CN-122020262-A - Industrial enterprise-oriented emergency fire event identification method and system

CN122020262ACN 122020262 ACN122020262 ACN 122020262ACN-122020262-A

Abstract

The application relates to the technical field of intelligent fire control, in particular to an emergency fire control event identification method and system for industrial enterprises, which provides the following scheme, the on-site perception data are collected through the visible light, thermal infrared and gas spectrum multi-mode sensing module, brightness change, temperature gradient and spectral line intensity characteristics are extracted under a unified space-time reference frame, a multi-mode characteristic diagram is constructed, and a fire suspected area is extracted. Aiming at the suspected region, a thermal plume field model is constructed based on thermal convection and smoke plume dynamics mechanism, a multi-mode prediction track is generated under the assumption of true fire, the prediction track and actual observation data are subjected to space and time registration, and a cross-mode consistency score is calculated to judge whether the fire is true or false. The application can integrate multisource physical information to realize the counter facts disambiguation of pseudo fires such as welding flowers, steam, hot exhaust and the like, and obviously improve the accuracy of fire identification and response timeliness in complex industrial environments.

Inventors

ZHANG JINFENG
LI CONGCONG
LI WENJING
HUANG DONGMEI
HE HAO
ZHU KAIMING

Assignees

浙江省应急管理科学研究院(浙江省安全生产技术检测检验中心、浙江省危险化学品登记中心)

Dates

Publication Date: 20260512
Application Date: 20260414

Claims (10)

