A highway tunnel intelligent fire safety early warning system

Abstract

The application relates to the technical field of expressway tunnel safety monitoring, and discloses an intelligent fire safety early warning system for an expressway tunnel, which comprises a data acquisition and preprocessing module, a dynamic risk assessment module, an emergency resource elastic adaptation module and a distributed edge computing node; the data acquisition and preprocessing module is used for collecting video monitoring data, environmental sensor data and vehicle positioning data in the tunnel in real time; the dynamic risk assessment module is used for constructing a risk field model based on the spatiotemporal coupling of vehicle trajectories and smoke diffusion; the emergency resource elastic adaptation module is used for automatically adjusting an induction screen and fire-fighting equipment according to the risk intensity; and the distributed edge computing node is used for realizing adaptive migration of a computing task; the application avoids the limitations of traditional threshold alarms, significantly advances fire warning nodes through dynamic fusion of multi-source data, guarantees the long-term stability of a monitoring benchmark by using an optical flow compensation and pollution self-repairing mechanism, and significantly improves the initiative and reliability of tunnel safety prevention and control.

Description

TECHNICAL FIELD

[0001] The present application relates to a highway tunnel intelligent fire safety early warning system, belonging to the technical field of highway tunnel safety monitoring. BACKGROUND

[0002] As a semi-closed traffic facility, fire prevention and control of highway tunnels need to sense environmental parameters, vehicle status and emergency resource distribution in real time. Traditional systems usually use threshold trigger type alarm mechanisms, such as starting fire response when the temperature or smoke concentration exceeds the preset threshold. Although this method can deal with open fire scenes, it has significant defects in early smoldering identification, dynamic resource scheduling and complex data fusion:

[0003] 1. Existing systems rely on single-dimensional parameters such as temperature and smoke concentration to make independent judgments, without considering the spatio-temporal dislocation of smoke diffusion path and vehicle trajectory caused by airflow disturbance in the tunnel. For example, when video monitoring captures smoke texture, traditional methods lack coupling analysis with gas diffusion path, and are easily affected by light refraction or local airflow interference, leading to false alarms or missed alarms. To solve this problem, the industry usually increases the density of sensors, but the problem of multi-source data islandization is aggravated, and the timestamp alignment error of different modal data is further magnified, making it difficult to build a global risk field model.

[0004] 2. Current fire linkage is mostly based on preset static path planning, without integrating the dynamic game relationship between real-time traffic flow and fire spread speed. For example, when heavy trucks are blocked, fire trucks cannot pass according to the preset route, and existing systems lack the ability to coordinate and schedule vehicle-mounted mobile fire extinguishing devices and fixed fire hydrants. Although some solutions introduce path optimization algorithms, they do not consider the physical constraints of the vehicle-mounted water tank capacity decaying with driving speed, leading to an imbalance between fire extinguishing agent delivery efficiency and path feasibility.

[0005] 3. Existing distributed systems mostly use independent nodes to process local data, without realizing model parameter migration and task takeover across nodes. For example, when a node drifts due to hardware failure or pollution, neighboring nodes cannot quickly reconstruct the data repair matrix due to the lack of topology awareness mechanism, causing the interruption of local risk field deduction. The industry tries to improve reliability through redundancy deployment, but the problem of computing resource fragmentation is prominent, and it cannot adapt to the characteristics of long-distance linear distribution of tunnels.

[0006] Therefore, how to realize spatio-temporal coupling analysis of multi-source data, dynamic resource elastic adaptation and edge node collaborative fault tolerance has become a technical problem to be solved by the present application. SUMMARY

[0007] The present application provides a highway tunnel intelligent fire safety early warning system, which mainly aims to solve the problems of early warning lag, emergency resource mismatch and insufficient system robustness.

[0008] To achieve the above object, the application provides a highway tunnel intelligent fire safety early warning system, which comprises:

[0009] A data acquisition and preprocessing module is used for collecting multi-source heterogeneous data in the tunnel in real time, and the multi-source heterogeneous data at least comprises video monitoring data, environmental sensor data and vehicle positioning data; vehicle motion trajectory and smoke texture features are extracted from the video monitoring data, time and space interpolation compensation is performed on the environmental sensor data, and when the video pixel change rate ΔP in the video monitoring data and the carbon monoxide concentration change gradient in the environmental sensor data satisfy the following conditions: cross-modal early smoldering feature recognition is performed, and when the following conditions are satisfied: early smoldering is determined.

[0010] A dynamic risk assessment module is in communication connection with the data acquisition and preprocessing module, and is used for constructing a fire risk field model based on time and space coupling of multi-source heterogeneous data, and specifically comprises the following steps: generating a risk propagation vector field based on the vehicle motion trajectory and the smoke texture features, generating a dynamic accessibility heat map in combination with static parameters such as the positions of fire hydrants and the states of smoke exhaust ports in the tunnel and real-time vehicle distribution, and establishing a game model of fire spread speed and traffic flow speed to predict the best rescue time window.

[0011] An emergency resource elastic adaptation module is in communication connection with the dynamic risk assessment module, and is used for automatically switching the display strategy of the induction screen according to the intensity gradient of the fire risk field model, and forming a negative feedback loop of fire pump pressure regulation and smoke exhaust fan speed to suppress the back-burning of smoke.

