Non-key unit fire risk assessment method fusing human factor analysis

Abstract

The invention discloses a non-key unit fire risk assessment method fusing human factor analysis, and the method comprises the steps: building a fire human factor analysis time sequence analysis model, dividing assessment subspaces, and collecting human factor dynamic data; extracting fire risk parameters such as electrical equipment use frequency and the like, and fusing the fire risk parameters with corresponding human factor data features to form vectors; a vector element relationship is captured by means of a harmful multi-head attention mechanism architecture, and an attention weighted feature vector is generated. And deducing a human factor change trend by combining a time sequence analysis model according to the vector, and determining risk influence factors of each subspace. A risk propagation network is established, weights are set by considering physical positions of subspaces and personnel flow, and risk diffusion is simulated. And finally, through a comprehensive evaluation model, in combination with a risk diffusion simulation result, a human factor model deduction result and an attention weighted feature vector, determining an overall fire risk level of a non-key unit, thereby providing a scientific basis for fire prevention and control.

Description

Technical Field

[0001] The present invention relates to the field of fire risk assessment, and in particular to a fire risk assessment method for non-key units integrating human factor analysis. Background Art

[0002] With the growing demand for fire safety, non-key units face significant challenges in fire prevention and control due to their widespread distribution and weak management. Traditional fire risk assessments focus primarily on key units, with insufficient attention paid to non-key units. Existing assessment methods often overlook the crucial role of human factors in the occurrence and development of fires, making it difficult to comprehensively and accurately assess fire risks in non-key units.

[0003] On the one hand, existing assessment techniques lack a dynamic consideration of human factors. Factors such as human behavior, operational errors, and management negligence exhibit significant time series characteristics and constantly change over time. However, traditional assessment methods rely solely on static data or simple indicators, failing to capture the dynamic evolution of human factors over time. This leads to discrepancies between predicted fire risk and actual conditions.

[0004] On the other hand, existing assessment models fail to effectively integrate multidimensional risk parameters and human factors. Fire risk is influenced by a variety of parameters, including electrical equipment, combustible materials, firefighting facilities, and personnel density, all of which are interrelated and interact with human factors. However, traditional models often analyze each parameter in isolation, failing to fully explore the potential connections between them. This makes it difficult for assessment results to truly reflect the complex nature of fire risks in non-key units and to provide accurate and effective risk prevention and control strategies. Summary of the Invention

[0005] In order to overcome the shortcomings and deficiencies of the prior art, the present invention provides a fire risk assessment method for non-key units that integrates human factor analysis.

[0006] The technical solution adopted by the present invention is a non-key unit fire risk assessment method integrating human factor analysis, comprising the following steps:

[0007] Step S1: Based on historical fire data, a time series analysis model for fire human factors analysis is constructed, including different dimensions such as personnel behavior, operational errors, and management negligence. Each area within the non-key unit is divided into multiple evaluation subspaces, and dynamic data of human factors in different time series are collected for each evaluation subspace;

[0008] Step S2: Extract various parameters related to fire risk in the assessment subspace, including the frequency of use of electrical equipment, the amount of combustible material accumulation, the response time of firefighting facilities, and the density of personnel, and fuse the parameters with the human factor data in the corresponding time series to form a fused feature vector;

[0009] Step S3: constructing a harmful multi-head attention mechanism architecture, inputting the fused feature vector into the harmful multi-head attention mechanism architecture, capturing the relationship between the elements in the fused feature vector from different angles through multi-head parallel computing, and generating an attention weighted feature vector;

[0010] Step S4: Based on the attention weighted feature vector, a fire human factor analysis time series analysis model is used to deduce the change trend of human factors in different evaluation subspaces at each time node, and the risk impact factor of each evaluation subspace at the corresponding time node is determined in combination with various fire risk parameters;

[0011] Step S5: Establish a risk propagation network, use each evaluation subspace as a network node, set node connection weights based on the physical location relationship between subspaces and personnel flow paths, and perform risk diffusion simulation on the risk propagation network based on the risk impact factors and node connection weights;

[0012] Step S6: Determine the overall fire risk level of non-key units by synthesizing the risk diffusion simulation results, combining the deduction results of the fire human factor analysis time series analysis model and the attention weighted feature vector processed by the harmful multi-head attention mechanism.

[0013] Furthermore, in step S2, the parameters are fused with the human factor data in the corresponding time series, using the following fusion model formula:

[0014]

[0015] Among them, F vec is the fusion feature vector, n is the total number of parameters and factor data, α i is the weight coefficient of the fusion of the i-th group of parameters and factor data, P i is the i-th fire risk parameter vector, H i is the human factor vector of the i-th fire in the corresponding time series, Represents a vector fusion operation.

[0016] Furthermore, in step S3, in the harmful multi-head attention mechanism architecture, the following formula is used to calculate each attention head:

[0017]

[0018] Among them, A head is the output of a single attention head, Q, K, and V are respectively the query vector, key vector, and value vector obtained by different linear transformations of the input fusion feature vector, and W q 、W k 、W vis the corresponding learnable weight matrix, d k is the dimension of the key vector, σ is the activation function, which is used to normalize the calculation results. Finally, the outputs of multiple attention heads are concatenated and linearly transformed to obtain the attention weighted feature vector.

