Method for monitoring fire in highway tunnel

Abstract

The invention discloses a method for monitoring fire in a highway tunnel, and relates to the technical field of safety monitoring. The method comprises the following steps: constructing a multi-modal monitoring network in a highway tunnel; the multi-modal monitoring network consists of a distributed temperature sensing network, a composite gas detection module, a video shooting module and auxiliary fire extinguishing equipment; based on the multi-modal monitoring network, utilizing a trained fire early warning model to dynamically analyze real-time conditions in a target area, warning workers according to a preset grading early warning mechanism, and outputting corresponding auxiliary fire extinguishing equipment control signals; wherein the fire early warning model is constructed based on a Bayesian network model and space-time correlation analysis. According to the invention, the accuracy and timeliness of fire monitoring can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of safety monitoring, and particularly to a method for monitoring fires in highway tunnels. Background Art

[0002] As a key node of the transportation network, highway tunnels have a significant fire risk due to their enclosed and narrow special structure. Traditional fire monitoring methods mostly rely on a single sensor (such as a point-type temperature sensor or a smoke sensor), which have significant limitations: on the one hand, interference such as vehicle exhaust and mechanical friction is likely to cause false alarms, and the lack of accurate fire source positioning leads to response delays; on the other hand, the lack of multi-parameter fusion analysis makes it difficult to effectively identify risks in the initial stage of a fire (smoldering stage). In recent years, with the development of intelligent sensing technology, multi-modal monitoring has gradually become a research direction. Although existing technologies have tried to combine temperature and video data, there are generally problems such as insufficient data fusion depth and poor adaptability of the early warning model. For example, some systems only judge the fire situation through simple thresholds and cannot dynamically adapt to the complex environmental changes in the tunnel; or rely on manual review of alarms, which is difficult to meet the real-time requirements. Summary of the Invention

[0003] The purpose of the present invention is to provide a method for monitoring fires in highway tunnels, which can improve the accuracy and timeliness of fire monitoring.

[0004] To achieve the above purpose, the present invention provides the following solution:

[0005] A method for monitoring fires in highway tunnels, comprising:

[0006] Constructing a multi-modal monitoring network in the highway tunnel; the multi-modal monitoring network is composed of a distributed temperature sensing network, a composite gas detection module, a video shooting module, and auxiliary fire extinguishing equipment;

[0007] Based on the multi-modal monitoring network, using a trained fire early warning model to dynamically analyze the real-time situation in the target area, give an alarm to the staff according to a preset hierarchical early warning mechanism, and output a corresponding control signal for the auxiliary fire extinguishing equipment; wherein, the fire early warning model is constructed based on a Bayesian network model and spatio-temporal correlation analysis.

[0008] Optionally, the construction method of the multi-modal monitoring network is:

[0009] Laying high-temperature-resistant optical fibers along the tunnel top / sidewall at a set interval, and installing infrared cameras at intervals on the tunnel top to cover the entire cross-section, constructing the distributed temperature sensing network, and real-time monitoring the temperature distribution of the whole line to locate abnormal temperature rise areas;

[0010] Set multi-parameter sensor nodes near traffic intersections, and configure sensors according to the placement of each node to construct a composite gas detection module to distinguish between fire and vehicle exhaust interference;

[0011] Install flame / smoke recognition cameras according to the fire risk locations and construct a video shooting module;

[0012] Set auxiliary fire extinguishing equipment according to the layout positions of the distributed temperature sensing network, the composite gas detection module, and the video shooting module.

[0013] Optionally, the composite gas detection module specifically includes: a CO sensor, a CO2 sensor, a smoke particle sensor, and a VOCs sensor.

[0014] Optionally, before dynamically analyzing the real-time situation in the target area using the trained fire warning model, it further includes: training the fire warning model.

[0015] Optionally, the process of training the fire warning model includes:

[0016] Obtain training data; the training data includes historical relevant data and corresponding prediction labels; the prediction labels include whether there is a fire in the current range, whether there is a fire source, and the fire source location information when there is a fire source;

[0017] Construct a pre-training network based on the Bayesian network model and the spatio-temporal correlation analysis;

[0018] Input the training data into the pre-training network, aim at minimizing the loss between the network output and the prediction labels, perform training according to the gradient descent strategy, and determine the network that meets the accuracy evaluation as the final fire warning model.

[0019] Optionally, during the training process of the pre-training network, the loss function used is the loss function.

[0020] Optionally, the accuracy evaluation includes overall accuracy, F1 score, Kappa coefficient, and recall rate.

