Fire accident investigation method, device, equipment, storage medium and product

Abstract

The invention relates to the technical field of fire tracing, in particular to a fire accident investigation method, device and equipment, a storage medium and a product. The method comprises the following steps: generating a digital model of a fire accident based on BIM, GIS and Internet of Things technologies; key data of a fire scene are collected and analyzed based on the digital model, and the key data comprise a three-dimensional live-action model, a point cloud model, a trace image, a fire scene video, an electronic material evidence and monitoring data; performing feature extraction on the key data to obtain fire feature data; constructing a fire feature knowledge graph based on the fire feature data; based on the fire feature knowledge graph, the deep learning model is trained, pre-judgment of fire risks in specific areas, time and types is formed, and the accuracy and efficiency of fire accident investigation are improved.

Description

Technical Field

[0001] The present application relates to the technical field of fire source tracing, and in particular to a fire accident investigation method, device, equipment, storage medium and product. Background Art

[0002] The investigation and analysis of fire accidents is an important part of the field of fire safety. Its purpose is to find out the cause of the fire, clarify the responsibility for the accident, and propose corresponding prevention and improvement measures through the investigation and data analysis of the fire accident scene. At present, traditional fire investigations mainly rely on manual on-site inspections, eyewitness interviews, and physical evidence collection, lacking systematic and digital tool support. These methods have great limitations when dealing with complex fire scenes, resulting in poor accuracy and efficiency in fire accident investigations. Therefore, how to improve the accuracy and efficiency of fire accident investigations has become a technical problem that needs to be solved urgently. Summary of the invention

[0003] The main purpose of this application is to provide a fire accident investigation method, device, equipment, storage medium and product, aiming to solve the technical problem of how to improve the accuracy and efficiency of fire accident investigation.

[0004] To achieve the above objectives, the present application provides a fire accident investigation method, the method comprising the following steps:

[0005] Generate digital models of fire accidents based on BIM, GIS and IoT technologies;

[0006] Based on the digital model, analyzing the key data of the fire scene, the key data includes a three-dimensional real scene model, a point cloud model, a trace image, a fire scene video, electronic evidence and monitoring data;

[0007] Extract features from the analysis results to obtain fire feature data;

[0008] Based on the fire characteristic data, construct a fire characteristic knowledge graph;

[0009] Based on the fire characteristic knowledge graph, a deep learning model is trained to form a prediction of fire risks in specific areas, times, and types.

[0010] In one embodiment, the step of generating a digital model of a fire accident based on BIM, GIS and Internet of Things technologies includes:

[0011] Acquire basic spatial data from the CIM data base, wherein the basic spatial data includes building unit data, GIS base map, and thematic spatial data;

[0012] Digitally processing the basic spatial data to obtain GIS data;

[0013] Build a BIM model based on the BIM data of the on-site building;

[0014] Superimposing the GIS data with the BIM model to obtain a preliminary model;

[0015] The preliminary model is integrated with the IoT data, and based on a preset data base, the GIS data is associated with the IoT data to obtain a digital model of the fire accident.

[0016] In one embodiment, before the step of extracting features from the analysis results to obtain fire feature data, the step further includes:

[0017] Optimizing the analysis results to obtain optimized data, wherein the optimization processing includes one or more of transcoding processing, format parsing and recovery processing, and cleaning and denoising processing;

[0018] Synchronizing the optimized data to the preset data base;

[0019] Based on the preset data base, a fire scene report is generated.

[0020] In one embodiment, the step of extracting features from the analysis results to obtain fire feature data includes:

[0021] Extracting fire visual data, key person data, equipment operation data, damage feature data, and fire trace data from the analysis results based on a preset feature extraction strategy;

[0022] Performing curve fitting and anomaly detection on the environmental characteristics in the analysis results to extract key monitoring data;

[0023] The fire visual data, the key person data, the equipment operation data, the damage feature data, the fire trace data and the key monitoring data are standardized to obtain the fire feature data, wherein the standardization processing includes one or more of normalization processing, data denoising processing and time synchronization processing.

[0024] In one embodiment, the step of constructing a fire characteristic knowledge graph based on the fire characteristic data includes:

[0025] Preprocessing the fire characteristic data to obtain preprocessed data, wherein the preprocessing includes one or more of data classification, data cleaning, and format conversion;

[0026] Based on a preset graph grid structure, extracting data from the preprocessed data, and generating graph nodes based on the extracted data;

[0027] Determine the relationship between the graph nodes based on the causal relationship and the correlation relationship between the extracted data;

[0028] Timestamp information is added to the graph nodes, and the fire characteristic knowledge graph is constructed based on the relationship between the graph nodes.

[0029] In one embodiment, after the step of adding timestamp information to the graph nodes and constructing the fire characteristic knowledge graph based on the relationship between the graph nodes, the step further includes:

[0030] Performing deduplication processing on the fire characteristic knowledge graph, wherein the deduplication processing includes one or more of consistency checking, redundant relationship removal, and knowledge reasoning;

[0031] The deduplicated fire characteristic knowledge graph is stored in a preset graph database and displayed through a graphical interface;

[0032] Acquire new data in real time, and update the fire characteristic knowledge graph based on the new data.

[0033] In addition, to achieve the above-mentioned purpose, the present application also proposes a fire accident investigation device, the fire accident investigation device comprising:

[0034] Digital model generation module, used to generate digital models of fire accidents based on BIM, GIS and IoT technologies;

[0035] A data analysis module, used to analyze key data of the fire scene based on the digital model, wherein the key data includes a three-dimensional real scene model, a point cloud model, a trace image, a fire scene video, electronic evidence and monitoring data;

[0036] A feature extraction module is used to extract features from the analysis results to obtain fire feature data;

[0037] A graph construction module, used to construct a fire characteristic knowledge graph based on the fire characteristic data;

[0038] The target module is used to train a deep learning model based on the fire characteristic knowledge graph to form a prediction of the fire risk in a specific area, time, and type.

[0039] In addition, to achieve the above-mentioned purpose, the present application also proposes a fire accident investigation device, which includes: a memory, a processor, and a fire accident investigation program stored on the memory and executable on the processor, wherein the fire accident investigation program is configured to implement the steps of the fire accident investigation method described above.

[0040] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, on which a fire accident investigation program is stored, and when the fire accident investigation program is executed by a processor, the steps of the fire accident investigation method described above are implemented.

[0041] In addition, to achieve the above-mentioned purpose, the present application also proposes a computer program product, which includes a computer program, and when the computer program is executed by a processor, the steps of the fire accident investigation method described above are implemented.

[0042] This application generates a digital model of a fire accident based on BIM, GIS and IoT technologies; based on the digital model, analyzes key data of the fire scene, including three-dimensional real-scene models, point cloud models, trace images, fire scene videos, electronic evidence and monitoring data; extracts features from the analysis results to obtain fire feature data; based on the fire feature data, constructs a fire feature knowledge graph; based on the fire feature knowledge graph, trains a deep learning model to form a prediction of fire risks in specific areas, times and types. This application generates a digital model of fire accidents based on BIM, GIS and Internet of Things technologies, integrating three-dimensional real-life models, point cloud models, trace images, fire scene videos, electronic evidence and monitoring data, and realizing the comprehensive digitization of fire scene information; based on the model, the key data of the fire scene is analyzed, and the fire feature data is extracted through feature extraction, providing accurate and structured core information for subsequent analysis; a fire feature knowledge graph is constructed through fire feature data, which systematically represents fire features and their correlations, thereby improving the efficiency of data analysis and the ability of knowledge management; further based on the knowledge graph, a deep learning model is trained to realize the prediction of fire risks in specific areas, times and types, thereby enhancing the accuracy and initiative of fire prevention and control, and improving the accuracy and efficiency of fire accident investigations. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a flow chart of the first embodiment of the fire accident investigation method of the present application;

[0044] Figure 2 This is a schematic diagram of a sub-process in the second embodiment of the fire accident investigation method of the present application;

[0045] Figure 3 This is a schematic diagram of a sub-flow in the third embodiment of the fire accident investigation method of the present application;

[0046] Figure 4 This is a functional framework diagram of fire accident investigation in an embodiment of the fire accident investigation method of this application;

[0047] Figure 5 This is a schematic diagram of the module structure of the fire accident investigation device according to an embodiment of the present application;

[0048] Figure 6 Schematic diagram of the equipment structure of the hardware operating environment involved in the fire accident investigation method in the embodiment of the present application.

