Rapid detection method for tunnel moving fire

Abstract

The invention relates to the technical field of tunnel fire detection, and provides a rapid tunnel moving fire detection method, which comprises the following steps of: calculating signals received by sensors in parallel by an intelligent algorithm, extracting undistorted characteristic signals in time, and combining data from a plurality of sensors through a data fusion technology; false alarms possibly generated by a single sensor are reduced, so that the fire detection accuracy is improved, and the time from fire occurrence to alarm giving is shortened; by fusing data of a plurality of sensors, the system is more robust to faults or abnormities of a single sensor, and the stability and reliability of the system are improved; the intelligent algorithm can be combined with the time domain oscillogram of a plurality of detectors, space and time information can be extracted in time, the specific position of the fire can be positioned more accurately, and fire extinguishing measures can be taken quickly.

Description

Technical Field

[0001] The invention relates to the technical field of tunnel fire detection, in particular to a method for quickly detecting moving fire in a tunnel. Background Art

[0002] With the rapid development of urbanization and transportation construction, tunnels are becoming more and more important in the transportation network, carrying a large number of personnel and material transportation tasks. However, the special closed space and complex environment of tunnels make their fire safety issues serious.

[0003] Once a fire occurs in a tunnel, smoke and heat accumulate quickly, making evacuation and rescue difficult, and easily causing heavy casualties and property losses. There are many flammable materials in the tunnel, such as vehicle fuel and cargo packaging, which poses a high risk of fire. Moreover, the internal environment is complex, with high temperature, high humidity, strong electromagnetic interference and other factors coexisting, which seriously affects the performance of fire detection equipment.

[0004] Traditional fire detection technologies, such as smoke, temperature, and flame detectors, are difficult to meet the needs of tunnel fire detection. Smoke detectors are easily interfered by dust and water mist, and false alarms are frequent; temperature detectors respond slowly and are difficult to detect early fires in time; flame detectors generally respond in more than 10 seconds, and to prevent false alarms, signals need to be accumulated. When encountering large flame signals, they need to be downgraded, further delaying the response. In addition, these detectors have poor positioning accuracy and cannot accurately lock the fire source, making it difficult to effectively respond to complex and changeable fire scenes in tunnels.

[0005] Therefore, a rapid detection method for tunnel mobile fire is proposed to solve the problems of slow response, many false alarms and inaccurate positioning of traditional technologies, improve the accuracy and timeliness of tunnel fire detection, enhance the safety protection capability of tunnels, and ensure the safety of life and property. Summary of the invention

[0006] Technical issues solved

[0007] Insufficient response time: Traditional fire detection systems rely on single sensors such as smoke, temperature or flame. Due to their sequential downshifting mechanism and the need to prevent false alarms, it usually takes some time to detect a fire, resulting in a delayed response time. High false alarm rate: Due to the high speed of the car during driving, the inspection time of a single detector is limited, resulting in limited data for judging and analyzing the fire. Insufficient accuracy: The detection accuracy of a single sensor is limited, especially in the complex and changeable tunnel environment, where it is difficult to accurately judge the severity and location of the fire.

[0008] Technical Solution

[0009] To achieve the above-mentioned solution objectives, the present invention provides the following technical solutions: A rapid detection method for mobile tunnel fires, applied to tunnel fire monitoring and prevention and control, includes the following steps:

[0010] S1. Multi-sensor data fusion: Based on the tunnel length, width, height, traffic flow, number of bends, ventilation conditions, and distribution of obstacles, determine the layout of flame detectors, select detectors based on the ultraviolet-infrared dual-band detection principle, collect data at a frequency of 200 Hz, and upload the data via LoRa wireless transmission technology.

[0011] S2. Intelligent algorithm analysis: Use the built-in 12-stage parallel amplification module in the flame detector to amplify the flame signal by multiple times, automatically switch the amplification multiple based on the signal intensity threshold and signal-to-noise ratio, process the data using a convolutional neural network, and use a hardware watchdog circuit and software fault tolerance mechanism to ensure device operation and data quality.

