Search

CN-120599764-B - Tunnel fire disaster early warning method and system based on deep learning

CN120599764BCN 120599764 BCN120599764 BCN 120599764BCN-120599764-B

Abstract

The invention relates to the technical field of fire early warning and discloses a tunnel fire early warning method and a system based on deep learning, wherein the tunnel fire early warning method based on deep learning comprises the steps of deploying a sensor and constructing a wireless transmission network to acquire multi-mode data; preprocessing, feature extraction and feature fusion are carried out on the multi-mode data, a tunnel fire early warning model adapting to the construction environment is built, a dynamic threshold mechanism is designed, a multi-stage early warning response mechanism is designed, and early warning is carried out. The invention solves the problem of easy dust interference by arranging the dustproof smoke sensor, automatically adjusts the sampling frequency and the communication frequency according to the monitoring requirement during data transmission, ensures the monitoring effect and prolongs the service life of a battery, builds a TCNN-based tunnel fire early-warning model and builds a large-scale fire data set through FDS digital simulation to train the model, designs a dynamic threshold value which can be automatically adjusted according to the environment, and solves the problem that a fixed threshold value system is not suitable for changing the environment.

Inventors

  • LIU XIAOPING
  • AN ZIYING
  • YE ZICHAO

Assignees

  • 合肥工业大学

Dates

Publication Date
20260508
Application Date
20250623

Claims (7)

  1. 1. The tunnel fire early warning method based on deep learning is characterized by comprising the following steps of: deploying sensors and constructing a low-power wireless transmission network to acquire multi-mode data; performing data preprocessing, feature extraction and feature fusion on the acquired multi-mode data; constructing a tunnel fire early warning model adapting to a construction environment and designing a dynamic threshold mechanism; the tunnel fire early warning model is designed into a multi-task output layer, and the specific output is as follows: outputting fire risk scores between 0 and 1; Fire type identification, namely distinguishing different types of fires, outputting the fire as a vector, and enabling the ith component of the vector to represent the probability of the fire as the ith type; The development trend prediction comprises the steps of predicting the development path and speed of a fire, and respectively representing the growth rate, the spreading speed, the main direction, the probability distribution of each stage of the fire and the predicted dangerous time by a vector; The fire risk score is used for judging whether fire occurs for the first time, the judging mode is that the fire risk score is compared with a dynamic threshold value, if the fire risk score is higher than the dynamic threshold value, an adaptive dynamic threshold algorithm is designed according to environmental characteristics of different construction stages, the calculated dynamic threshold value is used for judging whether fire occurs, and the calculation formula of the dynamic threshold value is as follows: Wherein: A dynamic threshold value (0-1) at time t; 、 、 、 Is a first, a second, a third fourth dynamic threshold adjustment coefficient; Is a basic threshold value, and is statistically determined by historical data; adjusting the item for the environmental factors at the time t; A construction activity adjustment item at the time t; A trend adjustment item at time t; the environmental factor adjustment item calculating method comprises the following steps: Wherein: The ambient temperature at time t; Is the reference temperature; The environmental humidity at the time t; is the reference humidity; Adjusting weight coefficients for the first and second environmental factors; The construction activity adjustment item calculating method comprises the following steps: Wherein: an indication function for the ith construction activity; the weight coefficient of the ith construction activity; The total number of types for the construction activity; the trend adjustment item calculation method comprises the following steps: Wherein: Is the rate of change of temperature; Is the time window width; controlling the trend to influence the intensity for adjusting the coefficient; and designing a multi-stage early warning response mechanism and carrying out early warning.
  2. 2. The tunnel fire early warning method based on deep learning according to claim 1, wherein the deployed sensor comprises a photoelectric smoke sensor with self-cleaning function, and the photoelectric smoke sensor comprises the following core components: the photoelectric smoke detection part adopts an infrared LED light source with the wavelength of 850nm and a photoelectric receiver to form a scattered light path; The micro compression air pump has the maximum pressure of 0.1MPa, the flow rate of 2L/min and the volume of less than 50 multiplied by 30 multiplied by 20mm; The air flow control valve is used for controlling the air flow intensity and direction through PWM; the dust concentration detector is used for monitoring the environmental dust concentration in real time and triggering a self-cleaning function.
  3. 3. The tunnel fire early warning method based on deep learning according to claim 2 is characterized in that the physical layer of the low-power wireless transmission network adopts LoRa technology, the link layer adopts TDMA-based protocol, the network layer realizes AODV-based self-organizing Mesh network, and the power consumption control formula of the node equipment is as follows: Wherein: the total power consumption of the nodes; is the base power consumption; power consumption for single sensing; Is the sampling frequency; Power consumption for single communication; Is the communication frequency.
  4. 4. The tunnel fire early warning method based on deep learning according to claim 3, wherein the sampling frequency and the communication frequency realize power consumption balance through dynamic adjustment, and the automatic switching of three working modes of a low power consumption mode, an alert mode and an emergency mode is realized through the local computing environment risk level through dynamic adjustment.
  5. 5. The tunnel Fire early warning method based on deep learning according to claim 4, wherein the feature extraction includes extracting thermal anomaly region features by using a modified YOLOv-Fire model, enhancing the capability of detecting small target thermal anomalies by adding a attention mechanism, adding a convolution block attention module in a backbone network, and the thermal anomaly target detection formula is as follows: Wherein: A conditional probability that the target belongs to the i-th class, representing the probability that the detected object belongs to flame, hot spot or smoke; the probability of the target in the image is represented by the confidence that the detection frame contains the actual target; measuring positioning accuracy for the intersection ratio of the prediction boundary frame and the real boundary frame; a confidence threshold is detected for the image, and the confidence threshold is used for judging whether fire features exist in the thermal image; The image feature detection confidence threshold value improves the accuracy of thermal image feature extraction by dynamically adjusting the image feature detection confidence threshold value under different environmental conditions, reduces the false detection rate of feature level, and adopts the following strategies: threshold=0.6 in normal construction environment; Threshold=0.5 at night or under low light conditions; Threshold=0.7 at high risk operation.
  6. 6. The tunnel fire early warning method based on deep learning according to claim 5, wherein the fire data set required by the tunnel fire early warning model training is constructed by establishing a target tunnel model by using FDS software, performing fire dynamics numerical simulation, and acquiring simulated fire temperature data and smoke images.
  7. 7. A tunnel fire early warning system based on deep learning, which is used for executing the tunnel fire early warning method based on deep learning as claimed in any one of claims 1 to 6, comprising: The multi-mode data acquisition module is used as a front end sensing layer of the system and is responsible for collecting related parameters in a tunnel construction environment; The low-power wireless transmission module solves the problems of unstable power supply and easy damage of cables in a tunnel construction environment and realizes reliable transmission of data; The data preprocessing and feature extraction module is responsible for converting the original sensing data into feature characterization with discriminant; The tunnel fire early warning model module is constructed based on a deep learning technology, adopts a transposed convolutional neural network architecture and is a core decision unit of the system; the dynamic threshold mechanism module adopts a deep Q network to realize a reinforcement learning framework, and can automatically adjust the early warning threshold according to the environment; The multi-stage early warning response module is responsible for triggering corresponding-level emergency measures according to the risk scores output by the early warning model; and the system integration and management module integrates all the functional modules into a unified system platform and provides management interfaces, data storage and analysis and system maintenance functions.

