Fire-fighting potential safety hazard intelligent evaluation system and evaluation method

Abstract

The invention relates to an intelligent fire-fighting potential safety hazard evaluation system and evaluation method, and relates to the technical field of fire-fighting safety management, and the system comprises a multi-dimensional sensing module, a state vectorization module, a regional risk quantification module, a global risk evaluation module and a model optimization module. Multi-dimensional data is converted into a unified state vector through the multi-dimensional sensing module and the state vectorization module, and the unified state vector is compared with a dynamic normal state, so that the accuracy of hidden danger recognition is improved. The regional risk quantification module calculates the internal risk index of each region, and the global risk assessment module integrates the indexes to form a comprehensive risk assessment result. The model optimization module dynamically adjusts model parameters based on historical data and real-time feedback, and the prediction accuracy is improved. According to the invention, comprehensive monitoring and intelligent assessment of fire risks are realized, the efficiency and practicability of fire safety work are significantly improved, and it is ensured that potential fire risks are effectively coped with in a complex and changeable environment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fire safety management, in particular to a fire safety hidden danger intelligent evaluation system and method. BACKGROUND

[0002] In recent years, with the acceleration of urbanization and the continuous expansion of building scale, fire safety hidden dangers have gradually shown complexity and diversity. The existing fire safety monitoring system faces many challenges in effectiveness and real-time performance.

[0003] Existing technologies often rely on single sensor monitoring, which cannot comprehensively and systematically analyze various potential risks in buildings. Although some systems can monitor electrical parameters or environmental temperature, they often lack comprehensive consideration of other important data such as humidity, image information, etc. This single data source results in inaccurate assessment of fire hazards, and cannot timely discover hidden dangers and issue warnings. This is in sharp contrast to the design of the integrated multi-dimensional perception module of the present application, which allows real-time collection of multiple data, improving the scientificity and effectiveness of fire safety management.

[0004] In addition, existing systems usually only base their risk assessment on static baseline data, lacking dynamic adjustment capabilities. Since environmental and fire risk factors often change, this static assessment method is prone to assessment bias, making the issued warnings or decisions not meaningful in reality. This problem further leads to slow response of fire safety management in handling complex emergencies. The present application uses real-time data and dynamic normal baseline comparison through the state vectorization module to achieve more accurate risk assessment, improving the sensitivity of the system to risk changes to better support subsequent decision-making processes.

[0005] Finally, existing technologies often rely on manual adjustment for model optimization, lacking intelligent and adaptive capabilities, making it difficult for the evaluation model to adapt to changing dangers. This limits the response speed and processing capacity of fire safety management. The model optimization module of the present application can dynamically adjust the evaluation model based on historical data and real-time feedback through continuous learning and automatic tuning, significantly improving the accuracy and response capability of the overall system in dealing with fire risks. This adaptive mechanism provides strong support for overall fire safety management, ensuring that the system can effectively respond to changing safety risks. SUMMARY

[0006] The purpose of the present application is to provide a fire safety hidden danger intelligent evaluation system and method, solving the technical problems of the existing fire safety monitoring system in real-time, comprehensiveness and intelligence.

[0007] In a first aspect, the application provides a fire safety hazard intelligent evaluation system, comprising: a multi-dimensional perception module, which collects multi-dimensional real-time perception data of a plurality of physical space regions, including at least electrical parameters and image information; a state vectorization module, which is connected with the multi-dimensional perception module, processes the real-time perception data into real-time state vectors corresponding to each physical space region, and calculates a state deviation vector by comparing with a dynamic normal state baseline model; a regional risk quantification module, which is connected with the state vectorization module, configures an independent hazard correlation matrix for each physical space region, and calculates an intrinsic risk index of each physical space region based on the state deviation vector and the hazard correlation matrix; a global risk assessment module, which is connected with the regional risk quantification module, constructs a weighted space adjacency graph describing the adjacency relationship between the plurality of physical space regions, and calculates a global comprehensive risk index of at least one target physical space region in combination with the intrinsic risk index of all physical space regions and the space adjacency graph.

