A Smart Fire Risk Assessment and Early Warning Method Based on Multi-Source Sensor Data Fusion

Abstract

This application discloses a smart fire risk assessment and early warning method based on multi-source sensor data fusion, belonging to the field of smart fire protection and emergency safety technology. It includes: acquiring a spatiotemporally aligned sensor dataset; performing fire anomaly identification based on the spatiotemporally aligned sensor dataset and extracting fire perception feature data characterizing the fire state; constructing smoke propagation probability distribution, temperature field probability distribution, and gas concentration field probability distribution respectively, and jointly fusing these probability distributions to obtain a multi-physics field joint probability heatmap; determining the fire source location and generating corresponding location confidence; predicting fire spread based on the fire source location and location confidence to obtain the evolution results of the danger zone; generating a dynamic evacuation plan; generating early warning information based on the fire source location, the evolution results of the danger zone, and the dynamic evacuation plan, and executing fire early warning and emergency response linkage control based on the early warning information.

Description

Technical Field

[0001] This application relates to the field of smart fire protection and emergency safety technology, and more specifically, to a smart fire risk assessment and early warning method based on multi-source sensor data fusion. Background Technology

[0002] Building fires pose a significant public safety risk, threatening lives and property. With rapid urbanization, high-rise buildings, large complexes, and underground spaces are emerging in large numbers, leading to increasingly complex internal structures. Once a fire breaks out, smoke spreads rapidly, evacuation is difficult, and the risk of serious casualties is extremely high. Traditional fire warning systems typically rely on a single type of sensor (such as point smoke detectors or heat detectors) for fire detection. Limited by the inherent limitations of a single sensing mechanism, these systems suffer from significant shortcomings in early fire identification accuracy, environmental adaptability, and response speed, making it difficult to meet the precise fire warning needs of complex building environments.

[0003] To improve fire detection accuracy, existing technologies have incorporated multi-sensor fusion and machine learning methods. For example, Chinese patent application CN120299162A discloses an intelligent electrical fire identification system based on multi-dimensional sensor fusion. This system uses temperature sensors, smoke sensors, gas sensors, and pyrolysis particle sensors to construct distributed sensing nodes. It dynamically adjusts sensor response errors using a Bayesian optimization algorithm and performs dimensionality reduction preprocessing on the collected data through principal component analysis. At the data fusion level, Kalman filtering is used to suppress edge noise, and dynamic time warping (DTW) is introduced to achieve time-series alignment of multiple sensors. At the feature modeling level, an LSTM network is used to extract long-term temporal dependency features, and mutual information analysis is combined to mine multi-dimensional feature correlations. Finally, fuzzy logic decision-making is used to assess the level of fire risk, and emergency response is achieved in conjunction with graded alarm and reinforcement learning fire suppression strategies. However, the above solutions only address the single scenario of electrical fires, using sensor signals from a single location as input and outputting a scalar risk level. They lack the ability to model the physical field distribution of multiple areas in a building space, cannot determine the spatial distribution of fire source locations and hazardous areas at the building scale, and do not have the function of dynamic evacuation route planning based on building topology.

[0004] In the field of multimodal fusion and uncertain reasoning, Chinese patent application CN121389039A discloses a fire detection method and device based on the fusion of multimodal perception and DS evidence theory. This method uses precise multimodal perception, probabilistic feature representation, and evidence-level decision fusion as its core framework. It identifies smoke and fire features in video images using an improved Faster-YOLOv8 target detection model. The average detection confidence within a time window is mapped to the fire support probability of the visual modality using a Sigmoid function. Simultaneously, temperature sensing data is mapped to the fire probability of the temperature modality based on a triangle-trapezoidal combined fuzzy membership function. Then, DS evidence theory (PCR6 combination rule) is used to fuse the visual and temperature evidence, and the final fire determination probability is output after a Pignistic probability transformation. This method effectively reduces the false alarm rate of single-modal detection and improves early fire response time. However, the DS evidence fusion of this scheme only covers two modalities: visual and temperature. The output conclusion is the probability of fire occurrence in a local area of ​​the building. It does not involve the spatial probability distribution modeling of multiple physical fields such as smoke propagation and gas concentration, nor does it perform spatial mapping with Building Information Modeling (BIM). Therefore, it cannot output fire source location confidence and hazard area evolution prediction results with spatial semantics, making it difficult to support dynamic fire protection decision-making at the entire building scale.

[0005] In the area of ​​building-level fire alarm linkage control, Chinese patent application CN121500838A discloses an intelligent linkage control system for video surveillance, access control, and fire alarms. This system integrates three types of heterogeneous data: video images, access control status, and fire alarm signals. It uses DS evidence theory to fuse the confidence levels of multiple types of evidence, classifying the fire situation into four levels. Regarding evacuation planning, a weighted graph model based on building information modeling (BIM) with rooms, corridors, and safety exits as nodes is constructed, employing an improved A / B algorithm. The algorithm calculates the optimal evacuation route and coordinates with video tracking, access control, and fire alarm broadcasts. However, the multi-source data fusion output in this scheme is a discrete fire situation level rather than a continuous spatial probability distribution, lacking quantitative modeling of the physical evolution of smoke propagation, temperature field, and gas concentration field. Its evacuation route planning uses a static graph model, failing to construct a time-step-based directed graph of building topology, thus unable to dynamically update the passability status of nodes at each moment during fire spread, and also not supporting the solution of the time-series shortest path under traffic capacity constraints. The generated evacuation scheme fails to accurately reflect the impact of dynamic fire evolution on the timeliness of evacuation routes. Furthermore, the system lacks a closed-loop dynamic verification mechanism for the prediction results of hazardous area evolution, and cannot correct the prediction results in real time during the execution of coordinated control.

[0006] Based on the above-mentioned existing technologies, the main shortcomings of current fire risk assessment and early warning methods are as follows: First, multi-sensor data fusion is limited to perception and judgment of a single location or limited modes, lacking the ability to jointly model the probability distribution of smoke propagation, temperature field, and gas concentration field in multiple spatial units of a building, and thus failing to generate a multi-physics joint probability heat map with spatial resolution. Second, existing solutions generally lack a mechanism for deep integration of physical field probability analysis results with building information models, making it difficult to combine building geometric constraints, access control status, and visual detection information to finely correct the fire probability of each spatial unit, thereby outputting fire source location confidence with spatial semantics. Third, existing technologies are insufficient in dynamically predicting the fire spread process, and evacuation route planning fails to fully consider the constraints of fire temporal evolution on route accessibility, making it difficult to generate dynamic evacuation plans that meet temporal feasibility under accessibility limitations. Fourth, existing systems lack a closed-loop dynamic verification and stable convergence judgment mechanism for prediction results, failing to guarantee the continuous accuracy of early warning output. Therefore, it is necessary to provide a smart fire risk assessment and early warning method based on multi-source sensor data fusion to overcome the shortcomings of the existing technologies. Summary of the Invention

[0007] To overcome a series of shortcomings in existing technologies, the purpose of this application is to provide a smart fire risk assessment and early warning method based on multi-source sensor data fusion, which includes the following steps: Acquire multi-source sensor data within the building, and perform standardization, time synchronization, and spatial correlation fusion processing on the multi-source sensor data to obtain a spatiotemporally aligned sensor dataset; Fire anomaly identification is performed based on spatiotemporal aligned sensor datasets, and fire perception feature data representing fire status are extracted; Based on fire perception feature data, the probability distributions of smoke propagation, temperature field, and gas concentration field are constructed respectively, and the probability distributions are jointly fused to obtain a multi-physics field joint probability heat map. The multiphysics joint probability heatmap is mapped to the corresponding spatial unit of the building information model, and the fire probability corresponding to each spatial unit is corrected by combining building geometric constraint information, access control status information and visual detection information. Based on the correction results, the fire source location is determined and the corresponding location confidence is generated. Fire spread prediction is performed based on the location of the fire source and the location reliability, and the evolution results of the danger zone are obtained; Based on the evolution results of hazardous areas, building topology information, and personnel distribution information, evacuation routes are planned to generate dynamic evacuation plans. Early warning information is generated based on the location of the fire source, the evolution of the danger zone, and the dynamic evacuation plan, and fire early warning and emergency response control are executed based on the early warning information.

[0008] In some embodiments, the method for constructing the smoke propagation probability distribution includes: Obtain the smoke concentration value, smoke concentration change gradient, and building ventilation system operation parameters for each spatial unit; A smoke diffusion model was constructed based on smoke concentration values, smoke concentration variation gradients, and building ventilation system operating parameters. The smoke concentration distribution in each spatial unit within a preset time window is predicted using a smoke diffusion model. Based on the smoke diffusion model and combined with preset parameter perturbation rules, multiple simulation calculations are performed to generate a smoke concentration sample set for each spatial unit. Based on the smoke concentration sample set and the preset visibility hazard threshold, the smoke propagation probability of each spatial unit at each predicted time is calculated. Based on the probability of smoke hazard state corresponding to each spatial unit at each prediction time, a smoke propagation probability distribution is constructed.

[0009] In some embodiments, the method for constructing the probability distribution of the temperature field and the probability distribution of the gas concentration field is as follows: Acquire temperature monitoring data and gas concentration monitoring data for each spatial unit within the building; The temperature field distribution of the building space is constructed based on temperature monitoring data, and the gas concentration field distribution of the building space is constructed based on gas concentration monitoring data. Based on the temperature field distribution and gas concentration field distribution, the evolution process of the temperature field and the evolution process of the gas concentration field within a preset time window are predicted respectively, and the prediction results of the temperature field and the gas concentration field are obtained. The temperature hazard probability of each space unit is calculated based on the temperature field prediction results and the preset temperature hazard threshold, and the gas hazard probability of each space unit is calculated based on the gas concentration field prediction results and the preset gas hazard threshold. A temperature field probability distribution is constructed based on the temperature hazard probability, and a gas concentration field probability distribution is constructed based on the gas hazard probability.

[0010] In some embodiments, the method for obtaining the multiphysics joint probability heatmap is as follows: Based on the probability of smoke propagation, the probability of temperature hazard, and the probability of gas hazard, the hazard confidence data corresponding to each space unit is determined. Calculate the degree of conflict of evidence among various risk factors based on risk confidence data; Based on the degree of conflict of evidence, conflict correction and fusion processing are performed on the hazard credibility data to obtain the comprehensive hazard credibility. The joint probability value corresponding to each spatial unit is determined based on the comprehensive risk confidence level, and a joint probability field is constructed. A multiphysics joint probability heatmap is generated based on the joint probability field and the spatial positional relationships between each spatial unit.

[0011] In some embodiments, the method for determining the location of the fire source and generating location confidence is as follows: Candidate fire source areas are determined based on the corrected fire probability of each spatial unit; The initial fire source location is determined based on the probability distribution characteristics of the candidate fire source areas; When there are multiple local probability extremes in the candidate fire source area, the visual detection results of the corresponding area are obtained, and the initial fire source position is corrected based on the visual detection results. The location reliability is calculated based on the fire probability corresponding to the corrected fire source location, the consistency information of multi-physics evidence, and the visual detection results. The fire source location result is generated based on the corrected fire source location and the location reliability.

[0012] In some embodiments, the method for obtaining the evolution results of hazardous areas is as follows: Determine the initial spatial unit for the spread of fire based on the location of the fire source; Determine the confidence zone of the fire source based on its location and location confidence level; A fire spread model is constructed based on the connectivity and attribute information between building space units; Based on the fire source confidence zone and fire spread model, the fire spread process within a preset time window is predicted, and the fire evolution results of each spatial unit are obtained. Determine the probability of danger for each spatial unit based on the fire evolution results; The danger zones are determined based on the danger probabilities corresponding to each prediction time, and the evolution results of the danger zones are generated.

[0013] In some embodiments, the method for generating a dynamic evacuation plan is as follows: Based on the evolution results of hazardous areas and building topology information, a time-series directed graph of building topology based on time steps is constructed. Based on the danger zone boundaries corresponding to each time step, the traversability status of nodes and edges in the building topology time-series directed graph is dynamically updated. Starting from the current location of each person in the building and ending at each safety exit, the shortest path in time is solved on the dynamically updated directed graph of the building topology. Based on the capacity constraints of each directed edge, the shortest path in time sequence is modified for path feasibility to obtain evacuation paths that satisfy the capacity constraints. Generate evacuation route plans based on individuals or groups of individuals, and output the corresponding route node sequence and the estimated arrival time of each node.

