Intelligent Fire Alarm Safety Control Method and Control System for Rail Transit

Abstract

The present invention relates to the technical field of cables, and particularly to an intelligent fire alarm safety control method and control system for rail transit. The intelligent fire alarm safety control method for rail transit is disclosed, including the following steps: setting a plurality of groups of detection components of different types for synchronous monitoring, taking the detection component that first detects the presence of a fire point in the space as the initial detection, and arranging the remaining detection components in a random order to perform follow-up detection on the fire point to determine the fire point, so as to avoid detection errors caused by dust coverage, and performing corresponding fire extinguishing operations on the fire point according to the detection results of step S2. The present invention performs detection through the set step-by-step follow-up detection method, so that each detection is based on the detection results obtained in the previous detection, realizing a step-by-step multi-angle verification method, which can ensure the accuracy of the flame detection results.

Description

Technical Field

[0001] The present invention relates to the field of rail transit safety, and particularly to an intelligent fire alarm safety control method and control system for rail transit. Background Art

[0002] Due to the long-term running of rail vehicles, a large amount of dust infiltrates into the vehicle body. The electrical cabinet needs to ensure long-term operation, and due to the heat dissipation requirement, it is necessary to ensure the circulation of internal and external airflows. This leads to the following problems in the electrical cabinet or the interior of the carriage:

[0003] 1. Due to the circulation of internal and external airflows, dust is extremely likely to enter the interior of the electrical cabinet along with the airflow. This situation will cause a large amount of dust to adhere to the outer surface of the flame detector, resulting in a decrease in the detection sensitivity of the flame detector, and further leading to false alarms.

[0004] 2. Since the existing fire extinguishing systems usually adopt the cooperation between the power bottle group and the fire extinguishing agent bottle group, when the flame detector gives a false alarm of fire, the power bottle group will pressurize the interior of the fire extinguishing agent bottle group. Due to the extremely fast pressurization speed and the irreversible pressurization process of the power bottle group, even if the system response is a false alarm, it is impossible to prevent the fire extinguishing agent bottle group from spraying the mist. Summary of the Invention

[0005] In view of the deficiencies of the existing technology, the present invention provides the following technical solutions:

[0006] An intelligent fire alarm safety control method for rail transit, comprising the following steps:

[0007] S1: Preset: Set several groups of different types of detection components for synchronous monitoring.

[0008] S2: Initial detection: Take the detection component that first detects the presence of a fire point in the space as the initial detection, and use the remaining detection components to perform follow-up detection on the fire point in a random order to determine the fire point, so as to avoid detection errors caused by dust coverage.

[0009] S3: Fire extinguishing: Perform corresponding fire extinguishing operations on the fire point according to the detection results of step S2.

[0010] As an improvement of the above technical solution, the step S1 includes the following steps:

[0011] S11: Take thermal images and panoramic images of the space through an imaging system, locate the high-temperature positions on the thermal images and the highlighted positions on the panoramic images, and mark the high-temperature and / or highlighted positions as fire points.

[0012] S12: Identify and locate the positions of fire points in the thermal images and / or panoramic images through a convolutional neural network.

[0013] S13: Detect and sense the space through a flame detection system, and locate the position of the ignition point.

[0014] As an improvement of the above technical solution, the step S2 includes the following steps:

[0015] S21: When it is determined in any one of the steps S11, S12 or S13 that there is an ignition point, randomly arrange the remaining detection components in sequence.

[0016] S22: Perform image recognition or detection in sequence on the basis of the previous step of detection according to the arranged order to determine the ignition point, so as to avoid detection errors caused by dust coverage.

[0017] As an improvement of the above technical solution, when the detection results of the step S2 are consistent, it is determined that the detection is accurate and the step S3 is executed; if the detection results are inconsistent, the following steps are executed:

[0018] S201: Randomly arrange several groups of detection components and perform several detections on the position determined as the ignition point in the step S2 to obtain several groups of detection results.

[0019] S202: If all the detection results are consistent, it is determined that the detection results in the step S2 are accurate, and report a fault for the detection components with inconsistent detection results in the step S2 and wait for maintenance; if all the detection results are inconsistent, it is determined that the detection components with inconsistent detection results in the step S2 are faulty, and perform an emergency inspection.

