Point-mode extremely-early-stage optical gas composite smoke detection system

Abstract

The invention, which relates to the technical field of modern intelligent fire-fighting equipment, discloses a point-mode extremely-early optical gas composite smoke detection system comprising a power supply communication unit, a main control unit, a detection unit and an air pump unit. The detection unit comprises a CO sensor, a CO2 sensor, an oxygen sensor, four light intensity sensors, two laser light sources with different wavelengths and a laser control and light intensity acquisition module, and is used for transmitting acquired light intensity signals and various gas concentration data to the main control unit; the power supply communication unit can realize the functions of power supply isolation, power supply conversion and RS485 communication, supplies power to other units of the system and provides communication interfaces. According to the invention, a plurality of gas sensors, such as a CO sensor, a CO2 sensor and an oxygen sensor, as well as a light intensity sensor and a laser light source, are integrated, so that gas and smoke parameters related to a plurality of fire disasters can be monitored at the same time, and the comprehensiveness and accuracy of fire detection are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of modern intelligent fire fighting equipment, and in particular to a point-type very early optical gas composite smoke detection system. Background Art

[0002] Very early fire detectors are a type of fire detection equipment developed using advanced science and technology. Their purpose is to conduct effective detection at the very early stage of a fire so that timely measures can be taken to prevent the spread and diffusion of the fire. In the current context of social development, fire safety issues are increasingly receiving attention and attention, especially in densely populated, complex buildings, and densely populated urban environments. Fire prevention and early detection are particularly important. On the one hand, with the acceleration of urbanization, there are more and more high-rise buildings, large commercial complexes, industrial production facilities, and other places. Once a fire occurs in these places, it often causes huge economic losses and casualties. Therefore, higher requirements are placed on fire safety in these places. On the other hand, with the development of science and technology, people's living standards are constantly improving, and the demand for fire safety is also increasing. People pay more attention to quality of life and safety, and hope to achieve early detection and effective prevention and control of fires through advanced technical means. The following problems exist in the existing technology:

[0003] Because existing fire detectors mainly rely on changes in physical phenomena such as smoke and temperature to detect fires, but these detectors have slow detection speeds and limited sensitivity. They can often only detect fires when they have already formed a certain scale, which makes it difficult to meet the needs of modern fire safety. Through very early fire detectors, the source of fire can be discovered in time in the early stages of the fire, which buys precious time for fire fighting and evacuation of personnel, and minimizes the loss of life and property caused by fire. At the same time, the research and application of very early fire detectors still face many challenges, such as environmental interference, sensor stability, data accuracy, etc. This requires researchers to continuously explore and innovate, and improve the performance and reliability of very early fire detectors through experiments and practice. At the same time, it is also necessary to strengthen cross-cooperation with other fields, such as architecture and environmental engineering, to achieve comprehensive protection and optimization of fire safety. Summary of the invention

[0004] The present invention provides a point-type very early optical gas composite smoke detection system to solve the problems raised in the above background technology.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] A point-type very early optical gas composite smoke detection system includes a power communication unit, a main control unit, a detection unit and an air pump unit. The detection unit includes a CO sensor, a CO2 sensor, an oxygen sensor, four light intensity sensors, two laser light sources with different wavelengths, and a laser control and light intensity acquisition module, which is used to transmit the collected light intensity signal and various gas concentration data to the main control unit. The power communication unit can realize the functions of power isolation, power conversion and RS485 communication, supply power to other units of the system and provide a communication interface. The air pump unit includes an air inlet nozzle, a filter device, and a mounting base, which is used to extract air from the protected area for detection. The main control unit can be connected to the power communication unit, powered by the power communication unit, and communicate with the power communication unit; connected to the air pump to control the air pump to extract air detection samples; built-in Bluetooth protocol stack, connected to the onboard PCB Bluetooth antenna, used to update the main control unit firmware and wireless wake-up device self-learning function; external sound and light alarm, used for local smoke alarm; connected to the detection unit, through the laser control and light intensity acquisition module to control 2 laser light sources of different wavelengths to emit lasers, and through UART to collect data from 4 light intensity sensors and gas concentration data, finally use the built-in deep learning model to analyze the data to realize the smoke alarm function.

