Control system and method for AI micro-blasting glass-blasting escape

Abstract

The invention discloses a control system and method for AI micro-blasting glass-breaking escape, and relates to the technical field of glass-breaking escape, and the method comprises the following steps: S1, a sensor collects and uploads various sensor data; s2, performing data processing on the acquired various sensor data through a main control board controller, and uploading the processed sensor data to a server for processing through a network; s3, the server processes and outputs a danger coefficient and feeds the danger coefficient back to the main control board controller; and S4, when the danger coefficient exceeds a preset threshold value, the main control board controller starts a local escape system. According to the method, the danger coefficient is judged and evaluated through various sensor data, when the danger coefficient exceeds a preset threshold value, the method can send a glass breaking triggering instruction to each node of the glass breaking escape system, and a micro-explosion ignition device breaks glass after receiving the instruction, so that an escape channel is opened up for trapped people.

Description

Technical Field

[0001] The present invention relates to the technical field of glass breaking escape, and in particular to a control system method for AI micro-blasting glass breaking escape. Background Art

[0002] When emergencies such as fires, gas leaks, and floods occur, traditional escape methods rely on manual or simple mechanical manipulation. In emergencies, people may be unable to effectively and promptly break windows and escape due to panic, injury, and other factors. Existing emergency escape tools often include a safety hammer, which is used to break glass in emergencies or disasters to ensure escape. However, this method has drawbacks. It requires a relatively high level of skill, is difficult to shatter high-strength tempered glass in one go, and is easily stolen, making it difficult to quickly open an escape route. Traditional escape safety hammers fail due to factors such as time, space, physiology, and psychology, as well as factors like fog, darkness, and potential loss. Consequently, they fail to effectively and promptly create an escape route, preventing people from escaping quickly. This often results in numerous casualties, endangering life and property, and impacting overall social stability. Consequently, there is an urgent need for a fast, efficient, safe, and reliable glass-breaking device to replace the traditional "safety hammer" escape tool.

[0003] Window-breaking escape will develop towards intelligent technology. Specific aspects of monitoring, window breaking, and escape will become increasingly intelligent, enabling escape through window-breaking without human intervention. Window-breaking devices will be integrated into the internet, sending an alarm via internet phone or other means simultaneously with window breaking, minimizing casualties by providing the fastest possible means or requiring outside assistance. However, existing technologies suffer from the problem of underreporting with traditional single-sensor systems, necessitating the coordinated processing of data from multiple sensors.

[0004] For example, Chinese patent publication number CN116279258A discloses a system and method for automatically breaking windows after a car collision. The system includes an electrically connected window breaking control module, a collision sensing module, a body control module, a driver monitoring module, a passenger monitoring module, a 360-degree panoramic view module, a voice control module, and a window breaking device. After a collision occurs, the driver monitoring module and the passenger monitoring module start personnel motion monitoring, the door lock status monitoring unit in the body control module monitors the door lock status, the 360-degree panoramic view module monitors the door lock status, the window breaking control module determines and identifies whether window breaking is required, and starts voice confirmation according to user settings, and controls the window breaking device installed inside the side door to break the window glass. In the extreme case where the side door cannot be opened after a car collision, the car window glass is automatically broken, thereby increasing the escape route, reducing the risk of secondary injury to people, and improving the active safety performance of the car.

[0005] The above patents all have the problems raised by this background technology: the low accuracy of traditional single sensors will lead to missed reports; at the same time, determining whether the glass is broken based solely on sensor data is not comprehensive enough, and requires multiple sensor data processing and the installation of special authorization devices for manual control. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to address the deficiencies of the existing technology and provide a control system method for AI micro-blasting glass escape.

[0007] In order to achieve the above object, the technical solution adopted by the present invention is:

[0008] A control system method for AI micro-blasting glass escape includes the following steps:

[0009] Step S1: The sensor collects and uploads a variety of sensor data, wherein the various sensor data include: smoke data, temperature data, gas data and image data;

[0010] Step S2: Process the collected sensor data through the main control board controller, and upload the processed sensor data to the server through the network for processing;

[0011] Step S3: The server processes and outputs the risk factor and feeds it back to the main control board controller;

[0012] Step S4: When the risk factor exceeds a preset threshold, the main control board controller activates the local escape system.

