A tunnel fire dynamic risk instant alarm and advanced prediction method and system

Abstract

The application discloses a tunnel fire dynamic risk instant alarm and advanced prediction method and system, relates to the technical field of tunnel safety monitoring and intelligent early warning, and comprises the following steps: dividing a tunnel into a plurality of evaluation intervals, identifying the types of vehicles passing through the tunnel, collecting the driving speed of the vehicles and the distance between the vehicles and the front vehicles in real time, and calculating the interval vehicle risk value of the evaluation intervals; detecting the road surface humidity and identifying the road surface damage condition, and calculating the interval road surface risk value of the evaluation intervals; obtaining the comprehensive fire risk index of each evaluation interval, calculating the whole tunnel fire risk index, constructing a time series dynamic risk trend analysis model, predicting the whole tunnel prediction risk index at a future moment, and outputting a fire risk prediction alarm information. The application realizes early and accurate early warning through comprehensive vehicle and road surface risk analysis, makes up for the defects of the lagging early warning of the traditional system, significantly improves the reliability of the fire risk evaluation, and ensures the real-time safety situation of the tunnel.

Description

Technical Field

[0001] This invention relates to the field of tunnel safety monitoring and intelligent early warning technology, specifically a method and system for real-time alarm and advance prediction of dynamic risks of tunnel fires. Background Technology

[0002] Highway tunnels are characterized by enclosed spaces, large longitudinal spans, limited lateral evacuation routes, and easy longitudinal spread of smoke. Once a vehicle catches fire inside a tunnel, the flames and smoke will spread rapidly, seriously threatening traffic safety and personnel evacuation safety in a short period of time. Although existing highway tunnels are generally equipped with various sensing devices, most of them can only use a fixed threshold alarm mechanism after a fire is detected. The alarm is triggered when indicators such as smoke and temperature reach a set threshold. This is mainly based on threshold triggering after a fire occurs and post-event linkage and response, which makes it difficult to achieve early warning of potential fire risks, unable to use continuous time-series monitoring data to assess the fire risk of a section, and lacking pre-fire intervention measures. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for real-time warning and advance prediction of dynamic risks of tunnel fires, so as to solve the problems raised in the prior art.

[0004] To solve the above-mentioned technical problems, the present invention provides the following technical solution: A method for real-time early warning and forecasting of dynamic risks of tunnel fires includes the following steps: Establish a spatial model of the tunnel to be analyzed, and divide the tunnel into several evaluation intervals along the longitudinal direction; Multiple types of sensing devices are deployed along the tunnel. Video surveillance equipment is used to identify the types of vehicles passing through the tunnel and to collect vehicle speed and distance from the vehicle in front in real time. The vehicle monitoring data is collected into the corresponding assessment interval and the interval vehicle risk value is calculated. Tunnel pavement condition monitoring is carried out using humidity sensors and video surveillance equipment. The pavement humidity is detected and the pavement damage is identified. The pavement risk value corresponding to the assessment interval is calculated comprehensively. Multiply the vehicle risk value and road surface risk value of each assessment interval to obtain the comprehensive fire risk index of each assessment interval. Based on the safety exits in the tunnel, set the hazard weight of each assessment interval, and sum the comprehensive fire risk indices to calculate the fire risk index of the entire tunnel. By sampling and storing the full tunnel fire risk index at fixed time steps, a time series dynamic risk trend analysis model is constructed to predict the full tunnel predicted risk index at future times. Based on the full tunnel predicted risk index, fire risk forecast and alarm information is output.

[0005] Furthermore, the risk value of the vehicle in the corresponding assessment interval is calculated, including: Using radar sensors and monitoring equipment deployed inside the tunnel, the speed V of vehicle i at time t is captured and evaluated within the evaluation interval j. i And the distance D between the front of the vehicle and the nearest vehicle in the same lane ahead. i , spacing D i Divide by the driving speed V i The minimum distance T between the front of vehicle i and the vehicle in front is obtained. i ; In this scheme, the radar sensor and monitoring equipment are first synchronized with clock and calibrated with spatial coordinates. Then, the millimeter-wave radar detects the position and movement of vehicles within the lane. The vehicle is identified and its lane is assigned through video surveillance. The two types of perception information are fused through target association matching. Finally, for vehicle i, its speed V at time t is obtained from the millimeter-wave radar. i Based on the monitoring equipment, the nearest vehicle in the same lane ahead of vehicle i is identified, and the distance D between the two vehicles is obtained. i ; Based on the driving speed V i and minimum time interval T i Identify speed anomalies and time-distance anomalies; The model is trained using historical traffic accident samples in the tunnel. The speed anomaly weight and the time-distance anomaly weight are obtained. The speed anomaly weight is used to weight the speed anomaly value, and the time-distance anomaly weight is used to weight the time-distance anomaly value. The results of the two weighting processes are summed. The sum is multiplied by the maximum driving state amplification and then 1 is added to obtain the driving anomaly coefficient of vehicle i at time t. Based on historical traffic accident samples, the basic risk coefficient of each vehicle type is determined. The basic risk coefficient of vehicle i is multiplied by the driving abnormality coefficient to obtain the vehicle risk coefficient of vehicle i. The average of the vehicle risk coefficients of all vehicles in the assessment interval j is calculated to obtain the interval vehicle risk value of the assessment interval j.

