Artificial intelligence-based air flow regulation type backdraft intelligent prevention system and method

Abstract

The application discloses a backdraft intelligent prevention system and method based on airflow regulation, and relates to the technical field of fire prevention.Compared with the previous fire prevention system, the application solves the problem that the traditional backdraft prevention method cannot adaptively adjust the strategy to cope with different fire development stages and complex environmental changes, and is difficult to meet the needs of modern fire safety protection; by fusing multi-sensor data to comprehensively consider the environmental factors in the region, combining the backdraft prediction model of the knowledge graph and the optimized extreme learning machine to analyze and predict backdraft, and intelligently regulating the airflow based on the backdraft risk, the system can automatically adjust the ventilation equipment parameters to reduce the possibility of backdraft occurrence.The system has adaptability and reliability, can self-learn and adapt to different scenes, and improves energy utilization efficiency.Meanwhile, the system is convenient to integrate and manage, integrated with other systems, provides a convenient interactive interface, facilitates remote monitoring and management, and improves the overall safety and efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fire prevention, and particularly relates to an air flow regulation type backdraft intelligent prevention system and method based on artificial intelligence. BACKGROUND

[0002] Backdraft is a dangerous phenomenon that may occur during a fire. When a building or enclosed space catches fire, combustible materials may undergo incomplete combustion under insufficient oxygen, producing a large amount of combustible gas and unburned smoke, which accumulates in the space. If fresh air suddenly enters at this time, backdraft may occur. When backdraft occurs, the fire will suddenly and violently erupt, the temperature will rise rapidly, and the pressure will increase rapidly, often accompanied by strong explosion and flame injection, which poses a serious threat to the building structure and the safety of the people inside. For example, in some commercial building fires, backdraft led to serious damage to the building structure and casualties.

[0003] The occurrence of backdraft is influenced by multiple factors, including indoor temperature, oxygen concentration, combustible gas concentration, ventilation conditions, etc. These factors are interrelated and dynamically changing, making it very complex to accurately predict and prevent backdraft. For example, the increase in temperature will accelerate the pyrolysis of combustible materials, increasing the production of combustible gas, while the ventilation conditions will affect the supply of oxygen and the accumulation of combustible gas, which together determine whether backdraft will occur and the intensity of the occurrence.

[0004] Traditional air flow regulation methods mainly control the flow of air through simple ventilation equipment operations, such as manually adjusting the opening of the ventilation valve or turning on the fixed power fan, etc. These methods lack comprehensive monitoring and intelligent analysis of backdraft-related factors, and cannot accurately regulate air flow according to real-time fire conditions and backdraft risks. For example, if the ventilation equipment does not reasonably regulate the air flow according to the actual situation in the early stages of the fire, it may lead to excessive oxygen entering, accelerating the spread of the fire and increasing the risk of backdraft.

[0005] For the prevention of backdraft, traditional methods often rely on fixed empirical thresholds and simple alarm devices. For example, an alarm is triggered when the temperature or smoke concentration reaches a certain set value, but this method cannot accurately predict the likelihood of backdraft and cannot take timely and effective preventive measures. Moreover, traditional prevention methods cannot adaptively adjust strategies to cope with different stages of fire development and complex environmental changes, making it difficult to meet the needs of modern fire safety protection.

[0006] To solve the above problems, the present application proposes an air flow regulation type backdraft intelligent prevention system and method based on artificial intelligence. SUMMARY

[0007] The present application aims to propose an air flow regulation type backdraft intelligent prevention system and method based on artificial intelligence to solve the problems proposed in the background art:

[0008] Traditional backdraft prevention methods cannot adaptively adjust strategies to cope with different fire development stages and complex environmental changes, and are difficult to meet the needs of modern fire safety protection.

[0009] In order to achieve the above-mentioned purpose, the present application adopts the following technical solutions:

[0010] The air flow regulation type backdraft intelligent prevention system based on artificial intelligence comprises:

[0011] Sensor layer: for collecting environmental data of the fire area through a plurality of sensors;

[0012] Processing and analysis layer: for receiving sensor layer data and conducting backdraft analysis and prediction based on knowledge graph and optimized extreme learning machine;

[0013] Decision control layer: making corresponding decisions according to the analysis and prediction results of the processing and analysis layer, and issuing corresponding air flow regulation instructions to the hardware execution layer;

[0014] Hardware execution layer: receiving the control instructions of the decision control layer, and controlling the corresponding ventilation equipment execution instructions, and also feeding back the air flow regulation results in real time through the sensor layer.

