Emergency evacuation place intelligent management system and method

Abstract

The invention relates to the technical field of chemical plant area intelligent evacuation, and discloses an emergency evacuation place intelligent management system and method, and the method comprises the steps: constructing a chemical risk semantic unified representation space through a risk semantic mapping algorithm, and generating a mapping relation between a risk state and semantic expression; applying a feature portrait algorithm to generate an employee feature representation model; on the basis of the chemical risk semantic personalized adaptation model, personalized semantic guidance for employees with different characteristics is generated; generating a multilevel chemical risk semantic model by using a regional hierarchical clustering algorithm; processing real-time position and risk change data through a collective intelligent optimization algorithm, and calculating a dynamic evacuation path planning scheme; differentiated safety management based on employee characteristics, hazard source distribution and semantic understanding is realized, the problem of low efficiency caused by one-step instructions in traditional evacuation management is solved, and the emergency evacuation efficiency and safety of the chemical plant area are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent evacuation of chemical plant areas, and more specifically, to an intelligent management system and method for emergency evacuation sites. Background Art

[0002] As a high-risk industrial site, chemical plants face potential safety accident risks such as toxic gas leakage, fire and explosion. When an emergency occurs, the existing emergency evacuation management technology in chemical plants has the following main problems: Unified evacuation instructions cannot adapt to the risk differences in different areas. High-risk areas and low-risk areas receive the same instructions, resulting in confusion in evacuation priorities. Failure to consider differences in individual employee characteristics (such as responsibilities, training levels, physical conditions, etc.); employees with different cognitive abilities may interpret the same instructions differently; The semantic expression is out of touch with the actual risk situation, and there is a lack of dynamic association mechanism between risk and evacuation guidance; It is impossible to optimize collective evacuation behavior, and it is difficult to coordinate individual behavior with overall safety goals.

[0003] Therefore, there is an urgent need for an emergency evacuation management method that can provide personalized semantic guidance based on risk status and employee characteristics to solve the above technical problems. Summary of the invention

[0004] The present invention provides an intelligent management system and method for emergency evacuation sites, which solves the technical problem that the evacuation management systems in the related art generally adopt unified evacuation instructions and routes, all employees receive the same safety guidance, and lack personalized processing for individual characteristics of different employees and risk differences in different regions.

[0005] The present invention provides an intelligent management method for emergency evacuation sites, comprising the following steps: The risk semantic mapping algorithm is used to construct a unified semantic representation space for chemical industry risk, generating a mapping relationship between risk status and semantic expression; Apply feature profiling algorithms to process individual employee data and generate employee feature representation models; Generate personalized semantic guidance for employees with different characteristics based on the personalized adaptation model of chemical risk semantics; Apply regional hierarchical clustering algorithm to process plant layout and personnel distribution data to generate a multi-level chemical risk semantic model; Process real-time location and risk change data through collective intelligent optimization algorithms to calculate dynamic evacuation path planning solutions; Among them, the chemical risk semantic personalized adaptation model calculates the optimal semantic expression based on the guiding utility function, and the guiding utility function comprehensively considers the safety function, understanding function, behavior deviation and risk space diffusion characteristics.

[0006] Furthermore, the risk semantic mapping algorithm includes the following steps: Collect and analyze data on various risk events that may occur in the chemical plant area and build a risk status set: ; in , , Respectively represent the first, second, and kth risk states (such as toxic gas leakage, fire, and explosion), Indicates the number of risk status categories; Based on the characteristics of risk events, construct feature vectors: ; in Indicates risk status The characteristic vector of , , Represent the first, second, and qth features respectively, represents the dimension of the feature vector; Using spatial distribution modeling technology, calculate the risk spatial distribution function: ; in Indicates risk status In Location and time The intensity distribution of Indicates location, Indicates time; Create a set of semantic expressions based on expert knowledge and security specifications: ; in represents a set of semantic expressions, , , They represent the first, second, and mth security guidance semantics respectively, Indicates the number of types of semantic expressions; Use the mapping algorithm to create a function: ; in represents the mapping function, represents the risk state set, Represents a set of semantic expressions, which represents the risk status Mapping to corresponding semantic expressions .

[0007] Furthermore, the feature profiling algorithm includes the following steps: Collect and organize individual employee data, including job responsibilities, training records, professional background, and physical condition, to build a set of employee characteristics; Through data analysis methods, feature vectors are constructed for each type of employee characteristics; Use deep neural networks to learn employee feature representations and map high-dimensional features to low-dimensional semantic understanding space; Through the analysis of historical emergency drill data, a behavior prediction function is established.

[0008] Furthermore, the chemical risk semantic personalized adaptation model adopts a multi-objective optimization method to calculate the optimal personalized semantic guidance for each employee and the current risk status through the following objective function: ; in represents the optimal personalized semantic guidance, Represented in the semantic expression set Finding the optimal semantic guidance , is the guiding utility function, which represents the risk state Downward, the characteristics are Provide semantic guidance to employees The validity value of Represents an element in a semantic expression set, Indicates the employee number.