[0007] A further improvement of the technical solution of the present invention is that: the four light intensity sensors are used to collect laser light intensity data in different directions and angles; the two laser light sources with different wavelengths are controlled by the main control unit to emit lasers of different wavelengths; the laser control and light intensity acquisition module controls the emission of the laser light source and the acquisition of light intensity sensor data; the CO sensor, CO2 sensor, and oxygen sensor are used to collect the gas composition and concentration extracted by the pump, and transmit them to the main control unit through the UART interface, wherein the detection range of the CO sensor is 25 to 500 ppm, the concentration range of the CO2 sensor is 0 to 5%, and the concentration range of the oxygen sensor is 0 to 25%.

[0008] A further improvement of the technical solution of the present invention is that: the main control unit has a built-in deep learning model, and the specific implementation process includes the following steps:

[0009] Step 1: Data preparation;

[0010] Step 2: Model building;

[0011] Step 3: Clustering;

[0012] Step 4: Classification.

[0013] A further improvement of the technical solution of the present invention is that: the step 1 includes collecting an algorithm training data set by an experimental method, igniting combustibles, such as wood, under different temperature and humidity conditions, obtaining smoke generated by the wood, and then emitting a laser with a wavelength of 1 to obtain light intensity data Data1{x1, x2, x3, x4}1 captured by four sensors under the scattering effect of the smoke particles, turning off the laser light source with a wavelength of 1, emitting a laser with a wavelength of 2, and obtaining light intensity data Data2{y1, y2, y3, y4}1 of the same combustible under the laser light source with a wavelength of 2, and in the same way, obtaining the light intensity data Data2{y1, y2, y3, y4}1 of the combustion products of sponges, sawdust, cotton thread, circuit boards, plastics, cloth, diesel, kerosene, gasoline, cartons, and toilet paper in turn. The corresponding light intensity data of smoke generated by two laser light sources with different wavelengths are obtained in the experiment. 10,000 groups of data are prepared as training data sets, Data1{x1,x2,x3,x4}1,Data1{x1,x2,x3,x4}……Data1{x1,x2,x3,x4}10000,Data2{y1,y2,y3,y4}1,Data2{y1,y2,y3,y4}……Data2{y1,y2,y3,y4}10000. In this way, the light intensity data generated by water vapor particles and dust particles under the action of laser light sources are obtained. After preprocessing the obtained training data set, it is sent to the artificial intelligence algorithm for training to obtain the prediction model.

[0014] A further improvement of the technical solution of the present invention is that: the step 2 refers to the light intensity classification first being divided by the unsupervised learning K-means clustering algorithm to obtain the number of clusters, and then the supervised learning KNN classification algorithm is used to train the prediction model, and finally the prediction model is used to predict the unknown particles. Since the K-means algorithm and the KNN algorithm are both based on the principle of minimum distance between samples for division, after the K-means algorithm is used for clustering, the KNN classification is used, the effect is better, the prediction accuracy is higher, and the sample separability is stronger.

[0015] The further improvement of the technical solution of the present invention is that: the K-means algorithm in step 3 is a commonly used clustering algorithm, which is characterized in that the more similar the objects in the cluster are, the better the clustering effect is. The input data of the algorithm is used as a sample set, and the samples are clustered by the algorithm so that samples with similar characteristics are clustered into one category. Then, K objects are randomly selected as initial clustering centers, and the data set is divided into K groups. The distance between each object and the cluster center is calculated and assigned to the cluster center closest to it. Each time an object is assigned, the cluster center is recalculated until the cluster center no longer changes. Assuming that the clusters are divided into (C1, C2, ..., C k ), new cluster centers:

[0016]

[0017] Among them, μ j It is cluster C j The mean vector of .

[0018] A further improvement of the technical solution of the present invention is that the KNN classification algorithm in step 4 is a common supervised learning method. The main idea is that if a sample has k nearest neighbors in the feature area, and most of these samples belong to a certain category, then the sample also belongs to this category. This method determines the sample category to be classified according to the category of the nearest one or more samples while considering the classification decision.

[0019] The further improvement of the technical solution of the present invention is that: regarding the distance measurement value, the Euclidean distance is commonly used. The Euclidean distance measures the absolute distance between two points in a multidimensional space, that is, the real distance between two points in an m-dimensional space or the natural length of a vector. 11 ,x 12 ,…,x n ) and b(x 11 ,x 12 ,…,x n ) is:

[0020]

[0021] Due to the adoption of the above technical solution, the present invention has the following technical advances compared with the prior art:

[0022] 1. The present invention provides a point-type very early optical gas composite smoke detection system. The detector integrates multiple gas sensors such as CO sensor, CO2 sensor, oxygen sensor, as well as light intensity sensor and laser light source. It can simultaneously monitor multiple fire-related gas and smoke parameters, thereby improving the comprehensiveness and accuracy of fire detection.