[0008] Preferably, the regional risk quantification module is specifically configured to calculate the intrinsic risk index of region k at time t by the following quadratic equation: wherein R int (k,t) is the intrinsic risk index; is the state deviation vector of region k at time t; H k is the hazard correlation matrix of region k, and the non-diagonal elements of the matrix are used to quantify the synergistic risk between different dimensional perception data.

[0009] Preferably, the global risk assessment module is used to: calculate a conduction risk index conducted to the target physical space region from other regions based on the risk conduction coefficient defined in the space adjacency graph between the physical space regions.

[0010] Preferably, the global risk assessment module calculates the conduction risk index of target physical space region k at time t by weighted summation of the intrinsic risk indexes of all adjacent physical space regions: R cnd (k,t)=∑ j≠k γ kj ·R int (j,t); wherein R cnd (k,t) is the conduction risk index; ∑ j≠k denotes summation of all adjacent physical space regions j of the target physical space region k; γkj is the risk transmission coefficient from the adjacent physical space region j to the target physical space region k; R int (j, t) is the intrinsic risk index of the adjacent physical space region j at time t.

[0011] Preferably, the global risk assessment module obtains the global comprehensive risk index of the target physical space region k at time t by adding the intrinsic risk index and the transmission risk index: R glb (k, t) = R int (k, t) + R cnd (k, t) ; Wherein, R glb (k, t) is the global comprehensive risk index; R int (k, t) is the intrinsic risk index of the target physical space region k; R cnd (k, t) is the transmission risk index of the target physical space region k.

[0012] Preferably, the dynamic normal state baseline model configured in the state vectorization module is constructed based on a long short-term memory network autoencoder, and is used to generate a normal state baseline vector that dynamically changes over time according to time information.

[0013] Preferably, it further comprises a model optimization module, which is used to receive and record manual feedback labels for historical warning events, and adjust and optimize the hazard correlation matrix in the regional risk quantification module based on the feedback labels using a gradient descent algorithm.

[0014] Preferably, the model optimization module is used to adaptively adjust the risk transmission coefficient of the spatial adjacency graph in the global risk assessment module according to the actual spreading path captured by the sensor after a real fire spreading event occurs.

[0015] Preferably, the global risk assessment module identifies and outputs one or more high-risk spreading paths based on the distribution gradient of the global comprehensive risk index calculated on the spatial adjacency graph.

[0016] In a second aspect, the present application provides a fire safety hidden danger intelligent evaluation method, comprising the following steps: Collecting multi-dimensional real-time perception data of a plurality of physical space regions, including at least electrical parameters and image information; Processing the real-time perception data into real-time state vectors corresponding to each physical space region, and calculating a state deviation vector by comparing with a preset dynamic normal state baseline model; An internal risk index of each physical space region is calculated based on a hazard correlation matrix independently configured for each physical space region and using the state deviation vector; A global comprehensive risk index of at least one target physical space region is calculated based on a pre-constructed weighted space adjacency graph describing adjacency relationships between the plurality of physical space regions and in combination with the internal risk index of all physical space regions, to achieve the evaluation of fire safety hazards.

[0017] In summary, the present application includes at least one of the following beneficial technical effects: 1. The present application can realize comprehensive monitoring of fire safety hazards by integrating multi-dimensional perception modules to collect various data such as electrical parameters, environmental temperature, humidity, image information, etc. within the building in real time. The collection and analysis of multi-dimensional data enable the system to timely identify and warn potential risks, significantly improving the scientificity and effectiveness of fire safety management; 2. The state vectorization module in the present application converts multi-dimensional perception data into a unified state vector and compares it with a dynamic normal state baseline, providing accurate risk assessment basis. It can adapt to different environments and conditions, enhancing the accuracy of risk assessment and better supporting the subsequent decision-making process; 3. The model optimization module of the present application can continuously learn and self-optimize, dynamically adjusting the evaluation model through analysis of historical data and real-time feedback. This adaptive ability enables the system to maintain high evaluation accuracy when dealing with changing fire risks, ultimately improving the response speed and processing capacity of overall fire safety management. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 is a system architecture diagram of the present application; Figure 2 is a method flowchart of the present application. DETAILED DESCRIPTION

[0019] The present application is further described in detail in conjunction with the accompanying Figure 1

