CN-122024396-A - Online monitoring and advanced early warning method for fire hazards in enclosed space

CN122024396ACN 122024396 ACN122024396 ACN 122024396ACN-122024396-A

Abstract

The invention discloses an on-line monitoring and advanced early warning method for hidden danger of fire in a closed space, which relates to the technical field of environmental safety of a power distribution room and comprises the steps of collecting particle information of each closed space through an optical sensor to generate particle sampling data of each closed space, carrying out feature extraction on the particle sampling data of each closed space to generate a corresponding particle feature vector of the closed space, analyzing the feature vector of the particle sampling data of the space through an isolated forest model, dynamically updating a normal particle feature baseline of the closed space based on the analyzed stable normal particle sampling data of the closed space, and further ensuring the accuracy of normal identification by ensuring that the particle feature baseline of the normal closed space can be adjusted in real time according to factors such as working conditions, environment and the like, and comprehensively judging the particle feature baseline of the normal closed space and the abnormal closed space by setting the normal particle feature baseline of the normal closed space.

Inventors

MENG CHAO
CHEN YI
ZHU NENGAN

Assignees

安徽健驰智能科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260228

Claims (10)

1. The on-line monitoring and advanced early warning method for the hidden danger of the fire disaster in the closed space is characterized by comprising the following steps: S1, collecting particle information of each enclosed space through an optical sensor, and generating particle sampling data of each enclosed space; s2, extracting features of the sampling data of each closed space particle to generate a corresponding closed space particle feature vector; S3, analyzing and processing the feature vectors of the particles in each closed space based on the isolated forest model to obtain the feature vectors of the particles in the normal closed space and the feature vectors of the particles in the abnormal closed space; S4, constructing a normal closed space particle characteristic baseline based on the normal closed space particle characteristic vector; s5, marking the actual abnormal historical closed space particle feature vectors in the historical closed space particle feature vectors to obtain historical abnormal feature samples, and constructing an abnormal closed space particle feature baseline based on the historical abnormal feature samples; s6, generating first analysis data based on deviation from the characteristic vector of the closed space particles to a characteristic baseline of the normal closed space particles; S7, outputting the maximum cosine similarity based on cosine similarity from the feature vector of the closed space particle to the feature base line of each abnormal closed space particle, and generating second analysis data; And S8, carrying out normalization processing on the first analysis data, carrying out weighted calculation on the normalized first analysis data and the normalized second analysis data based on the set confidence weight data to obtain comprehensive analysis data, judging whether the comprehensive analysis data is greater than or equal to a set comprehensive analysis threshold value, and if so, generating abnormal early warning data.
2. The method for on-line monitoring and advanced warning of fire hazards in an enclosed space according to claim 1, wherein the step S1 comprises the following steps: s1.1, one end of a plurality of capillary sampling tubes are arranged to respectively extend into each enclosed space, and the other end of each capillary sampling tube is connected with a high-speed gas circuit switching valve and is connected with an optical sensor through the high-speed gas circuit switching valve; s1.2, the high-speed gas circuit switching valve sequentially sucks and transmits particles in the air of the closed space communicated with each capillary sampling tube to the optical sensor according to the setting sequence, and then sampling data of the particles in each closed space are generated.
3. The method for on-line monitoring and advanced warning of fire hazards in an enclosed space according to claim 1, wherein the step S2 comprises the following steps: And calculating and analyzing the mean value, standard deviation, peak value, valley value, kurtosis, skewness, first derivative, average difference index, linear fitting slope, particle size distribution characteristic and channel ratio of the sampling data of each closed space particle in the first sliding window, and arranging the calculation results according to the setting sequence to generate corresponding closed space particle characteristic vectors.
4. The method for on-line monitoring and advanced warning of fire hazards in an enclosed space according to claim 1, wherein the step S4 comprises the following steps: S4.1, judging whether the proportion of the feature vectors of the particles in each closed space in the second sliding window which is recently set is a normal feature vector of the particles in the closed space is larger than a normal judgment threshold value of the set feature vector; And S4.2, if so, setting the normal closed space particle characteristic vector as a stable normal closed space particle characteristic vector, and constructing a normal closed space particle characteristic baseline based on the stable normal closed space particle characteristic vector.