[0023] 2. The present invention provides a point-type very early optical gas composite smoke detection system. Through a light intensity sensor and laser light sources of different wavelengths, the detector can detect tiny smoke changes produced in the smoldering stage of a fire, thereby sounding an alarm in the early stages of a fire, which helps to take timely firefighting measures and prevent the spread of fire.

[0024] 3. The present invention provides a point-type very early optical gas composite smoke detection system. The detector draws air through an air pump and removes dust through a filtering device, thereby reducing the possibility of false alarms and missed alarms. This design enables the detector to maintain high sensitivity and accuracy in complex environments, such as industrial sites, underground garages, etc.

[0025] 4. The present invention provides a point-type very early optical gas composite smoke detection system. The main control unit of the detector has a built-in artificial intelligence prediction model, which can analyze the collected light intensity data and gas concentration data to make more accurate fire judgments. In addition, the equipment supports single-machine self-learning calibration tasks, can adapt to learning different application environments, and improve the accuracy and reliability of the product under various scenario conditions.

[0026] 5. The present invention provides a point-type very early optical gas composite smoke detection system. Through RS485 communication and Bluetooth wireless communication, the user can remotely monitor the status of the detector and perform maintenance and calibration, thereby improving the usability and maintainability of the equipment.

[0027] 6. The present invention provides a point-type very early optical gas composite smoke detection system. Since the detector can monitor a variety of fire-related parameters and conduct comprehensive analysis in combination with an artificial intelligence model, it can more effectively distinguish between real fires and false alarms, reduce the false alarm rate, and at the same time improve the accuracy of fire detection and reduce the possibility of missed alarms.

[0028] 7. The present invention provides a point-type very early optical gas composite smoke detection system. The design of the detector allows customization and expansion according to different application requirements and environmental conditions, such as adding sensor types or adjusting the wavelength of the laser light source, so that it can adapt to more types of fire scenes. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 is a comprehensive schematic block diagram of the present invention;

[0030] Figure 2 It is a schematic block diagram of the detection unit in the present invention. DETAILED DESCRIPTION

[0031] The present invention is further described in detail below in conjunction with embodiments:

[0032] Example 1

[0033] like Figure 1 , Figure 2As shown, the present invention provides a point-type very early optical gas composite smoke detection system, including a power communication unit, a main control unit, a detection unit and an air pump unit. The detection unit includes a CO sensor, a CO2 sensor, an oxygen sensor, four light intensity sensors, two laser light sources with different wavelengths, and a laser control and light intensity acquisition module, which is used to transmit the collected light intensity signal and various gas concentration data to the main control unit. The power communication unit can realize the functions of power isolation, power conversion and RS485 communication, supply power to other units of the system and provide a communication interface. The air pump unit includes an air inlet nozzle, a filter device, and a mounting base, which is used to extract air from the protected area. The main control unit can be connected to the power communication unit, powered by the power communication unit, and communicate with the power communication unit; connected to the air pump to control the air pump to extract air detection samples; built-in Bluetooth protocol stack, connected to the onboard PCB Bluetooth antenna, for updating the main control unit firmware and wireless wake-up device self-learning function; external sound and light alarm for local smoke alarm; connected to the detection unit, through the laser control and light intensity acquisition module to control 2 laser light sources of different wavelengths to emit lasers, and through UART to collect data from 4 light intensity sensors and gas concentration data, finally use the built-in deep learning model to analyze the data to realize the smoke alarm function.

[0034] In this embodiment, it is suitable for various complex environments, such as places with high air flow, large open places, low temperature places, concealed detection places, key department places, explosion-proof places, strong electromagnetic and strong radiation places, etc. Its excellent performance and wide applicability make it an ideal choice in the field of modern intelligent fire-fighting equipment. Through continuous learning and optimization, the detector can achieve extremely early detection and early warning of fire, and maximize the protection of people's lives and property safety.

[0035] Example 2

[0036] like Figure 1 , Figure 2 As shown, on the basis of Example 1, the present invention provides a technical solution: preferably, four light intensity sensors are used to collect laser light intensity data in different directions and angles; two laser light sources with different wavelengths are controlled by the main control unit to emit lasers with different wavelengths; a laser control and light intensity acquisition module controls the emission of the laser light source and the acquisition of light intensity sensor data; a CO sensor, a CO2 sensor, and an oxygen sensor are used to collect the gas composition and concentration extracted by the pump, and transmit them to the main control unit through a UART interface, wherein the detection range of the CO sensor is 25 to 500 ppm, the concentration range of the CO2 sensor is 0 to 5%, and the concentration range of the oxygen sensor is 0 to 25%.

