A risk monitoring and early warning system for toxic and harmful gases at fire emergency rescue sites

Through the sensors and Reformer models carried by firefighters, combined with grid data processing, the problem of difficulty in obtaining information in fire accidents in petrochemical plants and hazardous chemical production sites is solved, real-time monitoring and early warning of toxic and harmful gas concentrations on site is achieved , improves the safety and accuracy of fire rescue.

CN118658270BActive Publication Date: 2024-11-22SHENYANG FIRE RES INST OF MEM
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411137009.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-19
Publication Date
2024-11-22
Estimated Expiration
2044-08-19

AI Technical Summary

Technical Problem

In fire accidents in petrochemical plant areas and hazardous chemical production and storage sites, it is difficult for fire rescue teams to obtain the concentration information of toxic and harmful gases on site in real time and other related factors, resulting in difficulties in deducing situations and rescue deployment at the disaster site.

Method used

Using the flexible toxic and harmful gas sensor and neural network Reformer model carried by firefighting individual soldiers, combined with the combat trajectory information of firefighting individual soldiers, the trend forecast and real-time early warning of toxic and harmful gas concentration in the field grid area is realized, and through the firefighting individual soldiers' physical domain network and combat The integration of the network and the command and dispatch network provides real-time life security guarantee and combat data support.

Benefits of technology

Real-time monitoring and early warning of toxic and harmful gas concentrations between firefighters and command centers, improve the safety and accuracy of the rescue site, reduce the risks caused by difficulties in obtaining information, and is suitable for complex petrochemical plants and hazardous chemicals. Product production site.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118658270B_ABST
    Figure CN118658270B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of emergency rescue of fire rescue teams, and discloses a toxic and harmful gas risk monitoring and early warning system for fire emergency rescue sites. The sensing unit obtains the real-time sensing data of the toxic and harmful gas concentration information of the toxic and harmful gas sensors carried by all participating fire soldiers and the location information of the communication terminal carried by the individual soldiers, and then transmits it to the fire fighting network, and then connects to the fire command and dispatch network. The prediction unit uses the neural network Reformer model to predict the trend of toxic and harmful gas concentration on site through the rescue site data information, and updates the individual soldier's early warning level for the predicted data according to the early warning level determination method of the area where the individual soldier is located. Based on the existing emergency communication system and equipment of the fire rescue team, the present invention realizes the real-time sensing, transmission, and processing of the toxic and harmful gas concentration around the body area of ​​the fire rescue soldier, realizes the prediction and early warning of the spread concentration of toxic and harmful gases, and provides data support for the scientific command of the on-site commander.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of emergency rescue of fire rescue teams, especially rescue scenes such as petrochemical plant areas, hazardous chemicals production and storage sites, etc., and specifically to a toxic and harmful gas risk monitoring and early warning system at a fire emergency rescue site. Background Art

[0002] Petrochemical plants, hazardous chemicals production and storage sites and other places involve a variety of chemical raw materials in daily production processes. Flammable, explosive and toxic gases are easily produced during the production process. Once a leak occurs, it can easily lead to a major production safety accident. Therefore, my country has extremely high requirements for such enterprises or places, especially for the production, storage, transportation and other links of such enterprises. Higher-level regulatory requirements are set.

[0003] Although the state has put forward high requirements for the production safety of petrochemical plants and hazardous chemical production and storage enterprises, due to the overly complex process and too many uncontrollable factors, many major fire accidents have occurred in recent years, resulting in irreparable losses. As the main force in handling petrochemical and hazardous chemical fire accidents, the fire rescue team scientifically handles such disasters to avoid casualties and reduce property losses. The premise of scientific handling is to obtain various effective information on the scene in real time, such as: types of hazardous chemicals, ignition points / leakage sources, and the development trend of accidents. This information is directly related to the fire rescue team's use of fire fighting and rescue methods, how to deploy the next rescue, etc. Therefore, how to obtain various effective information on the scene through technical means when disasters occur and effectively use various information for risk assessment is an important problem that needs to be solved urgently.