[0015] Preferably, the sensors of the sensor layer include temperature sensors, oxygen concentration sensors, smoke sensors, gas flow rate sensors, infrared imaging sensors and chemical sensors.

[0016] Preferably, the sensor layer periodically automatically calibrates all types of sensors through self-calibration technology, simultaneously performs self-diagnosis, and sends an alarm to the system according to the diagnosed abnormal results.

[0017] Preferably, the ventilation equipment of the hardware execution layer comprises intelligent ventilation valves, air circulation fans, smoke exhaust fans and adjustable ventilation ducts; the hardware execution layer uses intelligent liquid crystal material to prepare the shell and part of the structural members of the ventilation equipment, changes the transparency or color according to the control signal of the system, and is used to directly display the running state or backdraft risk level of the equipment.

[0018] Preferably, the fan blades of the intelligent ventilation valve and the air circulation fan are made of shape memory alloy material, which automatically changes shape according to the change of temperature and air flow pressure, so as to adjust the ventilation volume and wind direction of the ventilation equipment.

[0019] Preferably, a heat exchanger is also installed at the smoke exhaust fan, which recovers part of the energy in the high-temperature smoke and exhaust gas when it is discharged, for preheating fresh air entering the room; a nanofiber filter with self-cleaning function is used at the air inlet and smoke outlet of the smoke exhaust fan, an ultrasonic transducer is installed on the frame of the nanofiber filter for converting electrical energy into ultrasonic vibration energy to transfer to the filter material for self-cleaning; the surface of the nanofiber filter is also coated with a super-hydrophobic coating.

[0020] Preferably, the ventilation duct is internally provided with a ceramic matrix composite material with flame-retardant and adsorbing functions, and with a microporous structure.

[0021] The intelligent backdraft prevention method based on airflow regulation includes the following steps:

[0022] S1: Collecting environmental data of the fire area through a plurality of sensors;

[0023] S2: Based on the received sensor data, conducting backdraft analysis and prediction based on a knowledge graph and an optimized extreme learning machine;

[0024] S3: Making corresponding decisions according to the analysis and prediction results, and issuing corresponding airflow regulation instructions to each ventilation device;

[0025] S4: Each ventilation device receives and executes the corresponding instructions, and feedbacks the airflow regulation results in real time through the sensors.

[0026] Preferably, in S2, the environmental data collected by the sensors in S1 is collected, a backdraft knowledge graph is constructed through association and integration, and intelligent fault diagnosis and maintenance are conducted based on the knowledge graph; the knowledge graph is built based on an HMM model, entity recognition and relationship extraction are taken as sequence labeling problems, the HMM model defines hidden states and observable states, uses a probability model to predict hidden state sequences, thereby constructing the backdraft knowledge graph, calculating backdraft prediction indexes, and conducting feature extraction and backdraft prediction based on an extreme learning machine; the extreme learning machine is as follows:

[0027] When the training set of a single-hidden layer neural network model has m attack samples, the output function of ELM is as follows:

[0028]

[0029] Where, u i is the input vector; f(·) is the output vector; m is the total number of hidden layer neurons of ELM; w j is the weight vector connecting j hidden layer nodes and output layer nodes; σ(·) is the activation function of the hidden layer; ω jis the weight vector connecting the jth input layer node and the hidden layer node; δ j is the threshold value of the jth hidden layer neuron; σ(ω j ·u i +g j is the output of the jth hidden layer neuron with respect to the input sample u i ;

[0030] a preset output threshold value, based on the value of the output vector to classify the risk of backfiring;

[0031] The ELM hidden layer parameters are also optimized based on an improved butterfly optimization algorithm, which is specifically as follows:

[0032] The butterfly fragrance d is calculated as follows:

[0033]

[0034]

[0035] wherein h is a sensory factor; p is a stimulus intensity; α t is the equilibrium index of the tth iteration; α s , α e are the initial value and the final value of α t , respectively; τ is the maximum number of iterations;

[0036] The population is initialized in a manner based on chaotic mapping, at the beginning of iteration, each butterfly is randomly arranged to appear in the search area, the fitness function of the butterfly at this time is calculated; the size of the fragrance generated by each butterfly and the position at this time are calculated by the following formula; the global search is as follows:

[0037]

[0038] wherein r1 is a random number between 0 and 1; are the position vectors of the kth butterfly in the tth iteration and the t+1th iteration, respectively; n k is the fragrance generated by the kth butterfly; g is the optimal position vector of the current butterfly;

[0039] The local search is as follows:

[0040]

[0041] wherein r2 is a random number between 0 and 1; and are the ath and the bth butterflies in the search area, respectively;

[0042] The butterfly searches for food with global search and local search, distinguishes global search and local search by conversion probability p, and uses a random number between (0, 1) for comparison with the conversion probability in each iteration to distinguish global search and local search.