[0020] As an improvement of the above technical solution, the emergency inspection method includes the following steps:

[0021] S203: Mark the position of the ignition point with inconsistent detection results and alarm at this position.

[0022] S204: Send the panoramic image of the ignition point area with inconsistent detection results to the host computer, and manually determine whether there is a fire.

[0023] As an improvement of the above technical solution, the training method of the convolutional neural network depends on the following steps:

[0024] S121: Periodically capture thermal images and panoramic images through an imaging system.

[0025] S122: Separate the areas without ignition points in the thermal image and the panoramic image, and send this image area into the convolutional neural network for non-ignition point recognition training.

[0026] S123: Send the data of the planar image and the thermal image storing the flame into the convolutional neural network for flame recognition training.

[0027] The intelligent fire alarm safety control system for rail transit is controlled by using the intelligent fire alarm safety control method described in any one of the foregoing technical solutions, and includes: several groups of detection components and a fire extinguishing system.

[0028] Specifically, several groups of detection components are used to detect the ignition point in the space by different methods, and the fire extinguishing system is used to perform corresponding fire extinguishing operations following the detection results of the detection components.

[0029] As an improvement of the above technical solution, the detection components include: an imaging system, a convolutional neural network, and a flame detection system.

[0030] Specifically, the imaging system is used to capture the thermal image and panoramic image in the space, and locate the high-temperature position on the thermal image and the highlighted position on the panoramic image. The convolutional neural network is used to identify and locate the ignition point position on the thermal image and panoramic image. The flame detection system is used to perform inductive detection on the space and locate the position of the ignition point.

[0031] As an improvement of the above technical solution, the fire extinguishing system includes a starting bottle group, a fire extinguishing agent bottle group, and a buffer system provided on the connecting pipeline between the starting bottle group and the fire extinguishing agent bottle group.

[0032] As an improvement of the above technical solution, the buffer system includes a buffer tank, two groups of connecting pipes, and two groups of adjusting rods. Each group of adjusting rods has several. The two groups of connecting pipes respectively penetrate into the interior of the buffer tank from both ends of the buffer tank. The end of the connecting pipe located inside the buffer tank is provided with an inner cylinder that fits the inner wall of the buffer tank. The end of the connecting pipe located outside the buffer tank is provided with a flange. The adjusting rods penetrate through the buffer tank and the inner cylinder. A synchronous belt is synchronously connected between several adjusting rods on the same side. At least one end of the adjusting rod is connected to a motor, and the motor is fixed on the inner wall of the inner cylinder. Several bearing sleeves are provided inside the inner cylinder, and the adjusting rod is fixed to the inner ring of the bearing sleeve.

[0033] The beneficial effects of the present invention:

[0034] By setting a step-by-step following detection method for detection, it is ensured that each detection is based on the previous detection result, realizing a step-by-step multi-angle verification method. This method can still ensure the accuracy of the flame detection result when some structures have false detection judgments due to failures (such as dust covering causing some structures to be unable to detect accurately).

[0035] Furthermore, a buffer system is provided. When false alarms occur, the buffer system expands in time to absorb part of the air pressure, so that the air pressure sent by the starting cylinder group cannot open the container valve of the fire extinguishing agent cylinder group. In this way, emergency operations can be provided when false alarms occur, preventing the fire extinguishing agent cylinder group from being wrongly opened and further reducing the possibility of incorrect operations. Description of the Drawings

[0036] Figure 1 It is a three-dimensional structure diagram of the buffer system of the present invention;

[0037] Figure 2 It is a front view of the buffer system of the present invention;

[0038] Figure 3 is Figure 2 The sectional structure diagram at A-A in

[0039] Reference numerals in the drawings: 10, buffer tank; 20, connecting pipe; 21, flange; 22, inner cylinder; 23, bearing sleeve; 30, adjusting rod; 31, synchronous belt; 32, motor. Detailed Embodiments

[0040] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0041] Due to the circulation of internal and external airflows, dust is very likely to enter the electrical cabinet along with the airflows. This situation will cause more dust to adhere to the outer surface of the flame detector, resulting in a decrease in the detection sensitivity of the flame detector and further leading to false alarms.

