Flue gas front dynamic locking control method for closed traffic barrier structure

Abstract

The invention relates to the technical field of fire engineering and automatic control, and discloses a flue gas front dynamic locking control method for a closed traffic barrier structure, which comprises the following steps: acquiring environmental data in the closed traffic barrier structure through a multi-modal sensor, and fusing data of different modals to calculate the actual position of a flue gas backflow front; determining a target position of a smoke backflow front according to the position of the fire source; the position deviation between the actual position and the target position is calculated, the equivalent leakage area of the closed traffic barrier structure is estimated in real time, and the feedforward compensation amount is generated based on the equivalent leakage area; dynamically adjusting operation parameters of the ventilation equipment according to the position deviation and the feed-forward compensation amount, so that the actual position tends to the target position; when a plurality of potential fire sources are detected, an upstream priority scanning strategy is executed, and the first smoke backflow front distributed in the smoke exhaust direction along the evacuation area is locked as the actual position. According to the invention, the accurate control of the smoke backflow front position is realized, and the safety of the evacuation passage is ensured.

Description

A method for dynamic locking control of smoke front in closed traffic barrier structures Technical Field

[0001] This invention relates to the fields of fire protection engineering and automatic control technology, and in particular to a dynamic locking control method for smoke front in enclosed traffic barrier structures. Background Technology

[0002] With the rapid development of urban transportation infrastructure, enclosed traffic barriers (such as sound barrier tunnels and fully enclosed viaducts) are widely used in urban expressways and highways to reduce noise pollution and improve the driving environment. These structures typically use a steel frame and translucent sound-insulating panels (such as polymethyl methacrylate (PMMA) or polycarbonate (PC) sheets) to form a semi-enclosed passage, providing both sound insulation and light transmission. However, this structural form faces unique challenges in the event of a fire: the melting temperature of the translucent sound-insulating panels is typically 160-200°C, making them prone to softening and detachment under the high temperatures of a fire, which can compromise the airtightness of the passage structure and create large-area, unintended leakage channels.

[0003] For smoke control in enclosed or semi-enclosed spaces, existing technologies mainly rely on critical wind speed control methods widely used in tunnel construction. The core idea of ​​this method is to generate a longitudinal airflow exceeding the critical wind speed (typically 3 m / s) within the passage using a longitudinal jet fan, thereby preventing high-temperature smoke from spreading upstream of the fire source (i.e., smoke backflow or flue gas recirculation). However, existing critical wind speed control methods have the following technical drawbacks: First, the control objective of this method is to maintain a constant longitudinal wind speed, rather than directly controlling the spatial position of the smoke front. This indirect control strategy works when the structure is intact, but it cannot address structural failures caused by the melting of light-transmitting and sound-insulating panels in enclosed traffic barriers. Specifically, when large-area light-transmitting panels detach, numerous air leaks appear in the previously sealed space, causing drastic changes in the system's pressure field. The original wind speed setpoint becomes ineffective, and the smoke front may break through the original defense line and spread towards the evacuation area, but the control system cannot detect this positional change.

[0004] Secondly, existing methods typically rely on a single type of sensor (such as a wind speed sensor or a temperature sensor) for environmental perception. In complex fire conditions, a single sensing mode is prone to failure: for example, in the early stages of a fire or in low-temperature smoke scenarios caused by enhanced ventilation, smoke temperature stratification is not obvious, and temperature gradient-based detection methods will fail; while in dense smoke scenarios, video surveillance systems may be unable to provide effective image information due to smoke obstruction or light interference. This risk of single-point failure makes existing systems insufficiently robust.

[0005] Furthermore, existing methods assume that the airtightness of the passage structure remains unchanged during a fire, thus failing to consider the issue of air leakage compensation due to structural damage. This assumption is reasonable for rigid, non-combustible highway tunnels; however, for enclosed traffic barriers made of fusible materials, the structural integrity changes dynamically as the fire progresses, causing the system's equivalent leakage area to continuously increase. Existing control strategies cannot respond to this disturbance in real time, thus losing effective control over smoke spread.

[0006] Furthermore, in complex scenarios with multiple potential ignition sources, existing technologies lack a clear logic for prioritizing ignition sources. From a fire safety perspective, priority should be given to protecting evacuation routes, i.e., controlling the smoke front closest to the evacuation area; however, existing methods often treat all ignition points equally or only focus on the location with the most intense fire, which may lead to unreasonable situations where evacuation areas are invaded by smoke while the control system is still dealing with downstream secondary ignition points.

[0007] Therefore, there is an urgent need for a new flue gas control method that can directly locate the position of the flue gas front, has multimodal fusion sensing capabilities, can compensate for the impact of structural damage in real time, and prioritize the protection of evacuation routes. Summary of the Invention

[0008] The purpose of this application is to provide a dynamic locking control method for the smoke front in closed traffic barrier structures, in order to solve the problems of indirect control targets, inability to cope with structural damage, insufficient robustness of single sensors, and improper handling logic for multiple fire sources in the prior art, so as to achieve precise control of the position of the smoke backflow front and ensure the safety of evacuation channels.

[0009] To achieve the aforementioned objectives, the present invention provides a technical solution comprising: a dynamic locking control method for smoke fronts in enclosed traffic barrier structures, comprising: acquiring environmental data within the enclosed traffic barrier structure using multimodal sensors, and fusing data from different modalities to calculate the actual position of the smoke recirculation front; determining the target position of the smoke recirculation front based on the location of the fire source; calculating the positional deviation between the actual position and the target position, estimating the equivalent leakage area of ​​the enclosed traffic barrier structure in real time, and generating a feedforward compensation amount based on the equivalent leakage area; dynamically adjusting the operating parameters of the ventilation equipment according to the positional deviation and the feedforward compensation amount, so that the actual position tends towards the target position; wherein, when multiple potential fire sources are detected, an upstream priority scanning strategy is executed to lock the first smoke recirculation front distributed along the evacuation area towards the smoke exhaust direction as the actual position.