[0013] Furthermore, after the local escape system is started, the main control board controller will send a glass breaking trigger instruction to each node of the glass breaking escape system. The node is equipped with a micro-explosion ignition device. After receiving the instruction, the micro-explosion ignition device will quickly start to break the glass, opening an escape route for trapped people.

[0014] Furthermore, the local escape system also includes a special authorization device, which performs manual control according to the development status of the local event status, wherein the manual control specifically includes: the staff authorizes access to the local escape system through password verification, fingerprint recognition or facial recognition, presses the detonation button, and starts the local escape system in advance.

[0015] Furthermore, the step S2 specifically includes:

[0016] Step S2.1, time-synchronizing the collected data from various sensors;

[0017] Step S2.2, performing denoising on the time-synchronized multiple sensor data;

[0018] Step S2.3, respectively extracting features from the denoised multi-sensor data;

[0019] Step S2.4, uploading the extracted sensor data features to the server for processing.

[0020] Further, the step S2.1 specifically comprises:

[0021] The total compensation time amount is calculated, and the specific formula is:

[0022]

[0023] Wherein, Δt represents the total compensation time amount, N represents the number of sensor types, i represents the i-th sensor, t i represents the inherent delay of the i-th sensor, t trig represents the global trigger signal timestamp, and α represents the drift gain coefficient, represents the clock drift rate;

[0024] The original timestamp is compensated to obtain the synchronization time, and the calculation formula is:

[0025] T=Δt+t raw

[0026] Wherein, T represents the synchronization time, t raw represents the original timestamp.

[0027] Further, the step S2.3 specifically comprises:

[0028] The multi-scale scattering feature of the smoke sensor is extracted, and the specific formula is:

[0029]

[0030] Wherein, S represents the multi-scale scattering feature of the smoke sensor, I 450 and I 650 respectively represent the blue light band scattering intensity and the red light band scattering intensity, σ(·) represents the Sigmoid activation function, represents the Laplace Gaussian operator, I scatter represents the scattering light intensity distribution image matrix, and * represents convolution operation;

[0031] The dynamic temperature change energy feature of the temperature sensor is extracted, and the specific formula is:

[0032]

[0033] Wherein, E represents the dynamic temperature change energy feature of the temperature sensor, Δm represents the energy integration window with respect to time, d T represents the high temperature region temperature instantaneous change rate, represents the instantaneous rate of change of background temperature, and x represents the time variable of energy integration;

[0034] Extract the leakage risk characteristics of the gas sensor. The specific formula is:

[0035]

[0036] Among them, R represents the leakage risk characteristic of the gas sensor, Δn represents the risk integration window with respect to time, C represents the current gas concentration, and C th represents the gas concentration safety threshold, λ represents the risk time attenuation factor, and y represents the time variable of the risk integral;

[0037] Extract the flame visual features of the image sensor. The specific formula is:

[0038] V=ReLU(W c *CNN(I)

[0039] Where V represents the flame visual feature of the image sensor, ReLU(·) represents the ReLU activation function, and W c Represents the flame feature enhancement convolution kernel, I represents the input image, and CNN(I) represents the image feature matrix extracted by the CNN neural network.

[0040] Furthermore, the step S3 specifically includes:

[0041] Step S3.1: Input the multi-scale scattering characteristics S of the smoke sensor, the dynamic temperature change energy characteristics E of the temperature sensor, the leakage risk characteristics R of the gas sensor, and the image data into the risk assessment model;

[0042] Step S3.3, the risk assessment model outputs the risk coefficient;

[0043] Step S3.4: The server risk factor is fed back to the main control board controller.

[0044] Furthermore, the risk assessment model specifically includes: an input layer, an output layer, a hidden layer 1 and a hidden layer 2;

[0045] Among them, the input vector of the input layer is expressed as:

[0046]

[0047] The calculation formula for hidden layer 1 is:

[0048] h1=ReLU(W1X+b1)

[0049] Where h1 represents the output of hidden layer 1, W1 represents the input layer weight, and b1 represents the bias of hidden layer 1;

[0050] The calculation formula for hidden layer 2 is:

[0051] h2=ReLU(W2h1+b2)

[0052] Where h2 represents the output of hidden layer 1, W2 represents the weight of hidden layer, and b2 represents the bias of hidden layer 2;

[0053] The calculation formula of the output layer is:

[0054] Risk = σ(W3h2+b3)

[0055] Among them, Risk represents the risk coefficient, W3 represents the output layer weight, and b3 represents the output layer bias.