[0049] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0050] It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0051] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0052] It should be noted that the investigation and analysis of fire accidents is an important part of the field of fire safety. Its purpose is to find out the cause of the fire, clarify the responsibility for the accident, and propose corresponding prevention and improvement measures through the investigation and data analysis of the fire accident scene. At present, traditional fire investigations mainly rely on manual on-site investigations, eyewitness interviews, and physical evidence collection, and lack systematic and digital tool support. These methods have great limitations when dealing with complex fire scenes, resulting in poor accuracy and efficiency in fire accident investigations. Therefore, how to improve the accuracy and efficiency of fire accident investigations has become a technical problem that needs to be solved urgently.

[0053] The main solutions of this application are: based on BIM, GIS and Internet of Things technologies, a digital model of fire accidents is generated; based on the digital model, key data of the fire scene is analyzed, and the key data includes three-dimensional real-scene models, point cloud models, trace images, fire scene videos, electronic evidence and monitoring data; feature extraction is performed on the analysis results to obtain fire feature data; based on the fire feature data, a fire feature knowledge graph is constructed; based on the fire feature knowledge graph, a deep learning model is trained to form a prediction of fire risks in specific areas, times and types.

[0054] This application generates a digital model of fire accidents based on BIM, GIS and Internet of Things technologies, integrating three-dimensional real-life models, point cloud models, trace images, fire scene videos, electronic evidence and monitoring data, and realizing the comprehensive digitization of fire scene information; based on the model, the key data of the fire scene is analyzed, and the fire feature data is extracted through feature extraction, providing accurate and structured core information for subsequent analysis; a fire feature knowledge graph is constructed through fire feature data, which systematically represents fire features and their correlations, thereby improving the efficiency of data analysis and the ability of knowledge management; further based on the knowledge graph, a deep learning model is trained to realize the prediction of fire risks in specific areas, times and types, thereby enhancing the accuracy and initiative of fire prevention and control, and improving the accuracy and efficiency of fire accident investigations.

[0055] It should be noted that the execution subject of the method of this embodiment can be a computing service device with data processing, network communication and program running functions, or it can be the above-mentioned fire accident investigation device with the same or similar functions. This embodiment and the following embodiments will be described by taking the fire accident investigation device as an example.

[0056] Based on this, the first embodiment of the fire accident investigation method of this application is proposed, please refer to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the fire accident investigation method of the present application.

[0057] In this embodiment, the fire accident investigation method includes the following steps:

[0058] S1: Generate a digital model of fire accidents based on BIM, GIS and IoT technologies;

[0059] It should be noted that BIM is a three-dimensional digital expression technology that integrates multi-professional information such as architecture, structure, and equipment. By creating a digital model of a virtual building, the physical and functional characteristics of the building are visualized in three-dimensional space, which facilitates the management and analysis of the entire life cycle of the building. GIS is a system for collecting, managing, analyzing and displaying spatial geographic data, and can process various information related to geographic location. GIS technology can help associate the geographic location of a building with the surrounding environment and provide accurate geospatial analysis for the fire scene. The Internet of Things refers to the interconnection of objects in the physical world through various sensors, devices and network technologies to form an intelligent network that can sense, collect and transmit data. In the investigation of fire accidents, the Internet of Things technology can collect environmental data at the fire scene in real time, such as temperature, smoke concentration, etc., to provide dynamic data support for fire analysis.

[0060] Specifically, firstly, the detailed structural information of the building at the fire scene is obtained through BIM technology, including the specific layout of walls, doors and windows, stairs, fire-fighting facilities, etc., and a three-dimensional digital model of the building is generated. By integrating the information of various equipment inside the building, the building structure is combined with the data of fire-fighting systems such as sprinklers and alarms, providing an accurate spatial framework for subsequent fire simulation and analysis. At the same time, the geographic information data of the location where the fire occurred, such as roads, water sources, and refuge spaces around the building, are obtained through GIS technology, and these data are integrated with the BIM model. GIS data not only helps to determine the precise coordinates of the fire location, but also provides a comprehensive display of the geographical environment, allowing investigators to observe the overall picture of the fire scene in virtual space.

[0061] Furthermore, by connecting to IoT data and connecting with various sensors arranged in the building, dynamic data of the fire scene can be collected in real time, including status data of temperature sensors, smoke detectors, surveillance videos, and other safety equipment. These IoT data are associated with BIM and GIS models, integrated and dynamically displayed through a unified data base, to achieve a comprehensive and multi-dimensional reconstruction of the fire accident scene. Ultimately, the digital model of the fire accident can not only show the static building structure of the fire scene, but also reflect the dynamic changes in the development of the fire in real time, providing reliable digital support for subsequent data analysis and fire cause determination.

[0062] Through the deep integration of BIM, GIS and IoT technologies, a digital model of fire accidents was constructed, realizing the all-round digital reconstruction of the fire scene. This comprehensive information integration not only improves the accuracy and efficiency of fire accident investigation, but also brings a series of beneficial effects. First, the combination of BIM and GIS makes the spatial layout, building structure, and surrounding environment of the fire scene clear at a glance, providing investigators with accurate scene reproduction, which is convenient for quickly grasping the accident situation. Secondly, the real-time access of IoT data enhances the dynamic nature of the model, so that key data such as temperature changes and smoke diffusion during the fire can be monitored and recorded in real time, providing accurate data support in time and space for the analysis of the cause of the fire. Thirdly, through the visual display of the digital model, investigators can quickly simulate the development path of the fire and the evacuation route of personnel, thereby optimizing the rescue strategy at the fire scene. Finally, the digital model also provides data basis for the post-analysis of fire accidents, the determination of legal responsibilities, and the improvement of fire protection facilities, greatly improving the scientific nature and decision-making level of fire accident investigation and prevention work.

[0063] S2: Analyze key data of the fire scene based on the digital model, wherein the key data includes a three-dimensional real scene model, a point cloud model, a trace image, a fire scene video, electronic evidence and monitoring data;

[0064] It should be noted that fire scene video refers to real-time or recorded video data collected by fire scene monitoring equipment, cameras or mobile phones. These videos contain dynamic images of fire occurrence and spread, and are an important source of information for analyzing the fire process, identifying the source of fire and its impact. Electronic evidence refers to the electronic equipment involved in the fire scene and the data stored in it, such as electronic files, log records, communication records, etc. of computers, mobile phones, routers and other devices. These data can provide the use of related equipment and suspicious activities before and after the fire, providing potential digital evidence for fire investigation. Monitoring data refers to various types of data collected in real time by monitoring systems installed on site (such as CCTV, alarm systems, sensor networks), including video, alarm information, temperature, smoke concentration, etc. These data can reflect the environmental status and fire development process when the fire occurs, and provide important basis for fire cause analysis and on-site reconstruction. The three-dimensional real scene model is a visual model of the fire scene generated by three-dimensional scanning or photogrammetry technology, which accurately reflects the real spatial layout, structure and surface characteristics of the scene, and is an important tool for restoring and analyzing the fire scene. The point cloud model is a three-dimensional model composed of spatial data collected by laser scanning or other sensing equipment. The point cloud contains the spatial coordinates and attributes (such as reflection intensity) of each point, which is used to accurately analyze the geometric shape and damaged area of ​​the fire scene. Trace images refer to static images of the fire scene, including blackened marks, molten material marks, high-temperature cracks, etc., which record the physical and chemical changes on the surface of objects during the fire and are key evidence for analyzing the development of fire and thermal effects.