[0037] In this embodiment, in places such as telecommunication rooms and clean rooms where strict control of air quality is required, air flow is usually large, and traditional fire detectors may find it difficult to accurately detect fire signals. The detector of the present invention extracts air samples through an air pump and combines advanced sensor technology to overcome the interference caused by high air flow and achieve early detection of fires. In places with large areas such as warehouses, factories, and exhibition halls, early detection of fires is crucial to reducing losses. The detector of the present invention has a wide monitoring range and high sensitivity, and can achieve comprehensive monitoring of these places, timely detect fire sources, and prevent the spread of fires. In places such as cultural heritage buildings where the original structure and appearance need to be protected, traditional fire detectors may cause damage to the buildings. The detector of the present invention is exquisitely designed and can be concealedly installed without affecting the appearance of the building. At the same time, it provides efficient fire detection and is very useful in bank data centers, power distribution rooms of power departments, and airport control rooms. In key facilities such as tower control, early detection of fire is crucial to ensuring the safety of the facilities. The detector of the present invention can sound an alarm at the initial stage of the fire, provide timely fire warnings for these key departments, and ensure the safe operation of the facilities. In explosion-proof places such as oil depots and chemical facilities, the entry of personnel may bring safety risks. The detector of the present invention can realize unmanned fire detection in these places through remote monitoring and maintenance, thereby improving safety. In strong electromagnetic and strong radiation environments such as microwave rooms, traditional detectors may be interfered with, affecting their detection performance. The detector of the present invention adopts advanced anti-interference design and can maintain stable and accurate fire detection in these complex environments. In addition to the above-mentioned places, the detector of the present invention can also be widely used in other places that require efficient fire detection, such as high-rise buildings, shopping centers, schools, hospitals, etc. Its high sensitivity and intelligent decision-making ability enable it to provide comprehensive fire safety protection for various places.

[0038] Example 3

[0039] like Figure 1 , Figure 2 As shown, based on Example 1, the present invention provides a technical solution: preferably, the main control unit has a built-in deep learning model, and the specific implementation process includes the following steps:

[0040] Step 1: Data preparation;

[0041] Step 2: Model building;

[0042] Step 3: Clustering;

[0043] Step 4: Classification;

[0044] Step 1 includes collecting algorithm training data sets through experiments. Under different temperature and humidity conditions, combustibles, such as wood, are ignited to obtain the smoke produced by the wood. Then, a laser with wavelength 1 is emitted to obtain the light intensity data Data1{x1,x2,x3,x4}1 captured by four sensors under the scattering effect of the smoke particles. Then, the laser light source with wavelength 1 is turned off and a laser with wavelength 2 is emitted to obtain the light intensity data Data2{y1,y2,y3,y4}1 of the same combustible under the laser light source with wavelength 2. In the same way, sponge, sawdust, cotton thread, circuit board, plastic, cloth, and diesel are obtained in turn. , kerosene, gasoline, cardboard, toilet paper combustion of combustibles under the action of two different wavelength laser light sources corresponding to the light intensity data, the experiment prepared 10000 sets of data as the training data set, Data1{x1,x2,x3,x4}1,Data1{x1,x2,x3,x4}……Data1{x1,x2,x3,x4}10000,Data2{y1,y2,y3,y4}1,Data2{y1,y2,y3,y4}……Data2{y1,y2,y3,y4}10000. So as to obtain water vapor particles and dust particles The light intensity data generated by the interference under the action of the laser light source is preprocessed and sent to the artificial intelligence algorithm for training to obtain the prediction model. The second step refers to the light intensity classification. First, it is divided by the unsupervised learning K-means clustering algorithm to obtain the number of clusters, and then the supervised learning KNN classification algorithm is used to train the prediction model. Finally, the prediction model is used to predict the unknown particles. Since both the K-means algorithm and the KNN algorithm are divided based on the principle of minimum distance between samples, the K-means algorithm is used for clustering, and then the KNN classification is used, the effect is better and the prediction is better. The accuracy is also higher, and the sample is more separable. The K-means algorithm in step 3 is a commonly used clustering algorithm. Its characteristic is that the more similar the objects in the cluster are, the better the clustering effect is. The input data of the algorithm is used as a sample set. The samples are clustered through the algorithm so that samples with similar characteristics are clustered into one category. Then K objects are randomly selected as the initial clustering centers, and the data set is divided into K groups. The distance between each object and the clustering center is calculated and assigned to the clustering center closest to it. Each time an object is assigned, the clustering center is recalculated until the clustering center no longer changes. Assume that the clusters are divided into (C1, C2, ..., C k ), new cluster centers:

[0045]

[0046] Among them, μ j It is cluster C j The mean vector of .