[0042] Embodiment 1

[0043] To solve the foregoing problems, a method for intelligent fire alarm safety control in rail transit is provided, including the following steps:

[0044] S1: Presetting: Setting several groups of different types of detection components for synchronous monitoring.

[0045] By setting several groups of different types of detection components, the monitoring accuracy is enhanced. Different from the existing simple superposition, a mutually following monitoring method is adopted. Specifically, step S1 includes the following steps:

[0046] S11: Taking thermal images and panoramic images in the space through an imaging system, positioning the high-temperature positions on the thermal images and the highlighted positions on the panoramic images, and marking the high-temperature and / or highlighted positions as ignition points.

[0047] Since the most obvious features of a flame are its color, the heat of the infrared reaction, and the degree of radiation of the ultraviolet reaction.

[0048] There are errors in the flame colors and color differences produced by the combustion caused by different ignition sources. Therefore, color and color difference alone cannot guarantee the detection accuracy. On this basis, the thermal images of thermal imaging are combined to reflect the heat radiation of infrared rays. The failure rate of the imaging system is relatively low. Usually, as long as it can take normal pictures, the accuracy of the pictures can be guaranteed. Therefore, the high-temperature and / or high-brightness positions are marked as ignition points.

[0049] Here, the expression in the form of "and / or" is used to avoid missing suspicious ignition points. Substances catch fire because the temperature is higher than the ignition point of the substance, and the ignition points of different substances are different. Therefore, there may be some substances with relatively low ignition points and not high temperatures. In this case, color difference needs to be used for marking.

[0050] Among them, "high brightness" does not refer to high light intensity, but to the degree of color difference between the color and the surrounding area. Since the purpose of this embodiment is to detect when the flame just appears, therefore, except for the cases of explosion or extremely rapid spontaneous combustion, the fire spreads in a progressive form. In this case, the initial fire is actually not high. And because the color of the flame is inconsistent with the color of the surrounding items, this forms a certain color difference, and the degree of this color difference represents the conspicuousness of the flame color relative to the surrounding environment. Therefore, only by marking the high-brightness positions in the environment can all suspicious ignition points be found.

[0051] For this reason, the positions that are both in a high-temperature state and a high-brightness state can basically be determined as ignition points. And in order to avoid errors in this embodiment, the high-brightness, high-temperature, and positions that are both in a high-temperature state and a high-brightness state are all marked as ignition points.

[0052] S12: Identify and locate the ignition point positions in the thermal image and / or panoramic image through a convolutional neural network.

[0053] In addition to the method in step S11, this embodiment also discloses a method for identifying ignition points in a thermal image or a panoramic image through a convolutional neural network. This method is realized through a large number of image trainings. Specifically, the training method of the convolutional neural network depends on the following steps:

[0054] S121: Periodically capture thermal images and panoramic images through the imaging system.

[0055] Capturing images through the imaging system here is to make the convolutional neural network adapt to the environment to be monitored, that is, to enable the convolutional neural network to accurately identify the positions where abnormal states occur in the monitored environment.

[0056] S122: Separate the area without ignition points in the thermal image and the panoramic image, and send this image area into the convolutional neural network for the recognition training of areas without ignition points.

[0057] When the convolutional neural network has a certain understanding of the monitoring environment, the normal areas in the thermal image can be separated and input into the convolutional neural network for targeted recognition training.

[0058] S123: Send the data of the planar image storing the flame and the thermal image into the convolutional neural network for the recognition training of the flame.

[0059] In addition to the above peeling of the image area, it is also necessary to enable the convolutional neural network to accurately recognize the flame. Here, a database is usually used for training, and by recognizing a large number of data with different flame forms, it is determined whether there is a flame in the separated images of the monitoring area.

[0060] S13: Inductively detect the space through the flame detection system and locate the position of the ignition point

[0061] The flame detection system is a conventional device such as a sensor, such as a smoke and fire detection sensor, an ultraviolet detection sensor, etc. This part is a direct detection device.

[0062] After determining the three detection methods proposed in this embodiment, execute step S2

[0063] S2: Initial detection: Use the detection component that first detects the existence of an ignition point in the space as the initial detection, and use the remaining detection components to perform follow-up detection on the ignition point in a random order to determine the ignition point, so as to avoid detection errors caused by dust coverage.