5. The method for on-line monitoring and advanced warning of fire hazards in an enclosed space according to claim 1, wherein the step S4 comprises the following steps: s4.A, performing cluster analysis on the normal closed space particle feature vectors based on a DBSCAN clustering algorithm to obtain a plurality of normal closed space particle feature clusters; S4.B, constructing baselines based on the normal closed space particle characteristic vectors in each normal closed space particle characteristic cluster respectively, and obtaining a plurality of normal closed space particle characteristic baselines; And S4.C, associating each normal closed space particle characteristic baseline with a corresponding closed space.
6. The method for on-line monitoring and advanced warning of fire hazards in an enclosed space according to claim 4, further comprising dynamically updating a normal enclosed space particle characteristic baseline, comprising the steps of: Marking the normal closed space particle characteristic baseline corresponding to the second sliding window as a standby normal closed space particle characteristic baseline; Judging whether the characteristic difference between the standby normal closed space particle characteristic baseline and the normal closed space particle characteristic baseline in use is larger than a set normal baseline updating difference threshold value or whether the time interval of a second sliding window which is currently set and corresponds to the normal closed space particle characteristic baseline in use is larger than a set normal baseline updating time threshold value; If any judgment is yes, the normal closed space particle characteristic baseline to be used is used for updating the normal closed space particle characteristic baseline in use.
7. The method for on-line monitoring and advanced warning of fire hazards in an enclosed space according to claim 1, wherein the step S5 comprises the following steps: S5.1, carrying out cluster analysis on the historical abnormal characteristic samples based on a DBSCAN clustering algorithm to obtain a plurality of closed space particle abnormal characteristic clusters; s5.2, respectively setting corresponding abnormal labels to be associated with abnormal characteristic clusters of the particles in each enclosed space; S5.3, constructing baselines based on abnormal characteristic clusters of each closed space particle respectively to obtain a plurality of abnormal closed space particle characteristic baselines; S5.4, associating each abnormal closed space particle characteristic baseline with the corresponding closed space.
8. The method for on-line monitoring and advanced warning of fire hazards in an enclosed space according to claim 5, wherein the step S6 comprises the following steps: s6.1, searching out a normal closed space particle characteristic baseline associated with the closed space according to the closed space corresponding to the closed space particle characteristic vector; s6.2, calculating deviation from the characteristic vector of the closed space particles to the searched characteristic baseline of the normal closed space particles, and generating first analysis data.
9. The method for on-line monitoring and advanced warning of fire hazards in an enclosed space according to claim 7, wherein the step S7 comprises the following steps: s7.1, judging whether abnormal closed space particle feature vectors exist or not; S7.2, if so, calculating cosine similarity from the abnormal closed space particle characteristic vector to each abnormal closed space particle characteristic baseline with the same corresponding closed space, and generating the maximum cosine similarity of each abnormal closed space particle characteristic vector and the corresponding closed space into second analysis data; and S7.3, calculating and analyzing the cosine similarity from the uncomputed closed space particle feature vector to the corresponding closed space identical abnormal closed space particle feature base line, and generating the cosine similarity with the maximum value of each closed space particle feature vector and the corresponding closed space into second analysis data.
10. The method for on-line monitoring and advanced warning of fire hazards in an enclosed space according to claim 1, wherein the step S8 comprises the following steps: S8.1, normalizing the first analysis data to a (0, 1) interval to obtain first analysis normalized data, wherein the first analysis normalized data is more normal when the first analysis normalized data is closer to 0; s8.2, weighting the first analysis normalized data by using the set normal channel weight, weighting the second analysis data by using the set abnormal channel weight, and summing the weighted results to obtain comprehensive analysis data; s8.3, judging whether the comprehensive analysis data is larger than or equal to the set comprehensive analysis data, if so, collecting the comprehensive analysis data and the corresponding closed space, and generating abnormal early warning data; and S8.4, executing alarm feedback operation according to the abnormal early warning data.