[0056] Furthermore, in step S4, the preset threshold is determined according to the specific escape scenario.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] 1. This invention uses a variety of professional and special-purpose designed sensors to form a perception system for the AI ​​control + micro-explosion escape system. It can obtain surrounding environmental information comprehensively, accurately and timely, adapt to more complex scenarios, combine the signal processing and judgment capabilities of the main control board controller and the near-end / remote main server, and use AI algorithms for in-depth analysis, thereby improving the accuracy and timeliness of emergency judgment.

[0059] 2. The present invention improves detection accuracy through the collaborative detection of multimodal sensors and the fusion of four-dimensional features, including: smoke spectral scattering characteristics + temperature change energy characteristics + gas leakage risk characteristics + flame visual characteristics, thereby reducing the false alarm rate and the missed alarm rate.

[0060] 3. The activation of the glass-breaking escape system of the present invention and the escape triggering of related glass-breaking nodes within the system are fully AI+ automatic activation and triggering, which can accurately identify dangerous situations and respond quickly. This automated feature greatly improves escape efficiency, ensures a rapid response in emergency situations, and provides timely escape routes for victims.

[0061] 4. To adapt to diverse real-world application scenarios and environments, this system incorporates encrypted authorization for remote and local glass-break escape control. During data transmission, advanced encryption algorithms are employed to encrypt sensor data, control instructions, and user identity information, ensuring the confidentiality, integrity, and authenticity of data during network transmission. For local control, the system provides administrators with a convenient and secure user interface. Through encrypted authentication mechanisms such as password verification, fingerprint recognition, and facial recognition, only authorized personnel can access the local control system. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments made with reference to the following drawings:

[0063] Figure 1 Schematic diagram of a flow chart of an embodiment of the present invention;

[0064] Figure 2 A data processing structure diagram of an embodiment of the present invention;

[0065] Figure 3 This is a structural diagram of a risk assessment model according to an embodiment of the present invention;

[0066] Figure 4 2 is a system structure diagram of an embodiment of the present invention. DETAILED DESCRIPTION

[0067] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0068] like Figure 1 As shown, a control system method for AI micro-blasting glass escape includes the following steps:

[0069] Step S1: The sensor collects and uploads a variety of sensor data, wherein the various sensor data include: smoke data, temperature data, gas data and image data;

[0070] Step S2: Process the collected sensor data through the main control board controller, and upload the processed sensor data to the server through the network for processing;

[0071] Step S3: The server processes and outputs the risk factor and feeds it back to the main control board controller;

[0072] Step S4: When the risk factor exceeds a preset threshold, the main control board controller activates the local escape system.

[0073] After the local escape system is started, the main control board controller will send a glass breaking trigger instruction to each node of the glass breaking escape system. The node is equipped with a micro-explosion ignition device. After receiving the instruction, the micro-explosion ignition device will quickly start to break the glass, opening an escape route for trapped people.

[0074] The local escape system also includes a special authorization device, which performs manual control according to the development status of the local event status. The manual control specifically includes: the staff authorizes access to the local escape system through password verification, fingerprint recognition or facial recognition, presses the detonation button, and starts the local escape system in advance.

[0075] like Figure 2As shown, the step S2 specifically includes:

[0076] Step S2.1, time-synchronizing the collected data from various sensors;

[0077] Step S2.2, performing denoising on the time-synchronized multiple sensor data;

[0078] Step S2.3, performing feature extraction on the denoised sensor data;

[0079] Step S2.4: Upload the extracted sensor data features to the server for processing.