[0010] Preferably, the dynamic adjustment of the operating parameters of the ventilation equipment adopts a two-layer control architecture, including: training a strategy matching neural network model based on a digital twin simulation model; using the strategy matching neural network model to generate an initial control command based on the current state vector, and using PID closed-loop control to generate a real-time correction command based on the position deviation; and superimposing the initial control command, the real-time correction command, and the feedforward compensation amount to generate the final operating parameters.

[0011] Preferably, the formula for estimating the equivalent leakage area is: ;in, Let be the equivalent leakage area at time t. This refers to the total airflow entering the enclosed traffic barrier structure. The total exhaust volume, For flow coefficient, The static pressure difference between the inside and outside of the enclosed traffic barrier structure. This refers to air density.

[0012] Preferably, the multimodal sensor includes a distributed fiber optic temperature sensing system and a video surveillance system; the formula for calculating the actual position is: ;in, The actual location, The location of the temperature front is calculated based on the distributed fiber optic temperature sensing system. The visual front position is calculated based on the video surveillance system. Temperature weighting, For visual weights, and + =1, the temperature weight and the visual weight are dynamically adjusted according to the fire conditions.

[0013] Preferably, the dynamic adjustment rules for the temperature weight and the visual weight include: real-time monitoring of the maximum temperature difference within the enclosed traffic barrier structure; when the maximum temperature difference is less than a preset threshold, it is determined to be a low-temperature flue gas condition, and the temperature weight is reduced while the visual weight is increased.

[0014] Preferably, it further includes: real-time monitoring of the image quality of the video surveillance system; when smoke obstruction or light interference is detected, causing the image quality to be lower than a preset standard, the temperature weight is forcibly set to 1 and the visual weight is set to 0.

[0015] Preferably, the upstream priority scanning strategy specifically includes: defining the coordinates of the evacuation area as the origin; scanning along the path from the evacuation area to the smoke exhaust direction; determining the position of the first flue gas recirculation front on the scanning path that meets the anomaly determination conditions as the actual position; the anomaly determination conditions include the temperature gradient exceeding a preset gradient threshold or the visibility being lower than a preset visibility threshold.

[0016] Preferably, the target location is set to K meters upstream of the fire source location, where K is a safety distance parameter that is dynamically adjusted according to the length of the enclosed traffic barrier structure and evacuation requirements.

[0017] Preferably, the training process of the strategy matching neural network model includes: establishing a digital twin simulation model of the closed traffic barrier structure; generating a training dataset containing various fire conditions through computational fluid dynamics simulation; and training the strategy matching neural network model using the training dataset; wherein the state vector includes the fire source location, heat release rate, external environmental wind speed, and external environmental wind direction.

[0018] Preferably, the PID closed-loop control is calculated based on the position deviation, and its output is the frequency adjustment amount of the variable frequency longitudinal jet fan, which is used to correct the initial control command.

[0019] Beneficial Effects: This invention can accurately determine whether the smoke front threatens the evacuation area even when the light-transmitting and sound-insulating panels melt and fall off or the structural airtightness is compromised. It can then dynamically adjust the ventilation power to keep the smoke front in a safe position, thus having the ability to cope with structural failure and significantly improving the safety of enclosed traffic barriers in fire scenarios.

[0020] This invention proactively addresses sudden pressure field changes caused by the melting of light-transmitting and sound-insulating panels. When a sudden increase in the equivalent leakage area is detected, it immediately increases the operating parameters of the ventilation equipment to compensate for pressure losses caused by air leakage, without waiting for the flue gas front position deviation to accumulate. This feedforward compensation mechanism, combined with closed-loop control based on position feedback, constitutes a rapidly responding and highly stable control system, avoiding the lag response problem of traditional pure feedback control under large disturbances.

[0021] This invention significantly enhances the robustness of smoke front detection by integrating data from a distributed fiber optic temperature sensing system and a video surveillance system, and dynamically adjusting their weights according to fire conditions. Under low-temperature smoke conditions, the system automatically reduces the temperature weight and increases the visual weight, using the sensitivity of the visual sensor to smoke boundaries to compensate for the lack of clear temperature stratification. Conversely, when smoke obstruction or light interference causes the video surveillance system to fail, the system forcibly switches to the temperature sensor, ensuring that at least one reliable sensing method continues to operate. Attached Figure Description

[0022] Figure 1 is a flowchart illustrating a preferred embodiment of the dynamic locking control method for smoke fronts in a closed traffic barrier structure according to the present invention. Detailed Implementation

[0023] The present application will now be described in detail with reference to specific embodiments. It should be understood that the following embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit the scope of protection of the present application. Those skilled in the art can make various equivalent modifications and changes to the implementation methods under the guidance of the present application without departing from the spirit of the present application, and all such modifications and changes should fall within the scope of protection defined by the claims of the present application.

[0024] The flue gas front dynamic locking control method described in this application can be implemented in a control system including a computer processor, memory, and communication interface. This control system is configured to interact with a sensor network and ventilation equipment within a closed traffic barrier structure. Specifically, the control system can be an edge computing controller deployed on-site or a cloud server connected to on-site equipment via a network. To achieve real-time control, an industrial-grade edge controller is preferred to reduce the impact of network latency on control performance.

[0025] As shown in Figure 1, the dynamic locking control method for the flue gas front provided in this application includes the following steps: Step S1: Obtain environmental data within the closed traffic barrier structure through a multimodal sensor, and fuse data from different modes to calculate the actual position of the flue gas recirculation front.

[0026] In this embodiment, the multimodal sensor includes at least a temperature sensor and other auxiliary sensors. The system collects data from different types of sensors and uses a data fusion algorithm to comprehensively determine the current spatial position of the flue gas recirculation front. The design principle of this multimodal fusion is that a single type of sensor may fail or have reduced accuracy under specific operating conditions. For example, under low-temperature flue gas conditions, the temperature sensor may have difficulty identifying the flue gas front; and when obscured by dense smoke, the visual sensor may fail. By fusing multiple sensing modalities, the system can complement each other, improving the robustness and accuracy of detection.

