Regional fire risk analysis method and system based on multi-modal AI analysis

Abstract

The invention relates to the technical field of regional fire-fighting analysis, in particular to a regional fire-fighting risk analysis method and system based on multi-modal AI analysis. The method comprises the following steps: collecting a multi-modal fire-fighting related sensing source in an area, carrying out cross-modal fusion processing and abnormal event detection, and identifying a fire-fighting risk mutation point; carrying out risk propagation trend evolution on the fire risk mutation points, and carrying out fire source high-dimensional traceability analysis to obtain regional fire risk traceability points; performing risk evolution time estimation on the regional fire-fighting risk tracing point, and then performing fire risk situation evolution prediction to obtain a fire-fighting simulation situation map; and performing dynamic risk level change tracking according to the fire-fighting simulation situation map, and performing comprehensive emergency demand adjustment to obtain comprehensive risk intervention decision parameters. Through refined resource matching, the utilization rate of fire-fighting resources is improved, and the regional fire prevention and control quality and safety efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the field of regional fire analysis technology, and in particular to a regional fire risk analysis method and system based on multimodal AI analysis. Background Art

[0002] With the accelerating pace of urbanization and rapid socioeconomic development, the density of various types of buildings continues to rise, posing increasingly severe challenges to public safety. Fires are particularly frequent in densely populated areas such as urban residential areas, commercial complexes, and industrial parks, threatening not only the safety of people's lives and property but also placing enormous pressure on public resources. Against this backdrop, improving the intelligence of regional fire management and enabling accurate identification and real-time response to fire risks have become critical issues that urgently need to be addressed in the fire management field.

[0003] Currently, regional fire management primarily relies on traditional fire alarm systems, video surveillance, manual patrols, and regular inspections. These methods can detect and respond to fires to a certain extent, but because they often rely on a single data source, such as smoke alarm signals or camera images, they have significant limitations in complex environments and scenarios involving the fusion and judgment of multi-source information. This makes it difficult to fully perceive, accurately identify, and dynamically predict potential fire risks. Furthermore, traditional systems generally lack the ability to deeply analyze historical data, making it difficult to uncover the distribution patterns of potential fire hazards and making rapid decision-making support for sudden fires. Summary of the Invention

[0004] In order to solve the above technical problems, the present invention proposes a regional fire risk analysis method and system based on multimodal AI analysis to solve at least one of the above technical problems.

[0005] To achieve the above objectives, the present invention provides a regional fire risk analysis method based on multimodal AI analysis, comprising the following steps: Step S1: Collect multimodal fire-related sensor sources in the area, perform cross-modal fusion processing and abnormal event detection, and identify fire risk mutation points; Step S2: Evolve the risk propagation trend of the fire risk mutation point and conduct high-dimensional fire source tracing analysis to obtain the regional fire risk tracing point; Step S3: Estimate the risk evolution time of the regional fire risk traceability point, and then predict the evolution of the fire risk situation to obtain a fire simulation situation map; Step S4: Tracking the dynamic risk level changes based on the fire simulation situation map, and making comprehensive emergency demand adjustments to obtain comprehensive risk intervention decision parameters; Step S5: Obtain the time series data stream of the available status of emergency resources, and perform optimal resource allocation calculation and regional immediate fire intervention decision-making on the comprehensive risk intervention decision parameters to build a dynamic fire risk intervention strategy; Step S6: Calculate the secondary fire risk probability based on the dynamic fire risk intervention strategy, perform coordinated scheduling optimization across the entire region, and build an intelligent fire decision-making and control model.

[0006] In this specification, a regional fire risk analysis system using multimodal AI analysis is provided, which is used to perform the regional fire risk analysis method using multimodal AI analysis as described above, including: The risk mutation detection module is used to collect multimodal fire-related sensor sources in the area, perform cross-modal fusion processing and abnormal event detection, and identify fire risk mutation points; The risk tracing module is used to analyze the risk propagation trend evolution of fire risk mutation points and conduct high-dimensional fire source tracing analysis to obtain regional fire risk tracing points; The risk situation prediction module is used to estimate the risk evolution time of regional fire risk traceability points, and then predict the evolution of fire risk situation to obtain a fire simulation situation map; The emergency demand decision module is used to track the dynamic risk level changes based on the fire simulation situation map and make comprehensive emergency demand adjustments to obtain comprehensive risk intervention decision parameters; The optimal resource allocation module is used to obtain the time series data stream of the available status of emergency resources, and to perform optimal resource allocation calculations and regional immediate fire intervention decisions based on comprehensive risk intervention decision parameters, thereby building a dynamic fire risk intervention strategy; The full-area scheduling module is used to calculate the secondary fire risk probability based on the dynamic fire risk intervention strategy, perform full-area collaborative scheduling optimization, and build an intelligent fire decision-making and control model.

[0007] The beneficial effects of the present invention include: By collecting and integrating multimodal sensing sources (such as video surveillance, infrared thermal imaging, smoke sensors, environmental monitoring equipment, and social media data), comprehensive awareness of fire risk factors is achieved, avoiding information silos arising from a single sensor source. The integration of multimodal data, including images, sounds, text, and numerical values, effectively improves the accuracy of abnormal event detection and reduces missed and false alarm rates. It can detect and identify fire risk mutation points, such as fire, smoke, and toxic gas leaks, in real time, enabling early fire warning. By modeling the evolutionary trends of mutation points, the spatial and temporal spread of fire risk can be understood, providing a basis for the proactive deployment of firefighting resources. High-dimensional feature tracing and analysis techniques (such as multidimensional features such as time, location, temperature, and airflow direction) can accurately locate fire sources and triggering factors, assisting in investigation and prevention. A clear understanding of the origin of fires and their impact on their spread facilitates accurate decision-making and timely control of fire development. By estimating the evolutionary time windows of different risk sources, valuable time is gained for subsequent intervention and response. Construct a firefighting simulation situation map to visualize the spatial distribution and temporal evolution of fire risks, enabling managers to intuitively understand the overall firefighting situation. Use the simulation map to guide personnel deployment and routing, achieving scientific and visual emergency dispatch. Track risk level trends across different regions and time periods to achieve dynamic risk management. Dynamically adjust the allocation of firefighting forces, equipment, and rescue resources based on risk level changes to avoid resource waste or imbalanced distribution. Optimize emergency resource allocation and routing to improve fire response efficiency and reduce response time. Integrate real-time resource status stream data to fine-tune resource matching for intervention strategies, improving resource utilization. Make real-time decisions on firefighting intervention plans as the fire develops, ensuring targeted deployment and precise response, avoiding over- or under-response. Build an intervention strategy system with real-time update and feedback capabilities to support firefighting intervention decision-making within large, complex areas. Assess the probability of secondary risks (such as cascading fires, power outages, and traffic disruptions) based on the current fire development situation and intervention strategies. Optimize dispatch mechanisms through cross-regional and cross-unit collaboration to enhance overall risk mitigation and prevent isolated, isolated local responses. Ultimately, a firefighting intelligent decision-making and control model that is adaptive, self-updating, and iteratively optimized will be formed, comprehensively improving the level of intelligent fire prevention and control in the region. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 A schematic flow chart of the steps of a regional fire risk analysis method based on multimodal AI analysis according to the present invention; Figure 2 Detailed implementation flow chart of step S1; Figure 3 Detailed implementation flow chart of step S2; Figure 4 Schematic diagram of the detailed implementation steps of step S3. DETAILED DESCRIPTION

[0009] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0010] This application provides a method and system for regional fire risk analysis using multimodal AI analysis. The execution entities of this method and system include, but are not limited to, the following: mechanical equipment, data processing platforms, cloud server nodes, network upload devices, etc., which can be considered general computing nodes of this application. The data processing platform includes, but is not limited to, at least one of an audio and image management system, an information management system, and a cloud data management system.