[0014] In some embodiments, the method for performing fire early warning and emergency response linkage control based on early warning information is as follows: Based on the early warning information, the location of the fire source and the extent of the danger zone are determined; The building space is divided into dangerous areas, adjacent areas, and safe areas based on the location of the fire source and the extent of the danger zone. Based on the zoning results, the audible and visual alarm system is controlled in a hierarchical manner to achieve differentiated alarm prompts; Evacuation guidance information is generated based on the dynamic evacuation plan, and the evacuation guidance information is output to each area; Fire compartmentation facilities are linked and controlled based on the location of the fire source and the extent of the hazardous area in order to achieve fire compartmentation and isolation. The smoke exhaust and ventilation systems are linked and controlled based on the location of the fire source and the range of the hazardous area in order to achieve directional emission and diffusion suppression of smoke. The elevator system is linked and controlled based on the location of the fire source and the range of the danger zone to perform emergency elevator landing and switch operating modes.

[0015] In some embodiments, the intelligent fire risk assessment and early warning method further includes: During the execution of fire early warning and emergency response control, the evolution results of the dangerous area are dynamically verified based on real-time multi-source sensor data to determine whether the evolution results of the dangerous area meet the preset stable convergence conditions. If not satisfied, the fire perception feature data, multi-physics joint probability heat map, fire source location and location confidence are updated based on real-time multi-source sensor data, and the fire spread prediction and evacuation route planning are re-executed to update the evolution results of the danger zone and the dynamic evacuation plan. If the conditions are met, the final dynamic evacuation plan and early warning information will be output.

[0016] In some embodiments, the method for determining whether the evolution result of the dangerous area meets the preset stability convergence condition is as follows: During the early warning and linkage control process, real-time multi-source sensor data and the corresponding danger zone evolution results are obtained according to the preset dynamic verification cycle. The real-time multi-source sensor data and the evolution results of the dangerous area are time-series aligned, and a spatial unit mapping relationship between the predicted state and the measured state is established. Based on the spatial unit mapping relationship, the consistency error of the hazardous area range, temperature field distribution and smoke propagation boundary is verified; Determine whether the current verification cycle meets the preset stability threshold based on the consistency error verification results; When the stability requirements are met for multiple consecutive verification cycles, the evolution result of the dangerous region is determined to meet the preset stable convergence condition; otherwise, the evolution result of the dangerous region is determined to not meet the convergence condition, and the dangerous region evolution model update or recalculation process is triggered.

[0017] Compared with the prior art, this application has the following beneficial effects: This application achieves precise fire source location, dynamic prediction of hazardous areas, and adaptive optimization of personnel evacuation routes through multi-source sensor data fusion and multi-physics joint probabilistic modeling. Combined with hierarchical early warning and multi-system linkage control, it significantly improves the accuracy of building fire situation awareness, the reliability of risk assessment, and the real-time performance and safety of emergency response. Attached Figure Description

[0018] Figure 1 This is the overall flowchart of the intelligent fire risk assessment and early warning method based on multi-source sensor data fusion proposed in this application.

[0019] Figure 2 This is a flowchart illustrating the method for acquiring multi-source sensor data and constructing a spatiotemporally aligned sensor dataset in this application.

[0020] Figure 3 This is a flowchart of the method for acquiring fire perception feature data and generating multiphysics joint probability heatmaps in this application.

[0021] Figure 4 This is a flowchart illustrating the method for determining the location of the fire source, obtaining the evolution results of the hazardous area, and generating a dynamic evacuation plan in this application.

[0022] Figure 5 This is a flowchart illustrating the method for determining the stable convergence condition of the evolution results of the hazardous area in this application. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the embodiments of this invention will be described in more detail below with reference to the accompanying drawings. In the drawings, the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The described embodiments are some embodiments of this invention, but not all embodiments.

[0024] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] The embodiments and directional terms described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0026] Example 1 This embodiment provides an overall process for a smart fire risk assessment and early warning method based on multi-source sensor data fusion, targeting a multi-story office building scenario. The configured sensor network includes smoke detectors, temperature sensors, gas concentration sensors, infrared thermal imaging cameras, and visible light cameras, and is integrated with an access control management system. The description of this embodiment focuses on the overall step architecture of the method.

[0027] like Figure 1 As shown, the method described in this embodiment includes the following steps: The multi-source sensor data collected inside the building are sequentially processed through standardization, time synchronization, and spatial correlation fusion to obtain a spatiotemporally aligned sensor dataset. Fire anomaly identification is performed using a spatiotemporally aligned sensing dataset as input, and fire perception feature data representing the fire status are extracted from it; Based on fire perception feature data, the probability distributions of smoke propagation, temperature field, and gas concentration field are constructed respectively, and the probability distributions are jointly fused to obtain a multi-physics field joint probability heat map. After mapping the multiphysics joint probability heatmap to the corresponding spatial units in the building information model, building geometric constraint information, access control status information and visual detection information are introduced to correct the fire probability corresponding to each spatial unit. Based on the correction results, the fire source location is determined and the corresponding location confidence is generated. Using the location of the fire source and the location confidence level as inputs, fire spread prediction is performed to obtain the evolution results of the danger zone; Based on the evolution results of hazardous areas, building topology information, and personnel distribution information, evacuation route planning is carried out to generate dynamic evacuation plans; Early warning information is generated based on the location of the fire source, the evolution of the danger zone, and the dynamic evacuation plan, and the execution of fire early warning and emergency response control is carried out based on this early warning information; During the execution of fire early warning and emergency response control, real-time multi-source sensor data is continuously used to dynamically verify the evolution results of the dangerous area to determine whether it meets the preset stable convergence conditions. For cases where the preset stable convergence conditions are not met, the fire perception feature data, multi-physics joint probability heat map, fire source location and location confidence are updated based on real-time multi-source sensor data, and the fire spread prediction and evacuation route planning process is re-triggered to update the evolution results of the danger zone and the dynamic evacuation plan. If the preset stable convergence conditions have been met, the final dynamic evacuation plan and early warning information will be output.

[0028] The aforementioned intelligent fire risk assessment and early warning method based on multi-source sensor data fusion in this application achieves spatiotemporal consistency in the expression of heterogeneous multi-source data within a building by standardizing, synchronizing, and spatially associating multi-source sensor data such as smoke, temperature, gas, and visual data, thereby improving the completeness and consistency of fire perception. Furthermore, by constructing fire perception features and generating a multi-physics joint probability heatmap of smoke propagation, temperature field, and gas concentration field, it achieves probabilistic characterization and fusion enhancement of multi-dimensional risks in the early stages of a fire, improving the accuracy of fire identification and location. Finally, it incorporates building information model constraints. Access control status and visual detection information are used to correct the probability of fire in spatial units, achieving high-precision location and confidence assessment of fire source positions; dynamic evacuation route planning is performed by combining fire source positions with the evolution prediction of dangerous areas and personnel distribution information, enabling real-time optimization and dynamic updates of evacuation strategies; and dynamic verification and convergence judgment of dangerous area evolution results are continuously performed based on real-time multi-source sensor data during the linkage control process, and the model results in the non-converged state are iteratively updated, thereby significantly improving the stability of fire situation prediction, the reliability of early warning decisions, and the adaptive capability of emergency linkage control.

[0029] Example 2 This embodiment, based on Embodiment 1, further explains the acquisition method of multi-source sensor data and the construction method of spatiotemporally aligned sensor dataset. It is applicable to large complex buildings with distributed sensor networks and mainly addresses the problems of heterogeneous data from multiple types of sensors, time asynchrony, and inconsistent spatial mapping.

[0030] like Figure 2 As shown, in this embodiment, the method for acquiring multi-source sensor data within the building is as follows: The smoke monitoring signals collected inside the building are preprocessed and fused sequentially to obtain smoke sensing data; The temperature monitoring signals collected inside the building are preprocessed to obtain temperature sensing data; The gas concentration monitoring signals collected inside the building are preprocessed and time-aligned sequentially to obtain gas concentration sensing data. Infrared thermal imaging data and visible light video data collected inside the building are preprocessed in a unified manner to obtain visual perception data; Access control status data and personnel access record data collected within the building are processed in a time-series manner to obtain access control status data; Smoke sensor data, temperature sensor data, gas concentration sensor data, visual perception data, and access control status data are aggregated and fused to obtain multi-source sensor data within the building.

[0031] For example, consider a building with a floor area of ​​30,000 m². 2Taking a commercial complex as an example, a total of 120 smoke sensors, 80 temperature sensors, 40 combustible gas sensors, 20 infrared thermal imaging cameras, 30 visible light cameras, and 15 access control devices are deployed in the underground parking lot, equipment room, power distribution room, and public corridors. Within a certain monitoring period, the smoke sensor collected smoke concentration monitoring signals at a sampling interval of 1 second. The smoke concentration at a monitoring point in the underground parking lot increased from 0.03 dB / m³ to 0.18 dB / m³. After filtering, noise reduction, and multi-point fusion processing, smoke sensing data was generated. The temperature sensor collected ambient temperature monitoring signals, where the temperature in the corresponding area increased from 28.5℃ to 42.3℃. After outlier removal and standardization processing, temperature sensing data was generated. The gas sensor collected combustible gas concentration monitoring signals, where the carbon monoxide concentration increased from 12 ppm to 68 ppm. After data correction and time alignment processing, gas concentration sensing data was generated. The infrared thermal imaging camera collected thermal imaging data, and analysis revealed that the highest temperature in a local area reached 85.6℃. The visible light camera simultaneously acquired corresponding video footage, which, after image enhancement and time synchronization processing, formed visual perception data. The access control system recorded a total of 32 personnel passage records and 3 access control opening records within 5 minutes in the corresponding area. After time-series processing, access control status data was generated.

[0032] Subsequently, the smoke sensor data, temperature sensor data, gas concentration sensor data, visual perception data, and access control status data were correlated and fused according to a unified timestamp and spatial location identifier to construct a multi-source sensor data system for the building. This system includes data such as smoke concentration of 0.18 dB / m³, ambient temperature of 42.3℃, carbon monoxide concentration of 68 ppm, maximum thermal imaging temperature of 85.6℃, and real-time personnel flow information. This multi-source sensor data comprehensively reflects the environmental status, fire risk status, and personnel activity status of the target area, providing a data foundation for subsequent fire anomaly identification, risk level assessment, and early warning linkage control.

[0033] The method for acquiring multi-source sensor data within a building as described in this application generates high-quality smoke sensor data by preprocessing and fusing smoke monitoring signals collected within the building; obtains stable temperature sensor data by preprocessing temperature monitoring signals; achieves temporally consistent expression of gas concentration data by performing preprocessing and time alignment processing on gas concentration monitoring signals; forms visual perception data with spatial and visual semantic consistency by uniformly preprocessing infrared thermal imaging data and visible light video data; constructs a structured expression of personnel activity and spatial passage status by temporally organizing access control status data and personnel access record data; and, based on this, converges and fuses smoke sensor data, temperature sensor data, gas concentration sensor data, visual perception data, and access control status data to form a unified multi-source sensor dataset within the building, thereby significantly improving the multi-dimensional information integrity, spatiotemporal consistency of data, and the quality and reliability of basic data for subsequent fire risk assessment in building fire perception.

[0034] See also Figure 2 In this embodiment, the method for constructing the spatiotemporal aligned sensing dataset is as follows: Standardized sensor data is obtained by performing dimensional unification processing and abnormal data correction processing on multi-source sensor data within the building. The standardized sensor data is timestamped based on a unified clock source, and resampling is performed according to a preset reference sampling period to obtain time-aligned sensor data. Missing data in the time-aligned sensing data is filled in, and corresponding data validity identifiers are generated for each data point to obtain complete time-series sensing data. Based on building information modeling, establish the spatial relationship between sensor installation locations and building space units; Based on spatial relationships, complete temporal sensing data is mapped to its corresponding building space unit to obtain spatial unit sensing data; Spatial fusion processing is performed on the spatial unit sensing data to obtain a spatiotemporally aligned sensing dataset.

