A fire-fighting hidden danger real-time monitoring method based on multi-sensor fusion

Abstract

The application discloses a kind of fire-fighting hidden danger real-time monitoring method based on multi-sensing fusion, it is related to fire safety monitoring technical field, comprising: collecting multi-source fire status data in monitoring area, generate standardized multi-source fire status data;Divide space risk unit, extract smoke candidate area data;Improved Gaussian smoke model is constructed, and smoke diffusion state and field source coupling intensity are generated;Smoke phase change state field is constructed, and smoke phase change state is generated;Multi-dimensional risk state sequence is constructed and PELT algorithm is executed, and risk state variable point set is generated;Risk grade result is generated by calculating continuous diffusion score, and fire linkage response processing is executed.The application realizes the continuous monitoring, accurate identification and risk grade evaluation of fire hazard evolution process by introducing improved Gaussian smoke model and PELT algorithm.

Description

Technical Field

[0001] This invention relates to the field of fire safety monitoring technology, and in particular to a method for real-time monitoring of fire hazards based on multi-sensor fusion. Background Technology

[0002] With the increasing number of electrical devices in industrial plants, data centers, warehouses, and commercial complexes, fire safety monitoring systems have become a crucial technical means for fire prevention and control. Existing fire monitoring technologies primarily use smoke detectors, heat detectors, infrared thermal imaging equipment, electricity monitoring equipment, and video surveillance equipment to collect data on the status of the monitored area, and then use the collected results to identify fire hazards and trigger alarms. Specifically, smoke detectors monitor changes in smoke concentration, heat detectors monitor changes in ambient temperature, infrared thermal imaging equipment monitors the distribution of heat sources, electricity monitoring equipment monitors current, voltage, and fault arc conditions, and video surveillance equipment provides on-site image information. By analyzing the monitoring data and combining it with alarm strategies, fire risk detection and fire-fighting linkage control can be achieved, leading to its widespread application in the field of fire safety.

[0003] Currently, research on fire hazard monitoring is gradually shifting from single-sensor monitoring to multi-source information fusion. Existing solutions typically aggregate smoke data, temperature data, infrared thermal imaging data, electricity consumption monitoring data, and video surveillance data, and determine hazard based on threshold comparison, rule matching, or simple fusion algorithms. However, different data sources are often processed independently, lacking a unified spatiotemporal correlation analysis mechanism, making it difficult to accurately describe the evolution of fire hazards from local anomalies to continuous development. Furthermore, existing technologies mostly rely on instantaneous abnormal states for judgment, failing to adequately utilize risk change trends and diffusion characteristics, and are easily affected by environmental interference factors, resulting in room for further improvement in the accuracy of hazard identification and early warning capabilities.

[0004] Therefore, how to provide a real-time monitoring method for fire hazards based on multi-sensor fusion is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a real-time fire hazard monitoring method based on multi-sensor fusion. This invention utilizes multi-source information fusion technologies such as smoke detection, temperature monitoring, infrared thermal imaging, power status monitoring, and video image analysis. It details a complete technical process from spatial risk unit construction, smoke candidate region extraction, recursive smoke plume modeling, smoke-heat-electricity three-field coupling analysis, smoke plume phase change state identification, risk state change point detection, risk level assessment, and fire-fighting linkage response. This achieves continuous monitoring and dynamic judgment of fire hazards from the generation of local anomalies, their continuous diffusion and evolution, to risk escalation. By establishing the coupling relationship between smoke plume diffusion state and field sources, combined with risk propagation topology analysis and change point detection mechanisms, it improves the ability to identify real fire hazards in complex environments. It effectively reduces false alarms caused by interference factors such as dust, water vapor, and changes in light intensity, and enhances the early detection and continuous diffusion identification capabilities of fire hazards. It possesses advantages such as high monitoring accuracy, strong anti-interference ability, timely risk identification, and high efficiency in fire-fighting linkage response.

[0006] A real-time monitoring method for fire hazards based on multi-sensor fusion according to an embodiment of the present invention includes: Collect multi-source fire status data within the monitoring area, and perform preprocessing on the multi-source fire status data to generate standardized multi-source fire status data; Based on the installation locations of cameras and sensors and the boundaries of the monitoring area, spatial risk units are divided. Standardized multi-source fire status data are mapped to the corresponding spatial risk units, and smoke candidate area data are extracted. An improved Gaussian smoke plume model is constructed, which converts smoke candidate region data into multi-scale recursive smoke plume objects, establishes a recursive association between the main smoke plume and the sub-smoke plumes, and constructs a smoke-heat-electric three-field coupled field source to perform coupling correction on the smoke plume diffusion state and generate the smoke plume diffusion state and field source coupling strength. Based on the smoke diffusion state and the field-source coupling strength, a smoke phase transition state field is constructed, the smoke phase transition threshold is calculated, and a phase transition determination is performed to generate the smoke phase transition state. A multidimensional risk state sequence is constructed based on the smoke puff diffusion state and smoke puff phase transition state. The PELT algorithm is then executed to construct a multidimensional consistent change point detection structure. Dynamic programming segmentation and candidate segmentation path pruning are performed based on the risk state confidence to generate a set of risk state change points. Based on the smoke plume diffusion state, smoke plume phase change state, and risk state change point set, a continuous diffusion score is calculated, a risk level result is generated, and a fire-fighting linkage response is executed.

[0007] Optionally, the multi-source fire status data specifically includes smoke concentration data, ambient temperature data, infrared calorific value data, current status data, fault arc status data, video frame data, spatial location data, and acquisition timestamp data.

[0008] Optionally, the preprocessing of multi-source fire status data specifically includes time alignment processing, spatial mapping processing, outlier removal processing, image frame denoising processing, and standardization processing.

[0009] Optionally, the extraction of smoke candidate region data includes: Read the camera installation location, sensor installation location, and monitoring area boundary; establish a spatial coordinate system according to the monitoring area boundary; divide the monitoring area into spatial risk units based on the camera's field of view coverage and the sensor's acquisition coverage; and generate a set of spatial risk units. Based on the spatial risk unit set, the spatial location data and collection timestamp data in the standardized multi-source fire status data are read. The smoke concentration data, ambient temperature data, infrared calorific value data, current status data and fault arc status data that fall into the same spatial risk unit and whose collection timestamp difference is less than or equal to 1 second are written into the same spatial risk unit to generate spatial risk unit status data. Read the spatial risk unit status data and corresponding video frames. Based on the camera installation location and the camera's field of view coverage, map the pixel areas in the video frames to the corresponding spatial risk units. Extract areas with a grayscale attenuation value greater than 0.18, an edge blur value greater than 0.25, and an area change rate greater than 0.10 for three consecutive frames as smoke candidate areas, and generate smoke candidate area data.

[0010] Optionally, the generated smoke cloud diffusion state and field-source coupling strength include: An improved Gaussian puff model is constructed, which includes a recursive puff module, a three-field recirculation coupling module, a coherent diffusion analysis module, and an entropy collapse correction module. The recursive smoke cloud module reads the smoke candidate region data corresponding to the same collection timestamp within the same spatial risk unit, determines the smoke candidate region with the largest area as the main smoke cloud node, and determines the remaining smoke candidate regions as child smoke cloud nodes. It also establishes recursive associated edges and parent-child mapping relationships based on the center distance and region overlap status, generating a tree-like recursive topology structure. The three-field recirculation coupling module reads smoke concentration data, ambient temperature data, infrared calorific value data, current status data, and fault arc status data to construct smoke field, thermal field, and electric field. It analyzes the temporal driving relationship between the changing states of the three fields. When the electric field drives the thermal field, the thermal field drives the smoke field, and the smoke field inversely enhances the thermal field or electric field, a smoke-heat-electric causal recirculation structure is generated, and the field-source coupling strength is calculated. The coherent diffusion analysis module reads the change in the center coordinates of each plume node between adjacent acquisition timestamps, generates the plume diffusion trajectory, calculates the diffusion direction, diffusion velocity, and concentration change rate, and constructs a cooperative diffusion state matrix based on the difference in direction, velocity, and concentration change rate between each plume node. When the degree of synchronous diffusion is greater than 0.65, a coherent diffusion structure of the plume is generated. The entropy collapse correction module reads the smoke plume concentration distribution state in the tree-like recursive topology, calculates the smoke plume information entropy change state, and generates a smoke plume entropy collapse structure when the smoke plume information entropy decreases for three consecutive collection timestamps and a smoke plume coherent diffusion structure exists. It then performs collapse correction on the initial smoke plume concentration peak, the initial smoke plume diffusion radius, and the initial smoke plume duration to generate the smoke plume diffusion state. The improved Gaussian smoke model was trained using the smoke diffusion state prediction error, field-source coupling strength error, and smoke coherent diffusion state error as joint optimization objectives. The parameters of the recursive smoke module, the three-field backflow coupling module, the coherent diffusion analysis module, and the entropy collapse correction module were continuously optimized. The training was completed when the change in joint loss corresponding to five consecutive iterations was less than 0.001.