[0006] Furthermore, identify speed anomalies and time-distance anomalies, including: Get the tunnel speed limit V set by the system. lim The critical overspeed increment ΔV is used. If the vehicle speed does not exceed the tunnel speed limit, the speed anomaly value is set to 0. If it exceeds the tunnel speed limit but does not exceed the sum of the tunnel speed limit and the critical overspeed increment, the speed anomaly value (V) is obtained. i -V lim If the speed exceeds the sum of the tunnel speed limit and the critical overspeed increment, the abnormal speed value is set to 1. ΔV is the critical overspeed increment, which refers to the speed margin corresponding to a significant increase in driving risk after the vehicle speed exceeds the tunnel speed limit. Obtain the system-set danger time interval threshold T lim If the minimum time interval exceeds the dangerous time interval threshold, the time interval anomaly value is set to 0; if the minimum time interval does not exceed the dangerous time interval threshold, the time interval anomaly value (T) is obtained. lim -T i ) / T lim .

[0007] Danger distance threshold T lim The minimum headway required for a vehicle to safely avoid a sudden situation; when the minimum headway exceeds the danger headway threshold T. lim To determine if the following distance is safe, the outlier value is set to 0. If the minimum distance does not exceed the dangerous distance threshold, the following distance is considered to pose a certain risk. The smaller the minimum distance, the higher the risk of a rear-end collision. Therefore, when the minimum distance does not exceed the dangerous distance threshold, the outlier value (T) is obtained. lim -T i ) / T lim .

[0008] Furthermore, the velocity anomaly weights and time-distance anomaly weights are obtained by solving, including: Traffic accident samples of front-to-back collisions within the same lane in a tunnel are collected. Using surveillance footage and radar sensors, the speeds and distances between vehicles before and after the accident are extracted. Simultaneously, normal traffic flow data within the tunnel without incidents is collected as negative samples. Speed ​​anomaly values ​​δ are calculated based on both speed and distance. speed and time interval outlier δ spacing Using whether a vehicle has been involved in a traffic accident as a binary classification label, a logistic regression model is constructed: logit(p) = b0 + b v ×δ speed +b s ×δ spacing The intercept b0 and the characteristic regression coefficient b of the vehicle speed deviation are obtained by using maximum likelihood estimation. v and time interval deviation characteristic regression coefficient b s Calculate the speed anomaly weight k1=|b v | / (|b v |+|b s |) and time-distance anomaly weight k2=|b s | / (|b v |+|b s |), where k1+k2=1.

[0009] Furthermore, the basic risk coefficient for each vehicle model is determined, including: The number of occupants in vehicles of vehicle type q in historical traffic accidents is statistically analyzed. The average number of occupants for vehicle type q is obtained by calculating the average number of occupants. The average number of occupants is then normalized to obtain the personnel hazard value for vehicle type q. The probability of transporting various types of goods in the historical passage records of vehicle model q is statistically analyzed. The system retrieves the risk index values ​​corresponding to various types of goods, multiplies the transport probability of each type of goods by the corresponding risk index value, and sums them up to obtain the risk value of goods for vehicle model q. Preset personnel hazard weights and cargo hazard weights. Use personnel hazard weights to weight personnel hazard values ​​and use cargo hazard weights to weight cargo hazard values. Sum the results of the two weighting processes and use the sum as the basic risk coefficient of the vehicle type q. Repeat this process to obtain the basic risk coefficient of each vehicle type.

[0010] Furthermore, the corresponding pavement risk values ​​for the assessment intervals are obtained, including: Obtain the tunnel surface of evaluation interval j, deploy a humidity sensor in the tunnel, collect the sensor values ​​at time t, normalize the sensor values, and obtain the sensing slip index of evaluation interval j. The system acquires monitoring footage recorded by monitoring equipment, extracts road surface area images from the monitoring footage, divides the road surface area images into dry road surface, wet road surface, and puddle road surface, constructs a pixel-level semantic segmentation neural network and completes model training, collects monitoring footage at time t, inputs the trained neural network model, and segments the road surface area, dry road surface, wet road surface, and puddle road surface in the monitoring footage. The visual slip index of assessment interval j is obtained by dividing the sum of the areas of wet pavement and puddle pavement by the total area of ​​the road surface. Preset sensor slippage weights and visual slippage weights. Use the sensor slippage weights to weight the sensor slippage index, and use the visual slippage weights to weight the visual slippage index. Sum the results of the two weighting processes, and add 1 to the sum to obtain the road slippage degree of the evaluation interval j. Multiply the road slippage degree by the pre-set road damage index of the evaluation interval j to obtain the interval road risk value of the evaluation interval j.