[0043] Preferably, in S3, based on the backdraft risk predicted in S2, an improved butterfly optimization algorithm is used to optimize the airflow regulation strategy, and the optimal working state of each ventilation device is obtained.

[0044] Compared with the prior art, the present application provides an artificial intelligence-based airflow regulation type backdraft intelligent prevention system and method, which has the following beneficial effects:

[0045] The present application can early warn the possible backdraft situation by fusing multi-sensor data to comprehensively consider the environmental factors in the region, combining the knowledge graph and the backdraft prediction model of the optimized extreme learning machine for backdraft analysis and prediction, and intelligently regulating the airflow according to the backdraft risk to automatically adjust the ventilation equipment parameters and reduce the possibility of backdraft. The system has adaptability and reliability, can self-learn and adapt to different scenes, and the redundancy of the sensor ensures accurate risk assessment when failure occurs. It also improves energy utilization efficiency. At the same time, it is convenient to integrate and manage, integrate with other systems, provide a convenient interactive interface, facilitate remote monitoring and management, and improve overall safety and efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 The system block diagram mentioned in embodiment 1 of the present application;

[0047] Figure 2 The contact angle diagram of water droplets contacting the super-hydrophobic coating mentioned in embodiment 1 of the present application;

[0048] Figure 3 The schematic diagram of the lotus leaf surface papillary structure mentioned in embodiment 1 of the present application;

[0049] Figure 4 The schematic diagram of the surface microporous structure of the ceramic matrix composite material mentioned in embodiment 1 of the present application;

[0050] Figure 5 The method flow chart mentioned in embodiment 2 of the present application.

[0051] Significance of marks in the figure:

[0052] 1, sensor layer; 2, processing and analysis layer; 3, decision control layer; 4, hardware execution layer. DETAILED DESCRIPTION

[0053] The technical solutions in the embodiments of the present application will be clearly and completely described with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application.

[0054] The present application can early warn the possible backfire situation by comprehensively considering the environmental factors in the region through fusing multi-sensor data, and combining the knowledge graph and the backfire prediction model of the optimized extreme learning machine to analyze and predict the backfire, and intelligently regulate the airflow according to the backfire risk to automatically adjust the ventilation equipment parameters, thereby reducing the possibility of backfire. The system has self-adaptability and reliability, can self-learn to adapt to different scenes, and the redundancy of the sensor ensures accurate risk assessment when a fault occurs. It also improves energy utilization efficiency. At the same time, it is convenient to integrate and manage, integrate with other systems, provide a convenient interactive interface, facilitate remote monitoring and management, and improve the overall safety and efficiency. Specifically, the following contents are included.

[0055] Embodiment 1:

[0056] Please refer to Figures 1-4 The present application is based on an artificial intelligence-based airflow regulation type backfire intelligent prevention system, which comprises:

[0057] Sensor layer 1: This is the basic layer for the system to obtain data. It contains multiple types of sensors, such as temperature sensors, oxygen concentration sensors, smoke sensors, gas flow rate sensors, infrared imaging sensors, and chemical sensors. These sensors are distributed in areas where fire or backfire risk may occur, and real-time environmental data is collected and transmitted to the data processing center. The above-mentioned sensors have a self-calibration function, which enables them to automatically calibrate over a long period of use, reducing data errors caused by factors such as sensor drift. For example, the temperature sensor can be automatically calibrated every certain time (such as every week) according to the preset standard temperature source, ensuring that the measurement accuracy remains at a high level at all times. It also has a self-diagnosis function that can automatically detect and send an alarm to the system when the sensor fails or performance decreases. For example, if the gas flow rate sensor detects abnormal fluctuations in its own measurement data and exceeds the reasonable range, it will automatically diagnose possible causes of failure, such as probe blockage, and report to the system for timely maintenance or replacement of the sensor to ensure the reliability of the system.

[0058] Temperature sensor: Distributed in various key locations throughout the room, real-time monitoring of indoor temperature changes. These sensors can accurately measure the temperature of different areas, with an accuracy of ±0.5℃, so as to timely discover local high-temperature areas.