[0004] The change of toxic and harmful gas concentration over time at the fire emergency rescue site is a complex process with the action of multiple factors. It is related to the location of the leakage source, the amount of leakage source, wind direction, wind force, temperature, humidity, air pressure, etc., but these related factors are difficult to obtain in real time when a disaster occurs. Traditional fire simulation mainly relies on FiresDynamics Simulator (FDS) software. As a basic tool software for fire dynamics and combustion, this software is mainly used for smoke control design, detector start-up time simulation research and fire reconstruction. The basic principle of the software is mainly based on fluid mechanics models, turbulence models, combustion models, radiation models, etc.; FDS software is currently mainly used to review the fire scene, reconstruct the fire development process and analyze the cause of the fire. In the fire simulation process, a large number of data sources need to be used as input, such as: scene conditions, burning materials, the amount of materials, weather conditions when the fire occurs, etc., but it is difficult to quickly obtain a large amount of information at the disaster site. Therefore, it is difficult to use FDS for situation deduction at the disaster accident site. Summary of the invention

[0005] In view of the shortcomings of the existing methods, the present invention provides a toxic and harmful gas risk monitoring and early warning system for fire emergency rescue sites. Relying on the flexible toxic and harmful gas sensors carried by individual firefighters, real-time perception data of toxic and harmful gas concentration information at the site is realized. Combined with the combat trajectory information of individual firefighters, a neural network Reformer model is applied to realize the trend prediction of toxic and harmful gas concentration in the grid area at the site, providing real-time life safety protection for the operations of individual firefighters at the site, and also providing basic data support for combat command.

[0006] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0007] A toxic and harmful gas risk monitoring and early warning system for fire emergency rescue sites, wherein the system operating environment includes a fire individual body area network, a fire fighting network and a fire command and dispatch network, and the system includes a sensing unit, a transmission unit and a prediction unit, wherein the sensing unit operates in the fire individual body area network, the transmission unit operates in the fire fighting network, and the prediction unit operates in the fire command and dispatch network of the on-site command center;

[0008] The sensing unit acquires the real-time sensing data of the toxic and harmful gas concentration information of the toxic and harmful gas sensors carried by all participating firefighters and the location information of the communication terminal carried by the individual soldiers and transmits it to the firefighting network;

[0009] The transmission unit of the fire fighting network receives the data transmitted by the perception unit and connects to the fire command and dispatch network through Mesh. The prediction unit uses the neural network Reformer model to predict the concentration trend of toxic and harmful gases in the grid area of ​​the scene based on the data sent by the transmission unit, the real-time perception data of the toxic and harmful gas concentration information at the rescue site, and the firefighter's combat trajectory information. The prediction unit updates the individual soldier's warning level based on the predicted data according to the warning level determination method of the area where the individual soldier is located.

[0010] The prediction unit predicts the concentration trend of toxic and harmful gases at the rescue site, including the following steps:

[0011] S1. Collect historical data on the diffusion of toxic and hazardous gases and fire laboratory test data to establish a training data set;

[0012] S2. For the training data set, the neural network Reformer model is trained using an optimization algorithm to obtain an optimized Reformer model;

[0013] Step S2 specifically includes the following steps:

[0014] S2.1. Preprocess the toxic and harmful gas data of the training data set, including the following steps:

[0015] S2.1.1. Normalize the data to ensure that the data is on a uniform scale to facilitate model training;

[0016] S2.1.2. Divide the normalized data into training set and test set according to the time series to ensure that the model can learn the time characteristics of the data;

[0017] The input matrix after data normalization is recorded as

[0018] ;

[0019] in, Including toxic and hazardous gas concentration data;

[0020] S2.2, using the optimization algorithm to train the neural network Reformer model, specifically including the following steps:

[0021] S2.2.1. Model initialization: Set multiple hyperparameters to build the Reformer model to adapt to different task requirements. The hyperparameters include: number of layers, number of attention heads, number of hidden units, sequence length, and embedding dimension.