[0011] Furthermore, the risk weight for each assessment interval is set, including: obtaining the locations of all safety exits in the tunnel, calculating the longitudinal distance from the center of each assessment interval to the nearest safety exit, and setting the risk weight for each assessment interval according to the rule that the closer the distance, the smaller the risk weight, and the farther the distance, the larger the risk weight, and the sum of the risk weights of all assessment intervals is 1.

[0012] A dynamic risk real-time alarm and advance prediction system for tunnel fires includes a memory and a processor, as well as a computer program stored in the memory and running on the processor. The processor is coupled to the memory, and the processor implements the aforementioned dynamic risk real-time alarm and advance prediction method for tunnel fires when executing the computer program.

[0013] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention provides a method and system for real-time warning and advance prediction of dynamic risks in tunnel fires, including: dividing the tunnel into several assessment intervals, identifying the types of vehicles traveling in the tunnel, collecting vehicle speeds and distances to vehicles ahead in real time, and calculating the interval vehicle risk value for each assessment interval; detecting road surface humidity and identifying road surface damage, and calculating the interval road surface risk value for each assessment interval; obtaining the comprehensive fire risk index for each assessment interval, and calculating the overall tunnel fire risk index; constructing a time-series dynamic risk trend analysis model, predicting the overall tunnel predicted risk index for future moments, and outputting fire risk prediction and warning information. This invention achieves early and accurate early warning through comprehensive vehicle and road surface risk analysis, overcoming the shortcomings of traditional systems with delayed warnings, significantly improving the reliability of fire risk assessment, and ensuring the real-time safety status of the tunnel; it achieves multi-dimensional refined risk assessment, significantly improving assessment accuracy; differentiated vehicle risk identification, effectively addressing high-risk vehicles; for tunnel highways, it has strong advance prediction capabilities, providing sufficient pre-response time, offering tunnel management personnel ample advance response time windows, and effectively compensating for the shortcomings of traditional systems with delayed warnings and high energy consumption. Attached Figure Description

[0014] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a flowchart illustrating a method for real-time warning and advance prediction of dynamic risks of tunnel fires according to the present invention. Detailed Implementation

[0016] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0017] Example: Figure 1 As shown, this invention provides a technical solution for real-time early warning and forecasting of dynamic risks in tunnel fires, comprising the following steps: A spatial model of the tunnel to be analyzed is established, and the tunnel is divided into several evaluation intervals along the longitudinal direction.

[0018] In this scheme, the evaluation intervals can be divided with a fixed length. For example, for a tunnel with a length of 1000m, if the length of a single evaluation interval is 50m, it can be divided into 20 evaluation intervals. Alternatively, the tunnel can be divided into several continuous and non-overlapping evaluation intervals along the longitudinal direction by combining the tunnel station coordinates and the spatial monitoring range of the sensing equipment. For example, for monitoring targets that cross the boundary of the interval, such as water pits, they can be grouped into the evaluation interval with a larger area proportion based on their area proportion in each interval for calculation, so as to simplify the calculation process.

[0019] Multiple types of sensing devices are deployed along the tunnel. Video surveillance equipment is used to identify the types of vehicles traveling in the tunnel and to collect vehicle speed and distance from the vehicle in front in real time. The vehicle monitoring data is then collected into the corresponding assessment interval, and the interval vehicle risk value is calculated.

[0020] Calculate the vehicle risk value for the corresponding assessment interval, including: Using radar sensors and monitoring equipment deployed inside the tunnel, the speed V of vehicle i at time t is captured and evaluated within the evaluation interval j. i And the distance D between the front of the vehicle and the nearest vehicle in the same lane ahead. i , spacing D i Divide by the driving speed V i The minimum distance T between the front of vehicle i and the vehicle in front is obtained. i ; In this scheme, the radar sensor and monitoring equipment are first synchronized with clock and calibrated with spatial coordinates. Then, the millimeter-wave radar detects the position and movement of vehicles within the lane. The vehicle is identified and its lane is assigned through video surveillance. The two types of perception information are fused through target association matching. Finally, for vehicle i, its speed V at time t is obtained from the millimeter-wave radar. i Based on the monitoring equipment, the nearest vehicle in the same lane ahead of vehicle i is identified, and the distance D between the two vehicles is obtained. i ; Based on the driving speed V i and minimum time interval T i Identify speed anomalies and time-distance anomalies, including: Get the tunnel speed limit V set by the system. limThe critical overspeed increment ΔV is used. If the vehicle speed does not exceed the tunnel speed limit, the speed anomaly value is set to 0. If it exceeds the tunnel speed limit but does not exceed the sum of the tunnel speed limit and the critical overspeed increment, the speed anomaly value (V) is obtained. i -V lim If the speed exceeds the sum of the tunnel speed limit and the critical overspeed increment, the abnormal speed value is set to 1. ΔV is the critical overspeed increment, which refers to the speed margin corresponding to a significant increase in driving risk after the vehicle speed exceeds the tunnel speed limit. For example, if the tunnel speed limit is 80 km / h, the critical overspeed increment is set to 15 km / h. That is to say, the danger of a vehicle is maximized when its speed exceeds 95 km / h, at which point the speed anomaly value is 1. Generally speaking, a critical risk speed of 10%-20% of the tunnel speed limit is more appropriate. The specific value can be determined according to the actual working conditions, and no restrictions are imposed here.