[0020] Embodiment: An intelligent evaluation system for fire safety hazards, comprising: A multi-dimensional perception module collects multi-dimensional real-time perception data of a plurality of physical space regions, including at least electrical parameters and image information; The multi-dimensional perception module comprehensively monitors a plurality of physical space regions in a target building in real time. Its core function is to synchronously collect multi-dimensional data such as electrical parameters, image information, temperature data, humidity readings, etc. through various sensor devices, thereby building a comprehensive and accurate state database. These data will provide the basis for the intelligent evaluation of the system.​

[0021] In a specific implementation, the multi-dimensional perception module will be equipped with a combination of various sensors, including but not limited to current sensors, temperature sensors, image cameras, etc., in order to comprehensively cover different types of safety hazards. For example, the load condition of the electrical system can be monitored in real time through the current sensor to identify potential overload risks; at the same time, the collection of image information can provide real-time visual monitoring to assist in identifying fire hazards and other safety issues.

[0022] After real-time data collection of electrical parameters, the module will use the configured algorithm to convert the data into a state vector suitable for system processing. This state vector will include information reflecting the current electrical state, environmental temperature and humidity, and feature values of processed image data. Specifically, the state vector can be represented as: where V k represents the voltage value of region k, I k represents the current value, T k represents the environmental temperature, H k represents the environmental humidity, and F k represents the feature vector of image information.

[0023] In order to enhance the intelligent judgment ability of the multi-dimensional perception module, the module adopts a dynamic normal state baseline model. This baseline model is mainly generated through analysis and pattern recognition of historical data, in order to provide an effective reference for real-time collected data. In the dynamic normal state baseline model, the module will adaptively adjust according to time changes, for example, under certain environmental conditions, the normal range of electrical and temperature factors may be different. Therefore, the establishment process of the dynamic normal state baseline involves the use of deep learning technologies such as long short-term memory network (LSTM), through training on historical data, so that the system can be continuously optimized.

[0024] By comparing the real-time collected state vector with the above dynamic baseline model, the module can calculate the state deviation vector of each physical space region. This state deviation vector reflects the difference between the current state of the region and the normal operating state. The state deviation vector can be represented as: where, is the state deviation vector of region k at time t, and is the normal state baseline vector at the corresponding time point. In this way, the comparison of real-time monitored data with the dynamic baseline lays the foundation for subsequent risk quantification and comprehensive evaluation.

[0025] In this embodiment, the design of the multi-dimensional perception module not only improves the comprehensiveness of data collection, but also enhances the ability to identify abnormal states. In the entire system, this module serves as part of the information source, making the intelligent evaluation process of fire safety hazards more scientific and reliable. Through this multi-level, multi-dimensional data monitoring and analysis approach, the overall effective identification, real-time monitoring, and risk assessment of fire safety hazards are achieved. This enables the intelligent evaluation system to issue timely warnings and take appropriate measures, thereby effectively reducing the likelihood of fire accidents.

[0026] The state vectorization module, connected to the multi-dimensional perception module, processes real-time perception data into real-time state vectors corresponding to each physical space region, and calculates the state deviation vector by comparing with the dynamic normal state baseline model. The main function of the state vectorization module is to process multi-dimensional real-time data collected from the multi-dimensional perception module and integrate these data into a unified state representation. The design of this module is highly flexible and adaptable, and can be optimized for different monitoring needs to ensure accurate evaluation of the fire safety status of each physical space region of the target building.

[0027] In specific implementation, the state vectorization module first receives real-time perception data from the multi-dimensional perception module. These data include electrical parameters, environmental temperature, humidity, and image information, etc. The module needs to preprocess these data, such as denoising, standardization, etc., to eliminate errors caused by unstable collection environment or equipment.

[0028] Subsequently, the processed data will be converted into state vectors, which can be represented as: Where, V k represents the voltage value of region k, I k represents the current value, T k represents the environmental temperature, H k represents the environmental humidity, and F k represents the processed image feature vector. These components together constitute the state vector of region k at time t

[0029] In the process of constructing the state vector, the state vectorization module also uses the dynamic normal state baseline model as a reference for subsequent state comparison and abnormality identification. This dynamic normal state baseline model is constructed based on the analysis and learning of historical data using deep learning algorithms (such as LSTM). By learning the features of historical state data, this model can dynamically adjust and adapt to changes in normal state, ensuring the accuracy of the evaluation.