[0011] Optionally, the phase transition state of the generated smoke cloud includes: Read the smoke plume diffusion state and field-source coupling strength corresponding to the continuous acquisition timestamps within the same spatial risk unit, extract the peak smoke plume concentration, smoke plume diffusion radius, smoke plume duration, synchronous diffusion degree and smoke plume information entropy change state, and calculate the smoke plume concentration growth rate, diffusion radius growth rate and duration growth rate; A smoke plume phase transition state field was constructed using spatial risk units and collection timestamps as indices. The smoke plume concentration growth rate, diffusion radius growth rate, duration growth rate, field-source coupling strength, synchronous diffusion degree, and smoke plume information entropy change state were multiplied by 0.25, 0.20, 0.15, 0.20, 0.10, and 0.10, respectively, and then added together to obtain the smoke plume phase transition order value, which was then written into the smoke plume phase transition state field. Read the smoke phase transition order values ​​corresponding to the first 12 collection timestamps in the smoke phase transition state field, calculate the average and standard deviation, and add the average to 1.50 times the standard deviation to obtain the smoke phase transition threshold; The current phase transition order value of the smoke plume in the phase transition state field is read and compared with the phase transition threshold of the smoke plume. When the phase transition order value of the smoke plume is greater than the phase transition threshold of the smoke plume and there is a coherent diffusion structure and an entropy collapse structure of the smoke plume, a phase transition transition is determined to have occurred, and a phase transition state of the smoke plume is generated.

[0012] Optionally, the generation of the risk state change point set includes: Based on the collection timestamp, the smoke plume diffusion state and smoke plume phase transition state within the same spatial risk unit are read, and the peak smoke plume concentration, smoke plume diffusion radius, smoke plume duration, field-source coupling strength, synchronous diffusion degree, smoke plume information entropy change state, smoke-heat-electric causal backflow structure, and smoke plume phase transition order value are extracted to construct a multidimensional risk state sequence; A risk propagation topology graph is established based on the positional adjacency relationship between spatial risk units, the direction of smoke cloud diffusion, and the field-source coupling strength. The multidimensional risk state sequence is written into the nodes of the risk propagation topology graph, and the consistency of smoke cloud diffusion direction, the difference in field-source coupling strength, and the difference in smoke cloud phase transition order value between adjacent spatial risk units are written into the edges of the risk propagation topology graph to generate a topological risk state graph. The PELT algorithm is executed based on the topological risk state diagram. According to the phase transition state of the smoke plume, the topological risk state diagram is divided into random disturbance state segment, phase transition state segment, and continuous diffusion state segment. Corresponding segmental cost weights are set for each segment to generate a phase transition driven segmental cost function. Hierarchical recursive PELT processing is performed based on a tree-like recursive topology. Recursive propagation paths between main smoke cloud nodes and child smoke cloud nodes are generated according to the parent-child mapping relationship. Change point detection is performed on the main smoke cloud nodes, child smoke cloud nodes, and recursive propagation paths according to the phase change driven piecewise cost function. Change points of the main smoke cloud, child smoke cloud, and recursive propagation paths are written into the same hierarchical change point group to generate the hierarchical topology joint change point cost. Based on the hierarchical topology joint change point cost and phase transition driven segmentation cost function, dynamic programming segmentation and candidate segmentation path pruning are performed. Candidate segmentation paths that do not conform to the risk propagation topology direction, parent-child recursive propagation order, and phase transition state are deleted. The change point location, collection timestamp, spatial risk unit number, hierarchical topology joint change point cost, and risk state confidence of the candidate segmentation path are retained to generate a risk state change point set.

[0013] Optionally, the execution of the fire alarm linkage response process includes: Read the set of smoke cloud diffusion state, smoke cloud phase transition state and risk state change points corresponding to the same judgment window within the same spatial risk unit, and extract the field source coupling strength, synchronous diffusion degree, smoke cloud phase transition order value, phase transition judgment result, risk state confidence and change point continuity ratio to generate a set of continuous diffusion judgment indicators. The continuous diffusion score is obtained by multiplying the source-field coupling strength, synchronous diffusion degree, smoke phase transition order value, risk state confidence, change point continuity ratio, and phase transition judgment result by 0.20, 0.20, 0.20, 0.15, and 0.05, respectively. When the continuous diffusion score is less than 0.45, a Level 1 risk is generated and a risk warning is issued. When the continuous diffusion score is greater than or equal to 0.45 and less than 0.65, a Level 2 risk is generated and an early warning is issued. When the continuous diffusion score is greater than or equal to 0.65 and less than 0.80, a Level 3 risk is generated and a coordinated action is taken to activate area smoke extraction, audible and visual alarms, and electricity monitoring. When the continuous diffusion score is greater than or equal to 0.80, a Level 4 risk is generated and a fire alarm, area power outage, smoke extraction activation, and emergency notification are issued.

[0014] The beneficial effects of this invention are: Compared with existing technologies, this invention provides a real-time fire hazard monitoring method based on multi-sensor fusion, which introduces an improved Gaussian smoke cloud model. Building upon traditional Gaussian smoke cloud diffusion analysis, it adds a recursive smoke cloud structure, a smoke-heat-electric three-field recirculation coupling mechanism, a smoke cloud coherent diffusion analysis mechanism, and entropy collapse correction. The recursive smoke cloud structure establishes a hierarchical relationship between main smoke cloud nodes and sub-smoke cloud nodes. The smoke-heat-electric three-field recirculation coupling mechanism establishes the propagation relationship between changes in the electric field, thermal field, and smoke field. This allows smoke cloud diffusion analysis to not only reflect the spatial distribution of smoke but also the evolutionary process of electrical anomalies, heat accumulation, and smoke diffusion during hazard formation, improving the ability to characterize fire hazard states and the accuracy of risk identification in complex scenarios.

[0015] This invention constructs a smoke plume phase transition state field based on smoke plume diffusion analysis results, and combines risk propagation topology PELT, hierarchical recursive PELT, and phase transition driven PELT. Risk propagation topology PELT describes the risk diffusion path using the propagation relationship between spatial risk units; hierarchical recursive PELT describes the hierarchical evolution process of the smoke plume using the recursive propagation relationship between the main smoke plume node and the sub-smoke plume node; and phase transition driven PELT utilizes the smoke plume phase transition state to participate in the change point detection process, enabling the change point detection results to simultaneously possess spatial propagation characteristics, hierarchical propagation characteristics, and phase transition evolution characteristics. Through the synergistic effect of the above structures, continuous identification of fire hazards from local anomalies and diffusion development to continuous diffusion processes is achieved. Furthermore, risk level results are generated by combining continuous diffusion scoring, improving the stability of risk status determination, the ability to identify risk evolution, and the accuracy of fire-fighting linkage response. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a real-time monitoring method for fire hazards based on multi-sensor fusion proposed in this invention; Figure 2This is a schematic diagram of the structure of an improved Gaussian smoke cloud model for a real-time monitoring method of fire hazards based on multi-sensor fusion proposed in this invention. Figure 3 This is a data flow diagram of the PELT algorithm for a real-time monitoring method for fire hazards based on multi-sensor fusion proposed in this invention. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0018] refer to Figure 1 , Figure 2 and Figure 3 A real-time monitoring method for fire hazards based on multi-sensor fusion includes: Collect multi-source fire status data within the monitoring area, and perform preprocessing on the multi-source fire status data to generate standardized multi-source fire status data; Based on the installation locations of cameras and sensors and the boundaries of the monitoring area, spatial risk units are divided. Standardized multi-source fire status data are mapped to the corresponding spatial risk units, and smoke candidate area data are extracted. An improved Gaussian smoke plume model is constructed, which converts smoke candidate region data into multi-scale recursive smoke plume objects, establishes a recursive association between the main smoke plume and the sub-smoke plumes, and constructs a smoke-heat-electric three-field coupled field source to perform coupling correction on the smoke plume diffusion state and generate the smoke plume diffusion state and field source coupling strength. Based on the smoke diffusion state and the field-source coupling strength, a smoke phase transition state field is constructed, the smoke phase transition threshold is calculated, and a phase transition determination is performed to generate the smoke phase transition state. A multidimensional risk state sequence is constructed based on the smoke puff diffusion state and smoke puff phase transition state. The PELT algorithm is then executed to construct a multidimensional consistent change point detection structure. Dynamic programming segmentation and candidate segmentation path pruning are performed based on the risk state confidence to generate a set of risk state change points. Based on the smoke plume diffusion state, smoke plume phase change state, and risk state change point set, a continuous diffusion score is calculated, a risk level result is generated, and a fire-fighting linkage response is executed.

[0019] In this embodiment, the multi-source fire status data specifically includes smoke concentration data, ambient temperature data, infrared calorific value data, current status data, fault arc status data, video frame data, spatial location data, and acquisition timestamp data.

[0020] In this embodiment, the preprocessing of multi-source fire status data specifically includes time alignment processing, spatial mapping processing, outlier removal processing, image frame denoising processing, and standardization processing.