[0022] S2.2.2.2. Definition of loss function: The loss function is defined as the mean square error (MSE) between the predicted value and the true value to measure the prediction accuracy of the model;

[0023] ;

[0024] in, is the number of samples, For the i The predicted value of samples, y i is the true sample value;

[0025] S2.2.3, model training: first, input the training set obtained in step S2.1.2 into the model in batches, calculate the predicted value through forward propagation, and use the loss function to calculate the error; then calculate the gradient through the back propagation algorithm, and use the optimization algorithm to update the model parameters; through multiple iterations, continuously adjust the model parameters to minimize the loss function value, and use the test set to evaluate the model performance; the obtained optimized Reformer model has the time series characteristics of the concentration of toxic and harmful gases, which improves the accuracy and reliability of the prediction;

[0026] S3, establish a rescue site coordinate system, establish a toxic and harmful gas risk monitoring data matrix for the emergency rescue site and establish a corresponding relationship with the site coordinate system, initialize the types and concentrations of toxic and harmful gases in the matrix units, and obtain a real-time data set on site;

[0027] S4. Collect the real-time sensing data of the toxic and harmful gas concentration information of the toxic and harmful gas sensor carried by the firefighters every second and the location information of the communication terminal carried by the firefighters and record them into the input parameter matrix to form a real-time data set based on time series;

[0028] S5. Input the real-time data set based on time series into the optimized Reformer model to obtain the toxic and harmful gas concentration values ​​in each area and the toxic and harmful gas concentration trends at the rescue site predicted based on the time series.

[0029] Step S3 specifically includes the following steps:

[0030] S3.1、Define the fire emergency rescue scene as , according to 1 m ×1 m The rescue site is gridded and the center of the rescue site is selected as the origin using the GIS system of the fire brigade. The representative coordinates are ( i, j ) has a site area of ​​1 m ×1 m area;

[0031] S3.2. Define the toxic and hazardous gas risk monitoring data matrix at the fire emergency rescue site as;

[0032] ;

[0033] ;

[0034] Establish fire emergency rescue scene The concentration attribute of toxic and harmful gases in the area is vector,

[0035] ;

[0036] The internal elements are Eight typical fire-scene harmful gases and The concentration of two combustible gases, in ppm. If the value is 0, it means that the gas does not exist in the area.

[0037] Step S5 is specifically as follows: input the real-time data set based on the time series into the optimized Reformer model to obtain the concentration values ​​of toxic and harmful gases in each region predicted based on the time series, and then perform denormalization on the predicted values ​​to restore them to the original concentration scale; the prediction matrix output by the model is expressed as follows:

[0038] ;

[0039] in Indicates that based on the current moment t predict t+ The area after T time ( i, j ) of toxic and harmful gas concentrations.

[0040] The specific method for determining the warning level according to the area where the individual soldier is located is as follows: the emergency rescue site area of ​​the fire brigade is divided into 4 levels: ERPG-1 corresponds to that the individual fire soldier can conduct emergency rescue in this area without wearing protective equipment; ERPG-2 corresponds to that the individual fire soldier needs to wear protective equipment and conduct emergency rescue in this area; ERPG-3 corresponds to that the individual fire soldier needs to wear protective equipment, and the length of emergency rescue in this area should not be more than 1 hour; IDLH corresponds to that the individual fire soldier needs to immediately exit the area even if wearing protective equipment.

[0041] The specific update of the individual soldier warning level for the predicted data is as follows:

[0042] Establish a firefighting individual early warning model for emergency rescue sites of firefighting teams, which is defined as:

[0043] ;

[0044] Where: Representing the first i firefighters, where i is less than the total number of firefighters on site, n. Representative i The warning level of each firefighter’s location, level j Representative j The warning level is j Less than 4;

[0045] level 0 The concentrations of the 10 toxic and hazardous gases that the individual soldier should carry are within the concentration range specified in ERPG-1;

[0046] level 1 The concentrations of the 10 toxic and hazardous gases carried by the individual soldier are within the concentration range specified in ERPG-2;

[0047] level 2 The concentrations of the 10 toxic and hazardous gases carried by the individual soldier are within the concentration range specified in ERPG-3;

[0048] level 3 The concentrations of the 10 toxic and hazardous gases carried by the individual soldier reach the concentration specified by IDLH or are greater than the concentration specified by ERPG-3.

[0049] At the current warning level of the individual soldier level j The timer will start counting down. When the timer exceeds 1 hour, the existing warning level will be increased by 1 level. When the warning level of the area where the individual soldier is located changes, the timer will restart.