[0027] Specifically, the system first collects raw data from various sensors, including but not limited to temperature data, video image data, pressure data, and wind speed data distributed longitudinally along the closed traffic barrier. Then, the system preprocesses this data, such as filtering and denoising, and coordinate mapping. Finally, based on the current operating conditions, the system dynamically adjusts the fusion weights of different modal data to calculate the actual position of the flue gas recirculation front. This actual position is typically expressed as the longitudinal distance from a reference point on the closed traffic barrier (such as the starting point of the evacuation area), in meters.

[0028] Step S2: Determine the target position of the flue gas recirculation front based on the location of the fire source.

[0029] The core of this step lies in transforming the control objective from the abstract "preventing smoke backflow" to the concrete "locking the smoke front in a safe location." The system first needs to identify the location of the fire source within the enclosed traffic barrier. The fire source location can be determined in several ways: firstly, based on temperature sensor data, identifying the temperature peak point or the location with the fastest temperature rise rate; secondly, combining the location of the flame or smoke center identified by the video surveillance system; and thirdly, combining the trigger point information of the fire alarm system.

[0030] Once the location of the fire source is determined, the system can calculate the target location of the smoke backflow front. This target location is set according to the principle of "ensuring the safety of evacuation routes," and is typically set at a certain distance upstream of the fire source. The selection of this distance requires comprehensive consideration of several factors: the speed of movement of evacuees, the speed of smoke spread, and the length of enclosed traffic barriers. By controlling the smoke front upstream of the fire source, a smoke-free safe passage can be ensured between the fire source and the evacuation area, buying time for personnel to evacuate.

[0031] Step S3: Calculate the positional deviation between the actual position and the target position, estimate the equivalent leakage area of ​​the closed traffic barrier structure in real time, and generate a feedforward compensation amount based on the equivalent leakage area.

[0032] The calculation of position deviation is a typical closed-loop control concept. The system compares the actual position of the flue gas recirculation front with the target position to obtain the position deviation. This position deviation reflects the gap between the current control effect and the desired control effect: when the actual position is downstream of the target position (closer to the evacuation area), it indicates that the flue gas front is spreading upstream, and the system needs to increase the ventilation power; conversely, when the actual position is upstream of the target position, it indicates that the flue gas is being excessively pushed downstream, and the system can appropriately reduce the ventilation power to save energy.

[0033] The equivalent leakage area refers to treating all unintended openings in the enclosed traffic barrier structure (such as gaps created by the detachment of light-transmitting and sound-insulating panels) as a unified virtual area to quantify the degree of air leakage in the structure. The physical meaning of this parameter is that when a light-transmitting and sound-insulating panel detaches due to high temperature, a gap appears in the originally sealed passage, causing airflow generated by the ventilation equipment to escape from the gap, reducing the effective airflow used to propel the flue gas. By estimating the equivalent leakage area, the system can quantify this disturbance factor and take timely compensatory measures.

[0034] The estimation of the equivalent leakage area is based on the principle of mass conservation and orifice flow theory in fluid mechanics. The system monitors the total airflow entering and exiting the enclosed traffic barrier structure, as well as the static pressure difference inside and outside the structure, and uses the orifice flow formula to inversely calculate the equivalent leakage area. This estimation process is performed in real time, and whenever a sudden increase in the equivalent leakage area is detected, the system can determine that a light-transmitting and sound-insulating panel detachment event may have occurred.

[0035] Based on the equivalent leakage area, the system generates feedforward compensation. Feedforward compensation refers to taking control actions to offset a disturbance before it significantly affects the controlled object, based on predictions or estimates of the disturbance. In this application, when the system detects an increase in the equivalent leakage area, it means that more airflow is about to escape from the gap, and the flue gas front may advance upstream. At this time, the system does not need to wait for the flue gas front to actually move (i.e., wait for the position deviation to increase), but immediately increases the operating parameters of the ventilation equipment to compensate for the impending disturbance in advance. This feedforward control mechanism makes the system response faster and its dynamic performance better.

[0036] Step 4: Dynamically adjust the operating parameters of the ventilation equipment according to the position deviation and the feedforward compensation amount, so that the actual position tends to the target position.

[0037] This step involves the generation and execution of control commands. The system comprehensively considers position deviation (from feedback control) and feedforward compensation (from feedforward control) to calculate the adjustment of the operating parameters of the ventilation equipment. The ventilation equipment typically refers to a variable frequency longitudinal jet fan, whose operating parameters include operating frequency and speed. By adjusting the fan's operating frequency, the air volume and air pressure generated by the fan can be changed, thereby affecting the airflow velocity and distribution within the enclosed traffic barrier.

[0038] The system's adjustment strategy follows this logic: when the positional deviation is positive (the flue gas front is too close to the evacuation area) or the equivalent leakage area increases, the system increases the operating parameters of the ventilation equipment, raising the longitudinal air velocity to push the flue gas front downstream; conversely, when the positional deviation is negative (the flue gas front is excessively pushed downstream) and the equivalent leakage area is stable, the system can appropriately reduce the operating parameters to avoid energy waste caused by over-ventilation and excessive heating in the downstream area. Through this dynamic adjustment, the system gradually brings the actual position of the flue gas return front closer to the target position and maintains stability near the target position.

[0039] Multi-fire source scenario handling: When multiple potential fire sources are detected, an upstream priority scanning strategy is executed to lock the first smoke return front distributed along the evacuation area towards the smoke exhaust direction as the actual location.

[0040] In real-world fire scenarios, there may be multiple ignition points or multiple temperature peaks caused by the spread of fire. If the system simply uses the point with the highest temperature or the highest smoke concentration as the control target, it may lead to evacuation routes being invaded by smoke from secondary fire points while the control system is inappropriately dealing with the primary fire point downstream. The upstream priority scanning strategy proposed in this application embodies the fire safety concept of "protecting personnel evacuation is more important than protecting structural facilities."

[0041] Specifically, the system first defines the evacuation area as the origin of the reference coordinate system, and then scans the sensing data point by point along the path from the evacuation area towards the smoke exhaust direction. The system does not focus on which fire point has the highest temperature or the highest smoke concentration, but rather seeks the first anomaly encountered along the longitudinal direction from the evacuation area. The location of the smoke recirculation front corresponding to this first anomaly point is the system's control target. The advantage of this design is that as long as the upstream smoke front (closest to the evacuation area) is controlled, the evacuation route can remain smoke-free, and even if a larger fire exists downstream, it will not threaten the safety of personnel evacuation.