[0021] In this embodiment, the extraction of smoke candidate region data includes: The system reads the camera installation location, sensor installation location, and monitoring area boundary. A spatial coordinate system is established according to the monitoring area boundary. Based on the camera's field of view coverage and the sensor's acquisition coverage, the monitoring area is divided into spatial risk units, generating a set of spatial risk units, where: A spatial coordinate system is established according to the boundary of the monitoring area, specifically as follows: Read the boundary coordinates corresponding to the boundary of the monitoring area. Take the point with the smallest x-coordinate and the smallest y-coordinate among the boundary coordinates as the origin of the spatial coordinate system. Take the direction of the horizontal boundary where the origin is located as the x-axis direction. Take the direction of the vertical boundary where the origin is located as the y-axis direction. Transform all camera installation positions and sensor installation positions to the spatial coordinate system and generate corresponding spatial coordinates. Generate a set of spatial risk units, specifically as follows: Read the camera's field of view coverage and the sensor's acquisition coverage, take the overlapping area of ​​the camera's field of view coverage and the sensor's acquisition coverage as the basic area for division, perform grid division on the basic area with a grid size of 5m×5m, determine each grid as a spatial risk unit, and assign spatial risk unit numbers in the order from left to right and from top to bottom to generate a set of spatial risk units; Based on the spatial risk unit set, the spatial location data and collection timestamp data in the standardized multi-source fire status data are read. The smoke concentration data, ambient temperature data, infrared calorific value data, current status data and fault arc status data that fall into the same spatial risk unit and whose collection timestamp difference is less than or equal to 1 second are written into the same spatial risk unit to generate spatial risk unit status data. Read the spatial risk unit status data and corresponding video frames. Based on the camera installation location and the camera's field of view coverage, map the pixel regions in the video frames to the corresponding spatial risk units. Extract regions with a grayscale attenuation value greater than 0.18, an edge blur value greater than 0.25, and an area change rate greater than 0.10 for three consecutive frames as smoke candidate regions, generating smoke candidate region data, where: Preprocessing of multi-source fire status data specifically includes time alignment, spatial mapping, outlier removal, and standardization, as follows: The system reads smoke concentration data, ambient temperature data, infrared calorific value data, current status data, fault arc status data, video frame data, spatial location data, and acquisition timestamp data. Using the acquisition timestamp data as a benchmark, all data undergoes time alignment processing, and data with an acquisition time difference greater than 1 second is removed. Spatial location data is then read, and each type of data is mapped to its corresponding spatial risk unit. The average value and standard deviation of the smoke concentration data, ambient temperature data, infrared calorific value data, current status data, and fault arc status data are calculated separately. Data deviating from the average value by more than three times the standard deviation are identified as abnormal data and removed. The smoke concentration data, ambient temperature data, infrared calorific value data, current status data, and fault arc status data after removing abnormal data are then read separately and converted to the 0-1 range using minimum value normalization, generating standardized multi-source fire status data.