[0012] S3. Fire risk assessment: Integrate multi-sensor data, fire data of the tunnel and similar tunnels in the past 10 years, and tunnel structure data, and use the analytic hierarchy process and fuzzy comprehensive evaluation method to assess the fire risk and divide the grades.

[0013] S4. Real-time alarm and positioning: Use image recognition technology and machine learning technology to trigger the alarm system after confirming a fire, combine the time-domain waveform diagrams of multiple detectors, use the triangulation method and the time difference of arrival algorithm to locate the fire, and predict and track the fire location through a Kalman filter.

[0014] S5. Dynamically adjust the detection strategy: Set a monitoring device group including temperature and humidity sensors, electromagnetic interference sensors, and light sensors every 50 - 100 meters in the tunnel, adjust the detection parameters according to changes in environmental parameters, adjust the detection strategy according to the characteristics of different time periods, and optimize it using a genetic algorithm.

[0015] Preferably, in a tunnel with a straight section, low traffic flow, good ventilation, and no large obstacles, the layout spacing of the flame detectors is 20 - 25 meters; in a tunnel with many bends, high traffic flow, ventilation dead ends, or many large obstacles, the layout spacing is 15 - 20 meters, and the installation height is 0.5 - 1 meter from the tunnel top.

[0016] Preferably, the sensitivity of the flame detector can detect signals with a flame radiation intensity as low as 0.1 W / ㎡, the mean time between failures is not less than 50,000 hours, and it can work normally in an environment with a high temperature of 80℃, a relative humidity of 95%RH, and an electromagnetic interference intensity of 100 μT.

[0017] Preferably, when transmitting data, signal repeaters are set every 200 - 300 meters in the tunnel, and an adaptive channel selection algorithm and AES-128 encryption algorithm are used.

[0018] Preferably, the 12-stage parallel amplification module can amplify the flame signal by 1 time, 2 times, 5 times, 10 times, 20 times, 50 times, 100 times, 200 times, 500 times, 800 times, 1000 times, and 1500 times respectively.

[0019] Preferably, the training data of the convolutional neural network is sourced from real tunnel fire scene simulation experiments, historical fire data, and normal tunnel working condition data, and is trained using transfer learning, few-shot learning techniques combined with the Adam optimizer, and runs on a server equipped with an NVIDIA Tesla V100 GPU, with a memory of no less than 64GB and a hard disk capacity of no less than 1TB.

[0020] Preferably, in the fire risk assessment, the historical fire data is evaluated for accuracy and integrity and cleaned, the tunnel structure information is updated using 3D laser scanning and BIM technology, the factor weights are determined through the analytic hierarchy process, a judgment matrix is constructed and passed through a consistency test, and the fuzzy comprehensive evaluation method is used to quantify the factors to obtain a comprehensive evaluation value.

[0021] Preferably, in the fire confirmation and alarm, image recognition uses the YOLOv5 object detection algorithm combined with thermal imaging information, machine learning uses a support vector machine with a radial basis function as the kernel function, the sound intensity of the audible and visual alarm in the tunnel is not less than 85dB, the light flashing frequency is 3Hz, and standardized format fire information is sent through the SMS platform and the network platform.

[0022] Preferably, when locating the fire location, a multi-signal source location algorithm and signal reflection recognition technology are used to optimize the result, and the positioning accuracy reaches ±3 meters.

[0023] Preferably, in the environmental monitoring equipment, the temperature and humidity sensor has a measurement range of -20°C - 80°C and a relative humidity measurement range of 0% - 100%RH, with measurement accuracies of ±0.5°C and ±3%RH respectively; the electromagnetic interference sensor has a measurement range of 0 - 200μT and a measurement accuracy of ±1μT; the light sensor has a measurement range of 0 - 100000lux and a measurement accuracy of ±5%.