[0064] Specifically, step S2 includes the following steps:

[0065] S21: When any one of steps S11, S12 or S13 determines the existence of an ignition point, randomly arrange the remaining detection components in sequence.

[0066] Due to differences or variations in factors such as detection location, detection accuracy, detection reaction speed, and detection form, the time for each method given in this embodiment to detect the ignition point is uncertain. Therefore, a guiding method is adopted for detection, that is, as long as any one method detects an ignition point, the remaining methods will follow this method to detect this ignition point, and specifically execute step S22.

[0067] S22: Sequentially perform image recognition or detection on the basis of the previous step of detection according to the arranged order to determine the ignition point, so as to avoid detection errors caused by dust coverage.

[0068] The following detection here is not a simple direct detection, but rather a detection based on the previous step, as exemplified below:

[0069] When the imaging system captures a thermal image and a panoramic image, after imaging, the imaging system locates abnormal situations in the thermal image and the panoramic image. For example, there are high-temperature positions, high-brightness positions, or both superimposed on the image. Among them, the determination of high temperature and high brightness is to preset a threshold in advance. When the data exceeds this threshold, it is determined as high temperature or high brightness.

[0070] When the above situation occurs, the imaging system determines that there is a suspicious ignition point in the monitored area. At this time, the imaging system is the component that first discovers the ignition point. In this embodiment, there are also two methods, namely the flame detection system and the convolutional neural network. Therefore, in the subsequent detection, the two are randomly arranged in order. For example, the obtained random order is to first use the convolutional neural network for detection.

[0071] When it is determined that the convolutional neural network will perform detection immediately afterwards, the imaging system will separate the images around the marked suspicious ignition point to form several suspicious area images. These suspicious area images will be directly sent into the convolutional neural network for recognition and detection, and the convolutional neural network will be used to determine whether it is an ignition point. Here, the convolutional neural network will mark the position determined as the ignition point in space.

[0072] Here, the separation of the thermal image is usually the area with high-temperature anomalies, while for the panoramic image, it is the separation of the high-brightness area. The panoramic image is not applicable to the carriage, but is applicable to parts such as electrical cabinets where the long-term environment does not change. Due to the complexity of people in the carriage and the light reflection caused by the items carried by people, etc., it will affect the recognition of the high-brightness area. The recognition of the high-brightness area of the panoramic image can only be used as an auxiliary, and cannot make an accurate detection judgment.

[0073] After the convolutional neural network finishes its determination, the position of the ignition point is immediately passed to the flame detection system. The flame detection system will call the sensor closest to the position of the ignition point for detection, and finally determine whether the ignition point is accurate. If it is accurate, it will control the fire extinguishing system to extinguish the fire.

[0074] Correspondingly, when the imaging system captures an image, it will also directly pass the image to the convolutional neural network. If the convolutional neural network determines that there is an ignition point in the image before the imaging system, that is, if step S12 first determines that there is an ignition point in the image, then the subsequent steps S11 and S13 will also be random, and will be executed in sequence.

[0075] Although these three make data judgments based on the previous ones, there will still be certain error problems. For example, dust on the surface of the imaging system causes inaccurate shooting. For instance, thermal imaging is blocked by dust, resulting in an incomplete thermal image, or the flame detection system is covered by dust, leading to a low reception efficiency of light or smoke. These situations will all cause errors in the structure.

[0076] When the detection results in step S2 are consistent, it is determined that the detection is accurate and step S3 is executed. If the detection results are inconsistent, the following steps are executed:

[0077] S201: Randomly arrange a number of detection components and perform several detections on the positions determined as the ignition points in step S2 to obtain several groups of detection results.

[0078] The random arrangement and detection method here are the same as those in the aforementioned step S2, but the difference is that the first detection is determined randomly, rather than determined by the first detected step. This method is to ensure that several detections can cover all sequential arrangements.

[0079] Using different methods as the first detection is to reduce the error caused by subsequent all detections being based on faulty data due to a malfunction of the first detection device.

[0080] After several detections are completed, several situations may occur, as shown in step S202 specifically.

[0081] S202: If all detection results are consistent, it is determined that the detection results in step S2 are accurate, and a fault is reported for the detection components with inconsistent detection results in step S2, waiting for maintenance.