[0022] In this embodiment, the generated smoke cloud diffusion state and field-source coupling strength include: An improved Gaussian puff model is constructed, comprising a recursive puff module, a three-field recirculation coupling module, a coherent diffusion analysis module, and an entropy collapse correction module, wherein: The improved Gaussian smoke model is constructed as follows: This paper improves the Gaussian puff model by extracting the source term description module, source strength calculation module, diffusion calculation module, and concentration decay module from the traditional Gaussian puff model. It adds main puff nodes, child puff nodes, and parent-child mapping relationships to the source term description module, resulting in a recursive puff module. It also adds the temporal driving relationship between the smoke field, thermal field, and electric field to the source strength calculation module, resulting in a three-field recirculation coupling module. Furthermore, it adds the diffusion direction and diffusion velocity correlation relationships between puffs to the diffusion calculation module, resulting in a coherent diffusion analysis module. Finally, it adds the puff information entropy change state to the concentration decay module, resulting in an entropy collapse correction module, thus forming an improved Gaussian puff model. The recursive smoke plume module reads smoke candidate region data corresponding to the same collection timestamp within the same spatial risk unit, determines the smoke candidate region with the largest area as the main smoke plume node, and determines the remaining smoke candidate regions as child smoke plume nodes. It then establishes recursive edges and parent-child mapping relationships based on center distance and region overlap, generating a tree-like recursive topology structure, where: The recursive smoke module includes: Main smoke cloud generator: Generates the main smoke cloud node; Sub-puff generator: generates sub-puff nodes; Recursive associator: Establishes recursive associative edges; Parent-child mapping table: Stores the parent-child mapping relationship between the main smoke cloud node and the child smoke cloud nodes; Based on the center distance and region overlap, a recursive associative edge and parent-child mapping relationship are established, specifically as follows: Read the center coordinates of the main smoke cluster node and the center coordinates of the child smoke cluster nodes, calculate the center distance between the main smoke cluster node and each child smoke cluster node, read the area of ​​the main smoke cluster node region, the area of ​​the child smoke cluster node region, and the area of ​​the overlapping region between them, divide the overlapping region area by the sum of the area of ​​the main smoke cluster node region and the area of ​​the child smoke cluster node region and then subtract the overlapping region area to obtain the region overlap value, divide the center distance by the sum of the square root of the area of ​​the main smoke cluster node region and the square root of the area of ​​the child smoke cluster node region, and add the result of 1 minus the region overlap value to obtain the recursive association value. When the recursive association value is less than or equal to 1.20, establish a recursive association edge between the main smoke cluster node and the corresponding child smoke cluster node, and form a parent-child mapping relationship between the main smoke cluster node number and the corresponding child smoke cluster node number to generate a tree-like recursive topology structure; The three-field recirculation coupling module reads smoke concentration data, ambient temperature data, infrared calorific value data, current state data, and fault arc state data to construct smoke, thermal, and electric fields. It analyzes the temporal driving relationship between the changing states of these three fields. When the electric field drives the thermal field, the thermal field drives the smoke field, and the smoke field inversely enhances either the thermal or electric field, a smoke-heat-electric causal recirculation structure is generated, and the field-source coupling strength is calculated. The three-field reflow coupling module includes: Smoke Field Constructor: Generates a smoke field; Thermal field constructor: generates thermal fields; Electric field constructor: generates an electric field; Time-driven analyzer: Analyzes the time-driven relationships between the changing states of three fields; Reflux structure table: stores the causal reflux relationship between smoke, heat, and electricity; Coupling Strength Calculator: Calculates the coupling strength between field sources; The smoke field, thermal field, and electric field are constructed as follows: Read the smoke concentration data corresponding to the same acquisition timestamp within the same spatial risk unit, arrange the smoke concentration data in the order of acquisition timestamps to form a smoke field sequence, read the ambient temperature data and infrared calorific value data, arrange the ambient temperature data and infrared calorific value data in the order of acquisition timestamps to form a thermal field sequence, read the current status data and fault arc status data, arrange the current status data and fault arc status data in the order of acquisition timestamps to form an electric field sequence, and store the smoke field sequence, thermal field sequence and electric field sequence according to the spatial risk unit number to generate smoke field, thermal field and electric field; The temporal driving relationship between the three changing states is analyzed as follows: Read the smoke field, thermal field, and electric field corresponding to three consecutive acquisition timestamps. Subtract the value corresponding to the previous acquisition timestamp from the value corresponding to the subsequent acquisition timestamp to obtain the smoke field change value, thermal field change value, and electric field change value. When the electric field change value increases before the thermal field change value and the thermal field change value increases before the smoke field change value, it is determined that the electric field drives the thermal field and the thermal field drives the smoke field. When the smoke field change value increases and the corresponding thermal field change value or electric field change value continues to increase, it is determined that the smoke field inversely enhances the thermal field or electric field. The generation of a smoke-heat-electricity causal reflux structure is as follows: A reflux structure table is established, which includes a start node field, a target node field, a spatial risk unit number field, and a data acquisition timestamp field. When it is determined that the electric field drives the thermal field, the electric field node is written in the start node field and the thermal field node is written in the target node field. When it is determined that the thermal field drives the smoke field, the thermal field node is written in the start node field and the smoke field node is written in the target node field. When it is determined that the smoke field inversely enhances the thermal field, the smoke field node is written in the start node field and the thermal field node is written in the target node field. When it is determined that the smoke field inversely enhances the electric field, the smoke field node is written in the start node field and the electric field node is written in the target node field. All node connection relationships are written into the reflux structure table in the order of data acquisition timestamps to generate a smoke-heat-electricity causal reflux structure. The calculation of the field-source coupling strength is as follows: The smoke field, thermal field, electric field, and smoke-heat-electric causal return structure are read. The smoke concentration data is used as the smoke field intensity. The ambient temperature data is multiplied by 0.50 and then added to the infrared calorific value data multiplied by 0.50 to obtain the thermal field intensity. The current state data is multiplied by 0.60 and then added to the fault arc state data multiplied by 0.40 to obtain the electric field intensity. The node connection relationship in the smoke-heat-electric causal return structure is read. When there are connections where electric field nodes point to thermal field nodes, thermal field nodes point to smoke field nodes, and smoke field nodes point to either thermal field nodes or electric field nodes, the return flag value is set to 1. When any of the above connections are missing, the return flag value is set to 0. The smoke field intensity is multiplied by 0.35, the thermal field intensity by 0.30, the electric field intensity by 0.25, and the return flag value by 0.10, and then added together to obtain the field-source coupling strength. The coherent diffusion analysis module reads the changes in the center coordinates of each plume node between adjacent acquisition timestamps, generates the plume diffusion trajectory, calculates the diffusion direction, diffusion velocity, and concentration change rate, and constructs a cooperative diffusion state matrix based on the differences in direction, velocity, and concentration change rate between each plume node. When the degree of synchronous diffusion is greater than 0.65, a coherent diffusion structure of the plume is generated, where: The coherent diffusion analysis module includes: Trajectory generator: Generates the trajectory of smoke plume diffusion; Diffusion Characteristics Calculator: Calculates diffusion direction, diffusion rate, and concentration change rate; Co-diffusion state matrix: stores the co-diffusion state values ​​between smoke plumes; Coherent structure generator: generates coherent diffusion structures for smoke plumes; The smoke plume diffusion trajectory is generated as follows: Read the tree-like recursive topology corresponding to adjacent collection timestamps, match the center coordinates of the same smoke puff node under the previous and next collection timestamps according to the smoke puff node number, take the center coordinate under the previous collection timestamp as the trajectory start point, take the center coordinate under the next collection timestamp as the trajectory end point, and connect the trajectory start point and trajectory end point of the same smoke puff node according to the collection timestamp order to generate the smoke puff diffusion trajectory of the corresponding smoke puff node. The diffusion direction, diffusion rate, and concentration change rate are calculated as follows: Read the coordinates of the starting point and ending point of the smoke plume diffusion trajectory. Subtract the coordinates of the starting point from the coordinates of the ending point to obtain the smoke plume displacement vector. Determine the direction of the smoke plume displacement vector as the diffusion direction. Divide the length of the smoke plume displacement vector by the time difference between the two collection timestamps to obtain the diffusion velocity. Read the peak smoke plume concentration corresponding to the two collection timestamps. Subtract the peak smoke plume concentration of the previous collection timestamp from the peak smoke plume concentration of the later collection timestamp, and then divide by the time difference between the two collection timestamps to obtain the concentration change rate. The construction of the cooperative diffusion state matrix is ​​as follows: The total number of smoke cloud nodes in the tree-like recursive topology is counted. A two-dimensional table with the same number of rows and columns as the total number of smoke cloud nodes is established. This two-dimensional table is defined as the cooperative diffusion state matrix. The difference in diffusion direction, diffusion velocity, and concentration change rate between any two smoke cloud nodes are read. The normalized values ​​of the diffusion direction difference, diffusion velocity difference, and concentration change rate difference are multiplied by 0.40, 0.30, and 0.30, respectively. The results are then added together and subtracted from the sum by 1 to obtain the cooperative diffusion state value between the corresponding two smoke cloud nodes. The cooperative diffusion state value between the corresponding smoke cloud nodes is written into the corresponding position in the cooperative diffusion state matrix according to the order of the smoke cloud node numbers, thus completing the construction of the cooperative diffusion state matrix. The coherent diffusion structure of the generated smoke plume is as follows: Read all cooperative diffusion state values ​​except those on the main diagonal from the cooperative diffusion state matrix. Sum all cooperative diffusion state values ​​and divide by the total number of cooperative diffusion state values ​​to obtain the synchronization diffusion degree. When the synchronization diffusion degree is greater than 0.65, read the tree recursive topology, cooperative diffusion state matrix, and synchronization diffusion degree under the corresponding acquisition timestamp. Associate and store the smoke puff nodes, recursive associated edges, and cooperative diffusion state values ​​in the tree recursive topology and the cooperative diffusion state matrix to generate a smoke puff coherent diffusion structure. The entropy collapse correction module reads the smoke plume concentration distribution state in the tree-like recursive topology, calculates the smoke plume information entropy change state, and when the smoke plume information entropy decreases for three consecutive collection timestamps and a coherent diffusion structure of the smoke plume exists, it generates a smoke plume entropy collapse structure and performs collapse correction on the initial smoke plume concentration peak, initial smoke plume diffusion radius, and initial smoke plume duration to generate the smoke plume diffusion state, where: The entropy collapse correction module includes: Concentration distribution set: stores the concentration distribution status of smoke plumes; Information Entropy Calculator: Calculates the change in information entropy of a smoke cloud; Entropy collapse structure generator: generates entropy collapse structures for smoke plumes; Collapse Corrector: Corrects peak smoke concentration, smoke diffusion radius, and smoke duration; The calculation of the information entropy change state of the smoke plume is as follows: Read the peak smoke concentration of all smoke cloud nodes in the tree-like recursive topology under the same collection timestamp, divide the peak smoke concentration of each smoke cloud node by the sum of the peak smoke concentrations of all smoke cloud nodes to obtain the concentration ratio of that smoke cloud node, multiply each concentration ratio by its natural logarithm and take the negative value, and add the corresponding negative values ​​of all smoke cloud nodes to obtain the smoke cloud information entropy under the current collection timestamp, and subtract the smoke cloud information entropy under the previous collection timestamp from the smoke cloud information entropy under the current collection timestamp to obtain the smoke cloud information entropy change state; The entropy collapse structure of the generated smoke plume is as follows: Read the smoke plume information entropy change status corresponding to three consecutive collection timestamps. When the smoke plume information entropy corresponding to three consecutive collection timestamps decreases sequentially, and the synchronization diffusion degree corresponding to the current collection timestamp is greater than 0.65, it is determined that a smoke plume coherent diffusion structure exists. Read the smoke plume coherent diffusion structure, smoke plume information entropy, and smoke plume information entropy change status under the corresponding collection timestamp. Combine the smoke plume nodes, recursive associated edges, synchronization diffusion degree, smoke plume information entropy, and smoke plume information entropy change status in the smoke plume coherent diffusion structure to generate a smoke plume entropy collapse structure. The initial smoke plume concentration peak refers to the concentration peak determined by the smoke plume node based on the smoke concentration data and the grayscale attenuation value of the smoke candidate region before the entropy collapse correction is performed. The initial smoke plume diffusion radius refers to the equivalent circle radius of the smoke candidate region corresponding to the smoke plume node before the entropy collapse correction is performed. The equivalent circle radius is obtained by reading the area of ​​the smoke candidate region, dividing the area of ​​the smoke candidate region by pi, and then taking the square root. The initial smoke plume duration refers to the duration from the first appearance timestamp to the current timestamp before the entropy collapse correction is performed. Collapse corrections are applied to the initial plume concentration peak, initial plume diffusion radius, and initial plume duration, specifically as follows: The source-field coupling strength, synchronous diffusion degree, and smoke plume entropy change state are read. When the smoke plume entropy change state is less than 0, the absolute value of the smoke plume entropy change state is determined as the entropy collapse strength. The source-field coupling strength multiplied by 0.20 is added to the entropy collapse strength multiplied by 0.15 to obtain the concentration correction coefficient. The synchronous diffusion degree multiplied by 0.15 is added to the entropy collapse strength multiplied by 0.10 to obtain the radius correction coefficient. The result of multiplying the source-field coupling strength by 0.10 and the synchronous diffusion degree multiplied by 0.10 is... Add them together to get the duration correction coefficient. Multiply the initial peak concentration of the plume by 1 and add the concentration correction coefficient to get the corrected peak concentration of the plume. Multiply the initial plume diffusion radius by 1 and add the radius correction coefficient to get the corrected diffusion radius of the plume. Multiply the initial plume duration by 1 and add the duration correction coefficient to get the corrected duration of the plume. Combine the corrected peak concentration of the plume, the corrected diffusion radius of the plume, the corrected duration of the plume, the coherent diffusion structure of the plume, and the entropy collapse structure of the plume to generate the plume diffusion state. The improved Gaussian puff model was trained using the puff diffusion state prediction error, field-source coupling strength error, and puff coherent diffusion state error as joint optimization objectives. The parameters of the recursive puff module, the three-field backflow coupling module, the coherent diffusion analysis module, and the entropy collapse correction module were continuously optimized. Training was completed when the change in joint loss after five consecutive iterations was less than 0.001. The improved Gaussian smoke model was trained as follows: The system reads historical smoke candidate area data, smoke concentration data, ambient temperature data, infrared calorific value data, current state data, and fault arc state data. It inputs the historical smoke candidate area data into the recursive smoke plume module and the smoke concentration data, ambient temperature data, infrared calorific value data, current state data, and fault arc state data into the three-field return coupling module. After processing by the coherent diffusion analysis module and the entropy collapse correction module, it outputs the predicted smoke plume diffusion state and the predicted field source coupling strength. It reads the actual smoke plume diffusion state and the actual field source coupling strength at the corresponding acquisition timestamp. It squares the difference between the predicted and actual smoke plume diffusion states and then averages the results to obtain the smoke plume diffusion state prediction error. It squares the difference between the predicted and actual field source coupling strengths and then averages the results to obtain the field source coupling strength error. It reads the smoke plume coherent diffusion structure in the predicted and actual smoke plume diffusion states, extracts the corresponding synchronous diffusion degree, and calculates the average of the squared differences to obtain the smoke plume coherent diffusion state error. The joint loss value is obtained by multiplying the smoke puff diffusion state prediction error by 0.50, the field-source coupling strength error by 0.30, and the smoke puff coherent diffusion state error by 0.20. Based on the joint loss value, the gradient values ​​corresponding to each parameter in the recursive smoke puff module, the three-field backflow coupling module, the coherent diffusion analysis module, and the entropy collapse correction module are calculated using gradient descent. The parameters are updated in the opposite direction of the gradient value with a learning rate of 0.001. The joint loss value is recorded after each parameter update. When the absolute value of the difference between the current joint loss value and the previous joint loss value is less than 0.001 and occurs consecutively for 5 times, the joint loss value is considered to have reached the convergence state, and the training of the improved Gaussian smoke puff model is completed.

