CN-122024429-A - Fire intelligent early warning system and method applied to power system

CN122024429ACN 122024429 ACN122024429 ACN 122024429ACN-122024429-A

Abstract

The invention discloses an intelligent fire early warning system and method applied to a power system, and relates to the technical field of knowledge maps; the method comprises the steps of constructing a power equipment fault-fire knowledge graph by utilizing entity and entity relations, constructing a space-time correlation clustering mechanism to obtain comprehensive alarm events, generating a comprehensive evidence set according to the power equipment fault-fire knowledge graph, calculating and outputting posterior probability by utilizing a Bayesian network to obtain the fault type of the comprehensive alarm events, carrying out path reasoning in the power equipment fault-fire knowledge graph according to the fault type, outputting all follow-up nodes related to the fault type to form a result chain, automatically generating an interpretable comprehensive alarm report, calculating a reasoning confidence score, and setting a verification-update mechanism to update the power equipment fault-fire knowledge graph.

Inventors

SHAO HONGFEI
TIAN YUQING
XU BING
BI JINGHAO
ZHU QIANQIAN
HOU XUTONG

Assignees

国网山东省电力公司济宁供电公司

Dates

Publication Date: 20260512
Application Date: 20251215

Claims (10)

1. The intelligent fire early warning method applied to the power system is characterized by comprising the following steps of: S100, collecting all fire measuring points and corresponding sensor equipment in a power system, collecting fire events occurring in the power system in the past, defining a standard data model of equipment-measuring points-events, collecting fire data of the power system in real time, preprocessing the fire data, and converting the fire data into the standard data model to form a space-time data lake; S200, defining entity types and entity relations based on fire events occurring in a past power system, and constructing a power equipment fault-fire knowledge graph by utilizing the entity and the entity relations; S300, when different sensors send out fire alarms, acquiring the alarm time and the alarm position of the sensors according to a standard data model in a space-time data lake, constructing a space-time correlation clustering mechanism to acquire a comprehensive alarm event, and generating a comprehensive evidence set according to a power equipment fault-fire knowledge graph; S400, projecting the comprehensive alarm event into a power equipment fault-fire knowledge graph to obtain a correlation subgraph by inference, matching the correlation subgraph with a comprehensive evidence set, and calculating and outputting posterior probability to obtain the fault type of the comprehensive alarm event by using a Bayesian network; S500, path reasoning is carried out in a power equipment fault-fire knowledge graph according to fault types, all subsequent nodes related to the fault types are output to form a result chain, and an interpretable comprehensive alarm report is automatically generated; And S600, calculating an inference confidence score, presetting a confidence score threshold, manually verifying when the inference confidence score is lower than the confidence score threshold, and setting a verification-updating mechanism to update the power equipment fault-fire knowledge graph.
2. The fire intelligent early warning method applied to the power system according to claim 1, wherein the specific steps of forming the spatio-temporal data lake in S100 are as follows: S101, collecting all fire measuring points and corresponding sensor equipment in a power system, and collecting fire events occurring in the past power system, wherein an uploaded monitoring data standard data model is defined as equipment-measuring point-event, and each data comprises a space-time tag which comprises a sensor unique ID, a time stamp and a position coordinate; s101, each measuring point in the power system monitors fire conditions in real time through a sensor, and uploads monitoring data; preprocessing the monitoring data, wherein the preprocessing comprises standardization, transient spike removal and missing value filling by an average method; And converting the preprocessed monitoring data into a standard data model, uploading the standard data model into a data lake, and forming the space-time data lake based on space-time labels of all the monitoring data.
3. The fire intelligent early warning method for the power system according to claim 2, wherein the specific steps of constructing the power equipment fault-fire knowledge graph by using the entity and the entity relationship in S200 are as follows: S201, based on fire events occurring in a past power system, extracting fault power equipment, fault types, fire symptoms and event consequences in the power system in the historical fire events as different types of entities; extracting a fault type generated by fault power equipment in a fire event, a fire sign caused by the fault type, an event result caused by the fault type, a spatial adjacent relation of the power equipment and a power connection relation as entity relations; S202, constructing a power equipment fault-fire knowledge graph by taking different types of entities as nodes and entity relations as edges, extracting structural knowledge triples in equipment specifications and professional overhaul experience, wherein the format of the knowledge triples is that the phenomena of cause, judgment and influence are generated, inputting the knowledge triples into the power equipment fault-fire knowledge graph, adding the entity nodes and the entity relation edges, and extracting the number of events with the same entity relation as the historical fire events as the edge weight of the power equipment fault-fire knowledge graph.