[0012] An edge computing node is distributedly arranged in the tunnel, and is used for executing the calculation tasks of the data acquisition and preprocessing module and the dynamic risk assessment module, and realizing adaptive migration learning of model parameters through topological perception, so that the calculation tasks of the adjacent node can be seamlessly taken over in the case of single point failure.

[0013] Preferably, the data acquisition and preprocessing module further comprises a light flow compensation mechanism, specifically: the light flow field of the fixed light source in the tunnel side wall monitoring video is used to inverse the real-time air flow velocity field, and the time and space mapping relationship between the smoke diffusion path and the vehicle trajectory is dynamically corrected based on the real-time air flow velocity field to compensate the nonlinear time and space distortion caused by air flow disturbance.

[0014] Preferably, the data acquisition and preprocessing module further comprises a local pollution perception and image self-repairing mechanism, which specifically comprises the following steps: identifying the pollution features on the surface of the fixed light source based on a lightweight model to generate a pollution area mask; constructing a multi-frame reference repair matrix of the pollution area by using the spatial symmetry and temporal illumination consistency of adjacent fixed light sources; and when the pollution area ratio S polluted / S total >θ2 and lasts for a preset time Tpolluted At this time, the repair algorithm is activated, and non-emergency repair is suppressed according to the humidity data in the tunnel, wherein S polluted S is the area of the contaminated area total S is the total area of the light source.

[0015] Preferably, the local pollution perception and image self-repair mechanism further comprises: when it is identified that the contaminated area overlaps with the predicted path of the vehicle trajectory, the area is preferentially repaired; and the sampling frequency of the repair area is dynamically adjusted based on the predicted speed of the trajectory; and the repaired optical flow data is fed back for optimizing the vehicle trajectory prediction model.

[0016] Preferably, the dynamic risk assessment module establishes a game model of fire spread speed and traffic flow speed, also considers the influence of heavy truck congestion on the route of the fire truck, and combines the cooperative scheduling ability of the on-board mobile fire extinguishing device and the fixed fire hydrant to generate an optimal fire extinguishing resource scheduling scheme.

[0017] Preferably, the dynamic risk assessment module introduces a water tank capacity attenuation factor when generating the optimal fire extinguishing resource scheduling scheme, and calculates the Pareto front solution set of the optimal path and the injection pressure in real time.

[0018] Preferably, the emergency resource elastic adaptation module embeds a second derivative feedback term of the fire water pressure sensor in the exhaust fan control loop when forming a negative feedback loop of fire water pump pressure regulation and exhaust fan speed, to dynamically suppress air turbulence caused by sudden changes in water pressure.

[0019] Preferably, the dynamic risk assessment module performs microsecond-level time sequence alignment between the vehicle motion trajectory in the video stream and the smoke diffusion direction when generating the risk propagation vector field, to compensate for the time sequence misalignment of the smoke diffusion path and the video motion trajectory caused by tunnel air flow disturbance.

[0020] Preferably, the dynamic risk assessment module introduces a fire truck turning radius constraint when calculating the time-space game equilibrium point of the evacuation path and the optimal route of the fire truck, to generate a curvature-continuous optimal path.

[0021] Preferably, the edge computing nodes periodically broadcast their respective local risk assessment results and resource state information, to realize collaborative perception of the overall risk situation of the tunnel and distributed collaborative scheduling of emergency resources.

[0022] Compared with the background art, the present application has the following advantages:

[0023] 1. Through the dynamic coupling analysis of video pixel change characteristics and environmental parameter gradient, the system can capture the implicit correlation between micro heat release and gas diffusion that cannot be recognized by traditional single sensor. When the space-time propagation path of smoke texture and the concentration field form a certain phase difference, the system automatically triggers the early intervention strategy, making the fire warning node advance to the smoldering stage. This mechanism effectively compensates for the monitoring distortion caused by airflow disturbance in the tunnel, and realizes the self-consistency verification of multi-physical field data.

[0024] 2. The dynamic game space of emergency resource supply and demand is constructed by fusing vehicle trajectory, fire fighting facility state and smoke diffusion vector. The system automatically generates the curvature continuous fire truck travel path and the optimal solution set of pressure-flow by real-time deducing the speed game relationship between fire spread and traffic flow. This dynamic mapping mechanism not only avoids the path conflict risk of fixed preplan, but also effectively suppresses the turbulent disturbance caused by high-pressure water mist injection through the closed-loop feedback of pump pressure and smoke exhaust speed.

[0025] 3. The distributed nodes trigger cross-node data repair and model migration based on the spatial symmetry characteristics of LED array in local pollution or hardware failure. By reusing the optical flow information of traffic cameras to reverse environmental parameters, the system maintains the integrity of the reference data while achieving dynamic load balancing of computing tasks. This self-healing mechanism ensures the continuity of risk deduction process under single point failure, forming a robust control network with spatial and temporal redundancy characteristics.