[0059] Oxygen Concentration Sensor: Responsible for detecting the oxygen content in the room. Its detection range can cover from normal atmospheric oxygen content to extremely low oxygen content, with an accuracy of ±0.1%, ensuring accurate grasp of oxygen concentration changes and providing key data for airflow regulation.

[0060] Smoke Sensor: Monitors the smoke concentration in the room. It can send an alarm in time when the smoke concentration reaches a certain threshold, with high sensitivity and can detect smoke at an early stage, providing early warning for preventing backdraft.

[0061] Gas Flow Rate Sensor: Installed in the ventilation duct and key parts of indoor air circulation, measures air flow rate. Its measurement accuracy can reach ±0.1m / s, used to monitor air flow rate so that the artificial intelligence system can accurately regulate according to the actual flow rate.

[0062] Infrared thermal imaging sensor, which can more intuitively detect the temperature distribution of object surface, identify potential high temperature areas, with an accuracy of ±0.1℃. By integrating it with traditional temperature sensors, it can more comprehensively and accurately monitor indoor temperature conditions, improve the detection capability of local high temperature points, and thus more effectively prevent backdraft.

[0063] Chemical sensor array can detect a variety of trace combustible gas components, not just common carbon monoxide, but also some specific organic volatile chemicals produced during the fire process. By analyzing the concentration changes of these chemicals and combining artificial intelligence algorithms, it can more accurately predict the possibility of backdraft and improve the accuracy of the system's early warning.

[0064] For example, in a warehouse, temperature sensors are installed at different heights and locations in the goods storage area to monitor temperature changes comprehensively; oxygen concentration sensors are installed near ventilation openings and areas where smoldering materials may exist.

[0065] Processing and Analysis Layer 2: This layer is responsible for receiving data from Sensor Layer 1 and processing and analyzing it. Advanced data processing algorithms and artificial intelligence techniques are used to filter, denoise, and feature extract data to obtain valuable information.

[0066] For example, data processing algorithms are used to remove noise interference in temperature sensor data, making temperature data more accurate; feature extraction techniques are used to extract smoke concentration change characteristics from smoke sensor data. At the same time, machine learning algorithms are used to analyze processed data to predict the likelihood of backdraft occurrence.

[0067] Decision Control Layer 3: Based on the results of the processing and analysis layer 2, this layer makes corresponding decisions and controls hardware facilities such as ventilation equipment to regulate air flow. When the risk of backdraft is high, the optimal air flow regulation strategy is automatically generated, such as adjusting the opening of the ventilation valve, changing the speed and direction of the air circulation fan, controlling the power of the exhaust fan, etc.

[0068] For example, if the system determines that the oxygen concentration in a certain area is too high and the temperature is rising, the decision and control layer will immediately reduce the opening of the ventilation valve in that area while increasing the speed of the air circulation fan to reduce the oxygen concentration and temperature, preventing backdraft from occurring.

[0069] Hardware Execution Layer 4: This layer mainly includes ventilation equipment such as intelligent ventilation valves, air circulation fans, exhaust fans, and adjustable ventilation ducts. These devices perform actual operations according to the instructions of the decision and control layer to regulate air flow.

[0070] Intelligent Ventilation Valve: It can automatically adjust the opening according to the instructions of the artificial intelligence system to control air flow and direction. Its adjustment precision is high, and it can achieve small-scale precise adjustment to meet the fine control requirements of air flow in different situations.

[0071] Air Circulation Fan: It has adjustable speed and direction. It can adjust the speed to change the air circulation speed according to the indoor temperature distribution and air flow conditions. The direction adjustment range is wide, and it can achieve different angles of air supply to promote the uniform mixing of indoor air.

[0072] Exhaust Fan: It has strong exhaust capacity and automatically adjusts the exhaust power according to the concentration of smoke and flammable gas. Its exhaust efficiency is high, and it can quickly remove a large amount of flammable gas and smoke outside the room to ensure indoor air quality.

[0073] Flexible and bendable ventilation ducts: In the event of a fire, these ventilation ducts can automatically change shape and direction according to the instructions of the artificial intelligence system to guide fresh air to the appropriate location and more effectively remove smoke and flammable gas outside the room.

[0074] For example, when receiving instructions to reduce the opening of the ventilation valve, the motor drive device of the ventilation valve will automatically adjust the angle of the valve to achieve the purpose of reducing air flow; the air circulation fan will change the speed and direction according to the instructions to promote the reasonable flow of air.