[0009] Furthermore, the calculation formula of the guidance utility function is: ; in is the guiding utility function, , , , is the weight coefficient; is a safety function, which indicates the safety level of employees under the current risk status; is the understanding function, which represents the employee's understanding of the semantic expression; KL divergence is a measure of the degree of deviation between individual behavior and optimal behavior; is the attenuation function of risk and distance, reflecting the spatial diffusion characteristics of chemical risks. Indicates the risk status of the employee's location, Indicates the distance value from the employee to the risk source. Represents employee behavioral decisions.

[0010] Furthermore, the regional hierarchical clustering algorithm comprises the following steps: Analyze the layout data of chemical plant area and divide it into different functional areas; Collect and analyze risk characteristic data of each region and construct regional risk function; Integrate employee distribution data to construct the distribution of employee group characteristics in each region and their corresponding weight values; Create a region-specific semantic template library based on the characteristic data and safety regulations of each region; The swarm optimization algorithm is applied to calculate the optimal semantic expression of the region.

[0011] Furthermore, the optimal semantic expression of the region is calculated by the following formula: ; in, represents the optimal semantic expression of the region, Indicates area A semantic template library, Indicates area The distribution of employee group characteristics within The characteristic is of employees in the region The weight value in Indicates area In time risk status, represents the guiding utility function, Indicates the area The sum of all employee characteristic types in Indicates Regions, The region number.

[0012] Furthermore, the collective intelligence optimization algorithm comprises the following steps: Establish a global optimization objective function to incorporate individual safety and overall evacuation efficiency into the same framework; Use multi-agent modeling technology to build a collective behavior model of employees; Apply information entropy calculation method to construct collective behavior entropy function; Using dynamic programming algorithm, the optimal evacuation path for each employee is calculated.

[0013] Furthermore, the global optimization objective function is: ; in, represents the global optimization objective, is the total number of employees, Indicates the employee number. Indicates employees The guiding utility value of Indicates employees Features, Indicates employees The risk status of the location, Indicates employees Personalized semantic guidance, represents the collective behavior entropy function, is the balance parameter; represents the collective behavior model of employees, , , They represent the behavioral decisions of the 1st, 2nd, and Nth employees respectively.

[0014] An intelligent management system for emergency evacuation sites, used to implement the above-mentioned intelligent management method for emergency evacuation sites, comprising: The risk semantic mapping module is used to construct a unified risk semantic representation space and generate a mapping relationship between risk status and semantic expression; The feature profiling module is used to process individual personnel data and generate a personnel feature representation model; The personalized semantic adaptation module is used to calculate the optimal semantic expression based on the guidance utility function and generate personalized semantic guidance for people with different characteristics; The regional hierarchical clustering module processes the site layout and personnel distribution data to generate a multi-level risk semantic model for calculating the optimal semantic expression of the region; The collective intelligence optimization module is used to process real-time location and risk change data and calculate dynamic evacuation path planning solutions. The module is based on the global optimization objective function, integrates individual safety and overall evacuation efficiency, and uses a dynamic programming algorithm to calculate the optimal evacuation path for each person: The beneficial effects of the present invention are as follows: the risk-semantic mapping model constructed by the risk semantic mapping algorithm realizes the accurate conversion of chemical risk status to semantic expression, solves the technical problem of the separation of risk expression and semantic understanding in the traditional chemical safety system, and improves the accuracy and pertinence of safety guidance; The employee feature representation model generated by the feature profiling algorithm achieves accurate quantification of individual characteristics of employees in different positions, provides data support in the personnel dimension for personalized safety guidance, and solves the technical defect of ignoring individual differences of employees in traditional systems. The personalized adaptation model of chemical risk semantics built by multi-objective optimization algorithm integrates safety, understanding and behavior deviation into a unified framework for optimization, solving the technical contradiction of the difficulty in balancing safety and understanding in traditional systems, and enabling the system to automatically calculate the optimal personalized safety guidance plan; The multi-level chemical risk semantic model generated by the regional hierarchical clustering algorithm realizes differentiated safety guidance based on regional characteristics and personnel distribution, solves the technical contradiction of "unified evacuation instructions and risk differences in different regions" in the traditional system, and improves the regional adaptability of evacuation instructions; The dynamic evacuation path planning scheme calculated by the collective intelligent optimization algorithm realizes the optimal control of the global evacuation behavior, solves the technical problem that individual optimization in the traditional system may lead to overall suboptimality, and enables employees in different areas to form a coordinated and orderly evacuation order. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a flowchart of the main steps of the semantic-risk fusion personalized emergency evacuation management method for chemical plants of the present invention; Figure 2 is a flow chart of the sub-steps of the risk semantic mapping algorithm of the present invention; Figure 3 It is a sub-step flow chart of the feature profiling algorithm of the present invention; Figure 4 It is a sub-step flow chart of the chemical industry risk semantic personalized adaptation model of the present invention; Figure 5 is a flow chart of the sub-steps of the regional hierarchical clustering algorithm of the present invention; Figure 6 It is a sub-step flow chart of the collective intelligence optimization algorithm of the present invention. DETAILED DESCRIPTION

[0016] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that the discussion of these implementations is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the present specification. Various examples may omit, replace, or add various processes or components as needed. In addition, the features described in some examples may also be combined in other examples.