[0080] The step S2.1 specifically includes:

[0081] Calculate the total compensation time. The specific formula is:

[0082]

[0083] Where Δt represents the total compensation time, N represents the number of sensor types, i represents the i-th sensor, and t i represents the intrinsic delay of the i-th sensor, t trig represents the global trigger signal timestamp, α represents the drift gain coefficient, Indicates the clock drift rate;

[0084] The sensor type and delay calibration values ​​are: smoke: 50ms, image: 100ms, gas: 40ms; the drift gain coefficient is set to 0.2-0.8; the clock drift rate is measured using the PTP protocol;

[0085] The original timestamp is compensated to get the synchronization time. The calculation formula is:

[0086] T=Δt+t raw

[0087] Where T represents the synchronization time, t raw Represents the original timestamp.

[0088] The step S2.3 specifically includes:

[0089] Extract the multi-scale scattering characteristics of the smoke sensor. The specific formula is:

[0090]

[0091] Where S represents the multi-scale scattering characteristics of the smoke sensor, I 450 and I 650 They represent the scattering intensity of the blue light band and the red light band, σ(·) represents the Sigmoid activation function, represents the Laplace Gaussian operator, I scatter represents the scattered light intensity distribution image matrix, * represents the convolution operation;

[0092] Smoke particles scatter short wavelength light (blue light) more strongly and long wavelength light (red light) less strongly. This ratio quantifies the wavelength dependence of smoke:

[0093] Positive value increases → smoke probability is high (blue light scattering is enhanced)

[0094] Close to 0 → non-smoke particles (such as dust that scatters blue and red evenly).

[0095] Output characteristics:

[0096] Positive → Bright area (center of smoke)

[0097] Negative value → dark edge (smoke boundary)

[0098] Zero value → uniform area.

[0099] Extract the dynamic temperature change energy characteristics of the temperature sensor. The specific formula is:

[0100]

[0101] Where E represents the dynamic temperature change energy characteristic of the temperature sensor, Δm represents the energy integration window with respect to time, and d T Indicates the instantaneous rate of change of temperature in the high temperature area, represents the instantaneous rate of change of background temperature, and x represents the time variable of energy integration;

[0102] The typical value of Δm is 60s, which deducts the environmental background changes, specifically measures abnormal temperature rise, and integrates to capture slowly developing fire hazards;

[0103] Extract the leakage risk characteristics of the gas sensor. The specific formula is:

[0104]

[0105] Among them, R represents the leakage risk characteristic of the gas sensor, Δn represents the risk integration window with respect to time, C represents the current gas concentration, and C th represents the gas concentration safety threshold, λ represents the risk time attenuation factor, and y represents the time variable of the risk integral;

[0106] Δn typical 30s, λ typical 0.1s -1 , exponential decay: intensifies recent high risks; continuous accumulation: the integral captures low-concentration continuous leakage, the larger the integral value → the higher the leakage risk.

[0107] Extract the flame visual features of the image sensor. The specific formula is:

[0108] V=ReLU(W c *CNN(I)

[0109] Where V represents the flame visual feature of the image sensor, ReLU(·) represents the ReLU activation function, and W c Represents the flame feature enhancement convolution kernel, I represents the input image, and CNN(I) represents the image feature matrix extracted by the CNN neural network.

[0110] W c It is a learnable feature enhancement filter that is specifically designed to enhance the visual features of flames (such as flame edges, brightness gradients, color distribution, etc.

[0111] Example: Wc=[

[0112] [[0.1,-0.2,0.1], / / Edge detection kernel (red channel)

[0113] [-0.2,0.8,-0.2],

[0114] [0.1,-0.2,0.1]],

[0115] [[0.0,0.3,0.0], / / Brightness enhancement kernel (green channel)

[0116] [0.3,1.2,0.3],

[0117] [0.0,0.3,0.0]],

[0118] [[-0.1,0.4,-0.1], / / Color selection kernel (blue channel)

[0119] [0.4,0.6,0.4],

[0120] [-0.1,0.4,-0.1]]

[0121] ] / / Actually a 3×3 tensor with 576 input channels and 64 output channels.

[0122] Function:

[0123] Positive weight areas: Enhance flame characteristics (such as highlight areas)

[0124] Negative weight area: suppress background interference (such as reflective objects)

[0125] Automatically learn the spatial-spectral characteristics of flames through training.

[0126] * represents a discrete convolution operation, which is essentially a weighted summation of local areas and is used to extract spatial features.