[0024] Beneficial effects

[0025] Compared with the prior art, the present invention provides a rapid detection method for tunnel mobile fires, having the following beneficial effects:

[0026] 1. The rapid detection method for mobile fires in tunnels. The intelligent algorithm calculates the signals received by sensors in parallel, extracts undistorted characteristic signals in a timely manner. By using data fusion technology, data from multiple sensors can be combined to reduce false alarms that may occur in a single sensor, thereby improving the accuracy of fire detection and shortening the time from fire occurrence to alarm issuance. By fusing data from multiple sensors, the system is more robust to faults or abnormalities in a single sensor, improving the stability and reliability of the system. The intelligent algorithm can combine the time-domain waveform diagrams of multiple detectors, extract spatial and temporal information in a timely manner, and more accurately locate the specific location of the fire, which helps to quickly take fire extinguishing measures.

[0027] 2. The rapid detection method for mobile fires in tunnels. The intelligent algorithm can reduce false alarms by learning normal and abnormal situations. The tunnel environment is complex and changeable, and data fusion and intelligent algorithms can better adapt to these environments and improve the performance of the system under different conditions. Through data fusion and intelligent algorithms, a more general fire detection system can be developed, which is applicable to different types of tunnels and infrastructures. The intelligent system can automatically perform fault diagnosis and self-optimization, reducing the need for manual maintenance, thereby reducing long-term operating costs. The system can quickly identify fires and provide detailed information, which helps the emergency response team make more effective decisions, reducing casualties and property losses. By providing more accurate and timely fire alarms, it enhances users' confidence in the safety of the tunnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 is a flowchart of the present invention;

[0029] Figure 2 is a module diagram of the present invention;

[0030] Figure 3 is a framework diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0031] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without creative efforts shall fall within the protection scope of the present invention.

[0032] Please refer to Figures 1 to 3 , the present invention provides a rapid detection method for mobile fires in tunnels, including the following:

[0033] 1. Multi-sensor data fusion

[0034] 1.1 Detector layout planning

[0035] Inside the tunnel, comprehensively considering factors such as the length, width, height, traffic flow, curve conditions, ventilation conditions, and distribution of obstacles, reasonably determine the layout spacing of the flame detectors. By simulating the diffusion paths of flames and smoke and the detection ranges of the detectors under different working conditions, construct a mathematical model to optimize the layout. For a straight tunnel section with low traffic flow, good ventilation, and no large obstacles, arrange a flame detector every 20 - 25 meters; for a tunnel with many curves, large traffic flow, or ventilation dead zones and many large obstacles, arrange one every 15 - 20 meters. Ensure that the detectors can fully cover the tunnel space and avoid detection blind spots.

[0036] The installation height of the flame detectors should be adjusted according to the actual situation of the tunnel, and generally installed at a position 0.5 - 1 meter from the tunnel ceiling. At the same time, use laser scanning technology to obtain the three-dimensional space information around the installation position of the detector, ensuring that the detection angle of the detector can cover a certain area around its location and is not affected by obstacles.

[0037] 1.2 Requirements for detector selection

[0038] Select a flame detector with high sensitivity and high reliability. This detector is based on the ultraviolet - infrared dual - band detection principle and can simultaneously detect the ultraviolet and infrared radiation signals emitted by the flame, effectively improving the ability to identify flame signals. Through multiple rounds of testing in a simulated tunnel environment laboratory, verify the performance of the detector in complex environments. The sensitivity of the detector should be able to detect weak flame signals with a radiation intensity as low as 0.1 W / ㎡, and the average trouble - free working time should not be less than 50,000 hours.

[0039] The detector should have good anti - interference ability and be able to work normally in harsh environments such as high temperature (maximum tolerable temperature 80°C), high humidity (relative humidity 95% RH), and strong electromagnetic interference (electromagnetic interference intensity not exceeding 100 μT). For interference sources such as vehicle exhaust and dust in the tunnel, special filtering and signal processing algorithms are integrated inside the detector to effectively suppress interference signals.

[0040] 1.3 Data acquisition and transmission

[0041] The flame detector is built - in with a high - speed data acquisition module, which real - time collects the flame signals in the tunnel at a frequency of 200 Hz. The acquisition module uses a high - precision 16 - bit A / D converter to convert the analog flame signals into digital signals.