[0017] At least one embodiment of the present invention discloses an intelligent management method for emergency evacuation sites, such as Figure 1 As shown, the following steps are included: Step 100, constructing a unified chemical risk semantic representation space using a risk semantic mapping algorithm, and generating a mapping relationship between risk status and semantic expression; like Figure 2 As shown, specifically including: Step 101, collect and analyze data on various risk events that may occur in the chemical plant area, and construct a risk status set: ; in , , Respectively represent the first, second, and nth risk states (such as toxic gas leakage, fire, and explosion). Indicates the number of risk status categories; Step 102, constructing a feature vector based on the risk event characteristics; ; in , , They represent the first, second, and qth characteristics of the risk status (including risk type, severity, diffusion speed, and impact range), Indicates the number of types of features, Indicates The characteristic vector of the risk state, The value range is 1 to ; Step 103, using spatial distribution modeling technology, calculate the risk spatial distribution function: ; in Indicates risk status In Location and time The intensity distribution of Step 104: Create a semantic expression set based on expert knowledge and security specifications, including various possible security guidance semantics: ; in , , represent the first, second, and mth semantic expressions respectively (e.g., "evacuate immediately", "evacuate along the designated route", "close the valve"). Indicates the number of types of semantic expressions, Indicates the type of semantic expression; Step 105, using a mapping algorithm to establish a function to map the risk status to the corresponding semantic expression: ; in Indicates mapping the risk status to the corresponding semantic expression, represents the set of risk states, Represents a set of semantic expressions; the risk state set Each element in Associated with a set of semantic expressions The corresponding elements in , through the attention matrix The association strength between different risk states and semantic expressions is quantified, taking into account the risk feature vector , risk spatial distribution function And the hierarchical structure of semantic expression is essentially a conditional probability distribution, which can determine the most suitable semantic guidance for describing and responding to the current risk status according to the risk type, severity, and diffusion speed, and realize the intelligent conversion of risk status to safety guidance.

[0018] Among them, the specific steps of the risk semantic mapping algorithm are: Step 1051, first use the bidirectional attention mechanism to transform the risk state feature vector Align with the semantic expression features and calculate the attention matrix. The calculation formula is: ; in is the attention matrix (the matrix Line Column Elements Indicates risk status and semantic expression The specific structure is: Row index: represents different risk status ,in From 1 to (Total number of risk states); Column index: represents different semantic expressions ,in From 1 to (total number of semantic expressions); and are risk status feature vector and semantic expression feature vector respectively, is the dimension of the feature vector, is the normalization function.

[0019] Step 1052: Use a spatial information propagation method based on a graph convolutional network to perform spatial distribution of risk To model: Discretize the factory space into nodes, and the edges represent the spatial connectivity; Through the time-series graph convolution layer, the dynamic process of risk spreading from the source to the surrounding area is captured, and the trend of risk intensity changes is identified; Attention weighting is performed on node features to highlight high-risk areas; Using the message transmission mechanism, establish an inter-regional risk transmission network and predict the risk transmission path; Finally, a location-sensitive risk representation is generated (including the spatial distribution of risk, spatiotemporal evolution characteristics, risk intensity at key locations, and risk propagation associations between regions); Step 1053, combining semantic knowledge graph technology to construct semantic expression The association relationship between them forms a hierarchical semantic expression structure; Step 1054, using the probabilistic graphical model method, learn the conditional probability distribution between the risk state and the semantic expression, and generate a mapping function .

[0020] This step inputs the risk event data of the chemical plant area (including risk status, semantic expression and spatial distribution) and outputs a risk-semantic mapping model (the model includes risk status feature vector, semantic expression feature vector, attention matrix, risk space distribution function and semantic expression). This model can accurately map various risk states to corresponding semantic expressions, providing a basic representation space for subsequent personalized safety guidance.

[0021] Step 200, applying a feature profiling algorithm to process employee individual data to generate an employee feature representation model; like Figure 3 As shown in the figure, the specific steps of the feature profiling algorithm are: Step 201, collect and organize individual employee data, including job responsibilities, training records, professional background and physical condition, and construct an employee feature set; Step 202, construct a feature vector for each type of employee characteristics through data analysis method: Adopt multimodal feature fusion technology to extract features from multi-source data (including employee basic data, historical behavior data, cognitive test data), and construct employee feature vectors through data analysis methods. (including responsibility level, training level, professional background, physical condition and work area) Use deep neural networks to learn employee feature representations and map high-dimensional features to low-dimensional semantic understanding space; Step 203, generating a comprehension ability function based on the cognitive assessment test results; Based on fuzzy set theory, a comprehension ability function is constructed to quantify the degree of understanding of different types of employees on different semantic expressions. The calculation formula is: ; in The characteristic is of employees use semantic expression ability to understand, is the number of types of employee characteristics, is the number of types of semantic expressions, For employees' possible behaviors, represents the summation symbol, Indicates that the feature is The sum of all possible behaviors of employees, Characterized by The set of possible employee behaviors.

[0022] Step 204: Establish a behavior prediction function by analyzing historical emergency drill data: Combined with the reinforcement learning method, the behavior prediction model is trained through historical exercise data to generate a behavior prediction function, which can preview the actual effects of various instructions before a dangerous event occurs, so as to select the optimal personalized guidance strategy in actual emergency response. The calculation formula is: ; in, The characteristic is of employees are at risk Receive semantic guidance After taking action The conditional probability of is the state-action value function, which is learned by the reinforcement learning algorithm through historical practice data. represents the exponential function, The characteristic is The set of possible employee behaviors, represents the sum of all possible actions, For semantic expression, For employees' possible behaviors, For all possible behaviors, is a risky state.