[0065] Specifically, a preliminary analysis of key data at the fire scene is conducted based on the digital model. First, the spatial layout and structure of the fire scene are evaluated through the three-dimensional real-scene model and the point cloud model, and the damaged area of ​​the building, the scope of collapse and its relationship with the fire propagation path are accurately identified. Combined with the real-scene model, the degree of thermal damage caused by the fire to objects at the scene is analyzed, such as blackened walls, cracks on the ground and the distribution of molten materials. The point cloud model is used for three-dimensional reconstruction to quantify the deformation or fracture characteristics of building components and clarify the possible location of the heat center of the fire. At the same time, trace image analysis further extracts physical trace features, such as material melting and surface carbonization caused by high temperature, to provide a basis for judging the burning intensity and propagation direction of the fire.

[0066] Furthermore, the dynamic data (fire scene video, electronic evidence and monitoring data) are deeply analyzed. The video processing algorithm is used to analyze the combustion dynamics, flame shape and smoke diffusion path in the fire scene video frame by frame, and the flame change speed, smoke diffusion direction and its interaction information with the spatial structure are extracted. Through the operation log and communication records of electronic evidence, abnormal behaviors that may exist before the fire is triggered, such as equipment failure or human operation errors, are identified. Finally, the environmental parameters such as temperature and smoke concentration in the monitoring data are analyzed in time series, and the key moments and diffusion trends of the fire are extracted in combination with the alarm signal to form an accurate description of the timeline of the fire event.

[0067] Through multi-dimensional analysis of key data at the fire scene based on digital models, this step significantly improves the comprehensiveness of information, analysis accuracy and decision-making support capabilities of fire investigation. First, the three-dimensional real-scene model and point cloud model provide an accurate reproduction of the fire scene, enabling investigators to fully grasp the scope of fire impact and propagation path. The refined spatial analysis of the point cloud model further quantifies the degree of building damage, providing a scientific basis for determining the center of the fire and the heat-affected area. Secondly, trace image analysis reveals the high temperature effect and physical damage characteristics during the fire process, helping investigators identify the physical mechanism of fire propagation; the dynamic analysis of the fire scene video restores the entire process from the occurrence to the spread of the fire through a time series, providing intuitive evidence for the analysis of the cause of the fire and the determination of responsibility. At the same time, the in-depth analysis of electronic evidence and monitoring data reveals the equipment behavior and environmental changes before and after the fire, especially the extraction of abnormal events and key parameters, which provides reliable support for the precise positioning of the cause of the fire trigger. In addition, multi-source data analysis constructs a complete timeline and spatial logical relationship of the fire event through the deep association of time and space, significantly improving the scientificity and comprehensiveness of the analysis of the cause of the fire. This multi-dimensional and multi-level analysis method not only improves the accuracy and efficiency of fire accident investigations, but also provides strong technical support for fire prevention, control and emergency response.

[0068] S3: Extract features from the analysis results to obtain fire feature data;

[0069] It should be noted that the analysis results refer to the intermediate results extracted after analyzing the key data of the fire scene (such as three-dimensional real-scene models, point cloud models, trace images, fire scene videos, electronic evidence and monitoring data), including information such as the dynamic characteristics of the fire, environmental characteristics and physical traces. Feature extraction refers to the process of extracting core information from the analysis results through algorithms or models. These core information (fire characteristic data) can reflect the cause, development path and impact of the fire, including combustion intensity, smoke diffusion, damage, etc. Fire characteristic data is the core attribute information extracted from multi-source data that can describe the occurrence and development process of the fire, including visual characteristics (such as flame shape), environmental characteristics (such as temperature changes) and physical characteristics (such as melting marks).

[0070] Specifically, the geometric analysis algorithm is used to extract the spatial characteristics of the fire scene, including the location, size, collapse shape and burnt area of ​​the damaged area of ​​the building. The three-dimensional deformation characteristics and melting parts of the damaged components are further analyzed through point cloud data to provide support for the judgment of the center location and impact range of the fire. The texture characteristics of the molten material, the distribution of high-temperature cracks and the morphology of the surface carbonization area are extracted through image processing algorithms to infer the combustion intensity and propagation path of the fire. The deep learning model is used to extract the flame morphological characteristics (such as height, color, brightness distribution) and smoke diffusion trajectory from the fire video to generate time series feature data to intuitively reflect the dynamic change process of the fire. The equipment startup / shutdown time, abnormal operation behavior and possible fault records are extracted from the equipment operation log to provide clues for locating the cause of the fire trigger. Through curve fitting and anomaly detection, key feature parameters such as temperature peak, heating rate and smoke concentration change rate are extracted from monitoring data such as temperature and smoke concentration to identify the critical moment and high-risk area of ​​the fire.

[0071] Furthermore, the features extracted from the 3D real-life model, point cloud model, trace image, fire scene video, electronic evidence and monitoring data are integrated into multi-source data. For example, spatial features (damaged area) are associated with temporal features (combustion process) to form a complete data description of the fire development process. The extracted fire feature data is normalized, denoised and time-synchronized. Normalization adjusts the feature data to a uniform scale, denoising removes noise and invalid information in multi-source data, and time synchronization ensures the consistency of different data sources on the timeline. These processing steps ensure that the feature data has good quality and availability.

[0072] By extracting the core characteristic data of fire (such as flame morphology, smoke diffusion path and temperature change, etc.), accurate quantitative data is provided for fire cause analysis, which reduces the subjectivity of manual analysis and improves the scientificity and credibility of the analysis conclusions. By integrating the spatial, temporal and behavioral characteristics in multi-source data, a complete description of the development of the fire is formed, which is convenient for analyzing the causal relationship between the characteristics, such as the association between the combustion path and smoke diffusion, and provides key support for the reconstruction of the fire process. The extracted fire characteristic data is a structured and high-quality input, which lays a solid foundation for the construction of the fire characteristic knowledge graph and the training of the deep learning model, and significantly improves the efficiency and accuracy of subsequent analysis and prediction. The features extracted from multi-source data cover the static (such as blackened marks, damaged areas) and dynamic (such as flame dynamics, smoke diffusion path) information of the fire, comprehensively reflecting the entire process of fire occurrence and development, and providing comprehensive data support for fire responsibility identification and prevention and control strategies. The automated feature extraction process avoids the inefficiency and errors in manual identification. Through systematic and intelligent feature processing, the time of fire investigation is greatly shortened and the overall investigation efficiency is improved.

[0073] S4: constructing a fire characteristic knowledge graph based on the fire characteristic data;

[0074] It should be noted that the fire characteristic knowledge graph is a fire information system built based on knowledge graph technology. It structures fire characteristic data into a network form of nodes and relationships, represents fire characteristics, events and physical evidence through nodes, and represents the causal relationship and temporal logic between fire characteristics through relationships, providing systematic knowledge management and reasoning capabilities for fire analysis.

[0075] Specifically, according to the type of fire feature data, define the node categories in the knowledge graph, such as "flame shape", "smoke diffusion path", "temperature change characteristics", "physical evidence location", etc. Nodes represent key entities or attributes in fire events. Set the relationship types between fire features, such as "causal relationship" (such as the relationship between the direction of smoke diffusion and the fire spread path), "temporal relationship" (such as the time sequence between flame height change and temperature peak) and "spatial association" (such as the spatial association between damaged areas and melting marks). Clean the fire feature data and standardize the format, unify the data expression (such as converting it into a triple format: node 1-relationship-node 2), and ensure the data quality of the input knowledge graph.

[0076] Furthermore, node information is extracted from the fire feature data and added to the knowledge graph. For example, nodes can be "flame shape-height", "smoke diffusion-path", "physical evidence location-damaged area", etc. The edges in the graph are generated according to the known associations in the fire feature data. For example, "temperature peak" is connected to "melting trace" through the "influence" relationship, indicating that high temperature causes the material to melt. The fire feature data is associated according to the time dimension to generate a time series knowledge graph. For example, the change in flame height and the change in smoke diffusion direction are connected through a time series to form a logical chain of the dynamic development of the fire. The constructed fire feature knowledge graph is stored in a graph database (such as Neo4j) to support subsequent queries, reasoning, and dynamic updates. The nodes and relationships in the graph are verified for consistency, and duplicate nodes or logically conflicting relationships are removed to ensure the integrity and accuracy of the graph. Knowledge reasoning algorithms (such as rule-based or probability-based reasoning methods) are used to mine potential knowledge from existing nodes and relationships. For example, based on the "smoke diffusion path" and "flame propagation direction", the possible ignition point and diffusion trend of the fire are inferred. The nodes, relationships and time chains of the knowledge graph are displayed through a graphical interface, making it easier for fire investigators to intuitively understand fire characteristics and their correlations.