[0075] In the ventilation equipment, the energy recovery device is integrated, and the heat exchanger is installed at the exhaust fan. When the high-temperature smoke and exhaust gas are discharged, the heat exchanger can recover part of the heat therein for preheating the fresh air entering the room, thereby improving the energy utilization efficiency and reducing the energy consumption caused by the introduction of fresh cold air. In this way, it is expected that 10%-20% of the energy can be recovered, and the overall energy cost of the building can be reduced.

[0076] The kinetic energy generated by the air circulation fan can also be used to generate electricity. A special small power generation device is designed, which can convert the mechanical energy of the rotating fan into electrical energy and store it when the fan is running, to power the sensors or some low-power control devices. This can reduce the dependence on external power supply, improve the self-sufficiency of the system, and also meet the concept of energy saving and environmental protection.

[0077] A new type of high-efficiency filter material is also used at the air inlet and exhaust outlet of the ventilation system: a nanofiber filter with self-cleaning function. This filter not only can more effectively filter dust, particulate matter and some harmful gases in the air, improving the quality of air entering the room, but also installs an ultrasonic transducer on the frame of the filter material to convert electrical energy into ultrasonic vibration energy and transfer it to the filter material. For example, an ultrasonic transducer with a frequency of 20-40 kHz is used. When the transducer is working, the filter material will vibrate at a high frequency with a small amplitude, causing the dirt particles on the surface to loosen and fall off. For the mechanical vibration self-cleaning structure, a cam or eccentric wheel mechanism driven by a motor can be used to make the filter material vibrate at a low frequency with a large amplitude. For example, the motor drives the eccentric wheel to rotate, and the eccentric wheel is connected to the filter material frame through a connecting rod. When the eccentric wheel rotates, it will make the filter material frame vibrate up and down or left and right. The surface of the nanofiber filter is also coated with a super-hydrophobic coating. The super-hydrophobic coating has a special microstructure and chemical composition, which makes the water droplets on its surface have a very high contact angle usually greater than 150° and a very low rolling angle less than 10°. For details, please refer to Figure 2 When the water droplets containing dust or impurities come into contact with the surface of the coating, the water droplets will easily roll away and take away the dirt on the surface. The super-hydrophobic coating usually has a rough structure at the nanometer or micrometer level, such as a papillary structure similar to the surface of a lotus leaf. For details, please refer to Figure 3 On these tiny protruding structures, a layer of low-surface-energy chemicals such as fluorine-containing polymers is coated. When the tiny droplets or water vapor carried in the airflow come into contact with the surface of the coating, they will form nearly spherical droplets. Due to their very small rolling angle, the droplets can quickly roll away under the action of airflow or gravity, taking away the dirt particles and other dirt attached to the surface of the coating. Compared with traditional filters, its filtering efficiency can be improved by 30%-50%, and its service life can be extended by 2-3 times, reducing the need for frequent replacement due to filter clogging and reducing maintenance costs.

[0078] A new composite material with fire-retardant and adsorption functions: ceramic matrix composite, is used for the inner wall of the ventilation duct. This material can prevent the spread of fire in the duct during a fire, while adsorbing some combustible gases and smoke particles that may cause backdraft; it has good high-temperature resistance and fire-retardant properties, and the surface has a microporous structure, which can be referred to Figure 4 , which can adsorb small particles and combustible gas molecules in the smoke, reducing the risk of backdraft.

[0079] Shape memory alloy materials are used in components such as ventilation valves and fan blades. This material can automatically change shape according to changes in temperature and air pressure, thereby adjusting the ventilation volume and direction of the ventilation equipment. For example, when the temperature rises to a certain extent, the ventilation valve blade made of shape memory alloy will automatically bend and reduce the opening to control the airflow and prevent backdraft; when the temperature returns to normal, the blade will automatically return to its original shape.

[0080] Intelligent liquid crystal materials are used to make the shell or part of the structure of the ventilation equipment. This material can change transparency or color according to the control signal of the system, which is used to visually display the running state of the equipment or the backdraft risk level. For example, when the backdraft risk is high, the equipment shell will turn red to remind the staff to pay attention; when the equipment is running normally and the backdraft risk is low, the shell will turn green. At the same time, the intelligent liquid crystal material also has certain heat and sound insulation performance, which can improve the overall performance of the ventilation equipment.

[0081] Example 2:

[0082] Referring to Figure 5 , the airflow regulation type backdraft intelligent prevention method based on artificial intelligence includes the following steps:

[0083] S1: Collect environmental data of the fire area through a plurality of sensors; specifically as follows:

[0084] Based on temperature sensors, oxygen concentration sensors, smoke sensors, gas flow rate sensors, infrared imaging sensors, and chemical sensors, environmental data of the area where a fire may occur or there is a backdraft risk is collected.