[0019] Furthermore, in step S4, the fire human factor analysis time series analysis model is used to deduce the changing trends of human factors in different assessment subspaces at each time node, and the risk impact factor of each assessment subspace at the corresponding time node is determined in combination with various parameters of fire risk, using the following calculation model:

[0020]

[0021] Among them, R factor is the risk impact factor, β is the trend impact coefficient, ΔH trend is the changing trend of human factors within the time interval ΔT, m is the number of selected fire risk parameters, P j is the j-th fire risk parameter value, γ j is the weight index of the j-th fire risk parameter.

[0022] Furthermore, in step S5, the risk propagation network is established, each evaluation subspace is used as a network node, and the node connection weight is set according to the physical position relationship between the subspaces and the personnel flow path, using the following weight calculation model:

[0023]

[0024] Among them, W conn is the node connection weight, Dis space is the physical distance between the two evaluation subspaces, s total is the total number of evaluation subspaces, ρ is the distance weight adjustment coefficient, Flow rate is the personnel flow rate between the two subspaces.

[0025] Furthermore, in step S6, the comprehensive risk diffusion simulation results are combined with the deduction results of the fire human factor analysis time series analysis model and the attention weighted feature vector processed by the harmful multi-head attention mechanism to determine the overall fire risk level of non-key units, using the following comprehensive assessment model:

[0026]

[0027] Among them, R level is the overall fire risk level, S diffuse is the quantitative value of the risk diffusion simulation result, T trend A is the quantitative value of the deduction result of the time series analysis model for fire human factors analysis,vec is the quantized value of the attention weighted feature vector, ω1, ω2, and ω3 are the weight coefficients of the corresponding results, and Map is the mapping function that maps the calculation results to the corresponding risk level interval.

[0028] Furthermore, in step S1, the fire human factor analysis time series analysis model adopts an autoregressive integrated moving average model structure to model the collected dynamic data of human factors in different time series, and fits and predicts the dynamic change law of human factors through iterative optimization of model parameters.

[0029] Furthermore, in step S3, a gating mechanism is introduced into the harmful multi-head attention mechanism architecture to selectively fuse the outputs of different attention heads. The gating calculation model is as follows:

[0030]

[0031] Among them, G gate is the gate vector, H is the number of attention heads, θ h is the gating weight output by the h-th attention head, is the output of the hth attention head. The output of each attention head is weighted by the gating vector to focus and fuse the key features.

[0032] Furthermore, in step S5, when simulating risk diffusion on the risk propagation network, the evacuation influence coefficient is introduced in combination with the inhibitory effect of personnel evacuation behavior on risk diffusion, and the speed and scope of risk diffusion are corrected. The corrected risk diffusion calculation model is as follows:

[0033] S new =S old ·(1-δ·Evac eff )

[0034] Among them, S new is the risk diffusion state after correction, S old is the original risk diffusion state, δ is the evacuation impact weight coefficient, Evac eff It is an indicator of personnel evacuation efficiency.

[0035] Beneficial Effects: The present invention proposes a fire risk assessment method for non-key units that integrates human factor analysis. This method constructs a time series analysis model for fire human factor analysis, collects dynamic human factor data under different time series, and uses the model to deduce the changing trends of human factors, so that the assessment can track the evolution of factors such as personnel behavior and operational errors over time in real time, significantly improving the accuracy of risk prediction. In terms of addressing the lack of integration between multi-dimensional risk parameters and human factors, by integrating various parameters such as the frequency of electrical equipment use and the amount of combustible material accumulation with the human factor data under the corresponding time series, the harmful multi-head attention mechanism architecture is used to capture the relationship between each element, deeply explore the potential connections between parameters, and achieve the organic integration of multi-dimensional data. This method establishes a risk propagation network, simulates risk diffusion based on the physical location and personnel flow paths between subspaces, and combines a comprehensive assessment model to determine the overall risk level. It can comprehensively and accurately reflect the complex characteristics of fire risks in non-key units. At the same time, the multiple innovative model formulas introduced in the method further quantify the influencing factors of each link, provide a scientific basis for risk assessment, and provide strong technical support for non-key units to formulate effective fire prevention and control strategies and reduce fire risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is a flow chart of the method steps of the present invention;

[0037] Figure 2 This is a diagram of the unit composition of the method implementation of the present invention. DETAILED DESCRIPTION

[0038] It should be noted that, unless there is a conflict, the embodiments in this application and the features described in the embodiments can be combined with each other. The application is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0039] like Figure 1 As shown in the figure, the fire risk assessment method for non-key units integrating human factors analysis includes the following steps:

[0040] Step S1: Based on historical fire data, a time series analysis model for fire human factors analysis is constructed, including different dimensions such as personnel behavior, operational errors, and management negligence. Each area within the non-key unit is divided into multiple evaluation subspaces, and dynamic data of human factors in different time series are collected for each evaluation subspace;