[0011] See also Figures 1 to 4 The present invention provides a regional fire risk analysis method based on multimodal AI analysis, which includes the following steps: Step S1: Collect multimodal fire-related sensor sources in the area, perform cross-modal fusion processing and abnormal event detection, and identify fire risk mutation points; Step S2: Evolve the risk propagation trend of the fire risk mutation point and conduct high-dimensional fire source tracing analysis to obtain the regional fire risk tracing point; Step S3: Estimate the risk evolution time of the regional fire risk traceability point, and then predict the evolution of the fire risk situation to obtain a fire simulation situation map; Step S4: Tracking the dynamic risk level changes based on the fire simulation situation map, and making comprehensive emergency demand adjustments to obtain comprehensive risk intervention decision parameters; Step S5: Obtain the time series data stream of the available status of emergency resources, and perform optimal resource allocation calculation and regional immediate fire intervention decision-making on the comprehensive risk intervention decision parameters to build a dynamic fire risk intervention strategy; Step S6: Calculate the secondary fire risk probability based on the dynamic fire risk intervention strategy, perform coordinated scheduling optimization across the entire region, and build an intelligent fire decision-making and control model.

[0012] In the embodiment of the present invention, see Figure 1 , is a schematic flow chart of the steps of a regional fire risk analysis method using multimodal AI analysis according to the present invention. In this example, the steps of the regional fire risk analysis method using multimodal AI analysis include: Step S1: Collect multimodal fire-related sensor sources in the area, perform cross-modal fusion processing and abnormal event detection, and identify fire risk mutation points; In this example, appropriate sensors and cameras are selected. Smoke sensors, temperature sensors, visible light cameras, and infrared thermal imagers are deployed to ensure coverage of key areas. Within a 1,000-square-meter area, 10 smoke sensors and 5 temperature sensors, as well as 2 visible light cameras and 1 infrared thermal imager, are deployed. The data collection frequency is set. Assume that the smoke and temperature sensors collect data once per second, while the visible light video and infrared thermal imager are recorded at 30 frames per second. This ensures real-time and complete data. The collected data is transmitted to the central processing unit in real time. To ensure data security, an encrypted transmission protocol is used, and the data is stored on a cloud server for subsequent analysis. The operating status of each sensor and the amount of data collected are recorded. Assume that in one hour, the smoke sensor collects 600 data points and the temperature sensor collects 3,600 data points. Ensuring data integrity and reliability is crucial. Key information is extracted from the collected data. Feature extraction is performed on smoke concentration, temperature, visible light video, and infrared thermal imager videos. The smoke concentration and temperature values ​​at each time point are extracted, along with motion features (such as the number and speed of moving objects) extracted from the video. The extracted features are integrated into a multidimensional parameter matrix. Assume that the feature vector for each time point contains smoke concentration, temperature, visible light characteristics, and thermal imaging characteristics, for a total of four dimensions. Data normalization ensures comparability between different modalities. Use weighted averaging or principal component analysis (PCA) to fuse the data from different modalities to increase the overall information content. Assume that after fusion, the resulting parameter matrix contains rich multimodal information at each time point. Based on historical data and normal operating standards, set anomaly detection thresholds. For example, if the normal smoke concentration is 100 ppm and the normal temperature range is 70°C, then exceeding these values ​​is considered an anomaly. Select an appropriate anomaly detection algorithm, such as the Z-score, isolation forest, or a machine learning-based anomaly detection model. Apply these algorithms to analyze the fused multimodal data and identify time points that exceed the set thresholds. At a certain time point, if the smoke concentration reaches 200 ppm and the temperature reaches 95°C, this is marked as an anomaly. Record identified anomalies, including timestamps, corresponding smoke concentration, temperature, and other information. Generate an abnormal event detection report, describing the detected risk mutation points and their possible impacts, and provide a basis for subsequent emergency response.

[0013] Step S2: Evolve the risk propagation trend of the fire risk mutation point and conduct high-dimensional fire source tracing analysis to obtain the regional fire risk tracing point; In this example, after data preparation, time series analysis was used to analyze the evolution of risk propagation trends at fire risk mutation points. The autoregressive integrated moving average (ARIMA) model was selected for modeling, with parameters set to (p=2, d=1, q=2), where p is the number of autoregressive terms, d is the number of differencing, and q is the number of moving average terms. The data was trained, validated, and tested, and the model was updated weekly using a rolling window technique. By analyzing historical data, peak periods and trends in risk propagation were identified. For example, certain weather conditions (such as high temperature and low humidity) significantly increase fire risk. The generated trend charts display the changing risk and help decision makers identify potential fire peaks. After identifying risk propagation trends, high-dimensional fire source tracing analysis was performed. This process requires correlation analysis of multidimensional fire source data (such as environmental factors, fire occurrence time and location, etc.) with historical fire events. Experimental parameters were set, such as the analysis window covering all fire events that occurred within the past three months. Machine learning algorithms, such as random forests or support vector machines (SVMs), were used to model the relationships between various factors and fire occurrence. Through feature importance analysis, the factors that have the greatest impact on the occurrence of fire are identified. It is found that temperature and smoke concentration are the key factors affecting the source of fire. After completing the fire source analysis, the regional fire risk traceability points are determined. Through cluster analysis (such as K-means), similar fire source data are grouped to find risk points with the same characteristics. The experimental parameters are set, and the number of clusters is 3, representing high-risk, medium-risk and low-risk areas respectively. The clustering results are verified, and the silhouette coefficient is used to evaluate the effectiveness of the clustering to ensure that the identified risk traceability points are significant and reasonable. Finally, a detailed regional fire risk traceability point report is generated, which includes the characteristics of each traceability point, the influencing factors and the corresponding fire risk level.

[0014] Step S3: Estimate the risk evolution time of the regional fire risk traceability point, and then predict the evolution of the fire risk situation to obtain a fire simulation situation map; In this example, before predicting the evolution of fire risk, it is first necessary to estimate the risk evolution time for the regional fire risk origin points. This step aims to assess the time required from activation of the risk origin point to reaching a specific risk level. Real-time monitoring data, including environmental factors such as temperature, humidity, and smoke concentration, is collected and combined with historical fire data to set experimental parameters. For example, fire events from the past six months may be analyzed. Data should be recorded on an hourly basis to ensure the continuity and accuracy of the time series. A dynamic time warping (DTW) algorithm is used to analyze the risk evolution over different time periods to determine the rate of risk change under different conditions. A baseline risk level (smoke concentration reaching 150 mg / m³) is set, and the time required to reach this level is calculated. By comparing the evolution time under different environmental conditions, the main factors affecting the risk evolution time are identified, such as the faster risk evolution under high temperature and low humidity conditions. After completing the risk evolution time estimation, the fire risk evolution prediction model is constructed. Experimental parameters are set, such as selecting a prediction time range of the next 48 hours and using time series analysis techniques. The model is constructed using an autoregressive moving average (ARIMA) model or a long short-term memory (LSTM) network. Model training is performed by dividing historical data into training and test sets. During training, model parameters are set, such as (p, d, q) for ARIMA to (2, 1, 2). After training, the test set is used to evaluate model accuracy. Model fitting yields an estimate of future fire risk and generates a forecast, including the predicted risk level, the likely time of fire occurrence, and its intensity. This forecast provides important support for fire department warning and response. Based on the output of the forecast model, a fire simulation situation map is generated. First, the predicted risk level is combined with a regional map and spatially visualized using GIS (Geographic Information System) technology. Experimental parameters are set, such as a map resolution of 500 meters, to clearly display the risk distribution in different areas. Different risk levels (low, medium, and high) are indicated on the map using color coding. High-risk areas are indicated in red, medium-risk areas in yellow, and low-risk areas in green. The simulation situation map is dynamically updated with real-time monitoring data to ensure real-time information. Furthermore, the simulation situation map should include the predicted location of the fire source, possible danger zones, and the distribution of emergency resources around them, facilitating rapid decision-making by the command center. Ultimately, the resulting fire simulation situation map will serve as a decision-making support tool, helping fire departments effectively allocate and dispatch resources in the event of a sudden fire. After the fire simulation situation map is completed, the results are analyzed to identify potential high-risk areas and time periods. The accuracy and reliability of the prediction model are evaluated by comparing historical fire data with current prediction results. The risk assessment report generated based on the simulation situation map will provide a scientific basis for fire prevention and emergency response.If the forecast shows that the risk level in a certain area will increase in the next 24 hours, the fire department can deploy emergency resources in advance and strengthen monitoring and inspections in the area.