[0035] For example, taking the aforementioned commercial complex building as an example, the multi-source sensor data acquired at a certain monitoring time includes: smoke concentration of 0.18 dB / m³ in area A of the underground parking lot, ambient temperature of 42.3℃, carbon monoxide concentration of 68 ppm, highest temperature detected by infrared thermal imaging of 85.6℃, and 32 records of personnel passage within the corresponding area within 5 minutes. First, the smoke concentration, temperature, gas concentration, and visual perception data are processed to unify their dimensions, and a sliding window anomaly detection algorithm is used to correct abnormal data. For example, the abnormal temperature value of 128℃ caused by communication interference is corrected to the interpolated result of 43.1℃ at a nearby time, thus obtaining standardized sensor data.

[0036] Subsequently, using the unified clock source provided by the building fire monitoring center as a benchmark, the standardized sensor data was timestamped and resampled according to a 1-second benchmark sampling period. Specifically, the original smoke sensor sampling period was 1 second, the temperature sensor sampling period was 2 seconds, the gas sensor sampling period was 5 seconds, and the video analysis result output period was 0.5 seconds. After resampling, time-aligned sensor data with a 1-second interval was uniformly generated.

[0037] Furthermore, missing data in the time-aligned sensing data is completed. For example, if gas concentration data is missing at a certain moment due to network transmission delay, linear interpolation is performed using data from adjacent moments to complete the missing data, resulting in a completed gas concentration value of 66 ppm. An "interpolation completion" validity label is generated for this data. Normally acquired data is assigned an "original valid" label, thus forming complete time-series sensing data.

[0038] In the spatial association processing stage, a mapping relationship between sensors and building space units is established based on the Building Information Model (BIM). For example, the SM-15 smoke sensor, TM-08 temperature sensor, GM-06 gas sensor, and their corresponding infrared thermal imaging camera are associated with the space unit of area A in the underground parking lot. Subsequently, the complete time-series sensor data is mapped to the corresponding space units, forming sensor data for space units such as area A of the underground parking lot, area B of the power distribution room, and area C of the equipment room.

[0039] Finally, spatial fusion processing is performed on multi-source sensor data within the same spatial unit. For example, for area A of the underground parking lot, smoke concentration of 0.18 dB / m³, ambient temperature of 42.3℃, carbon monoxide concentration of 68 ppm, maximum thermal imaging temperature of 85.6℃, and personnel flow information are correlated and fused to form a comprehensive state feature vector for this spatial unit. After summarizing the comprehensive state feature vectors of all spatial units, a spatiotemporally aligned sensor dataset is obtained, providing a unified data foundation for subsequent fire anomaly identification, fire source location, and hazardous area evolution analysis.

[0040] The method for constructing the spatiotemporally aligned sensor dataset described in this application achieves standardized representation of heterogeneous sensor data by performing dimensional unification processing and anomaly correction on multi-source sensor data within a building; it achieves strict alignment and consistent representation of multi-source data in the time dimension by performing timestamp correction on the standardized sensor data based on a unified clock source and performing resampling processing in conjunction with a preset benchmark sampling period; it constructs a complete and traceable time-series sensor data system by performing missing data completion processing on the time-aligned sensor data and generating data validity identifiers; it establishes spatial associations between sensor installation locations and building spatial units based on building information models and maps time-series sensor data to corresponding spatial units, achieving spatial semantic binding of data; and it generates a spatiotemporally consistent spatiotemporally aligned sensor dataset by performing spatial fusion processing on the spatial unit sensor data, thereby significantly improving the expressive ability of multi-source sensor data in terms of time consistency, spatial consistency, and data integrity, providing highly reliable basic data support for subsequent fire identification and risk assessment.

[0041] Example 3 Based on Example 2, this embodiment further explains the method for acquiring fire perception feature data, the method for constructing multi-physics probability distribution, and the joint fusion processing method, and is applicable to application scenarios that require accurate characterization of complex fire conditions.

[0042] like Figure 3 As shown, in this embodiment, the method for obtaining fire perception feature data is as follows: Input the spatiotemporal alignment sensing dataset into a pre-trained multivariate time series anomaly detection model and obtain its output reconstructed time series; The reconstruction error sequence is calculated using the deviation between the spatiotemporally aligned sensing dataset and the reconstructed time series. The adaptive anomaly detection threshold is determined based on the reconstruction error sequence during the historical normal operation phase. By combining the reconstructed error sequence with the adaptive anomaly detection threshold, abnormal candidate events are identified; Perform multi-source consistency verification on abnormal candidate events to identify abnormal fire events; Determine the start time of the anomaly and the corresponding set of anomaly spatial unit numbers based on the fire anomaly event; The smoke concentration change rate, temperature rise rate, and gas concentration offset are extracted from the abnormal spatial cells, and the above data are organized and processed to obtain fire perception feature data.

[0043] For example, based on the spatiotemporally aligned sensor dataset obtained from the aforementioned commercial complex building, fire anomaly identification is performed on sensor data from areas A of the underground parking lot, area B of the power distribution room, and area C of the equipment room. First, a continuous 24-hour spatiotemporally aligned sensor dataset is input into a pre-trained multivariate time-series anomaly detection model. The model input features include multidimensional feature parameters such as smoke concentration, ambient temperature, carbon monoxide concentration, thermal imaging temperature, and the intensity of human activity. The model outputs the corresponding reconstructed time series.

[0044] Subsequently, the actual spatiotemporal alignment sensor dataset was compared with the reconstructed time series output by the model, and the reconstruction error sequence at each time point was calculated. For example, during the normal operation phase of area A in the underground parking lot, the mean reconstruction error of the model was 0.08, and the standard deviation was 0.03. When the smoke concentration rapidly increased from 0.05 dB / m³ to 0.18 dB / m³, the ambient temperature increased from 31.2℃ to 42.3℃, and the carbon monoxide concentration increased from 15 ppm to 68 ppm at a certain moment, the corresponding reconstruction error increased to 0.62.

[0045] Furthermore, using the reconstruction error sequence from the historical 30-day normal operation phase as a reference, an adaptive anomaly detection threshold was determined using the mean plus three standard deviations method. The calculated anomaly detection threshold was 0.17. When the real-time reconstruction error reached 0.62 and remained there for more than 10 seconds, it was identified as an anomaly candidate event.

[0046] Subsequently, a multi-source consistency check was performed on the candidate abnormal event. Specifically, it was determined whether the rate of change of smoke concentration, the rate of temperature rise, and the deviation of gas concentration simultaneously exceeded preset thresholds. The smoke concentration change rate reached 0.013 dB / m·s, exceeding the preset threshold of 0.005 dB / m·s; the rate of temperature rise reached 1.11℃ / min, exceeding the preset threshold of 0.5℃ / min; and the carbon monoxide concentration deviation reached 53 ppm, exceeding the preset threshold of 20 ppm. Simultaneously, infrared thermal imaging data showed that the highest temperature in the local area reached 85.6℃. Therefore, this candidate abnormal event was determined to be a fire-related abnormal event.

[0047] After confirming the fire anomaly, the time of its first occurrence was recorded as 14:26:18, and the corresponding set of anomaly spatial unit numbers was identified as {A-03}, where A-03 represents the building spatial unit corresponding to area A of the underground parking lot. Subsequently, fire-related characteristic parameters were extracted from the anomaly spatial unit, including the smoke concentration change rate of 0.013 dB / m·s, the temperature rise rate of 1.11℃ / min, the carbon monoxide concentration shift of 53 ppm, the highest temperature in thermal imaging of 85.6℃, and the duration of the anomaly. These parameters were then organized and processed according to a unified data structure to form fire perception characteristic data.

[0048] The method for acquiring fire perception feature data described in this application involves inputting a spatiotemporally aligned sensing dataset into a pre-trained multivariate time series anomaly detection model to obtain a reconstructed time series. A reconstruction error sequence is calculated based on the deviation between the spatiotemporally aligned sensing dataset and the reconstructed time series, achieving a preliminary quantitative characterization of abnormal behavior. An adaptive anomaly judgment threshold is constructed based on the reconstruction error sequence from historical normal operation phases, and anomaly candidate events are identified using the reconstruction error sequence, improving the adaptability and robustness of anomaly detection. Multi-source consistency verification is performed on the anomaly candidate events to filter out fire anomaly events and determine the anomaly initiation time and corresponding set of anomaly spatial units, achieving precise spatiotemporal localization of fire anomalies. Based on this, key evolutionary features such as smoke concentration change rate, temperature rise rate, and gas concentration shift are extracted from the anomaly spatial units and structurally organized to generate fire perception feature data. This significantly improves the accuracy of early fire anomaly identification, spatiotemporal localization capability, and multi-source feature expression capability, providing highly reliable feature support for subsequent fire risk assessment and early warning.

[0049] See also Figure 3 In this embodiment, the method for constructing the smoke propagation probability distribution includes: Obtain the smoke concentration value, smoke concentration change gradient, and building ventilation system operation parameters for each spatial unit; A smoke diffusion model is constructed by taking smoke concentration values, smoke concentration change gradients, and building ventilation system operating parameters as inputs. The smoke concentration distribution in each spatial unit within a preset time window is predicted using a smoke diffusion model. Based on the smoke diffusion model, multiple simulation calculations are performed in combination with preset parameter perturbation rules to generate a set of smoke concentration samples for each spatial unit. Based on the smoke concentration sample set and the preset visibility hazard threshold, the smoke propagation probability of each spatial unit at each predicted time is calculated; Based on the probability of smoke hazard state corresponding to each spatial unit at each prediction time, a smoke propagation probability distribution is constructed.

[0050] For example, in the aforementioned commercial complex building, a fire incident has been identified in spatial unit A-03 of area A in the underground parking garage. Using A-03 as the starting spatial unit and combining the spatial topology constructed by the building information model, the smoke diffusion situation within the next 10 minutes after the fire occurs is predicted and analyzed.

[0051] First, the smoke concentration values, smoke concentration gradients, and building ventilation system operating parameters for each spatial unit were obtained. Specifically, the current smoke concentration in spatial unit A-03 is 0.18 dB / m³, with a smoke concentration gradient of 0.013 dB / m³·s; the smoke concentrations in its adjacent spatial units A-02, A-04, and A-05 are 0.07 dB / m³, 0.05 dB / m³, and 0.03 dB / m³, respectively. The building's mechanical exhaust system is currently operational, with an exhaust volume of 18,000 m³ / s. 3 / h, air supply volume is 12000m³ 3 / h, the main ventilation duct wind speed is 3.5m / s.

[0052] Subsequently, the smoke concentration value, smoke concentration change gradient, and building ventilation system operating parameters are input into the smoke diffusion model to predict the smoke diffusion process of each spatial unit within the next 10 minutes. For example, at the prediction time T+3 minutes, the predicted smoke concentration for spatial unit A-03 is 0.31 dB / m³, for spatial unit A-02 it is 0.18 dB / m³, and for spatial unit A-04 it is 0.15 dB / m³; at the prediction time T+5 minutes, the predicted smoke concentrations for the above spatial units reach 0.42 dB / m³, 0.26 dB / m³, and 0.23 dB / m³, respectively.

[0053] Furthermore, to account for uncertainties such as fluctuations in the ignition source release rate, changes in ventilation system airflow, and environmental disturbances, 1000 Monte Carlo simulations were conducted based on the smoke diffusion model according to preset parameter perturbation rules. The perturbation ranges for the ignition source smoke release rate, exhaust volume, and wind speed were set to ±15%, ±10%, and ±8%, respectively. After multiple simulations, smoke concentration sample sets were generated for each spatial unit.

[0054] Subsequently, using a smoke concentration of 0.25 dB / m, corresponding to the visibility hazard threshold, as the preset visibility hazard threshold, statistical analysis was performed on the smoke concentration sample sets of each spatial unit. For example, at the prediction time T+5 min, 980 out of 1000 simulation samples in spatial unit A-03 exceeded 0.25 dB / m, corresponding to a smoke hazard probability of 98%; 820 samples in spatial unit A-02 exceeded the threshold, corresponding to a smoke hazard probability of 82%; 760 samples in spatial unit A-04 exceeded the threshold, corresponding to a smoke hazard probability of 76%; and 410 samples in spatial unit A-05 exceeded the threshold, corresponding to a smoke hazard probability of 41%.