[0050] The present invention has the following beneficial effects and advantages:

[0051] 1) The present invention provides a risk monitoring and early warning system for toxic and harmful gases at fire emergency rescue sites. Based on the existing emergency communication system and equipment of the fire rescue team, it realizes the real-time perception, transmission and processing of the concentration of toxic and harmful gases around the body of a single firefighter. By gridding the rescue site and matching it with the positioning data, the Reformer model is used to predict and early warning the spread concentration of toxic and harmful gases in the grid area, providing data support for the scientific command of the on-site commander, and effectively ensuring the life safety of the firefighters participating in the battle on the scene;

[0052] 2) This invention is different from the traditional toxic and harmful gas concentration spread model based on fluid mechanics FDS. It abandons the adverse impact of the inability to obtain the type and quantity of leakage sources and the on-site climate and environmental data in real time at the disaster site on prediction and early warning. It establishes a method for predicting trends based on time series data based on a machine learning model, which has higher practicality in the actual emergency rescue process;

[0053] 3) The patent of this invention is different from the traditional method of information perception, transmission and background processing based on the public wireless mobile communication network. It is based on the integration of three networks: the firefighting individual body area network, the firefighting combat network and the firefighting command and dispatch network. It has higher applicability in the application of firefighting emergency rescue teams, especially in petrochemical plant areas, hazardous chemicals production and storage sites and other firefighting and rescue sites.

[0054] 4) The Reformer model used in the patent of this invention is an efficient attention mechanism model, which is suitable for predicting and processing long sequence data. In this system, the toxic and harmful gas concentration data of the on-site grid is processed with strong pertinence. Compared with other machine learning models, it can achieve better prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 This is a network schematic diagram of the operating environment of the system of the present invention;

[0056] Figure 2 This is a flow chart of the risk prediction system for toxic and harmful gases at fire emergency rescue sites of the present invention. DETAILED DESCRIPTION

[0057] The present invention is described in detail below with reference to the accompanying drawings and specific examples.

[0058] The present invention provides a risk monitoring and early warning system for toxic and harmful gases at fire emergency rescue sites. Figure 1 As shown, the system operating environment includes a firefighting individual body area network, a firefighting combat network and a firefighting command and dispatch network. The system includes a sensing unit, a transmission unit and a prediction unit.

[0059] The sensing unit runs in the firefighter body area network and is composed of a communication terminal held / worn by a firefighter and a gas sensor attached to the firefighter's protective clothing. The gas sensors include 8 typical fire-scene harmful gases and The sensor collects the type and concentration of gas and transmits it to the individual communication terminal wirelessly. The individual communication terminal supports BeiDou satellite positioning. The individual communication terminal loads the location information and transmits it wirelessly to the fire fighting network. The individual body area network is a local area network within 10m around the individual soldier. Bluetooth and Wi-Fi radio frequency can be used for data transmission between the gas sensor and the communication terminal. The fire fighting network is a local area network covering the entire battle scene. At present, the fire fighting team mostly adopts Mesh strategy wireless networking. The communication frequency is mostly 600Mhz and 1.4Ghz. The communication terminal can access the fire fighting network by direct access / back-to-back access (that is, the individual soldier carries another fire fighting network networking device, and the communication terminal accesses the networking device via Wi-Fi).

[0060] The transmission unit runs on the fire fighting network, which is mainly composed of multiple individual firefighting communication terminals and multiple fire fighting network networking devices. The initiator is the individual firefighting communication terminal, and the receiving end is the firefighting network networking device adjacent to the individual firefighting soldier. The firefighting network networking is a local area network based on the Mesh strategy. The firefighting network networking device can wirelessly access the fire command and dispatch network, and the access method is Mesh direct access.

[0061] The prediction unit runs on the hardware computing and storage device in the on-site command center, and the operating network is the fire command and dispatch network. The fire command and dispatch network can be connected to the brigade-level fire communication command center through a wireless public network / private network to achieve data sharing and interconnection.

[0062] like Figure 2 As shown in the figure, through the real-time perception data of toxic and harmful gas concentration information at the rescue site and combined with the combat trajectory information of individual firefighters, the neural network Reformer model is used to realize the trend prediction of toxic and harmful gas concentration in the grid area of ​​the site and update the individual soldier warning level according to the warning level determination method of the area where the individual soldier is located.