[0042] Example 2 This example is based on Example 1.

[0043] The dynamic adjustment of the ventilation equipment's operating parameters adopts a two-layer control architecture, including: training a strategy matching neural network model based on a digital twin simulation model; using the strategy matching neural network model to generate initial control commands based on the current state vector, and using PID closed-loop control to generate real-time correction commands based on the position deviation; and superimposing the initial control commands, the real-time correction commands, and the feedforward compensation to generate the final operating parameters.

[0044] The design philosophy of this two-tier architecture lies in combining the advantages of artificial intelligence methods and classical control theory. The role of the policy matching neural network model is rapid response: when a fire occurs, the system needs to determine the initial operating strategy of the fans within a very short time (e.g., seconds). Because the smoke flow in enclosed traffic barriers is affected by various factors (fire source location, heat release rate, external ambient wind, etc.), traditional control methods struggle to quickly find the optimal initial parameters. Through extensive offline simulation training, the policy matching neural network model learns the mapping relationship between various operating conditions and optimal control parameters, enabling it to provide reasonable initial control commands within milliseconds.

[0045] However, while neural network models offer fast responses, their outputs exhibit some uncertainty and struggle to handle real-time disturbances. Therefore, this application introduces PID closed-loop control for fine-tuning, building upon the coarse-tuning of the neural network. The PID controller receives the position deviation of the flue gas recirculation front as input and outputs real-time correction commands. The advantages of PID control lie in its mature theory, strong stability, and good interpretability. Through its proportional (P), integral (I), and derivative (D) components, the PID controller effectively eliminates position deviations: the proportional component provides immediate response, the integral component eliminates steady-state errors, and the derivative component predicts and dampens the deviation's changing trend.

[0046] The final operating parameters are composed of three superimposed parts: initial control commands (from the neural network, providing a coarse control reference), real-time correction commands (from the PID controller, eliminating deviations), and feedforward compensation (from ELA estimation, offsetting structural damage disturbances). This combined control architecture of "neural network + PID + feedforward" ensures both rapid response and stability and disturbance rejection capability, which is the innovation of the control strategy in this application.

[0047] In implementation, a digital twin simulation model refers to a virtual model built for a specific enclosed traffic barrier structure. This model can accurately simulate the geometry, ventilation equipment configuration, and environmental conditions of the real barrier. Using computational fluid dynamics (CFD) simulation software (such as ANSYS Fluent and OpenFOAM), various fire conditions can be simulated on this digital twin model: changing parameters such as the fire source location, heat release rate, and ambient wind speed and direction. For each condition, the optimal combination of ventilation parameters that can successfully control the smoke front is found through multiple trials. A large number of simulation conditions and their corresponding optimal control parameters constitute the training dataset for the neural network.

[0048] The neural network training process employs supervised learning: the input layer receives a state vector (including fire source location, heat release rate, ambient wind speed and direction, etc.), the output layer provides initial control commands (including target location, upstream fan frequency, downstream fan frequency, etc.), and the hidden layer adjusts the weights through backpropagation, gradually bringing the network output closer to the optimal value obtained from simulation. The trained neural network model can quickly map any input state to a reasonable initial control command without requiring complex online optimization calculations, thus meeting the time requirements of real-time control.

[0049] Example 3 This example provides the following formula for estimating the equivalent leakage area: ;in, Let be the equivalent leakage area at time t. This refers to the total airflow entering the enclosed traffic barrier structure. The total exhaust volume, For flow coefficient, The static pressure difference between the inside and outside of the enclosed traffic barrier structure. This refers to air density.

[0050] This formula is based on the law of conservation of mass and orifice flow theory. Under steady-state conditions, the total air volume entering a closed traffic barrier should equal the total air volume exiting the barrier plus the air volume escaping from the vent. The air volume escaping from the vent can be approximated as orifice flow, and its flow rate is proportional to the orifice area and the square root of the pressure difference. By measuring the difference between the entering and exiting air volumes and combining this with real-time monitoring of the static pressure difference, the equivalent leakage area can be calculated.

[0051] Specifically, It can be calculated from the operating parameters and performance curves of the upstream air supply fan, or measured directly by installing an anemometer at the fan outlet; The total air volume is usually calculated by measuring the cross-sectional wind speed using an array of wind speed sensors installed at the smoke exhaust end, and then combining the cross-sectional integral. Real-time monitoring via differential pressure sensor; The flow coefficient is an empirical parameter, typically ranging from 0.6 to 0.65. The specific value can be adjusted based on factors such as the material of the light-transmitting and sound-insulating panel and the shape of the damaged opening. This is the air density, which can be corrected for temperature and air pressure. For air at room temperature under standard atmospheric pressure, it is usually taken as 1.2 kg / m³. 3 .

[0052] The advantages of this formula are: the required input parameters (inlet air volume, outlet air volume, and pressure difference) can all be collected in real time by conventional sensors without the need for additional complex measurement methods; the calculation process is simple and suitable for online real-time calculation; and the output equivalent leakage area has a clear physical meaning, which facilitates subsequent feedforward compensation calculation.

[0053] It should be understood that, in addition to the above-mentioned estimation method based on flow balance, other methods can also be used to estimate the equivalent leakage area. For example, a video surveillance system can be used to detect the detached area of ​​the light-transmitting sound insulation panel through image recognition algorithms and calculate its area; alternatively, an acoustic sensor array can be used to infer the location and size of the leak based on the airflow noise characteristics generated by the leak. This application does not limit the specific estimation method for the equivalent leakage area; as long as the degree of structural damage can be quantified in real time, it falls within the protection scope of this application.

[0054] Example 4 This example provides a specific example of temperature-visual multimodal fusion.

[0055] The multimodal sensor includes a distributed fiber optic temperature sensing system and a video surveillance system; the formula for calculating the actual position is: ;in, The actual location, The location of the temperature front is calculated based on the distributed fiber optic temperature sensing system. The visual front position is calculated based on the video surveillance system. Temperature weighting, For visual weights, and + =1, the temperature weight and the visual weight are dynamically adjusted according to the fire conditions.