[0055] Finally, a smoke propagation probability distribution is constructed based on the smoke hazard probability of each spatial unit at each prediction time. For example, at prediction time T+5min, spatial units A-03, A-02, and A-04 are all in high-probability smoke hazard areas, while spatial unit A-05 is in a medium-probability smoke hazard area. As the prediction time progresses to T+10min, the smoke hazard area further expands to adjacent spatial units, thus forming a smoke propagation probability distribution map covering the entire underground parking area. This provides a basis for subsequent hazard area evolution analysis, personnel evacuation route planning, and fire linkage control.

[0056] The method for constructing the smoke propagation probability distribution described in this application achieves a multi-dimensional characterization of factors influencing smoke diffusion within a building by acquiring the smoke concentration value, smoke concentration change gradient, and building ventilation system operating parameters corresponding to each spatial unit. By using these parameters as inputs to construct a smoke diffusion model, the method predicts the smoke concentration distribution of each spatial unit within a preset time window, achieving dynamic simulation and temporal evolution characterization of the smoke propagation process. Multiple simulation calculations are performed based on preset parameter perturbation rules introduced into the smoke diffusion model to generate a smoke concentration sample set for each spatial unit, enhancing the model's adaptability to uncertainties. The smoke propagation probability of each spatial unit at different prediction times is calculated by combining the smoke concentration sample set with a preset visibility hazard threshold, achieving a probabilistic expression of the risk state. Based on this, a smoke propagation probability distribution is constructed according to the smoke hazard state probability corresponding to each spatial unit at each prediction time, thereby significantly improving the predictability of the smoke diffusion process, the robustness of risk assessment, and the accuracy of early fire spread pattern modeling.

[0057] See also Figure 3 In this embodiment, the method for constructing the temperature field probability distribution and the gas concentration field probability distribution is as follows: Acquire temperature monitoring data and gas concentration monitoring data for each spatial unit within the building; The temperature field distribution of the building space is constructed based on temperature monitoring data, and the gas concentration field distribution of the building space is constructed based on gas concentration monitoring data. Based on the temperature field distribution and gas concentration field distribution, the evolution process of the temperature field and the evolution process of the gas concentration field within the preset time window are predicted respectively, and the prediction results of the temperature field and the gas concentration field are obtained. The temperature hazard probability of each space unit is calculated based on the temperature field prediction results and the preset temperature hazard threshold, and the gas hazard probability of each space unit is calculated based on the gas concentration field prediction results and the preset gas hazard threshold. A temperature field probability distribution is constructed based on the probability of temperature hazard, and a gas concentration field probability distribution is constructed based on the probability of gas hazard.

[0058] For example, after an abnormal fire event occurs in area A of the underground parking lot of the aforementioned commercial complex, the temperature field evolution and gas concentration field evolution within the next 10 minutes are predicted and analyzed based on the real-time temperature monitoring data and gas concentration monitoring data collected from each spatial unit.

[0059] First, temperature and gas concentration monitoring data were obtained for each spatial unit within the building. Specifically, the current ambient temperature of the abnormal spatial unit A-03 was 42.3℃, while the ambient temperatures of adjacent spatial units A-02, A-04, and A-05 were 35.6℃, 34.2℃, and 31.8℃, respectively. Simultaneously, the carbon monoxide concentration in spatial unit A-03 was 68 ppm, while the carbon monoxide concentrations in adjacent spatial units A-02, A-04, and A-05 were 32 ppm, 28 ppm, and 15 ppm, respectively.

[0060] Subsequently, the temperature field distribution of the building space is constructed using temperature monitoring data from each spatial unit, and the gas concentration field distribution of the building space is constructed using gas concentration monitoring data from each spatial unit. For example, based on the spatial topology and heat conduction relationship in the building information model, A-03 is determined as the central region of the spatial temperature field; at the same time, the gas concentration diffusion relationship is established based on the building ventilation conditions and airflow direction to form the gas concentration field distribution at the current moment.

[0061] Furthermore, based on the temperature field distribution and gas concentration field distribution, the evolution of the temperature field and gas concentration field within the next 10 minutes is predicted. Specifically, at the prediction time T+5 minutes, the predicted temperature of space unit A-03 reaches 78.5℃, the predicted temperature of space unit A-02 reaches 61.8℃, and the predicted temperature of space unit A-04 reaches 58.3℃; at the prediction time T+10 minutes, the predicted temperature of space unit A-03 reaches 112.6℃, the predicted temperature of space unit A-02 reaches 86.4℃, and the predicted temperature of space unit A-04 reaches 81.7℃.

[0062] Meanwhile, at prediction time T+5 min, the carbon monoxide concentration predicted by space unit A-03 reached 165 ppm, the carbon monoxide concentration predicted by space unit A-02 reached 118 ppm, and the carbon monoxide concentration predicted by space unit A-04 reached 103 ppm; at prediction time T+10 min, the carbon monoxide concentrations of the above space units reached 286 ppm, 214 ppm and 198 ppm respectively, thus obtaining the temperature field prediction results and the gas concentration field prediction results.

[0063] Subsequently, to account for uncertainties such as variations in the heat release rate of the fire source, differences in heat transfer within the building envelope, and fluctuations in the ventilation system, 1000 random perturbation simulations were performed on both the temperature field prediction results and the gas concentration field prediction results, generating corresponding temperature and gas concentration sample sets. A preset temperature hazard threshold of 60℃ and a preset gas hazard threshold of 120ppm were used to calculate the probability of hazard.

[0064] For example, at the prediction time T+5min, 972 out of 1000 temperature simulation results for space element A-03 exceeded 60℃, corresponding to a temperature danger probability of 97.2%; 815 out of 1000 results for space element A-02 exceeded 60℃, corresponding to a temperature danger probability of 81.5%; and 736 out of 1000 results for space element A-04 exceeded 60℃, corresponding to a temperature danger probability of 73.6%.

[0065] Meanwhile, at the prediction time T+5min, among the 1000 sets of gas concentration simulation results for space element A-03, 986 sets exceeded 120ppm, with a gas hazard probability of 98.6%; for space element A-02, 842 sets exceeded 120ppm, with a gas hazard probability of 84.2%; for space element A-04, 791 sets exceeded 120ppm, with a gas hazard probability of 79.1%; and for space element A-05, 358 sets exceeded 120ppm, with a gas hazard probability of 35.8%.

[0066] Finally, a temperature field probability distribution is constructed based on the temperature hazard probability corresponding to each space unit, and a gas concentration field probability distribution is constructed based on the gas hazard probability corresponding to each space unit. For example, at the prediction time T+5min, space unit A-03 has both a 97.2% temperature hazard probability and a 98.6% gas hazard probability, and can be identified as a high-risk core area; space units A-02 and A-04 have relatively high temperature and gas hazard probabilities, and can be identified as high-risk diffusion areas; space unit A-05 has a relatively low hazard probability, and can be identified as a potential impact area. As the prediction time progresses, each high-hazard probability area gradually expands to the surrounding space units, thus forming the temperature field probability distribution and the gas concentration field probability distribution.

[0067] The method for constructing the temperature field probability distribution and gas concentration field probability distribution described in this application acquires temperature monitoring data and gas concentration monitoring data corresponding to each spatial unit within a building, thereby collecting basic information on the thermal and gas environments of multiple spaces. It constructs temperature field distributions and gas concentration field distributions based on the temperature and gas concentration monitoring data, respectively, to achieve spatial characterization of the building's internal thermal and gas environments. Based on the constructed temperature and gas concentration field distributions, it predicts the evolution of the temperature field and gas concentration field within a preset time window, obtaining prediction results for both, thus achieving temporal dynamic evolution modeling of the environmental field. By combining the temperature field prediction results with preset temperature hazard thresholds and the gas concentration field prediction results with preset gas hazard thresholds, it calculates the temperature hazard probability and gas hazard probability of each spatial unit, achieving a probabilistic and quantitative expression of hazardous states. Based on this, it constructs temperature field probability distributions using temperature hazard probabilities and gas concentration field probability distributions using gas hazard probabilities, thereby significantly improving the refinement, dynamic prediction capability, and reliability of risk assessment in multi-physics field fire risk characterization.

[0068] See also Figure 3 In this embodiment, the method for obtaining the multiphysics joint probability heatmap is as follows: Using the probability of smoke propagation, the probability of temperature hazard, and the probability of gas hazard as inputs, the hazard confidence data corresponding to each spatial unit is determined; Calculate the degree of conflict of evidence among various risk factors based on risk confidence data; Based on the degree of evidence conflict, conflict correction and fusion processing are performed on the hazard credibility data to obtain a comprehensive hazard credibility. The joint probability value corresponding to each spatial unit is determined by the comprehensive risk confidence level, and a joint probability field is constructed based on this. By integrating the joint probability field with the spatial positional relationships between each spatial unit, a multiphysics joint probability heatmap is generated.

[0069] For example, after an abnormal fire event occurs in area A of the underground parking lot of the aforementioned commercial complex, a joint assessment of the danger zone at time T+5 minutes is conducted based on the probability distribution of smoke propagation, the probability distribution of temperature field, and the probability distribution of gas concentration field, and a multi-physics field joint probability heat map is generated.

[0070] First, the smoke propagation probability, temperature hazard probability, and gas hazard probability for each space unit were obtained. Specifically, at prediction time T+5min, the smoke hazard probability for space unit A-03 was 98.0%, the temperature hazard probability was 97.2%, and the gas hazard probability was 98.6%; for space unit A-02, the smoke hazard probability was 82.0%, the temperature hazard probability was 81.5%, and the gas hazard probability was 84.2%; for space unit A-04, the smoke hazard probability was 76.0%, the temperature hazard probability was 73.6%, and the gas hazard probability was 79.1%; and for space unit A-05, the smoke hazard probability was 41.0%, the temperature hazard probability was 26.8%, and the gas hazard probability was 35.8%.

[0071] Subsequently, using the probability of smoke propagation, the probability of temperature hazard, and the probability of gas hazard as inputs, hazard confidence data corresponding to each spatial unit is constructed. For example, for spatial unit A-03, a hazard confidence set {0.980, 0.972, 0.986} can be formed; for spatial unit A-05, a hazard confidence set {0.410, 0.268, 0.358} can be formed.

[0072] Furthermore, the degree of conflict of evidence among various hazard factors was calculated based on the hazard credibility data. For example, in space unit A-03, the probability of smoke propagation, the probability of temperature hazard, and the probability of gas hazard are all at a high level, and the evidence is highly consistent, with a conflict coefficient of 0.021; while in space unit A-05, since the probability of temperature hazard is significantly lower than the probability of smoke hazard and gas hazard, there are certain differences among the evidence, with a conflict coefficient of 0.186.

[0073] Subsequently, conflict correction and fusion processing were performed on the hazard credibility data based on the degree of evidence conflict. For example, an evidence credibility weighted correction mechanism was adopted to reduce the impact of conflicting evidence on the fusion results. After fusion, the comprehensive hazard credibility of spatial unit A-03 was 0.981; after fusion, the comprehensive hazard credibility of spatial unit A-02 was 0.829; after fusion, the comprehensive hazard credibility of spatial unit A-04 was 0.763; and after fusion, the comprehensive hazard credibility of spatial unit A-05 was 0.344.

[0074] After obtaining the overall hazard confidence level, it is used as the joint probability value of the corresponding spatial units, and a joint probability field is constructed. For example, at the prediction time T+5min, the joint probability values ​​of each spatial unit in the underground parking area are as follows: A-03 spatial unit: 0.981; A-02 spatial unit: 0.829; A-04 spatial unit: 0.763; A-05 spatial unit: 0.344; A-06 spatial unit: 0.127; A-07 spatial unit: 0.082. This forms a joint probability field reflecting the spatial distribution characteristics of the hazard level.