[0063] Specifically, the prediction unit predicts the concentration trend of toxic and harmful gases at the rescue site, including the following steps:

[0064] S1. Collect historical data on the diffusion of toxic and hazardous gases and fire laboratory test data to establish a training data set;

[0065] S2. For the training data set, an optimization algorithm is used to train the neural network Reformer model to obtain an optimized Reformer model; the Reformer model is an efficient attention mechanism model suitable for processing long sequence data. This model is used to predict the changes in toxic and harmful gas concentrations in various areas of the fire emergency rescue site. Its workflow includes data preprocessing, model training, prediction output and other steps.

[0066] The specific steps include:

[0067] S2.1. Preprocess the toxic and harmful gas data of the training data set, including the following steps:

[0068] S2.1.1. Normalize the data to ensure that the data is on a uniform scale to facilitate model training;

[0069] S2.1.2. Divide the normalized data into training set and test set according to the time series to ensure that the model can learn the time characteristics of the data;

[0070] The input matrix after data normalization is recorded as

[0071] ;

[0072] in, Including toxic and hazardous gas concentration data;

[0073] S2.2, using the optimization algorithm to train the neural network Reformer model, specifically including the following steps:

[0074] S2.2.1. Model initialization: Set multiple hyperparameters to build the Reformer model to adapt to different task requirements. The hyperparameters include: number of layers, number of attention heads, number of hidden units, sequence length and embedding dimension; the number of layers is usually between 6 and 12 layers, each layer contains a self-attention module and a feedforward neural network module; the number of attention heads is usually between 8 and 16, each head independently captures different features in the input sequence and merges them; the number of hidden units is usually between 512 and 2048, which are used to process complex feature representations; the sequence length determines the number of time steps that the model can process, and Reformer processes long sequence data through local sensitive hashing (LSH) technology to reduce computational complexity; the embedding dimension is usually the same as the number of hidden units, which determines the size of each input feature vector; through the setting of these hyperparameters, the model can adapt to different task requirements and achieve efficient time series data processing.

[0075] S2.2.2.2. Definition of loss function: The loss function is defined as the mean square error (MSE) between the predicted value and the true value to measure the prediction accuracy of the model;

[0076] ;

[0077] in, is the number of samples, For the i The predicted value of samples, y i is the true sample value;

[0078] S2.2.3, model training: first, input the training set obtained in step S2.1.2 into the model in batches, calculate the predicted value through forward propagation, and use the loss function to calculate the error; then calculate the gradient through the backpropagation algorithm, and use the optimization algorithm (such as Adam) to update the model parameters; through multiple iterations, continuously adjust the model parameters to minimize the loss function value, and use the test set to evaluate the model performance, and adjust the hyperparameters when necessary; the optimized Reformer model obtained has the time series characteristics of toxic and harmful gas concentrations, which improves the accuracy and reliability of the prediction.

[0079] S3, establish a rescue site coordinate system, establish a toxic and harmful gas risk monitoring data matrix for the emergency rescue site and establish a corresponding relationship with the site coordinate system, initialize the types and concentrations of toxic and harmful gases in the matrix units, and obtain a real-time data set on site; specifically includes the following steps:

[0080] S3.1. Define the fire emergency rescue scene as , according to 1 m ×1 m The rescue site is gridded and the center of the rescue site is selected as the origin using the GIS system of the fire brigade. The representative coordinates are ( i, j ) has a site area of ​​1 m ×1 m area;

[0081] S3.2. Define the toxic and hazardous gas risk monitoring data matrix at the fire emergency rescue site as;

[0082] ;

[0083] ;

[0084] Establish fire emergency rescue scene The concentration attribute of toxic and harmful gases in the area is vector,

[0085] ;

[0086] The internal elements are Eight typical fire-scene harmful gases and The concentration of two combustible gases, in ppm. If the value is 0, it means that the gas does not exist in the area.