[0023] Through these data analyses, an accurate employee characteristic characterization model (including responsibility level, training level, professional background, physical condition and work area) was constructed, providing a basis for subsequent personalized safety guidance.

[0024] Step 300, generating personalized semantic guidance for employees with different characteristics based on the chemical risk semantic personalized adaptation model; like Figure 4 As shown, the specific implementation process is as follows: Step 301, generate questions for personalized semantic guidance of employees and construct an objective function: First, for employees The personalized semantic guidance generation problem is constructed to construct the objective function: ; in For optimal personalized semantic guidance; is the guiding utility function, which represents the risk state The downward feature is Provide semantic guidance to employees effectiveness; is a preset set of semantic expressions, For employees Characteristic portrait of For employees The risk status of the location, Indicates the selection of the guiding utility function in the preset semantic expression set Maximum semantic expression.

[0025] Step 302, setting the guidance utility function, using a multi-factor weighting method: Guidance Utility Function The calculation formula is: ; in , , , is the weight coefficient, which is determined by model training; is a safety function, which indicates the safety level of employees under the current risk status; is the understanding function, which indicates the employee’s understanding of the semantic expression; is the KL divergence, which measures the degree of deviation of individual behavior from the optimal behavior; Indicates that employees take actions under given conditions probability; represents the probability distribution of the optimal behavior; is the attenuation function of risk and distance, reflecting the spatial diffusion characteristics of chemical risks, where Indicates the distance from the employee to the risk source.

[0026] Step 303, solving the optimal semantic guidance by a multi-objective optimization method: Based on the objective function constructed in step 301, the optimal semantic guidance is solved by a multi-objective optimization method. , the specific solution process is as follows: First, through the risk-semantic mapping model Filter out the current risk status Semantic expressions with high relevance, building a set of candidate semantic expressions ( ), ensure that the semantic expressions in the set can accurately describe the characteristics, degree of harm and countermeasures of the current risk status, and the number of semantic expressions in the set does not exceed the preset threshold (the threshold is 10% of the number of semantic expressions in the preset semantic expression set); For each candidate semantic expression , calculate its guiding utility value ; Select the semantic expression with the largest utility value as the optimal personalized semantic guidance .

[0027] Step 304, calculate the optimal personalized safety guidance plan based on employee characteristics and risk status: Precise guidance for employees with different characteristics and different risk status, including: The core of the security guidance solution is to select the optimal semantic expression: Evacuation instructions (e.g., "evacuate immediately," "evacuate along designated routes"); Operating instructions (e.g., “Close the valve,” “Put on a gas mask”); Risk warnings (such as "toxic gas leakage", "explosion risk"); Action recommendations (e.g., “keep a low profile,” “avoid the north side”); Personalized semantic adaptation: the guidance content received by each employee will be adjusted according to their personal characteristics: Professional terminology level: Use accurate professional terms for employees with strong professional backgrounds and use simplified expressions for ordinary employees; Level of detail: Adjust the level of detail of instructions based on employee training levels; Expression method: Consider the employees’ understanding ability and choose the appropriate expression difficulty and complexity; Priority information: Emphasize different action priorities based on employee responsibilities; Risk context related information. The safety guidance plan will contain specific information related to the current risk status: Risk type description: clearly identify the current risk type (such as gas leak, fire) Risk level indication: Indicates the risk level and urgency Spatial distribution of risks: providing information on the location and diffusion direction of risk sources Time evolution tips: Warning of possible development trends of risks Behavior guidance strategy. Based on the behavior prediction model, the guidance plan will include strategies that can guide employees to take the best behavior: Clear instructions for action: clearly state the specific actions that need to be performed Sequence prompt: Provides the sequence of multi-step operations Expected results: Describe the expected results after following the instructions Avoidance Tips: Clearly indicate wrong actions that should not be taken Security measures, based on security function evaluation, the guidance plan will include: Personal protection tips: reminders on what protective equipment to use Dangerous area signs: clearly mark areas that should be avoided Safe Route Suggestion: Recommends the safest evacuation route based on your current location Support Information: Provides locations or contact information where you can get help.

[0028] The input data of this step are the risk-semantic mapping model of step 100 and the employee characteristic representation model of step 200, and the output is a chemical risk semantic personalized adaptation model (including risk-semantic mapping model, employee characteristic representation model, guidance utility function, objective function and optimal semantic guidance). This model can automatically calculate the optimal personalized safety guidance plan according to employee characteristics and risk status.

[0029] Step 400, applying a regional hierarchical clustering algorithm to process plant layout and personnel distribution data to generate a multi-level chemical risk semantic model; Based on the chemical risk semantic personalized adaptation model output in step 300, such as Figure 5 As shown, regionalized differential security semantics is implemented through the following operations: Step 401, analyze the layout data of the chemical plant area and divide it into different functional areas: ; in It is a collection of chemical plant areas. , , Respectively represent the first, second, and Areas in a chemical plant (area types such as high-risk production areas, control rooms, and storage areas).