[0021] Obtain the system-set danger time interval threshold T lim If the minimum time interval exceeds the dangerous time interval threshold, the time interval anomaly value is set to 0; if the minimum time interval does not exceed the dangerous time interval threshold, the time interval anomaly value (T) is obtained. lim -T i ) / T lim .

[0022] Danger distance threshold T lim The minimum headway required for a vehicle to safely avoid a sudden situation; when the minimum headway exceeds the danger headway threshold T. lim To determine if the following distance is safe, the outlier value is set to 0. If the minimum distance does not exceed the dangerous distance threshold, the following distance is considered to pose a certain risk. The smaller the minimum distance, the higher the risk of a rear-end collision. Therefore, when the minimum distance does not exceed the dangerous distance threshold, the outlier value (T) is obtained. lim -T i ) / T lim For example, assume a danger time distance threshold T. lim =5 seconds, when the calculated minimum distance T i If the following distance exceeds 5 seconds, it means the vehicle's following distance meets the safety avoidance requirements. At this time, the time-distance anomaly value is set to 0, while the calculated minimum distance T... i A time interval of 4 seconds indicates a relatively dangerous following distance, at which point an outlier (T) is obtained. lim -T i ) / T lim =(5-4) / 5=0.2, while when the calculated minimum distance T i A time interval of 1.5 seconds indicates that following another vehicle is more dangerous, and this is when an outlier (T) is obtained. lim -T i ) / T lim=(5-1.5) / 5=0.7.

[0023] The model is trained using historical traffic accident samples from tunnels, and the speed anomaly weight and time-distance anomaly weight are obtained. The sum of the two weights is 1, including: Traffic accident samples of front-to-back collisions within the same lane in a tunnel are collected. Using surveillance footage and radar sensors, the speeds and distances between vehicles before and after the accident are extracted. Simultaneously, normal traffic flow data within the tunnel without incidents is collected as negative samples. Speed ​​anomaly values ​​δ are calculated based on both speed and distance. speed and time interval outlier δ spacing Using whether a vehicle has been involved in a traffic accident as a binary classification label, a logistic regression model is constructed: logit(p) = b0 + b v ×δ speed +b s ×δ spacing Specifically, if a sample vehicle is involved in a rear-end collision in the same lane, the label y=1; if the vehicle is passing normally without incident, the label y=0. Then, maximum likelihood estimation is used to solve for the model intercept b0 and the speed deviation characteristic regression coefficient b. v and time interval deviation characteristic regression coefficient b s Calculate the speed anomaly weight k1=|b v | / (|b v |+|b s |) and time-distance anomaly weight k2=|b s | / (|b v |+|b s |), where k1+k2=1, and the relative magnitudes of k1 and k2 are determined by |b v | and |b s The decision is used to characterize the relative contributions of speed anomalies and time-distance anomalies to rear-end collision risk; since constructing a logistic regression model is an existing technology, it will not be elaborated on here.

[0024] Speed ​​anomaly values ​​are weighted using speed anomaly weighting, and time distance anomaly values ​​are weighted using time distance anomaly weighting. The results of the two weighting processes are summed, multiplied by the maximum driving state amplification factor, and then 1 is added to obtain the driving anomaly coefficient of vehicle i at time t. The formula is: β i =1+A β (k1×δ speed i +k2×δ spacing i ), where β i Let k1 be the driving anomaly coefficient of vehicle i at time t, k2 be the speed anomaly weight, k2 be the time-distance anomaly weight, and δ be the driving anomaly coefficient. speed iLet δ be the outlier speed value of vehicle i at time t. spacing i Let A be the time-distance anomaly of vehicle i at time t. β To achieve the maximum amplification under the maximum driving conditions, in this scheme, A β A is not included in the relative weight calculation, but is uniformly applied to the weighted result as the maximum driving state amplification. β It can be 0 β It takes values ​​within the range of ≤1; Based on historical traffic accident samples, the basic risk coefficient for each vehicle type is determined, including: The number of occupants in vehicles of vehicle type q in historical traffic accidents is statistically analyzed. The average number of occupants for vehicle type q is obtained by calculating the average number of occupants. The average number of occupants is then normalized to obtain the personnel hazard value for vehicle type q. The normalization formula here is f(x) = 1 - e -x This formula is a function where f(x) takes the value between 0 and 1 when x is greater than 0, and f(x) increases as x increases. This is consistent with the logic that the greater the average number of passengers, the greater the risk to personnel.

