasoning and fire early warning module is configured to establish an electrical fire situation perception model based on the knowledge graph, adaptively optimize the early warning threshold by dynamically coupling real-time monitoring data and historical disaster-causing patterns, and generate an analysis report containing direct technical responsibility and indirect management responsibility based on a multi-level disaster-causing factor backtracking mechanism.

Images