[0015] Step S4: Tracking the dynamic risk level changes based on the fire simulation situation map, and making comprehensive emergency demand adjustments to obtain comprehensive risk intervention decision parameters; In this embodiment, based on the generated fire simulation situation map, the first step is to track dynamic risk level changes. This process aims to monitor and assess changes in fire risk levels within the area in real time. Experimental parameters are set, such as a monitoring frequency of every 30 minutes, to ensure timely updates of risk information. A dynamic risk assessment model is established using a real-time data monitoring system, combining sensor data (such as temperature, humidity, and smoke concentration) with historical fire data. Machine learning algorithms (such as random forests or support vector machines) are used to perform a weighted analysis of different risk factors and dynamically adjust the risk score. Risk level thresholds are set, for example, categorizing risk levels into low (0-30), medium (31-70), and high (71-100). By comparing real-time monitoring data with the output of the simulation situation map, risk level trends are tracked and a dynamic risk level report is generated. This report reflects changes in fire risk within the area in real time, providing data support for subsequent adjustments to emergency response needs. After dynamic risk level tracking is completed, a comprehensive emergency response needs analysis is conducted. Based on the current risk level and historical emergency response data, the required emergency resources and response measures for different risk levels are assessed. Set experimental parameters, such as analyzing the emergency response data of the past 12 fire incidents, considering factors such as the number of fire trucks dispatched, the number of firefighters, and the amount of fire extinguishing agents used. Use a demand forecasting model combined with multiple regression analysis to quantify emergency demand under different risk levels. Set the emergency demand under high risk level to x units of fire resources, medium risk level to y units, and low risk level to z units. By analyzing historical data, establish a relationship model between emergency demand and risk level to ensure the accuracy and reliability of the forecast. Based on the results of the comprehensive emergency demand analysis, dynamically adjust the emergency demand. When the risk level changes, the configuration of emergency resources needs to be adjusted in time. Set experimental parameters, for example, when the risk level increases from medium to high risk, the resource allocation should increase by 30%. Use optimization algorithms (such as linear programming or genetic algorithms) to calculate the optimal configuration of emergency resources. Establish an objective function to minimize the cost of resource use while ensuring the effectiveness of the emergency response. Set the goals as: ;in, are the cost coefficients of different resources, The allocation amount of each type of resource. Through optimization calculation, a new emergency resource allocation plan is generated to ensure a rapid response in high-risk situations. The allocation amount of each type of resource. Through optimization calculation, a new emergency resource allocation plan is generated to ensure a rapid response in high-risk situations. Based on the dynamically adjusted emergency needs, comprehensive risk intervention decision parameters are generated. These parameters include specific resource allocation plans, response time, dispatched personnel and equipment types, etc. Through summary analysis, a comprehensive risk intervention decision report is formed to provide scientific decision-making support for the fire department. The report lists in detail the specific allocation plans for each type of resource and the expected response time. When the risk level is high, it is recommended to dispatch 10 fire trucks, 50 firefighters, and use 200 liters of firefighting foam. In low-risk situations, it can be adjusted to 5 fire trucks and 20 firefighters.

[0016] Step S5: Obtain the time series data stream of the available status of emergency resources, and perform optimal resource allocation calculation and regional immediate fire intervention decision-making on the comprehensive risk intervention decision parameters to build a dynamic fire risk intervention strategy; In this embodiment, the availability of emergency resources is monitored in real time to obtain a time-series data stream. This step relies on sensors and monitoring equipment deployed at each fire station and key areas to collect real-time data on emergency resource status, including the availability of fire trucks, the number of firefighting equipment, and the presence of firefighters. Experimental parameters are set, such as a monitoring frequency of every five minutes, to ensure data timeliness and accuracy. A data acquisition system transmits various resource status information to a central database in real time. This data includes resource type, current status (e.g., available, busy, down), and location. During data processing, data cleaning techniques are used to remove outliers and incomplete records to ensure the accuracy and consistency of the information in the database. This creates a complete time-series data set for emergency resource status, providing foundational data for subsequent resource optimization and allocation analysis. After obtaining the time-series data stream for emergency resource availability, comprehensive risk intervention decision parameters are analyzed to assess emergency demand at different risk levels. Based on the dynamic risk level report generated in the previous step and combined with real-time resource status, experimental parameters are set and emergency response data from 12 past fire incidents is analyzed. A demand forecasting model is used, combined with historical data, to quantify emergency demand at different risk levels. In high-risk situations, the required number of fire trucks, firefighters, and extinguishing agent quantities are determined. Through multivariate regression analysis, a model is established to model the relationship between emergency demand and risk level, generating specific emergency demand quantities for each risk category. Based on the comprehensive risk intervention decision parameters and the emergency demand analysis results, optimal resource allocation is calculated. The objective function is set to minimize resource usage costs while ensuring the effectiveness of the emergency response. The specific method uses linear programming or other optimization algorithms to set experimental parameters, such as resource usage costs and response time. A model is established to establish availability constraints for various emergency resources. The total resource quantity is limited to 100 units to ensure that the required resources do not exceed the available resources. By solving the optimization model, the optimal resource allocation plan is obtained, ensuring efficient resource scheduling in high-risk situations. Based on the optimized resource allocation results, a regional immediate fire intervention decision is generated. This decision will include specific information such as the number of fire trucks dispatched, the number of firefighters, and the type and quantity of extinguishing agent used. Experimental parameters are set to account for different risk levels and resource allocation flexibility. When the risk level is high, the decision might call for 10 fire trucks, 50 firefighters, and 200 liters of firefighting foam. In a medium-risk situation, this might be adjusted to 5 fire trucks and 20 firefighters. These decisions are dynamically adjusted based on real-time monitoring data to adapt to the evolving risk situation. Based on these analyses and decisions, a dynamic fire risk intervention strategy is constructed. This strategy should take into account the real-time status of resources and the dynamic changes in risk to ensure a rapid response in the event of a sudden fire. By integrating real-time monitoring data with historical data, the fire risk intervention strategy is dynamically updated.Establishing a sustainable emergency response process, including resource allocation, personnel deployment, and action guidance, will provide scientific and effective decision-making support for fire departments.

[0017] Step S6: Calculate the secondary fire risk probability based on the dynamic fire risk intervention strategy, perform coordinated scheduling optimization across the entire region, and build an intelligent fire decision-making and control model.

[0018] In this embodiment, based on the development of a dynamic fire risk intervention strategy, the probability of secondary fires is first calculated. This step aims to assess the potential risk of secondary fires occurring in a region after an initial fire. Experimental parameters are set, such as selecting fire data from the past year, combined with historical fire frequency and environmental conditions. Statistical analysis methods, such as the Poisson or binomial distribution, are used to model the probability of fire occurrence. Parameters are set, such as the incidence of secondary fires within the monitored region within 72 hours of the initial fire. Historical data is analyzed to determine the average probability of secondary fires occurring under specific climatic conditions (such as high temperature and dryness). This data is applied to the model to calculate the probability of secondary fires under different environmental conditions. It is assumed that within 24 hours after the initial fire, the probability of secondary fires increases significantly if the temperature exceeds 30°C and the humidity is below 20%. A secondary fire risk probability curve is ultimately generated to help identify risk trends over different time periods. After the secondary fire risk probability calculation is completed, coordinated scheduling optimization is performed across the entire region. This step aims to ensure that various emergency resources are effectively deployed to address potential secondary fire risks. Experimental parameters are set, such as resource limits, response time requirements, and resource priorities. Optimization algorithms, such as linear programming or mixed integer programming, are used to establish a resource scheduling model. Objective functions are set, such as minimizing response time and resource usage costs while meeting early warning requirements for secondary fire risk probability. Historical fire response data is analyzed to determine the required emergency resource allocation for different risk levels. The secondary fire risk probability is incorporated into the model as a constraint to ensure that resources are prioritized to potentially high-risk areas during high-risk periods. When the secondary fire risk probability exceeds a certain threshold, the system automatically adjusts resource allocation to minimize fire losses. Based on the optimization results, an intelligent firefighting decision-making and control model is constructed. This model will dynamically adjust emergency response strategies based on real-time monitoring data, historical data, and secondary fire risk probability. Experimental parameters are set, such as a model update frequency of every 30 minutes, to ensure rapid response to environmental changes. Machine learning algorithms, such as decision trees or reinforcement learning, are used to train the model, continuously optimizing decision rules based on historical fire data. The model will learn the optimal resource allocation plan for different scenarios and can self-adjust based on real-time data. When detecting changes in temperature and humidity in a specific area, the model can automatically adjust the deployment of firefighting resources and personnel. The constructed intelligent firefighting decision-making and control model will be integrated with the regional fire management system. This ensures that the decision-making model receives real-time sensor data and dynamically adjusts the emergency response plan based on the latest fire risk assessment results. Experimental parameters are set, such as a target response time of less than 15 minutes after each fire. This system enables fire departments to quickly and effectively dispatch resources and respond to sudden fires, reducing fire losses.Ultimately, the generated intelligent decision-making and control model will provide a scientific basis for regional fire management and improve the efficiency and effectiveness of emergency response.