[0075] Finally, combining the spatial relationships of each spatial unit in the Building Information Model (BIM), the joint probability field is mapped to the corresponding building plan space. Specifically, areas with a joint probability value greater than 0.90 are marked as extremely high-risk areas, areas with a joint probability value between 0.70 and 0.90 are marked as high-risk areas, areas with a joint probability value between 0.30 and 0.70 are marked as medium-risk areas, and areas with a joint probability value less than 0.30 are marked as low-risk areas. This generates a multiphysics joint probability heatmap. The heatmap shows that spatial unit A-03 is located in the danger center area, spatial units A-02 and A-04 form a danger diffusion zone, while spatial units A-05 and the surrounding spatial units belong to the potentially affected areas, thus visually demonstrating the spatial distribution and evolution trend of the danger zones.

[0076] The method for obtaining the multiphysics joint probability heatmap described in this application determines the hazard credibility data corresponding to each spatial unit by using smoke propagation probability, temperature hazard probability, and gas hazard probability as inputs, thereby achieving a unified quantitative expression of multi-source fire risk information. It characterizes the consistency and uncertainty relationships between different hazard factors by calculating the degree of evidence conflict between hazard credibility data. It obtains a comprehensive hazard credibility by performing conflict correction and fusion processing on the hazard credibility data based on the degree of evidence conflict, achieving collaborative correction and fusion enhancement of multiphysics risk information. It determines the joint probability value corresponding to each spatial unit from the comprehensive hazard credibility and constructs a joint probability field, achieving unified probability modeling of multiple risk factors. Based on this, it generates a multiphysics joint probability heatmap by combining the joint probability field with the spatial positional relationship of each spatial unit, thereby significantly improving the fusion consistency, spatial expression capability, and overall risk visualization and assessment accuracy of multi-source fire risk information.

[0077] Example 4 Based on Example 3, this embodiment further explains the method for determining the fire source location, the method for obtaining the evolution results of the dangerous area, the representation method of building topology information, the generation method of dynamic evacuation plan, and the hierarchical mechanism of early warning information. It is applicable to complex building scenarios that require accurate fire source location and real-time evacuation guidance.

[0078] like Figure 4 As shown, in this embodiment, the method for correcting the fire probability corresponding to each spatial unit is as follows: Based on the multiphysics joint probability heatmap, read the joint probability value corresponding to each spatial unit in the building information model; Spatial constraint correction is performed on the joint probability value using the structural information of each spatial unit as a constraint. By combining the real-time status information of the access control system, the joint probability value after spatial constraint correction is adjusted for connectivity. Visual detection results are incorporated to perform credibility correction on the joint probability value after connectivity correction. Normalization is performed on the corrected joint probability values ​​to obtain the corrected fire probability for each spatial unit.

[0079] For example, in the fire scenario of the underground parking lot of the aforementioned commercial complex, the fire probability corresponding to each spatial unit is further corrected based on the multi-physics joint probability heat map, so as to improve the consistency between the dangerous area identification results and the actual building environment.

[0080] First, the joint probability values ​​corresponding to each spatial unit are read from the multiphysics joint probability heatmap. For example, at the prediction time T+5min, the joint probability values ​​corresponding to spatial units A-03, A-02, A-04, A-05 and A-06 are 0.981, 0.829, 0.763, 0.344 and 0.127, respectively.

[0081] Subsequently, the joint probability values ​​were spatially constrained and corrected using structural information from the Building Information Model (BIM). The BIM showed that A-03 and A-02 were separated only by a standard partition wall, while A-03 and A-06 were separated by a firewall structure with a fire resistance rating of 3 hours. Therefore, a structural attenuation factor of 0.45 was applied to the A-06 spatial unit isolated by the firewall, correcting its joint probability value from 0.127 to 0.057. The A-02 and A-04 spatial units, being directly connected to the fire source area, received only a slight correction factor of 0.95, correcting their joint probability values ​​to 0.788 and 0.725, respectively.

[0082] Furthermore, by combining the real-time status information of the access control system, the joint probability value after spatial constraint correction is adjusted for connectivity. For example, access control records show that the fire door between A-03 and A-05 is in a normally open state, and there are 17 personnel passage records in the last 3 minutes. Therefore, the hazard propagation weight of spatial unit A-05 is increased, and its probability value is corrected from 0.344 to 0.418. Meanwhile, the fire door between A-04 and the adjacent area has been automatically closed and remains locked, so the hazard propagation capability of the adjacent area is suppressed.

[0083] Subsequently, visual inspection results were used to correct the reliability of the joint probability values ​​after connectivity correction. For example, the visible light camera detected significant smoke accumulation in spatial cell A-03, with a flame target recognition confidence level of 96.8%; infrared thermal imaging detected a local hotspot temperature of 138.5℃, thus increasing the reliability weight of spatial cell A-03, and adjusting its probability value from 0.981 to 0.992. Meanwhile, spatial cell A-02 detected slight smoke diffusion but no open flame target, and its probability value was adjusted from 0.788 to 0.812; spatial cell A-05 detected smoke but no high-temperature hotspot, so its probability value was adjusted from 0.418 to 0.392. For spatial cell A-06, visual inspection did not detect any abnormal smoke or heat source, and its probability value was further reduced from 0.057 to 0.041.

[0084] After completing the confidence correction, the probability values ​​corresponding to each spatial unit are normalized to obtain the corrected fire probability. For example: spatial unit A-03: 0.992; spatial unit A-02: 0.812; spatial unit A-04: 0.741; spatial unit A-05: 0.392; spatial unit A-06: 0.041; spatial unit A-07: 0.026. This forms a corrected fire probability distribution that reflects the actual structural characteristics of the building, traffic conditions, and visual observation results.

[0085] The method described in this application for correcting the fire probability corresponding to each spatial unit achieves spatial mapping acquisition of fire probability by reading the joint probability value corresponding to each spatial unit in the building information model based on the multiphysics joint probability heat map; it performs spatial constraint correction on the joint probability value by using the structural information of each spatial unit as a constraint to eliminate the influence of building structure differences on probability distribution; it performs connectivity correction on the joint probability value after spatial constraint correction by combining the real-time status information of the access control system to achieve dynamic constraints on personnel passage and spatial accessibility; it performs credibility correction on the joint probability value after connectivity correction by introducing visual inspection results to improve the multi-source consistency and reliability of fire judgment; and it performs normalization processing on the corrected joint probability value to obtain the corrected fire probability corresponding to each spatial unit, thereby significantly improving the accuracy, robustness and multi-source information fusion consistency of fire spatial probability estimation.

[0086] See also Figure 4 In this embodiment, the method for determining the location of the fire source and generating location confidence is as follows: Candidate fire source areas are determined based on the corrected fire probability of each spatial unit; The initial fire source location is determined based on the probability distribution characteristics of the candidate fire source areas; When multiple local probability extrema exist in the candidate fire source region, the visual detection results of the corresponding region are obtained, and the initial fire source position is corrected accordingly. The location reliability is calculated by combining the fire probability of the fire source location after comprehensive correction, the consistency information of multi-physics field evidence, and the visual detection results. The fire source location results are generated based on the corrected fire source location and location reliability.

[0087] For example, in the fire scenario of the underground parking lot of the aforementioned commercial complex, the location of the fire source is determined and the corresponding location confidence is generated based on the corrected fire probability distribution.

[0088] First, candidate fire source areas are determined based on the corrected fire probability of each spatial unit. For example, after probability correction, the fire probabilities for spatial units A-03, A-02, A-04, and A-05 are 0.992, 0.812, 0.741, and 0.392, respectively. Spatial units with a fire probability greater than 0.70 are defined as candidate fire source areas, thus the set of candidate fire source areas is determined as {A-03, A-02, A-04}.

[0089] Subsequently, the spatial distribution of fire probability within the candidate fire source areas was analyzed, and the initial fire source location was determined based on the probability distribution characteristics. Among them, spatial unit A-03 had the highest fire probability at 0.992, and the probability of surrounding spatial units showed a significant decreasing trend. Therefore, spatial unit A-03 was initially determined as the initial fire source location. Combining the spatial coordinate information in the building information model, the initial fire source location can be represented as the 3rd grid unit in area A of the underground parking lot, with its center coordinates of (28.4m, 16.7m, 0m).

[0090] Furthermore, in some implementation scenarios, multiple local probability extremes may exist in the candidate fire source area. For example, if the fire probability in space unit A-03 is 0.992, while the fire probability in adjacent space unit B-01 reaches 0.965, multiple local probability extreme areas are formed. In this case, the visual detection results of the corresponding areas are obtained for auxiliary judgment. Among them, the visible light image corresponding to space unit A-03 detected an open flame target, with a flame recognition confidence level of 96.8% and a flame area of ​​approximately 3.2m². 2 Infrared thermal imaging detected a maximum temperature of 138.5℃. However, only a high-temperature hotspot was detected in space unit B-01, with a maximum temperature of 81.4℃; no clear flame target was detected. Therefore, space unit A-03 was determined as the final fire source location, and the initial fire source location was corrected and confirmed.

[0091] Subsequently, the fire probability of the corrected fire source location, the consistency information of multiphysics evidence, and the visual detection results were used to calculate the location confidence level. Specifically, the smoke hazard probability for spatial unit A-03 was 98.0%, the temperature hazard probability was 97.2%, and the gas hazard probability was 98.6%, with a consistency coefficient of 0.964 among the three types of evidence. Meanwhile, the flame recognition confidence level in the visual detection results was 96.8%, and the thermal imaging anomaly confidence level was 98.1%. Based on a preset weighted fusion rule, the above information was fused and calculated, resulting in a fire source location confidence level of 97.6%.

[0092] Finally, the fire source location result is generated based on the corrected fire source location and location confidence level. For example, the output result is: the fire source is located in spatial unit A-03 of area A in the underground parking lot, with corresponding spatial coordinates of (28.4m, 16.7m, 0m), a location confidence level of 97.6%, and a location status of "high confidence fire source confirmed". Simultaneously, the fire source location can be highlighted in the 3D scene of the building information model, and the fire source location result can be output to the subsequent hazardous area evolution analysis module.

[0093] The method for determining the location of a fire source and generating location reliability described in this application achieves preliminary screening of potential fire source distribution areas by determining candidate fire source regions based on the corrected fire probability of each spatial unit; it achieves probabilistic positioning of the fire source spatial location by determining the initial fire source location based on the probability distribution characteristics of the candidate fire source regions; it corrects the initial fire source location by introducing visual detection results when multiple local probability extremes exist in the candidate fire source region, achieving disambiguation and optimization of the fire source location under multi-peak interference conditions; it calculates the location reliability by comprehensively considering the fire probability of the corrected fire source location, the consistency information of multi-physics evidence, and the visual detection results, achieving a quantitative assessment of the reliability of the fire source positioning results; and it generates fire source positioning results based on the corrected fire source location and location reliability, thereby significantly improving the accuracy, anti-interference ability, and positioning credibility under multi-source information fusion conditions of fire source positioning.

[0094] See also Figure 4 In this embodiment, the method for obtaining the evolution results of the danger zone is as follows: Determine the initial spatial unit for the spread of fire based on the location of the fire source; The confidence zone of the fire source is determined by combining the location of the fire source and the location confidence level. A fire spread model is constructed based on the connectivity and attribute information between building space units; Based on the fire source confidence zone and fire spread model, the fire spread process within a preset time window is predicted to obtain the fire evolution results of each spatial unit. Determine the probability of danger for each spatial unit based on the fire evolution results; Danger zones are defined based on the danger probability corresponding to each predicted time, and the evolution results of the danger zones are generated.

[0095] For example, in the fire scenario of the underground parking lot of the aforementioned commercial complex, the fire source has been identified as being located in spatial unit A-03, with corresponding spatial coordinates of (28.4m, 16.7m, 0m), and the fire source location confidence level is 97.6%. Based on this, the evolution process of the danger zone within the next 15 minutes is predicted and analyzed.

[0096] First, spatial unit A-03, where the fire source is located, is taken as the starting spatial unit for fire spread. Since the fire source location confidence level is 97.6%, exceeding the preset high confidence threshold of 95%, spatial unit A-03 is directly identified as the fire spread initiation area. Simultaneously, based on the location error range, a fire source confidence region is established within a 5m radius of A-03, forming a fire source confidence region set that includes A-03 and its adjacent boundary areas.