[0082] All detection results being consistent here means that there is a malfunction or error in the inconsistent detection step in step S2, resulting in an incorrect result. In this case, it can usually be solved by maintenance and debugging.

[0083] If all detection results are inconsistent, it is determined that the detection components with inconsistent detection results in step S2 are faulty, and an emergency inspection is executed.

[0084] That is, after the detection in step S201, when there are still inconsistent detection results, it means that there are still errors even after reducing the first device failure. This indicates that there are faulty devices in the detection components. In this case, it cannot be simply solved by debugging. Replacement or repair is required, and an emergency inspection also needs to be executed to avoid impacts.

[0085] The situations that can lead to emergency inspections occur with an extremely low probability. Since triple detection and several random cyclic detections can basically ensure inspections in various accident situations, once an emergency inspection is required, the machine usually cannot detect it in time.

[0086] In one embodiment, the emergency inspection method includes the following steps:

[0087] S203: Mark the location of the ignition point where the detection results are inconsistent and alarm at this location.

[0088] Here, it is necessary to first alarm at the location where the detection results are inconsistent, indicating that there is a problem with the entire fire detection system and it needs to be repaired. On this basis, step S204 is further executed.

[0089] S204: Send the panoramic image of the ignition point area where the detection results are inconsistent to the host computer, and manually determine whether it is a fire.

[0090] Here, usually only the panoramic image is sent. The fastest way for manual review is to review through the panoramic image, which can intuitively see whether the marked position is a flame. Only the position where the detection results are inconsistent needs to be sent. Whether it is determined manually that fire extinguishing is required or the detection result indicates that fire extinguishing is required, step S3 is executed.

[0091] S3: Extinguish the fire: Perform corresponding fire extinguishing operations on the ignition point according to the detection results of step S2.

[0092] That is, extinguish the fire at the location of the ignition point determined by the detection.

[0093] Embodiment 2

[0094] To cooperate with Embodiment 1, an intelligent fire alarm safety control system for rail transit is also provided, which is controlled by using the intelligent fire alarm safety control method of any one of Embodiments 1, including: several groups of detection components, a fire extinguishing system.

[0095] Several groups of detection components are used to detect the ignition points in the space by different methods, and the fire extinguishing system is used to perform corresponding fire extinguishing operations following the detection results of the detection components.

[0096] That is, the space is monitored by the detection components. When an ignition point appears in the space, the fire extinguishing system extinguishes the fire at the location of the ignition point.

[0097] The detection components include: an imaging system, a convolutional neural network, a flame detection system;

[0098] The imaging system is used to capture thermal images and panoramic images in space, and locate the high-temperature positions on the thermal images and the highlighted positions on the panoramic images. The convolutional neural network is used to identify and locate the ignition points on the thermal images and panoramic images. The flame detection system is used to perform induction detection in the space and locate the positions of the ignition points.

[0099] Embodiment III

[0100] Since the fire extinguishing system in the existing design usually adopts the cooperation between the power bottle group and the fire extinguishing agent bottle group, when the flame detector gives a false alarm of fire in this design, the power bottle group will pressurize the inside of the fire extinguishing agent bottle group. Since the pressurization speed is extremely fast and the pressurization process of the power bottle group is irreversible, even if the system reaction is a false alarm, it is impossible to prevent the fire extinguishing agent bottle group from spraying the spray.

[0101] To solve this problem, the difference from Embodiment II is that it includes: the fire extinguishing system includes a starting bottle group, a fire extinguishing agent bottle group, and a buffer system arranged on the connecting pipeline between the starting bottle group and the fire extinguishing agent bottle group.

[0102] Whether it is Embodiment I or Embodiment II, a certain detection time is required during execution, and this time is particularly obvious when detection errors occur and cyclic detection is needed. This leads to the inability to prevent the fire extinguishing agent bottle group from spraying the spray when the reaction in Embodiment I and Embodiment II is a false alarm. Based on this, a buffer system is arranged between the starting bottle group and the fire extinguishing agent bottle group for buffering.