[0042] The collected data is uploaded to the host through LoRa wireless transmission technology. To address signal occlusion and interference issues, signal repeaters are installed at regular intervals in the tunnel (determined according to the specific tunnel structure and signal attenuation, generally 200 - 300 meters), enhancing the signal transmission capacity. Meanwhile, an adaptive channel selection algorithm is adopted to automatically switch to the channel with the least interference for data transmission based on the real-time monitored channel quality. During the transmission process, the AES-128 encryption algorithm is used to encrypt the data to ensure the security and integrity of the data.

[0043] 2. Intelligent algorithm analysis

[0044] 2.1 Design of parallel amplification module

[0045] Each flame detector is equipped with a 12-stage parallel amplification module, which can amplify the flame signal by 1, 2, 5, 10, 20, 50, 100, 200, 500, 800, 1000, and 1500 times simultaneously. The amplification module adopts a high-precision operational amplifier and a stable circuit design. The gain-bandwidth product of the operational amplifier is not less than 10 MHz, and the offset voltage does not exceed 1 mV, ensuring that the amplified signal is distortion-free.

[0046] Based on the analysis of the flame signal intensity threshold and signal-to-noise ratio, a mechanism for automatically switching the amplification factor is designed. When the flame signal intensity is lower than the set low threshold, a larger amplification factor is selected; when the signal intensity is within a certain range and the signal-to-noise ratio is good, a moderate amplification factor is chosen; when the signal intensity exceeds the high threshold, a smaller amplification factor is selected to avoid signal saturation and distortion. The output end of the parallel amplification module is connected to a multiplexer, which selects the appropriate signal for subsequent processing according to the above switching mechanism.

[0047] 2.2 Application of artificial intelligence algorithms

[0048] A convolutional neural network (CNN) is used to perform fusion analysis on multiple groups of undistorted feature data collected. The training data of the CNN model comes from multiple real tunnel fire scenario simulation experiments, historical fire data, and a large amount of normal tunnel working condition data. The data is labeled to clarify categories such as flame signals and interference signals. Transfer learning and few-shot learning techniques are adopted, and the model is trained in combination with the Adam optimizer to improve the training efficiency and generalization ability of the model.

[0049] The input layer of the CNN model receives the time-domain waveform diagram of the flame signal after preprocessing. After being processed by multiple convolutional layers, pooling layers, and fully connected layers, the features of the flame signal are extracted, and classification and recognition are carried out.

[0050] On the server in the tunnel control center, an NVIDIA Tesla V100 GPU is equipped to accelerate the operation of the CNN algorithm, improving the data processing speed. The memory of the server is not less than 64 GB, and the hard disk capacity is not less than 1 TB to meet the storage and processing requirements of a large amount of data.

[0051] The analysis content includes: evaluating the possibility of a fire by analyzing the characteristics of the flame signal such as intensity, frequency, duration, and flicker pattern; comparing the signal changes collected by different detectors in a short period of time to determine whether it is a moving fire; and predicting the change trend of the fire location based on the signal change trends of multiple detectors.

[0052] 2.3 Data processing flow

[0053] During the data acquisition process, a hardware watchdog circuit and a software fault tolerance mechanism are used to ensure the stable operation of the device. The timing time of the hardware watchdog circuit is set to 100 ms. When the device does not send a watchdog feeding signal to the watchdog within 100 ms, the watchdog will reset the device. At the same time, a dual watchdog redundancy design is adopted. When one watchdog fails, the other watchdog can still work normally to ensure the reliability of the system.

[0054] The software fault tolerance mechanism adopts a redundant backup and error recovery strategy. When a data error or device failure is detected, it automatically switches to a standby device for data acquisition and records the fault information. At the same time, the data during the switching process is cached and verified to ensure the continuity and accuracy of the data.

[0055] The collected data is first preprocessed, including filtering, denoising, and normalization. Filtering uses a combination of median filtering and Gaussian filtering. The window size of the median filtering is 3×3, and the standard deviation of the Gaussian filtering is 1.5 to remove noise interference. Denoising is based on a wavelet transform-based denoising algorithm, and the Daubechies wavelet basis is selected with 3 decomposition layers to further improve the signal quality. Normalization maps the data to the [0,1] interval and uses a linear normalization method. The formula is: where x is the original data, x min and x max are the minimum and maximum values of the data respectively.