[0056] The working principle of a distributed fiber optic temperature sensing system (DTS) is based on the Raman scattering effect. An optical fiber is laid longitudinally along a closed traffic barrier. A light pulse emitted by a laser propagates through the fiber and undergoes Raman scattering with the fiber molecules. The frequency shift and intensity ratio of the scattered light are related to the local temperature. By demodulating the scattered light signal, the temperature distribution at various points along the fiber can be obtained, with a spatial resolution typically reaching 1-5 meters and a temperature accuracy of 1°C. Based on the temperature distribution curve T(x) obtained by DTS, the system calculates the temperature gradient. The location where the gradient exceeds a preset threshold is identified as the temperature front. The temperature front typically corresponds to the flue gas front, because the flue gas temperature is significantly higher than the ambient temperature, and there is a noticeable temperature jump at the interface between the flue gas and the ambient air.

[0057] Video surveillance systems typically consist of multiple cameras deployed along a closed traffic barrier. The system runs a video smoke recognition algorithm to analyze images captured by the cameras and identify smoke areas. This video smoke recognition algorithm can be based on traditional image processing methods (such as edge detection and texture analysis) or deep learning methods (such as semantic segmentation using convolutional neural networks). After identifying the smoke area, the system extracts the pixel coordinates of the smoke's foreground and converts them into actual spatial coordinates within the closed traffic barrier using the camera's calibration parameters, thus obtaining the visual foreground position. .

[0058] Temperature sensing and visual sensing each have their advantages and disadvantages: temperature sensing is highly reliable in high-temperature smoke scenarios, but in low-temperature smoke conditions (such as early-stage fires or cold smoke caused by enhanced ventilation), temperature stratification is not obvious, making it difficult to accurately locate the smoke front; visual sensing is sensitive to smoke morphology and can identify smoke even at low temperatures, but in dense smoke scenarios, the camera may fail to acquire effective images due to smoke obstruction or light interference. Therefore, this application adopts a weighted fusion approach, dynamically adjusting the weights of the two sensing modes according to different operating conditions to achieve complementary advantages.

[0059] The dynamic adjustment rules for the fusion weights are as follows: 1. For low-temperature flue gas conditions: The dynamic adjustment rules for the temperature weight and the visual weight include: real-time monitoring of the maximum temperature difference within the closed traffic barrier structure; when the maximum temperature difference is less than a preset threshold, it is determined to be a low-temperature flue gas condition, the temperature weight is reduced and the visual weight is increased, and the preset threshold can be set to 50°C.

[0060] Maximum temperature difference This refers to the difference between the highest temperature monitored inside a closed traffic barrier and the ambient temperature. This parameter reflects the fire intensity and smoke temperature. When the temperature is significantly higher (e.g., exceeding 50°C), it indicates a high flue gas temperature and significant temperature stratification. The temperature sensor can accurately identify the flue gas leading edge, at which point the system increases the temperature weighting. ;when When the temperature is low (e.g., below 50°C), it indicates that the flue gas temperature is low or the ventilation effect is good, resulting in the flue gas being diluted and cooled, and the temperature stratification is not obvious. In this case, the system reduces the temperature weight. Increase visual weight accordingly They rely more on visual sensors to identify smoke boundaries.

[0061] The selection of the preset threshold of 50°C is based on engineering experience: generally, when the temperature difference between flue gas and ambient temperature reaches 50°C or more, the temperature gradient is sufficiently significant, and temperature gradient-based detection methods have high reliability; however, when the temperature difference is below 50°C, the accuracy of temperature detection methods decreases, requiring the introduction of other sensing methods for assistance. It should be understood that this threshold is not an absolute value and can be adjusted according to specific application scenarios, sensor performance, and other factors. For example, it can be set to 40°C, 60°C, etc., and this application does not impose any restrictions on this.

[0062] 2. For visual sensor failure scenarios: In some preferred embodiments, the method further includes: real-time monitoring of the image quality of the video surveillance system; when smoke obstruction or light interference is detected, causing the image quality to be lower than a preset standard, the temperature weight is forcibly set to 1 and the visual weight is set to 0.

[0063] Image quality monitoring can be achieved in several ways: Firstly, objective indicators such as contrast, sharpness, and signal-to-noise ratio can be calculated. When these indicators fall below preset thresholds, the image quality is considered poor. Secondly, it can detect whether the image contains completely white (overexposed) or completely black (underexposed) areas, which typically correspond to sensor failure caused by smoke obstruction or light interference. When the image quality is below the preset standard, the smoke boundary identified based on this image is unreliable, and the system forcibly adjusts the visual weights. Set to 0, temperature weight Setting it to 1 means that the system relies entirely on temperature sensors for flue gas front location. This fault-tolerant mechanism ensures that even if one sensing mode fails, the system can still maintain basic sensing capabilities through other modes, avoiding a complete paralysis of the control system.

[0064] In addition to the aforementioned dual-modal fusion based on temperature and vision, those skilled in the art can introduce other sensing methods, such as smoke concentration sensors and gas composition analyzers, to construct a multimodal fusion system. The logic for weight adjustment can be implemented based on fuzzy rules, expert systems, machine learning models, etc., and this application does not limit the specific fusion algorithm.

[0065] Example 5 This example provides the specific content of the upstream priority scanning strategy, including: defining the coordinates of the evacuation area as the origin; scanning along the path from the evacuation area to the smoke exhaust direction; determining the position of the first flue gas recirculation front that meets the anomaly judgment condition on the scanning path as the actual position; the anomaly judgment condition includes the temperature gradient exceeding a preset gradient threshold or the visibility being lower than a preset visibility threshold.

[0066] This strategy is designed based on the core principle of fire safety: "Saving lives is more important than protecting structures." In actual fire scenarios, multiple ignition points may exist within an enclosed traffic barrier: the initial fire source may ignite one vehicle, and the fire may spread to other vehicles downstream, creating multiple temperature peaks. If the control system simply uses the highest temperature point or the point with the highest smoke concentration as the control target, the following situation may occur: in order to deal with the main fire point downstream, the system increases the ventilation power, but this causes the smoke from the secondary fire point upstream to spread in the evacuation direction, blocking the evacuation route.