[0019] In this embodiment, refer to Figure 2 , is a flowchart of the detailed implementation steps of step S1. In this embodiment, the detailed implementation steps of step S1 include: Collecting multimodal fire-related sensing sources in the area, wherein the multimodal fire-related sensing sources in the area include smoke concentration, ambient temperature, visible light video and infrared thermal imaging video; Performing multi-scale time synchronization processing on the multimodal fire-related perception source to obtain a time-synchronized multimodal perception source; Perform cross-modal fusion processing on time-synchronized multimodal perception sources and construct a multimodal perception parameter matrix; Calculating the temperature gradient, smoke concentration, thermal imaging characteristics, and acoustic spectrum of the multidimensional spatiotemporal sequence matrix to generate multidimensional fire situation characteristics; Perform multi-scale time-frequency decomposition on the multi-dimensional fire situation characteristics and construct a dynamic time-frequency spectrum of the regional fire situation; Abnormal event detection is performed on the dynamic time-frequency spectrum of the regional fire situation to identify fire risk mutation points.

[0020] In this example, sensors are deployed within a designated area to ensure coverage of key areas with potential fire risks. Assume the selected area is 1,000 square meters, with 10 smoke sensors, 5 temperature sensors, and 2 cameras deployed. The data collection frequency is set, for example, once per second to collect smoke concentration and temperature data. Visible light video and infrared thermal imaging video are simultaneously captured at 30 frames per second to ensure real-time data integrity. The collected data is transmitted in real time to a central processing unit for storage and processing, ensuring data security and reliability during the collection process. The smoke sensor monitors smoke concentration (in ppm) in real time. The temperature sensor monitors ambient temperature (in °C). The visible light camera captures real-time video of the area. The infrared thermal imager captures thermal video of the area to reflect temperature distribution. Accurate timestamps are set on each sensor and camera to accurately record the acquisition time of each data point. Interpolation algorithms (such as linear or spline interpolation) are used to synchronize data at different frequencies. Assuming smoke concentration and temperature data are collected every second, and the video frame rate is 30 frames per second, the temperature and smoke data need to be interpolated to the time of each video frame. Record the time synchronization results for each modality's data to generate a dataset of time-synchronized multimodal perception sources. Time synchronization ensures effective comparison and analysis of data from different modalities. Extract the time-synchronized smoke density, temperature, visible light video, and infrared thermal imaging video data as feature vectors. Extract the smoke density and temperature values ​​at each time point, along with motion features (such as the number of moving objects) extracted from the video. Combine these feature vectors to form a multimodal perception parameter matrix. Assume that the feature vector for each time point contains smoke density, temperature, visible light features, and thermal imaging features, for a total of four dimensions. Dimensionality reduction is performed using data fusion algorithms (such as principal component analysis (PCA)) to reduce data complexity while retaining important information. The resulting multimodal perception parameter matrix will be used for subsequent multidimensional spatiotemporal series analysis. Temperature gradient calculation: Calculate the temperature difference using data from adjacent temperature sensors (set a threshold of 1°C; any temperature change exceeding this value is recorded as an anomaly). Smoke density extraction: Analyze the changing trend of the smoke density data and calculate its standard deviation to assess volatility. Thermal imaging feature extraction: Extract thermal imaging features from infrared thermal imaging video to identify high-temperature areas (e.g., areas above 75°C). Acoustic spectrum calculation: If an acoustic wave sensor is available, extract acoustic wave spectrum features and analyze abnormal acoustic wave patterns (e.g., those generated by flames). These features are integrated to generate a multidimensional fire situation feature dataset, recording the changes in each feature for subsequent analysis. Select an appropriate wavelet basis (e.g., Daubechies wavelet) for time-frequency decomposition to decompose the multidimensional fire situation feature data. Assume that the number of decomposition levels selected is three. Decompose each feature (e.g., temperature, smoke concentration, etc.) to obtain feature representations at different frequencies and time scales.High-frequency components of temperature may indicate rapidly changing fire conditions, while low-frequency components may indicate a stable state. Generate a dynamic time-frequency spectrum to display the time-frequency characteristics of the fire situation, recording changes in these characteristics over time, and visually display the dynamics of the fire situation within the area. Determine anomaly detection thresholds and set normal ranges based on historical data. Set smoke concentration exceeding a certain value (e.g., 200 ppm) or temperature exceeding a certain threshold (e.g., 100°C) as an anomaly. Use machine learning algorithms (e.g., random forests or support vector machines) to train an anomaly detection model and perform classification based on multi-dimensional fire situation characteristics. Assume the training set has an accuracy of 90%. Monitor real-time data, compare new data with the model's predicted normal range, and identify sudden changes. If a sudden increase in smoke concentration to 300 ppm is detected, the system will flag it as an abnormal event. Detected abnormal events are recorded and an abnormal event detection report is generated, describing the identified fire risk sudden changes and their potential impacts to support subsequent emergency response decisions.

[0021] In this embodiment, refer to Figure 3 , is a flowchart of the detailed implementation steps of step S2. In this embodiment, the detailed implementation steps of step S2 include: Evolve the risk propagation trend of fire risk mutation points and construct the dynamic evolution characteristics of risk diffusion; Mining the risk propagation path based on the dynamic evolution characteristics of risk diffusion, extracting the diffusion trajectory of risk points; Identify the temporal variation characteristics of temperature fluctuation, smoke diffusion and harmful gas concentration at the fire risk mutation point, and generate the change pattern of the risk point; Deeply identify risk patterns based on the changing patterns of risk points, classify risk types, and generate risk point pattern features; Analyze the spatiotemporal distribution characteristics of risk point pattern characteristics to obtain the spatiotemporal distribution characteristics of the current risk pattern; According to the diffusion and propagation trajectory of risk points, a high-dimensional fire source tracing analysis is performed on the spatiotemporal distribution characteristics to obtain the regional fire risk tracing points.

[0022] In this example, we analyze fire risk mutation points. We first collect comprehensive historical data, including fire records, environmental monitoring data (such as temperature, humidity, smoke concentration, and hazardous gas concentration), and real-time sensor data. Fire records from 2019 to 2023 are selected, with monitoring data sampled hourly. Time series analysis is performed using an autoregressive integrated moving average (ARIMA) model. Parameters are set as a 1-day time window, a model order of (2, 1, 2), and a 95% confidence interval. By analyzing historical data, we identify peak periods of risk transmission, such as specific summer months (such as June and July), when fire incidence increases significantly. Using a rolling window technique, we update the model weekly to capture the latest trends. After identifying risk transmission trends, we construct a risk transmission network, with nodes representing different risk points and edges representing risk transmission paths. Simulations are performed using the SIR model, with experimental parameters set as an infection rate of 0.3 and a recovery rate of 0.1. By adjusting these parameters, we simulate risk transmission under different scenarios.