[0103] Specifically, as Figures 1-3 shown, the buffer system includes a buffer tank 10, two groups of connecting pipes 20, and two groups of adjusting rods 30. Each group of adjusting rods 30 has several. The two groups of connecting pipes 20 respectively penetrate into the inside of the buffer tank 10 from both ends of the buffer tank 10. One end of the connecting pipe 20 located inside the buffer tank 10 is provided with an inner cylinder 22 that fits the inner wall of the buffer tank 10. One end of the connecting pipe 20 located outside the buffer tank 10 is provided with a flange 21. The adjusting rods 30 penetrate through the buffer tank 10 and the inner cylinder 22. A synchronous belt 31 is synchronously connected between several adjusting rods 30 on the same side. At least one end of the adjusting rod 30 is connected to a motor 32, and the motor 32 is fixed on the inner wall of the inner cylinder 22. Several bearing sleeves 23 are arranged inside the inner cylinder 22, and the adjusting rods 30 are fixed to the inner rings of the bearing sleeves 23.

[0104] By driving the adjusting rods 30 and the buffer tank 10 to perform threaded rotation in the way of the rotation of the motor 32, this way can control the movement of the inner cylinder 22 inside the buffer tank 10, so as to freely control the distance between the inner cylinders 22. The flange is used to connect the pipelines of the starting bottle group and the fire extinguishing agent bottle group.

[0105] Under normal conditions, as Figure 3As shown, the inner cylinders 22 on both sides are in a separated state. When high-pressure gas enters the interior, it expands to form a speed reduction. When more fire extinguishing agent bottle groups are opened through the activation bottle group, the initial air pressure generated by the activation bottle group is stronger. In this case, the distance between the two inner cylinders 22 is larger, and the cavity formed by separating the interior of the buffer tank 10 from between the two inner cylinders 22 is larger. This is an adjustment following the air pressure. When the entire system operation ends and it is determined that there is no false alarm, the distance between the inner cylinders 22 shrinks until it returns to the normal operation state.

[0106] During the process of detecting the ignition point in the technical solution of Embodiment 1 or Embodiment 2, first determine the number of fire extinguishing agent bottle groups that need to be activated in the case of the most ignition points, and determine the distance between the two inner cylinders 22 according to this number. After the detection result comes out, automatically control the two inner cylinders 22 to approach according to the detection result. For example, if some ignition points are marked as non-ignition points, at this time, it is necessary to control the two inner cylinders 22 to approach, so as to ensure that the air pressure intensity is sufficient to open the container valves of the remaining ignition points.

[0107] If all the detected ignition points are correct ignition points, then the two inner cylinders 22 are in contact. At this time, there is still a cavity for buffering between the two inner cylinders 22. This part of the cavity can reduce the air flow speed while ensuring that the air flow can still open the container valves of the fire extinguishing agent bottle groups. This is to provide a reaction time for the entire fire extinguishing system, and to extend the time for the high-pressure gas to reach the container valve by reducing the gas flow rate. This time is short, but it is sufficient for the system to react.

[0108] Because a certain buffer time is provided, when the state of ignition is judged, the high-pressure air flow flows normally. If the system detects the above result as a misjudgment during the startup process, it only needs to control the two inner cylinders 22 to move away from each other. In this way, the volume of the internal buffer cavity increases, generating a pressure relief effect, which can cause the air flow to be unable to open the container valves of the fire extinguishing agent bottle groups, realizing the blocking in the case of false alarms. In this case, usually an air extraction device can be used to recover the air flow that has been sent in, or control the two inner cylinders 22 to approach to increase the internal pressure until it returns to the normal state pressure (usually an air pressure sensor is also set to detect the internal air pressure), so as to ensure that when the system judges the appearance of an ignition point next time, the high-pressure gas given by the activation bottle group can normally open the container valve.

[0109] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it.

Claims

1. An intelligent fire alarm safety control method for rail transit, characterized in that, It includes the following steps: S1: Preset: Set several groups of different types of detection components for synchronous monitoring; S2: Initial detection: Take the detection component that first detects the presence of a fire point in the space as the initial detection, and use the remaining detection components to follow and detect the fire point in a random order to determine the fire point, so as to avoid detection errors caused by dust coverage; S3: Fire extinguishing: Perform corresponding fire extinguishing operations on the fire point according to the detection results of step S2.