[0077] The fire feature knowledge graph converts fire feature data into a structured network of nodes and relationships, intuitively showing the causal relationship, temporal logic and spatial distribution between fire features, making up for the problem of information fragmentation and isolation in traditional fire analysis, and significantly improving the systematic nature of fire information. The knowledge graph greatly reduces the time for manual data collation and analysis by standardizing and associating feature data, enabling investigators to quickly locate key nodes and relationships, and accelerating the process of fire cause analysis and development path restoration. Through knowledge reasoning algorithms, the knowledge graph can discover implicit associations or patterns from existing data, such as predicting the starting point, diffusion path or possible responsible party of a fire, providing a scientific basis for fire cause analysis and risk prediction. The knowledge graph displays fire features and their relationships in a visual form, making complex data relationships easy to understand and analyze, and helping investigators quickly grasp the overall picture and key factors of the fire. The knowledge graph provides high-quality structured data input for deep learning models, significantly improving the efficiency and accuracy of model training, making fire risk prediction more accurate and reliable.

[0078] S5: Based on the fire characteristic knowledge graph, a deep learning model is trained to form a prediction of fire risks in specific areas, times, and types.

[0079] It should be noted that the deep learning model is an algorithm model built on artificial neural networks, which is good at learning complex features and patterns from large amounts of data. Through training, the deep learning model can form the ability to predict fire risks, such as identifying high-risk areas for fires, possible fire ignition times and fire types. Fire risk prediction refers to the use of deep learning models to predict the possibility, potential causes and scope of fire, and provide support for fire prevention and control and emergency management.

[0080] Specifically, extract structured data from the fire feature knowledge graph and convert it into an input form suitable for deep learning models. For example, embed nodes and their relationships into feature vectors (such as graph embedding methods) to represent the associations and patterns between fire features. According to historical fire cases, add labels to the data (such as fire area risk level, possible fire time, fire type, etc.) to provide a labeled data set for supervised learning of the model. Divide the fire feature data set into training set, validation set, and test set to ensure that the model can be effectively evaluated and generalized. Select a deep learning model architecture suitable for fire risk prediction tasks, such as a model based on graph neural network (GNN) to process knowledge graph data, or combine convolutional neural network (CNN) and time series models (such as LSTM) to analyze dynamic features. Input the training data into the model, calculate the predicted value through forward propagation, compare it with the actual value (label data), calculate the error (loss function), and update the model weights through the back-propagation algorithm to continuously optimize the model. Adjust hyperparameters such as learning rate, network depth, and regularization parameter during training to ensure that the model achieves optimal performance.

[0081] Furthermore, the trained model is cross-validated to evaluate the performance of the model on unseen data to ensure its generalization ability. The accuracy, recall rate, F1 score and other indicators of the model are verified on the test set to ensure that the model can accurately predict fire risks. According to the verification and test results, the model structure or input features are continuously adjusted to improve the prediction ability. Based on the model prediction output, high-risk points for fire in specific areas are identified, such as specific floors or areas in a building. Based on historical and real-time data, the possibility of fire in a specific time period is predicted to support emergency preparedness during high-risk periods. Based on the fire characteristic pattern, possible fire types (such as electrical fires, chemical fires, combustible fires, etc.) are predicted to provide targeted suggestions for prevention and control measures.

[0082] By training a deep learning model based on the knowledge graph of fire characteristics, this step can learn the complex associations and development patterns between fire characteristics, thereby achieving accurate prediction of high-risk areas, time and types of fires, significantly improving the pertinence and scientificity of fire prevention and control. The model can predict risks before a fire occurs, provide early warning information for fire managers, help deploy resources in advance, optimize the layout of fire facilities, and shift from passive response to active prevention and control to reduce the incidence of fires. By identifying high-risk areas and high-risk time periods for fires, the model provides a scientific basis for the optimal configuration of fire resources (such as fire-fighting equipment and emergency personnel) and improves the efficiency of emergency response. The deep learning model automatically analyzes fire characteristic data, replacing traditional manual experience judgment, which not only improves the accuracy of prediction, but also greatly reduces the workload of manual analysis, making fire prediction more intelligent and efficient. By learning the patterns of historical fire data and knowledge graphs, the model can explore the potential laws and characteristics of fire occurrence, providing new data support and method tools for future fire risk management research. The output results of the model can support the overall strategic planning of fire safety, such as optimizing building fire protection design, formulating targeted fire prevention measures and emergency evacuation plans, and improving the overall fire resistance of the region.

[0083] This embodiment generates a digital model of a fire accident based on BIM, GIS and Internet of Things technologies; based on the digital model, analyzes key data of the fire scene, including three-dimensional real-scene models, point cloud models, trace images, fire scene videos, electronic evidence and monitoring data; extracts features from the analysis results to obtain fire feature data; based on the fire feature data, constructs a fire feature knowledge graph; based on the fire feature knowledge graph, trains a deep learning model to form a prediction of fire risks in specific areas, times and types. This embodiment generates a digital model of fire accidents based on BIM, GIS and Internet of Things technologies, integrates three-dimensional real-scene models, point cloud models, trace images, fire scene videos, electronic evidence and monitoring data, and realizes the comprehensive digitization of fire scene information; based on the model, the key data of the fire scene is analyzed, and the fire feature data is extracted through feature extraction, providing accurate and structured core information for subsequent analysis; a fire feature knowledge graph is constructed through the fire feature data, and the fire features and their correlations are systematically represented, thereby improving the efficiency of data analysis and the ability of knowledge management; further based on the knowledge graph, a deep learning model is trained to realize the prediction of fire risks in specific areas, times and types, thereby enhancing the precision and initiative of fire prevention and control, and improving the accuracy and efficiency of fire accident investigations.

[0084] Based on the above first embodiment, a second embodiment of the fire accident investigation method of the present application is proposed. Figure 2 , Figure 2 This is a schematic diagram of a sub-process in the second embodiment of the fire accident investigation method of the present application.

[0085] like Figure 2 As shown, in this embodiment, step S1 includes:

[0086] S11: Acquire basic spatial data from the CIM data base, wherein the basic spatial data includes building unit data, GIS base map, and thematic spatial data;

[0087] S12: Digitally process the basic spatial data to obtain GIS data;

[0088] S13: Build a BIM model based on the BIM data of the on-site building;

[0089] S14: superimposing the GIS data and the BIM model to obtain a preliminary model;

[0090] S15: Integrate the preliminary model with the IoT data, and based on a preset data base, associate the GIS data with the IoT data to obtain a digital model of the fire accident.

[0091] It should be noted that the CIM data base is a city-level information model database that integrates spatial data, building information, geographic information and related thematic data of urban infrastructure to support urban management, planning and emergency response. CIM provides comprehensive basic spatial data for fire accident investigations, helping to accurately restore accident scenes. Building monomer data refers to the detailed data of a single building contained in the CIM data base, including information such as the structure, size, material, and functional zoning of the building, which is the core data for building digital building models in fire accident investigations. The GIS base map is a basic map based on the geographic information system, which contains spatial data of elements such as terrain, roads, water systems, buildings, and greening. The GIS base map provides background data support for the geographical environment analysis and spatial positioning of fire accidents. Thematic spatial data refers to geographic spatial data related to specific topics, such as the layout of fire-fighting facilities, the location of water sources, and evacuation channels, which are important references for emergency analysis and decision-making of fire accidents. BIM is a technology for three-dimensional digital expression of buildings, covering the geometric shape, spatial relationship, material and component information of buildings, and providing digital support for building design, construction and management. In fire investigations, BIM provides detailed building structure information, which facilitates the analysis of the propagation path and impact range of fire in buildings.