[0087] The change of toxic and harmful gas concentration over time at the fire emergency rescue site is a complex process of multiple factors, which is related to the location of the leakage source, the inventory of the leakage source, wind direction, wind force, temperature, humidity, air pressure, etc., but these related factors are difficult to obtain in real time when a disaster occurs, and the FDS software based on fluid mechanics is suitable for laboratory modeling, and its accuracy for on-site simulation is questionable. Based on these analyses, in the present invention, environmental factors such as wind direction, wind force, temperature, humidity, and air pressure are not included in the input of the prediction model for gas diffusion. The present invention is based on the dynamic changes of time series data, and extracts trend features through a machine learning model, which is then used to predict the concentration changes of toxic and harmful gases in the region.

[0088] S4. Real-time sensing data of toxic and harmful gas concentration information from toxic and harmful gas sensors carried by individual firefighters and location information of communication terminals carried by individual firefighters are collected every second and recorded in an input parameter matrix to form a real-time data set based on time series.

[0089] S5. Input the real-time data set based on time series into the optimized Reformer model to obtain the concentration values ​​of toxic and harmful gases in each region predicted based on the time series, and then perform denormalization on the predicted values ​​to restore them to the original concentration scale; the prediction matrix output by the model is expressed as follows:

[0090] ;

[0091] in Indicates that based on the current moment t predict t+ The area after T time ( i, j ) of toxic and harmful gas concentrations.

[0092] The national standard GB / T 50493-2019 "Petrochemical Combustible Gas and Toxic Gas Detection Alarm Design Standard" is mainly used to ensure the personal safety and production safety of petrochemical enterprises, monitor the leakage of flammable or toxic gases in the production process and storage and transportation facilities, and promptly alarm to prevent personal injury and fire and explosion accidents.

[0093] Since the application scenario of the present invention is the emergency rescue scene of the fire rescue team, the scenes involved are mainly petrochemical plant areas, hazardous chemicals production and storage places, etc., by real-time monitoring of the types and concentrations of toxic and harmful gases on site, toxic and harmful gas monitoring information guarantee is provided for the fire rescue process. Therefore, the first and second level alarm functions and parameters in GB / T 50493-2019 are not applicable to the patent of this invention.

[0094] The Emergency Response Planning Guidelines (ERPG) for air pollution published by the American Industrial Hygiene Association (AIHA) provides a detailed principle for regional hazard classification based on the types and concentrations of toxic and hazardous gases on site:

[0095] ERPG-1: The maximum concentration of a chemical in the air that a person can be exposed to for one hour without experiencing any symptoms;

[0096] ERPG-2: The concentration of a chemical in the air to which people can be exposed for one hour without experiencing irreversible or severe health effects that render them incapable of taking protective measures;

[0097] ERPG-3: The concentration of chemicals in the air to which a person can be exposed for one hour without causing life-threatening effects.

[0098] In addition to the above classifications, the immediately lethal concentration (IDLH) is also defined as the maximum permissible concentration at which a person can be exposed to a toxic gas environment for 30 minutes and still be able to escape without suffering adverse symptoms or irreversible health effects.

[0099] Therefore, in the present invention patent, the fire brigade emergency rescue site area is divided into 4 levels:

[0100] ERPG-1 corresponds to that a single firefighter can perform emergency rescue in the area without wearing protective equipment; ERPG-2 corresponds to that a single firefighter needs to wear protective equipment to perform emergency rescue in the area;

[0101] ERPG-3: Firefighters must wear protective equipment and emergency rescue in this area should not exceed 1 hour.

[0102] IDLH requires individual firefighters to immediately exit the area even if they are wearing protective equipment.

[0103] Furthermore, the individual soldier warning level is updated for the predicted data as follows:

[0104] Establish a firefighting individual early warning model for emergency rescue sites of firefighting teams, which is defined as:

[0105] ;

[0106] Where: Representing the first i firefighters, where i is less than the total number of firefighters on site, n. Representative i The warning level of each firefighter’s location, level j Representative j The warning level is j Less than 4;

[0107] level 0 The concentrations of the 10 toxic and hazardous gases that the individual soldier should carry are within the concentration range specified in ERPG-1;

[0108] level 1 The concentrations of the 10 toxic and hazardous gases carried by the individual soldier are within the concentration range specified in ERPG-2;

[0109] level 2 The concentrations of the 10 toxic and hazardous gases carried by the individual soldier are within the concentration range specified in ERPG-3;

[0110] level 3 The concentrations of the 10 toxic and hazardous gases carried by the individual soldier reach the concentration specified by IDLH or are greater than the concentration specified by ERPG-3.