[0097] Subsequently, a fire spread model was constructed based on the spatial topology, building structure information, and combustible material distribution information in the Building Information Modeling (BIM). Specifically, A-03 is directly connected to A-02 and A-04 via a vehicle passageway; A-03 and A-06 are separated by a fire-resistant firewall with a fire resistance rating of 3 hours; and the vehicle density near A-03 is approximately 0.75 vehicles / m². 2 The combustible load density is approximately 820 MJ / m³. 2 The mechanical ventilation system has a wind speed of 3.5 m / s, and the airflow direction is from area A-03 to area A-02. These parameters are used together as input conditions for the fire spread model.

[0098] Furthermore, based on the fire source confidence area and fire spread model, the fire spread process within the next 15 minutes is predicted. For example, at the predicted time T+3 minutes, the fire is mainly concentrated in space cell A-03, with an affected area of ​​approximately 48m². 2 At the predicted time T+5 minutes, the fire began to spread to space units A-02 and A-04, increasing the affected area to 113m². 2 At the predicted time T+10min, space unit A-02 exhibited continuous burning characteristics, with the overall fire-affected area reaching 268m². 2 At the predicted time T+15 minutes, the fire further spread towards area A-05, affecting an area of ​​425m². 2 This allows us to obtain the fire evolution results for each spatial unit.

[0099] Subsequently, the danger probability corresponding to each space unit was calculated based on the fire evolution results. For example, at the prediction time T+10min: the fire danger probability of space unit A-03 was 99.3%; the fire danger probability of space unit A-02 was 92.7%; the fire danger probability of space unit A-04 was 84.6%; the fire danger probability of space unit A-05 was 63.8%; due to the isolation of the firewall, the fire danger probability of space unit A-06 was only 11.5%; and the fire danger probability of space unit A-07 was 6.8%.

[0100] Furthermore, hazardous areas are classified and delineated based on the probability of danger. Specifically, areas with a probability of danger greater than 90% are defined as Level 1 hazardous areas, areas with a probability of danger between 70% and 90% are defined as Level 2 hazardous areas, and areas with a probability of danger between 40% and 70% are defined as Level 3 hazardous areas.

[0101] For example, at the predicted time T+10min: Level 1 hazard area: A-03, A-02; Level 2 hazard area: A-04; Level 3 hazard area: A-05; Safety monitoring area: A-06, A-07 and other peripheral areas.

[0102] As the prediction time progressed to T+15min, the hazard probability of space unit A-05 rose to 81.2%, upgrading it from a Level III hazard area to a Level II hazard area; the hazard probability of space unit A-04 rose to 91.8%, upgrading it to a Level I hazard area, thus forming a dynamic evolution process of the hazard areas.

[0103] Finally, the hazardous area division results corresponding to each prediction time are organized in chronological order to generate hazardous area evolution results. For example, hazardous area evolution sequence data for the next 0-15 minutes is generated and displayed in the 3D scene of the building information model as a dynamic heat map to show the expansion process of hazardous areas, providing a basis for emergency evacuation route planning, fire resource scheduling, and linkage control decisions.

[0104] The method for obtaining the aforementioned hazardous area evolution results in this application achieves spatial anchoring of the initial conditions for fire propagation by determining the initial spatial unit of fire spread starting from the fire source location; by determining the fire source confidence area by comprehensively considering the fire source location and location confidence, it achieves probabilistic expression and spatial expansion modeling of fire source uncertainty; by constructing a fire spread model based on the connectivity and attribute information between building spatial units, it achieves constrained modeling of the fire propagation path by the building spatial topology; by predicting the fire spread process within a preset time window based on the fire source confidence area and the fire spread model, it obtains the fire evolution results of each spatial unit, achieving temporal dynamic simulation of the fire diffusion process; by calculating the hazard probability corresponding to each spatial unit based on the fire evolution results, it achieves quantitative assessment of the risk level; and on this basis, it delineates hazardous areas according to the hazard probability corresponding to each prediction time and generates hazardous area evolution results, thereby significantly improving the continuity, spatial consistency, and dynamic risk characterization capability of fire spread trend prediction.

[0105] See also Figure 4 In this embodiment, the building topology information is represented in the form of a weighted directed graph, where each node corresponds to a spatial unit in the building information model, and the node attributes cover the area, floor height, usage type and rated personnel capacity of the spatial unit. Each directed edge corresponds to a connecting passage between two adjacent spatial units. The types of connecting passages include corridor connections, staircase connections, elevator connections, and evacuation door connections. The attributes of the directed edge include passage type, clear width, maximum passage capacity, normal passage direction, and emergency reverse passage permission indicator. The initial weights of directed edges are assigned according to the following rules: corridor connection edges are weighted based on the reciprocal of the net width of the corridor; stair connection edges are weighted according to the direction of the evacuation target floor. In above-ground buildings, the weight of stair edges pointing towards the ground floor is lower than that of stair edges pointing away from the ground floor. In underground buildings, the weight of stair edges pointing towards the ground floor is also lower than that of stair edges pointing away from the ground floor, in order to reflect the principle of prioritizing evacuation to the nearest safe exit on the ground; elevator connection edges are set to infinite weight in the event of a fire, so as to prohibit the use of elevators for evacuation route planning. In the event of a fire, the weights of each edge in the weighted directed graph are dynamically updated: when the spatial unit corresponding to a certain directed edge enters the danger zone, the weight of that directed edge is adjusted to the level of prohibiting passage; when the fire door is closed, the weight of the corresponding directed edge is corrected according to the type of fire door. When a Class A fire door is closed, the weight of its corresponding directed edge for ordinary evacuees is adjusted to the level of prohibiting passage, and the passable attribute is only retained in the evacuation route planning for firefighters.

[0106] The method for representing and weighting the aforementioned building topology information in this application represents spatial units in the building information model as nodes using a weighted directed graph, and sets node attributes such as area, floor height, usage type, and rated occupancy capacity to achieve multi-dimensional semantic modeling of the building spatial structure. By modeling the connecting channels between adjacent spatial units as directed edges and introducing attributes such as channel type, net width, maximum passage capacity, normal passage direction, and emergency reverse passage permission indicators, it achieves refined expression and computable modeling of building spatial connectivity relationships. The method assigns initial weights to the directed edges based on channel type and spatial constraints, with the reciprocal of the net channel width used as the basic weight for corridor connections. Based on the differences in evacuation target directions of staircase connections, directional weighting is applied. In the elevator connection edges, the weight is set to infinity under fire conditions, thus reflecting the constraint rule that elevators are prohibited in fire scenarios. By dynamically updating the weights of the weighted directed graph edges under fire conditions, when the spatial unit where the passage is located enters the danger zone, the weight of the corresponding edge is adjusted to the level of prohibition of passage. Under the constraint of fire door status, the passage weight is differentiated according to the door type. This ensures that Class A fire doors, when closed, impose a prohibition of passage constraint on ordinary evacuation route planning, while retaining the passage attribute for fire rescue routes. This significantly improves the authenticity, dynamic adaptability, and consistency of fire scenario constraints in the topological structure expression of building evacuation route planning.

[0107] See also Figure 4 In this embodiment, the method for generating the dynamic evacuation plan is as follows: By combining the evolution results of hazardous areas with building topology information, a time-series directed graph of building topology based on time steps is constructed. Based on the danger zone boundaries corresponding to each time step, the passability status of each node and directed edge in the building topology time-series directed graph is dynamically updated. Starting from the current location of each person in the building and ending at each safety exit, solve for the shortest path in the time sequence on the dynamically updated building topology time sequence directed graph; Based on the capacity constraints of each directed edge, the shortest path in time is modified to obtain an evacuation path that satisfies the capacity constraints. Evacuation route plans are generated for individuals or groups of individuals, and the corresponding route node sequence and the estimated arrival time of each node are output.

[0108] For example, in the fire scenario of the underground parking garage of the aforementioned commercial complex, the evolution of the danger zone over the next 15 minutes has been obtained. Specifically, at the predicted time T+10 minutes, spatial units A-03 and A-02 are identified as Level 1 danger zones, A-04 as a Level 2 danger zone, and A-05 as a Level 3 danger zone; the remaining areas are considered safe. Based on this, a dynamic evacuation plan for personnel within the building is generated.

[0109] First, a time-series directed graph of the building topology is constructed based on time steps, integrating the evolution results of hazardous areas with building topology information. The underground parking garage is divided into 18 spatial unit nodes, including parking area nodes A-01 to A-12, stairwell node C-01, safety passage node C-02, and two safety exit nodes E-01 and E-02. A time-series topology graph is established with a time step of 30 seconds, forming a time-series directed graph of the building topology for 30 time layers in the next 15 minutes.

[0110] Subsequently, the traversability status of nodes and directed edges in the graph is dynamically updated based on the danger zone boundaries corresponding to each time step. For example, at T+5 min, the danger probability of spatial unit A-03 reaches 96.4%, and the corresponding node is marked as impassable; the traversability weights of the edges directly connected to A-03 (A-03→A-02) and (A-03→A-04) are increased to their maximum values; at T+10 min, the danger probability of spatial unit A-02 rises to 92.7%, and the corresponding node is further marked as closed; at T+15 min, the danger probability of spatial unit A-04 reaches 91.8%, and the relevant traversable edges are closed simultaneously, thus realizing the dynamic update of the temporal directed graph of the building topology.

[0111] Furthermore, the real-time location of people inside the building is obtained through access control and visual detection systems. For example, it is identified that there are currently 86 people in the underground parking lot, of which: group G1 is located in area A-01 with 35 people; group G2 is located in area A-05 with 28 people; and group G3 is located in area A-08 with 23 people.

[0112] Starting from the current location of each group of people, and ending at the nearest safety exit E-01 or E-02, solve for the shortest path in the time series on the dynamically updated building topology time series directed graph.

[0113] For example, for group G1 located in area A-01, the optimal evacuation route is calculated as: A-01→A-07→C-01→C-02→E-01; the estimated total evacuation time is 148s.

[0114] For group G2 located in area A-05, since area A-04 will enter a high-risk state within the next 5 minutes, the evacuation route should be avoided and the planned evacuation route should be: A-05→A-09→A-10→C-01→E-02; the total evacuation time is expected to be 176 seconds.

[0115] For group G3 located in area A-08, the planned route is: A-08→A-11→C-02→E-02; the total evacuation time is expected to be 122 seconds.

[0116] Subsequently, the feasibility of the time-series shortest path is adjusted based on the capacity constraints of each directed edge. For example, the maximum capacity of stairwell C-01 is 45 people / min, and both paths G1 and G2 pass through node C-01, which is predicted to become congested at time T+3 minutes. Therefore, the 12 people in group G2 are reassigned to the backup path: A-05→A-06→A-11→C-02→E-02.

[0117] After the correction, the instantaneous passenger flow at node C-01 decreased from 58 people / min to 41 people / min, meeting the traffic capacity constraint requirements.

[0118] Finally, evacuation route plans are generated for each group of people, and the corresponding route node sequence and estimated arrival time are output. For example, for group G1, the output is: Path node A-01, estimated arrival time 0s; Path node A-07, estimated arrival time 38s; Path node C-01, estimated arrival time 82s; Path node C-02, estimated arrival time 118s; Path node E-01, estimated arrival time 148s. For groups G2 and G3, corresponding path node sequences and estimated arrival times are also generated and pushed to relevant personnel in real time via mobile terminals, emergency broadcasting systems, and evacuation guidance equipment.

[0119] The method for generating dynamic evacuation schemes described in this application constructs a time-series directed graph of building topology based on time steps by integrating the evolution results of hazardous areas and building topology information, thereby achieving unified modeling of building spatial connectivity and the dynamic evolution process of fire. It dynamically updates the passability status of nodes and directed edges in the time-series directed graph according to the hazardous area boundaries corresponding to each time step, realizing real-time constraint expression of evacuation routes during fire evolution. It solves for the shortest path in the time-series directed graph, starting from the current location of personnel and ending at the safety exit, achieving preliminary planning of optimal evacuation routes for dynamic risk environments. It corrects the feasibility of the shortest path by combining the passability constraints of each directed edge, achieving comprehensive correction of capacity limitations and congestion constraints. Based on this, it generates evacuation route schemes for individuals or groups of individuals, and outputs the path node sequence and the estimated arrival time of each node, thus significantly improving the real-time performance, feasibility, and safety reliability of evacuation decisions and route planning in complex fire scenarios.