[0041] Specifically, in step S1, a time-series analysis model for analyzing human factors in fires is constructed based on historical fire data. This is the initial, critical step in the entire assessment method. Historical fire data covers multi-dimensional information on human behavior, operational errors, and management negligence from various past fire incidents. Through in-depth mining and analysis of this data, it is possible to summarize the changing patterns and underlying patterns of human factors during fire events. Dividing areas within non-key units into multiple assessment subspaces enables a refined assessment. The subspaces are divided based on factors such as the unit's architectural layout and functional zoning, such as by floor or room usage, to ensure relative consistency and uniqueness between human factors and fire risks within each subspace. For each assessment subspace, dynamic data on human factors over different time series is collected. This dynamic data includes personnel activity trajectories, operational frequency, and safety awareness during different time periods. This data can be collected through the installation of monitoring equipment, sensors, and other technical means to continuously record and collect data, providing rich and accurate foundational information for subsequent assessments.

[0042] This step, constructing a time-series analysis model, combines human factors with the time dimension, fully considering the dynamic nature of human factors and avoiding the limitations of traditional assessments that focus solely on static data. Dividing the assessment subspace and collecting dynamic data makes the assessment more relevant to the actual conditions of non-key units, accurately capturing the differences in human factors across different regions at different times, and laying a solid foundation for subsequent in-depth analysis of fire risks. This approach enables a comprehensive and systematic understanding of the full scope of human factors within non-key units, providing strong support for accurate assessments of fire risks in subsequent steps and helping to promptly identify potential fire risk hazards, allowing for the implementation of targeted preventive measures.

[0043] Step S2: Extract various parameters related to fire risk in the assessment subspace, including the frequency of use of electrical equipment, the amount of combustible material accumulation, the response time of firefighting facilities, and the density of personnel, and fuse the parameters with the human factor data in the corresponding time series to form a fused feature vector;

[0044] Specifically, step S2 primarily extracts parameters related to fire risk within the assessment subspace and fuses these parameters with the human factor data from the corresponding time series. Numerous parameters are associated with fire risk, such as the frequency of electrical equipment use (frequent use increases the probability of line aging and short circuits); the accumulation of combustible materials (large accumulations of combustibles are the material basis for the occurrence and spread of fires); the response time of firefighting facilities, which directly impacts the effectiveness of initial firefighting; and the density of personnel, which influences the difficulty and efficiency of evacuation during a fire. These parameters reflect the fire risk situation within the assessment subspace from different perspectives. Fusion of these parameters with the human factor data from the corresponding time series is designed to integrate multi-source information and explore potential connections between parameters and human factors. The fusion process uses specific technical means to organically combine parameter data and human factor data to form a fused feature vector, making the data more comprehensive and representative.

[0045] This step, by extracting various parameters, comprehensively quantifies and assesses the factors influencing fire risk within the subspace, providing specific data metrics for subsequent analysis. Feature fusion, however, breaks down the isolation between data, allowing different types of data to complement and validate each other, more accurately reflecting the essential characteristics of fire risk. The fused feature vector contains rich information, preserving the original features of the various parameters and human factors while also generating new feature dimensions through fusion. This provides high-quality data input for subsequent in-depth analysis using the multi-head attention mechanism architecture, facilitating more accurate fire risk assessments and improving the reliability and validity of assessment results.

[0046] Step S3: constructing a harmful multi-head attention mechanism architecture, inputting the fused feature vector into the harmful multi-head attention mechanism architecture, capturing the relationship between the elements in the fused feature vector from different angles through multi-head parallel computing, and generating an attention weighted feature vector;

[0047] Specifically, step S3 constructs a harmful multi-head attention mechanism architecture, and inputs the fused feature vector into it for processing. The harmful multi-head attention mechanism architecture is an advanced information processing structure that can deeply capture the relationship between the elements in the fused feature vector from multiple different angles through multi-head parallel computing. In this architecture, each attention head is equivalent to an independent information processing unit, which analyzes and calculates the input fused feature vector in different ways, focusing on different aspects of information in the fused feature vector. Through multi-head parallel computing, the complex relationships hidden in the fused feature vector can be fully explored, avoiding information omissions or misjudgments that may be caused by a single perspective. Finally, the outputs of multiple attention heads are integrated and processed to generate an attention-weighted feature vector, which highlights the key information in the fused feature vector that is more relevant to fire risk assessment and enhances the expressive power of the data.

[0048] This step plays a core role in the entire assessment method. Traditional data processing methods often struggle to fully capture the complex relationships between data elements. However, the multi-head attention mechanism architecture, through multi-head parallel computing and unique processing logic, can accurately analyze the degree of correlation between elements in the fused feature vector and effectively extract key information. Compared to the original fused feature vector, the attention-weighted feature vector better reflects the core elements of fire risk assessment. This provides more targeted and effective data support for subsequent risk deduction and identification of risk influencing factors using the fire human factor analysis time series analysis model, thereby enhancing the overall assessment method's ability to analyze and judge fire risks.