[0025] The probability of transporting various types of goods in the historical passage records of vehicle model q is statistically analyzed. The system retrieves the risk index values ​​corresponding to various types of goods, multiplies the transport probability of each type of goods by the corresponding risk index value, and sums them up to obtain the risk value of goods for vehicle model q. JT / T 617 is used to determine the standard classification and transport attributes of dangerous goods, but it does not directly specify the hazard index values ​​adopted in this scheme. The hazard index values ​​in this scheme are normalized risk parameters of 0 to 1 set by the system. The larger the value, the greater the hazard to fire development and personnel evacuation caused by combustion, explosion, toxicity, or corrosiveness of the corresponding goods in a tunnel scenario. The system performs initial calibration based on the historical transport records of the target tunnel, the consequences of accidents or emergencies, and management requirements, and then uses subsequent accumulated data for calibration, as shown in Table 1. Here are the hazard index values ​​corresponding to several common goods. The values ​​listed in Table 1 are only one example and do not belong to the standard values ​​specified in JT / T 617. Table 1. Mapping Relationship between Dangerous Goods Categories and Example Hazard Index Values

[0026] Preset personnel hazard weights and cargo hazard weights, with the sum of the two weights being 1. The personnel hazard weight is used to weight the personnel hazard value, and the cargo hazard weight is used to weight the cargo hazard value. The results of the two weighting processes are summed, and the sum is used as the basic risk coefficient of the vehicle type q. This process is repeated to obtain the basic risk coefficient of each vehicle type.

[0027] ​Multiply the basic risk coefficient of vehicle i by the driving anomaly coefficient to obtain the vehicle risk coefficient of vehicle i. Then, average the vehicle risk coefficients of all vehicles in the evaluation interval j to obtain the interval vehicle risk value of the evaluation interval j.

[0028] Tunnel pavement condition monitoring is carried out using humidity sensors and video surveillance equipment. The pavement humidity is detected and the pavement damage is identified. The pavement risk value of the corresponding assessment section is calculated by comprehensively calculating the pavement risk value of the assessment section.

[0029] The corresponding pavement risk value for the assessment interval is obtained, including: Obtain the tunnel surface of evaluation interval j, deploy humidity sensors in the tunnel, collect sensor values ​​at time t, normalize the sensor values, and obtain the sensing slip index of evaluation interval j. The higher the sensor value, the closer the sensing slip index is to 1. The normalization process here can be determined based on the humidity sensor values ​​U1 when the tunnel surface is dry and U2 when it is wet. Specifically, the sensing slippage index is: (U... j -U1) / (U2-U1), when the exponent is less than 0, set the exponent to 0; when the exponent is greater than 1, set the exponent to 1. j This refers to the sensor value of the evaluation interval j at time t.

[0030] The system acquires monitoring footage recorded by monitoring equipment, extracts road surface area images from the monitoring footage, divides the road surface area images into dry road surface, wet road surface, and puddle road surface, constructs a pixel-level semantic segmentation neural network and completes model training, collects monitoring footage at time t, inputs the trained neural network model, and segments the road surface area, dry road surface, wet road surface, and puddle road surface in the monitoring footage. The general steps for training a neural network model in this scheme include: First, we acquired on-site monitoring footage of the tunnel. Images of the tunnel surface area were cropped from the footage and extracted as the original sample set. The training, validation, and test datasets were then divided in a 7:2:1 ratio. Pixel-level manual annotation was performed on the road surface images, classifying each pixel into three semantic labels: dry road surface, wet road surface, and puddle surface. A lightweight U-Net pixel-level semantic segmentation neural network was built, using the RGB image of the road surface as input and outputting a semantic segmentation prediction map with the same resolution as the input image. Each pixel in the prediction map corresponds to a category label. During training, the cross-entropy loss function was used, with the Adam optimizer selected. The initial learning rate was set to 0.0001, and a learning rate decay strategy was employed, reducing the learning rate to 0.8 times its original value every 10 training epochs. The batch size was 8, and the maximum number of training epochs was 80. Since the specific training process is based on existing technology, it will not be detailed here.

[0031] The visual slip index of assessment interval j is obtained by dividing the sum of the areas of wet pavement and puddle pavement by the total area of ​​the road surface. Preset sensor slippage weights and visual slippage weights, with the sum of the two weights being 1. Use the sensor slippage weights to weight the sensor slippage index, and use the visual slippage weights to weight the visual slippage index. Sum the results of the two weighting processes, and add 1 to the sum to obtain the road slippage degree of the evaluation interval j. Multiply the road slippage degree by the pre-set road damage index of the evaluation interval j to obtain the interval road risk value of the evaluation interval j.