4. The fire intelligent early warning method applied to the power system according to claim 3, wherein the specific step of generating the comprehensive evidence set according to the power equipment fault-fire knowledge graph in S300 is as follows: s301, when different sensors send out fire alarms, obtaining sensor alarm time stamps and position coordinates according to a standard data model in a space-time data lake, wherein the construction of a space-time correlation clustering mechanism is specifically as follows: ; in the formula, the corelation represents a comprehensive alarm event after space-time Correlation clustering, deltaT represents absolute values of alarm time differences of different sensors, deltaS represents Euclidean distances of different sensors, symptomType represents alarm types, and f represents a clustering function, wherein the clustering rule of the clustering function is specifically as follows: Calculating absolute difference values by utilizing different sensor alarm time stamps according to the absolute value of the alarm time difference, presetting a time window threshold T th , and judging that the sensor time correlation of two corresponding alarms is high when DeltaT < T th , and the two alarms belong to the same alarm event; Aiming at Euclidean distance, the position coordinates of the alarm sensors are substituted into the Euclidean distance formula to be calculated, a distance threshold S th is preset, and when DeltaS < S th , the sensor space correlation of the two corresponding alarms is judged to be high, and the two alarms belong to the same alarm event; Aiming at the alarm types, extracting fire symptoms when different sensors are used for alarming, reasoning the fault types with the relationship sides of the fire symptoms of the different sensors in a power equipment fault-fire knowledge graph, and judging that the sensor corresponding to the two alarms has high causal correlation when the deduced fault types are the same, and belongs to the same alarm event; s302, after the different sensor alarms are subjected to association clustering to obtain comprehensive alarm events, clustered sensor alarm information, alarm values and fire symptoms are packaged to be used as an initial evidence chain.
5. The fire intelligent early warning method for the electric power system according to claim 4, wherein the step of calculating the output posterior probability of the evidence matching result by using the Bayesian network in S400 to obtain the fault type of the comprehensive alarm event comprises the following specific steps: S401, taking equipment of all sensor measuring points in the comprehensive alarm event as fault equipment nodes, taking each fault equipment node as a center in a power equipment fault-fire knowledge graph, and extracting nodes with relation edges with the fault equipment nodes to construct a correlation subgraph; matching the fire symptoms in the initial evidence chain with fire symptom nodes in the associated subgraph to obtain the same fire symptoms, and extracting fault types with relationship sides of the matched fire symptoms; aiming at each fault type, giving prior probability based on the historical equipment fault records, wherein the prior probability is the duty ratio of different fault types in the total times of the historical fault records; s402, constructing a Bayesian network, inputting the matched symptoms into the Bayesian network, and calculating the posterior probability of each fault type through a confidence coefficient propagation algorithm, wherein the formula is as follows: ; In the formula, P (F i |E) represents the posterior probability of the fault type F i , P (E|F i ) represents the possibility that the fire symptom E is observed when the fault type F i is true, the probability is learned by expert experience, P (F i ) represents the prior probability of the fault type F i , and F i represents the ith fault type.
6. The intelligent fire early warning method for electric power system according to claim 5, wherein the specific steps of automatically generating the interpretable comprehensive alarm report in S500 are as follows: S501, taking the fault type with the highest posterior probability as the fault type of the comprehensive alarm event, inputting the fault type into a power equipment fault-fire knowledge graph, extracting all event results caused by the fault type of the comprehensive alarm event, and forming a result chain according to the relationship edges of the event results in the power equipment fault-fire knowledge graph; S502, generating a comprehensive alarm report of a comprehensive alarm event according to a power equipment fault-fire knowledge graph reasoning result, wherein the comprehensive alarm report comprises an event root, an equipment fault type, key evidence, a fire sign corresponding to the fault type and a deduction occurrence path, namely a result chain.