[0023] In this embodiment, the phase transition state of the generated smoke cloud includes: Read the smoke plume diffusion state and field-source coupling strength corresponding to consecutive acquisition timestamps within the same spatial risk unit, extract the peak smoke plume concentration, smoke plume diffusion radius, smoke plume duration, synchronous diffusion degree, and smoke plume information entropy change status, and calculate the smoke plume concentration growth rate, diffusion radius growth rate, and duration growth rate, where: The peak concentration of the smoke plume, the diffusion radius of the smoke plume, the duration of the smoke plume, the degree of synchronous diffusion, and the change state of the smoke plume information entropy are extracted, specifically as follows: Read the smoke cloud diffusion state corresponding to the continuous acquisition timestamp, extract the corrected smoke cloud concentration peak value from the smoke cloud diffusion state as the smoke cloud concentration peak value, extract the corrected smoke cloud diffusion radius as the smoke cloud diffusion radius, extract the corrected smoke cloud duration as the smoke cloud duration, read the synchronous diffusion degree in the smoke cloud coherent diffusion structure as the synchronous diffusion degree, and read the smoke cloud information entropy change state in the smoke cloud entropy collapse structure as the smoke cloud information entropy change state. The growth rates of smoke plume concentration, diffusion radius, and duration are calculated as follows: Read the peak smoke concentration corresponding to the current collection timestamp and the peak smoke concentration corresponding to the previous collection timestamp, divide the difference between the two by the peak smoke concentration corresponding to the previous collection timestamp to obtain the smoke concentration growth rate; read the smoke diffusion radius corresponding to the current collection timestamp and the smoke diffusion radius corresponding to the previous collection timestamp, divide the difference between the two by the smoke diffusion radius corresponding to the previous collection timestamp to obtain the diffusion radius growth rate; read the smoke duration corresponding to the current collection timestamp and the smoke duration corresponding to the previous collection timestamp, divide the difference between the two by the smoke duration corresponding to the previous collection timestamp to obtain the duration growth rate. A smoke plume phase transition state field is constructed using spatial risk units and acquisition timestamps as indices. The smoke plume concentration growth rate, diffusion radius growth rate, duration growth rate, field-source coupling strength, synchronous diffusion degree, and smoke plume information entropy change state are multiplied by 0.25, 0.20, 0.15, 0.20, 0.10, and 0.10 respectively, and then summed to obtain the smoke plume phase transition order value, which is then written into the smoke plume phase transition state field. Where: The phase transition state field of the smoke plume is constructed using spatial risk units and acquisition timestamps as indexes, specifically as follows: A smoke plume phase transition state table is established, which includes a spatial risk unit number field, a collection timestamp field, and a smoke plume phase transition order value field. The spatial risk unit number is used as the row index and the collection timestamp is used as the column index. The smoke plume phase transition order values ​​corresponding to the same spatial risk unit under different collection timestamps are written into the corresponding positions to generate a smoke plume phase transition state field. Read the smoke phase transition order values ​​corresponding to the first 12 collection timestamps in the smoke phase transition state field, calculate the average and standard deviation, and add the average to 1.50 times the standard deviation to obtain the smoke phase transition threshold; The current phase transition order value of the plume in the phase transition state field is read and compared with the phase transition threshold of the plume. When the phase transition order value of the plume is greater than the phase transition threshold of the plume and there is a coherent diffusion structure and an entropy collapse structure in the plume, a phase transition is determined to have occurred, and a phase transition state of the plume is generated, wherein: The determination of a phase transition is as follows: Read the smoke phase transition order value, smoke phase transition threshold, smoke coherent diffusion structure and smoke entropy collapse structure corresponding to the current collection timestamp. When the smoke phase transition order value is greater than the smoke phase transition threshold and the synchronous diffusion degree is greater than 0.65, and the smoke information entropy corresponding to 3 consecutive collection timestamps continues to decrease, it is determined that a phase transition has occurred. The phase transition state of the generated smoke plume is as follows: A smoke plume phase transition state record table is established, which includes a spatial risk unit number field, a collection timestamp field, a smoke plume phase transition order value field, a smoke plume phase transition threshold field, and a phase transition judgment field. The spatial risk unit number, smoke plume phase transition order value, smoke plume phase transition threshold, and phase transition judgment result corresponding to the current collection timestamp are written into the smoke plume phase transition state record table to generate the smoke plume phase transition state.

[0024] In this embodiment, generating the risk state change point set includes: Based on the collection timestamp, the smoke plume diffusion state and phase transition state within the same spatial risk unit are read. Peak smoke plume concentration, smoke plume diffusion radius, smoke plume duration, field-source coupling strength, synchronous diffusion degree, smoke plume information entropy change state, smoke-heat-electric causal backflow structure, and smoke plume phase transition order value are extracted to construct a multidimensional risk state sequence, where: Constructing a multidimensional risk state sequence, specifically: The peak concentration of smoke plume, smoke plume diffusion radius, smoke plume duration, field-source coupling strength, synchronous diffusion degree, smoke plume information entropy change state, smoke-heat-electric causal reflux structure and smoke plume phase transition order value of the same spatial risk unit are read in the order of collection timestamps. The above data are arranged in the order of collection timestamps, and the data corresponding to the same collection timestamp are combined to form a risk state vector. All risk state vectors are connected in the order of collection timestamps to generate a multidimensional risk state sequence. A risk propagation topology graph is established based on the positional adjacency relationships between spatial risk units, the smoke plume diffusion direction, and the field-source coupling strength. A multidimensional risk state sequence is written into the nodes of the risk propagation topology graph. The consistency of smoke plume diffusion direction, the difference in field-source coupling strength, and the difference in smoke plume phase transition order value between adjacent spatial risk units are written into the edges of the risk propagation topology graph, generating a topological risk state graph, where: A risk propagation topology map is established based on the positional adjacency relationships between spatial risk units, the direction of smoke plume diffusion, and the coupling strength of the field sources, specifically as follows: Read the location coordinates of all spatial risk units, calculate the center distance between spatial risk units, and establish a location adjacency relationship when the center distance is less than or equal to 1.50 times the side length of the spatial risk unit. Read the smoke cloud diffusion direction corresponding to the adjacent spatial risk units. When the smoke cloud diffusion direction of one spatial risk unit points to another spatial risk unit, establish a directed connection edge between the two. Treat the spatial risk units as nodes, write the location adjacency relationship and the directed connection edge into the topology structure, and generate a risk propagation topology graph. The consistency of the smoke plume diffusion direction refers to the cosine value of the angle between the smoke plume diffusion directions of adjacent spatial risk units; the difference in field-source coupling strength refers to the absolute value of the difference in the field-source coupling strength of adjacent spatial risk units; and the difference in smoke plume phase transition order value refers to the absolute value of the difference in the smoke plume phase transition order value of adjacent spatial risk units. The PELT algorithm is executed based on the topological risk state diagram. According to the phase transition state of the smoke plume, the topological risk state diagram is divided into random perturbation state segments, phase transition state segments, and continuous diffusion state segments. Corresponding segmental cost weights are assigned to each segment, generating a phase transition-driven segmental cost function, where: Based on the phase transition state of the smoke plume, the topological risk state diagram is divided into three segments: random disturbance state segment, phase transition state segment, and persistent diffusion state segment, as follows: Read the smoke phase transition state corresponding to each node in the topological risk state diagram. When the phase transition judgment result is negative and the smoke phase transition order value is less than the smoke phase transition threshold, it is divided into a random disturbance state segment. When the phase transition judgment result is positive, it is divided into a phase transition state segment. When the phase transition judgment result is positive and the smoke phase transition order value corresponding to 3 consecutive collection timestamps is greater than the smoke phase transition threshold, it is divided into a continuous diffusion state segment. Set the corresponding segment cost weights as follows: Set the segmented cost weight corresponding to the random disturbance state segment to 0.20, the segmented cost weight corresponding to the phase transition state segment to 0.35, and the segmented cost weight corresponding to the continuous diffusion state segment to 0.45. The phase transition-driven piecewise cost function is generated as follows: Read the node state values ​​and corresponding segment cost weights in the topology risk state diagram. Multiply the node state values ​​by the corresponding segment cost weights and sum them to obtain the state-weighted cost value. Count the number of collection timestamps in the current segment. Multiply the natural logarithm of the number of collection timestamps by 0.50 to obtain the change point penalty term. Add the state-weighted cost value to the change point penalty term to generate the phase change driven segment cost function. Hierarchical recursive PELT processing is performed based on a tree-like recursive topology. Recursive propagation paths between main smoke cloud nodes and child smoke cloud nodes are generated according to parent-child mapping relationships. Change point detection is then performed on the main smoke cloud nodes, child smoke cloud nodes, and recursive propagation paths based on the phase transition-driven piecewise cost function. Change points in the main smoke cloud, child smoke cloud, and recursive propagation paths are written into the same hierarchical change point group, generating a hierarchical topology joint change point cost, where: Hierarchical recursive PELT processing is performed based on a tree-like recursive topology, specifically as follows: Read the main smoke cluster node, child smoke cluster node, recursive associated edges, and parent-child mapping relationship in the tree-like recursive topology; take the multidimensional risk state sequence corresponding to the main smoke cluster node as the first-level detection sequence, and take the multidimensional risk state sequence corresponding to the child smoke cluster node as the next-level detection sequence. Connect the main smoke cluster node and child smoke cluster node in sequence according to the parent-child mapping relationship to form a hierarchical detection chain. Read the multidimensional risk state sequence in the hierarchical detection chain according to the collection timestamp order, calculate the Euclidean distance between the multidimensional risk state data corresponding to the current collection timestamp and the multidimensional risk state data corresponding to the previous collection timestamp. When the Euclidean distance is greater than 1.50 times the average of the Euclidean distances corresponding to the previous 12 collection timestamps, determine the current collection timestamp as a candidate change point and generate hierarchical recursive change point results. The parent-child mapping relationship refers to the set of correspondences between the node numbers of the main smoke cluster and the node numbers of the child smoke cluster in a tree-like recursive topology. Based on the parent-child mapping relationship, a recursive propagation path is generated between the main smoke cloud node and the child smoke cloud node, specifically as follows: Read the main smoke cluster node number and child smoke cluster node number in the parent-child mapping relationship. Using the main smoke cluster node as the starting node and the child smoke cluster node as the ending node, connect the corresponding nodes in the order of the recursive associated edges to generate the recursive propagation path between the main smoke cluster node and the child smoke cluster node. Based on the phase transition-driven piecewise cost function, change point detection is performed on the main smoke cloud node, sub-smoke cloud nodes, and recursive propagation path, specifically as follows: Read the multidimensional risk state sequence corresponding to the main smoke plume node, and determine the data between adjacent candidate change points as a candidate segment; calculate the average values ​​of smoke plume concentration peak, smoke plume diffusion radius, smoke plume duration, field-source coupling strength, synchronous diffusion degree, smoke plume information entropy change state, and smoke plume phase transition order value for each collection time stamp within the candidate segment; then calculate the squared differences between the peak smoke plume concentration and the average peak smoke plume concentration, the squared differences between the smoke plume diffusion radius and the average smoke plume diffusion radius, the squared differences between the smoke plume duration and the average smoke plume duration, the squared differences between the field-source coupling strength and the average field-source coupling strength, the squared differences between the synchronous diffusion degree and the average synchronous diffusion degree, the squared differences between the smoke plume information entropy change state and the average smoke plume information entropy change state, and the squared differences between the smoke plume phase transition order value and the average smoke plume phase transition order value; sum all the squared differences to obtain the state-weighted cost value, and then add the state-weighted cost value to the change point penalty term to obtain the phase transition-driven segment cost value; When the sum of the phase change-driven segment values ​​of all candidate segments corresponding to the newly added candidate variable point is less than the sum of the phase change-driven segment values ​​of all candidate segments corresponding to the segments without the newly added candidate variable point, the candidate variable point is determined as the main smoke cloud variable point. The same processing is performed on the sub-smoke cloud nodes and recursive propagation paths to generate sub-smoke cloud variable points and recursive propagation path variable points. The main smoke plume change point, the sub-smoke plume change point, and the recursive propagation path change point are written into the same level change point group, specifically: Establish a hierarchical variable point group record table, which includes a main smoke cloud variable point field, a child smoke cloud variable point field, and a recursive propagation path variable point field. Write the main smoke cloud variable point, child smoke cloud variable point, and recursive propagation path variable point corresponding to the same parent and child levels into the corresponding fields to generate a hierarchical variable point group. The value of generating the hierarchical topology joint variable point is as follows: Read the number of main smoke cloud variables, sub-smoke cloud variables, and recursive propagation path variables in the hierarchical variable group. Multiply the number of main smoke cloud variables by 0.40, the number of sub-smoke cloud variables by 0.30, and the number of recursive propagation path variables by 0.30, and add the results to generate the hierarchical topology joint variable value. S55. Based on the hierarchical topology joint change point cost and phase transition driven piecewise cost function, perform dynamic programming piecewise segmentation and candidate segmentation path pruning. Delete candidate segmentation paths that do not conform to the risk propagation topology direction, parent-child recursive propagation order, and phase transition state. Retain the change point location, acquisition timestamp, spatial risk unit number, hierarchical topology joint change point cost, and risk state confidence score of the candidate segmentation paths to generate a risk state change point set, where: Based on the hierarchical topology joint variable point cost function and the phase transition driven piecewise cost function, dynamic programming piecewise segmentation and candidate split path pruning are performed, specifically as follows: Read the phase change-driven segmentation value and the hierarchical topology joint change point value corresponding to all candidate segmentation paths, add the two together to obtain the total path value, compare the total path value of adjacent candidate segmentation paths according to the collection timestamp order, delete the current candidate segmentation path when the total path value of the current candidate segmentation path is greater than the total path value of the previous candidate segmentation path, and retain the current candidate segmentation path when the total path value of the current candidate segmentation path is less than or equal to the total path value of the previous candidate segmentation path, and generate retained candidate segmentation paths. The risk propagation topology direction refers to the propagation direction of directed connecting edges in the risk propagation topology graph, and the parent-child recursive propagation order refers to the propagation order in the parent-child mapping relationship where the main smoke cluster node comes first and the child smoke cluster node comes last. Candidate segmentation paths that do not conform to the risk propagation topology, parent-child recursive propagation order, and phase transition states are deleted, specifically: Read the start and end nodes in the candidate segmentation path. Delete the candidate segmentation path when the propagation direction of the candidate segmentation path is opposite to the risk propagation topology direction. Delete the candidate segmentation path when the sub-puff node in the candidate segmentation path changes before the main puff node. Delete the candidate segmentation path when the phase transition judgment result in the phase transition state of the puff corresponding to the candidate segmentation path is negative and the puff phase transition order value is less than the puff phase transition threshold. Preserving the variable point position corresponding to the candidate segmentation path means preserving the collection timestamp sequence number corresponding to the variable point in the candidate segmentation path. Risk status confidence refers to the ratio of the number of retained candidate segmentation paths to the total number of candidate segmentation paths; The generation of the risk state change point set is as follows: A risk status change point record table is established, which includes fields for change point location, collection timestamp, spatial risk unit number, hierarchical topology joint change point value, and risk status confidence. The change point location, collection timestamp, spatial risk unit number, hierarchical topology joint change point value, and risk status confidence corresponding to the retained candidate segmentation path are written into the risk status change point record table to generate a risk status change point set.