[0120] In this embodiment, the early warning information is divided into three levels according to the degree of danger. The early warning level is automatically determined based on the evolution of the fire situation, as follows: The first level of warning is the attention level. Its triggering condition is that the confidence level of the anomaly in the fire perception feature data exceeds the preset low threshold and the location of the fire source has not yet been determined. The content of the warning information at this level includes a description of the spatial location of the abnormal perception area, the degree of deviation of the current physical field monitoring values, and a prompt instruction for personnel to enter the standby state. The target of the warning is the terminal of the building fire protection duty room. The second-level warning is an alarm level. Its triggering conditions are: the location of the fire source has been determined, and the evolution of the danger zone indicates that at least one spatial unit with personnel will enter the danger zone within a preset time window. The warning information at this level includes the spatial unit number and floor information corresponding to the fire source location, the current distribution range of the danger zone, the location of the threatened personnel and the corresponding evacuation route information, as well as a list of fire linkage control instructions. The warning information is simultaneously sent to the building broadcasting system and the fire control room, and forwarded to the local fire and rescue agency through the standard interface of the urban fire alarm network system. The third level of warning is the emergency level. Its triggering condition is: the fire evolution results indicate that the danger zone will cover the main evacuation routes of the building in a short period of time. The warning information of this level is further expanded on the basis of the second level warning information. The new content includes suggestions on the priority access direction for fire rescue, information on the location of hazardous materials stored inside the building, and control instructions for elevators to stop at the first floor.

[0121] The method for generating and outputting the aforementioned early warning information in this application automatically classifies the early warning levels based on the fire situation evolution results, realizing a hierarchical response mechanism from low to high fire risk. The first level of warning is the attention level, triggered when the confidence level of fire perception feature data anomalies exceeds a preset low threshold and the fire source location is not yet determined. It outputs the spatial location of the abnormal perception area, the degree of deviation in physical field monitoring, and personnel standby instructions to provide early warning of fire anomalies. The second level of warning is the alarm level, triggered when the fire source location is determined and the evolution results of the danger zone indicate that there are personnel distribution spatial units that will enter the danger zone within a preset time window. The system outputs information on the location of the fire source, the distribution of dangerous areas, the areas of threatened personnel, and evacuation routes. It also links with the building broadcasting system, the fire control room, and the city's fire alarm network system to achieve cross-system collaborative early warning and emergency response for fire situations. The third level of warning is the emergency level, which is triggered when the dangerous area will cover the main evacuation routes of the building within a short period of time. Based on the alarm level, it further expands the suggestions for priority access directions for fire rescue, the location information of dangerous materials, and the control commands for mandatory elevator stops. This significantly improves the timeliness, accuracy, and multi-linkage response capabilities of fire early warning classification, and enables differentiated emergency decision support under different risk stages.

[0122] Example 5 Based on Example 4, this embodiment further explains the execution method of fire early warning and emergency linkage control and the method for determining the stable convergence of the evolution results of dangerous areas. It is applicable to emergency management scenarios of large buildings with high requirements for the precision of early warning linkage and the convergence of the evolution model.

[0123] like Figure 5 As shown, in this embodiment, the method for executing fire early warning and emergency response linkage control based on early warning information is as follows: The location of the fire source and the extent of the danger zone are analyzed and determined from the early warning information; The building space is divided into dangerous areas, adjacent areas, and safe areas based on the location of the fire source and the extent of the danger zone. Based on the zoning results, the audible and visual alarm system is subjected to hierarchical linkage control to achieve differentiated alarm prompts; Based on the dynamic evacuation plan, evacuation guidance information is generated and distributed to various areas; Fire-prevention partitions are linked and controlled based on the location of the fire source and the extent of the hazardous area to achieve fire-prevention zone isolation. Based on the location of the fire source and the extent of the hazardous area, the smoke exhaust and ventilation systems are linked for control, thereby achieving directional smoke emission and diffusion suppression. Based on the location of the fire source and the range of the danger zone, the elevator system is linked for control, and elevator emergency landing and operation mode switching are executed.

[0124] For example, after a fire broke out in space unit A-03 of the underground parking garage of the aforementioned commercial complex, an early warning message containing the location of the fire source, the extent of the danger zone, and a dynamic evacuation plan was generated. The fire source location result showed that the fire source was located in space unit A-03, with a location confidence level of 97.6%. At the predicted time T+10 minutes, the first-level danger zone included space units A-03 and A-02, the second-level danger zone included space unit A-04, and the third-level danger zone included space unit A-05. Fire warning and emergency response control were implemented based on this early warning message.

[0125] First, the location of the fire source and the extent of the danger zone were analyzed from the early warning information. The fire source was determined to be located in the center of Zone A of the underground parking garage (28.4m, 16.7m, 0m). The danger zone covers spatial units A-02, A-03, A-04, and A-05, with a total danger area of ​​approximately 268m². 2 .

[0126] Subsequently, the building space was divided into zones based on the location of the fire source and the extent of the danger zone. Specifically, spaces A-02 and A-03 were designated as danger zones; spaces A-04, A-05, and A-06, which are directly connected to the danger zones, were designated as adjacent zones; and the remaining areas of the underground parking garage and other floors of the building were designated as safe zones.

[0127] Furthermore, based on the zoning results, the audible and visual alarm system is subject to tiered linkage control. Within hazardous areas, the highest-level alarm mode is activated, with the audible alarm operating at a continuous 95dB frequency and the red strobe light flashing at 2Hz. In adjacent areas, a medium-level alarm mode is activated, with the audible alarm operating at an intermittent 85dB frequency and simultaneously broadcasting evacuation prompts. Within safe areas, only a notification-level broadcast is activated to remind personnel to pay attention to evacuation information and maintain orderly evacuation, thus achieving differentiated alarm notifications.

[0128] Subsequently, evacuation guidance information is generated based on the dynamic evacuation plan. For example, for group G1 located in area A-01, guidance information is generated stating, "Please evacuate via area A-07 and stairwell C-01 to exit E-01"; for group G2 located in area A-05, guidance information is generated stating, "Please evacuate via area A-09 and passage A-10 to exit E-02". This evacuation guidance information is then disseminated in real time to the corresponding areas through the emergency broadcast system, electronic evacuation signs, mobile terminal push notifications, and building information dissemination.

[0129] Furthermore, fire-resistant partitions are linked and controlled based on the location of the fire source and the extent of the hazardous area. For example, the fire-resistant roller shutter door between A-03 and A-04 is automatically closed, as is the fire door between A-02 and A-06; at the same time, the electromagnetic release device of the fire door is activated to keep the fire door closed, thereby forming an independent fire compartment and suppressing the spread of fire to the outer area.

[0130] At the same time, the smoke extraction and ventilation systems are controlled in a coordinated manner based on the location of the fire source and the extent of the hazardous area. For example, the mechanical smoke extraction system corresponding to smoke extraction zone A in the underground parking lot is activated, increasing the smoke extraction air volume to 22,000 m³ / h. 3 / h; shut down the ventilation system around the hazardous area to prevent airflow from fueling the fire's spread; open the top smoke exhaust valves of areas A-02 and A-03, and close the connecting ventilation valves of adjacent areas to achieve directional smoke emission. Simultaneously, dynamically adjust the speed of the smoke exhaust fans based on smoke diffusion prediction results to maintain the smoke concentration in the hazardous area within a controlled range.

[0131] Subsequently, the elevator system was controlled in conjunction with the location of the fire source and the extent of the danger zone. For example, if it was detected that four passenger elevators were operating in the underground parking garage, two of which were located near the underground level, an emergency landing control was immediately implemented, automatically returning all passenger elevators to the ground floor refuge floor and ceasing to respond to external calls; simultaneously, the fire-fighting elevators were switched to fire-fighting operation mode, allowing only authorized operation from the fire control center, thereby preventing people from accidentally entering the danger zone by using the elevators.

[0132] Finally, the system continuously receives the evolution results of hazardous areas and real-time sensor data, and updates the linkage control strategy according to a 30-second dynamic verification cycle. For example, when the prediction results show that the hazard probability of space unit A-05 increases from 63.8% to 81.2%, the system automatically adjusts area A-05 from a level three hazardous area to a level two hazardous area, and simultaneously upgrades the alarm level, adjusts evacuation routes, and optimizes the smoke exhaust control strategy to achieve dynamic fire linkage control.

[0133] The method described in this application for implementing fire early warning and emergency response linkage control based on early warning information achieves structured parameter extraction of the fire situation by analyzing and determining the location of the fire source and the scope of the dangerous area from the early warning information; it achieves hierarchical spatial expression of the fire impact range by dividing the building space into dangerous areas, adjacent areas, and safe areas based on the location of the fire source and the scope of the dangerous area; it achieves differentiated alarm prompts and precise alarms for different risk areas by implementing hierarchical linkage control of the audible and visual alarm system based on the zoning results; it achieves dynamic guidance and real-time updates of personnel evacuation paths by generating evacuation guidance information based on dynamic evacuation plans and distributing it to each area; it achieves automatic isolation of fire compartments and suppression of fire spread by implementing linkage control of fire-resistant partitions based on the location of the fire source and the scope of the dangerous area; it achieves directional emission and diffusion control of smoke by implementing linkage control of smoke exhaust and ventilation systems; and it achieves elevator forced landing and operation mode switching by implementing linkage control of elevator systems, thereby significantly improving the coordination of fire emergency response, linkage response speed, and safety and reliability of multi-system integrated control.

[0134] In this embodiment, the method for determining whether the evolution result of the dangerous region meets the preset stable convergence condition is as follows: During the execution of early warning and linkage control, real-time multi-source sensor data and the evolution results of the danger zone at the corresponding time are obtained according to the preset dynamic verification cycle. Time-series alignment of real-time multi-source sensor data with the evolution results of hazardous areas is performed to establish a spatial unit mapping relationship between predicted and measured states; Based on the spatial unit mapping relationship, consistency error checks are performed on the hazardous area range, temperature field distribution, and smoke propagation boundary, respectively. Based on the consistency error verification results, determine whether the current verification cycle meets the preset stability threshold; If the stability determination is passed for multiple consecutive verification cycles, the evolution result of the dangerous region is determined to meet the preset stability convergence condition; otherwise, the evolution result of the dangerous region is determined to not meet the convergence condition, and the update or recalculation process of the dangerous region evolution model is triggered.

[0135] For example, in the fire scenario of the underground parking lot of the aforementioned commercial complex, the hazard zone evolution model has been continuously running, generating hazard zone evolution results for the next 15 minutes based on the fire source location, smoke propagation probability distribution, temperature field probability distribution, and fire spread results. To ensure that the hazard zone prediction results are consistent with the actual fire development, the hazard zone evolution results are dynamically verified during the early warning and linkage control execution process, and it is determined whether they meet the preset stable convergence conditions.

[0136] First, during the execution of early warning and coordinated control, real-time multi-source sensor data and the corresponding hazardous area evolution results are acquired according to a preset dynamic verification cycle. For example, the dynamic verification cycle is set to 30 seconds. At a certain verification time T1, the real-time smoke concentration, ambient temperature, carbon monoxide concentration, thermal imaging temperature, and visual detection results of each spatial unit in the underground parking lot are acquired. Simultaneously, the prediction results generated by the hazardous area evolution model for that time are acquired, where A-03 and A-02 are predicted to be Level 1 hazardous areas, and A-04 is predicted to be a Level 2 hazardous area.

[0137] Subsequently, the real-time multi-source sensor data and the evolution results of the hazardous area are time-series aligned, and a spatial unit mapping relationship between the predicted state and the measured state is established. For example, by unifying the timestamp and the spatial numbering system in the building information model, the predicted spatial units A-03, A-02, and A-04 are mapped to their corresponding measured spatial units, forming a one-to-one correspondence between the predicted state and the measured state.

[0138] Furthermore, based on the spatial unit mapping relationship, consistency error checks are performed on the range of dangerous areas, temperature field distribution, and smoke propagation boundaries.

[0139] For the verification of the danger zone range, the model predicts that the coverage area of ​​the Level 1 danger zone is 275m. 2 The actual danger zone area, calculated from real-time sensor data, is 261m². 2 The error in the area of ​​the danger zone is 5.36%.