[0067] The upstream-priority scanning strategy avoids this problem through explicit logic: the system first establishes a coordinate system, defining the evacuation area (the direction of personnel evacuation) as the origin and the smoke exhaust direction (the direction away from the evacuation area) as the positive direction. Then, starting from the origin, the system scans the sensed data point by point along the positive direction. During the scan, the system checks whether each location meets the anomaly detection criteria: the temperature gradient exceeds a preset gradient threshold (indicating a temperature jump at this location, possibly a smoke front), or the visibility is below a preset visibility threshold (indicating the presence of smoke at this location). Once a location that meets the criteria is found, the system immediately locks it as the current smoke return front location and terminates the scan, no longer continuing the downstream search.

[0068] The advantage of this design is that regardless of the number of fire points downstream or the temperature distribution, the system only focuses on the abnormal location closest to the evacuation area. Controlling this upstream smoke front ensures the safety of the evacuation routes. Even at the cost of uncontrollable downstream fires, this aligns with the priority principle of fire rescue: ensuring the safe evacuation of personnel first, then considering the protection of property and structures.

[0069] The preset gradient threshold needs to be adjusted based on the actual situation. Typically, the temperature gradient at the smoke front can reach 10-50°C / m, while the normal temperature distribution gradient is usually less than 5°C / m. Setting the preset gradient threshold to 8-10°C / m can effectively identify the smoke front while avoiding misjudgments of normal temperature fluctuations. The visibility threshold also needs to be determined based on sensor performance and engineering experience; for example, it can be set to 10 meters, meaning that when visibility is below 10 meters, it is considered a smoke area.

[0070] Example 6 This example provides a specific example of dynamically setting the target position.

[0071] The target location is set to be K meters upstream of the fire source, where K is a safety distance parameter, which is dynamically adjusted according to the length of the closed traffic barrier structure and evacuation requirements.

[0072] Setting the target location is one of the key parameters of the control strategy. Setting the target location a certain distance upstream of the fire source is to establish a smoke-free safety buffer zone between the fire source and the evacuation area. The selection of this safety distance K requires consideration of several factors: First, K should be greater than the minimum safe evacuation distance. According to ergonomics and fire safety regulations, the movement speed of people in a smoke environment is approximately 0.5-1 m / s, and the evacuation time depends on the length of the enclosed traffic barrier and the location of the exits. To ensure that even if the smoke front fluctuates slightly, evacuees will not come into contact with the smoke, the K value should generally not be less than the distance that evacuees can move within a few minutes, for example, 5-10 meters.

[0073] Secondly, K should not be too large. An excessively large K value means that the smoke front needs to be controlled far away from the fire source, which requires a large ventilation power. This not only results in high energy consumption but may also lead to excessive heating in downstream areas, causing greater thermal damage to the structure. Therefore, the K value should be selected as small as possible while ensuring safety.

[0074] The K value can be dynamically adjusted according to actual working conditions: when the length of the closed traffic barrier is short and the evacuation distance is short, the K value can be appropriately reduced; when a large number of people are detected to be stranded in the evacuation area, the K value can be appropriately increased to prolong the evacuation time; when the fire is small and the smoke spreads slowly, the K value can be reduced to save energy; when the fire is fierce and the smoke spreads quickly, the K value should be increased to leave sufficient safety margin.

[0075] In practical implementation, the K value can be determined through table lookup, empirical formulas, or optimization algorithms. For example, a mapping table can be established between the K value and parameters such as the length of the enclosed traffic barrier, the heat release rate of the fire source, and the number of evacuees, and a suitable K value can be obtained by looking up the table based on the current state; alternatively, an online optimization algorithm can be used to solve for the optimal K value that minimizes ventilation energy consumption while meeting safety constraints.

[0076] Example 7 This example provides a training implementation example of the policy matching neural network model.

[0077] The training process of the strategy matching neural network model includes: establishing a digital twin simulation model of the closed traffic barrier structure; generating a training dataset containing various fire conditions through computational fluid dynamics simulation; and training the strategy matching neural network model using the training dataset; wherein the state vector includes the fire source location, heat release rate, external environmental wind speed, and external environmental wind direction.

[0078] Building a digital twin simulation model is a systematic project. First, detailed geometric parameters of the enclosed traffic barrier need to be obtained: length, width, height, arrangement of light-transmitting and sound-insulating panels, location and performance parameters of ventilation equipment, etc. Then, a three-dimensional geometric model is built in computational fluid dynamics (CFD) software, and a computational mesh is generated. The quality and density of the mesh directly affect the simulation accuracy: in areas with intense smoke flow (such as near fire sources or ventilation equipment outlets), a denser mesh is needed, while in areas with gentle flow, a sparser mesh can be used to save computational resources.

[0079] After establishing the geometric model, boundary conditions and a physical model need to be set. Boundary conditions include: the heat release rate of the fire source (which can be set based on statistical data of different types of vehicle fires, such as the peak heat release rate of a car fire being approximately 3-5 MW, and a bus fire reaching 15-20 MW), the wind speed and direction of the external environment (which can be set based on meteorological data, such as prevailing southeast winds in summer and northwest winds in winter), and the operating parameters of the ventilation equipment (fan speed, air volume, etc.). Physical models include: turbulence models (such as the k-ε model, large eddy simulation, etc.), combustion models, and radiative heat transfer models. By solving the Navier-Stokes equations, the energy conservation equations, and the component transport equations, CFD software can simulate the spatiotemporal evolution of the flow field, temperature field, and smoke concentration field within a closed traffic barrier after a fire.

[0080] To construct the training dataset, a large number of simulation scenarios need to be designed. These scenarios should cover various situations that might actually occur: the fire source location can be set to different positions along the longitudinal direction of the closed traffic barrier (e.g., 100m, 200m, 500m from the starting point); the heat release rate can be set to different values ​​(e.g., 2 MW, 5 MW, 10 MW); the external wind speed can be set to different levels (e.g., 0 m / s calm, 2 m / s light breeze, 5 m / s moderate wind); and the wind direction can be set to tailwind, headwind, crosswind, etc. Through orthogonal experimental design or Latin hypercube sampling, the parameter space can be efficiently covered, generating representative combinations of scenarios.