[0030] Step 402, collect and analyze the risk characteristic data of each region, and construct a regional risk function: ; in Indicates area In time The comprehensive risk status value of Indicates area In time The comprehensive risk status value of represents the time step; Step 403, integrate the employee distribution data, and construct the characteristic distribution of employee groups in each region and their corresponding weight values: ; in For Region The set of employee group characteristics, , , Respectively represent the first, second, and The characteristics of the employee groups in the chemical plant area The total number of characteristics of the chemical plant area employee group; The corresponding weight value is , Characteristics of the employee group. Characterized by of employees in the region The weight value in .

[0031] Step 404: Create a region-specific semantic template library based on the characteristic data of each region and the safety specification requirements: ; in For Region The semantic expression set of , , Respectively represent the first, second, and The regional semantic expression of a chemical plant area, is the total number of semantic expressions of the chemical plant area; Step 405, apply the group optimization algorithm to calculate the optimal semantic expression of the region: ; in For Region The optimal semantic expression of Characterized by of employees in the region The weight value in is the guidance utility function constructed in step 302, For Region In time The comprehensive risk status value of For semantic expression, For Region A collection of employee group characteristics.

[0032] The input of this step is the plant layout data, regional risk characteristic data, personnel distribution data, and the chemical risk semantic personalized adaptation model from step 300. It outputs a multi-level chemical risk semantic model (including regional function division, time-varying risk state function, regional employee group characteristic distribution and weight, regional exclusive semantic template library and regional optimal semantic expression). The model can automatically generate regional differentiated safety guidance plans based on different regional characteristics and personnel distribution.

[0033] Step 500, processing real-time location and risk change data through a collective intelligence optimization algorithm to calculate a dynamic evacuation path planning solution; This step integrates all the above models, such as Figure 6 As shown in the figure, the globally optimized evacuation management is achieved through the following operations: Step 501, establish a global optimization objective function, incorporating individual safety and overall evacuation efficiency into the same framework: The improved particle swarm optimization algorithm is used to construct the global optimization objective function: ; in is the global optimization objective function, is the guidance utility function constructed in step 302, is the total number of employees, is the employee number, For employees Characteristic portrait of For employees The risk status of the location, For employees Receive personalized semantic guidance, is the collective behavior entropy function, which represents the order value of the overall behavior. is a balance parameter.

[0034] Step 502, using multi-agent modeling technology to construct a collective behavior model of employees; Among them, each agent represents an employee, which has the ability to perceive the environment, receive instructions, make decisions and execute actions. The collective behavior set of employees is: ; in Collective behavior collection for employees, , , They represent the behavioral decisions of the first, second, and Nth employees respectively (such as the selected evacuation path and action speed).

[0035] Step 503, applying the information entropy calculation method to construct a collective behavior entropy function: ; in is the collective behavior entropy function; The characteristic is of employees are at risk Receive semantic guidance Post-selection behavior The joint probability of The characteristic is The set of possible employee behaviors; Represents a set of employee feature types; Indicates risk status; It represents semantic expression; Indicates possible employee behavior; Indicates the employee characteristic type; represents the logarithmic function; represents the risk state set constructed in step 400; represents the semantic expression set constructed in step 300; The characteristic is The set of possible employee behaviors; Represents the employee feature type set constructed in step 200.

[0036] Step 504, using a dynamic programming algorithm, calculate the optimal evacuation path for each employee: ; in Indicates employees The optimal evacuation path, Indicates employees Optional evacuation routes, represents the path safety function, represents the path efficiency function, represents the congestion function, represents the employee collective behavior set constructed in step 502, Indicates time.

[0037] Step 505, dynamically adjust the evacuation plan according to the real-time personnel location and risk changes: Dynamic adjustment of evacuation plans includes personalized optimal path planning, global coordination and scheduling strategies, congestion avoidance mechanisms, and real-time adjustment plans: Among them, the global coordination scheduling strategy is implemented based on the global optimization objective function to balance individual safety and collective efficiency; Congestion avoidance mechanism uses congestion function Calculate the risk of path congestion; The real-time adjustment plan is based on a dynamic programming algorithm, which seeks the optimal balance between safety, efficiency and congestion to achieve dynamic response to channel load balancing and changes in risk status.

[0038] The input of this step is real-time personnel location data, risk status change data and the aforementioned constructed models (including the risk-semantic mapping model in step 100, the employee feature representation model in step 200, the chemical risk semantic personalized adaptation model in step 300, and the multi-level chemical risk semantic model in step 400); the output is a dynamic evacuation path planning scheme; the scheme can adjust the optimal evacuation path for each employee in real time, avoid the diffusion path of the hazard source, and reserve channels for emergency rescue to achieve the global optimal evacuation effect.

[0039] An intelligent management system for emergency evacuation sites, used to implement the above-mentioned intelligent management method for emergency evacuation sites, comprising: The risk semantic mapping module is used to construct a unified risk semantic representation space and generate a mapping relationship between risk status and semantic expression; The feature profiling module is used to process individual personnel data and generate a personnel feature representation model; The personalized semantic adaptation module is used to calculate the optimal semantic expression based on the guidance utility function and generate personalized semantic guidance for people with different characteristics; The regional hierarchical clustering module processes the site layout and personnel distribution data to generate a multi-level risk semantic model for calculating the optimal semantic expression of the region; The collective intelligent optimization module is used to process real-time location and risk change data and calculate dynamic evacuation path planning solutions. The module is based on the global optimization objective function, integrates individual safety and overall evacuation efficiency, and uses a dynamic programming algorithm to calculate the optimal evacuation path for each person.