[0049] Step S4: Based on the attention weighted feature vector, a fire human factor analysis time series analysis model is used to deduce the change trend of human factors in different evaluation subspaces at each time node, and the risk impact factor of each evaluation subspace at the corresponding time node is determined in combination with various fire risk parameters;

[0050] Specifically, step S4 utilizes the attention-weighted eigenvectors and the fire human factor analysis time series analysis model to deduce the changing trends of human factors at various time points in different assessment subspaces. Combined with various fire risk parameters, the risk impact factor for each assessment subspace at the corresponding time point is determined. The fire human factor analysis time series analysis model, constructed based on step S1, predicts and deduces the changing trends of human factors at different time points in the future based on historical data and current human factor information. By analyzing the temporal variations of human factors, such as the impact of changes in personnel operating habits and adjustments to safety management measures, these trends are predicted. Combining various fire risk parameters such as the frequency of electrical equipment use and the amount of combustible material accumulated, the risk impact factor for each assessment subspace at the corresponding time point is determined by comprehensively considering the changes in human factors and their impact. The risk impact factor is an important indicator for measuring the degree of fire risk in a given assessment subspace at a specific time. It comprehensively reflects the combined effects of human factors and other risk parameters.

[0051] By deducing the changing trends of human factors, this step proactively considers the dynamic impact of human factors on fire risk, avoiding the limitations of assessments based solely on current conditions. By combining various fire risk parameters to determine risk influencing factors, human factors are organically integrated with other objective risk factors to comprehensively assess fire risk from multiple dimensions. This assessment method can more accurately reflect the actual fire risk status of the assessment subspace at different time points, providing a scientific basis for the subsequent development of targeted risk prevention and control measures. It helps to promptly identify high-risk areas and time periods, take preventative measures in advance, and reduce the likelihood of fire.

[0052] Step S5: Establish a risk propagation network, use each evaluation subspace as a network node, set node connection weights based on the physical location relationship between subspaces and personnel flow paths, and perform risk diffusion simulation on the risk propagation network based on the risk impact factors and node connection weights;

[0053] Specifically, step S5 establishes a risk propagation network, using each assessment subspace as a network node, setting node connection weights based on the physical location relationship and personnel flow paths between subspaces, and performing risk diffusion simulation on the risk propagation network based on the risk impact factor and the node connection weight. The risk propagation network uses the assessment subspace as the basic unit to construct a network structure that reflects the propagation relationship of fire risk within non-key units. The connection weights between nodes are determined based on the physical location relationship between subspaces, such as adjacent floors, adjacent rooms, etc., as well as the flow paths of personnel within the unit. The closer the physical location and the more frequent the flow of personnel, the greater the connection weight between subspaces, which means that the fire risk is more likely to spread between these subspaces. Based on the risk impact factor and node connection weight of each assessment subspace, the diffusion process of fire risk is simulated on the risk propagation network, and the propagation path, speed and range of risk between different subspaces are predicted through calculation and analysis.

[0054] Fire risk does not exist in isolation within each assessment subspace; instead, it propagates and diffuses within a unit. Establishing and simulating a risk propagation network fully accounts for the propagation characteristics of risk and provides a holistic understanding of the development of fire risk. By setting appropriate node connection weights and incorporating risk influencing factors into the simulation, we can more realistically reflect the spread of fire risk in a real-world environment, providing comprehensive information support for developing effective risk prevention and control strategies. This allows for the early identification of areas where risk may spread, enabling the implementation of isolation and evacuation measures to prevent further spread and minimize fire losses.

[0055] Step S6: Determine the overall fire risk level of non-key units by synthesizing the risk diffusion simulation results, combining the deduction results of the fire human factor analysis time series analysis model and the attention weighted feature vector processed by the harmful multi-head attention mechanism.

[0056] Specifically, step S6 combines the results of the risk diffusion simulation with the derivation results of the fire human factor analysis time series analysis model and the attention-weighted eigenvectors processed by the harmful multi-head attention mechanism to determine the overall fire risk level for non-key units. The risk diffusion simulation results demonstrate the propagation of fire risk across the various assessment subspaces of non-key units and the ultimate impact range; the derivation results of the fire human factor analysis time series analysis model reflect the changing trends of human factors at different time points and their impact on fire risk; and the attention-weighted eigenvectors highlight key information relevant to fire risk assessment. These three results are comprehensively analyzed, and through specific assessment algorithms and rules, a quantitative assessment of the overall fire risk of non-key units is conducted, ultimately determining the overall fire risk level, such as low risk, medium risk, or high risk.

[0057] This step is the link to achieve the ultimate goal of the entire assessment method. It is difficult to comprehensively and accurately assess the overall fire risk of non-key units based on the results of any one aspect alone. Only by combining the risk diffusion simulation results, the human factor deduction results and the attention weighted eigenvector, and comprehensively considering the various influencing factors of fire risk from multiple angles, can we draw objective and reliable assessment conclusions. Determining the overall fire risk level provides a clear basis for the fire safety management of non-key units. Units can formulate corresponding fire safety measures according to the risk level, reasonably allocate fire resources, focus on prevention and control of high-risk areas, improve the efficiency and effectiveness of fire safety management, and effectively reduce the risk of fire.