[0021] Optionally, the hierarchical warning mechanism includes a three-level response protocol, namely a first-level yellow warning, a second-level orange warning, and a third-level red warning.

[0022] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0023] The present invention discloses a method for fire monitoring in a highway tunnel. The method includes constructing a multimodal monitoring network in the highway tunnel. The multimodal monitoring network consists of a distributed temperature sensing network, a composite gas detection module, a video shooting module, and auxiliary fire extinguishing equipment. Based on the multimodal monitoring network, a trained fire warning model is used to dynamically analyze the real-time situation in the target area, give warnings to the staff according to a preset hierarchical warning mechanism, and output corresponding control signals for the auxiliary fire extinguishing equipment. Among them, the fire warning model is constructed based on a Bayesian network model and spatio-temporal correlation analysis. The present invention can improve the accuracy and timeliness of fire monitoring. Description of the Drawings

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0025] Figure 1 It is a schematic flowchart of the method for fire monitoring in a highway tunnel of the present invention. Detailed Embodiments

[0026] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0027] The purpose of the present invention is to provide a method for fire monitoring in a highway tunnel, which can improve the accuracy and timeliness of fire monitoring.

[0028] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the drawings and specific embodiments.

[0029] As Figure 1 shown, the present invention provides a method for fire monitoring in a highway tunnel, including:

[0030] Step 100: Construct a multimodal monitoring network in the highway tunnel. The multimodal monitoring network consists of a distributed temperature sensing network, a composite gas detection module, a video shooting module, and auxiliary fire extinguishing equipment. Among them, the composite gas detection module specifically includes: a CO sensor, a CO2 sensor, a smoke particle sensor, and a VOCs sensor.

[0031] Step 200: Based on the multi-modal monitoring network, use the trained fire warning model to dynamically analyze the real-time situation in the target area, give warnings to the staff according to the preset hierarchical warning mechanism, and output the corresponding control signals for auxiliary fire extinguishing equipment; wherein, the fire warning model is constructed based on the Bayesian network model and spatio-temporal correlation analysis. The hierarchical warning mechanism includes three-level response protocols, namely, the first-level yellow warning, the second-level orange warning, and the third-level red warning.

[0032] As a specific implementation manner, the construction method of the multi-modal monitoring network is as follows:

[0033] Lay high-temperature resistant optical fibers along the top / side wall of the tunnel at a set interval, and install infrared cameras at intervals on the top of the tunnel to cover the entire cross-section, construct the distributed temperature sensing network, monitor the temperature distribution of the whole line in real time, and locate the abnormal temperature rise area; set up multi-parameter sensor nodes near the traffic intersection, and install sensors according to the settings of each node to construct a composite gas detection module to distinguish between fire and vehicle exhaust interference; install flame / smoke recognition cameras according to the fire risk positions, and construct a video shooting module; set auxiliary fire extinguishing equipment according to the layout positions of the distributed temperature sensing network, the composite gas detection module, and the video shooting module.

[0034] As a specific implementation manner, before using the trained fire warning model to dynamically analyze the real-time situation in the target area, it further includes: training the fire warning model.

[0035] The process of training the fire warning model includes:

[0036] Obtain training data; the training data includes historical relevant data and corresponding prediction labels; the prediction labels include whether there is a fire in the current range, whether there is a fire source, and the fire source position information when there is a fire source; construct a pre-training network based on the Bayesian network model and the spatio-temporal correlation analysis; input the training data into the pre-training network, aim at minimizing the loss between the network output and the prediction labels, train according to the gradient descent strategy, and determine the network that meets the accuracy evaluation as the final fire warning model.

[0037] In a further technical solution, during the training process of the pre-training network, the loss function used is the loss function. The accuracy evaluation includes overall accuracy, F1 score, Kappa coefficient, and recall rate.

[0038] Based on the above technical solutions, the following embodiments are provided.

[0039] Taking a two-way four-lane highway tunnel in the mountainous area as an example, the tunnel is 3.2 kilometers long, with a daily traffic flow of about 8,000 vehicles. There are long steep sections, and the fire risk is relatively high. It is necessary to deploy a multi-modal monitoring network inside the tunnel and link it with the existing ventilation, lighting, and fire protection systems in the tunnel.

[0040] First, deploy the multi-modal monitoring network.

[0041] 1. Distributed temperature sensing network

[0042] Lay high-temperature resistant optical fibers (model: FBG-T100) at 20-meter intervals along the top of the tunnel sidewall to form a continuous temperature measurement chain, covering the entire tunnel.