[0056] 3. Fire risk assessment

[0057] 3.1 Data fusion and analysis

[0058] Combined with the results of multi-sensor data fusion analysis (including signal characteristics, location information, etc. collected by flame detectors), historical fire data (collecting detailed data such as the time, location, cause, fire size, casualties, and property losses of fires in the tunnel and similar tunnels in the past 10 years), and tunnel structure information (such as the length, width, height, ventilation system layout, fire-fighting facilities distribution, slope, curve radius, etc.), the analytic hierarchy process (AHP) and the fuzzy comprehensive evaluation method are combined to conduct fire risk assessment.

[0059] For historical fire data, a data quality assessment system is established to evaluate and clean the accuracy and integrity of the data. For missing data, interpolation methods or prediction models based on data of similar tunnels are used for supplementation. For tunnel structure information, 3D laser scanning and BIM technology are used for real-time updating and management.

[0060] The analytic hierarchy process is used to determine the weights of various factors, construct a judgment matrix, and ensure the rationality of the weights through consistency tests. Then, the fuzzy comprehensive evaluation method is used to fuzzily quantify various factors, establish a fuzzy evaluation matrix, and finally obtain the comprehensive evaluation value of fire risk.

[0061] 3.2 Risk level classification

[0062] Considering the functions (such as passenger transportation, freight transportation, mixed) and importance of different tunnels, the fire risk level classification standard is refined. The fire risk is divided into five levels: low, relatively low, medium, relatively high, and high. For tunnels mainly for passenger transportation, when the comprehensive evaluation value is lower than 0.2, it is determined as low risk; when it is between 0.2 - 0.35, it is determined as relatively low risk; when it is between 0.35 - 0.55, it is determined as medium risk; when it is between 0.55 - 0.7, it is determined as relatively high risk; when it is higher than 0.7, it is determined as high risk. For tunnels mainly for freight transportation or with mixed functions, the risk level classification thresholds are appropriately adjusted according to factors such as the danger of transported goods and the degree of personnel density.

[0063] According to the evaluation results, a detailed risk report is generated, including information such as the areas where fires may occur, the possibility of fire spread, the threat level to personnel and property, and recommended preventive measures, providing a basis for subsequent decision-making.

[0064] 4. Real-time alarm and positioning

[0065] 4.1 Fire confirmation and alarm

[0066] When the fire risk assessment result shows a risk, the system immediately initiates a further fire confirmation procedure. Using image recognition technology and machine learning technology, it conducts real-time analysis on the surveillance video images in the tunnel. Image recognition adopts the YOLOv5 object detection algorithm and combines multi-modal information (such as thermal imaging information) to improve the recognition accuracy under harsh visual conditions such as low light and high smoke concentration. Machine learning technology uses support vector machines (SVMs) to classify and judge sensor data and image features. The kernel function of the SVM is selected as the radial basis function (RBF), and the optimal parameters are determined through the method of cross-validation.

[0067] Once a fire is confirmed, the fire alarm system is immediately triggered. A sufficient number of audible and visual alarms are set in the tunnel. The sound intensity of the audible and visual alarms is not less than 85 dB, and the light flashing frequency is 3 Hz. At the same time, text messages containing detailed information such as the fire location, fire size, and expected spread direction are sent to relevant personnel such as tunnel management staff and the fire department through the text message platform. The content of the text message adopts a standardized format for easy understanding by the recipients. In addition, through the network platform of the tunnel monitoring system, the fire information is pushed to the computer and mobile phone clients of relevant personnel in real time. The client is developed using HTML5+CSS3+JavaScript technology, with good compatibility and user experience.

[0068] 4.2 Improvement of positioning accuracy

[0069] Combining the time-domain waveform diagrams of multiple detectors, using the triangulation method and the time difference of arrival (TDOA) algorithm, the specific location of the fire occurrence is more accurately located. Multiple reference points are preset in the tunnel. By measuring the time difference and angle information of the flame signal arriving at different detectors, the location of the fire is calculated. To address the problems of multiple fire sources or signal reflection interference, a multi-signal source positioning algorithm and signal reflection recognition technology are adopted to optimize the positioning results, and the positioning accuracy can reach ±3 meters.