[0023] Using community detection methods from graph theory (such as the Louvain algorithm), key transmission paths were identified. It was discovered that fire risk within a certain community spread rapidly under specific weather conditions (such as high temperature and low humidity). Integrating GIS technology, transmission paths were visualized, identifying high-risk areas, such as commercial districts in city centers. The focus was on analyzing the environmental changes at fire risk mutation points, using real-time sensor data collection. The monitoring frequency was set to once per minute, recording temperature, smoke concentration, and hazardous gas concentrations (such as CO and NO2). The dynamic time warping (DTW) algorithm was used to compare time series data from different time periods. Analysis of temperature fluctuations at a specific mutation point before and after the fire revealed a 5°C increase in temperature within the 24 hours before the fire, and a sharp increase in smoke concentration within the two hours before the fire. These changes provided important evidence for subsequent risk pattern identification. Risk pattern identification and classification were performed based on the changing patterns of risk points. Cluster analysis (such as K-means algorithm) was used to select characteristic parameters such as the temperature change amplitude (set to 1°C, 3°C, and 5°C) and the smoke concentration change rate (set to 0.5 mg / m³, 1 mg / m³, and 2 mg / m³ per hour).

[0024] Through cross-validation, model parameters were optimized, and the number of clusters was set to three to identify different types of fire risk. Experimental results showed that certain risk patterns were more likely to occur under high temperatures and high smoke concentrations. The resulting risk point pattern features will facilitate subsequent spatiotemporal distribution analysis. The spatiotemporal distribution of risk point pattern features was analyzed using heat maps and spatial clustering methods. With a spatial resolution of 500 meters and a temporal resolution of 1 hour, the risk distribution over different time periods was analyzed to identify high-risk areas and peak times. It was found that the fire rate in a certain region was significantly higher on weekends during the summer than on weekdays. The spatiotemporal distribution analysis revealed that the peak fire period was between 3:00 PM and 6:00 PM. This finding provides a basis for the rational allocation of regional firefighting resources. Based on the diffusion and propagation trajectories of risk points, a high-dimensional source tracing analysis of the spatiotemporal distribution features was conducted. Combining multidimensional data analysis techniques, a random forest algorithm was used to identify fire source points. A cross-analysis of environmental monitoring data with historical fire data revealed that the average CO concentration in a certain region reached 50 mg / m³ in the 24 hours prior to the fire, which was used as a key indicator. Adjust the model parameters, set the feature selection threshold to 0.2, and finally determine the source point of regional fire risk. This analysis result provides a scientific basis for formulating effective fire prevention measures. According to the risk propagation path and spatiotemporal distribution characteristics, a retrospective model is established to analyze the possible location of the fire source. If the smoke concentration at a monitoring point rises significantly before the fire occurs, this point may be the source of the fire. Combining historical data and real-time monitoring data, a Bayesian network is used to perform source analysis to estimate the probability of different sources. Record the results of the source analysis and generate a regional fire risk source analysis report, describing the identified risk sources and their possible impacts, and providing decision support for emergency management.

[0025] In this embodiment, refer to Figure 4 , is a flowchart of the detailed implementation steps of step S3. In this embodiment, the detailed implementation steps of step S3 include: According to the regional fire situation dynamic time-frequency spectrum diagram, the risk situation of the regional fire risk traceability point is fitted to generate the risk evolution curve; Conduct morphological trend evolution analysis on the risk evolution curve and extract morphological trend characteristics; Extract the abnormal timestamp of regional fire risk traceability points; Estimating the risk evolution time of the abnormal timestamp based on the morphological trend characteristics to obtain a risk response time curve; The fire risk situation evolution is predicted based on the risk response time curve to obtain a fire simulation situation diagram.

[0026] In this embodiment, a dynamic time-frequency spectrum of regional fire situation is an important tool for analyzing fire risk. First, regional time-frequency spectrum data, including multidimensional data such as temperature, smoke concentration, and hazardous gas concentration, must be acquired from the monitoring system. This data should cover a specific time period, such as minute-by-minute data for the past 24 hours. Spectral analysis methods (such as short-time Fourier transform) are used to process the data to generate a time-frequency spectrum. Key band features are extracted from the time-frequency spectrum and converted into risk situation indicators using a weighted average method. Experimental parameters are set, such as a 5-minute spectral window size and a 50% overlap rate, to ensure the accuracy of the time-frequency features. The extracted risk situation indicators are then fitted to generate a risk evolution curve. Smoothing curve fitting methods (such as B-splines) can be used to smooth data fluctuations, ensuring the readability and accuracy of the curve. The resulting risk evolution curve provides the foundational data for subsequent analysis. To analyze the morphological trend evolution of the generated risk evolution curve, key features of the curve, such as peaks, valleys, and inflection points, must first be identified. Signal processing techniques can be used to identify abnormal fluctuations by setting a threshold (e.g., a standard deviation of 0.1).

[0027] During the analysis process, trend analysis methods (such as the moving average method) are applied to determine the long-term trend of the curve. A 30-minute time window is set, and the average value for each time period is calculated to capture changes in the long-term trend. This process extracts morphological trend features, such as the frequency, duration, and magnitude of risk fluctuations. Once identified, these trend features are further quantified to form a system of indicators, such as risk volatility and trend intensity. These features provide a basis for subsequent anomaly timestamp extraction. Based on the risk evolution curve, anomaly timestamps are extracted at regional fire risk source points. Statistical methods are used to establish anomaly detection criteria, such as the Z-score method, with an anomaly threshold of 3 standard deviations. Point-by-point analysis of the risk evolution curve identifies time points that exceed the threshold. If smoke concentration fluctuates within the normal range over a certain period but suddenly reaches 200 mg / m³ at a certain moment, this is marked as an anomaly timestamp. All anomaly timestamps are recorded and visualized for intuitive display. These anomaly timestamps serve as the basis for subsequent risk evolution time estimation, helping to identify potential sources of fire risk and their impact. Based on the morphological trend characteristics, the risk evolution time of the extracted anomaly timestamps is estimated. First, the relationship between the anomaly timestamps and the risk evolution curve is analyzed to determine the time delay of risk change after the occurrence of an anomaly event.

[0028] Using regression analysis, experimental parameters are set, such as using linear regression as the regression model, and analyzing the mean and standard deviation of time delays. Risk response time is estimated by analyzing risk changes before and after the anomaly timestamp. If the risk evolution curve changes significantly within 10 minutes after the anomaly occurs, the response time is 10 minutes. The generated risk response time curve will provide an important basis for subsequent fire risk situation evolution prediction. Fire risk situation evolution prediction is performed based on the risk response time curve. A time series prediction model (such as an ARIMA or LSTM model) is used as input for risk response time data to predict risk changes over a period of time. Experimental parameters are set, such as a 1-hour prediction window and a 72-hour model training cycle. Through model training, patterns in risk changes are identified and the potential future risk level is predicted. The risk level within the next 30 minutes is predicted to be medium, based on reasonable inferences based on historical data patterns. The generated fire simulation situation map will visually demonstrate the future fire risk situation, providing effective decision support for fire departments. This process combines real-time monitoring data with historical data to ensure accurate and timely predictions.

[0029] In this embodiment, the specific steps of predicting the evolution of the fire risk situation based on the risk response time curve to obtain a fire simulation situation diagram are as follows: Conduct dynamic risk topology mapping based on the risk response time curve and construct a dynamic risk propagation topology network; Perform deep fire semantic analysis on the dynamic risk propagation topology network to obtain fire risk semantic data; Perform multimodal semantic fusion simulation based on fire risk semantic data to obtain smoke concentration distribution simulation maps, fire source spread simulation maps, and trapped area distribution maps; The fire risk situation evolution is predicted based on the smoke concentration distribution simulation map, fire source spread simulation map and trapped area distribution map to obtain a fire simulation situation map.