[0043] Install 3 infrared thermal imagers (resolution 640×480) at the entrance, middle, and exit of the tunnel respectively. The viewing angle covers the entire cross-section, and focuses on monitoring the vehicle engine compartment and tire area.

[0044] 2. Composite gas detection module

[0045] Set gas sensor nodes 50 meters after the tunnel entrance, in the middle, and 50 meters before the tunnel exit. Each node integrates:

[0046] CO sensor (range 0 - 500 ppm, accuracy ±1 ppm)

[0047] CO2 sensor (range 0 - 5000 ppm, accuracy ±50 ppm)

[0048] Smoke particle sensor (PM2.5 detection range 0 - 1000 μg / m 3 )

[0049] VOCs sensor (PID principle, detection range 0 - 20 ppm)

[0050] Calibrate the baseline through the hydrogen compensation algorithm to exclude vehicle exhaust interference.

[0051] 3. Video shooting module

[0052] Install 1 AI camera (supporting flame / smoke recognition, frame rate 30 fps) every 50 meters on the top of the tunnel, arranged staggeredly with the position of the infrared thermal imager.

[0053] The camera is built-in with an edge computing unit and runs a lightweight YOLOv5 fire detection model in real time.

[0054] 4. Auxiliary fire extinguishing equipment

[0055] Synchronously deploy high-pressure fine water mist sprinkler devices at the positions of the sensor nodes, with a coverage radius of 15 meters, and link them with the gas detection module.

[0056] Secondly, training of the fire warning model

[0057] 1. Data collection

[0058] Historical data: Extract the operation data of the tunnel in the past 3 years, including temperature field distribution, gas concentration, video frames and fire records.

[0059] Simulation data: Generate 1000 groups of virtual scene data (including different fire source positions and intensities) through CFD fire simulation.

[0060] Label definition:

[0061] Label 1: Whether a fire has occurred (binary classification)

[0062] Label 2: Fire source position (divided into 10 regions)

[0063] Label 3: Fire development stage (initial / mid-stage / out of control)

[0064] 2. Model construction and training

[0065] Bayesian network layer: Establish a joint probability distribution model of temperature - gas concentration - video features.

[0066] Spatio-temporal correlation layer: Learn time series features through the LSTM network and extract spatial fire spread patterns through the CNN network.

[0067] Loss function: Adopt weighted cross-entropy loss (fire sample weight + 2) + fire source position regression loss (MSE).

[0068] Training process:

[0069] Input data: 80% training set + 20% validation set

[0070] Optimizer: Adam (learning rate 0.001)

[0071] Termination condition: The F1 score of the validation set improves by < 0.1% for 5 consecutive rounds

[0072] 3. Accuracy evaluation

[0073] Overall accuracy: 92.3%

[0074] F1 score: 0.89 for fire recognition, 0.82 for fire source location

[0075] Kappa coefficient: 0.85

[0076] Recall rate: 98.7% for fire events (false negative rate < 1.3%)

[0077] Finally, conduct actual operation and emergency response

[0078] 1. Initial detection

[0079] 14:25:12: The infrared thermal imager in the middle of the tunnel detected abnormal temperature (180°C) in the tire area of the truck with license plate number Shanghai A12345, triggering a first-level yellow warning.

[0080] 14:25:15: The gas sensor detected a sudden increase in CO concentration (320 ppm) in this area, and the smoke particle concentration exceeded the standard (450 μg / m 3 ), and the model calculated FPI = 0.78.

[0081] 2. Dynamic analysis

[0082] 14:25:18: Video AI identified an open fire at the tire. The model comprehensively determined the fire source location (K3+120), and upgraded to a second-level orange warning.

[0083] 14:25:20: The sound and light alarm was activated, the tunnel entrance information board showed "Accident ahead, slow down", and the entrance traffic lights flashed yellow and red.

[0084] 3. Fire extinguishing linkage

[0085] 14:25:25: The fire spread to the carriage (temperature > 300°C). The model predicted that the probability of the fire getting out of control was > 85%, and it was automatically upgraded to a third-level red warning.

[0086] 14:25:28: The rolling shutter door at the tunnel entrance was closed, the sprinkler device at K3+100 was activated, and a fire report (including coordinates and video screenshots) was sent to the fire brigade at the same time.

[0087] 14:25:35: The ventilation system was switched to the reverse smoke exhaust mode (wind speed 8 m / s) to guide vehicles to evacuate from the nearest exit.