Description

Tunnel fire disaster early warning method and system based on deep learning Technical Field The invention relates to the technical field of fire early warning, in particular to a tunnel fire early warning method and system based on deep learning. Background The tunnel construction stage has a large amount of open fire operation and electrical equipment, and is a fire disaster high-rise area. Once the fire disaster occurs in the construction period of the tunnel, the accident of group death and group injury is extremely easy to be caused due to the closed space and limited escape passage, and the engineering delay and the huge economic loss can be caused. In construction, the tunnel is usually only provided with a unidirectional outlet, the temporary support structure is limited in strength, collapse is easy to cause after a fire disaster occurs, and the casualties risk is further increased. The mountain tunnel is remote in geographic position, the external rescue arrival time is long, and early warning and self-rescue are needed. In a highway network extension project, a mountain area long tunnel (the whole length is 8.3 km) is in a construction stage. The tunnel traverses a variety of complex geologic structures including fault zones, carbonaceous shale and coal seam regions, with abundance of groundwater. In the tunnel construction process, a large number of tunneling machines, transport vehicles, temporary power supply equipment and welding operations are involved. The average number of constructors per shift reaches 120, and the constructors can work continuously for 24 hours per day. Construction equipment and material stacking points in the tunnel are dispersed, the temporary fan provides limited ventilation, the ventilation effect of a part of areas is poor, and dust concentration in the air is high. The fire disaster early warning technology adopted at present mainly comprises: and the point type temperature sensor is arranged at the top of the tunnel according to fixed intervals, and an alarm is triggered when the local temperature exceeds a preset threshold value. The point type smoke detector is based on the photoelectric principle, and triggers an alarm when the smoke concentration reaches a certain degree. And the linear optical fiber temperature measuring system is used for paving a temperature sensing optical fiber along the tunnel and monitoring temperature abnormality by analyzing optical signal change. And (3) monitoring by a fixed camera, wherein the suspicious smoke and open fire are monitored by manual or basic video analysis software. And (3) manual inspection, namely arranging special persons to periodically inspect the construction site, and immediately reporting the discovered fire. In a complex tunnel construction environment, the existing fire early warning technology has the following obvious defects: the dust interference problem is that a large amount of dust generated in the construction process can cause frequent false alarm of the traditional photoelectric smoke detector, and meanwhile, the dust attached to the surface of the sensor can reduce the sensitivity and prolong the reaction time. The existing system cannot effectively distinguish normal dust in a construction environment from smoke generated by fire, lacks a self-cleaning function and needs frequent maintenance to keep a normal working state. The data acquisition and transmission are unstable, the power supply in the tunnel construction environment is unstable, the cable is easy to damage, and the reliability of the traditional wired sensing network is low. The sensor adopts a fixed power consumption working mode, has high energy consumption and short battery endurance, and is not suitable for long-time remote monitoring. The transmission network lacks self-organizing and self-recovering capabilities, and single-point faults easily lead to paralysis of the whole monitoring system. The multi-source heterogeneous data fusion is insufficient, and the existing system usually adopts a single sensing mode, for example, the temperature or smoke detection is only relied on, so that the collaborative analysis of the multi-source data can not be realized. The early warning model is too simplified, that is, a fixed threshold trigger mechanism is generally adopted in the traditional system, so that the system cannot adapt to the changing environment of different construction stages (blasting, tunneling, masonry and the like). The early warning decision lacks the ability to predict the development trend of fire, and key information such as the direction and speed of fire spreading cannot be provided. The response mechanism is single, and most of the existing systems are single-level alarms, and the mechanism for grading response according to the risk degree is lacking. The traditional fire early warning method has poor effect under complex construction environment, is difficult to meet the safety production requirement, and nee