[0030] In this example, starting from the risk response time curve, a dynamic risk propagation topological network is first constructed. Historical fire event data and real-time monitoring data (such as temperature, smoke concentration, and hazardous gas concentration) within the region are collected and converted into network nodes and edges. Nodes represent individual risk points, while edges represent risk propagation paths. Network construction methods from graph theory, such as the Dijkstra algorithm, are used to determine the connectivity and propagation weights between nodes. Experimental parameters are set, such as a connection strength threshold of 0.5, to ensure that only significant risk propagation paths are included in the topological network. By visualizing the topological network, key risk nodes and propagation paths are identified, and the network's connectivity and clustering characteristics are analyzed. Ultimately, a dynamic risk propagation topological network is constructed that reflects the dynamic changes in fire risk propagation within the region in real time. After constructing the dynamic risk propagation topological network, deep fire semantic parsing is performed. The real-time data generated during the risk propagation process is combined with historical fire data to extract semantic features of fire risk. These features include fire source location, smoke concentration changes, and temperature fluctuations. Natural language processing (NLP) techniques are used to analyze fire-related text data (such as alarm records and on-site reports) to extract key information. Experimental parameters were set, such as using the TF-IDF model to extract keywords and selecting the top 10 keywords for analysis. By combining spatiotemporal data, a semantic dataset for fire risk was constructed, ensuring data integrity and accuracy. This process laid the foundation for subsequent multimodal semantic fusion simulations. Based on the fire risk semantic data, a multimodal semantic fusion simulation was conducted to generate simulated smoke concentration distribution maps, fire source spread maps, and trapped area distribution maps. First, smoke concentration was spatially interpolated using the Kriging interpolation method. Experimental parameters were set, such as a sample size of 50, to ensure model accuracy. Next, a fire spread model (such as the FARSITE model) was used to simulate the spread of the fire. The fire spread speed was set to 1 meter per minute, and environmental factors such as wind speed and humidity were considered to generate a simulated fire spread map. Trapped areas were identified based on regional topography and building distribution, and a trapped area distribution map was generated. Using GIS technology, the simulation results for smoke concentration, fire source spread, and trapped areas were integrated to form a comprehensive fire risk situation map. The generated smoke concentration distribution simulation map, fire source spread simulation map, and trapped area distribution map are used to predict the evolution of fire risk situations. A time series prediction model (such as a long short-term memory (LSTM) network) is used to train and predict multimodal data. Experimental parameters are set, such as a 30-minute prediction window and a 72-hour model training cycle. The trained model identifies patterns in fire risk changes and predicts potential future risk levels. The resulting fire simulation situation map visually displays the future fire risk situation, including smoke spread paths, fire source spread areas, and potential trapped areas.This process will combine real-time monitoring data and historical data to ensure the accuracy and timeliness of predictions and provide effective decision-making support for fire departments.

[0031] In this embodiment, step S4 includes the following steps: Conduct fire risk level assessment on regional fire risk source points to obtain the current fire risk level; According to the fire simulation situation diagram, the current fire risk level is dynamically tracked to obtain the risk level evolution trajectory; Conduct fire intervention demand analysis on the risk level evolution trajectory to obtain emergency intervention demand information; Calculate the intensity and direction of emergency intervention demand information to obtain the amount and priority of intervention resources; Conduct timeliness response characteristic demand analysis on emergency intervention demand information to obtain the intervention response time window; Comprehensive emergency demand adjustments are made to the intervention response time window, intervention resource allocation and priority to obtain comprehensive risk intervention decision parameters.

[0032] In this example, fire risk level assessment for regional fire risk traceability points requires first collecting multidimensional fire-related data, including historical fire records, real-time monitoring data (such as temperature, smoke concentration, and hazardous gas concentration), and environmental factors (such as wind speed and humidity). Experimental parameters are set to limit data collection to the past 72 hours to ensure data timeliness and relevance. A risk assessment model (such as a quantitative risk assessment model) is used to comprehensively analyze the collected data. Scoring criteria are set to assign weights to fire risk factors (such as temperature and smoke concentration). Temperature is weighted 0.4, smoke concentration 0.3, hazardous gas concentration 0.2, and wind speed 0.1. The risk level of each risk point is calculated using a weighted average method, generating an assessment of the current fire risk level. Ultimately, based on the assessment results, the risk level is categorized into low, medium, and high to clarify the current fire risk status and provide a basis for subsequent decision-making. Dynamic changes in the current fire risk level are tracked using a fire simulation situation map. First, a dynamically updated risk level model is constructed by combining real-time monitoring data with historical data. The tracking period is set to update every hour to ensure timely reflection of changes in risk levels. A moving average method is used to smooth risk level data to eliminate the impact of short-term fluctuations and extract the evolution trajectory of the risk level. Experimental parameters are set, such as a 3-hour time window, to observe the changing trend of the risk level within this time period. Visualization tools are used to display the risk level evolution trajectory on a fire simulation situation map for easier understanding and analysis. Fire intervention needs are analyzed based on the risk level evolution trajectory. First, an intervention threshold is set to define the level at which intervention is required when the risk level exceeds a certain threshold. Rate of change analysis is used to calculate the magnitude of risk level changes over different time periods. When the risk level rises by more than 0.5 within one hour, it is marked as a high intervention need. Based on historical case studies, the intervention needs corresponding to different risk levels are analyzed to generate emergency intervention need information. If the risk level is high, additional firefighting resources are required for emergency response. Finally, emergency intervention need information is generated, including the required equipment, personnel, and resource types, to provide a basis for subsequent resource allocation. Based on the emergency intervention need information, the intensity and direction of the intervention need are calculated. An intensity calculation model is developed to quantify the intensity of intervention needs at different risk levels based on historical data and real-time monitoring data. For high risk levels, the required resource intensity is set at 100 units, for medium risk levels at 50 units, and for low risk levels at 20 units. Simultaneously, intervention targets are analyzed, specifically the priority areas and resource allocation. If the risk level in a particular area continues to rise, resources are allocated to that area first. A weighted allocation method is used to determine the priority of intervention resource allocation for each area, ultimately generating a list of intervention resource allocation quantities and priorities. Emergency intervention demand information is analyzed for its timeliness and response characteristics to determine the intervention response window. Target response times are set through historical data analysis.For high risk levels, the response time should be controlled within 15 minutes; for medium risk levels, it should be 30 minutes; and for low risk levels, it should be 60 minutes.

[0033] Using time series analysis, we analyze the distribution characteristics of historical response times and identify the optimal response time window for each risk level. Experimental parameters are set, such as analyzing response time data from the past 12 interventions, to ensure the accuracy and reliability of the analysis results. We generate evaluation results for the intervention response time window, providing a scientific basis for subsequent emergency decision-making. We adjust the intervention response time window, intervention resource allocation, and priority based on comprehensive emergency needs to develop comprehensive risk intervention decision parameters. Using a multi-objective optimization algorithm, we set an objective function to minimize response time and resource consumption while ensuring the effectiveness of emergency intervention. By combining real-time monitoring data with historical data, we dynamically adjust intervention resource allocation. When the risk level in a particular area increases, we immediately increase resources allocated to that area to ensure a timely and effective emergency response. By generating comprehensive risk intervention decision parameters, we provide the fire department with a complete emergency response plan, ensuring rapid and effective intervention in the event of a fire risk.

[0034] In this embodiment, step S5 includes the following steps: Conduct real-time status monitoring of regional emergency resource systems and obtain time-series data streams on the availability of emergency resources; Perform resource scheduling feasibility analysis on the time series data stream of emergency resource availability status and extract resource scheduling feasibility status; Conduct executable emergency capability mining on the feasibility status of resource scheduling and build a schedulable emergency resource model; Optimize resource allocation calculations for comprehensive risk intervention decision parameters based on the dispatchable emergency resource model to obtain the optimal emergency response resources; Make regional immediate fire intervention decisions based on optimized emergency response resources and build a dynamic fire risk intervention strategy.