[0085] S2: Based on the received sensor data, based on the knowledge graph and the optimized extreme learning machine, backdraft analysis and prediction are performed; specifically as follows:

[0086] The various environmental data collected by various sensors are associated and integrated to construct a backfire knowledge graph, and intelligent fault diagnosis and maintenance are performed based on the knowledge graph. Through the knowledge graph, the artificial intelligence system can more deeply understand the relationship between various factors, thereby making more accurate backfire risk prediction and airflow regulation decisions. The knowledge graph is built based on an HMM model, and entity recognition and relationship extraction are regarded as a sequence labeling problem. The HMM model defines hidden states and observable states, and uses a probability model to predict the sequence of hidden states, thereby constructing the backfire knowledge graph and calculating backfire prediction indexes. Based on extreme learning machines, feature extraction and backfire prediction are performed. The extreme learning machine is as follows:

[0087] When the training set of the single-hidden layer neural network model has m attack samples, the output function of the ELM is as follows:

[0088]

[0089] wherein, u i is an input vector; f(·) is an output vector; m is the total number of hidden layer neurons of the ELM; w j is a weight vector connecting j hidden layer nodes and output layer nodes; σ(·) is an activation function of the hidden layer; ω j is a weight vector connecting the jth input layer node and the hidden layer node; δ j is a threshold value of the jth hidden layer neuron; σ(ω j ·u i + g j ) is an output of the jth hidden layer neuron with respect to the input sample u i ;

[0090] A preset output threshold value is set, and backfire risk is divided based on the value of the output vector.

[0091] The ELM hidden layer parameters are also optimized based on an improved butterfly optimization algorithm. The improved butterfly optimization algorithm is as follows:

[0092] The butterfly fragrance d is calculated as follows:

[0093] d = hp αt

[0094]

[0095] wherein, h is a sensory factor; p is a stimulation intensity; α t is a balance index of the tth iteration; α s , α e are respectively α tthe initial value and the final value of the target function; τ is the maximum number of iterations;

[0096] The population is initialized based on the chaotic mapping. At the beginning of each iteration, each butterfly is randomly assigned to appear in the search area, and the fitness function of the butterfly at this time is calculated. The size of the scent produced by each butterfly and its current position are calculated by the following formula. Global search is as follows:

[0097]

[0098] where r1 is a random number between 0 and 1; are the position vectors of the kth butterfly in the tth iteration and the t+1th iteration, respectively; n k is the scent produced by the kth butterfly; g is the optimal position vector of the current butterfly;

[0099] Local search is as follows:

[0100]

[0101] where r2 is a random number between 0 and 1; and are the ath and bth butterflies in the search area, respectively;

[0102] Butterflies search for food with global search and local search. The conversion probability p is used to distinguish between global search and local search. A random number between 0 and 1 is compared with the conversion probability to distinguish between global search and local search at each iteration.

[0103] Extreme learning machine (ELM) is a new type of single-hidden layer feedforward neural network learning algorithm. It has the characteristics of extremely fast learning speed and can quickly process a large amount of data related to backflow, such as temperature, oxygen concentration, and smoke. Compared with traditional neural network algorithms, ELM can shorten the training time by several times or even dozens of times. ELM also has certain advantages in processing high-dimensional data. Backflow involves multiple complex physical and chemical processes, and the related data often has high-dimensional characteristics. ELM can effectively process these high-dimensional data by randomly initializing input weights and biases and then calculating output weights, and to some extent, avoid the problem of overfitting. For example, it can consider the distribution of temperature at different spatial positions, the trend of oxygen concentration, and the size and concentration of smoke particles, etc. multiple dimensions of data, accurately capture the potential relationship between these factors and backflow, and improve the accuracy of backflow prediction. Based on the above prediction method, the backflow in different scenarios is predicted, and the specific results can be referred to Table 1:

[0104] Table 1 Backflow prediction in different scenarios

[0105]

[0106]

[0107] Experimental studies have shown that, under appropriate parameter settings, ELM can achieve high accuracy in predicting backdraft. By combining ELM with actual backdraft experimental data for verification, it is found that the accuracy of predicting backdraft occurrence can reach about 70%-80%, which is significantly improved compared to some traditional prediction methods. This is because ELM can automatically learn the complex patterns and nonlinear relationships in the data, and for backdraft, which is influenced by multiple factors and has nonlinear characteristics, it can better model and predict. For example, when there are complex interactions between temperature, oxygen concentration, and smoke concentration, ELM can accurately capture these relationships through its unique learning mechanism, thus more accurately predicting whether backdraft will occur.