[0058] Preferably, in step S2, the parameters are fused with the human factor data in the corresponding time series using the following fusion model formula:

[0059]

[0060] Among them, F vec is the fusion feature vector, n is the total number of parameters and factor data, α i is the weight coefficient of the fusion of the i-th group of parameters and factor data, P i is the i-th fire risk parameter vector, H i is the human factor vector of the i-th fire in the corresponding time series, Represents a vector fusion operation, which includes vector concatenation, weighted summation, and other methods.

[0061] Specifically, a specific fusion model is used to organically combine fire risk parameters and human factor data. This model introduces a weight coefficient αi to adjust the importance of different parameter and factor data during the fusion process. The weight coefficient is set based on a correlation analysis of the impact of each parameter and factor on fire risk. The vector fusion operation ⊕ includes methods such as vector concatenation and weighted summation. It can select an appropriate fusion method based on data characteristics, achieving data dimension expansion and information complementarity. This feature fusion method not only retains the key features of the original data but also, through weighted combination, uncovers potential correlations between parameters and human factors, forming a more representative fused feature vector. This provides high-quality data input for subsequent attention calculations, improving the accuracy and comprehensiveness of fire risk assessments.

[0062] Preferably, in step S3, in the harmful multi-head attention mechanism architecture, the calculation of each attention head adopts the following formula:

[0063]

[0064] Among them, A headis the output of a single attention head, Q, K, and V are respectively the query vector, key vector, and value vector obtained by different linear transformations of the input fusion feature vector, and W q 、W k 、W v is the corresponding learnable weight matrix, d k is the dimension of the key vector, σ is the activation function, which is used to normalize the calculation results. Finally, the outputs of multiple attention heads are concatenated and linearly transformed to obtain the attention weighted feature vector.

[0065] Specifically, the harmful multi-head attention mechanism architecture in step S3 is explained in detail for the calculation process of a single attention head. This calculation process generates a query vector, a key vector, and a value vector by linearly transforming the input fused feature vector, and then captures the relationship between each element in the fused feature vector through matrix operations and normalization. Each attention head focuses on a different feature subspace. Through multi-head parallel calculation, it can comprehensively analyze the data from multiple angles and extract richer feature information. Finally, the outputs of multiple attention heads are spliced ​​and linearly transformed to generate an attention-weighted feature vector, which highlights the key information closely related to fire risk assessment, suppresses the interference of irrelevant information, enhances the model's ability to express fire risk characteristics, and provides more accurate data support for subsequent risk deduction.

[0066] Preferably, in step S4, the fire human factor analysis time series analysis model is used to deduce the changing trends of human factors in different assessment subspaces at each time node, and the risk impact factor of each assessment subspace at the corresponding time node is determined in combination with various parameters of fire risk, using the following calculation model:

[0067]

[0068] Among them, R factor is the risk impact factor, β is the trend impact coefficient, ΔH trend is the changing trend of human factors within the time interval ΔT, m is the number of selected fire risk parameters, P j is the j-th fire risk parameter value, γ j is the weight index of the j-th fire risk parameter.

[0069] Specifically, by comprehensively considering the changing trends of human factors and fire risk parameters, a quantitative assessment of the risk level of each assessment subspace is achieved. This calculation model introduces a trend influence coefficient β to adjust the weight of the impact of the changing trends of human factors on risk, reflecting the degree of influence of dynamic changes in human factors on fire risk. By calculating the changing trend of human factors within a time interval and combining the weighted product of multiple fire risk parameters, the synergistic effect of human factors and other risk parameters is fully considered. This calculation method can more accurately reflect the actual risk status of the assessment subspace at different time points, provide a scientific basis for the formulation of risk prevention and control measures, and help identify high-risk areas and time periods in advance so that targeted preventive measures can be taken.

[0070] Preferably, in step S5, the risk propagation network is established, each evaluation subspace is used as a network node, and the node connection weight is set according to the physical position relationship between the subspaces and the personnel flow path, and the following weight calculation model is adopted:

[0071]

[0072] Among them, W conn is the node connection weight, Dis space is the physical distance between the two evaluation subspaces, s total is the total number of evaluation subspaces, ρ is the distance weight adjustment coefficient, Flow rate is the personnel flow rate between the two subspaces.

[0073] Specifically, the method for calculating the connection weights of the nodes in the risk propagation network in step S5 comprehensively considers the physical distance and personnel flow factors between subspaces. The inverse of the physical distance is used as the basic weight, which reflects that the closer the subspaces are in space, the greater the possibility of risk propagation. The distance weight adjustment coefficient ρ is introduced to adjust the degree of influence of physical distance on the weight according to the actual environmental characteristics. The personnel flow rate reflects the frequency of personnel interaction between subspaces. The more frequent the personnel flow, the higher the probability of risk propagation. The node connection weights set in this way can more realistically simulate the propagation path and speed of fire risks in non-key units, provide an accurate network structure for risk diffusion simulation, help predict the scope of risk propagation, formulate response measures in advance, and reduce the losses caused by fire.