[0026] 4. In view of the monitoring reference drift caused by tunnel wall pollution, optical compensation is realized under the premise of no hardware modification through multi-frame reference repair of adjacent light sources and dynamic triggering mechanism of humidity sensing. The repaired video data reversely optimizes the vehicle trajectory prediction model, forming a closed-loop enhanced loop of perception-decision. This mechanism not only maintains the long-term stability of optical flow field inversion, but also synchronously improves the accuracy of derivative functions such as license plate recognition. In the heavy vehicle congestion scene, the system constructs the collaborative injection network of mobile fire extinguishing device and fixed fire hydrant through the joint optimization of vehicle-mounted water tank capacity attenuation model and path curvature constraint. This dynamic adaptation mechanism avoids the static resource allocation limitations of traditional preplan, maintains the water flow coverage density, and optimizes the fire extinguishing agent delivery efficiency and fire truck mobility to the optimal balance. BRIEF DESCRIPTION OF DRAWINGS

[0027] Fig. 1 The structural block diagram of the tunnel intelligent fire safety early warning system of the present application;

[0028] Fig. 2 The dynamic collaborative response timing diagram based on risk parameters in the tunnel intelligent fire safety system of the present application;

[0029] Fig. 3 The node collaboration and fault tolerance mechanism based on edge computing network of the present application.

[0030] The objectives, functional characteristics and advantages of the present application will be further explained in conjunction with the embodiments with reference to the accompanying drawings. DETAILED DESCRIPTION

[0031] It should be understood that the specific embodiments described herein are merely exemplary and are not intended to limit the present application.

[0032] The embodiment of the present application provides a highway tunnel intelligent fire safety early warning system, and the system comprises:

[0033] A data acquisition and preprocessing module is used for acquiring multi-source heterogeneous data in a tunnel in real time, wherein the multi-source heterogeneous data at least comprises video monitoring data, environmental sensor data and vehicle positioning data; vehicle motion trajectory and smoke texture features are extracted from the video monitoring data, time-space interpolation compensation is performed on the environmental sensor data, and when a video pixel change rate ΔP in the video monitoring data and a carbon monoxide concentration change gradient cross-modal early smoldering feature recognition is performed, and when early smoldering is determined.

[0034] A dynamic risk assessment module is in communication connection with the data acquisition and preprocessing module, and is used for constructing a fire risk field model based on time-space coupling of multi-source heterogeneous data, and specifically comprises the following steps: generating a risk propagation vector field based on the vehicle motion trajectory and the smoke texture features, generating a dynamic accessibility heat map in combination with static parameters such as positions of fire hydrants in the tunnel and states of smoke exhaust ports and real-time vehicle distribution, and establishing a game model of fire spread speed and traffic flow speed to predict a best rescue time window.

[0035] An emergency resource elastic adaptation module is in communication connection with the dynamic risk assessment module, and is used for automatically switching display strategies of an induction screen according to an intensity gradient of the fire risk field model, and forming a negative feedback loop of fire pump pressure regulation and smoke exhaust fan speed to suppress smoke backdraft.

[0036] An edge computing node is distributedly arranged in the tunnel, and is used for executing computing tasks of the data acquisition and preprocessing module and the dynamic risk assessment module, and realizing adaptive migration learning of model parameters through topology perception, so that the computing tasks of adjacent nodes can be seamlessly taken over in the case of single point failure.

[0037] Preferably, the data acquisition and preprocessing module further comprises an optical flow compensation mechanism, and specifically: the optical flow field of a fixed light source in a tunnel side wall monitoring video is used to inverse a real-time airflow velocity field, and a time-space mapping relationship between a smoke diffusion path and a vehicle trajectory is dynamically corrected based on the real-time airflow velocity field to compensate for nonlinear time-space distortion caused by airflow disturbance.

[0038] Preferably, the data acquisition and preprocessing module further comprises a local pollution perception and image self-repair mechanism, which specifically comprises: identifying the pollution characteristics of the fixed light source surface based on a lightweight model to generate a pollution area mask; using the spatial symmetry of adjacent fixed light sources and the time light consistency to construct a multi-frame reference repair matrix of the pollution area; and when the pollution area ratio S polluted / S total > θ2 and lasts for a preset time T polluted , activating the repair algorithm, and suppressing non-emergency repair according to the humidity data in the tunnel, wherein S polluted is the area of the pollution area, and S total is the total area of the light source.

[0039] Preferably, the local pollution perception and image self-repair mechanism further comprises: when it is identified that the pollution area overlaps with the predicted path of the vehicle, the area is repaired preferentially; and dynamically adjusting the sampling frequency of the repair area based on the trajectory prediction speed; and the repaired optical flow data is fed back to optimize the vehicle trajectory prediction model.

[0040] Preferably, the game model of fire spread speed and traffic flow speed established by the dynamic risk assessment module also considers the influence of heavy truck congestion on the route of the fire truck, and combines the cooperative scheduling ability of the vehicle-mounted mobile fire extinguishing device and the fixed fire hydrant to generate an optimal fire extinguishing resource scheduling scheme.

[0041] Preferably, the dynamic risk assessment module introduces a water tank capacity attenuation factor when generating the optimal fire extinguishing resource scheduling scheme, and calculates the Pareto front solution set of the optimal path and injection pressure in real time.