[0025] In this embodiment, the execution of fire alarm linkage response processing includes: Read the set of smoke plume diffusion states, smoke plume phase transition states, and risk state change points corresponding to the same judgment window within the same spatial risk unit. Extract the field-source coupling strength, synchronous diffusion degree, smoke plume phase transition order value, phase transition judgment result, risk state confidence, and change point continuity ratio to generate a set of continuous diffusion judgment indicators, including: The phase transition determination result refers to the determination result obtained by comparing the phase transition order value of the plume with the phase transition threshold of the plume. When the phase transition order value of the plume is greater than the phase transition threshold of the plume and there is a coherent diffusion structure and an entropy collapse structure of the plume, the value is 1; otherwise, the value is 0. The continuous change point ratio refers to the ratio of the number of timestamps collected for consecutive changes in the risk state change point set to the total number of timestamps collected for all changes. Generate a set of indicators for determining persistent diffusion, specifically as follows: Read the smoke plume diffusion state corresponding to the same judgment window within the same spatial risk unit, extract the field-source coupling strength and synchronous diffusion degree from the smoke plume diffusion state, read the smoke plume phase transition state, extract the smoke plume phase transition order value and phase transition judgment result, read the risk state change point set, extract the risk state confidence and change point continuity ratio, and establish a continuous diffusion judgment index table. The continuous diffusion judgment index table includes the field-source coupling strength field, synchronous diffusion degree field, smoke plume phase transition order value field, phase transition judgment result field, risk state confidence field, and change point continuity ratio field. Write the indexes into the corresponding fields to generate a continuous diffusion judgment index set. The continuous diffusion score is obtained by multiplying the source-field coupling strength, synchronous diffusion degree, smoke phase transition order value, risk state confidence, change point continuity ratio, and phase transition judgment result by 0.20, 0.20, 0.20, 0.15, and 0.05, respectively. When the continuous diffusion score is less than 0.45, a Level 1 risk is generated and a risk warning is issued. When the continuous diffusion score is greater than or equal to 0.45 and less than 0.65, a Level 2 risk is generated and an early warning is issued. When the continuous diffusion score is greater than or equal to 0.65 and less than 0.80, a Level 3 risk is generated and a coordinated action is taken to activate area smoke extraction, audible and visual alarms, and electricity monitoring. When the continuous diffusion score is greater than or equal to 0.80, a Level 4 risk is generated and a fire alarm, area power outage, smoke extraction activation, and emergency notification are issued.

[0026] Example 1: To verify the feasibility of this invention in practice, it was applied to a real-time fire hazard monitoring scenario in an enterprise data center server room. The server room contains server rack aisles, power distribution branches, air conditioning vents, and cable trays. The monitoring area is divided into 5m × 5m spatial risk units. Each spatial risk unit is equipped with smoke detectors, ambient temperature detectors, infrared thermal imaging detectors, power consumption monitoring modules, and video cameras to collect smoke concentration data, ambient temperature data, infrared calorific value data, current status data, fault arc status data, video frame data, spatial location data, and data collection timestamp data. Before the system went live, an improved Gaussian smoke cloud model was trained using 8000 sets of normal heat dissipation samples, 2200 sets of air conditioning airflow disturbance samples, 1200 sets of short-term dust disturbance samples, 900 sets of short-term current spike samples, 650 sets of local overheating samples, and 500 sets of continuously spreading smoke samples. Each sample set contains 120 seconds of continuous multi-source fire status data. At the start of training, the prediction error of the smoke diffusion state was 0.126, the field-source coupling strength error was 0.041, the smoke coherent diffusion state error was 0.017, and the joint loss was 0.079. By the 51st iteration, the joint loss had dropped to 0.011, and the change in joint loss for five consecutive iterations was less than 0.001, thus completing the model training.

[0027] Within a continuous monitoring segment, spatial risk unit A3 corresponds to a set of power distribution areas behind a cabinet. Initially, the smoke concentration in spatial risk unit A3 ranged from 0.018 to 0.024, the ambient temperature from 23.6°C to 24.4°C, the infrared calorimetry from 0.18 to 0.23, the current state value from 0.31 to 0.36, the fault arc state value was 0, the video frame grayscale attenuation was below 0.06, and the edge blurriness was below 0.08. No smoke candidate areas were extracted. This invention calculated a continuous diffusion score of 0.21, generating a Level 1 risk and recording a risk warning. Traditional smoke detection threshold methods, traditional infrared temperature threshold methods, and traditional video smoke recognition methods all failed to trigger alarms, indicating that this invention does not introduce false alarms under safe conditions.