[0040] Here, the present invention provides an implementation example: This implementation method has been actually deployed and tested in the production plant of a large petrochemical enterprise. The plant covers an area of ​​about 5 square kilometers, including high-risk production areas, central control rooms, storage areas, office areas, and different functional areas. The total number of employees is about 2,500, mainly including operators, technicians, managers, and different types of external visitors. The specific application process and effect verification of this implementation method in this petrochemical enterprise are described in detail below.

[0041] The petrochemical enterprise is mainly engaged in the production and processing of petrochemical products. There are many chemical hazards in the plant area, including flammable and explosive substances, toxic gases and corrosive substances. According to historical records, the enterprise has experienced many safety accidents, including chlorine gas leaks and tank fires, exposing many problems in traditional emergency evacuation management methods.

[0042] Through the analysis of 18 safety accidents and 42 emergency drills in the company in the past five years, it is found that the main problems of traditional emergency evacuation management are as shown in Table 1: Table 1: Analysis of the main problems and causes of traditional emergency evacuation management

[0043] Based on the above problem analysis, the company introduced the semantic-risk fusion personalized emergency evacuation management method for chemical plant areas in this implementation mode. By constructing a chemical risk semantic personalized adaptation model (CRSPAM), it realized differentiated safety management based on employee characteristics and risk distribution.

[0044] Firstly, a risk semantic mapping model was constructed for the specific risk type of the petrochemical enterprise. By collecting and analyzing the enterprise's past safety accident data, expert knowledge and industry norms, the risk state set and semantic expression set were defined, and the mapping relationship between them was established.

[0045] Based on expert knowledge and safety specifications, a set of semantic expressions is created, and corresponding weights and applicable conditions are assigned to each type of semantic expression according to different risk types and severity.

[0046] In the application example, for the chlorine gas leakage scenario, the algorithm implementation process is as follows: Collect historical data and simulation data of chlorine gas leakage and extract feature vectors ; The gas diffusion model was used to calculate , get the chlorine concentration distribution at different locations and times; Filter semantic expressions suitable for chlorine gas leakage scenarios from the semantic expression set, such as "evacuate immediately", "wear a gas mask", and "evacuate in the upwind direction"; A mapping function is generated through an algorithm to map different risk states of chlorine leakage (such as initial leakage, rapid diffusion, and reaching dangerous concentration) to corresponding semantic expressions.

[0047] Some examples of semantic expression data are shown in Table 2: Table 2: Some examples of semantic expression data

[0048] Next, based on the employee structure characteristics of the company, the feature profiling algorithm was used to build an employee feature representation model. First, the 2,500 employees in the factory were classified and relevant data was collected, including division of responsibilities, training records, professional background, and years of work experience.

[0049] In the application example, the characteristic portrait modeling for personnel in different positions in a chemical plant is as follows: Extract feature vectors for different types of employees (operators, technicians, managers, and external visitors)

[0050] ; Standardized tests are used to determine how well different types of employees understand various semantic expressions, such as: ; The above formula indicates that the operator's understanding of the semantic expression "evacuate immediately" is 95%; Analyze historical drill data and build behavior prediction functions, such as: ; The above formula indicates that in the event of a chlorine gas leak, when receiving an "immediate evacuation" command, the probability that the operator will evacuate along the designated route is 92%.

[0051] Based on the employee classification, the cognitive assessment test was used to quantify the degree of understanding of different types of employees on various semantic expressions, as shown in Table 3: Table 3: Understanding of semantic expressions by different types of employees (0-100)

[0052] Based on the aforementioned risk semantic mapping model and employee characteristic representation model, a chemical risk semantic personalized adaptation model (CRSPAM) was constructed. This model calculates the optimal semantic expression for each employee by comprehensively considering safety, understanding, behavioral deviation and risk space diffusion characteristics through the guiding utility function.

[0053] In the application example, for a toxic gas leak in a chemical production area, the application process of the model is as follows: Get the current risk status from the risk-semantic mapping model Characterization of the leak, including leak location, gas type, and concentration; Get employees from the employee feature representation model Characteristic portrait , including the employee’s duties, training level, and location information; Calculating security functions , assess the employee's safety level under the current risk; Calculate the understanding function , assess the employee’s understanding of different semantic expressions; Computational behavior prediction , predict the actions that employees may take after receiving different semantic guidance; Calculating the risk decay function , assess the distance relationship between employees and risk sources; Taking all the above factors into consideration, the optimal semantic expression is solved through a multi-objective optimization algorithm. .

[0054] For example, for operators in the production area, the system may generate an instruction to "evacuate 300 meters to the northwest along Channel 2", while for technicians in the same area, it may generate an instruction to "close the southeast corner valve and evacuate along Channel 2", thus realizing personalized safety guidance based on employee characteristics.

[0055] In a chlorine gas leak emergency drill, the optimal semantic expressions of different types of employees in different positions calculated by the CRSPAM model are shown in Table 4: Table 4. Example of personalized semantic guidance generated by the CRSPAM model (chlorine gas leak scenario)

[0056] In order to better adapt to the characteristics of different regions, the company applied a regional hierarchical clustering algorithm, divided the entire plant into eight main functional areas according to the plant layout and functional divisions, and built an exclusive semantic template library for each area.