[0108] S3: Based on the analysis and prediction results, make corresponding decisions and issue airflow control instructions to each ventilation device; the specific steps are as follows:

[0109] Based on the backdraft risk predicted by S2, the improved butterfly optimization algorithm is used to optimize the airflow control strategy to obtain the best working state for each ventilation device. The airflow control strategy generally includes controlling oxygen supply: maintaining the indoor oxygen concentration at a relatively low and stable level to suppress the intensity of combustion reaction. For example, the target oxygen concentration can be set to 18%-20%, close to the oxygen concentration range required for smoldering state, while avoiding sudden increase of oxygen concentration to trigger backdraft.

[0110] Adjusting temperature distribution: removing heat through airflow to reduce indoor temperature. The goal is to control the temperature within a certain range, for example, for common flammable materials, reduce the temperature below the self-ignition temperature. Assuming that the self-ignition temperature of a certain material is 300℃, the target temperature can be set to below 250℃.

[0111] Timely removal of flammable gas and smoke: reduce the concentration of flammable gas in the room. For example, the goal is to reduce the concentration of flammable gas to a certain safe level, such as carbon monoxide concentration below 50ppm, to reduce the possibility of backdraft.

[0112] S4: Each ventilation device receives and executes the corresponding instructions. The specific steps are as follows:

[0113] During the implementation of the airflow control strategy, the temperature, oxygen concentration, smoke concentration, and gas flow rate, etc. parameters in the room are continuously monitored in real time through sensors. For example, temperature data is collected every 1 second, oxygen concentration data is collected every 2 seconds, etc., to ensure that the latest changes in environmental parameters can be obtained in a timely manner.

[0114] The real-time monitored data is constantly fed back to the backdraft risk assessment model, and the input data of the model is updated. If the model finds that the backdraft risk assessment value has changed due to environmental changes, for example, the risk value increases or decreases, the corresponding adjustment mechanism will be triggered.

[0115] If the real-time monitored parameters indicate that the backdraft risk is reduced and is lower than the set low risk threshold, such as the risk value being lower than 0.4, the air flow regulation strategy can be appropriately adjusted to make the ventilation equipment operate in a more energy-saving state under the premise of safety. For example, slightly increasing the opening degree of the ventilation valve, reducing the rotation speed of the air circulation fan, etc.

[0116] If the backdraft risk is still high or further increases, the system will further optimize the air flow regulation strategy. For example, according to the real-time data, some steps of the genetic algorithm are re-run to quickly search for a more optimal strategy. Or according to some pre-set emergency adjustment rules, such as immediately increasing the exhaust fan power to the maximum value while further reducing the opening degree of the ventilation valve, etc., to deal with emergency situations and ensure that the backdraft risk is always effectively controlled.

[0117] The above is only the preferred specific implementation of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can make equivalent replacement or change within the technical range disclosed by the present application according to the technical solution and the inventive concept of the present application, which should be covered within the protection scope of the present application.