[0111] Furthermore, at the current warning level of the individual soldier level jWhen the timer exceeds 1 hour, the existing warning level is increased by 1, that is, when the area where a single soldier is located is level 0 When the timer reaches 1 hour, the current warning level should be raised level 0 → level 1 ;

[0112] When the warning level of the area where the individual soldier is located changes, the timer starts again.

[0113] The present invention relies on the flexible toxic and harmful gas sensors carried by individual firefighters to realize the real-time perception data of toxic and harmful gas concentration information on site, combines the combat trajectory information of individual firefighters, and applies the neural network Reformer model to realize the prediction of toxic and harmful gas concentration trend in the on-site grid area, providing real-time life safety protection for the combat of individual firefighters on site, and also providing basic data support for combat command. Different from the traditional method of information perception, transmission and background processing based on the public wireless mobile communication network, the method based on the integration of the three networks of the firefighter body area network, the firefighting combat network and the firefighting command and dispatch network has higher applicability in the application of fire emergency rescue teams, especially in petrochemical plant areas, hazardous chemical production and storage sites and other firefighting and rescue sites.

Claims

1. A toxic and harmful gas risk monitoring and early warning system for fire emergency rescue sites, characterized in that: The system operation environment includes a firefighting individual body area network, a firefighting battle network and a firefighting command and dispatch network. The system includes a sensing unit, a transmission unit and a prediction unit. The sensing unit runs in the firefighting individual body area network, the transmission unit runs in the firefighting battle network, and the prediction unit runs in the firefighting command and dispatch network of the on-site command center. The sensing unit acquires the real-time sensing data of the toxic and harmful gas concentration information of the toxic and harmful gas sensors carried by all participating firefighters and the location information of the communication terminal carried by the individual soldiers and transmits it to the firefighting network; After receiving the data transmitted by the sensing unit, the transmission unit of the fire fighting network is connected to the fire command and dispatch network through a Mesh method. Based on the data sent by the transmission unit, the prediction unit uses the real-time sensing data of the toxic and harmful gas concentration information at the rescue site and the firefighting individual combat trajectory information to apply the neural network Reformer model to predict the concentration trend of toxic and harmful gases in the on-site grid area and update the individual soldier warning level of the predicted data according to the warning level determination method of the area where the individual soldier is located; The prediction unit predicts the concentration trend of toxic and harmful gases at the rescue site, including the following steps: S1. Collect historical data on the diffusion of toxic and hazardous gases and fire laboratory test data to establish a training data set; S2. For the training data set, the neural network Reformer model is trained using an optimization algorithm to obtain an optimized Reformer model; The specific steps include: S2.

1. Preprocess the toxic and harmful gas data of the training data set, including the following steps: S2.1.

1. Normalize the data to ensure that the data is on a uniform scale to facilitate model training; S2.1.

2. Divide the normalized data into training set and test set according to the time series to ensure that the model can learn the time characteristics of the data; The input matrix after data normalization is recorded as ; in, Including toxic and hazardous gas concentration data; S2.2, using the optimization algorithm to train the neural network Reformer model, specifically including the following steps: S2.2.

1. Model initialization: Set multiple hyperparameters to build the Reformer model to adapt to different task requirements. The hyperparameters include: number of layers, number of attention heads, number of hidden units, sequence length, and embedding dimension. S2.2.2.