[0030] By comparing the real-time collected state vector with the dynamic baseline model, the state vectorization module can further calculate the state deviation vector of each physical space area. The calculation method of the state deviation vector is: wherein, is the state deviation vector of area k at time t, is the normal state baseline vector of the area at the corresponding time point. This calculation result is used to reflect the difference between the current state and the expected normal state, and further provides necessary data support for the area risk quantification module.

[0031] In combination with the above description, the state vectorization module ensures the efficiency and accuracy of the monitoring and evaluation process of fire safety hazards through the processing and comparison of real-time data. The effective operation of this module lays a solid foundation for subsequent risk identification and decision-making, forms an intelligent evaluation mechanism, and can timely discover and identify fire safety hazards, thereby providing scientific basis for fire safety management.

[0032] The area risk quantification module, connected with the state vectorization module, is configured with an independent hazard correlation matrix for each physical space area, and based on the state deviation vector and the hazard correlation matrix, the internal risk index of each physical space area is calculated; The primary task of the area risk quantification module is to calculate the internal risk index of the corresponding area based on the state deviation vector provided by the state vectorization module and the independently configured hazard correlation matrix for each physical space area. This process involves in-depth analysis of the collected data to identify potential fire safety hazards.

[0033] In specific implementation, the area risk quantification module first receives the state deviation vector from the state vectorization module, which is represented as: wherein, ΔV k , ΔI k , ΔT k , ΔH k , ΔF k respectively represent the voltage, current, temperature, humidity and image feature deviation of area k at time t. By analyzing these deviations, the module can identify the difference between the current state of the area and the preset normal state, and provide necessary information for subsequent risk quantification.

[0034] Next, the area risk quantification module will use the independently configured hazard correlation matrix H kThese deviations are weighted. The hazard correlation matrix is used to quantify the relationship between different perception data, such as the interaction between different electrical parameters, and the synergistic effect of environmental factors such as temperature and humidity. The construction of this matrix is based on historical data and domain knowledge, which can accurately reflect the contribution of various parameters to safety risks in a specific scenario.

[0035] Then, the module uses a quadratic equation to calculate the intrinsic risk index R int (k,t) of region k at time t, which is expressed as follows: The intrinsic risk index R int (k,t) in the above formula is the quantitative result of the regional risk. By transposing the state deviation vector and multiplying it with the hazard correlation matrix, and finally multiplying it with the state deviation vector, the interaction between different parameters is effectively quantified. This risk index can be used to determine whether there is a potential safety hazard in the region, and to take appropriate warning and processing according to the result.

[0036] The design of the regional risk quantification module ensures the real-time and accuracy of risk analysis. By continuously receiving and processing monitoring data, this module can constantly update the intrinsic risk index and reflect the changes in the regional safety state in real time. This dynamic updating capability enables the system to quickly respond and take measures when potential fire hazards occur, reducing the risk.

[0037] In summary, the regional risk quantification module provides scientific basis and accurate data support for the evaluation of fire safety hazards through intelligent data analysis and quantification methods. This intrinsic risk index calculation mechanism based on state deviation plays an important role in the entire intelligent evaluation system, laying a foundation for efficient and safe fire management.

[0038] The global risk assessment module, connected with the regional risk quantification module, constructs a weighted space adjacency graph describing the adjacency relationship between multiple physical space regions, and calculates the global comprehensive risk index of at least one target physical space region by combining the intrinsic risk index of all physical space regions and the space adjacency graph.

[0039] The core function of the global risk assessment module is to connect the intrinsic risk index of each physical space region and their adjacency relationship by constructing a weighted space adjacency graph. This design not only considers the individual risk of the region, but also effectively evaluates the transmission effect of risk between different regions, providing a more comprehensive analysis of the overall fire safety situation.