[0028] During the continued operation of the same monitoring segment, the current state value of the A3 space risk unit increased from 0.35 to 0.52, the fault arc state value increased from 0 to 0.16, while the smoke concentration was only 0.027, the ambient temperature was 24.8℃, the infrared calorimetry was 0.29, and there was no obvious smoke texture in the video frame. Traditional smoke detection, infrared temperature threshold, and video smoke recognition methods all failed to trigger an alarm. This invention writes the current state data and fault arc state data into the electric field sequence and performs time-series analysis with the smoke field sequence and the thermal field sequence. Since a complete backflow relationship is not formed where the electric field drives the thermal field, the thermal field drives the smoke field, and the smoke field inversely enhances the thermal field or electric field, the field-source coupling strength is 0.32, the continuous diffusion score is 0.36, and it still maintains a level one risk. This process shows that this invention can record the initial signs of electrical anomalies, but will not directly determine a single electrical disturbance as a high-level fire hazard.

[0029] Subsequently, the anomaly within the A3 space risk unit developed from the electrical side to the thermal and smoke sides. The ambient temperature rose from 25.1℃ to 32.8℃, the infrared calorific value increased from 0.30 to 0.61, the smoke concentration increased from 0.028 to 0.071, the current state value remained between 0.58 and 0.64, and the fault arc state value increased from 0.18 to 0.29. The system read the three field change states corresponding to three consecutive acquisition timestamps, finding that the electric field change value increased first, followed by the thermal field change value, and then the smoke field change value increased after the thermal field change. Furthermore, the thermal field intensity continued to increase after the smoke field change. Therefore, connections were written into the return flow structure table between electric field nodes pointing to thermal field nodes, thermal field nodes pointing to smoke field nodes, and smoke field nodes pointing to thermal field nodes, forming a smoke-heat-electric causal return flow structure. At this time, the smoke field intensity was 0.071, the thermal field intensity was 0.565, the electric field intensity was 0.482, the return flow identifier value was 1, and the field-source coupling strength was 0.415. The grayscale attenuation value in the video frame reached 0.21, the edge blurring reached 0.28, and the area change rate for three consecutive frames was 0.13. The system extracted a smoke candidate region with an area of ​​0.42 square meters. The traditional video smoke recognition method gave a smoke confidence score of 0.46, which is lower than the trigger condition of 0.70, and therefore no alarm was triggered. This invention determines the smoke candidate region as the main smoke cloud node and generates two child smoke cloud nodes. The recursive association values ​​between the main smoke cloud node and the two child smoke cloud nodes are 0.83 and 0.91, respectively, both less than 1.20, forming a tree-like recursive topology.

[0030] As the hidden danger continued to develop, the center coordinates of the main plume node moved from (12.4, 8.5) to (13.1, 8.9), the center coordinates of the first sub-puff node moved from (12.9, 8.7) to (13.7, 9.1), and the center coordinates of the second sub-puff node moved from (12.2, 8.9) to (12.8, 9.5). The system calculated the diffusion direction, diffusion rate, and concentration change rate of each plume node, and constructed a co-diffusion state matrix. The co-diffusion state values ​​of the non-main diagonal lines were 0.71, 0.69, 0.73, 0.69, 0.73, and 0.70, and the synchronous diffusion degree was 0.708, which is greater than 0.65, generating a coherent diffusion structure for the plumes. The information entropy of the plumes within the tree-like recursive topology decreased from 1.06 to 0.93, and then further to 0.81, decreasing over three consecutive collection timestamps, generating a plume entropy collapse structure. After collapse correction, the peak smoke concentration was corrected from 0.094 to 0.116, the smoke diffusion radius was corrected from 0.76m to 0.91m, and the smoke duration was corrected from 112s to 138s, indicating that the smoke has changed from a short-term disturbance to a smoke structure with synchronous diffusion and concentration accumulation trends.

[0031] The system continues to calculate the phase transition state of the plume. The current plume concentration growth rate is 0.34, the diffusion radius growth rate is 0.27, the duration growth rate is 0.18, the field-source coupling strength is 0.49, the synchronous diffusion degree is 0.708, and the plume information entropy change state is -0.12. The plume phase transition order value is calculated to be 0.339 using weights of 0.25, 0.20, 0.15, 0.20, 0.10, and 0.10. The average value of the plume phase transition order value corresponding to the first 12 acquisition timestamps is 0.164, the standard deviation is 0.071, and the plume phase transition threshold is 0.2705. Since 0.339 is greater than 0.2705, and there is a coherent diffusion structure and a plume entropy collapse structure in the plume, the system determines that a phase transition has occurred. Traditional methods at this stage can only produce isolated results such as temperature rise but not exceeding limits, smoke concentration rise but not exceeding limits, or slight haze but with insufficient confidence. They cannot determine whether local smoke has changed from random disturbance to a continuous diffusion type of fire hazard.

[0032] During the risk state change point detection process, A3 and its adjacent spatial risk unit A4 are adjacent, the smoke plume diffusion direction is from A3 to A4, the field-source coupling strength difference is 0.08, the smoke plume phase transition order value difference is 0.06, and the smoke plume diffusion direction consistency is 0.91. The system writes A3 and A4 into the risk propagation topology graph nodes and writes the point from A3 to A4 into the directed edge. The hierarchical recursive PELT reads the multidimensional risk state sequence of the main smoke plume node and the child smoke plume node. The average Euclidean distance of the first 12 collection timestamps is 0.118, and the current Euclidean distance is 0.204, which is greater than 0.177, so the current position is determined as a candidate change point. The sum of the phase transition driven segment cost after adding this candidate change point is 0.82, while it is 1.37 before adding it, so this candidate change point is retained. After risk propagation topology direction, parent-child recursive propagation order, and phase transition state pruning, 3 out of 4 candidate segmentation paths are retained, and the risk state confidence is 0.75.

[0033] When the persistent diffusion score is calculated, the system reads the field-source coupling strength (0.56), synchronous diffusion degree (0.708), smoke phase transition order value (0.376), risk state confidence level (0.75), change point continuity ratio (0.67), and phase transition judgment result (1). Based on weights of 0.20, 0.20, 0.20, 0.20, 0.15, and 0.05, a persistent diffusion score of 0.629 is calculated, generating a level-two risk and executing an early warning. At this point, the traditional smoke detection threshold method still does not trigger an alarm, the traditional infrared temperature threshold method only displays a temperature rise warning, and the traditional video smoke recognition method still does not trigger an alarm. After continued operation, the smoke concentration in the A3 space risk unit rose to 0.192, the ambient temperature rose to 43.6℃, ​​the infrared calorific value rose to 0.82, the current state value rose to 0.74, the fault arc state value rose to 0.41, and the area of ​​the smoke candidate region in the video frame reached 1.35 square meters. The system calculated a continuous diffusion score of 0.738, generating a level 3 risk, and executed the linkage processing of area smoke exhaust, audible and visual alarms, and power consumption monitoring. When the infrared calorific value reached 0.91 and the smoke concentration reached 0.263, the continuous diffusion score rose to 0.842, generating a level 4 risk, and executed the processing of fire alarm, area power outage, smoke exhaust activation, and emergency notification. The first level 2 risk warning of this invention is 116 seconds earlier than traditional smoke alarms, 101 seconds earlier than traditional video smoke recognition alarms, and 78 seconds earlier than traditional infrared temperature threshold alarms; the level 3 risk linkage is 56 seconds earlier than traditional smoke alarms.

[0034] When using the same test set for comparative verification, the test set included 600 sets of normal operation samples, 180 sets of air conditioning airflow disturbance samples, 120 sets of dust disturbance samples, 110 sets of short-term current spike samples, 90 sets of local overheating samples without smoke, and 100 sets of samples with continuous diffusion hazard. The traditional smoke detection threshold method identified 72 sets of samples with continuous diffusion hazard, with 28 missed detections and 31 false positives for dust disturbance samples, achieving an overall accuracy of 86.0%. The traditional infrared temperature threshold method identified 76 sets, with 24 missed detections and 27 false positives, achieving an overall accuracy of 87.3%. The traditional video smoke recognition method identified 81 sets, with 19 missed detections and 24 false positives, achieving an overall accuracy of 88.6%. This invention identified 96 sets of samples with continuous diffusion hazard, with 4 missed detections and a total of 7 false positives for air conditioning airflow disturbance, dust disturbance, and short-term current spike samples, achieving an overall accuracy of 97.1%. The simulation results above show that the present invention can identify the process of local smoke transforming from random disturbance to continuous diffusion fire hazard when a single sensor has not yet met the alarm conditions and the video image has not yet formed a clear smoke pattern, and can complete the real-time monitoring of fire hazards with earlier risk warnings, fewer false alarms and more accurate linkage levels.

[0035] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for real-time monitoring of fire hazards based on multi-sensor fusion, characterized in that, include: Collect multi-source fire status data within the monitoring area, and perform preprocessing on the multi-source fire status data to generate standardized multi-source fire status data; Based on the installation locations of cameras and sensors and the boundaries of the monitoring area, spatial risk units are divided. Standardized multi-source fire status data are mapped to the corresponding spatial risk units, and smoke candidate area data are extracted. An improved Gaussian smoke plume model is constructed, which converts smoke candidate region data into multi-scale recursive smoke plume objects, establishes a recursive association between the main smoke plume and the sub-smoke plumes, and constructs a smoke-heat-electric three-field coupled field source to perform coupling correction on the smoke plume diffusion state and generate the smoke plume diffusion state and field source coupling strength. Based on the smoke diffusion state and the field-source coupling strength, a smoke phase transition state field is constructed, the smoke phase transition threshold is calculated, and a phase transition determination is performed to generate the smoke phase transition state. A multidimensional risk state sequence is constructed based on the smoke puff diffusion state and smoke puff phase transition state. The PELT algorithm is then executed to construct a multidimensional consistent change point detection structure. Dynamic programming segmentation and candidate segmentation path pruning are performed based on the risk state confidence to generate a set of risk state change points. Based on the smoke plume diffusion state, smoke plume phase change state, and risk state change point set, a continuous diffusion score is calculated, a risk level result is generated, and a fire-fighting linkage response is executed.