2. Definition of loss function: The loss function is defined as the mean square error (MSE) between the predicted value and the true value to measure the prediction accuracy of the model; ; in, is the number of samples, For the i The predicted value of samples, y i is the true sample value; S2.2.3, model training: first, input the training set obtained in step S2.1.2 into the model in batches, calculate the predicted value through forward propagation, and use the loss function to calculate the error; then calculate the gradient through the back propagation algorithm, and use the optimization algorithm to update the model parameters; through multiple iterations, continuously adjust the model parameters to minimize the loss function value, and use the test set to evaluate the model performance; the obtained optimized Reformer model has the time series characteristics of the concentration of toxic and harmful gases, which improves the accuracy and reliability of the prediction; S3, establish a rescue site coordinate system, establish a toxic and harmful gas risk monitoring data matrix for the emergency rescue site and establish a corresponding relationship with the site coordinate system, initialize the types and concentrations of toxic and harmful gases in the matrix units, and obtain a real-time data set on site; S4. Collect the real-time sensing data of the toxic and harmful gas concentration information of the toxic and harmful gas sensor carried by the firefighters every second and the location information of the communication terminal carried by the firefighters and record them into the input parameter matrix to form a real-time data set based on time series; S5. Input the real-time data set based on time series into the optimized Reformer model to obtain the toxic and harmful gas concentration values ​​in each area and the toxic and harmful gas concentration trends at the rescue site predicted based on the time series.

2. A fire emergency rescue site toxic and harmful gas risk monitoring and early warning system according to claim 1, characterized in that Step S3 specifically includes the following steps: S3.1、Define the fire emergency rescue scene as , according to 1 m ×1 m The rescue site is gridded and the center of the rescue site is selected as the origin using the GIS system of the fire brigade. The representative coordinates are ( i, j ) has a site area of ​​1 m ×1 m area; S3.

2. Define the toxic and hazardous gas risk monitoring data matrix at the fire emergency rescue site as; ; ; Establish fire emergency rescue scene The concentration attribute of toxic and harmful gases in the area is vector, ; The internal elements are Eight typical fire-scene harmful gases and The concentration of two combustible gases, in ppm. If the value is 0, it means that the gas does not exist in the area.

3. A fire emergency rescue site toxic and harmful gas risk monitoring and early warning system according to claim 1, characterized in that Step S5 is specifically as follows: input the real-time data set based on the time series into the optimized Reformer model to obtain the concentration values ​​of toxic and harmful gases in each region predicted based on the time series, and then perform denormalization on the predicted values ​​to restore them to the original concentration scale; the prediction matrix output by the model is expressed as follows: ; in Indicates that based on the current moment t predict t+ The area after T time ( i, j ) of toxic and harmful gas concentrations.

4. A fire emergency rescue site toxic and harmful gas risk monitoring and early warning system according to claim 1, characterized in that The specific method for determining the warning level according to the area where the individual soldier is located is as follows: the emergency rescue site area of ​​the fire brigade is divided into 4 levels: ERPG-1 corresponds to that the individual firefighting soldier can conduct emergency rescue in this area without wearing protective equipment; ERPG-2 corresponds to that the individual firefighting soldier needs to wear protective equipment and conduct emergency rescue in this area; ERPG-3 corresponds to that the individual firefighting soldier needs to wear protective equipment, and the length of emergency rescue in this area should not be more than 1 hour; IDLH corresponds to that even if the individual firefighting soldier wears protective equipment, he or she must immediately exit the area.

5. A fire emergency rescue site toxic and harmful gas risk monitoring and early warning system according to claim 4, characterized in that The specific update of the individual soldier warning level for the predicted data is as follows: Establish a firefighting individual early warning model for emergency rescue sites of firefighting teams, which is defined as: ; Where: Representing the first i firefighters, where i is less than the total number of firefighters on site, n. Representative i The warning level of each firefighter’s location, level j Representative j The warning level is j Less than 4; level 0 The concentrations of the 10 toxic and hazardous gases that the individual soldier should carry are within the concentration range specified in ERPG-1; level 1 The concentrations of the 10 toxic and hazardous gases carried by the individual soldier are within the concentration range specified in ERPG-2; level 2 The concentrations of the 10 toxic and hazardous gases carried by the individual soldier are within the concentration range specified in ERPG-3; level 3 The concentrations of the 10 toxic and hazardous gases carried by the individual soldier reach the concentration specified by IDLH or are greater than the concentration specified by ERPG-3.

6. A fire emergency rescue site toxic and harmful gas risk monitoring and early warning system according to claim 5, characterized in that: At the current warning level of the individual soldier level j The timer will start counting down. When the timer exceeds 1 hour, the existing warning level will be increased by 1 level. When the warning level of the area where the individual soldier is located changes, the timer will restart.