[0140] For the temperature field distribution verification, the model predicted the temperature of space cell A-03 to be 112.6℃, while the real-time measured temperature was 108.9℃; the model predicted the temperature of space cell A-02 to be 86.4℃, while the real-time measured temperature was 83.7℃; the calculated average relative error of the temperature field was 3.4%.

[0141] For the smoke propagation boundary verification, the model predicted that the smoke influence range had extended to the boundary of spatial cell A-05, while real-time smoke monitoring results showed that the average offset distance between the smoke propagation boundary and the predicted boundary was 1.8m. Since the preset boundary offset threshold is 3m, the current smoke propagation boundary verification results meet the requirements.

[0142] Subsequently, based on the consistency error verification results, it is determined whether the current verification cycle meets the preset stability thresholds. For example, the preset thresholds are: 10% for the area error of the hazardous area; 8% for the average relative error of the temperature field; and 3m for the smoke propagation boundary offset.

[0143] In the current verification period, the area error of the dangerous area is 5.36%, the average relative error of the temperature field is 3.4%, and the offset distance of the smoke propagation boundary is 1.8m, all of which are lower than the corresponding thresholds. Therefore, the current verification period is determined to pass the stability test.

[0144] Furthermore, multiple dynamic verification cycles were executed consecutively. For example, in the subsequent four consecutive verification cycles, the area error of the hazardous area was 4.8%, 5.1%, 4.5%, and 5.3%, respectively; the average relative error of the temperature field was 3.2%, 3.8%, 3.5%, and 3.7%, respectively; and the offset distance of the smoke propagation boundary was 1.6m, 1.9m, 1.5m, and 1.7m, respectively, all of which continuously met the stability threshold requirements.

[0145] Therefore, after five consecutive verification cycles have passed the stability determination, the current evolution result of the dangerous area is determined to meet the preset stable convergence condition, and the current dangerous area evolution model operation status is maintained.

[0146] Conversely, in another implementation scenario, if a certain verification finds that the predicted dangerous area area is 420m² 2 The actual danger zone area reached 560m². 2 If the corresponding error reaches 33.3%, or the smoke propagation boundary offset distance reaches 8.4m, the current verification cycle is determined to have failed the stability test. When multiple consecutive verification cycles have errors exceeding the threshold, the evolution results of the hazardous area are deemed not to meet the convergence conditions, and the hazardous area evolution model update process is automatically triggered. For example, the latest multi-source sensor data is reloaded, fire spread parameters are updated, the smoke diffusion model and temperature field prediction model are corrected, and the evolution results of the hazardous area are recalculated.

[0147] The method for determining whether the evolution results of the aforementioned hazardous area meet the preset stability convergence conditions in this application achieves continuous monitoring of the fire situation evolution process by acquiring real-time multi-source sensor data and the corresponding hazardous area evolution results at the current moment according to a preset dynamic verification cycle during the early warning and linkage control execution process; by temporally aligning the real-time multi-source sensor data and the hazardous area evolution results, establishing a spatial unit mapping relationship between the predicted state and the measured state, and achieving a unified association expression between multi-source sensing data and model output results; by performing consistency error verification on the hazardous area range, temperature field distribution, and smoke propagation boundary based on the spatial unit mapping relationship, achieving a multi-dimensional consistency assessment of the multi-physics field evolution results; by determining whether the current verification cycle meets the preset stability threshold based on the consistency error verification results, achieving a periodic determination of the convergence of the evolution results; and on this basis, when the stability determination is met for multiple consecutive verification cycles, it is determined that the hazardous area evolution results meet the preset stability convergence conditions; otherwise, it is determined that the convergence conditions are not met and the hazardous area evolution model update or recalculation process is triggered, thereby significantly improving the accuracy of the stability determination of the hazardous area evolution results, the adaptive update capability of the model, and the reliability of fire situation prediction.

[0148] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A smart fire risk assessment and early warning method based on multi-source sensor data fusion, characterized in that, Includes the following steps: Acquire multi-source sensor data within the building, and perform standardization, time synchronization, and spatial correlation fusion processing on the multi-source sensor data to obtain a spatiotemporally aligned sensor dataset; Fire anomaly identification is performed based on spatiotemporal aligned sensor datasets, and fire perception feature data representing fire status are extracted; Based on fire perception feature data, the probability distributions of smoke propagation, temperature field, and gas concentration field are constructed respectively, and the probability distributions are jointly fused to obtain a multi-physics field joint probability heat map. The multiphysics joint probability heatmap is mapped to the corresponding spatial unit of the building information model, and the fire probability corresponding to each spatial unit is corrected by combining building geometric constraint information, access control status information and visual detection information. Based on the correction results, the fire source location is determined and the corresponding location confidence is generated. Fire spread prediction is performed based on the location of the fire source and the location reliability, and the evolution results of the danger zone are obtained; Based on the evolution results of hazardous areas, building topology information, and personnel distribution information, evacuation routes are planned to generate dynamic evacuation plans. Early warning information is generated based on the location of the fire source, the evolution of the danger zone, and the dynamic evacuation plan, and fire early warning and emergency response control are executed based on the early warning information.

2. The intelligent fire protection risk assessment and early warning method according to claim 1, characterized in that, Methods for constructing the smoke propagation probability distribution include: Obtain the smoke concentration value, smoke concentration change gradient, and building ventilation system operation parameters for each spatial unit; A smoke diffusion model was constructed based on smoke concentration values, smoke concentration variation gradients, and building ventilation system operating parameters. The smoke concentration distribution in each spatial unit within a preset time window is predicted using a smoke diffusion model. Based on the smoke diffusion model and combined with preset parameter perturbation rules, multiple simulation calculations are performed to generate a smoke concentration sample set for each spatial unit. Based on the smoke concentration sample set and the preset visibility hazard threshold, the smoke propagation probability of each spatial unit at each predicted time is calculated. Based on the probability of smoke hazard state corresponding to each spatial unit at each prediction time, a smoke propagation probability distribution is constructed.

3. The intelligent fire protection risk assessment and early warning method according to claim 2, characterized in that, The methods for constructing the probability distributions of the temperature field and the gas concentration field are as follows: Acquire temperature monitoring data and gas concentration monitoring data for each spatial unit within the building; The temperature field distribution of the building space is constructed based on temperature monitoring data, and the gas concentration field distribution of the building space is constructed based on gas concentration monitoring data. Based on the temperature field distribution and gas concentration field distribution, the evolution process of the temperature field and the evolution process of the gas concentration field within a preset time window are predicted respectively, and the prediction results of the temperature field and the gas concentration field are obtained. The temperature hazard probability of each space unit is calculated based on the temperature field prediction results and the preset temperature hazard threshold, and the gas hazard probability of each space unit is calculated based on the gas concentration field prediction results and the preset gas hazard threshold. A temperature field probability distribution is constructed based on the temperature hazard probability, and a gas concentration field probability distribution is constructed based on the gas hazard probability.

4. The intelligent fire protection risk assessment and early warning method according to claim 1, characterized in that, The method for obtaining the multiphysics joint probability heatmap is as follows: Based on the probability of smoke propagation, the probability of temperature hazard, and the probability of gas hazard, the hazard confidence data corresponding to each space unit is determined. Calculate the degree of conflict of evidence among various risk factors based on risk confidence data; Based on the degree of conflict of evidence, conflict correction and fusion processing are performed on the hazard credibility data to obtain the comprehensive hazard credibility. The joint probability value corresponding to each spatial unit is determined based on the comprehensive risk confidence level, and a joint probability field is constructed. A multiphysics joint probability heatmap is generated based on the joint probability field and the spatial positional relationships between each spatial unit.

5. The intelligent fire protection risk assessment and early warning method according to claim 1, characterized in that, The method for determining the location of the fire source and generating location confidence is as follows: Candidate fire source areas are determined based on the corrected fire probability of each spatial unit; The initial fire source location is determined based on the probability distribution characteristics of the candidate fire source areas; When there are multiple local probability extremes in the candidate fire source area, the visual detection results of the corresponding area are obtained, and the initial fire source position is corrected based on the visual detection results. The location reliability is calculated based on the fire probability corresponding to the corrected fire source location, the consistency information of multi-physics evidence, and the visual detection results. The fire source location result is generated based on the corrected fire source location and the location reliability.

6. The intelligent fire protection risk assessment and early warning method according to claim 1, characterized in that, The method for obtaining the evolution results of the danger zone is as follows: Determine the initial spatial unit for the spread of fire based on the location of the fire source; Determine the confidence zone of the fire source based on its location and location confidence level; A fire spread model is constructed based on the connectivity and attribute information between building space units; Based on the fire source confidence zone and fire spread model, the fire spread process within a preset time window is predicted, and the fire evolution results of each spatial unit are obtained. Determine the probability of danger for each spatial unit based on the fire evolution results; The danger zones are determined based on the danger probabilities corresponding to each prediction time, and the evolution results of the danger zones are generated.

7. The intelligent fire protection risk assessment and early warning method according to claim 1, characterized in that, The method for generating dynamic evacuation plans is as follows: Based on the evolution results of hazardous areas and building topology information, a time-series directed graph of building topology based on time steps is constructed. Based on the danger zone boundaries corresponding to each time step, the traversability status of nodes and edges in the building topology time-series directed graph is dynamically updated. Starting from the current location of each person in the building and ending at each safety exit, the shortest path in time is solved on the dynamically updated directed graph of the building topology. Based on the capacity constraints of each directed edge, the shortest path in time sequence is modified for path feasibility to obtain evacuation paths that satisfy the capacity constraints. Generate evacuation route plans based on individuals or groups of individuals, and output the corresponding route node sequence and the estimated arrival time of each node.

8. The intelligent fire risk assessment and early warning method according to claim 1, characterized in that, The method for implementing fire early warning and emergency response linkage control based on early warning information is as follows: Based on the early warning information, the location of the fire source and the extent of the danger zone are determined; The building space is divided into dangerous areas, adjacent areas, and safe areas based on the location of the fire source and the extent of the danger zone. Based on the zoning results, the audible and visual alarm system is controlled in a hierarchical manner to achieve differentiated alarm prompts; Evacuation guidance information is generated based on the dynamic evacuation plan, and the evacuation guidance information is output to each area; Fire compartmentation facilities are linked and controlled based on the location of the fire source and the extent of the hazardous area in order to achieve fire compartmentation and isolation. The smoke exhaust and ventilation systems are linked and controlled based on the location of the fire source and the range of the hazardous area in order to achieve directional emission and diffusion suppression of smoke. The elevator system is linked and controlled based on the location of the fire source and the range of the danger zone to perform emergency elevator landing and switch operating modes.

9. The intelligent fire risk assessment and early warning method according to any one of claims 1-8, characterized in that, The intelligent fire risk assessment and early warning method also includes: During the execution of fire early warning and emergency response control, the evolution results of the dangerous area are dynamically verified based on real-time multi-source sensor data to determine whether the evolution results of the dangerous area meet the preset stable convergence conditions. If not satisfied, the fire perception feature data, multi-physics joint probability heat map, fire source location and location confidence are updated based on real-time multi-source sensor data, and the fire spread prediction and evacuation route planning are re-executed to update the evolution results of the danger zone and the dynamic evacuation plan. If the conditions are met, the final dynamic evacuation plan and early warning information will be output.

10. The intelligent fire protection risk assessment and early warning method according to claim 9, characterized in that, The method for determining whether the evolution results of the dangerous region meet the preset stable convergence conditions is as follows: During the early warning and linkage control process, real-time multi-source sensor data and the corresponding danger zone evolution results are obtained according to the preset dynamic verification cycle. The real-time multi-source sensor data and the evolution results of the dangerous area are time-series aligned, and a spatial unit mapping relationship between the predicted state and the measured state is established. Based on the spatial unit mapping relationship, the consistency error of the hazardous area range, temperature field distribution and smoke propagation boundary is verified; Determine whether the current verification cycle meets the preset stability threshold based on the consistency error verification results; When the stability requirements are met for multiple consecutive verification cycles, the evolution result of the dangerous area is determined to meet the preset stable convergence condition. Otherwise, the evolution result of the dangerous region is determined to have failed to meet the convergence condition, and the dangerous region evolution model update or recalculation process is triggered.

Images