[0127] Input feature map F (3×3 area) and convolution kernel Wc (3×3)

[0128] [0.1,0.4,0.2] [0.1,-0.2,0.1]

[0129] [0.3,0.9,0.5]* [-0.2,0.8,-0.2]

[0130] [0.2,0.7,0.3][0.1,-0.2,0.1]

[0131] Calculation process:

[0132] =(0.1×0.1)+(0.4×-0.2)+...+(0.3×0.1)

[0133] =0.01-0.08+...+0.03=0.42

[0134] The step S3 specifically includes:

[0135] Step S3.1: Input the multi-scale scattering characteristics S of the smoke sensor, the dynamic temperature change energy characteristics E of the temperature sensor, the leakage risk characteristics R of the gas sensor, and the image data into the risk assessment model;

[0136] Step S3.3, the risk assessment model outputs the risk coefficient;

[0137] Step S3.4: The server risk factor is fed back to the main control board controller.

[0138] like Figure 3 As shown, the risk assessment model specifically includes: an input layer, an output layer, a hidden layer 1 and a hidden layer 2;

[0139] Among them, the number of neurons in the input layer is 4, the number of neurons in the output layer is 1, the number of neurons in the hidden layer 1 is 8, and the number of neurons in the hidden layer 2 is 4.

[0140] Among them, the input vector of the input layer is expressed as:

[0141]

[0142] The calculation formula for hidden layer 1 is:

[0143] h1=ReLU(W1X+b1)

[0144] Where h1 represents the output of hidden layer 1, W1 represents the input layer weight, and b1 represents the bias of hidden layer 1;

[0145] The calculation formula for hidden layer 2 is:

[0146] h2=ReLU(W2h1+b2)

[0147] Where h2 represents the output of hidden layer 1, W2 represents the weight of hidden layer, and b2 represents the bias of hidden layer 2. The calculation formula of the output layer is:

[0148] Risk = σ(W3h2+b3)

[0149] Among them, Risk represents the risk factor, W3 represents the output layer weight, and b3 represents the output layer bias. The specific settings are shown in Table 1:

[0150] Table 1

[0151] parameter Dimensions Physical meaning <![CDATA[W1]]> 8*4 Mapping of features to hidden layers <![CDATA[b1]]> 8*1 Neuron activation threshold <![CDATA[W2]]> 4*8 Feature Abstraction Transformation <![CDATA[b2]]> 4*1 High-order feature migration <![CDATA[W3]]> 1*4 Risk decision weight <![CDATA[b3]]> Scalar Basic risk value

[0152] In step S4, the preset threshold is determined according to the specific escape scenario.

[0153] For example: Threshold setting basis description

[0154] Time window principle

[0155] Vehicle falling into water: Lowest threshold (0.20) → Reaction time < 30 seconds

[0156] Casualty Cost Principle

[0157] Laboratory accident: Higher threshold (0.15) → Professional personnel present

[0158] Secondary risk principle

[0159] Stampede scenario: Medium risk threshold 0.45 → Prevent panic from spreading

[0160] Subway fire: High risk threshold 0.65 → Avoid escaping in the wrong direction

[0161] Dynamic adjustment mechanism:

[0162] Night mode: all thresholds lowered by 0.05-0.15

[0163] Severe weather: threshold reduced by an additional 0.1

[0164] Historical incidents: The threshold for similar scenarios is permanently lowered by 0.2

[0165] For example: In a shopping mall where a fire occurred, the original threshold is 0.45 → 0.25

[0166] like Figure 4As shown, the power module supplies power to the main control board controller. The sensor module, equipped with smoke sensors, temperature sensors, and gas sensors, monitors environmental changes in real time and converts the received signals into electrical signals for transmission to the main control board controller. The camera is primarily responsible for capturing images and transmitting them to the main control board controller. The main control board controller uploads complex image signals that it cannot process to the local / remote main server via the wireless transmission module. The main server uses powerful computing power and advanced AI algorithms to conduct in-depth analysis and judgment of the signals, and then feeds the results back to the main control board controller via the wireless network module. The main control board controller processes and analyzes these signals and determines whether to activate the AI ​​glass-breaking escape system. If it determines that danger exists and the glass-breaking escape conditions are met, the main control board sends a glass-breaking command to each node in the glass-breaking escape system. Upon receiving the command, the micro-explosion ignition device is quickly activated, rapidly shattering the glass and creating an escape route for victims. Furthermore, each AI micro-explosion glass-breaking escape control system is equipped with a special authorization device, which allows personnel to manually control it based on the development of the local event status. Human control is an encrypted authorization system. Staff can authorize access to the local control system through password verification, fingerprint recognition, facial recognition, etc., press the detonation button, and start the local escape system in advance.