2. The method for real-time monitoring of fire hazards based on multi-sensor fusion according to claim 1, characterized in that, The multi-source fire status data specifically includes smoke concentration data, ambient temperature data, infrared calorific value data, current status data, fault arc status data, video frame data, spatial location data, and acquisition timestamp data.

3. The method for real-time monitoring of fire hazards based on multi-sensor fusion according to claim 1, characterized in that, The preprocessing of multi-source fire status data specifically includes time alignment, spatial mapping, outlier removal, image frame denoising, and standardization.

4. The method for real-time monitoring of fire hazards based on multi-sensor fusion according to claim 1, characterized in that, The extraction of smoke candidate region data includes: Read the camera installation location, sensor installation location, and monitoring area boundary; establish a spatial coordinate system according to the monitoring area boundary; divide the monitoring area into spatial risk units based on the camera's field of view coverage and the sensor's acquisition coverage; and generate a set of spatial risk units. Based on the spatial risk unit set, the spatial location data and collection timestamp data in the standardized multi-source fire status data are read. The smoke concentration data, ambient temperature data, infrared calorific value data, current status data and fault arc status data that fall into the same spatial risk unit and whose collection timestamp difference is less than or equal to 1 second are written into the same spatial risk unit to generate spatial risk unit status data. Read the spatial risk unit status data and corresponding video frames. Based on the camera installation location and the camera's field of view coverage, map the pixel areas in the video frames to the corresponding spatial risk units. Extract areas with a grayscale attenuation value greater than 0.18, an edge blur value greater than 0.25, and an area change rate greater than 0.10 for three consecutive frames as smoke candidate areas, and generate smoke candidate area data.

5. The method for real-time monitoring of fire hazards based on multi-sensor fusion according to claim 1, characterized in that, The generated smoke cloud diffusion state and field-source coupling strength include: An improved Gaussian puff model is constructed, which includes a recursive puff module, a three-field recirculation coupling module, a coherent diffusion analysis module, and an entropy collapse correction module. The recursive smoke cloud module reads the smoke candidate region data corresponding to the same collection timestamp within the same spatial risk unit, determines the smoke candidate region with the largest area as the main smoke cloud node, and determines the remaining smoke candidate regions as child smoke cloud nodes. It also establishes recursive associated edges and parent-child mapping relationships based on the center distance and region overlap status, generating a tree-like recursive topology structure. The three-field recirculation coupling module reads smoke concentration data, ambient temperature data, infrared calorific value data, current status data, and fault arc status data to construct smoke field, thermal field, and electric field. It analyzes the temporal driving relationship between the changing states of the three fields. When the electric field drives the thermal field, the thermal field drives the smoke field, and the smoke field inversely enhances the thermal field or electric field, a smoke-heat-electric causal recirculation structure is generated, and the field-source coupling strength is calculated. The coherent diffusion analysis module reads the change in the center coordinates of each plume node between adjacent acquisition timestamps, generates the plume diffusion trajectory, calculates the diffusion direction, diffusion velocity, and concentration change rate, and constructs a cooperative diffusion state matrix based on the difference in direction, velocity, and concentration change rate between each plume node. When the degree of synchronous diffusion is greater than 0.65, a coherent diffusion structure of the plume is generated. The entropy collapse correction module reads the smoke plume concentration distribution state in the tree-like recursive topology, calculates the smoke plume information entropy change state, and generates a smoke plume entropy collapse structure when the smoke plume information entropy decreases for three consecutive collection timestamps and a smoke plume coherent diffusion structure exists. It then performs collapse correction on the initial smoke plume concentration peak, the initial smoke plume diffusion radius, and the initial smoke plume duration to generate the smoke plume diffusion state. The improved Gaussian smoke model was trained using the smoke diffusion state prediction error, field-source coupling strength error, and smoke coherent diffusion state error as joint optimization objectives. The parameters of the recursive smoke module, the three-field backflow coupling module, the coherent diffusion analysis module, and the entropy collapse correction module were continuously optimized. The training was completed when the change in joint loss corresponding to five consecutive iterations was less than 0.

001.

6. The method for real-time monitoring of fire hazards based on multi-sensor fusion according to claim 1, characterized in that, The phase transition states of the generated smoke plume include: Read the smoke plume diffusion state and field-source coupling strength corresponding to the continuous acquisition timestamps within the same spatial risk unit, extract the peak smoke plume concentration, smoke plume diffusion radius, smoke plume duration, synchronous diffusion degree and smoke plume information entropy change state, and calculate the smoke plume concentration growth rate, diffusion radius growth rate and duration growth rate; A smoke plume phase transition state field was constructed using spatial risk units and collection timestamps as indices. The smoke plume concentration growth rate, diffusion radius growth rate, duration growth rate, field-source coupling strength, synchronous diffusion degree, and smoke plume information entropy change state were multiplied by 0.25, 0.20, 0.15, 0.20, 0.10, and 0.10, respectively, and then added together to obtain the smoke plume phase transition order value, which was then written into the smoke plume phase transition state field. Read the smoke phase transition order values ​​corresponding to the first 12 collection timestamps in the smoke phase transition state field, calculate the average and standard deviation, and add the average to 1.50 times the standard deviation to obtain the smoke phase transition threshold; The current phase transition order value of the smoke plume in the phase transition state field is read and compared with the phase transition threshold of the smoke plume. When the phase transition order value of the smoke plume is greater than the phase transition threshold of the smoke plume and there is a coherent diffusion structure and an entropy collapse structure of the smoke plume, a phase transition transition is determined to have occurred, and a phase transition state of the smoke plume is generated.

7. The method for real-time monitoring of fire hazards based on multi-sensor fusion according to claim 1, characterized in that, The set of risk state change points includes: Based on the collection timestamp, the smoke plume diffusion state and smoke plume phase transition state within the same spatial risk unit are read, and the peak smoke plume concentration, smoke plume diffusion radius, smoke plume duration, field-source coupling strength, synchronous diffusion degree, smoke plume information entropy change state, smoke-heat-electric causal backflow structure, and smoke plume phase transition order value are extracted to construct a multidimensional risk state sequence; A risk propagation topology graph is established based on the positional adjacency relationship between spatial risk units, the direction of smoke cloud diffusion, and the field-source coupling strength. The multidimensional risk state sequence is written into the nodes of the risk propagation topology graph, and the consistency of smoke cloud diffusion direction, the difference in field-source coupling strength, and the difference in smoke cloud phase transition order value between adjacent spatial risk units are written into the edges of the risk propagation topology graph to generate a topological risk state graph. The PELT algorithm is executed based on the topological risk state diagram. According to the phase transition state of the smoke plume, the topological risk state diagram is divided into random disturbance state segment, phase transition state segment, and continuous diffusion state segment. Corresponding segmental cost weights are set for each segment to generate a phase transition driven segmental cost function. Hierarchical recursive PELT processing is performed based on a tree-like recursive topology. Recursive propagation paths between main smoke cloud nodes and child smoke cloud nodes are generated according to the parent-child mapping relationship. Change point detection is performed on the main smoke cloud nodes, child smoke cloud nodes, and recursive propagation paths according to the phase change driven piecewise cost function. Change points of the main smoke cloud, child smoke cloud, and recursive propagation paths are written into the same hierarchical change point group to generate the hierarchical topology joint change point cost. Based on the hierarchical topology joint change point cost and phase transition driven segmentation cost function, dynamic programming segmentation and candidate segmentation path pruning are performed. Candidate segmentation paths that do not conform to the risk propagation topology direction, parent-child recursive propagation order, and phase transition state are deleted. The change point location, collection timestamp, spatial risk unit number, hierarchical topology joint change point cost, and risk state confidence of the candidate segmentation path are retained to generate a risk state change point set.

8. The method for real-time monitoring of fire hazards based on multi-sensor fusion according to claim 1, characterized in that, The execution of fire alarm linkage response processing includes: Read the set of smoke cloud diffusion state, smoke cloud phase transition state and risk state change points corresponding to the same judgment window within the same spatial risk unit, and extract the field source coupling strength, synchronous diffusion degree, smoke cloud phase transition order value, phase transition judgment result, risk state confidence and change point continuity ratio to generate a set of continuous diffusion judgment indicators. The continuous diffusion score is obtained by multiplying the source-field coupling strength, synchronous diffusion degree, smoke phase transition order value, risk state confidence, change point continuity ratio, and phase transition judgment result by 0.20, 0.20, 0.20, 0.15, and 0.05, respectively. When the continuous diffusion score is less than 0.45, a Level 1 risk is generated and a risk warning is issued. When the continuous diffusion score is greater than or equal to 0.45 and less than 0.65, a Level 2 risk is generated and an early warning is issued. When the continuous diffusion score is greater than or equal to 0.65 and less than 0.80, a Level 3 risk is generated and a coordinated action is taken to activate area smoke extraction, audible and visual alarms, and electricity monitoring. When the continuous diffusion score is greater than or equal to 0.80, a Level 4 risk is generated and a fire alarm, area power outage, smoke extraction activation, and emergency notification are issued.

Images