[0042] Preferably, the emergency resource elastic adaptation module embeds a second derivative feedback term of the fire water pressure sensor in the exhaust fan control loop when forming a negative feedback loop of fire water pump pressure regulation and exhaust fan speed, to dynamically suppress air turbulence caused by sudden changes in water pressure.

[0043] Preferably, the dynamic risk assessment module aligns the vehicle motion trajectory in the video stream with the smoke diffusion direction in microsecond-level time sequence when generating the risk propagation vector field, to compensate for the time sequence misalignment of the smoke diffusion path and the video motion trajectory caused by tunnel air flow disturbance.

[0044] Preferably, the dynamic risk assessment module introduces a fire truck turning radius constraint when calculating the time-space game equilibrium point of the evacuation path and the optimal route of the fire truck, to generate a curvature-continuous optimal path.

[0045] Preferably, the edge computing nodes periodically broadcast their respective local risk assessment results and resource state information to realize collaborative perception of the overall risk situation of the tunnel and distributed collaborative scheduling of emergency resources.

[0046] In the expressway two-way tunnel scene, for early smoldering identification, risk field construction, resource dynamic scheduling, and edge fault tolerance, the system is deployed in a distributed edge node about one-third away from the tunnel entrance. Relying on local high-definition cameras, environmental sensor arrays, and edge computing modules, it realizes rapid identification of early signs of fire and adaptive response of resources. In the early smoldering identification, the system uses a multi-source heterogeneous data collaborative analysis mechanism. Specifically, video monitoring data is extracted by background modeling and foreground separation method to obtain local temporal brightness change rate, forming real-time video pixel change rate sequence. At the same time, local carbon monoxide concentration data obtained by sensors is processed by three-dimensional spline interpolation and gradient regression method to obtain spatially consistent carbon monoxide concentration change gradient field. The system uses a combination of criteria to perform point-to-point multiplication operation on real-time video pixel change rate and carbon monoxide concentration change gradient to evaluate the synchronicity of trace smoldering signs. When the product value continuously exceeds the set threshold, the system triggers early fire response. The threshold is set by referring to the cross-statistical distribution of gas release amount and visible light disturbance level in the early stage of smoldering in similar tunnel cases in the past three years, and is set as a quantile value that balances sensitivity and false alarm rate to ensure that the system has early response capability and can effectively suppress false alarms.

[0047] For the coupling modeling of smoke diffusion and vehicle trajectory, the system synchronously extracts vehicle moving trajectory and smoke texture boundary change information at the video pixel level, combines the airflow velocity field obtained by tunnel sidewall flow inversion, and constructs a risk propagation vector field containing multiple time points and multiple spatial levels in each processing period; the vector field takes the vehicle prediction path as the reference coordinate system, the smoke edge expansion trend as the vector direction, and introduces an airflow disturbance correction factor to dynamically compensate for nonlinear disturbances, thereby generating a risk migration path graph with time consistency and spatial accessibility; on this basis, the system further constructs a dynamic accessibility heat map to describe the spatial distribution of the influence of the fire source on different vehicles within the current time to the future estimated time window; the heat map is generated by superimposing the risk propagation vector and the real-time vehicle position, and combined with the static and semi-static parameters such as fire hydrant location, smoke exhaust port state, and traffic density to construct a weighted accessibility index for subsequent resource scheduling judgment. At the same time, to achieve efficient scheduling and adaptation of fire resources, the system introduces a game model of traffic flow and fire spread speed, in which the traffic flow speed is calculated from the average instantaneous speed of adjacent vehicles, and the fire spread speed is estimated from the smoke texture expansion speed and local temperature rise speed; both are used as opponents in the model, and the goal is to solve the shortest response time window of the fire vehicle on different paths; the game model is solved based on differential dynamic programming iteration, and the vehicle passable time period and the expansion boundary of the fire source influence area are updated in each iteration to finally solve the shortest response path set. In the path feasibility analysis, the water tank capacity decay factor of the vehicle-mounted fire extinguishing device is introduced, that is, as the driving distance increases, the carrying capacity of the fire extinguishing agent decreases linearly, so the system normalizes the decay curve in path planning to form a time-dose compound optimization target, and selects the optimal Pareto solution set as the scheduling suggestion.

[0048] In terms of environmental perception reliability, the system models and identifies fixed light source surface pollution. First, by periodically collecting light source image frames, it extracts brightness distribution and edge sharpness indicators, combines a lightweight pollution identification module based on convolutional neural networks to judge the degree of light source surface pollution, and generates a pollution mask image; when the pollution mask area overlaps with the vehicle trajectory prediction path, the system preferentially activates the image repair module. This module uses the symmetry of adjacent light sources in spatial position and the consistency of historical frame illumination to construct a multi-frame reference image set, and implements pollution area repair through weighted averaging and image reconstruction algorithms; if the pollution area ratio exceeds five percent of the total light-emitting area and the duration exceeds the set threshold (which is based on daily pollution evolution rate statistics at monitoring points and is set to more than 5 minutes), the system automatically triggers the repair action. The repaired image is fed back to the vehicle trajectory prediction module to update its prediction model parameters, preventing the prediction path from deviating due to perception errors.