[0040] In specific implementation, the global risk assessment module first receives the intrinsic risk index R (k,t) of multiple regions from the regional risk quantification moduleint (j, t), where j represents different regions and t is the current time. The module will utilize these intrinsic risk indices in conjunction with adjacency relationships to construct a weighted spatial adjacency graph. This graph visually reflects the inter-influence relationships between regions, with the weights of adjacency relationships determined by prior knowledge or historical data, representing the strength of risk transmission between different regions.

[0041] After the adjacency graph is constructed, the global risk assessment module needs to calculate the transmission risk index R cnd (k, t) of a specific region k at time t. This transmission risk index is obtained by weighting and summing the intrinsic risk indices of adjacent regions to region k with risk transmission coefficients. Its expression is: R cnd (k, t) = ∑ j≠k γ kj ·R int (j, t); where γ kj is the risk transmission coefficient from adjacent region j to target region k, representing the degree of mutual influence of fire safety risks between regions. Through the calculation of this formula, the module can effectively identify those adjacent regions that may pose a threat to the safety of region k and quantify their impact.

[0042] Next, the module will integrate the calculated transmission risk index with the intrinsic risk index of region k to generate the global comprehensive risk index R glb (k, t) of this region. The calculation expression of the global comprehensive risk index is: R glb (k, t) = R int (k, t) + R cnd (k, t); Through this formula, the global risk assessment module can combine internal and external transmission risks to form a comprehensive risk assessment result.

[0043] In addition, the global risk assessment module also has the ability of adaptive adjustment when processing information. In actual application, the module can dynamically optimize the calculation of risk transmission coefficients using feedback from historical events and real propagation path data captured by sensors. This makes the model more effective in capturing and analyzing actual risk propagation paths when facing emergencies, enhancing the flexibility and practicality of the system.

[0044] In summary, the global risk assessment module can effectively identify and assess the overall fire safety hazards of the building by comprehensively analyzing the inherent risks of each area and their risk transmission relationship with the surrounding environment. The introduction of this module not only improves the accuracy of risk assessment, but also provides strong data support for fire management decision-making, ensuring that the system can reflect the potential safety threats within the building in real time.

[0045] The model optimization module is used to receive and record manual feedback labels for historical warning events, and based on the feedback labels, the gradient descent algorithm is used to adjust and optimize the hazard correlation matrix in the regional risk quantification module.

[0046] The core function of the model optimization module is to use the collected historical data and real-time feedback information to adaptively adjust and optimize the parameters of the evaluation model to adapt to the changing environmental conditions and fire hazard characteristics. The design of this module ensures that the system can maintain good performance in different working environments to accurately predict and evaluate fire safety risks.

[0047] In specific implementation, the model optimization module first integrates relevant data from the multi-dimensional perception module, state vectorization module, regional risk quantification module, and global risk assessment module. By collecting the output data provided by these modules, the module optimization module can obtain the real fire risk state and compare it with the expected output of the model to identify possible deviations.

[0048] For parameter optimization, the module uses an interactive incremental learning algorithm that can dynamically adjust model parameters based on new data and feedback. Specifically, let the model output at the current time t be Y pred (t), and the corresponding actual observation data be Y true (t), the deviation between the two can be represented as: ΔY(t)=Y true (t)-Y pred (t); This deviation is used to guide the adjustment of model parameters.

[0049] To achieve adaptive optimization, the model optimization module introduces the gradient descent method to update the model parameters. The parameter update expression can be represented as: where θ is the model parameter, α is the learning rate, and represents the gradient of the current model parameters. Through iterative updates, the model can gradually approach the optimal state, effectively reducing the prediction error.

[0050] During the optimization process, the module also introduces an adaptive regularization strategy to prevent model overfitting. This regularization strategy relies on controlling the model's complexity by introducing a regularization term in the loss function, so that the model not only pursues the minimization of prediction bias during optimization, but also maintains a moderate model complexity. The optimized loss function can be expressed as: where R(θ) is the regularization term, and the value of λ is set according to specific circumstances.

[0051] In addition, the model optimization module also needs to regularly evaluate the model comprehensively to test its effectiveness in different environments. This can be done through cross-validation methods. For example, the performance of the model is calculated on the training set and the validation set, and then a comprehensive evaluation is made to ensure the reliability of the model in practical applications.