Description

Online monitoring and advanced early warning method for fire hazards in enclosed space Technical Field The invention relates to the technical field of environmental safety of power distribution rooms, in particular to an on-line monitoring and advanced early warning method for fire hazards in a closed space. Background Most of electric equipment operates in a strong electromagnetic environment, discharge faults are easy to form when insulating materials are degraded, the equipment is locally overheated due to discharge, the temperature is higher after an arc is formed, the equipment is easy to burn, and on the other hand, the overheating phenomenon is easy to occur when the through-flow equipment operates for a long time due to the thermal effect of current. Thus, discharge and overheat detection of electrical equipment is an important point in ensuring safe operation thereof. Aiming at overheat monitoring of electrical equipment, the prior art cannot realize active early warning, has a limited application scene, still cannot complement the technical short plates of power distribution station houses and transformer substations in safety monitoring, such as linear temperature sensing detectors which need to reach a certain temperature to trigger alarm and do not have hidden danger early warning capability, the infrared video detectors usually adopt two wave bands and three wave bands, are greatly influenced by ambient light and shielding, are easy to alarm by mistake or can not alarm due to saturation of the detectors when the ambient temperature is higher, have lower alarm sensitivity of point type smoke sensing and temperature sensing detectors, and are used for passive submerged type detection, a large amount of smoke is needed to gather to alarm, the false alarm rate is high, the influence of environment is easy, the insulating materials adopted by equipment such as an air suction type smoke sensing detector transformer, a reactor and a cable are needed to reach more than 300 ℃ to generate smoke, and the safety problem is not solved at the moment, and the like can not realize safety early warning. Disclosure of Invention The invention aims to provide an on-line monitoring and advanced early warning method for fire hazards in a closed space, which aims to solve at least one of the defects in the prior art. In order to achieve the purpose, the invention provides the technical scheme that the method for on-line monitoring and advanced early warning of the hidden danger of the fire disaster in the closed space comprises the following steps: S1, collecting particle information of each enclosed space through an optical sensor, and generating particle sampling data of each enclosed space; s2, extracting features of the sampling data of each closed space particle to generate a corresponding closed space particle feature vector; S3, analyzing and processing each closed space particle characteristic vector based on the isolated forest model to obtain a normal closed space particle characteristic vector and an abnormal closed space particle characteristic vector, and analyzing isolated closed space particle characteristic vectors in all the closed space particle characteristic vectors through the isolated forest model to set the isolated closed space particle characteristic vectors as the abnormal closed space particle characteristic vectors, wherein other closed space particle characteristic vectors are set as the normal closed space particle characteristic vectors. S4, constructing a normal closed space particle characteristic baseline based on the normal closed space particle characteristic vector, and further dynamically updating the normal closed space particle characteristic baseline based on the normal closed space particle characteristic vector in the second sliding window; s5, marking the actual abnormal historical closed space particle feature vectors in the historical closed space particle feature vectors to obtain historical abnormal feature samples, and constructing an abnormal closed space particle feature baseline based on the historical abnormal feature samples; S6, based on the deviation from the characteristic vector of the closed space particles to the characteristic baseline of the normal closed space particles (the deviation can be represented by calculating the Markov distance or the Z-score comprehensive deviation), analyzing whether the characteristic vector of the closed space particles is normal or not, and generating first analysis data; s7, based on cosine similarity from the feature vector of the closed space particle to the feature base line of each abnormal closed space particle, analyzing whether the feature vector of the closed space particle is abnormal, outputting the maximum cosine similarity, and generating second analysis data; And S8, carrying out normalization processing on the first analysis data, carrying out weighted calculation on the normalized first analysis data and th