[0057] In the application example, for different functional areas of a large petrochemical enterprise, the implementation process of the algorithm is as follows: Based on the plant floor plan and functional division, the entire plant is divided into production area, control room, storage area, and office area; Conduct risk assessments for each area, including risk functions for production areas Consider the inventory of toxic substances, equipment pressure, and temperature to form a risk status value; Analyze the distribution of employees in each region. For example, the production area is mainly composed of operators (weight 0.7) and technicians (weight 0.3); Create a unique semantic template for each area. For example, the semantic template for the production area contains highly technical professional instructions, while the semantic template for the office area is mainly simple and easy to understand. When an accident occurs in a certain area, the system automatically calculates the optimal regional semantic expression based on the characteristics of the employee group in that area.

[0058] In the chlorine gas leakage scenario, the production area may generate a regional instruction of "evacuate in the upwind direction and use a respirator when necessary", while the office area may generate a regional instruction of "stay away from the production area and evacuate along the designated route", realizing differentiated safety guidance based on regional characteristics as shown in Table 5: Table 5: Division and characteristics of main functional areas of the factory

[0059] In order to achieve global optimization of the overall evacuation process, the company applied a collective intelligence optimization algorithm to establish a global optimization objective function and adjust the evacuation path planning scheme in real time. The system combines an improved particle swarm optimization algorithm, multi-agent simulation technology, and path planning algorithm, and can dynamically adjust the evacuation strategy according to real-time personnel distribution and risk changes.

[0060] In the application example, for the emergency evacuation scenario of a large chemical plant, the implementation process of the algorithm is as follows: The system receives real-time data on the location of all factory employees, risk monitoring data, and evacuation channel status data; Generate personalized safety guidance for each employee, such as evacuation direction and speed, through the personalized adaptation model of chemical risk semantics; At the same time, through multi-agent simulation technology, the overall effect that each employee may produce after following the guidance is predicted; When a certain evacuation channel is detected to be congested, the system automatically adjusts the evacuation route of relevant employees to avoid the formation of congestion points; When the risk status changes (such as the diffusion direction of toxic gases changes), the system updates the risk assessment in real time and adjusts the evacuation plan.

[0061] For example, in an evacuation drill, the system initially recommended Channel 1 for employees in Area A and Channel 2 for employees in Area B. When the system detected that the crowd density in Channel 1 was too high, it automatically assigned Channel 3 to subsequent employees in Area A and adjusted the travel speed appropriately, avoiding the formation of congestion points and reducing the overall evacuation time by nearly 30%.

[0062] In a large-scale evacuation drill involving 1,200 employees, the system monitored the density and risk distribution of people in each evacuation channel in real time and automatically adjusted the evacuation route allocation as shown in Table 6: Table 6: Example of dynamic route adjustment in evacuation drill

[0063] To verify the technical effect of this implementation, the company conducted a series of comparative tests and actual applications after deploying the system, focusing on two core technical effects: the execution efficiency of personalized semantic guidance and the overall evacuation time.

[0064] To verify the impact of this system on overall evacuation efficiency, the company conducted 12 standardized evacuation drills, using both the traditional system and this system, and recorded the time required to complete the evacuation. The test scenarios included evacuations of different scales and complexities, as shown in Table 7: Table 7: Comparison of overall evacuation time in different evacuation scenarios (unit: minutes)

[0065] In addition to the above two core technical effects, this system also improves the following aspects: The probability of accidentally entering a dangerous area is reduced from 15.8% in the traditional system to 0.9%, a reduction of 94.3%; The time from the start of emergency response to the start of the first evacuation of personnel was reduced from an average of 2.3 minutes to 0.8 minutes; The difference in evacuation time between different areas was reduced from a standard deviation of 5.4 minutes to 1.2 minutes, reflecting the balance of the system; The system has strong adaptability and can complete plan adjustments within an average of 12 seconds in tests where risk status changes suddenly.

[0066] In summary, this implementation has verified the practicality and advancement of the semantic-risk fusion personalized emergency evacuation management method for chemical plants in actual application in the enterprise, and provided an effective solution for the safety management of chemical enterprises.

[0067] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation mode. The above-mentioned specific implementation mode is merely illustrative and not restrictive. Under the guidance of this embodiment, ordinary technicians in this field can also make more forms of equivalent embodiments, all of which are within the protection of this embodiment.

Claims

1. An intelligent management method for emergency evacuation sites, characterized in that: The following steps are involved: The risk semantic mapping algorithm is used to construct a unified semantic representation space for chemical industry risk, generating a mapping relationship between risk status and semantic expression; Apply feature profiling algorithms to process individual employee data and generate employee feature representation models; Generate personalized semantic guidance for employees with different characteristics based on the personalized adaptation model of chemical risk semantics; Apply regional hierarchical clustering algorithm to process plant layout and personnel distribution data to generate a multi-level chemical risk semantic model; The real-time location and risk change data are processed through collective intelligent optimization algorithms to calculate dynamic evacuation path planning solutions.