[0052] In addition, the module also introduces cross-validation technology to evaluate the generalization ability of the optimized model. This technology divides the data set into multiple subsets, ensuring that the model also maintains good performance on unseen data. By training and validating on different data subsets, the module will help identify overfitting and adjust the model complexity in real time to obtain the best model structure.

[0053] To achieve the above optimization process, the model optimization module also designs an automatic parameter tuning mechanism. This mechanism uses advanced techniques such as Bayesian optimization to automatically select the optimal hyperparameters without human intervention. This significantly improves the efficiency and accuracy of model optimization, enabling the module to quickly respond to changes in environmental and data conditions and continuously improve model performance.

[0054] Finally, the model optimization module will generate a new optimized model after optimization and feed it back to the fire safety hazard intelligent evaluation system for use in subsequent risk assessment, monitoring and early warning processes. This feedback loop helps continuously improve the overall intelligence level and practical value of the system.

[0055] In summary, the model optimization module achieves continuous optimization of the internal evaluation model by analyzing historical data and real-time information and using modern optimization algorithms for dynamic model adjustment. This multi-level optimization strategy not only significantly enhances the adaptability and accuracy of the model, but also provides strong decision support for the entire fire safety hazard intelligent evaluation system, thereby significantly improving the ability to respond to fire risks.

[0056] In combination with the Figure 2 Another embodiment of the present application provides a fire safety hazard intelligent evaluation method, comprising the following steps: S1, collect multi-dimensional real-time perception data of a plurality of physical space regions, including at least electrical parameters and image information; S2, process the real-time perception data into real-time state vectors corresponding to each physical space region, and calculate a state deviation vector by comparing with a preset dynamic normal state baseline model; S3, based on a hazard correlation matrix independently configured for each physical space region, and using the state deviation vector, calculate the intrinsic risk index of each physical space region; S4, based on a pre-constructed weighted space adjacency graph describing the adjacency relationship between a plurality of physical space regions, and combined with the intrinsic risk index of all physical space regions, calculate the global comprehensive risk index of at least one target physical space region to realize the evaluation of fire safety hidden dangers.

[0057] In the fire inspection scene, the system first conducts on-site investigation of fire safety hidden dangers through an artificial inspection module. The inspector can use a mobile terminal or a special device to identify and record hidden danger information. The module provides a friendly user interface, allowing the inspector to quickly access historical records and hidden danger types, improving inspection efficiency and accuracy.

[0058] During the hidden danger investigation process, the inspector can use a mobile terminal to take photos and upload hidden danger photos. The image data will be automatically stored by the system and associated with the inspection record, facilitating subsequent analysis and tracking. On this basis, the system will automatically identify and analyze the uploaded hidden danger images through image recognition technology, extracting relevant feature information to quickly confirm the hidden danger type and severity.

[0059] When the hidden danger is identified by the system, the hidden danger response module is triggered. The module will automatically generate processing suggestions and solutions based on the severity of the hidden danger, such as temporary rectification measures, maintenance detection requirements, or notifications to relevant responsible persons. In this processing scheme, the system will also introduce emergency plans and executable rectification tasks, which will be automatically assigned to responsible units to ensure the timeliness and effectiveness of hidden danger rectification.

[0060] Finally, the system will generate dynamic reports and feedback to relevant personnel according to the hidden danger type and rectification progress, ensuring the transparency and traceability of fire safety management. The entire process effectively combines artificial inspection with system intelligence, making the investigation, identification, processing, and tracking of fire safety hidden dangers more efficient and orderly, providing strong support for improving fire safety management level.

[0061] The embodiments of the specific implementation are the preferred embodiments of the present application, not limited to the protection scope of the present application, wherein the same parts are indicated by the same reference numerals. Therefore: any equivalent changes made according to the structure, shape, principle of the present application should be covered within the protection scope of the present application.