[0081] For each simulation scenario, the optimal ventilation strategy is sought by adjusting the operating parameters of the ventilation equipment to successfully control the flue gas front (i.e., stabilize the flue gas front at the target position). This process can be achieved through multiple trial calculations or algorithm optimization. Ultimately, pairs of data, representing "input state vector → optimal control parameters," are formed, constituting the training samples for the neural network.

[0082] The state vector, including the fire source location, heat release rate, external ambient wind speed, and external ambient wind direction, is a quantitative description of the current fire situation. The fire source location is typically normalized to a value between 0 and 1; for example, a normalized value of 0.5 is used when the fire source is located at the midpoint of a closed traffic barrier. The heat release rate can be estimated using the temperature rise rate or set based on prior knowledge of the fire type. External ambient wind speed and direction can be obtained from weather stations deployed outside the barrier. In addition to the above parameters, the state vector can also include other relevant information, such as traffic congestion coefficients (reflecting vehicle density and number of people on the evacuation side) and time parameters (fire duration), to further improve the prediction accuracy of the neural network.

[0083] The neural network structure can employ fully connected networks, convolutional neural networks, or recurrent neural networks, among others. For the scenario described in this application, since the input is a one-dimensional state vector and the output is a control parameter vector, a fully connected network is typically the appropriate choice. The number of layers and neurons per layer needs to be determined through cross-validation: too few layers may lead to insufficient fitting ability, while too many layers may result in overfitting and excessive computational burden. A typical network structure may include an input layer (5-10 neurons, corresponding to the components of the state vector), 2-3 hidden layers (20-50 neurons per layer), and an output layer (3-5 neurons, corresponding to the control parameters). The activation function can be ReLU or Tanh, and the output layer can use linear activation or Sigmoid activation (if the output needs to be restricted to a certain range).

[0084] The training process employs supervised learning: the state vector is used as input, and the corresponding optimal control parameters are used as labels. The network weights are adjusted through backpropagation to minimize the mean squared error between the predicted output and the label. During training, the dataset is divided into training, validation, and test sets. The training set is used to update the weights, the validation set is used to select hyperparameters and prevent overfitting, and the test set is used to evaluate the generalization performance of the final model. After training, the neural network model is deployed to an edge controller, enabling online real-time inference.

[0085] Example 8 This example provides the implementation details of PID control.

[0086] The PID closed-loop control is calculated based on the position deviation, and its output is the frequency adjustment amount of the variable frequency longitudinal jet fan. The frequency adjustment amount is used to correct the initial control command.

[0087] The PID controller is a classic feedback controller widely used in various industrial control systems. In this application, the input to the PID controller is the position deviation. This refers to the difference between the actual position and the target position of the flue gas recirculation front; the output is the fan frequency adjustment amount, used to correct the initial control command.

[0088] A PID controller consists of three components: 1. Proportional element (P): The output is proportional to the position deviation, i.e. in This is the proportional coefficient. The proportional control provides an immediate response: when the position deviation is positive (the flue gas front is too close to the evacuation area), the proportional control outputs a positive value, increasing the fan frequency; when the position deviation is negative, the proportional control outputs a negative value, decreasing the fan frequency. Proportional coefficient It determines the system's response speed: The larger the value, the faster the system response, but an excessively large value... This could lead to overshoot and oscillation.

[0089] Integral stage (I): The output is proportional to the time integral of the position deviation, i.e. in These are the integral coefficients. The function of the integrator is to eliminate steady-state error: even if the position deviation is small, as long as it persists, the output of the integrator will accumulate until the deviation is completely eliminated. Integrator coefficients This determines the speed at which steady-state error is eliminated: The larger the value, the faster the steady-state error is eliminated, but an excessively large value... This may lead to integral saturation and sluggish response.

[0090] Differential element (D): The output is proportional to the rate of change of the position deviation, i.e. ,in These are the differential coefficients. The function of the differential element is to predict the trend of deviation changes in advance and provide a damping effect: when the position deviation is increasing rapidly, the differential element increases the output in advance to accelerate the correction speed; when the position deviation is decreasing rapidly, the differential element decreases the output to prevent overshoot. Differential coefficients Determines the damping strength: The larger the value, the more stable the system, but an excessively large value... It may lead to sensitivity to noise.

[0091] The output of a PID controller is the sum of three components: This output is the fan frequency adjustment amount, in Hz. This adjustment amount is then added to the initial control command to obtain the final fan operating frequency.

[0092] PID parameters Tuning is crucial in PID controller design. Common tuning methods include the Ziegler-Nichols method, the critical proportional gain method, and simulation trial-and-error. In this application, based on a digital twin simulation model, the PID parameters that ensure the system response meets performance requirements (e.g., overshoot less than 10%, settling time less than 30 seconds) can be determined through simulation experiments, and then fine-tuned in the actual system. Alternatively, adaptive PID control can be used, adjusting the PID parameters in real time according to the system's operating state to address nonlinear and time-varying characteristics.

[0093] It should be understood that, in addition to classic PID control, those skilled in the art can also employ other feedback control methods, such as fuzzy control, model predictive control (MPC), and sliding mode control, to correct position deviations. This application does not limit the specific feedback control algorithm; any algorithm that can generate a reasonable control adjustment based on the position deviation falls within the scope of protection of this application.

[0094] In addition to the preferred embodiments described above, those skilled in the art, guided by the teachings of this application, can also employ the following alternative methods to achieve the same or similar technical effects: Alternative sensor selection: Besides distributed fiber optic temperature sensing systems and video surveillance systems, infrared thermal imager arrays can be used for temperature and smoke detection; LiDAR can be used to measure the smoke particle concentration distribution; and gas sensor arrays can be used to detect the spatial distribution of flue gas components (such as CO concentration). These sensors can all be used to identify the location of the flue gas recirculation front and can be incorporated into a multimodal fusion framework.