[0070] At the same time, by analyzing the changing trend of the fire location, the fire location information is updated in real time. The Kalman filter is used to predict and track the fire location. The state transition matrix and observation matrix of the Kalman filter are adjusted according to the actual situation of the tunnel to improve the prediction accuracy and provide accurate guidance for fire extinguishing and rescue work.

[0071] 5. Dynamic adjustment of detection strategy

[0072] 5.1 Environmental monitoring and parameter adjustment

[0073] Reasonably arrange environmental monitoring devices such as temperature and humidity sensors, electromagnetic interference sensors, and light sensors in the tunnel. According to the length, structure, and environmental change characteristics of the tunnel, set up a group of monitoring devices every 50 - 100 meters. Each group of devices includes multiple sensors to ensure that the overall environmental conditions of the tunnel can be accurately reflected. The measurement range of the temperature and humidity sensor is -20°C - 80°C, and the relative humidity measurement range is 0% - 100%RH. The measurement accuracies are ±0.5°C and ±3%RH respectively. The electromagnetic interference sensor can measure the electromagnetic interference intensity of 0 - 200μT, and the measurement accuracy is ±1μT. The light sensor can measure the light intensity of 0 - 100000lux, and the measurement accuracy is ±5%.

[0074] When the environmental parameters exceed the set thresholds, the system automatically adjusts the detection parameters. For example, when the temperature is higher than 45°C, increase the sensitivity of the flame detector and appropriately increase the magnification; when the humidity is higher than 90%RH, increase the data acquisition frequency from 200Hz to 300Hz; when the electromagnetic interference intensity exceeds 50μT, adjust the filtering parameters of the detector to enhance the anti-interference ability; when the light intensity is too high or too low, adjust the parameters of the image recognition algorithm to improve the accuracy of image recognition. At the same time, establish a correlation model between environmental parameters and detection parameters, and optimize the adjustment strategy through big data analysis to improve the accuracy and timeliness of the adjustment.

[0075] 5.2 Strategy Optimization and Adaptive Adjustment

[0076] According to the fire risk characteristics in different time periods (such as day, night, traffic peak period, trough period) and the actual situation in the tunnel, dynamically adjust the detection strategy. During the traffic peak period, appropriately reduce the detection range of the detector and increase the detection frequency to cope with the increased fire risk caused by vehicle density; at night, reduce the power consumption of the detector and adjust the alarm threshold at the same time to avoid false alarms caused by factors such as light changes.

[0077] Through the analysis of historical fire data and environmental monitoring data, continuously optimize the detection strategy. Use the genetic algorithm to optimize the detection parameters and strategies, and clarify the optimization objective function with detection accuracy, response time, and false alarm rate. The population size of the genetic algorithm is 50, the number of iterations is 100 times, the crossover probability is 0.8, and the mutation probability is 0.05. Enable the system to adapt to various complex tunnel environments and improve the detection accuracy and response speed.

[0078] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A rapid detection method for tunnel mobile fires, which is applied to tunnel fire monitoring and prevention and control, is characterized in that: It includes the following steps: S1. Multi-sensor data fusion: Based on the tunnel length, width, height, traffic flow, number of bends, ventilation conditions, and the distribution of obstacles, determine the layout of flame detectors. Select detectors based on the ultraviolet-infrared dual-band detection principle, collect data at a frequency of 200 Hz, and upload the data via LoRa wireless transmission technology; S2. Intelligent algorithm analysis: Use the 12-level parallel amplification module built into the flame detector to amplify the flame signal by multiple times, automatically switch the amplification factor based on the signal intensity threshold and signal-to-noise ratio, process the data using a convolutional neural network, and use a hardware watchdog circuit and software fault tolerance mechanism to ensure the operation of the device and the quality of the data; S3. Fire risk assessment: Integrate multi-sensor data, fire data of the tunnel and similar tunnels in the past 10 years, and tunnel structure data, and use the analytic hierarchy process and fuzzy comprehensive evaluation method to assess the fire risk and divide the levels; S4. Real-time alarm and positioning: Use image recognition technology and machine learning technology to trigger the alarm system after confirming a fire. Combine the time-domain waveform diagrams of multiple detectors, use the triangulation method and the time difference of arrival algorithm to locate the fire, and predict and track the fire location through a Kalman filter; S5. Dynamically adjust the detection strategy: Set up a monitoring device group including temperature and humidity sensors, electromagnetic interference sensors, and light sensors every 50 - 100 meters in the tunnel. Adjust the detection parameters according to the changes in environmental parameters, adjust the detection strategy according to the characteristics of different time periods, and optimize it using a genetic algorithm.