Claims

1. A fire safety hazard intelligent evaluation and inspection system, characterized in that, Comprise: A multi-dimensional perception module that collects multi-dimensional real-time perception data of multiple physical space regions, including at least electrical parameters and image information; A state vectorization module connected with the multi-dimensional perception module, which processes the real-time perception data into real-time state vectors corresponding to each physical space region, and calculates a state deviation vector by comparing with a dynamic normal state baseline model; A regional risk quantification module connected with the state vectorization module, which configures an independent hazard correlation matrix for each physical space region, and calculates an intrinsic risk index of each physical space region based on the state deviation vector and the hazard correlation matrix; A global risk assessment module connected with the regional risk quantification module, which constructs a weighted space adjacency graph describing the adjacency relationship between the multiple physical space regions, and calculates a global comprehensive risk index of at least one target physical space region in combination with the intrinsic risk index of all physical space regions and the space adjacency graph.

2. The fire safety hazard intelligent evaluation and inspection system according to claim 1, characterized in that, The regional risk quantification module is specifically configured to calculate the intrinsic risk index of region k at time t through the following quadratic equation: wherein R int (k, t) is the intrinsic risk index; is the state deviation vector of region k at time t; H k is the hazard correlation matrix of region k, and the off-diagonal elements of the matrix are used to quantify the synergistic risk between different dimensions of perception data.

3. The fire safety hazard intelligent evaluation and inspection system according to claim 1, characterized in that, The global risk assessment module is used to: Calculate a conduction risk index conducted to the target physical space region from other regions based on the risk conduction coefficient defined in the space adjacency graph.

4. The fire safety hazard intelligent evaluation and inspection system according to claim 1, characterized in that, The global risk assessment module calculates the conduction risk index of target physical space region k at time t by weighted summing the intrinsic risk indices of all adjacent physical space regions: R cnd (k,t) = ∑ j≠k γ kj ·R int (j,t); wherein R cnd (k, t) is a conduction risk index;∑ j≠k denotes the summation over all adjacent physical space regions j to the target physical space region k; γ kj is the risk conduction coefficient from the adjacent physical space region j to the target physical space region k; R int (j, t) is the intrinsic risk index of the adjacent physical space region j at time t.

5. The fire safety hazard intelligent evaluation and inspection system according to claim 4, characterized in that, The global risk assessment module obtains the global comprehensive risk index of target physical space region k at time t by adding the intrinsic risk index and the conduction risk index: R glb (k,t) = R int (k,t) + R cnd (k,t); wherein R glb (k,t) is the global composite risk index; R int (k,t) is the intrinsic risk index for the target physical space region k; R cnd (k,t) is the conduction risk index for the target physical space region k.

6. The fire safety hazard intelligent evaluation and inspection system according to claim 1, characterized in that, The dynamic normal state baseline model configured in the state vectorization module is constructed based on a long short-term memory network autoencoder, which is used to generate a normal state baseline vector that dynamically changes over time according to time information.

7. The fire safety hazard intelligent evaluation and inspection system according to claim 1, wherein, Further comprising a model optimization module, which is used to receive and record manual feedback labels for historical warning events, and adjust and optimize the hazard correlation matrix in the regional risk quantification module based on the feedback labels using a gradient descent algorithm.

8. The fire safety hazard intelligent evaluation and inspection system according to claim 7, characterized in that, The model optimization module is used to adaptively adjust the risk conduction coefficient of the space adjacency graph in the global risk assessment module according to the actual spread path captured by the sensor after a real fire spread event occurs.

9. The fire safety hazard intelligent evaluation and inspection system according to claim 1, characterized in that, The global risk assessment module identifies and outputs one or more high-risk spread paths based on the distribution gradient of the global comprehensive risk index calculated on the space adjacency graph.

10. A fire safety hazard intelligent evaluation method, according to any one of claims 1-9, a fire safety hazard intelligent evaluation system, characterized in that, Comprise the following steps: Collect multi-dimensional real-time perception data of multiple physical space regions, including at least electrical parameters and image information; Process the real-time perception data into real-time state vectors corresponding to each physical space region, and calculate a state deviation vector by comparing with a preset dynamic normal state baseline model; An inherent risk index of each physical space region is calculated based on a hazard correlation matrix independently configured for each physical space region and by using the state bias vector; A global comprehensive risk index of at least one target physical space region is calculated based on a pre-constructed weighted space adjacency graph describing adjacency relationships between the plurality of physical space regions and in combination with the inherent risk index of all physical space regions, so as to realize the evaluation of fire safety hazards.

Images