2. The intelligent fire alarm safety control method for rail transit according to claim 1, characterized in that: The step S1 includes the following steps: S11: Take thermal images and panoramic images of the space through the imaging system, locate the high-temperature positions on the thermal images and the highlighted positions on the panoramic images, and mark the high-temperature and / or highlighted positions as fire points; S12: Identify and locate the positions of fire points in the thermal images and / or panoramic images through a convolutional neural network; S13: Perform inductive detection on the space through the flame detection system and locate the position of the fire point.

3. The intelligent fire alarm safety control method for rail transit according to claim 2, characterized in that: The step S2 includes the following steps: S21: When any one of steps S11, S12 or S13 determines the existence of a fire point, randomly arrange the remaining detection components in order; S22: Determine the fire point by performing image recognition or detection in sequence on the basis of the previous step of detection, so as to avoid detection errors caused by dust coverage.

4. The intelligent fire alarm safety control method for rail transit according to claim 2, wherein: When the detection results of step S2 are consistent, it is determined that the detection is accurate and step S3 is executed. If the detection results are inconsistent, the following steps are executed: S201: Randomly arrange several groups of detection components and perform several detections on the positions determined as fire points in step S2 to obtain several groups of detection results; S202: If all the detection results are consistent, it is determined that the detection results in step S2 are accurate, and report a fault for the detection components with inconsistent detection results in step S2 and wait for maintenance; If all the detection results are inconsistent, it is determined that the detection components with inconsistent detection results in step S2 are faulty, and perform an emergency inspection.

5. The intelligent fire alarm safety control method for rail transit according to claim 4, wherein: The method of the emergency inspection includes the following steps: S203: Mark the positions of the fire points with inconsistent detection results and alarm at this position; S204: Send the panoramic images of the areas of the fire points with inconsistent detection results to the host computer, and manually determine whether there is a fire.

6. The intelligent fire alarm safety control method for rail transit according to any one of claims 2-5, characterized in that: The training method of the convolutional neural network depends on the following steps: S121: Periodically take thermal images and panoramic images through the imaging system; S122: Separate the areas without fire points in the thermal images and panoramic images, and send these areas into the convolutional neural network for non-fire point recognition training; S123: Send the data of the planar images with flames and the thermal images into the convolutional neural network for flame recognition training.

7. The intelligent fire alarm safety control system for rail transit is controlled by using the intelligent fire alarm safety control method for rail transit according to any one of claims 1-6, characterized in that, It includes: Several groups of detection components, used to detect the fire points in the space by different methods; A fire extinguishing system, used to perform corresponding fire extinguishing operations following the detection results of the detection components.

8. The intelligent fire alarm safety control system for rail transit according to claim 7, wherein: The detection components include: an imaging system, a convolutional neural network, and a flame detection system; The imaging system is used to take thermal images and panoramic images of the space, and locate the high-temperature positions on the thermal images and the highlighted positions on the panoramic images; The convolutional neural network is used to identify and locate the ignition point positions on the thermal image and the panoramic image; The flame detection system is used to perform induction detection on the space and locate the position of the ignition point.

9. The intelligent fire alarm safety control system for rail transit according to claim 7, wherein: The fire extinguishing system includes a starting bottle group, a fire extinguishing agent bottle group, and a buffer system arranged on the connecting pipeline between the starting bottle group and the fire extinguishing agent bottle group.

10. The intelligent fire alarm safety control system for rail transit according to claim 9, characterized in that: The buffer system includes a buffer tank (10), two groups of connecting pipes (20), and two groups of adjusting rods (30), and each group of the adjusting rods (30) is several; The two groups of the connecting pipes (20) respectively penetrate into the interior of the buffer tank (10) from both ends of the buffer tank (10). One end of the connecting pipe (20) located inside the buffer tank (10) is provided with an inner cylinder (22) that fits the inner wall of the buffer tank (10). One end of the connecting pipe (20) located outside the buffer tank (10) is provided with a flange (21). The adjusting rod (30) penetrates through the buffer tank (10) and the inner cylinder (22). A synchronous belt (31) is synchronously connected between several adjusting rods (30) on the same side. At least one end of the adjusting rod (30) is connected to a motor (32), and the motor (32) is fixed on the inner wall of the inner cylinder (22); Several bearing sleeves (23) are arranged inside the inner cylinder (22), and the adjusting rod (30) is fixed to the inner ring of the bearing sleeve (23).

Images