2. The intelligent management method for emergency evacuation sites according to claim 1, characterized in that: The risk semantic mapping algorithm comprises the following steps: Collect and analyze data on various risk events that may occur in the chemical plant area and build a risk status set: ； in , , They represent the first, second, and kth risk states respectively, Indicates the number of risk status categories; Based on the characteristics of risk events, construct feature vectors: ； in Indicates risk status The characteristic vector of , , Represent the first, second, and qth features respectively, represents the dimension of the feature vector; Using spatial distribution modeling technology, calculate the risk spatial distribution function: ； in Indicates risk status In Location and time The intensity distribution of Indicates location, Indicates time; Create a set of semantic expressions based on expert knowledge and security specifications: ； in represents a set of semantic expressions, , , They represent the first, second, and mth security guidance semantics respectively, Indicates the number of types of semantic expressions; Use the mapping algorithm to create a function: ； in represents the mapping function, represents the risk state set, Represents a set of semantic expressions, which represents the risk status Mapping to corresponding semantic expressions .

3. The intelligent management method for emergency evacuation sites according to claim 1, characterized in that: The feature profiling algorithm comprises the following steps: Collect and organize individual employee data, including job responsibilities, training records, professional background, and physical condition, to build a set of employee characteristics; Through data analysis methods, feature vectors are constructed for each type of employee characteristics; Use deep neural networks to learn employee feature representations and map high-dimensional features to low-dimensional semantic understanding space; Through the analysis of historical emergency drill data, a behavior prediction function is established.

4. The intelligent management method for emergency evacuation sites according to claim 1, characterized in that: The chemical risk semantic personalized adaptation model adopts a multi-objective optimization method to calculate the optimal personalized semantic guidance for each employee and the current risk status through the following objective function: ； in represents the optimal personalized semantic guidance, Represented in the semantic expression set Finding the optimal semantic guidance , is the guiding utility function, which represents the risk state Downward, the characteristics are Provide semantic guidance to employees The validity value of Represents an element in a semantic expression set, Indicates the employee number.

5. The intelligent management method for emergency evacuation sites according to claim 4, characterized in that: The calculation formula of the guidance utility function is: ； in is the guiding utility function, , , , is the weight coefficient; is a safety function, which indicates the safety level of employees under the current risk status; is the understanding function, which represents the employee's understanding of the semantic expression; KL divergence is a measure of the degree of deviation between individual behavior and optimal behavior; is the attenuation function of risk and distance, reflecting the spatial diffusion characteristics of chemical risks. Indicates the risk status of the employee's location, Indicates the distance value from the employee to the risk source. Represents employee behavioral decisions.

6. The intelligent management method for emergency evacuation sites according to claim 1, characterized in that: The regional hierarchical clustering algorithm comprises the following steps: Analyze the layout data of chemical plant area and divide it into different functional areas; Collect and analyze risk characteristic data of each region and construct regional risk function; Integrate employee distribution data to construct the distribution of employee group characteristics in each region and their corresponding weight values; Create a region-specific semantic template library based on the characteristic data and safety regulations of each region; The swarm optimization algorithm is applied to calculate the optimal semantic expression of the region.

7. The intelligent management method for emergency evacuation sites according to claim 6, characterized in that: The optimal semantic expression of the region is calculated by the following formula: ； in, represents the optimal semantic expression of the region, Indicates area A semantic template library, Indicates area The distribution of employee group characteristics within The characteristic is of employees in the region The weight value in Indicates area In time risk status, represents the guiding utility function, Indicates the area The sum of all employee characteristic types in Indicates Regions, The region number.

8. The intelligent management method for emergency evacuation sites according to claim 1, characterized in that: The collective intelligence optimization algorithm comprises the following steps: Establish a global optimization objective function to incorporate individual safety and overall evacuation efficiency into the same framework; Use multi-agent modeling technology to build a collective behavior model of employees; Apply information entropy calculation method to construct collective behavior entropy function; Using dynamic programming algorithm, the optimal evacuation path for each employee is calculated.

9. The intelligent management method for emergency evacuation sites according to claim 8, characterized in that: The global optimization objective function is: ； in, represents the global optimization objective, is the total number of employees, Indicates the employee number. Indicates employees The guiding utility value of Indicates employees Features, Indicates employees The risk status of the location, Indicates employees Personalized semantic guidance, represents the collective behavior entropy function, is the balance parameter; represents the collective behavior model of employees, , , They represent the behavioral decisions of the 1st, 2nd, and Nth employees respectively.

10. An intelligent management system for emergency evacuation sites, characterized in that: A method for intelligently managing an emergency evacuation site according to any one of claims 1 to 9, comprising: The risk semantic mapping module is used to construct a unified risk semantic representation space and generate a mapping relationship between risk status and semantic expression; The feature profiling module is used to process individual personnel data and generate a personnel feature representation model; The personalized semantic adaptation module is used to calculate the optimal semantic expression based on the guidance utility function and generate personalized semantic guidance for people with different characteristics; The regional hierarchical clustering module processes the site layout and personnel distribution data to generate a multi-level risk semantic model for calculating the optimal semantic expression of the region; The collective intelligent optimization module is used to process real-time location and risk change data and calculate dynamic evacuation path planning solutions. The module is based on the global optimization objective function, integrates individual safety and overall evacuation efficiency, and uses a dynamic programming algorithm to calculate the optimal evacuation path for each person.

Images