1. The utility model provides an emergent fire control incident identification method towards industrial enterprise, is applied to disaster prevention monitor platform, its characterized in that, disaster prevention monitor platform is used for gathering and handling industrial enterprise scene's multimode perception data, disaster prevention monitor platform includes visible light sensing module, thermal infrared sensing module and gas spectrum sensing module, the method includes: acquiring multi-mode sensing data from the visible light sensing module, the thermal infrared sensing module and the gas spectrum sensing module, and extracting a fire suspected area; Inputting the multi-mode sensing data corresponding to the fire suspected region into a preset thermal plume field model to generate a multi-mode prediction track under the condition of true fire, wherein the thermal plume field model is constructed based on the mechanism prior of thermal convection and smoke plume dynamics; calculating cross-modal consistency scores of multi-modal sensing data corresponding to the fire suspected areas according to the multi-modal prediction tracks; if the cross-mode consistency score is greater than or equal to a preset judgment threshold value, determining that the target event is a true fire and outputting an evidence pair comprising a space-time thermal plume and a spectral line alignment result, otherwise, determining that the target event is a false fire and outputting a corresponding false label.
2. The method for identifying an emergency fire event for an industrial enterprise according to claim 1, wherein the acquiring multi-modal sensing data from the visible light sensing module, the thermal infrared sensing module and the gas spectrum sensing module comprises: performing time synchronization and space calibration on the visible light image sequence acquired by the visible light sensing module, the thermal infrared temperature field sequence acquired by the thermal infrared sensing module and the gas spectrum response data acquired by the gas spectrum sensing module, and establishing a unified space-time coordinate reference frame; And under the space-time coordinate reference frame, carrying out feature extraction and joint analysis on the multi-modal sensing data to obtain a multi-modal feature map for representing brightness change, temperature gradient and spectral line intensity change.
3. The industrial enterprise-oriented emergency fire event recognition method of claim 2, wherein the feature extraction and joint analysis are performed on the multi-modal awareness data to obtain a multi-modal feature map for characterizing brightness variation, temperature gradient and spectral line intensity variation, comprising: Carrying out dynamic texture analysis on the visible light image sequence, extracting parameters of pixel brightness change rate, edge oscillation frequency and optical flow direction consistency, and obtaining image mode characteristics; Calculating non-Gaussian indexes of time temperature gradient, space heat conduction diffusivity and local temperature distribution for the thermal infrared temperature field sequence to identify a thermal plume region which is continuously heated and has buoyancy lifting characteristics, so as to obtain temperature modal characteristics; extracting the positions, half-width and intensity change rate of absorption peaks and radiation peaks of the gas spectral response data in the characteristic bandwidth range, and performing similarity matching with a preset fuel spectral line model to distinguish a characteristic spectral line of a combustion product from a steam or hot gas flow scattering spectral line so as to obtain a gas modal characteristic; and constructing a multi-mode fusion tensor based on each mode characteristic, and calculating a multi-mode characteristic diagram according to the multi-mode fusion tensor.
4. The industrial enterprise-oriented emergency fire event recognition method of claim 3, wherein computing a multi-modal feature map from the multi-modal fusion tensor comprises: Carrying out standardization and dimension unification processing on the multi-modal fusion tensor according to three dimensions of space, time and modes, constructing a common measurement domain taking brightness change, temperature gradient and spectral line intensity as main components, and giving initial weights to the main components according to on-site environment parameters; Performing robust tensor decomposition on the common measurement domain to obtain a low-rank component for representing cross-modal common variation and a sparse component for representing instantaneous noise, reflection flicker and local vapor interference, and taking the low-rank component as a candidate consistency response field; Calculating a modal correlation matrix and a typical correlation coefficient of the low-rank component in a preset sliding time window, and obtaining a correlation intensity map and a time mutation map by combining statistics of a change point test; The physical feasibility screening is carried out on the candidate consistency response field to obtain a physical consistency response diagram, wherein the physical feasibility screening comprises the steps of reserving a voxel region which simultaneously meets the heat convection and smoke plume dynamics, a voxel region with the ascending direction consistent with the visible light movement direction, a voxel region with the continuously enhanced spectral bandwidth intensity along with time and a voxel region which passes the approximate inspection of mass conservation and energy balance; performing weighted combination on the physical consistency response graph, the related intensity graph and the time mutation graph to obtain a primary characteristic graph, wherein the weight of the weighted combination is adjusted according to wind speed, temperature and humidity; And carrying out space-time connected domain screening and morphological constraint on the primary feature map to obtain a multi-mode feature map.
5. The method for identifying an emergency fire event for an industrial enterprise according to claim 2, wherein the extracting the suspected fire area comprises: In the multi-modal feature map, calculating a comprehensive response intensity map according to the brightness change rate, the temperature gradient amplitude and the spectral line intensity change rate, wherein the comprehensive response intensity map is used for representing the cooperative enhancement degree of the multi-modal features in the space region; Processing the comprehensive response intensity map through multi-threshold segmentation and a region growing algorithm, and extracting candidate response regions, wherein the multi-threshold segmentation is used for determining an initial range of response intensity, and the region growing algorithm is used for expanding response pixel groups of the initial range; detecting the time consistency of the candidate response area, calculating the cooperative correlation coefficient of the brightness change direction, the temperature rise rate and the spectral line change trend, and reserving the area corresponding to the positive correlation of the cooperative correlation coefficient to obtain a time consistency area; And carrying out dynamic stability judgment treatment on the time consistency region by combining with the on-site environment parameters, and outputting a fire suspected region through differential smoothing.
6. The industrial enterprise-oriented emergency fire event recognition method of claim 1, wherein the thermal plume field model comprises an input layer, an implied layer, and an output layer, wherein: The input layer is used for receiving multi-mode sensing data corresponding to the fire suspected region, establishing a characteristic sequence according to input parameters, and calculating a fitting value of the characteristic sequence; The hidden layer is used for carrying out coupling calculation on the fitting value based on a mechanism prior of thermal convection and smoke plume dynamics, wherein the mechanism prior comprises a buoyancy driving equation, a momentum conservation equation and an energy balance equation, and the hidden layer is used for solving an ascending speed field, a temperature distribution field and a plume top diffusion radius of the thermal plume in a time sequence in a numerical discrete mode and correcting the thermal plume field parameters in the calculation process through the ambient wind speed, the pressure difference and the obstacle distribution; The output layer is used for generating a multi-mode prediction track under the condition of true fire according to the thermal plume field parameters obtained by calculation of the hidden layer.
7. The method for identifying an emergency fire event for an industrial enterprise according to claim 6, wherein the generating the multi-modal predicted trajectory under the assumption of a real fire comprises: according to the thermal plume field parameters, calculating the ascending speed distribution of a thermal plume central axis, the height of a plume top and a temperature space-time evolution curve, and obtaining the radiation intensity distribution of a thermal plume surface layer based on an energy conservation relation; Projecting the radiation intensity distribution to a visible light and thermal infrared appearance measuring plane, and generating a corresponding visible light brightness prediction sequence and an infrared radiation brightness prediction sequence by combining a camera field angle, a distance attenuation coefficient and a sensor spectral response characteristic; According to the temperature distribution field and the fuel type information, calculating the characteristic spectral line emission intensity and the change rate with time of the combustion products through a preset radiation transmission model to obtain a time response curve of the spectral bandwidth; And synchronizing and registering the visible light brightness prediction sequence, the infrared radiation brightness prediction sequence and the time response curve of the spectral bandwidth under the unified space-time coordinates to generate a multi-mode prediction track.
8. The industrial enterprise-oriented emergency fire event recognition method of claim 1, wherein calculating a cross-modal consistency score of multi-modal awareness data corresponding to the fire suspected region according to the multi-modal prediction trajectory comprises: Carrying out spatial registration and time alignment on the multi-modal prediction track and corresponding multi-modal sensing data, and establishing an observation corresponding relation; According to the observation corresponding relation, calculating a consistency measurement result of each mode, wherein the consistency measurement result of each mode comprises a consistency measurement result of a visible light mode, a consistency measurement result of a thermal infrared mode and a consistency measurement result of a gas spectrum mode; And calculating the cross-modal consistency score according to the consistency measurement result.
9. An industrial enterprise-oriented emergency fire event recognition system for implementing an industrial enterprise-oriented emergency fire event recognition method according to any one of claims 1-8, the system comprising: the multi-mode sensing acquisition module is used for acquiring visible light images, thermal infrared temperature fields and gas spectrum response data of industrial enterprise sites, and performing time synchronization and space calibration on each mode sensing data; The feature fusion analysis module is used for carrying out feature extraction and joint analysis on the calibrated multi-modal sensing data, constructing a multi-modal fusion tensor and generating a multi-modal feature map for representing brightness change, temperature gradient and spectral line intensity change; the thermal plume field modeling module is used for receiving multi-mode sensing data corresponding to a fire suspected region, calculating thermal plume field parameters based on a mechanism priori of thermal convection and smoke plume dynamics, and generating a multi-mode prediction track under a true fire assumption; And the cross-modal consistency judging module is used for calculating cross-modal consistency scores according to the multi-modal prediction tracks and the actual observation data, judging whether the target event is a true fire or a false fire according to a preset threshold value, and outputting corresponding evidence pairs or pseudo labels.
10. The industrial enterprise-oriented emergency fire event recognition system of claim 9, wherein the feature fusion resolution module comprises: the modal feature extraction unit is used for respectively carrying out dynamic texture analysis, temperature gradient calculation and spectral line feature extraction on the visible light image sequence, the thermal infrared temperature field sequence and the gas spectral response data to obtain brightness variation parameters, temperature distribution parameters and spectral intensity variation parameters; the multi-modal tensor construction unit is used for constructing a multi-modal fusion tensor based on the brightness variation parameter, the temperature distribution parameter and the spectrum intensity variation parameter, and carrying out unified processing on space, time and dimension on the multi-modal fusion tensor; and the characteristic association calculation unit is used for calculating the correlation coefficient and the change trend among the modes according to the multi-mode fusion tensor and generating a multi-mode characteristic diagram for representing the cooperative change relation of brightness, temperature and spectral lines.