[0032] PCI is an abbreviation for Pavement Condition Index, a core indicator for evaluating the surface damage of road surfaces. Its value ranges from 0 to 100; a lower PCI value indicates more severe pavement damage. In this scheme, the PCI value can be obtained from existing road condition inspection results based on JTG 5210—2018 "Highway Technical Condition Assessment Standard," or it can be calculated according to this standard. During calculation, the type, degree, and area of ​​various types of pavement damage within the assessment interval are collected. The pavement damage rate (DR) is calculated according to the damage weights specified in the standard, and then the PCI value for the assessment interval is obtained according to the PCI calculation method specified in the standard. Since the specific calculation of the PCI value is existing technology, it will not be detailed here. Based on the calculated PCI values, Table 2 shows the grading boundaries of the highway PCI grading interval using this standard. According to this table, highways are divided into 5 levels based on PCI values. The pavement damage index is the normalized mapping value used by this system for risk calculation, as shown in Table 2. Table 2. Example Mapping Relationship between PCI Classification and Pavement Damage Index for Highways

[0033] Multiply the vehicle risk value and road surface risk value of each assessment interval to obtain the comprehensive fire risk index of each assessment interval. Based on the safety exits in the tunnel, set the hazard weight of each assessment interval, and sum the comprehensive fire risk indices to calculate the fire risk index of the entire tunnel. The process of setting a hazard weight for each assessment interval includes: obtaining the locations of all safety exits in the tunnel, calculating the longitudinal distance from the center of each assessment interval to the nearest safety exit, and setting the hazard weight for each assessment interval based on the rule that the closer the distance, the smaller the hazard weight, and the farther the distance, the larger the hazard weight, and the sum of the hazard weights of all assessment intervals is 1.

[0034] In this embodiment, the hazard weight of each assessment interval can be set based on the location of safety exits and smoke vents. Safety exits refer to facilities for emergency evacuation and refuge of personnel in the tunnel, including tunnel entrances, pedestrian cross passages, vehicular cross passages, evacuation shafts, and tunnel refuge rooms. Smoke vents refer to opening facilities used to extract high-temperature fumes. The specific steps are as follows: Obtain the center position of the evaluation interval j, and then obtain the spatial distance H from the center position to the nearest safe exit. j And the spatial distance G from the center location to the nearest smoke exhaust outlet j By obtaining the evacuation speed of personnel during historical accidents, the average evacuation speed V within the tunnel is obtained, and then the average time T required for personnel to evacuate to a safe exit is calculated. j =H j / V; Retrieve historical fire accident records in the tunnel to determine the location of the fire source and the time of the fire. Using the time of the fire as the starting point, collect the monitoring time series data of the smoke sensors deployed along the tunnel. Read the time t1 when the sensor at a distance L1 from the fire source first detects that the smoke concentration reaches the threshold concentration, and the time t2 when the sensor at a distance L2 from the fire source first detects that the smoke concentration reaches the threshold concentration. When selecting the sensor cross-section, ensure that there is no smoke exhaust port between the two monitoring cross-sections L1 and L2. The smoke threshold concentration is calculated using the 3σ criterion: First, smoke concentration values ​​from the smoke sensor are collected at several times when no accident has occurred, and the average value µ and standard deviation σ are calculated. µ+3σ is then used as the smoke threshold concentration. This is because, for tunnels, in the absence of accidents, the smoke concentration is generally stable within a stable range. The probability of the value falling within the 3σ criterion range [µ-3σ, µ+3σ] is 99.73%, while the probability of it falling outside this range is only 0.27%. Therefore, when the smoke concentration measured by the sensor exceeds µ+3σ, it can be determined that the section has been affected by fire smoke, and there is a safety risk in the area. Therefore, µ+3σ is used as the smoke threshold concentration here.

[0035] Then, according to the formula: Get the duration T0, and according to T j The larger T0 is, the greater the risk weight of the assessment interval, and the rule that the sum of the risk weights of all assessment intervals is 1 is used to set the risk weight of each assessment interval; this is because T j T0 represents the average escape time for people within the assessment zone to evacuate to the nearest safe exit, while T0 represents the time required for smoke to reach and diffuse from the smoke exhaust vent. j When T is significantly greater than T0, it indicates that personnel will be exposed to high temperatures or toxic fumes before evacuation, the effective escape window is extremely small, the fire hazard level in the assessment section is extremely high, and the escape time redundancy for personnel in this assessment section is low. Therefore, T jThe larger the -T0 value, the greater the risk weight should be.

[0036] By sampling and storing the full tunnel fire risk index at fixed time steps, a time series dynamic risk trend analysis model is constructed to predict the full tunnel predicted risk index at future times. Based on the full tunnel predicted risk index, fire risk forecast and alarm information is output.

[0037] In this embodiment, the predicted risk index of the entire tunnel at future times is obtained, including: A fixed time step is set, and data from several historical moments is collected to obtain the overall tunnel fire risk index corresponding to each moment in the time series. An LSTM time series prediction model is constructed, and the overall tunnel fire risk index sequence is input into the model to deduce the predicted overall tunnel risk index for future moments. The specific steps of using the LSTM time series prediction model include: The LSTM model uses a multivariate sequence of 60 time steps as input. The network consists of an input layer, a single-layer LSTM with 32 hidden units, a Dropout layer, and a fully connected output layer. The output layer has 4 nodes, corresponding to the risk values ​​in the next 30s, 60s, 120s, and 300s.