[0074] Preferably, in step S6, the comprehensive risk diffusion simulation results are combined with the deduction results of the fire human factor analysis time series analysis model and the attention weighted feature vector processed by the harmful multi-head attention mechanism to determine the overall fire risk level of non-key units, using the following comprehensive assessment model:

[0075]

[0076] Among them, R level is the overall fire risk level, S diffuse is the quantitative value of the risk diffusion simulation result, T trend A is the quantitative value of the deduction result of the time series analysis model for fire human factors analysis, vec is the quantized value of the attention weighted feature vector, ω1, ω2, and ω3 are the weight coefficients of the corresponding results, and Map is the mapping function that maps the calculation results to the corresponding risk level interval.

[0077] Specifically, the risk diffusion simulation results, human factor deduction results and attention weighted feature vectors are organically integrated through a comprehensive assessment model. The model introduces weight coefficients ω1, ω2, and ω3 to adjust the importance of the three results in the comprehensive assessment respectively. The setting of the weight coefficients is based on the analysis of the contribution of the three results in the fire risk assessment. The comprehensive assessment value is obtained by weighted averaging, and then mapped to the corresponding risk level interval using a mapping function to achieve a quantitative classification of the overall fire risk of non-key units. This comprehensive assessment method fully considers the multi-dimensional characteristics of fire risk and conducts a comprehensive assessment from multiple angles such as risk propagation, human factor changes and feature associations, making the assessment results more objective and reliable, and providing a scientific basis for fire safety management decisions.

[0078] Preferably, in step S1, the fire human factor analysis time series analysis model adopts an autoregressive integrated moving average model structure to model the collected dynamic data of human factors in different time series, and realizes the fitting and prediction of the dynamic change law of human factors through iterative optimization of model parameters.

[0079] Specifically, the specific structure of the time series analysis model for fire human factor analysis in step S1 uses the autoregressive integrated moving average (ARIMA) model to model the dynamic data of human factors. This model can capture the autocorrelation and trend in time series data and fit the dynamic changes of human factors by learning from historical data. The iterative optimization process of the model parameters is based on methods such as maximum likelihood estimation, and the model parameters are continuously adjusted to optimize the model's fit to the historical data, thereby improving the accuracy of predicting future trends in human factor changes. This model structure is suitable for processing human factor data with a certain degree of randomness and trend, and can provide reliable predictions of human factor changes for fire risk assessment, enhancing the foresight and adaptability of the assessment method.

[0080] Preferably, in step S3, a gating mechanism is introduced into the harmful multi-head attention mechanism architecture to selectively fuse the outputs of different attention heads. The gating calculation model is as follows:

[0081]

[0082] Among them, G gate is the gate vector, H is the number of attention heads, θ h is the gating weight output by the h-th attention head, is the output of the hth attention head. The output of each attention head is weighted by the gating vector to achieve focusing and fusion of key features.

[0083] Specifically, a gating mechanism is introduced into the harmful multi-head attention mechanism architecture in step S3 to selectively fuse the outputs of different attention heads. The gating mechanism calculates a gating vector and weights the outputs of each attention head, focusing on key features and suppressing irrelevant information. The gating weight θh is dynamically adjusted based on the importance of each attention head's output and mapped to the interval [0, 1] using a Sigmoid function to ensure that the gating vector's weighting of each output is reasonable and effective. This selective fusion approach adaptively adjusts the contribution of each attention head based on data characteristics, improving the model's ability to handle complex data and enabling the weighted attention feature vector to focus more closely on key information relevant to fire risk assessment, thereby enhancing the accuracy and robustness of the assessment.

[0084] Preferably, in step S5, when simulating risk diffusion on the risk propagation network, the evacuation influence coefficient is introduced in combination with the inhibitory effect of personnel evacuation behavior on risk diffusion, and the speed and scope of risk diffusion are corrected. The corrected risk diffusion calculation model is as follows:

[0085] S new =S old ·(1-δ·Evac eff )

[0086] Among them, S new is the risk diffusion state after correction, S old is the original risk diffusion state, δ is the evacuation impact weight coefficient, Evac eff It is an indicator of personnel evacuation efficiency.

[0087] Specifically, in the risk diffusion simulation process of step S5, the inhibitory effect of personnel evacuation behavior on risk diffusion is taken into account, and the evacuation influence coefficient is introduced to correct the risk diffusion state. The evacuation influence weight coefficient δ is used to adjust the degree of influence of personnel evacuation behavior on risk diffusion, and its value is determined according to factors such as the configuration of evacuation facilities and personnel evacuation capacity in the actual environment. The personnel evacuation efficiency index quantifies the evacuation capacity of personnel in the event of a fire, including aspects such as evacuation speed and rationality of evacuation paths. Through this correction, the role of personnel evacuation in risk control in actual fire scenarios can be more realistically reflected, making the risk diffusion simulation results closer to the actual situation, providing a scientific basis for formulating more effective evacuation strategies and risk prevention and control measures, and improving the fire safety level of non-key units.