[0049] In the linkage control of fire water pump pressure and smoke exhaust fan speed, the system constructs a negative feedback closed loop. The specific mechanism is: real-time monitoring of fire water pressure sensor data, extracting its second derivative to judge the pressure mutation trend; if the pressure increases rapidly, the system immediately reduces the speed of the smoke exhaust fan to suppress the local air turbulence backwash phenomenon caused by high-pressure water column; the weight of this feedback term is derived from the fluid mechanics model of the tunnel cross section, by adjusting the feedback sensitivity coefficient, the control system can quickly respond and avoid frequent oscillation, so as to realize stable air-water linkage control. At the same time, the system is deployed between the edge computing units of each node in the tunnel, and the state information is exchanged through the periodic broadcast mechanism. The local risk assessment results, resource remaining state and node running situation are exchanged every ten seconds. When any node appears hardware failure or computing resource overload, the adjacent node receives its stop broadcasting signal or state abnormal flag, and starts the parameter migration mechanism. Through the local cache and the central scheduling table, the model parameters are synchronized to take over. During the takeover process, the system selects the node with the lightest computing load as the takeover party according to the adjacency matrix and historical cooperation records, ensuring uninterrupted data processing and risk assessment, and realizing dynamic self-adaptation of disaster tolerance.

[0050] Embodiment 2: This embodiment combines Figs. 1 to 3 to realize a highway tunnel intelligent fire safety early warning system. As Fig. 1 shown, the system first receives video streams and sensor signals from the tunnel environment. These data are input into the data acquisition and preprocessing module. The data collected by the module include video monitoring data, environmental sensor data and vehicle positioning data. The video pixel change rate and carbon monoxide concentration gradient are analyzed to realize early smoldering feature recognition. On this basis, the system transmits the processed data to the dynamic risk assessment module to generate risk assessment parameters and risk field model gradient. This module uses vehicle trajectory and smoke diffusion path to construct fire risk propagation vector field, and considers static parameters such as fire hydrant and smoke exhaust port and dynamic distribution of vehicles to generate dynamic heat map. At the same time, based on the game model of fire spread speed and traffic flow speed, the rescue window is deduced. The generated risk assessment results are transmitted to the emergency resource elastic adaptation module. This module automatically adjusts the induction, fire extinguishing and smoke exhaust strategy according to the risk intensity, and outputs control instructions to the induction screen, pressure regulation instructions to the fire water pump and speed control instructions to the smoke exhaust fan, realizing intelligent regulation and linkage response of resources. At the same time, the edge computing nodes in the system work with the data acquisition and preprocessing module to process the optical flow compensation data, ensuring the stability and accuracy of the monitoring data, and taking over the task when the computing node fails, maintaining the continuity of system operation. Finally, based on the fire extinguishing resource scheduling scheme, the system can also control the path and injection strategy of the vehicle-mounted fire extinguishing device, realizing the cooperation of fixed facilities and mobile fire extinguishing equipment.

[0051] As Fig. 2As shown in the figure, first, the monitoring terminal pushes the environmental data (temperature / carbon monoxide / smoke) to the edge node; the edge node performs spatiotemporal data alignment to achieve accurate temporal and spatial matching of the video and sensor data; the edge node completes the risk parameter transmission (the product of the pixel change rate and the concentration gradient > threshold 1), indicating that the early smoldering signs of the fire have been identified; then, the decision center performs game strategy calculations based on the received risk parameters to construct a dynamic game model of the fire spread speed and traffic flow speed; according to the calculation results, the decision center issues a control instruction set (pressure 80 to 150 MPa) to the actuator; to ensure the accuracy of execution, the edge node issues a status feedback request, thereby completing the real-time working condition confirmation at the monitoring terminal; then, the actuator fine-tunes the execution parameters according to the current feedback to ensure that the pressure and smoke exhaust control accurately match the actual conditions; finally, the actuator returns an execution effect report (flow deviation is less than 5%) to the decision center to verify the system response effect and form a closed-loop control.

[0052] like Fig. 3 As shown, the edge computing network consists of nodes 1, 2, 3, and 4. All nodes can interact with the data acquisition module and support collaborative operations with the dynamic risk assessment module. During system operation, node 1 synchronizes model and risk field parameters to node 3 through parameter migration. Node 1 also periodically broadcasts its local state and resource information to node 2. When a single point of failure is detected, node 2 takes over the task, ensuring uninterrupted core functionality. Nodes 3 and 4 share state information based on a collaborative perception mechanism, supporting model synchronization and fault diagnosis. During task switching, node 4 can seamlessly take over some of node 3's computational tasks, enabling dynamic migration and adaptive balancing of computational load during fault recovery. This entire edge computing framework ensures that the system can maintain the steady state of the edge computing network through node collaboration and task takeover mechanisms in the event of emergencies or hardware anomalies, ensuring the continuous execution of dynamic risk assessment and effective support for the data acquisition module.