[0088] 4. Post-disaster recovery

[0089] 14:40:00: After the fire was extinguished, the system started self-diagnosis, detected that the survival rate of the sensors was 98%, and sent back the structural damage assessment report through the LoRa network.

[0090] 14:50:00: The drone inspection unit entered the tunnel, and after confirming that there was no risk of re-ignition, the traffic resumed.

[0091] As a summary of the system performance of this embodiment:

[0092]

[0093]

[0094] Therefore, it can be seen that this embodiment has the following beneficial effects:

[0095] This solution proposes a highway tunnel fire monitoring method based on a multi-modal monitoring network and an intelligent early warning model. By constructing a collaborative system of a distributed temperature sensing network, a composite gas detection module, a video shooting module, and auxiliary fire extinguishing equipment, multi-dimensional perception of fire characteristics is achieved. Among them, the fire early warning model uses a Bayesian network to fuse multi-source data and combines spatio-temporal correlation analysis to analyze the law of fire spread, significantly improving the accuracy and robustness of fire recognition. The hierarchical early warning mechanism can initiate targeted fire extinguishing measures when the fire source has not fully developed by dynamically adjusting the response strategy, providing a critical time window for personnel evacuation and emergency decision-making, and solving the technical bottlenecks of high false alarm rates and response lags in traditional monitoring technologies.

[0096] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same and similar parts among the various embodiments, reference can be made to each other.

[0097] Specific examples are used in this article to elaborate on the principles and implementation methods of the present invention. The descriptions of the above embodiments are only used to help understand the core idea of the present invention. At the same time, for those of ordinary skill in the art, based on the idea of the present invention, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A fire monitoring method in a highway tunnel, characterized in that, Including: Constructing a multi-modal monitoring network in a highway tunnel; The multi-modal monitoring network consists of a distributed temperature sensing network, a composite gas detection module, a video shooting module, and auxiliary fire extinguishing equipment; Based on the multi-modal monitoring network, using the trained fire warning model to dynamically analyze the real-time situation in the target area, warning the staff according to the preset hierarchical warning mechanism, and outputting the corresponding control signal for the auxiliary fire extinguishing equipment; among them, the fire warning model is constructed based on the Bayesian network model and spatio-temporal correlation analysis.

2. The method for monitoring fire in a highway tunnel according to claim 1, characterized in that, The construction method of the multi-modal monitoring network is as follows: Laying high-temperature-resistant optical fibers along the tunnel top / sidewall at a set interval, and installing infrared cameras at intervals on the tunnel top to cover the entire cross-section, constructing the distributed temperature sensing network to monitor the temperature distribution of the whole line in real time and locate the abnormal temperature rise area; Setting multi-parameter sensor nodes near traffic intersections and installing sensors according to the settings of each node to construct a composite gas detection module to distinguish between fire and vehicle exhaust interference; Installing flame / smoke recognition cameras according to the fire risk location and constructing a video shooting module; Setting auxiliary fire extinguishing equipment corresponding to the layout positions of the distributed temperature sensing network, the composite gas detection module, and the video shooting module.

3. The method for monitoring fire in a highway tunnel according to claim 2, wherein The composite gas detection module specifically includes: a CO sensor, a CO2 sensor, a smoke particle sensor, and a VOCs sensor.

4. The method for monitoring a fire in a highway tunnel according to claim 1, characterized in that, Before using the trained fire warning model to dynamically analyze the real-time situation in the target area, it also includes: training the fire warning model.

5. The method for monitoring fire in a highway tunnel according to claim 4, wherein The process of training the fire warning model includes: Obtaining training data; the training data includes historical relevant data and corresponding prediction labels; the prediction labels include whether there is a fire in the current range, whether there is a fire source, and the fire source location information when there is a fire source; Constructing a pre-training network based on the Bayesian network model and the spatio-temporal correlation analysis; Inputting the training data into the pre-training network, aiming at minimizing the loss between the network output and the prediction label, training according to the gradient descent strategy, and determining the network that meets the accuracy evaluation as the final fire warning model.

6. The method for monitoring fire in a highway tunnel according to claim 5, characterized in that, During the training process of the pre-training network, the loss function used is the loss function.

7. The method for monitoring fire in a highway tunnel according to claim 5, wherein The accuracy evaluation includes overall accuracy, F1 score, Kappa coefficient, and recall rate.

8. The method for monitoring fire in a highway tunnel according to claim 1, characterized in that, The hierarchical warning mechanism includes a three-level response protocol, namely a first-level yellow warning, a second-level orange warning, and a third-level red warning.

Images