[0088] like Figure 2 As shown in the figure, the non-key unit fire risk assessment method integrating human factors analysis is implemented through different units, including:

[0089] The data collection and division unit is used to build a time series analysis model for fire human factors analysis based on historical fire data, divide each area within non-key units into multiple evaluation subspaces, and collect dynamic data on human factors in different time series;

[0090] A feature fusion unit, connected to the data acquisition and division unit, is used to extract various fire risk parameters in the assessment subspace and perform feature fusion on them with the human factor data in the corresponding time series to form a fused feature vector;

[0091] An attention calculation unit, connected to the feature fusion unit, is used to construct a harmful multi-head attention mechanism architecture, input the fused feature vector and generate an attention weighted feature vector;

[0092] a risk factor determination unit connected to the attention calculation unit, configured to deduce the changing trend of human factors based on the attention weighted feature vector and utilize the fire human factor analysis time series analysis model, and determine the risk impact factor of each evaluation subspace in combination with the fire risk parameter;

[0093] a risk diffusion simulation unit, connected to the risk factor determination unit, for establishing a risk propagation network and performing risk diffusion simulation based on risk impact factors and node connection weights;

[0094] The comprehensive assessment unit is connected to the risk diffusion simulation unit and is used to determine the overall fire risk level of non-key units by integrating the risk diffusion simulation results, the fire human factor analysis time series analysis model deduction results and the attention weighted feature vector.

[0095] Through multi-dimensional technological innovation, the present invention effectively overcomes the shortcomings of the background technology and demonstrates significant advantages. Traditional assessment methods do not adequately consider the dynamic changes of human factors. This method constructs a time series analysis model for fire human factors analysis. Based on historical fire data, it collects dynamic data such as human behavior and operational errors in different time series, and uses the autoregressive integrated moving average model structure to fit and predict the changing patterns of human factors. This innovation enables the assessment process to accurately capture the evolution trend of human factors in the time dimension, breaking through the limitations of traditional static assessments, making the assessment results more in line with actual risk changes, and greatly improving the ability to predict fire risks.

[0096] To address the insufficient integration of multidimensional risk parameters and human factors, this method uses a feature fusion model to deeply integrate fire risk parameters such as the frequency of electrical equipment use and the amount of combustible material accumulated with human factor data. Through a multi-head attention mechanism architecture, it explores the potential connections between these elements from multiple perspectives and generates attention-weighted feature vectors. This fusion approach breaks away from the traditional model of isolated parameter analysis in assessments, achieving organic integration of data and more comprehensively reflecting the complex characteristics of fire risks in non-key units. Furthermore, by introducing a gating mechanism to selectively fuse the outputs of different attention heads, it further focuses on key features and improves assessment accuracy.

[0097] In addition, this method establishes a risk propagation network, sets node connection weights based on the physical location of the subspace and the flow of personnel, and makes corrections based on the impact of personnel evacuation behavior on risk diffusion. The overall risk level is derived by combining a comprehensive assessment model. This process comprehensively considers the propagation characteristics of fire risk in space and personnel flow, as well as the inhibitory effect of personnel evacuation behavior on risk, making the assessment results more systematic and reliable. By introducing a variety of innovative model formulas to quantify the influencing factors of each link, a scientific basis is provided for risk assessment. Compared with traditional methods, it has obvious advantages in accuracy, comprehensiveness, and dynamic adaptability, providing efficient and reliable technical support for fire prevention and control in non-key units.

[0098] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "disposed," "installed," "connected," "connected," and "fixed" should be understood in a broad sense. For example, they may refer to fixed connections, detachable connections, or integral connections; they may refer to mechanical connections or electrical connections; they may refer to direct connections or indirect connections through an intermediate medium; and they may refer to internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0099] While embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A fire risk assessment method for non-key units that integrates human factors analysis is characterized by: The steps include: Step S1: Based on historical fire data, a time series analysis model for fire human factors analysis is constructed, including different dimensions such as personnel behavior, operational errors, and management negligence. Each area within the non-key unit is divided into multiple evaluation subspaces, and dynamic data of human factors in different time series are collected for each evaluation subspace; Step S2: Extract various parameters related to fire risk in the assessment subspace, including the frequency of use of electrical equipment, the amount of combustible material accumulation, the response time of firefighting facilities, and the density of personnel, and fuse the parameters with the human factor data in the corresponding time series to form a fused feature vector; Step S3: constructing a harmful multi-head attention mechanism architecture, inputting the fused feature vector into the harmful multi-head attention mechanism architecture, capturing the relationship between the elements in the fused feature vector from different angles through multi-head parallel computing, and generating an attention weighted feature vector; Step S4: Based on the attention weighted feature vector, a fire human factor analysis time series analysis model is used to deduce the change trend of human factors in different evaluation subspaces at each time node, and the risk impact factor of each evaluation subspace at the corresponding time node is determined in combination with various fire risk parameters; Step S5: Establish a risk propagation network, use each evaluation subspace as a network node, set node connection weights based on the physical location relationship between subspaces and personnel flow paths, and perform risk diffusion simulation on the risk propagation network based on the risk impact factors and node connection weights; Step S6: Determine the overall fire risk level of non-key units by synthesizing the risk diffusion simulation results, combining the deduction results of the fire human factor analysis time series analysis model and the attention weighted feature vector processed by the harmful multi-head attention mechanism.