[0167] In this embodiment, the local controller uploads sensor signals that are difficult to process locally to the local / remote server via the network and waits for the server to feedback the processing results. The entire AI micro-blasting glass escape control system requires both software and hardware. The main control board controller connects to multiple sensors to ensure normal signal transmission. The main control board controller is installed with programs related to signal processing. When the main control board controller receives sensor input signals, it analyzes and processes the sensor input signals according to preset thresholds in the software program to determine whether the current environment is dangerous. The main control board controller also has a wireless network transmission port. By configuring the main control board controller's network connection parameters, it can communicate with the local / remote main server. If some situations cannot be handled, such as the presence of unusual activities, the camera will first transmit the real-time monitoring image to the main control board controller. The main control board controller then uploads the real-time monitoring image to the local / remote main server via the wireless network. The main server uses its powerful computing power and AI algorithms to conduct in-depth analysis and judgment on the signals, and then feedback the processing results to the main control board controller.

[0168] In this embodiment, the main control board controller decides whether the local escape system is ready to start based on the processing results output by the remote and local main servers: after receiving various signals input by the sensors and the signals fed back by the main server, the main control board controller processes and judges these signals according to the preset thresholds in the software program, and decides whether the local escape system needs to be started.

[0169] In this embodiment, the micro-explosion ignition device is the basis for the safe and reliable use of the AI ​​control + micro-explosion escape system. Compared with traditional electronic detonators, the micro-explosion ignition device has multiple safety protection mechanisms. On the one hand, it sets a strict ignition threshold. The ignition program will only be started when the main control board controller sends a trigger signal that meets the specific coding and voltage requirements, effectively preventing accidental detonation caused by external electromagnetic interference, misoperation and other factors. On the other hand, the device adopts a multi-layer explosion-proof and insulation design in its structure, which can withstand a certain degree of harsh environmental influences such as temperature, humidity, and vibration, ensuring that it can maintain stable performance in various complex practical application scenarios and will not fail or work abnormally due to environmental influences. The energy release of the micro-explosion ignition device can be precisely controlled. While ensuring the glass shattering effect, the impact force generated by the explosion and the range of fragment splashing are controlled within a safe range, avoiding accidental injury to surrounding personnel and providing reliable protection for personnel escape.

[0170] The computer-readable storage medium of this embodiment may be an internal storage unit of the terminal, such as a hard disk or memory of the terminal; the computer-readable storage medium of this embodiment may also be an external storage device of the terminal, such as a plug-in hard disk, a smart memory card, a secure digital card, a flash memory card, etc. equipped on the terminal; further, the computer-readable storage medium may also include both an internal storage unit of the terminal and an external storage device.

[0171] The computer-readable storage medium of this embodiment is used to store computer programs and other programs and data required by the terminal. The computer-readable storage medium can also be used to temporarily store data that has been output or is to be output.

[0172] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware through a computer program. The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0173] The examples described in the present application are merely to describe the preferred embodiments of the present application, and are not intended to limit the concept and scope of the present application. Without departing from the design idea of the present application, various modifications and improvements of the technical solutions of the present application made by the engineering technicians in the field shall fall within the protection scope of the present application.

Claims

1. A control system method for AI micro-blasting glass escape, characterized in that: The steps include: Step S1: The sensor collects and uploads a variety of sensor data, wherein the various sensor data include: smoke data, temperature data, gas data and image data; Step S2: Process the collected sensor data through the main control board controller, and upload the processed sensor data to the server through the network for processing; Step S3: The server processes and outputs the risk factor and feeds it back to the main control board controller; Step S4: When the risk factor exceeds a preset threshold, the main control board controller activates the local escape system.