[0095] Alternative control algorithms: In addition to the two-layer architecture of neural network + PID, fuzzy control, model predictive control (MPC), and adaptive control methods can also be used. For example, a simplified dynamic model of the smoke flow in a closed traffic barrier can be established, and the optimal control sequence can be solved online using the MPC algorithm; a fuzzy rule base can be designed to output control adjustment quantities based on fuzzy variables such as position deviation, rate of change of deviation, and equivalent leakage area.

[0096] Alternatives to equivalent leakage area estimation: In addition to flow balance-based methods, image recognition technology can be used to detect damaged areas of the light-transmitting and sound-insulating panels and directly calculate the damaged area; structural stress monitoring can be used to infer that panel detachment has occurred when a sudden drop in stress is detected; and acoustic sensor arrays can be used to locate the leakage point based on the airflow noise characteristics generated by the leakage.

[0097] Alternatives to multi-fire source handling: In addition to the upstream priority strategy, a weighted comprehensive strategy can be adopted: For multiple detected smoke fronts, different weights are assigned according to their distance from the evacuation area, and the weighted average position is calculated as the control target; Alternatively, a multi-objective optimization strategy can be adopted: Considering both the protection of evacuation routes and the control of downstream fire, a compromise solution is found through Pareto optimization.

[0098] It should be understood that the above are merely exemplary alternatives. Those skilled in the art can make various equivalent transformations and improvements based on specific application scenarios without departing from the core ideas of this application (taking the flue gas front position as the control target, adopting multimodal fusion sensing, and real-time compensation for structural damage). All such transformations and improvements fall within the protection scope of this application.

[0099] The specific embodiments of this application have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of this application without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of this application through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims of this application.

Claims

1. A method for dynamic locking control of the smoke front in a closed traffic barrier structure, characterized in that, include: Environmental data within the closed traffic barrier structure is acquired using multimodal sensors, and data from different modalities are fused to calculate the actual location of the flue gas recirculation front. The target location of the flue gas recirculation front is determined based on the location of the fire source; The positional deviation between the actual position and the target position is calculated, the equivalent leakage area of ​​the closed traffic barrier structure is estimated in real time, and a feedforward compensation amount is generated based on the equivalent leakage area. The operating parameters of the ventilation equipment are dynamically adjusted according to the positional deviation and the feedforward compensation amount to make the actual position approach the target position. When multiple potential fire sources are detected, an upstream priority scanning strategy is executed to lock the first smoke return front distributed along the evacuation area towards the smoke exhaust direction as the actual position.

2. The method for dynamic locking control of the smoke front in a closed traffic barrier structure according to claim 1, characterized in that, The dynamic adjustment of the ventilation equipment's operating parameters adopts a two-layer control architecture, including: training a strategy matching neural network model based on a digital twin simulation model; using the strategy matching neural network model to generate initial control commands based on the current state vector, and using PID closed-loop control to generate real-time correction commands based on the position deviation; and superimposing the initial control commands, the real-time correction commands, and the feedforward compensation to generate the final operating parameters.

3. The method for dynamic locking control of the smoke front in a closed traffic barrier structure according to claim 1, characterized in that, The formula for estimating the equivalent leakage area is: ;in, Let be the equivalent leakage area at time t. This refers to the total airflow entering the enclosed traffic barrier structure. The total exhaust volume, For flow coefficient, The static pressure difference between the inside and outside of the enclosed traffic barrier structure. This refers to air density.

4. The method for dynamic locking control of the smoke front in a closed traffic barrier structure according to claim 1, characterized in that, The multimodal sensor includes a distributed fiber optic temperature sensing system and a video surveillance system; the formula for calculating the actual position is: ;in, The actual location, The location of the temperature front is calculated based on the distributed fiber optic temperature sensing system. The visual front position is calculated based on the video surveillance system. Temperature weighting, For visual weights, and + =1, the temperature weight and the visual weight are dynamically adjusted according to the fire conditions.

5. The method for dynamic locking control of the smoke front in a closed traffic barrier structure according to claim 4, characterized in that, The dynamic adjustment rules for the temperature weight and the visual weight include: real-time monitoring of the maximum temperature difference within the enclosed traffic barrier structure; when the maximum temperature difference is less than a preset threshold, it is determined to be a low-temperature flue gas condition, and the temperature weight is reduced while the visual weight is increased.

6. The method for dynamic locking control of the smoke front in a closed traffic barrier structure according to claim 4, characterized in that, Also includes: Real-time monitoring of the image quality of the video surveillance system; When smoke obstruction or light interference is detected, causing the image quality to fall below the preset standard, the temperature weight is forcibly set to 1 and the visual weight is set to 0.

7. The method for dynamic locking control of the smoke front in a closed traffic barrier structure according to claim 1, characterized in that, The upstream priority scanning strategy specifically includes: defining the coordinates of the evacuation area as the origin; scanning along the path from the evacuation area to the smoke exhaust direction; determining the position of the first flue gas recirculation front that meets the anomaly determination conditions on the scanning path as the actual position; the anomaly determination conditions include the temperature gradient exceeding a preset gradient threshold or the visibility being lower than a preset visibility threshold.

8. The method for dynamic locking control of the smoke front in a closed traffic barrier structure according to claim 1, characterized in that, The target location is set to K meters upstream of the fire source, where K is a safety distance parameter that is dynamically adjusted according to the length of the enclosed traffic barrier structure and evacuation requirements.

9. The method for dynamic locking control of the smoke front in a closed traffic barrier structure according to claim 2, characterized in that, The training process of the strategy matching neural network model includes: establishing a digital twin simulation model of the closed traffic barrier structure; generating a training dataset containing various fire conditions through computational fluid dynamics simulation; and training the strategy matching neural network model using the training dataset; wherein the state vector includes the fire source location, heat release rate, external environmental wind speed, and external environmental wind direction.

10. The method for dynamic locking control of the smoke front in a closed traffic barrier structure according to claim 2, characterized in that, The PID closed-loop control is calculated based on the position deviation, and its output is the frequency adjustment amount of the variable frequency longitudinal jet fan. The frequency adjustment amount is used to correct the initial control command.

Images