[0092] Specifically, the basic spatial data of the fire scene is first extracted from the CIM data base, including building data, GIS base map and thematic spatial data. These data contain detailed structural information of the building, the surrounding geographical environment, road network and layout of firefighting facilities, laying the foundation for the digital modeling of fire accidents. The extracted basic spatial data is converted into GIS data through digital processing, so that it can be visualized and analyzed on a digital platform. The generation of GIS data includes the precise location of the building, its spatial distribution, and its relationship with the surrounding environment, providing geographical support for further reconstruction of the fire scene.

[0093] Furthermore, a BIM model is constructed based on the BIM data of the on-site building. The BIM model generates a three-dimensional digital model that reflects the actual building situation by integrating the building's three-dimensional structural information, internal space layout, and detailed location of fire-fighting facilities. Then, the processed GIS data is superimposed on the BIM model to generate a preliminary model. This preliminary model combines the internal structure of the building and the external geographical environment, providing a complete spatial framework for the scene reconstruction of the fire accident. Finally, the preliminary model is integrated with the IoT data, and the real-time monitored sensor data (such as temperature, smoke concentration, etc.) is associated with the spatial information in the model. Through the preset data base, the GIS data is dynamically associated with the IoT data. Finally, the digital model of the fire accident formed can not only statically display the building and geographic information, but also reflect the dynamic changes of the fire scene in real time.

[0094] By obtaining basic spatial data from the CIM data base and integrating it with the BIM model and IoT data, a digital model of fire accidents is formed, which brings significant beneficial effects. First, the digital model integrates detailed information about the internal and external environment of the building, allowing investigators to intuitively see the full picture of the fire scene, greatly improving the accuracy of fire scene reproduction. The addition of the BIM model provides a deep understanding of the internal structure of the building, helping to identify the propagation path of the fire in the building and the possible impact area. Through the superposition of GIS data, the model can show the spatial relationship between the building and its surrounding environment, such as escape routes, the location of fire-fighting facilities, etc., providing a scientific basis for the formulation of fire rescue strategies. Secondly, the dynamic access of IoT data enables the model to reflect the environmental status and development trend when the fire occurs in real time. Information such as temperature, smoke, and equipment status can help investigators quickly identify the starting point of the fire, the direction of fire spread, and the current dangerous area through the dynamic display of the model. This comprehensive display of multi-source data improves the comprehensiveness and accuracy of fire scene analysis. In addition, the digital model provides a reliable basis for the simulation and prediction of fire accidents. By simulating the propagation process of fire inside and outside the building, it optimizes firefighting and rescue decisions and enhances accident response capabilities. The application of this model also supports the loss assessment and liability tracing after the accident, provides detailed scene data, and provides objective evidence for fire liability identification and legal dispute resolution. By improving the efficiency and accuracy of fire scene analysis, this digital model has greatly improved the traditional fire investigation method and provided advanced technical support for fire safety management, fire prevention and emergency response.

[0095] Based on the first embodiment above, before step S3, the method further includes:

[0096] S3a: Optimizing the analysis results to obtain optimized data, wherein the optimization processing includes one or more of transcoding processing, format parsing and recovery processing, and cleaning and denoising processing;

[0097] S3b: Synchronizing the optimized data to the preset data base;

[0098] S3c: Generate a fire scene report based on the preset data base.

[0099] It should be noted that transcoding is the process of converting, compressing or decoding the format of video files so that the video can adapt to the requirements of different playback devices or analysis software. In fire investigation, transcoding can optimize the quality and compatibility of fire scene videos for further analysis and storage. Format parsing refers to the process of converting electronic evidence (such as computer files, communication records, etc.) from the original format to a readable and analyzable format. Recovery processing is the recovery of damaged or deleted data to extract important evidence that is hidden or lost. This process ensures the availability and integrity of electronic evidence in fire investigations. Data cleaning refers to the process of cleaning up erroneous, incomplete or invalid information in monitoring data; denoising is the process of extracting key signals and data by filtering out irrelevant or redundant information. This process improves the quality of monitoring data and makes the analysis results more accurate. The preset data base is a unified platform for storing and managing various data collected during fire investigations, including standardized structures and formats, which facilitates the synchronization, integration and analysis of multi-source data. A fire scene report is a detailed document generated based on data processed and analyzed during the fire investigation. It records the time, location, cause, development process, loss and responsibility analysis of the fire. It is an important outcome document of the fire investigation.

[0100] Specifically, the multimedia data collected at the fire scene (such as fire scene videos and surveillance videos) are converted into formats, and videos from different sources are unified into analyzable formats (such as MP4, AVI, etc.). The picture clarity is improved through resolution optimization to ensure data integrity during the video analysis process. Electronic evidence (such as equipment logs, communication records, etc.) is parsed and converted from original formats (such as log files, encrypted data, etc.) to standardized formats (such as CSV, JSON), and data recovery tools are used to extract damaged or lost key data, such as deleted communication records or damaged equipment operation logs. Sensor data (such as temperature, smoke concentration) and point cloud data are cleaned and denoised to filter out outliers, redundant data and interference signals, extract meaningful environmental features and spatial information, and ensure data accuracy.

[0101] Furthermore, the optimized fire data (including video data, electronic evidence data, sensor monitoring data, etc.) is uploaded to the preset data base for unified management and storage, and provides high-quality input for subsequent feature extraction and analysis. The data base ensures the consistency and availability of multi-source data. Based on the optimized data, a fire scene report is automatically generated. The report includes the time of fire occurrence, location of the fire point, flame spread path, damage scope and on-site evidence analysis, etc., providing a detailed description and preliminary conclusion of the fire, and providing data support for fire cause analysis, responsibility identification and emergency response decision-making.

[0102] Optimization processing significantly improves the integrity, accuracy and consistency of fire data through transcoding, format parsing and denoising, solves problems such as data format incompatibility, noise interference and missing, and lays a solid foundation for subsequent feature extraction and analysis. The optimized data is synchronized to the preset data base, centrally stored and managed, and forms a structured and standardized fire information library. This data integration method reduces the phenomenon of information islands, facilitates the subsequent analysis module to call multi-source data, and improves data utilization efficiency. By automatically generating fire scene reports based on optimized data, this step shortens the report generation time and ensures the comprehensiveness and accuracy of the report content. The analysis conclusions and data support in the report provide a scientific basis for fire investigation and emergency decision-making. The optimized data is standardized and cleaned to become high-quality input, which can be directly used for fire feature extraction, knowledge graph construction and deep learning model training, reducing redundant processing work in the intermediate links and significantly improving the efficiency of subsequent fire analysis and prediction. Through the optimized management and report generation of fire scene data, investigators can quickly obtain accurate and reliable data support, greatly reducing the errors and inefficiencies caused by insufficient data or scattered information in traditional fire investigations.

[0103] This embodiment generates a digital model of a fire accident based on BIM, GIS and Internet of Things technologies; based on the digital model, analyzes key data of the fire scene, including three-dimensional real-scene models, point cloud models, trace images, fire scene videos, electronic evidence and monitoring data; extracts features from the analysis results to obtain fire feature data; based on the fire feature data, constructs a fire feature knowledge graph; based on the fire feature knowledge graph, trains a deep learning model to form a prediction of fire risks in specific areas, times and types. This embodiment generates a digital model of fire accidents based on BIM, GIS and Internet of Things technologies, integrates three-dimensional real-scene models, point cloud models, trace images, fire scene videos, electronic evidence and monitoring data, and realizes the comprehensive digitization of fire scene information; based on the model, the key data of the fire scene is analyzed, and the fire feature data is extracted through feature extraction, providing accurate and structured core information for subsequent analysis; a fire feature knowledge graph is constructed through the fire feature data, and the fire features and their correlations are systematically represented, thereby improving the efficiency of data analysis and the ability of knowledge management; further based on the knowledge graph, a deep learning model is trained to realize the prediction of fire risks in specific areas, times and types, thereby enhancing the precision and initiative of fire prevention and control, and improving the accuracy and efficiency of fire accident investigations.

[0104] Based on the above second embodiment, a third embodiment of the fire accident investigation method of the present application is proposed. Figure 3 , Figure 3 This is a schematic diagram of a sub-process in the third embodiment of the fire accident investigation method of the present application.