[0053] In Example 3, for example, in a two-way six-lane highway tunnel located in mountainous terrain, to deal with the potential smoldering hazard during the night truck traffic peak, the system is deployed in a distributed edge node about 800 meters from the north entrance of the tunnel, which is equipped with high-definition video acquisition devices, multiple environmental parameter sensor arrays, high-speed computing modules, and local induction control actuators. In the video monitoring part, the system uses video cameras fixedly installed on the tunnel wall facing the lane to collect image frame sequences in real time and update the cache at a frequency of 25 frames per second. The image processing module uses a frame difference algorithm based on background modeling to extract the brightness change rate of each pixel on the time axis. In the calculation formula, the time window is selected as 2 seconds, the pixel change rate is based on the average brightness increment between frames, and the root mean square of pixel grayscale change is used as a local disturbance sensitivity indicator. Then, the global pixel change rate estimate value at the current time is obtained by clustering and summing on the image region. To avoid interference caused by single-point strong light or obstruction, the system divides the image into blocks and introduces a regional smoothing filter mechanism. After removing the influence of abnormal noise, the stable video pixel change rate time series is output. At the same time, the environmental parameter acquisition module collects data from multiple carbon monoxide concentration sensors installed on the tunnel ceiling and both sides. The data is organized using a three-dimensional spatial grid structure. The system constructs a dynamic carbon monoxide concentration change gradient field using spatial interpolation and temporal fitting methods. The specific method is as follows: In the three-dimensional spatial grid constructed with sensors as nodes, each node uses a polynomial curve fitting of its concentration change trend in the time series within a sliding window, and uses bilinear interpolation in different directions to obtain the change rate in space, which is combined into a local concentration change gradient vector set in a unit volume. The system samples the video pixel change rate and the carbon monoxide concentration change gradient field every 1 second, and calculates their spatial consistency product. In the specific implementation, the corresponding monitoring areas in the video image and the nodes in the environmental parameter grid are paired through the coordinate mapping mechanism, and the point-to-point product calculation is performed. When the product result continuously exceeds the response threshold (which is based on the statistical analysis of the cross-sectional features of the first 15 seconds of 100 smoldering cases in the tunnel) for 5 seconds, the system's internal smoldering warning mechanism is triggered.

[0054] In terms of smoke propagation and vehicle trajectory coupling modeling, the system reconstructs the air flow velocity field based on the light flow inversion mechanism. Specifically, fixed light sources installed on the side walls of the tunnel are selected as reference points. By detecting the shape distortion and brightness distribution changes of these light sources in the video, the system uses the sparse optical flow method combined with template matching technology to extract the optical flow vector. After dimension reduction and spatial fitting of the optical flow field, it is mapped to the air flow velocity estimation value on the cross section. Using this velocity field, the system performs nonlinear correction on the risk propagation path composed of vehicle trajectories and smoke boundaries. The vehicle trajectory is converted into an accessibility path extending along the air flow direction, and a vector layer stacking structure is established at multiple time points. The vehicle trajectory acquisition module is based on target detection and tracking technology in video images. It uses a lightweight convolutional neural network model to realize vehicle body detection and realizes continuous tracking through inter-frame feature matching. The vehicle position point set is projected into the corrected air flow coordinate system, and after superimposing the smoke boundary layer, a complete risk propagation vector field is formed. In each evaluation period, the system constructs a weighted accessibility heat map based on the risk vector field, vehicle distribution map, and static layout of fire fighting facilities, including fire hydrant, smoke exhaust port position and state. The generation method of the heat map includes: calculating the propagation path length from the current position of each vehicle to the potential fire source point, the average risk gradient of the region passed through, and the current fire fighting resource accessibility index. These factors are combined as weights as the basis for coloring the heat map. Further, to achieve emergency resource scheduling optimization, the system introduces a dynamic game model of traffic flow speed and fire spread speed. The traffic flow speed is obtained by continuously tracking the speed vector of the vehicle and averaging it in the lane area. The fire spread speed is derived by combining the smoke boundary expansion rate and the local heat accumulation estimate. Both are used as the limiting condition of path feasibility in the model. The path optimization algorithm uses a variable step size difference method to derive the rescue time window, and combines the vehicle turning restriction condition and the fire truck turning radius model to generate a set of optimal paths with continuous curvature. When generating the path set, the system considers the water tank capacity decay coefficient of the vehicle, assumes that the extinguishing agent decreases linearly with distance, and combines the location of the pipe network supply point to generate the extinguishing coverage area, thereby forming a Pareto optimal path and pressure allocation solution set.

[0055] In the pollution perception and image repair mechanism, the system automatically analyzes every 60 seconds whether there is brightness abnormality or edge blur in the light source area in the monitoring image. If the brightness decrease amplitude exceeds 15% and the edge gradient is lower than the set clarity threshold for 5 consecutive periods, the system automatically marks it as a pollution area. Combined with time consistency and adjacent light source symmetry characteristics, the system extracts the corresponding area segment from the adjacent light source historical image using multi-frame fusion reconstruction technology to generate a repaired image to replace the polluted area. This mechanism executes immediately when the pollution area coincides with the vehicle predicted path and feeds back the repair result to the vehicle trajectory prediction module to update the prediction offset correction coefficient, avoiding prediction distortion caused by image abnormalities.