[0038] The model training steps include: A sliding window of samples is constructed in chronological order. The input consists of 60 consecutive time steps, and the labels are the actual risk values ​​at the 6th, 12th, 24th, and 60th time steps after the window ends. The samples are divided into a 70% training set, a 15% validation set, and a 15% test set in chronological order to avoid random shuffling that could lead to future information leakage. Mean squared error is used as the loss function. The Adam optimizer has an initial learning rate of 0.001 and a batch size of 64. The validation set is stopped early if the loss does not decrease for 10 consecutive rounds, and the model with the lowest validation error is saved. During online operation, the latest data is added to the window and the oldest data is deleted every 5 seconds. Four time interval prediction values ​​are output, which are the prediction values ​​at 30 s, 60 s, 120 s, and 300 s, respectively.

[0039] This system divides risks into multiple levels for advance warning. When warning conditions are met, an early warning command is triggered in advance, and advance warning information is pushed to the tunnel management center according to the corresponding risk level. Specifically, this includes: Since the final predicted risk index is a value between 0 and 1, and the larger the value, the higher the probability of an accident and the greater the severity of the accident, this embodiment divides the danger into 5 levels and analyzes the predicted risk index for the next 30 seconds. When the predicted risk index is in [0, 0.2), it indicates a low risk, and normal operation and automatic monitoring should be maintained; when the predicted risk index is in [0.2, 0.4), it indicates a relatively low risk, and the frequency of video and data verification should be increased; when the predicted risk index is in [0.4, 0.6), it indicates a medium risk, and the management center should issue an early warning and prepare traffic and ventilation measures; when the predicted risk index is in [0.6, 0.8), it indicates a relatively high risk, and an immediate alarm should be issued, high-risk vehicles should be restricted from entering, and the contingency plan should be activated; when the predicted risk index is in [0.8, 1.0], it indicates a high risk, and emergency broadcasts, traffic closures, and fire and rescue coordination should be implemented.

[0040] This embodiment also provides a real-time alarm and advance prediction system for dynamic risks of tunnel fires, including a memory and a processor, and a computer program stored in the memory and running on the processor. The processor is coupled to the memory, and when the processor executes the computer program, it implements the aforementioned method for real-time alarm and advance prediction of dynamic risks of tunnel fires. Since this method for real-time alarm and advance prediction of dynamic risks of tunnel fires has been described in detail above, it will not be repeated here.

[0041] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in a different order than that shown in the embodiments and still achieve the desired result. Additionally, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0042] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0043] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for real-time early warning and advance prediction of dynamic risks of tunnel fires, characterized in that, Includes the following steps: Establish a spatial model of the tunnel to be analyzed, and divide the tunnel into several evaluation intervals along the longitudinal direction; Multiple types of sensing devices are deployed along the tunnel. Video surveillance equipment is used to identify the types of vehicles passing through the tunnel and to collect vehicle speed and distance from the vehicle in front in real time. The vehicle monitoring data is collected into the corresponding assessment interval and the interval vehicle risk value is calculated. Tunnel pavement condition monitoring is carried out using humidity sensors and video surveillance equipment. The pavement humidity is detected and the pavement damage is identified. The pavement risk value corresponding to the assessment interval is calculated comprehensively. Multiply the vehicle risk value and road surface risk value of each assessment interval to obtain the comprehensive fire risk index of each assessment interval. Based on the safety exits in the tunnel, set the hazard weight of each assessment interval, and sum the comprehensive fire risk indices to calculate the fire risk index of the entire tunnel. By sampling and storing the full tunnel fire risk index at fixed time steps, a time series dynamic risk trend analysis model is constructed to predict the full tunnel predicted risk index at future times. Based on the full tunnel predicted risk index, fire risk forecast and alarm information is output.

2. The method for real-time warning and advance prediction of dynamic risks of tunnel fires according to claim 1, characterized in that, Calculate the vehicle risk value for the corresponding assessment interval, including: Using radar sensors and monitoring equipment deployed inside the tunnel, the speed V of vehicle i at time t is captured and evaluated within the evaluation interval j. i And the distance D between the front of the vehicle and the nearest vehicle in the same lane ahead. i , spacing D i Divide by the driving speed V i The minimum distance T between the front of vehicle i and the vehicle in front is obtained. i ; Based on the driving speed V i and minimum time interval T i Identify speed anomalies and time-distance anomalies; The model is trained using historical traffic accident samples in the tunnel. The speed anomaly weight and the time-distance anomaly weight are obtained. The speed anomaly weight is used to weight the speed anomaly value, and the time-distance anomaly weight is used to weight the time-distance anomaly value. The results of the two weighting processes are summed. The sum is multiplied by the maximum driving state amplification and then 1 is added to obtain the driving anomaly coefficient of vehicle i at time t. Based on historical traffic accident samples, the basic risk coefficient of each vehicle type is determined. The basic risk coefficient of vehicle i is multiplied by the driving abnormality coefficient to obtain the vehicle risk coefficient of vehicle i. The average of the vehicle risk coefficients of all vehicles in the assessment interval j is calculated to obtain the interval vehicle risk value of the assessment interval j.