2. The rapid detection method for a tunnel mobile fire according to claim 1, wherein: In the tunnel with straight sections, low traffic flow, good ventilation, and no large obstacles, the layout spacing of the flame detectors is 20 - 25 meters; in the tunnel with many bends, large traffic flow, ventilation dead ends, or many large obstacles, the layout spacing is 15 - 20 meters, and the installation height is 0.5 - 1 meter from the tunnel top.

3. A rapid detection method for tunnel mobile fires according to claim 1, characterized in that: The sensitivity of the flame detector can detect signals with a flame radiation intensity as low as 0.1 W / ㎡, the mean time between failures is not less than 50,000 hours, and it can work normally in an environment with a high temperature of 80℃, a relative humidity of 95%RH, and an electromagnetic interference intensity of 100μT.

4. A rapid detection method for tunnel mobile fires according to claim 1, characterized in that: When transmitting data, signal repeaters are set every 200 - 300 meters in the tunnel, and an adaptive channel selection algorithm and AES-128 encryption algorithm are used.

5. A rapid detection method for tunnel mobile fires according to claim 1, characterized in that: The 12-level parallel amplification module can amplify the flame signal by 1 time, 2 times, 5 times, 10 times, 20 times, 50 times, 100 times, 200 times, 500 times, 800 times, 1000 times, and 1500 times respectively.

6. A rapid detection method for a tunnel mobile fire according to claim 1, characterized in that: The training data of the convolutional neural network comes from real tunnel fire scene simulation experiments, historical fire data, and normal tunnel working condition data. It is trained using transfer learning, few-shot learning techniques combined with an Adam optimizer, and runs on a server equipped with an NVIDIA Tesla V100 GPU, with a memory of not less than 64GB and a hard disk capacity of not less than 1TB.

7. A rapid detection method for tunnel mobile fires according to claim 1, characterized in that: In the fire risk assessment, the historical fire data is evaluated for accuracy and integrity and cleaned. The tunnel structure information is updated using 3D laser scanning and BIM technology. The factor weights are determined by the analytic hierarchy process, a judgment matrix is constructed and passed through a consistency test, and the fuzzy comprehensive evaluation method is used to quantify the factors to obtain a comprehensive evaluation value.

8. A rapid detection method for tunnel mobile fires according to claim 1, characterized in that: In the fire confirmation and alarm, image recognition uses the YOLOv5 object detection algorithm combined with thermal imaging information. Machine learning uses a support vector machine with a radial basis function as the kernel function. The sound intensity of the audible and visual alarms in the tunnel is not less than 85 dB, and the light flashing frequency is 3 Hz. Standardized format fire information is sent through the SMS platform and the network platform.

9. A rapid detection method for tunnel mobile fires according to claim 1, characterized in that: When locating the fire position, a multi-signal source location algorithm and signal reflection recognition technology are used to optimize the result, and the positioning accuracy reaches ±3 meters.

10. A rapid detection method for tunnel mobile fires according to claim 1, characterized in that: In the environmental monitoring equipment, the temperature and humidity sensor has a measurement range of -20°C to 80°C and a relative humidity measurement range of 0% to 100% RH, with measurement accuracies of ±0.5°C and ±3% RH respectively; the electromagnetic interference sensor has a measurement range of 0 - 200 μT and a measurement accuracy of ±1 μT; the light sensor has a measurement range of 0 - 100000 lux and a measurement accuracy of ±5%.

Images