[0056] In the water pump pressure and smoke exhaust fan linkage control mechanism, the system collects real-time fire water pressure data, and calculates the second-order change value in the 2-second sliding window to determine whether there is a tendency of pressure mutation. When the second-order change value exceeds the threshold value (which is obtained by tunnel fluid model simulation and is preset to be different in different cross-section structures), the system adjusts the speed of the smoke exhaust fan in real time through the proportional control mechanism to suppress the backflow turbulence effect caused by the high-pressure water column. In the control logic, the second-order change value is used as the feedback input of the fan speed regulator, combined with the current air humidity and temperature indicators to form a comprehensive response factor to ensure a balance between sensitivity and stability. At the same time, to ensure the continuous operation and fault tolerance of the system, the system establishes a periodic state broadcast mechanism between the edge computing modules at each node in the tunnel. Every 10 seconds, it synchronizes local risk assessment results, device status, and computing load information. When a node does not broadcast or send an abnormal state flag according to the period, the adjacent node immediately starts the model parameter migration mechanism according to the topology table and load priority to seamlessly transfer the current computing task to the node with the smallest load and historical stability record, realizing function takeover. The takeover process includes parameter state reconstruction, interrupted frame recalculation, and interface address mapping replacement, etc. steps to ensure that the overall evaluation of the system does not interrupt and the response strategy does not stop during task conversion, improving the robustness and reliability of the system in complex long tunnel environments. All belong to the extended implementation mode known to those skilled in the art.

[0057] In the risk propagation vector field construction, the system takes a sequence of video frames with a duration of five seconds as the input window, uses a pixel trajectory extraction algorithm based on foreground motion region segmentation to identify the shape change trend of the vehicle edge and the spatial expansion path of the smoke texture in each frame, and forms a texture gradient map after non-maximum suppression of the extracted edge information. Then, a local maximum value tracking algorithm is used to generate candidate boundary lines. The system establishes an inter-frame mapping relationship between the boundary lines and the identified vehicle contour through a feature point tracking algorithm. On this basis, a Kalman filter is used to construct a vehicle prediction trajectory curve, and a reference coordinate system is defined with the curve as the main axis. The direction of each smoke boundary expansion vector is determined by the boundary difference vector between the current frame and the previous frame, and the length is weighted according to the local texture gray level change rate. The final generated vector field covers the entire image effective area. To correct the offset error of the above vector field under air flow disturbance, the system estimates the air flow speed on the current cross section in real time based on the optical flow inversion mechanism of the fixed light source area. In specific implementation, the system selects two groups of fixed light sources on the tunnel wall surface with equal distance, extracts their brightness change and edge displacement information in consecutive frames through template matching and optical flow field sparse estimation method. The system uses Newton interpolation method to construct a brightness change surface, and converts the gradient component in the horizontal direction into an air flow speed estimate. This speed value is spatially expanded through bilinear interpolation to form a two-dimensional air flow speed field on the cross section. This speed field then acts on the smoke expansion vector field, the vector direction is rotated to compensate for the dominant direction of the local air flow, and the length is scaled according to the air flow speed distribution to form the corrected risk propagation vector layer.

[0058] The system superimposes the vector layer and the vehicle prediction trajectory to construct a risk propagation path set, and calculates the area covered by the risk propagation along each vehicle trajectory direction in a unit time step from the initial point of the fire source. Each vehicle path point will be assigned a propagation weight value according to the vector density, superposition direction consistency and historical risk record of the location, to reflect the intensity distribution of the risk field. The risk values of all path points are processed through Gaussian blur to form a risk heat map, providing a quantitative basis for subsequent emergency resource scheduling. In the pollution image identification and repair mechanism, the system analyzes the brightness histogram of the light source area in the image every ten seconds, extracts the brightness peak, edge contrast and definition indicators, and if any indicator decreases by more than 15% in three consecutive periods, the system determines that the area may be contaminated. At this time, the image repair module is started, the symmetric region fragments are extracted from the historical images of the adjacent light source areas, and the pixel-level repair is performed using a weighted fusion algorithm combined with the texture information of the non-polluted area in the current frame. After edge reconstruction and gray balance processing, the repaired image is returned to the vehicle trajectory prediction module to update its background modeling parameters, preventing the accumulation of trajectory errors caused by pollution.