[0035] In this embodiment, real-time status monitoring of the regional emergency resource system requires the deployment of sensors and monitoring equipment to collect data on the availability of emergency resources. This data includes information such as resource type (e.g., fire trucks, fire extinguishers, firefighters), current status (available, busy, down), and location. Experimental parameters are set, such as a monitoring frequency of every 5 minutes, to ensure data timeliness. The data acquisition system transmits real-time status data to a central database, forming a time-series data stream of emergency resource availability. During data processing, data cleaning techniques are applied to remove outliers and missing data to ensure data accuracy and completeness. This process provides basic data support for subsequent resource scheduling feasibility analysis. Resource scheduling feasibility analysis is performed on the time-series data stream of emergency resource availability. First, based on the real-time status data, the current availability of resources is assessed and scheduling feasibility criteria are set. The resource availability threshold is set to 80%. When the number of available resources falls below 80% of the total number of resources, the resource is marked as unschedulable. A decision tree model is used to evaluate the feasibility of resource scheduling and analyze the scheduling possibilities under different conditions. Using a classification algorithm, resource status is divided into three categories: schedulable, partially schedulable, and non-schedulable. Finally, the resource schedulability status is extracted and a feasibility report is generated, providing a basis for subsequent emergency capacity mining. After obtaining the resource schedulability status, executable emergency capacity mining is performed. Combined with historical emergency event data, emergency capacity under different resource states is analyzed. The success rate of fire rescue missions under specific resource combinations is evaluated. Experimental parameters are set, such as analyzing emergency event data within the past year, and selecting key performance indicators (KPIs) such as response time, success rate, and resource utilization. Multivariate regression analysis is used to identify the main factors affecting emergency capacity and construct a schedulable emergency resource model. Through model analysis, the optimal configuration scheme for different resource combinations is determined, ensuring a rapid and efficient response to emergency needs under specific risk levels. Based on the schedulable emergency resource model, the optimal resource allocation is calculated for comprehensive risk intervention decision parameters. First, the optimization objective is set, such as minimizing response time and resource cost while maximizing emergency capacity.

[0036] A linear programming algorithm was used to construct an optimization model. Resource availability and emergency response capacity data were input, along with constraints such as resource quantity and response time. The objective function for resource allocation was set as: minimizing Z = c1 * x1 + c2 * x2, where c1 and c2 represent resource cost coefficients and x1 and x2 represent the allocated quantities of each resource. The optimization model was run to obtain the optimal emergency response resource allocation plan. This plan included the specific allocation quantities and scheduling strategies for each resource, providing data support for regional immediate fire intervention decisions. Based on the optimized emergency response resources, regional immediate fire intervention decisions were made. First, the optimized resource allocation plan was compared with the current fire risk level to assess the effectiveness of resource allocation. Emergency response decision criteria were set: when the fire risk level was high, all available resources were immediately dispatched; when the risk level was medium, only a portion of the resources were dispatched. Integrating dynamic monitoring data, the resource scheduling strategy was adjusted in real time to ensure flexibility and effectiveness of the emergency response. Finally, a dynamic fire risk intervention strategy was developed, along with an emergency response manual that clarified the scheduling process and implementation standards for each resource. This strategy will help regional fire departments intervene quickly and effectively in sudden fire incidents and reduce fire losses.

[0037] In this embodiment, step S6 includes the following steps: Collect immediate fire intervention response parameters based on dynamic fire risk intervention strategies; Quantifying the response effect of the immediate fire intervention response parameters to obtain the effectiveness of the risk intervention response; Calculate regional analytical residuals based on the effectiveness of risk intervention responses; Calculating the secondary fire risk probability based on the regional analysis residue to obtain a secondary fire risk probability curve; Conduct dynamic risk cascade effect analysis on the secondary fire risk probability curve to generate a current secondary disaster risk assessment report; Based on the current secondary disaster risk assessment, the dynamic fire risk intervention strategy is coordinated and optimized across the entire region to build an intelligent fire decision-making and control model.

[0038] In this embodiment, according to the dynamic fire risk intervention strategy, the first step is to collect immediate fire intervention response parameters. These parameters include response time, number of resources dispatched, on-site personnel deployment, and amount of fire extinguishing agent used. Experimental parameters are set, such as a monitoring frequency of immediate collection after each intervention, and detailed records of each response are kept. These parameters are input into a central database through a data acquisition system to form a response parameter dataset. To ensure data accuracy and completeness, data cleaning techniques are used to eliminate incomplete or abnormal data records. By classifying the response parameters, a summary of response parameters for different fire scenarios is generated to facilitate subsequent quantitative analysis of the effects. To quantify the response effects of immediate fire intervention response parameters, evaluation criteria must first be determined. These criteria may include fire extinguishing success rate, the difference between response time and expected time, and resource utilization efficiency. Experimental parameters are set, such as a success rate threshold of 80%, meaning that completing the fire extinguishing task 80% of the time is considered an effective response.

[0039] A weighted scoring method is used to weight each response parameter to calculate the total score of the response effect. For example, the fire extinguishing success rate accounts for 60%, the response time accounts for 30%, and the resource utilization efficiency accounts for 10%. The calculation formula is: Total score = 0.6 × success rate + 0.3 × (1- ) + 0.1 × resource utilization efficiency; the resulting score will reflect the effectiveness of the risk intervention response and provide basic data for subsequent analysis. Based on the effectiveness of the risk intervention response, the regional analysis residual is calculated. The residual represents the fire risk that remains after the intervention and needs to be estimated based on the response effect and fire characteristics. Experimental parameters are set, such as the time range for residual calculation is within 24 hours after the intervention. The residual is calculated using the residual risk assessment model combined with the response effectiveness score. The residual risk is set as: Residual = Initial Risk × (1 − Response Effectiveness Score); by analyzing the residual after each intervention, the risk status of different time periods within the region is generated, providing data support for the subsequent calculation of the secondary fire risk probability. The secondary fire risk probability is calculated based on the regional analysis residual. Using statistical analysis methods and a probability model such as the Poisson distribution or binomial distribution, the probability of a secondary fire occurring within a specific time period is calculated.

[0040] Experimental parameters are set, such as an analysis timeframe of 48 hours after the intervention. Historical data is used to estimate the probability of secondary fire risk. If, during the past five interventions, an average of two secondary fires occurred each time, a corresponding probability curve can be generated. Generating a secondary fire risk probability curve facilitates monitoring and forecasting of future secondary fire risks, providing baseline data for dynamic risk cascade analysis. Dynamic risk cascade analysis is performed on the secondary fire risk probability curve. First, key factors influencing secondary fire risk are identified, such as environmental conditions, residual risk, and resource allocation. Experimental parameters are set to analyze dynamic changes within 72 hours after the intervention. Dynamic models (such as system dynamics models) are used to simulate secondary fire risk and analyze the impact of different factors on risk. A current secondary disaster risk assessment report is generated, which includes risk probability trends, influencing factors, and their interactions, helping decision makers understand risk propagation mechanisms. Based on the current secondary disaster risk assessment, dynamic fire risk intervention strategies are optimized and coordinated across the region. First, information on various emergency resources is integrated to assess the distribution of available resources within the region and set resource scheduling priorities. Optimization algorithms (such as genetic algorithms or particle swarm optimization) are used to determine the optimal resource scheduling plan. Optimization objectives are set, such as minimizing response time and resource costs while ensuring the risk of secondary fires is effectively controlled. Ultimately, an intelligent firefighting decision-making and control model is constructed, integrating real-time monitoring data with historical data to dynamically adjust resource scheduling and emergency response strategies. This model will provide fire departments with scientific and effective decision-making support, improving the efficiency and effectiveness of emergency response.

[0041] In this embodiment, a regional fire risk analysis system using multimodal AI analysis is provided, which is used to perform the regional fire risk analysis method using multimodal AI analysis as described above, including: The risk mutation detection module is used to collect multimodal fire-related sensor sources in the area, perform cross-modal fusion processing and abnormal event detection, and identify fire risk mutation points; The risk tracing module is used to analyze the risk propagation trend evolution of fire risk mutation points and conduct high-dimensional fire source tracing analysis to obtain regional fire risk tracing points; The risk situation prediction module is used to estimate the risk evolution time of regional fire risk traceability points, and then predict the evolution of fire risk situation to obtain a fire simulation situation map; The emergency demand decision module is used to track the dynamic risk level changes based on the fire simulation situation map and make comprehensive emergency demand adjustments to obtain comprehensive risk intervention decision parameters; The optimal resource allocation module is used to obtain the time series data stream of the available status of emergency resources, and to perform optimal resource allocation calculations and regional immediate fire intervention decisions based on comprehensive risk intervention decision parameters, thereby building a dynamic fire risk intervention strategy; The full-area scheduling module is used to calculate the secondary fire risk probability based on the dynamic fire risk intervention strategy, perform full-area collaborative scheduling optimization, and build an intelligent fire decision-making and control model.