7. The fire intelligent early warning method applied to the power system according to claim 6, wherein the step of setting a verification-update mechanism to update the power equipment fault-fire knowledge graph in S600 comprises the following specific steps: S601, calculating an inference confidence score, wherein the formula is as follows: ; In the formula, confidence represents an inference confidence score, evidence represents evidence intensity, and the evidence intensity represents the sum of all fire symptom nodes of the comprehensive alarm event and the relation side weights of the fault types; s602, presetting a confidence score threshold, and performing manual verification when the inference confidence score is lower than the confidence score threshold; The method comprises the steps of setting a verification-updating mechanism to update a power equipment fault-fire knowledge graph, specifically, after carrying out operation and maintenance processing according to a result chain of a comprehensive alarm event by a professional, feeding back and confirming a real fault type in a system, recording the comprehensive alarm event when the real fault type is consistent with power equipment fault-fire knowledge graph reasoning, when the real fault type is inconsistent with the power equipment fault-fire knowledge graph reasoning, if the real fault type is not judged to be a new fault type in the power equipment fault-fire knowledge graph and is supplemented to complete updating in the power equipment fault-fire knowledge graph, and if the real fault type is judged to be reasoning errors in the power equipment fault-fire knowledge graph, updating the relation edge weight of the real fault type and the comprehensive alarm event fire sign in the power equipment fault-fire knowledge graph.
8. The fire intelligent early warning system is applied to the power system and is characterized by comprising a data acquisition module, a power equipment fault-fire knowledge graph construction module, a comprehensive alarm event clustering module, a fault reasoning module, a report generation module and a verification updating module; the data acquisition module is used for acquiring all fire measuring points and corresponding sensor equipment in the power system, collecting fire events occurring in the past power system, defining a standard data model of equipment-measuring points-events, acquiring fire data of the power system in real time, preprocessing the fire data, and converting the fire data into the standard data model to form a space-time data lake; The power equipment fault-fire knowledge graph construction module is used for defining entity types and entity relations based on fire events occurring in a past power system and constructing a power equipment fault-fire knowledge graph by utilizing the entity and entity relations; The comprehensive alarm event clustering module is used for obtaining the alarm time and the alarm position of the sensors according to a standard data model in a space-time data lake when different sensors send out fire alarms, constructing a space-time correlation clustering mechanism to obtain a comprehensive alarm event, and generating a comprehensive evidence set according to a power equipment fault-fire knowledge graph; the fault reasoning module is used for projecting the comprehensive alarm event into a power equipment fault-fire knowledge graph to infer and obtain a correlation sub-graph, matching the correlation sub-graph with the comprehensive evidence set, and calculating and outputting posterior probability to obtain the fault type of the comprehensive alarm event by using a Bayesian network; The report generation module is used for carrying out path reasoning according to the fault type in the power equipment fault-fire knowledge graph, outputting all subsequent nodes related to the fault type to form a result chain, and automatically generating an interpretable comprehensive alarm report; the verification updating module is used for calculating the reasoning confidence score, presetting a confidence score threshold value, manually verifying when the reasoning confidence score is lower than the confidence score threshold value, and setting a verification-updating mechanism to update the power equipment fault-fire knowledge graph.
9. The fire intelligent early warning system applied to the power system according to claim 8, wherein the fault reasoning module comprises a priori probability unit and a posterior probability unit; the prior probability unit is used for projecting the comprehensive alarm event to the power equipment fault-fire knowledge graph to obtain a correlation subgraph by inference, matching the correlation subgraph with the comprehensive evidence set, and giving prior probability based on the historical equipment fault record aiming at each fault type; and the posterior probability unit is used for calculating and outputting posterior probability to obtain the fault type of the comprehensive alarm event by using the Bayesian network to the evidence matching result.
10. The fire intelligent early warning system applied to the power system according to claim 8, wherein the verification updating module comprises an inference verification unit and a map updating unit; The reasoning verification unit is used for calculating a reasoning confidence score, presetting a confidence score threshold value and carrying out manual verification when the reasoning confidence score is lower than the confidence score threshold value; the map updating unit is used for carrying out operation and maintenance processing according to the result chain of the comprehensive alarm event by professionals, feeding back and confirming the real fault type in the system, and updating the power equipment fault-fire knowledge map according to different judging results.