[0105] In this embodiment, step S3 includes:

[0106] S31: extracting fire visual data, key person data, equipment operation data, damage feature data, and fire trace data from the analysis results based on a preset feature extraction strategy;

[0107] S32: performing curve fitting and anomaly detection on the environmental features in the analysis results to extract key monitoring data;

[0108] S33: Standardize the fire visual data, the key person data, the equipment operation data, the damage feature data, the fire trace data and the key monitoring data to obtain the fire feature data, wherein the standardization processing includes one or more of normalization processing, data denoising processing and time synchronization processing.

[0109] It should be noted that fire visual data is visual feature information related to fire dynamics extracted from fire scene videos and images, including flame morphology, smoke diffusion path and combustion dynamics. Key person data is the identity and behavior data of personnel related to fire occurrence or rescue identified from fire scene videos or electronic evidence, including information such as the time, location and activity trajectory of personnel. Equipment operation data is the operation records, status data and abnormal behavior data of fire-related equipment, which comes from electronic evidence and the operation log of the monitoring system, and can reflect the equipment activities before and after the fire. Damage feature data is the fire scene damage information extracted through three-dimensional real-life models, point cloud models and trace images, including damaged areas of buildings, melting traces, high-temperature cracks, etc. Fire trace data is physical trace data reflecting the occurrence and development process of fire, such as the morphology of molten materials, the distribution of carbonized areas, smoke traces, etc. Environmental feature data is the environmental characteristics of the fire scene collected by the monitoring system sensors, including temperature change curves, smoke concentration changes, and trigger records of alarm signals. Key monitoring data are key parameters extracted from environmental characteristics through curve fitting and anomaly detection, including temperature peak, heating rate, smoke concentration change rate, alarm trigger time, etc. Standardization is the process of processing multi-source data to ensure the consistency of data in terms of dimension, time and format, including normalization, data denoising and time synchronization.

[0110] Specifically, deep learning models (such as convolutional neural networks) are used to analyze fire scene videos and images to extract the morphological characteristics of flames (height, color, brightness changes, etc.), smoke diffusion paths (direction, range), and combustion dynamics (speed, changes in burning areas). These data reflect the dynamic evolution of fires. Face recognition and behavior analysis algorithms are used to identify the activity trajectories of people related to fires from fire scene videos, and to extract the identities, appearance times, and behavioral characteristics of people (such as whether they participated in arson, whether they evacuated in time, etc.). Equipment operating status and abnormal behaviors are extracted from the operation logs of electronic evidence and monitoring system records, such as the power-off time of electrical equipment, overload records, and the start-up time of fire-fighting equipment. Based on point cloud models and three-dimensional real-life models, the degree of damage to building components caused by fire is analyzed, the location of damaged areas, the morphology of melting traces, and the distribution of cracks are extracted, and the physical damage caused by fire is quantified. By analyzing the trace images, the distribution characteristics of carbonized areas, blackened traces, and melting residues are extracted, and the combustion intensity and heat-affected range of the fire are inferred in combination with material properties.

[0111] Furthermore, the environmental characteristic data (such as temperature change and smoke concentration) are analyzed in time series. The temperature rise trend and concentration change rate are extracted through curve fitting to reveal the environmental change pattern before and after the fire. The monitoring data are analyzed by using anomaly detection algorithms to identify key events such as abnormal temperature rise points, smoke concentration mutation points, and alarm triggering time. These key monitoring data directly reflect the triggering points and development trends of fires and provide basic information for fire cause analysis. The extracted fire characteristic data are normalized to adjust the data of different dimensions (such as temperature, concentration, and spatial position) to a unified range to facilitate subsequent unified analysis and comparison. Filtering and noise reduction algorithms are used to remove noise signals in the fire characteristic data, such as false alarm data of monitoring sensors or interference information in videos, to ensure the accuracy and reliability of the data. Multi-source data are time-aligned according to timestamps, and fire visual data, environmental monitoring data, and equipment operation data are integrated into a unified time axis to form a complete time series of fire development, which is convenient for subsequent dynamic analysis.

[0112] By extracting fire visual data, key person data, equipment operation data, damage feature data, fire trace data, and key monitoring data, the fire scene static and dynamic information is fully covered, providing multi-dimensional data support for the reconstruction of the fire development process. The extracted feature data is standardized (including normalization, denoising and time synchronization) to eliminate the inconsistency of multi-source data in terms of dimension, noise and time, ensuring the accuracy and consistency of subsequent analysis and modeling. By combining key monitoring data (such as temperature peak, smoke diffusion path) with fire trace data (such as damaged area, melting trace), the cause, diffusion path and impact range of the fire can be effectively inferred, providing a scientific basis for the determination of fire responsibility. The standardized fire feature data has high quality and consistency, and can be directly used as input for knowledge graph construction and deep learning model training, significantly reducing the redundant links of data processing, and improving modeling efficiency and prediction performance. By automatically extracting and processing multi-source data, an efficient process from fire scene data analysis to feature extraction is achieved, reducing manual intervention and analysis time, and improving the overall efficiency and scientificity of fire accident investigation.

[0113] Based on the above second embodiment, in this embodiment, step S4 includes:

[0114] S41: preprocessing the fire characteristic data to obtain preprocessed data, wherein the preprocessing includes one or more of data classification, data cleaning, and format conversion;

[0115] S42: extracting data from the preprocessed data based on a preset graph grid structure, and generating graph nodes based on the extracted data;

[0116] S43: Determine the relationship between the graph nodes based on the causal relationship and the correlation relationship between the extracted data;

[0117] S44: Add timestamp information to the graph nodes, and construct the fire characteristic knowledge graph based on the relationship between the graph nodes.

[0118] It should be noted that the preprocessed data is standardized data generated after the fire characteristic data is classified, cleaned and formatted, ensuring the integrity, consistency and availability of the input data. The graph grid structure is the framework structure of the knowledge graph, which defines the organization of nodes in the graph and the type of relationship between nodes, and provides basic rules for constructing the fire characteristic knowledge graph. The graph node is the basic unit in the knowledge graph, representing the specific entity or attribute in the fire characteristic (such as flame shape, melting trace, smoke diffusion path, etc.). Causal relationship is the description of the causal relationship between fire characteristic data, such as the causal relationship between "smoke diffusion path" and "flame spread direction". The association relationship refers to the mutual correlation between fire characteristic data, such as the spatial association between "melting trace" and "high temperature area". Timestamp information is the time dimension information added to the graph node, indicating the time when a specific fire characteristic occurs, and is used to reconstruct the timeline of the fire development process.

[0119] Specifically, fire feature data is classified according to its type, such as dynamic data (flame shape, smoke diffusion), static data (damage characteristics, fire traces) and monitoring data (ambient temperature, alarm records, etc.) to facilitate subsequent processing and organization. Redundant, invalid or abnormal data is removed through cleaning algorithms, such as eliminating false alarm records in monitoring data and invalid frames in video data, to ensure the quality of feature data. Data from different sources (such as video, point cloud, sensor data) are converted into a unified format, such as converting image data into vector features, or parsing log files into a structured database format, to provide standardized input for knowledge graph construction. Based on the preset graph grid structure, core information is extracted from the preprocessed data. For example, key features such as "flame height" and "color distribution" are extracted from flame shape data, and "damaged area location" and "crack length" are extracted from damage feature data. Graph nodes are generated based on the extracted data, and each node represents an entity or attribute in the fire feature, such as "flame shape-height" and "temperature peak". Nodes need to be classified and stored according to the definition of the graph grid structure to ensure semantic consistency between nodes. Add attribute information (such as node type, value range, data source) to each node to further describe the characteristics of the node and its context.

[0120] Furthermore, the causal inference mechanism between data is used to identify the causal relationship between feature data. For example, the causal relationship that "high temperature area" leads to "melting trace" is inferred, and the logical association that "smoke diffusion path" promotes "flame spread". Based on the analysis of space, time or data logic, the association relationship between feature data is determined, such as the spatial overlapping association between "damaged area" and "smoke trace", and the temporal association between "alarm trigger time" and "temperature peak time". Weights are added to each relationship to reflect the importance or reliability of different relationships. For example, causal relationships with high confidence are given higher weights. Timestamp information is added to each graph node to record the time when the feature data corresponding to the node occurs. For example, the time corresponding to "smoke diffusion direction-north" is the 5th minute after the fire occurs. Timestamp information enables the knowledge graph to have dynamic representation capabilities. Based on the graph nodes and the causal and association relationships between them, network connections between nodes are constructed. For example, "flame height" is connected to "smoke concentration change" through causal relationships, and "damaged area" is connected to "melting trace" through spatial relationships. The constructed knowledge graph is stored in a graph database (such as Neo4j) to form a structured fire feature knowledge graph that supports efficient query, reasoning and dynamic update.