Claims

1. An intelligent backfire prevention system based on artificial intelligence and airflow regulation, characterized in that, include: Sensor layer (1): Used to collect environmental data of the fire area through several sensors; Processing and Analysis Layer (2): Used to receive data from the sensor layer (1) and perform relapse analysis and prediction based on the knowledge graph and optimized extreme learning machine; specifically including: collecting various environmental data collected by the sensors, constructing a relapse knowledge graph through association and integration, and performing intelligent fault diagnosis and maintenance based on the knowledge graph; the knowledge graph is built based on the HMM model, taking entity recognition and relation extraction as sequence labeling problems, the HMM model defines hidden states and observable states, uses a probability model to predict the hidden state sequence, thereby constructing a relapse knowledge graph, calculating relapse prediction indicators, and performing feature extraction and relapse prediction based on the extreme learning machine, the extreme learning machine is as follows: When the training set of a single hidden layer neural network model has When there are multiple attack samples, the ELM output function is as follows: in, The input vector; This is the output vector; This represents the total number of hidden layer neurons in the ELM; For connection The weight vectors of each hidden layer node and the output layer node; The activation function for the hidden layer; To connect the first The weight vectors of each input layer node and hidden layer node; For the first The threshold of each hidden layer neuron; For the first Each hidden layer neuron is relative to the input sample The output; Preset output thresholds and classify relapse risk based on the values ​​of the output vector; Furthermore, the ELM hidden layer parameters are optimized based on an improved butterfly optimization algorithm, which is as follows: Butterfly scent The calculation is as follows: in, For sensory factors; Stimulus intensity; For the first The balance index of the next iteration; , They are respectively The initial and final values; This represents the maximum number of iterations. The population is initialized using a chaotic mapping approach. At the start of the iteration, each butterfly is randomly placed within the search region, and its fitness function is calculated and preserved at that moment. The amount of fragrance produced by each butterfly and its current position are calculated using the following formula. The global search is as follows: in, for Random numbers between; , The first The second iteration and the first The iteration of the ... The position vector of the butterfly; For the first The scent produced by a butterfly; This represents the current optimal position vector for the butterfly. The local search is as follows: in, for Random numbers between; and The first and second in the search area, respectively Only with the first A butterfly; Butterflies employ both global and local searches when searching for food, which is achieved through probability transformation. Distinguish between global search and local search, and use it in each iteration. The random numbers and transformation probabilities are compared to distinguish between global search and local search; Decision control layer (3): Makes corresponding decisions based on the analysis and prediction results of the processing and analysis layer (2), and issues corresponding airflow control commands to the hardware execution layer (4); Hardware execution layer (4): Receives control commands from the decision control layer (3), controls the corresponding ventilation equipment to execute commands, and also provides real-time feedback of airflow regulation results through the sensor layer (1).

2. The airflow-controlled flashback intelligent prevention system based on artificial intelligence according to claim 1, characterized in that, The sensors in the sensor layer (1) include a temperature sensor, an oxygen concentration sensor, a smoke sensor, a gas flow rate sensor, an infrared imaging sensor, and a chemical sensor.

3. The airflow-controlled flashback intelligent prevention system based on artificial intelligence according to claim 2, characterized in that, The sensor layer (1) periodically performs automatic calibration of various sensors through self-calibration technology, performs self-diagnosis, and sends alarms to the system based on the abnormal results of the diagnosis.

4. The airflow-controlled flashback intelligent prevention system based on artificial intelligence according to claim 1, characterized in that, The ventilation equipment of the hardware execution layer (4) includes an intelligent ventilation valve, an air circulation fan, a smoke exhaust fan and an adjustable ventilation duct; the hardware execution layer (4) uses intelligent liquid crystal material to prepare the outer shell and some structural components of the ventilation equipment, and changes the transparency or color according to the control signal of the system to intuitively display the operating status of the equipment or the risk level of backfire.

5. The airflow-controlled flashback intelligent prevention system based on artificial intelligence according to claim 4, characterized in that, The blades of the intelligent ventilation valve and the air circulation fan are made of shape memory alloy material, which automatically changes shape according to changes in temperature and airflow pressure, thereby adjusting the ventilation volume and airflow direction of the ventilation equipment.

6. The airflow-controlled flashback intelligent prevention system based on artificial intelligence according to claim 5, characterized in that, A heat exchanger is also installed at the exhaust fan. When high-temperature smoke and exhaust gas are discharged, the heat exchanger recovers some of the energy to preheat the fresh air entering the room. The air inlet and exhaust outlet of the exhaust fan use nanofiber filters with self-cleaning function. An ultrasonic transducer is installed on the frame of the nanofiber filter to convert electrical energy into ultrasonic vibration energy and transfer it to the filter material for self-cleaning. The surface of the nanofiber filter is also coated with a superhydrophobic coating.

7. The airflow-controlled flashback intelligent prevention system based on artificial intelligence according to claim 6, characterized in that, The ventilation duct is internally constructed with a ceramic-based composite material that has flame-retardant and adsorption functions, and has a microporous structure.

8. The AI-based airflow regulation-based intelligent backfire prevention method applied to the system described in any one of claims 1-7, characterized in that, Includes the following steps: S1: Collect environmental data of the fire area using several sensors; S2: Based on received sensor data, backfire analysis and prediction are performed using knowledge graphs and optimized extreme learning machines; S3: Make corresponding decisions based on the analysis and prediction results, and issue corresponding airflow control instructions to each ventilation device; S4: Each ventilation device receives and executes the corresponding instructions, and provides real-time feedback on the airflow control results through sensors.

9. The intelligent backfire prevention method based on airflow regulation according to claim 8, characterized in that, In step S3, based on the backfire risk predicted in step S2, the airflow control strategy is optimized using an improved butterfly optimization algorithm to obtain the best working state for each ventilation device.

Images