[0042] The present invention is therefore intended to be illustrative and non-restrictive in all respects, with the scope of the invention being defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the application documents are intended to be embraced therein.

[0043] The foregoing description is intended only to provide specific embodiments of the present invention, which will enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be construed in the widest possible manner consistent with the principles and novel features disclosed herein.

Claims

1. A regional fire risk analysis method based on multimodal AI analysis, characterized in that: The following steps are involved: Step S1: Collect multimodal fire-related sensor sources in the area, perform cross-modal fusion processing and abnormal event detection, and identify fire risk mutation points; Step S2: Evolve the risk propagation trend of the fire risk mutation point and conduct high-dimensional fire source tracing analysis to obtain the regional fire risk tracing point; Step S3: Estimate the risk evolution time of the regional fire risk traceability point, and then predict the evolution of the fire risk situation to obtain a fire simulation situation map; Step S4: Tracking the dynamic risk level changes based on the fire simulation situation map, and making comprehensive emergency demand adjustments to obtain comprehensive risk intervention decision parameters; Step S5: Obtain the time series data stream of the available status of emergency resources, and perform optimal resource allocation calculation and regional immediate fire intervention decision-making on the comprehensive risk intervention decision parameters to build a dynamic fire risk intervention strategy; Step S6: Calculate the secondary fire risk probability based on the dynamic fire risk intervention strategy, perform coordinated scheduling optimization across the entire region, and build an intelligent fire decision-making and control model.

2. The regional fire risk analysis method based on multimodal AI analysis according to claim 1 is characterized in that: The specific steps of step S1 are: Collecting multimodal fire-related sensing sources in the area, wherein the multimodal fire-related sensing sources in the area include smoke concentration, ambient temperature, visible light video and infrared thermal imaging video; Performing multi-scale time synchronization processing on the multimodal fire-related perception source to obtain a time-synchronized multimodal perception source; Perform cross-modal fusion processing on time-synchronized multimodal perception sources and construct a multimodal perception parameter matrix; Calculating the temperature gradient, smoke concentration, thermal imaging characteristics, and acoustic spectrum of the multidimensional spatiotemporal sequence matrix to generate multidimensional fire situation characteristics; Perform multi-scale time-frequency decomposition on the multi-dimensional fire situation characteristics and construct a dynamic time-frequency spectrum of the regional fire situation; Abnormal event detection is performed on the dynamic time-frequency spectrum of the regional fire situation to identify fire risk mutation points.

3. The regional fire risk analysis method based on multimodal AI analysis according to claim 1 is characterized in that: The specific steps of step S2 are: Evolve the risk propagation trend of fire risk mutation points and construct the dynamic evolution characteristics of risk diffusion; Mining the risk propagation path based on the dynamic evolution characteristics of risk diffusion, extracting the diffusion trajectory of risk points; Identify the temporal variation characteristics of temperature fluctuation, smoke diffusion and harmful gas concentration at the fire risk mutation point, and generate the change pattern of the risk point; Deeply identify risk patterns based on the changing patterns of risk points, classify risk types, and generate risk point pattern features; Analyze the spatiotemporal distribution characteristics of risk point pattern characteristics to obtain the spatiotemporal distribution characteristics of the current risk pattern; According to the diffusion and propagation trajectory of risk points, a high-dimensional fire source tracing analysis is performed on the spatiotemporal distribution characteristics to obtain the regional fire risk tracing points.

4. The regional fire risk analysis method based on multimodal AI analysis according to claim 1 is characterized in that: The specific steps of step S3 are: According to the regional fire situation dynamic time-frequency spectrum diagram, the risk situation of the regional fire risk traceability point is fitted to generate the risk evolution curve; Conduct morphological trend evolution analysis on the risk evolution curve and extract morphological trend characteristics; Extract the abnormal timestamp of regional fire risk traceability points; Estimating the risk evolution time of the abnormal timestamp based on the morphological trend characteristics to obtain a risk response time curve; The fire risk situation evolution is predicted based on the risk response time curve to obtain a fire simulation situation diagram.

5. The regional fire risk analysis method based on multimodal AI analysis according to claim 4 is characterized in that: The specific steps of predicting the evolution of the fire risk situation based on the risk response time curve to obtain the fire simulation situation map are as follows: Conduct dynamic risk topology mapping based on the risk response time curve and construct a dynamic risk propagation topology network; Perform deep fire semantic analysis on the dynamic risk propagation topology network to obtain fire risk semantic data; Perform multimodal semantic fusion simulation based on fire risk semantic data to obtain smoke concentration distribution simulation maps, fire source spread simulation maps, and trapped area distribution maps; The fire risk situation evolution is predicted based on the smoke concentration distribution simulation map, fire source spread simulation map and trapped area distribution map to obtain a fire simulation situation map.

6. The regional fire risk analysis method based on multimodal AI analysis according to claim 1 is characterized in that: The specific steps of step S4 are: Conduct fire risk level assessment on regional fire risk source points to obtain the current fire risk level; According to the fire simulation situation diagram, the current fire risk level is dynamically tracked to obtain the risk level evolution trajectory; Conduct fire intervention demand analysis on the risk level evolution trajectory to obtain emergency intervention demand information; Calculate the intensity and direction of emergency intervention demand information to obtain the amount and priority of intervention resources; Conduct timeliness response characteristic demand analysis on emergency intervention demand information to obtain the intervention response time window; Comprehensive emergency demand adjustments are made to the intervention response time window, intervention resource allocation and priority to obtain comprehensive risk intervention decision parameters.

7. The regional fire risk analysis method based on multimodal AI analysis according to claim 1 is characterized in that: The specific steps of step S5 are: Conduct real-time status monitoring of regional emergency resource systems and obtain time-series data streams on the availability of emergency resources; Perform resource scheduling feasibility analysis on the time series data stream of emergency resource availability status and extract resource scheduling feasibility status; Conduct executable emergency capability mining on the feasibility status of resource scheduling and build a schedulable emergency resource model; Optimize resource allocation calculations for comprehensive risk intervention decision parameters based on the dispatchable emergency resource model to obtain the optimal emergency response resources; Make regional immediate fire intervention decisions based on optimized emergency response resources and build a dynamic fire risk intervention strategy.

8. The regional fire risk analysis method based on multimodal AI analysis according to claim 1 is characterized in that: The specific steps of step S6 are: Collect immediate fire intervention response parameters based on dynamic fire risk intervention strategies; Quantifying the response effect of the immediate fire intervention response parameters to obtain the effectiveness of the risk intervention response; Calculate regional analytical residuals based on the effectiveness of risk intervention responses; Calculating the secondary fire risk probability based on the regional analysis residue to obtain a secondary fire risk probability curve; Conduct dynamic risk cascade effect analysis on the secondary fire risk probability curve to generate a current secondary disaster risk assessment report; Based on the current secondary disaster risk assessment, the dynamic fire risk intervention strategy is coordinated and optimized across the entire region to build an intelligent fire decision-making and control model.

9. A regional fire risk analysis system based on multimodal AI analysis, characterized in that: A method for performing regional fire risk analysis using multimodal AI analysis according to claim 1, comprising: The risk mutation detection module is used to collect multimodal fire-related sensor sources in the area, perform cross-modal fusion processing and abnormal event detection, and identify fire risk mutation points; The risk tracing module is used to analyze the risk propagation trend evolution of fire risk mutation points and conduct high-dimensional fire source tracing analysis to obtain regional fire risk tracing points; The risk situation prediction module is used to estimate the risk evolution time of regional fire risk traceability points, and then predict the evolution of fire risk situation to obtain a fire simulation situation map; The emergency demand decision module is used to track the dynamic risk level changes based on the fire simulation situation map and make comprehensive emergency demand adjustments to obtain comprehensive risk intervention decision parameters; The optimal resource allocation module is used to obtain the time series data stream of the available status of emergency resources, and to perform optimal resource allocation calculations and regional immediate fire intervention decisions based on comprehensive risk intervention decision parameters, thereby building a dynamic fire risk intervention strategy; The full-area scheduling module is used to calculate the secondary fire risk probability based on the dynamic fire risk intervention strategy, perform full-area collaborative scheduling optimization, and build an intelligent fire decision-making and control model.

Images