[0121] Through preprocessing and graph node generation, this step classifies, cleans and standardizes the fire feature data to form structured knowledge graph nodes and relationship networks, realizing the systematic organization and management of fire feature information and avoiding the fragmentation and redundancy problems in traditional data analysis. By establishing causal and correlation relationships between graph nodes, the knowledge graph intuitively displays the logical associations and development paths between fire features, enabling investigators to quickly identify key factors and potential risks, significantly improving the efficiency and accuracy of fire cause analysis. After adding timestamp information, the knowledge graph can dynamically represent the time series logic of fire development, support the dynamic reproduction of the entire process from ignition to spread, and provide basic data support for the analysis of fire propagation paths and emergency response optimization. The knowledge graph is based on graph database storage and supports fast query and reasoning analysis of complex relationships. For example, investigators can use reasoning algorithms to predict the possible spread direction of a fire or identify possible ignition points, thereby improving the scientificity and accuracy of fire accident investigations. The constructed fire feature knowledge graph can provide high-quality input for deep learning models. By learning the node and relationship structure of the knowledge graph, the model can accurately predict fire risks and provide a scientific basis for the formulation of fire prevention and control strategies.

[0122] Based on the above second embodiment, in this embodiment, after step S44, the following is further included:

[0123] S44a: performing deduplication processing on the fire characteristic knowledge graph, wherein the deduplication processing includes one or more of consistency checking, redundant relationship removal, and knowledge reasoning;

[0124] S44b: storing the deduplicated fire characteristic knowledge graph into a preset graph database, and displaying it through a graphical interface;

[0125] S44c: Acquire new data in real time, and update the fire characteristic knowledge graph based on the new data.

[0126] It should be noted that deduplication is to check the nodes and relationships in the fire characteristics knowledge graph, remove duplicate, redundant or conflicting information, and ensure the uniqueness, accuracy and consistency of the graph structure and data. Graph database is a database system used to store and manage knowledge graphs. It supports efficient node and relationship queries. Commonly used databases include Neo4j and TigerGraph. The graphical interface refers to the interface that visually displays the knowledge graph. It intuitively presents fire characteristics and their associations in the form of graphical nodes and edges, which is convenient for users to understand and interact. Real-time data update is to monitor the newly added data stream and dynamically supplement the fire characteristics knowledge graph to ensure that the graph can reflect the latest fire characteristics and relationships.

[0127] Specifically, semantic verification is performed on all nodes and relationships in the knowledge graph to check whether the definitions of node names, attributes, and relationships conform to the preset rules, and nodes with naming conflicts or repeated attribute values ​​are removed. For example, for the "flame height" node that appears multiple times, the latest or most accurate version is retained by verifying its numerical value and timestamp information. The relationship edges in the graph are filtered to remove duplicate edges between two nodes. For example, if there are multiple identical "influence" relationships between the nodes "flame height" and "smoke diffusion direction", the edge with the highest weight is retained. Invalid or weakly associated relationships, such as edges with low connectivity and insufficient weight, are deleted to optimize the graph structure and improve query efficiency. Inference rules are used to generate implicit knowledge from existing nodes and relationships. For example, the "possible ignition point" is inferred based on the relationship between "damaged area" and "temperature peak", or the key development stages of a fire are identified through time series analysis. The inference results are added to the knowledge graph in the form of new nodes and relationships to enrich the graph content.

[0128] Furthermore, the deduplicated fire feature knowledge graph is stored in a preset graph database (such as Neo4j). The graph database supports efficient storage and retrieval of graph structure data, which is convenient for subsequent analysis and query operations. A unique identifier (ID) is added to each node and relationship to ensure data integrity and consistency during the graph storage process. The knowledge graph is displayed in a graphical interface using visualization tools (such as Gephi or Cytoscape). Each node is represented by a dot, and the relationship is represented by a line. The size or color of the node can be dynamically adjusted according to the attribute (such as flame intensity, severity of damage). The graphical interface provides interactive functions, such as clicking on a node to display attribute details, or filtering nodes and relationships of a specific time period, area or type, so that users can quickly understand and operate the graph. By connecting with IoT devices or monitoring systems, new data (such as sensor monitoring data, fire scene video analysis results, etc.) is continuously received. The new data contains new fire features (such as the latest temperature peak, smoke diffusion direction, etc.) and related information. The new data is compared with the existing knowledge graph. If the data contains new feature information, the corresponding node or relationship is generated and updated to the graph. If the new data modifies the information of existing nodes or relationships (such as temperature data changes), the corresponding nodes or relationships will be replaced or supplemented. The update process of the knowledge graph is version controlled and historical versions are retained to facilitate the backtracking and analysis of the evolution of fire knowledge.

[0129] Through consistency checking and redundant relationship removal, this step eliminates duplicate or conflicting information in the graph, making the knowledge graph structure more optimized, nodes and relationships more accurate, and improving overall reliability and scientificity. The deduplicated knowledge graph is more streamlined and has no redundant relationships, which significantly improves the query efficiency of the graph database, supports fast retrieval and complex relationship reasoning, and provides a convenient tool for fire investigation and analysis. Real-time acquisition of new data and updating of the graph enables the knowledge graph to dynamically reflect the latest fire feature information, providing accurate timeliness support for subsequent fire analysis, risk prediction, and emergency response. By displaying the knowledge graph through a graphical interface, users can intuitively see the association and development logic between fire features, making it easier to quickly grasp the core features and key nodes of the fire. The real-time updated knowledge graph combined with the new data can continuously improve the causal analysis and risk prediction capabilities of fire features, and provide more scientific guidance for fire management and prevention.

[0130] This embodiment generates a digital model of a fire accident based on BIM, GIS and Internet of Things technologies; based on the digital model, analyzes key data of the fire scene, including three-dimensional real-scene models, point cloud models, trace images, fire scene videos, electronic evidence and monitoring data; extracts features from the analysis results to obtain fire feature data; based on the fire feature data, constructs a fire feature knowledge graph; based on the fire feature knowledge graph, trains a deep learning model to form a prediction of fire risks in specific areas, times and types. This embodiment generates a digital model of fire accidents based on BIM, GIS and Internet of Things technologies, integrates three-dimensional real-scene models, point cloud models, trace images, fire scene videos, electronic evidence and monitoring data, and realizes the comprehensive digitization of fire scene information; based on the model, the key data of the fire scene is analyzed, and the fire feature data is extracted through feature extraction, providing accurate and structured core information for subsequent analysis; a fire feature knowledge graph is constructed through the fire feature data, and the fire features and their correlations are systematically represented, thereby improving the efficiency of data analysis and the ability of knowledge management; further based on the knowledge graph, a deep learning model is trained to realize the prediction of fire risks in specific areas, times and types, thereby enhancing the precision and initiative of fire prevention and control, and improving the accuracy and efficiency of fire accident investigations.

[0131] For example, in order to help understand the technical concept or technical principle of the fire accident investigation method of the above embodiment, please refer to Figure 4 , Figure 4 This is a functional framework diagram of fire accident investigation in an embodiment of the fire accident investigation method of the present application.

[0132] like Figure 4 As shown, a large database and decision-making analysis function module are built around the core evidence chain of fire accident investigation. The various data intersections of the CIM city digital base are utilized to access basic data such as the IoT data of the building fire protection Internet of Things, electronic data, power system data, gas system data, and video surveillance data. The data are cross-checked, and the data of the investigation records are accessed at the same time. The natural semantic model is used to extract key data to assist in generating conclusions from the fire scene investigation.