[0047] Example 4

[0048] like Figure 1 , Figure 2 As shown, on the basis of Example 1, the present invention provides a technical solution: Preferably, the KNN classification algorithm in step 4 is a common supervised learning method. The main idea is that if a sample has k nearest neighbors in the feature area, and most of these samples belong to a certain category, then the sample also belongs to this category. This method determines the sample category to be classified according to the category of the nearest one or more samples while considering the classification decision. Regarding the distance measurement value, the Euclidean distance is commonly used. The Euclidean distance measures the absolute distance between two points in a multidimensional space, that is, the real distance between two points in an m-dimensional space or the natural length of a vector. The n-dimensional space point a(x 11 ,x 12 ,…,x n ) and b(x 11 ,x 12 ,…,x n ) is:

[0049]

[0050] In this embodiment, in order to select the most suitable k value for the KNN model, this paper adopts K-fold cross validation to determine the size of k, uses cross-split to obtain different training sets and validation sets to test the accuracy of the model, and the mean of the accuracy is the result of cross validation. The hyperparameter with the highest accuracy is selected as the k value of the KNN model. Using the two training data sets obtained in the experiment, two prediction models are obtained through the artificial intelligence algorithm. The two prediction models are used to test unknown particles. The predicted results are consistent with the actual results, and the smoke particles and non-smoke particles are accurately distinguished. The algorithm ensures that the fire is discovered in the very early smoldering stage of the fire, and the alarm is issued in time to maximize the protection of human life and property.

[0051] The following is a detailed description of the working principle of this point-type very early optical gas composite smoke detection system.

[0052] like Figure 1 , Figure 2As shown in the figure, through the comprehensive application of optics, gas sensors and deep learning algorithms, the ability to detect and warn of fire at an extremely early stage is achieved. During the installation process, users can send self-learning commands to the detector through the mobile phone Bluetooth or the RS485 communication device of the remote monitoring center. The detector will process the air detection samples extracted from the protected area, irradiate the air samples with laser light sources of different wavelengths, and analyze the scattered light intensity data of smoke particles and the gas component concentration data collected by the gas sensor. The deep learning prediction model built into the main control unit is used for intelligent analysis and decision-making to determine whether the smoke alarm needs to be activated. In addition, the detector also has the ability of continuous self-learning and optimization. Through regular self-learning, the detector can continuously adapt to the data of the current detection environment, improve detection accuracy, and reduce the occurrence of false alarms. Users can also remotely monitor the status of the detector through Bluetooth or RS485 communication equipment, and perform maintenance and calibration, which greatly improves the ease of use and maintainability of the detector.

[0053] The above generally describes the present invention in detail, but it is obvious to a person skilled in the art that some modifications or improvements can be made to the present invention. Therefore, modifications or improvements that do not depart from the spirit of the present invention are within the scope of protection of the present invention.

Claims

1. A point-type very early optical gas composite smoke detection system, characterized by: It includes a power communication unit, a main control unit, a detection unit and an air pump unit. The detection unit includes a CO sensor, a CO2 sensor, an oxygen sensor, 4 light intensity sensors, 2 laser light sources with different wavelengths, and a laser control and light intensity acquisition module, which is used to transmit the collected light intensity signal and various gas concentration data to the main control unit. The power communication unit can realize the functions of power isolation, power conversion and RS485 communication, supply power to other units of the system and provide a communication interface. The air pump unit includes an air inlet nozzle, a filter device, and a mounting base, which are used to extract air detection samples from the protected area. The main control unit can It is connected to the power communication unit, powered by the power communication unit, and communicates with the power communication unit; it is connected to the air pump to control the air pump to extract air detection samples; it has a built-in Bluetooth protocol stack and is connected to the onboard PCB Bluetooth antenna to update the main control unit firmware and the wireless wake-up device self-learning function; it is connected to an external sound and light alarm for local smoke alarm; it is connected to the detection unit, controls two laser light sources of different wavelengths to emit lasers through the laser control and light intensity acquisition module, and collects data from four light intensity sensors and gas concentration data through UART. Finally, it uses the built-in deep learning model to analyze the data to realize the smoke alarm function.