2. The method according to claim 1, characterized in that After the local escape system is started, the main control board controller will send a glass breaking trigger instruction to each node of the glass breaking escape system. The node is equipped with a micro-explosion ignition device. After receiving the instruction, the micro-explosion ignition device will quickly start to break the glass, opening an escape route for trapped people.

3. The method according to claim 1, characterized in that The local escape system also includes a special authorization device, which performs manual control according to the development status of the local event status. The manual control specifically includes: the staff authorizes access to the local escape system through password verification, fingerprint recognition or facial recognition, presses the detonation button, and starts the local escape system in advance.

4. The method according to claim 1, wherein The step S2 specifically includes: Step S2.1, time-synchronizing the collected data from various sensors; Step S2.2, performing denoising on the time-synchronized multiple sensor data; Step S2.3, performing feature extraction on the denoised sensor data; Step S2.4: Upload the extracted sensor data features to the server for processing.

5. The method according to claim 4, characterized in that The step S2.1 specifically includes: Calculate the total compensation time. The specific formula is: Where Δt represents the total compensation time, N represents the number of sensor types, i represents the i-th sensor, and t i represents the intrinsic delay of the i-th sensor, t trig represents the global trigger signal timestamp, α represents the drift gain coefficient, Indicates the clock drift rate; The original timestamp is compensated to get the synchronization time. The calculation formula is: T=Δt+t raw Where T represents the synchronization time, t raw Represents the original timestamp.

6. The method according to claim 5, characterized in that The step S2.3 specifically includes: Extract the multi-scale scattering characteristics of the smoke sensor. The specific formula is: Where S represents the multi-scale scattering characteristics of the smoke sensor, I 450 and I 650 They represent the scattering intensity of the blue light band and the red light band, σ(·) represents the Sigmoid activation function, ▽ 2 G represents the Laplace Gaussian operator, I scatter represents the scattered light intensity distribution image matrix, * represents the convolution operation; Extract the dynamic temperature change energy characteristics of the temperature sensor. The specific formula is: Where E represents the dynamic temperature change energy characteristic of the temperature sensor, Δm represents the energy integration window with respect to time, and d T Indicates the instantaneous rate of change of temperature in the high temperature area, represents the instantaneous rate of change of background temperature, and x represents the time variable of energy integration; Extract the leakage risk characteristics of the gas sensor. The specific formula is: Among them, R represents the leakage risk characteristic of the gas sensor, Δn represents the risk integration window with respect to time, C represents the current gas concentration, and C th represents the gas concentration safety threshold, λ represents the risk time attenuation factor, and y represents the time variable of the risk integral; Extract the flame visual features of the image sensor. The specific formula is: V=ReLU(W c *CNN(I)) Where V represents the flame visual feature of the image sensor, ReLU(·) represents the ReLU activation function, and W c Represents the flame feature enhancement convolution kernel, I represents the input image, and CNN(I) represents the image feature matrix extracted by the CNN neural network.

7. The method according to claim 6, characterized in that The step S3 specifically includes: Step S3.1: Input the multi-scale scattering characteristics S of the smoke sensor, the dynamic temperature change energy characteristics E of the temperature sensor, the leakage risk characteristics R of the gas sensor, and the image data into the risk assessment model; Step S3.3, the risk assessment model outputs the risk coefficient; Step S3.4: The server risk factor is fed back to the main control board controller.

8. The method according to claim 7, characterized in that The risk assessment model specifically includes: an input layer, an output layer, a hidden layer 1 and a hidden layer 2; Among them, the input vector of the input layer is expressed as: The calculation formula for hidden layer 1 is: h1=ReLU(W1X+b1) Where h1 represents the output of hidden layer 1, W1 represents the input layer weight, and b1 represents the bias of hidden layer 1; The calculation formula for hidden layer 2 is: h2=ReLU(W2h1+b2) Where h2 represents the output of hidden layer 1, W2 represents the weight of hidden layer, and b2 represents the bias of hidden layer 2; The calculation formula of the output layer is: Risk = σ(W3h2+b3) Among them, Risk represents the risk coefficient, W3 represents the output layer weight, and b3 represents the output layer bias.

9. The method according to claim 5, characterized in that In step S4, the preset threshold is determined according to the specific escape scenario.

Images