3. The method for real-time warning and advance prediction of dynamic risks of tunnel fires according to claim 2, characterized in that, Identify speed anomalies and time-distance anomalies, including: Get the tunnel speed limit V set by the system. lim The critical overspeed increment ΔV is used. If the vehicle speed does not exceed the tunnel speed limit, the speed anomaly value is set to 0. If it exceeds the tunnel speed limit but does not exceed the sum of the tunnel speed limit and the critical overspeed increment, the speed anomaly value (V) is obtained. i -V lim If the speed exceeds the sum of the tunnel speed limit and the critical overspeed increment, the abnormal speed value is set to 1. Obtain the system-set danger time interval threshold T lim If the minimum time interval exceeds the dangerous time interval threshold, the time interval anomaly value is set to 0; if the minimum time interval does not exceed the dangerous time interval threshold, the time interval anomaly value (T) is obtained. lim -T i ) / T lim .

4. The method for real-time alarm and advance prediction of dynamic risk of tunnel fire according to claim 2, characterized in that, The velocity anomaly weights and time-distance anomaly weights are obtained by solving the problem, including: Traffic accident samples of front-to-back collisions within the same lane in a tunnel are collected. Using surveillance footage and radar sensors, the speeds and distances between vehicles before and after the accident are extracted. Simultaneously, normal traffic flow data within the tunnel without incidents is collected as negative samples. Speed ​​anomaly values ​​δ are calculated based on both speed and distance. speed and time interval outlier δ spacing Using whether a vehicle has been involved in a traffic accident as a binary classification label, a logistic regression model is constructed: logit(p) = b0 + b v ×δ speed +b s ×δ spacing The intercept b0 and the characteristic regression coefficient b of the vehicle speed deviation are obtained by using maximum likelihood estimation. v and time interval deviation characteristic regression coefficient b s Calculate the speed anomaly weight k1=|b v | / (|b v |+|b s |) and time-distance anomaly weight k2=|b s | / (|b v |+|b s |), where k1+k2=1.

5. The method for real-time warning and advance prediction of dynamic risks of tunnel fires according to claim 2, characterized in that, Determine the basic risk coefficient for each vehicle model, including: The number of occupants in vehicles of vehicle type q in historical traffic accidents is statistically analyzed. The average number of occupants for vehicle type q is obtained by calculating the average number of occupants. The average number of occupants is then normalized to obtain the personnel hazard value for vehicle type q. The probability of transporting various types of goods in the historical passage records of vehicle model q is statistically analyzed. The system retrieves the risk index values ​​corresponding to various types of goods, multiplies the transport probability of each type of goods by the corresponding risk index value, and sums them up to obtain the risk value of goods for vehicle model q. Preset personnel hazard weights and cargo hazard weights. Use personnel hazard weights to weight personnel hazard values ​​and use cargo hazard weights to weight cargo hazard values. Sum the results of the two weighting processes and use the sum as the basic risk coefficient of the vehicle type q. Repeat this process to obtain the basic risk coefficient of each vehicle type.

6. The method for real-time warning and advance prediction of dynamic risks of tunnel fires according to claim 1, characterized in that, The corresponding pavement risk value for the assessment interval is obtained, including: Obtain the tunnel surface of evaluation interval j, deploy a humidity sensor in the tunnel, collect the sensor values ​​at time t, normalize the sensor values, and obtain the sensing slip index of evaluation interval j. The system acquires monitoring footage recorded by monitoring equipment, extracts road surface area images from the monitoring footage, divides the road surface area images into dry road surface, wet road surface, and puddle road surface, constructs a pixel-level semantic segmentation neural network and completes model training, collects monitoring footage at time t, inputs the trained neural network model, and segments the road surface area, dry road surface, wet road surface, and puddle road surface in the monitoring footage. The visual slip index of assessment interval j is obtained by dividing the sum of the areas of wet pavement and puddle pavement by the total area of ​​the road surface. Preset sensor slippage weights and visual slippage weights. Use the sensor slippage weights to weight the sensor slippage index, and use the visual slippage weights to weight the visual slippage index. Sum the results of the two weighting processes, and add 1 to the sum to obtain the road slippage degree of the evaluation interval j. Multiply the road slippage degree by the pre-set road damage index of the evaluation interval j to obtain the interval road risk value of the evaluation interval j.

7. The method for real-time warning and advance prediction of dynamic risks of tunnel fires according to claim 1, characterized in that, The hazard weight for each assessment interval is set by: obtaining the locations of all safety exits in the tunnel, calculating the longitudinal distance from the center of each assessment interval to the nearest safety exit, and setting the hazard weight for each assessment interval according to the rule that the closer the distance, the smaller the hazard weight, and the farther the distance, the larger the hazard weight, and the sum of the hazard weights of all assessment intervals is 1.

8. A dynamic real-time alarm and advance prediction system for tunnel fire risks, characterized in that, It includes a memory and a processor, as well as a computer program stored in the memory and running on the processor. The processor is coupled to the memory, and when the processor executes the computer program, it implements a method for real-time warning and advance prediction of dynamic risks of tunnel fires as described in any one of claims 1-7.

Images