2. A point-type very early optical gas composite smoke detection system according to claim 1, characterized in that: The four light intensity sensors are used to collect laser light intensity data in different directions and angles; the two laser light sources with different wavelengths are controlled by the main control unit to emit lasers of different wavelengths; the laser control and light intensity acquisition module controls the emission of the laser light source and the acquisition of light intensity sensor data; the CO sensor, CO2 sensor, and oxygen sensor are used to collect the gas composition and concentration extracted by the pump, and transmit them to the main control unit through the UART interface, wherein the detection range of the CO sensor is 25-500ppm, the concentration range of the CO2 sensor is 0-5%, and the concentration range of the oxygen sensor is 0-25%.

3. The point-type very early optical gas composite smoke detection system according to claim 1, characterized in that: The main control unit has a built-in deep learning model, and the specific implementation process includes the following steps: Step 1: Data preparation; Step 2: Model building; Step 3: Clustering; Step 4: Classification.

4. A point-type very early optical gas composite smoke detection system according to claim 3, characterized in that: The step 1 includes collecting an algorithm training data set by an experimental method, igniting combustibles, such as wood, under different temperature and humidity conditions to obtain smoke generated by the wood, then emitting a laser with a wavelength of 1, obtaining light intensity data Data1{x1,x2,x3,x4}1 captured by four sensors under the scattering effect of the smoke particles, turning off the laser light source with wavelength 1, emitting a laser with a wavelength of 2, and obtaining light intensity data Data2{y1,y2,y3,y4}1 of the same combustible under the laser light source with wavelength 2, and in the same way, obtaining the smoke generated by the combustion of sponge, sawdust, cotton thread, circuit board, plastic, cloth, diesel, kerosene, gasoline, cardboard, and toilet paper under two different conditions. The corresponding light intensity data under the action of the same wavelength laser light source, the experiment prepares 10000 groups of data as the training data set, Data1{x1,x2,x3,x4}1,Data1{x1,x2,x3,x4}……Data1{x1,x2,x3,x4}10000,Data2{y1,y2,y3,y4}1,Data2{y1,y2,y3,y4}……Data2{y1,y2,y3,y4}10000. In this way, the light intensity data generated by water vapor particles and dust particles under the action of laser light source are obtained. After preprocessing the obtained training data set, it is sent to the artificial intelligence algorithm for training to obtain the prediction model.

5. The point-type very early optical gas composite smoke detection system according to claim 3 is characterized by: The step 2 refers to that the light intensity classification is first divided by the unsupervised learning K-means clustering algorithm to obtain the number of clusters, and then the supervised learning KNN classification algorithm is used to train the prediction model, and finally the prediction model is used to predict the unknown particles. Since the K-means algorithm and the KNN algorithm are both based on the principle of minimum distance between samples for division, the effect of using the K-means algorithm for clustering and then using the KNN classification is better, the prediction accuracy is higher, and the sample separability is stronger.

6. The point-type very early optical gas composite smoke detection system according to claim 3, characterized in that: The K-means algorithm in step 3 is a commonly used clustering algorithm. Its characteristic is that the more similar the objects in the cluster are, the better the clustering effect is. The input data of the algorithm is used as a sample set. The samples are clustered by the algorithm so that samples with similar characteristics are clustered into one category. Then K objects are randomly selected as the initial clustering centers, and the data set is divided into K groups. The distance between each object and the cluster center is calculated and assigned to the cluster center closest to it. Each time an object is assigned, the cluster center is recalculated until the cluster center does not change. Assume that the clusters are divided into (C1, C2, ..., C k ), new cluster centers: Among them, μ j It is cluster C j The mean vector of .

7. The point-type very early optical gas composite smoke detection system according to claim 3 is characterized by: The KNN classification algorithm in step 4 is a common supervised learning method. The main idea is that if a sample has k nearest neighbors in the feature area, and most of these samples belong to a certain category, then the sample also belongs to this category. This method determines the sample category to be classified based on the category of the nearest one or more samples while considering the classification decision.

8. The point-type very early optical gas composite smoke detection system according to claim 1, characterized in that: Regarding distance measurement, Euclidean distance is often used. Euclidean distance measures the absolute distance between two points in multidimensional space, that is, the real distance between two points in m-dimensional space or the natural length of a vector. 11 ,x 12 ,…,x n ) and b(x 11 ,x 12 ,…,x n ) is:

Images