[0059] In the modeling of fire spread and traffic flow game, the system takes the current position, predicted speed and current risk path map of the vehicle as input, constructs the expansion boundary of the fire source influence domain in each evaluation period, and derives the fire spread speed from the average displacement speed of the smoke boundary in consecutive frames and the local temperature sensor incremental curve. The traffic flow speed is calculated from the average instantaneous speed of the vehicle in front. The two form the risk domain boundary and the vehicle passable area, respectively, on the basis of which the path accessibility matrix is constructed. Each candidate path will consider the vehicle turning radius, passing width, fire extinguishing resource delivery capacity and other constraints, and the multi-objective differential evolution algorithm is used to solve the path set with the shortest response time and the smallest water agent attenuation. The water agent attenuation factor is estimated based on the linear function between the vehicle carrying water and the path length, and the water source supply node is generated combined with the fixed fire hydrant distribution, so that the path optimization result has physical realizability. Finally, the edge computing nodes realize model synchronization and task takeover function through 5-second period state broadcast, each node maintains the local topological structure graph and adjacent node parameter cache, when detecting that the adjacent node stops broadcasting or sends an abnormal state marker, the system selects the takeover node according to the historical calculation stability and the current calculation load priority, and completes the seamless migration of the calculation task through the three stages of parameter synchronization, state reconstruction and task switching. The parameter synchronization stage includes the update of risk map layer state, thermal map distribution and vehicle prediction parameters, the state reconstruction stage performs the backfill of incomplete data, and the task switching stage realizes the business continuity guarantee through interface address mapping. The whole mechanism ensures that the overall operation logic of the system is not interrupted when any node fails, and the continuity of risk assessment and control response is guaranteed. All belong to the extended implementation mode known to those skilled in the art.

[0060] It is obvious for those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application. Finally, it should be noted that the above examples are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. A highway tunnel intelligent fire safety early warning system, characterized by: The system comprises: The data acquisition and preprocessing module is used to collect multi-source heterogeneous data in the tunnel in real time. The multi-source heterogeneous data includes at least video surveillance data, environmental sensor data and vehicle positioning data; and extract vehicle motion trajectory and smoke texture features from the video surveillance data, perform spatiotemporal interpolation compensation on the environmental sensor data, and compare the video pixel change rate ΔP in the video surveillance data with the carbon monoxide concentration change gradient in the environmental sensor data. Perform cross-modal early smoldering feature recognition, when When the fire is burning, it is judged as early smoldering; The dynamic risk assessment module communicates with the data acquisition and preprocessing module and is used to build a fire risk field model based on the spatiotemporal coupling of multi-source heterogeneous data. Specifically, it generates a risk propagation vector field based on vehicle motion trajectories and smoke texture characteristics, generates a dynamic accessibility heat map based on the static parameters of fire hydrant locations and smoke exhaust vents in the tunnel, and combines them with real-time vehicle distribution. It also establishes a game model between fire spread speed and traffic flow speed to predict the optimal rescue time window. The emergency resource elastic adaptation module is connected to the dynamic risk assessment module and is used to automatically switch the induction screen display strategy according to the intensity gradient of the fire risk field model, and to form a negative feedback closed loop for fire pump pressure regulation and smoke exhaust fan speed; Edge computing nodes are distributed and deployed within the tunnel. They perform computing tasks for the data acquisition and preprocessing module and the dynamic risk assessment module. They also implement adaptive transfer learning of model parameters through topology awareness and seamlessly take over computing tasks from neighboring nodes in the event of a single point of failure. The data acquisition and preprocessing module also includes an optical flow compensation mechanism. Specifically, it uses the optical flow field of a fixed light source in the tunnel sidewall surveillance video to invert the real-time airflow velocity field. Based on the real-time airflow velocity field, it dynamically corrects the spatiotemporal mapping relationship between the smoke diffusion path and the vehicle trajectory to compensate for the nonlinear spatiotemporal distortion caused by airflow disturbances. When generating the risk propagation vector field, the dynamic risk assessment module aligns the vehicle motion trajectory in the video stream with the smoke diffusion direction at the microsecond level to compensate for the timing misalignment between the smoke diffusion path and the video motion trajectory caused by tunnel airflow disturbances.

2. The highway tunnel intelligent fire safety early warning system according to claim 1 is characterized in that: The data acquisition and preprocessing module also includes local pollution perception and image self-repair mechanism, which includes: identifying pollution features on the surface of fixed light sources based on lightweight models and generating pollution area masks; using the spatial symmetry and temporal illumination consistency of adjacent fixed light sources to construct a multi-frame reference repair matrix for the pollution area; and when the pollution area accounts for S polluted / S total >θ2 and lasts for the preset time T polluted When , the repair algorithm is activated and non-urgent repairs are suppressed according to the humidity data in the tunnel, where S polluted is the area of ​​the polluted area, S total is the total area of ​​the light source.

3. The highway tunnel intelligent fire safety early warning system according to claim 2 is characterized in that: The local pollution perception and image self-repair mechanism also includes: when a polluted area is identified to overlap with the vehicle trajectory prediction path, this area is repaired first; the sampling frequency of the repaired area is dynamically adjusted based on the trajectory prediction speed; and the repaired optical flow data is fed back to optimize the vehicle trajectory prediction model.

4. The highway tunnel intelligent fire safety early warning system according to claim 1 is characterized in that: When forming a negative feedback closed loop between the fire pump pressure regulation and the smoke exhaust fan speed, the emergency resource elastic adaptation module embeds the second-order derivative feedback term of the fire water pressure sensor in the smoke exhaust fan control loop to dynamically suppress the air turbulence caused by sudden changes in water pressure.

5. The highway tunnel intelligent fire safety early warning system according to claim 1 is characterized in that: Edge computing nodes periodically broadcast their respective local risk assessment results and resource status information to achieve collaborative perception of the overall risk situation of the tunnel and distributed collaborative scheduling of emergency resources.

Images