Description

Industrial enterprise-oriented emergency fire event identification method and system Technical Field The application relates to the technical field of intelligent fire protection, in particular to an emergency fire event identification method and system for industrial enterprises. Background Existing industrial fire monitoring relies on single-mode sensing means such as flame detection based on visible light video or anomaly identification based on infrared temperature. The method is often influenced by reflection, heat radiation, steam atomization and dust interference in complex industrial scenes, and is easy to misreport or miss report of fire. Particularly, when high similarity phenomena such as welding sparks, hot exhaust air or steam exhaust exist, the discrimination characteristics of a single mode cannot effectively distinguish real combustion from non-combustion phenomena, so that the system frequently and mistakenly triggers an alarm to influence the production continuity and the emergency decision reliability. Although different sensing information can be integrated by a part of multi-mode fusion schemes, consistency constraint of response of each mode cannot be realized in a physical layer, the problems of fuzzy judgment basis, longer response time delay and insufficient credibility still exist, and the requirement of high-risk industrial site on early-stage accurate identification of fire is difficult to meet. In order to solve the problems, the application designs an emergency fire event identification method and an emergency fire event identification system for industrial enterprises. Disclosure of Invention The application aims to solve the technical problems of the prior art, provides an emergency fire event identification method and system for industrial enterprises, the on-site perception data are collected through the visible light, thermal infrared and gas spectrum multi-mode sensing module, brightness change, temperature gradient and spectral line intensity characteristics are extracted under a unified space-time reference frame, a multi-mode characteristic diagram is constructed, and a fire suspected area is extracted. Aiming at the suspected region, a thermal plume field model is constructed based on thermal convection and smoke plume dynamics mechanism, a multi-mode prediction track is generated under the assumption of true fire, the prediction track and actual observation data are subjected to space and time registration, and a cross-mode consistency score is calculated to judge whether the fire is true or false. In order to achieve the above purpose, the present application provides the following technical solutions: The utility model provides an emergent fire control incident identification method towards industrial enterprise, is applied to disaster prevention monitor platform, disaster prevention monitor platform is used for gathering and handling industrial enterprise scene's multimode perception data, disaster prevention monitor platform includes visible light sensing module, thermal infrared sensing module and gaseous spectrum sensing module, the method includes: acquiring multi-mode sensing data from the visible light sensing module, the thermal infrared sensing module and the gas spectrum sensing module, and extracting a fire suspected area; Inputting the multi-mode sensing data corresponding to the fire suspected region into a preset thermal plume field model to generate a multi-mode prediction track under the condition of true fire, wherein the thermal plume field model is constructed based on the mechanism prior of thermal convection and smoke plume dynamics; calculating cross-modal consistency scores of multi-modal sensing data corresponding to the fire suspected areas according to the multi-modal prediction tracks; if the cross-mode consistency score is greater than or equal to a preset judgment threshold value, determining that the target event is a true fire and outputting an evidence pair comprising a space-time thermal plume and a spectral line alignment result, otherwise, determining that the target event is a false fire and outputting a corresponding false label. The obtaining multi-modal sensing data from the visible light sensing module, the thermal infrared sensing module and the gas spectrum sensing module comprises: performing time synchronization and space calibration on the visible light image sequence acquired by the visible light sensing module, the thermal infrared temperature field sequence acquired by the thermal infrared sensing module and the gas spectrum response data acquired by the gas spectrum sensing module, and establishing a unified space-time coordinate reference frame; And under the space-time coordinate reference frame, carrying out feature extraction and joint analysis on the multi-modal sensing data to obtain a multi-modal feature map for representing brightness change, temperature gradient and spectral line intensity change. Performing