[0133] The present application also provides a fire accident investigation device, please refer to Figure 5 , Figure 5 This is a schematic diagram of the module structure of a fire accident investigation device according to an embodiment of the present application, wherein the fire accident investigation device comprises:

[0134] A digital model generation module 501 is used to generate a digital model of a fire accident based on BIM, GIS and Internet of Things technologies;

[0135] A data analysis module 502 is used to analyze key data of the fire scene based on the digital model, wherein the key data includes a three-dimensional real scene model, a point cloud model, a trace image, a fire scene video, electronic evidence and monitoring data;

[0136] The feature extraction module 503 is used to extract features from the analysis results to obtain fire feature data;

[0137] A graph construction module 504, used to construct a fire characteristic knowledge graph based on the fire characteristic data;

[0138] The target module 505 is used to train a deep learning model based on the fire characteristic knowledge graph to form a prediction of the fire risk of a specific area, time, and type.

[0139] The fire accident investigation device provided in the embodiment of the present application adopts the fire accident investigation method in the above embodiment, which can solve the technical problem of how to improve the accuracy and efficiency of fire accident investigation. Compared with the prior art, the beneficial effects of the fire accident investigation device provided in the embodiment of the present application are the same as the beneficial effects of the fire accident investigation method provided in the above embodiment, and the other technical features in the fire accident investigation device are the same as the features disclosed in the above embodiment method, which will not be repeated here.

[0140] The present application provides a fire accident investigation device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the fire accident investigation method in the above-mentioned embodiment.

[0141] Reference below Figure 6 , Figure 6 It is a schematic diagram of the device structure of the hardware operating environment involved in the fire accident investigation method in the embodiment of the present application, which shows a schematic diagram of the structure of the fire accident investigation device suitable for implementing the embodiment of the present application. Figure 6 The fire accident investigation equipment shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0142] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. When the computer program is executed by the processing device 1001, the above functions defined in the method of the embodiments disclosed in the present application are executed.

[0143] The fire accident investigation equipment provided by the present application adopts the fire accident investigation method in the above embodiment, which can solve the technical problem of how to improve the accuracy and efficiency of fire accident investigation. Compared with the prior art, the beneficial effects of the fire accident investigation equipment provided by the present application are the same as the beneficial effects of the fire accident investigation method provided by the above embodiment, and the other technical features in the fire accident investigation equipment are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0144] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0145] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0146] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, and the computer-readable program instructions are used to execute the fire accident investigation method in the above-mentioned embodiment.

[0147] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by the fire accident investigation device, the fire accident investigation device generates a digital model of the fire accident based on BIM, GIS and Internet of Things technologies; analyzes the key data of the fire scene based on the digital model, and the key data includes a three-dimensional real scene model, a point cloud model, a trace image, a fire scene video, electronic evidence and monitoring data; extracts features from the analysis results to obtain fire feature data; constructs a fire feature knowledge graph based on the fire feature data; trains a deep learning model based on the fire feature knowledge graph to form a prediction of fire risks in a specific area, time and type. The computer program code for performing the operations of the present application can be written in one or more programming languages ​​or a combination thereof, and the programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" language or similar programming languages. The program code can be executed completely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or it can be connected to an external computer (for example, through the Internet using an Internet service provider).

[0148] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0149] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0150] The readable storage medium provided in this application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned fire accident investigation method, and can solve the technical problem of how to improve the accuracy and efficiency of fire accident investigation. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as the beneficial effects of the fire accident investigation method provided in the above-mentioned embodiment, and will not be repeated here.

[0151] An embodiment of the present application provides a computer program product, including a computer program, which implements the steps of the fire accident investigation method as described above when the computer program is executed by a processor.

[0152] The computer program product provided in this application can solve the technical problem of how to improve the accuracy and efficiency of fire accident investigation. Compared with the prior art, the beneficial effects of the computer program product provided in the embodiment of this application are the same as the beneficial effects of the fire accident investigation method provided in the above embodiment, which will not be repeated here.

[0153] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent processing scope of the present application.

Claims

1. A fire accident investigation method, characterized in that: The method comprises: Generate digital models of fire accidents based on BIM, GIS and IoT technologies; Based on the digital model, analyzing the key data of the fire scene, the key data includes a three-dimensional real scene model, a point cloud model, a trace image, a fire scene video, electronic evidence and monitoring data; Extract features from the analysis results to obtain fire feature data; Based on the fire characteristic data, construct a fire characteristic knowledge graph; Based on the fire characteristic knowledge graph, a deep learning model is trained to form a prediction of fire risks in specific areas, times, and types.

2. The method according to claim 1, characterized in that The steps of generating a digital model of a fire accident based on BIM, GIS and IoT technologies include: Acquire basic spatial data from the CIM data base, wherein the basic spatial data includes building unit data, GIS base map, and thematic spatial data; Digitally processing the basic spatial data to obtain GIS data; Build a BIM model based on the BIM data of the on-site building; Superimposing the GIS data with the BIM model to obtain a preliminary model; The preliminary model is integrated with the IoT data, and based on a preset data base, the GIS data is associated with the IoT data to obtain a digital model of the fire accident.

3. The method according to claim 2, characterized in that Before the step of extracting features from the analysis results to obtain fire feature data, the method further includes: Optimizing the analysis results to obtain optimized data, wherein the optimization processing includes one or more of transcoding processing, format parsing and recovery processing, and cleaning and denoising processing; Synchronizing the optimized data to the preset data base; Based on the preset data base, a fire scene report is generated.

4. The method according to claim 1, characterized in that The step of extracting features from the analysis results to obtain fire feature data includes: Extracting fire visual data, key person data, equipment operation data, damage feature data, and fire trace data from the analysis results based on a preset feature extraction strategy; Performing curve fitting and anomaly detection on the environmental characteristics in the analysis results to extract key monitoring data; The fire visual data, the key person data, the equipment operation data, the damage feature data, the fire trace data and the key monitoring data are standardized to obtain the fire feature data, wherein the standardization processing includes one or more of normalization processing, data denoising processing and time synchronization processing.

5. The method according to claim 1, characterized in that The step of constructing a fire characteristic knowledge graph based on the fire characteristic data includes: Preprocessing the fire characteristic data to obtain preprocessed data, wherein the preprocessing includes one or more of data classification, data cleaning, and format conversion; Based on a preset graph grid structure, extracting data from the preprocessed data, and generating graph nodes based on the extracted data; Determine the relationship between the graph nodes based on the causal relationship and the correlation relationship between the extracted data; Timestamp information is added to the graph nodes, and the fire characteristic knowledge graph is constructed based on the relationship between the graph nodes.

6. The method according to claim 5, characterized in that After the step of adding timestamp information to the graph nodes and constructing the fire characteristic knowledge graph based on the relationship between the graph nodes, the method further includes: Performing deduplication processing on the fire characteristic knowledge graph, wherein the deduplication processing includes one or more of consistency checking, redundant relationship removal, and knowledge reasoning; The deduplicated fire characteristic knowledge graph is stored in a preset graph database and displayed through a graphical interface; Acquire new data in real time, and update the fire characteristic knowledge graph based on the new data.

7. A fire accident investigation device, characterized in that: The device comprises: Digital model generation module, used to generate digital models of fire accidents based on BIM, GIS and IoT technologies; A data analysis module, used to analyze key data of the fire scene based on the digital model, wherein the key data includes a three-dimensional real scene model, a point cloud model, a trace image, a fire scene video, electronic evidence and monitoring data; A feature extraction module is used to extract features from the analysis results to obtain fire feature data; A graph construction module, used to construct a fire characteristic knowledge graph based on the fire characteristic data; The target module is used to train a deep learning model based on the fire characteristic knowledge graph to form a prediction of the fire risk in a specific area, time, and type.

8. A computer device, characterized in that: The device comprises: a memory, a processor, and a fire accident investigation program stored in the memory and executable on the processor, wherein the fire accident investigation program is configured to implement the steps of the fire accident investigation method according to any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium stores a fire accident investigation program, and when the fire accident investigation program is executed by the processor, the steps of the fire accident investigation method according to any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the fire accident investigation method according to any one of claims 1 to 6 are implemented.

Images