Description

Fire intelligent early warning system and method applied to power system Technical Field The invention relates to the technical field of knowledge maps, in particular to an intelligent fire early warning system and method applied to a power system. Background The power system is used as a core infrastructure of national economy and covers a plurality of links such as power generation, power transmission, transformation, power distribution and the like, and the safe and stable operation of the power system is directly related to industrial production, civil security and social order. The fire is one of the major potential safety hazards faced in the operation of the power system, and the power equipment is in a high-load, high-voltage and complex environment (such as outdoor high temperature, humidity, dust, corrosive gas and the like) working condition for a long time, so that the fire is easily caused by faults such as insulation aging, poor contact, short circuit, oil leakage, arc ignition and the like. The fire has the characteristics of strong burst, high spreading speed and wide influence range, and once the fire occurs, the fire can cause equipment burnout and power interruption, cause huge economic loss, can also cause chain reaction, even cause large-area power failure to surrounding power grid facilities, and seriously threaten public safety. The early stage takes 'civil air defense+single-point simple monitoring' as a core, relies on basic temperature sensing and smoke sensing detectors of key areas such as manual regular line inspection and transformer substation, is limited by human coverage and equipment functions, early warning is often delayed from fire development, and only can realize post-treatment or large-area fire warning. Along with the preliminary integration of an electrical technology and a communication technology, the fire early warning enters an automatic starting period, and the sensing monitoring device of parameters such as temperature, current and the like is combined with wired communication, so that the automatic reporting of abnormal data of equipment is realized, the limitation of purely manual recording is eliminated, the single-point data acquisition is still the main, and the system linkage capability is lacked. At present, the artificial intelligence development is that all subsystems in the traditional fire monitoring and early warning are independently operated, the information island phenomenon exists among different monitoring systems, all data cannot be integrated comprehensively, the traditional technology is that fire reasons, development trends and the like can not be prejudged and prevented in advance by judging fire reasons, development trends and the like according to experience on site by manpower due to various fire development changes, and a single sensor is easy to be interfered by environment (such as dust, water vapor and illumination), so that frequent false alarm is caused, and operation and maintenance resources are consumed. Disclosure of Invention The invention aims to provide an intelligent fire early warning system and method applied to a power system, so as to solve the problems in the prior art. In order to achieve the above purpose, the present invention provides the following technical solutions: an intelligent fire early warning method applied to a power system, the method comprises the following steps: S100, collecting all fire measuring points and corresponding sensor equipment in a power system, collecting fire events occurring in the power system in the past, defining a standard data model of equipment-measuring points-events, collecting fire data of the power system in real time, preprocessing the fire data, and converting the fire data into the standard data model to form a space-time data lake; further, the specific steps of forming the spatiotemporal data lake are as follows: S101, collecting all fire measuring points and corresponding sensor devices in a power system, collecting fire events occurring in the past power system, defining an uploaded monitoring data standard data model as device-measuring point-event, wherein each data comprises a space-time tag which comprises a sensor unique ID, a time stamp and a position coordinate, unifying data formats, solving the heterogeneous problem of data of different sensors and different historical events, realizing interconnection and intercommunication of the data, and laying a foundation for data analysis of subsequent cross-device and cross-scene. And the time and space attributes of the data are precisely positioned, so that the subsequent alarm clustering and fault positioning can be combined with space-time dimension, and the erroneous judgment caused by isolated analysis is avoided. S101, each measuring point in the power system monitors fire conditions in real time through a sensor, and uploads monitoring data; preprocessing the monitoring data, wherein the preprocessing compris