2. The non-key unit fire risk assessment method integrating human factor analysis according to claim 1 is characterized in that: In step S2, the parameters are fused with the human factor data in the corresponding time series, using the following fusion model formula: Among them, F vec is the fusion feature vector, n is the total number of parameters and factor data, α i is the weight coefficient of the fusion of the i-th group of parameters and factor data, P i is the i-th fire risk parameter vector, H i is the human factor vector of the i-th fire in the corresponding time series, Represents a vector fusion operation.

3. The non-key unit fire risk assessment method integrating human factor analysis according to claim 1 is characterized in that: In step S3, in the harmful multi-head attention mechanism architecture, the calculation for each attention head uses the following formula: Among them, A head is the output of a single attention head, Q, K, and V are respectively the query vector, key vector, and value vector obtained by different linear transformations of the input fusion feature vector, and W q 、W k 、W v is the corresponding learnable weight matrix, d k is the dimension of the key vector, σ is the activation function, which is used to normalize the calculation results. Finally, the outputs of multiple attention heads are concatenated and linearly transformed to obtain the attention weighted feature vector.

4. The non-key unit fire risk assessment method integrating human factor analysis according to claim 1 is characterized in that: In step S4, the fire human factor analysis time series analysis model is used to deduce the changing trends of human factors in different assessment subspaces at each time node, and the risk impact factor of each assessment subspace at the corresponding time node is determined in combination with various fire risk parameters. The following calculation model is used: Among them, R factor is the risk impact factor, β is the trend impact coefficient, ΔH trend is the changing trend of human factors within the time interval ΔT, m is the number of selected fire risk parameters, P j is the j-th fire risk parameter value, γ j is the weight index of the j-th fire risk parameter.

5. The non-key unit fire risk assessment method integrating human factor analysis according to claim 1 is characterized in that: In step S5, the risk propagation network is established, each evaluation subspace is used as a network node, and the node connection weight is set according to the physical position relationship between the subspaces and the personnel flow path. The following weight calculation model is used: Among them, W conn is the node connection weight, Dis space is the physical distance between the two evaluation subspaces, s total is the total number of evaluation subspaces, ρ is the distance weight adjustment coefficient, Flow rate is the personnel flow rate between the two subspaces.

6. The non-key unit fire risk assessment method integrating human factor analysis according to claim 1 is characterized in that: In step S6, the comprehensive risk diffusion simulation results are combined with the deduction results of the fire human factor analysis time series analysis model and the attention weighted feature vector processed by the harmful multi-head attention mechanism to determine the overall fire risk level of non-key units. The following comprehensive assessment model is used: Among them, R level is the overall fire risk level, S diffuse is the quantitative value of the risk diffusion simulation result, T trend A is the quantitative value of the deduction result of the time series analysis model for fire human factors analysis, vec is the quantized value of the attention weighted feature vector, ω1, ω2, and ω3 are the weight coefficients of the corresponding results, and Map is the mapping function that maps the calculation results to the corresponding risk level interval.

7. The non-key unit fire risk assessment method integrating human factor analysis according to claim 1 is characterized in that: In step S1, the fire human factor analysis time series analysis model adopts an autoregressive integrated moving average model structure to model the collected dynamic data of human factors in different time series, and fits and predicts the dynamic change law of human factors through iterative optimization of model parameters.

8. The non-key unit fire risk assessment method integrating human factor analysis according to claim 1 is characterized in that: In step S3, a gating mechanism is introduced into the harmful multi-head attention mechanism architecture to selectively fuse the outputs of different attention heads. The gating calculation model is as follows: Among them, G gate is the gate vector, H is the number of attention heads, θ h is the gating weight output by the h-th attention head, is the output of the hth attention head. The output of each attention head is weighted by the gating vector to focus and fuse the key features.

9. The non-key unit fire risk assessment method integrating human factor analysis according to claim 1 is characterized in that: In step S5, when simulating risk diffusion on the risk propagation network, the evacuation influence coefficient is introduced to correct the speed and range of risk diffusion in combination with the inhibitory effect of personnel evacuation behavior on risk diffusion. The corrected risk diffusion calculation model is as follows: S new =S old ·(1-δ·Evac eff ) Among them, S new is the risk diffusion state after correction, S old is the original risk diffusion state, δ is the evacuation impact weight coefficient, Evac eff It is an indicator of personnel evacuation efficiency.

Images