Vertical shaft type urban highway tunnel fire position and scale prediction method based on temperature data

Abstract

The invention relates to a shaft-type urban highway tunnel fire position and scale prediction method based on temperature data, which comprises the following steps: building a shaft-type urban highway tunnel (URTS) full-size physical model based on fire simulation software, and determining geometric parameters of a driving area and a shaft based on the actual tunnel size; various fire positions and scales are set in combination with fire scenes, grid sizes are reasonably divided, temperature measuring points are arranged, three-dimensional dynamic numerical simulation is carried out, and temperature data are obtained. After fire temperature data under various working conditions are collected, a database is established, data preprocessing is carried out, and a training set, a verification set and a test set are divided. A long short-term memory (LSTM) neural network is selected as a prediction model, training and evaluation of the LSTM model are realized through programming, real-time temperature data are input when a fire occurs, the position and scale of the fire are predicted, and iterative prediction is performed to obtain optimal state estimation. And changing the arrangement interval and the coverage range of the temperature sensors, and testing and comparing the R2 value to obtain an optimized arrangement scheme.

Description

Technical Field

[0001] The patent of this invention belongs to the field of tunnel fire fighting technology. Specifically, the invention relates to a method for predicting the location and scale of a shaft-type urban highway tunnel fire based on temperature data. Background Art

[0002] Driven by my country's rapid economic development and urbanization, more and more urban road tunnels with shafts (URTS) have been built to relieve traffic pressure. Shaft urban road tunnels can rely on natural ventilation to ensure the safety of the air environment in the tunnel, but due to the large traffic volume, high vehicle fire factors, narrow and long space, and complex smoke flow paths, the accurate judgment of the location and scale of the fire source (heat release rate) still faces great difficulties, which restricts the effective implementation of personnel evacuation and fire rescue. For example, in 2023, a van caught fire in the Xi'anmen Tunnel (a shaft urban road tunnel) in Nanjing. It took 660 seconds, a total of 11 minutes, from the fire to traffic control; it took 720 seconds, a total of 12 minutes, from the fire to the fire extinguishing by firefighters. In such a semi-enclosed public underground space, any fire incident that is not handled in time may be catastrophic. The probability of a real fire in a tunnel is low, data is difficult to obtain, the fire is highly destructive, and physical experiments are difficult to carry out, but numerical simulation provides an alternative. Due to the rapid emission of hot smoke and toxic gases in tunnel fires, timely and accurate identification of fire location and fire scale is of vital importance for guiding fire rescue operations and mitigating infrastructure losses and other losses. At present, various types of fire detection technologies are used in tunnel fire protection systems. For example, smoke particles produced during fires can be detected by extinction sensors, but the reliability is questionable in the case of poor air quality and visibility in tunnels. Video image processing systems have flourished in recent years and have shown excellent capabilities in early fire detection of tunnel flames and smoke. However, the low illumination caused by fire smoke limits the effectiveness of this system in tunnel fire detection. In summary, current fire detection systems are not smart enough to determine fire characteristics, such as location and size, which limits response options in tunnel fire emergencies. Summary of the invention

[0003] In view of the above technical deficiencies, the present invention provides a method for predicting the location and scale of fires in shaft-type urban highway tunnels based on temperature data. The required shaft-type urban highway tunnels are selected, and fire parameters are preset using fire simulation software to simulate different tunnel fire scenes (including different fire locations, fire scales (fire source heat release rate), ambient temperature, and ambient wind speed). Multiple temperature sensors are arranged under the ceiling and on the side walls of the shaft to obtain temperature data that changes over time, and then a fire database is constructed and divided into a training set, a validation set, and a test set. The database is used to train a long short-term memory recurrent neural network. When a real fire occurs in the tunnel, the smoke temperature of the driving area and the shaft is transmitted back to the prediction model to achieve rapid prediction of the location and scale of the fire. As time goes by, the prediction is continuously iterated until the ideal prediction result is achieved.

[0004] To achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0005] A method for predicting the location and scale of a shaft-type urban highway tunnel fire based on temperature data comprises the following steps:

[0006] (1) Select appropriate fire simulation software, build a full-scale physical model of an urban road tunnel with shafts (URTS), and select the driving area (height, width, length, number of lanes) and shafts (length, width, height, spacing, number) according to the actual tunnel size;

[0007] (2) Combined with the actual fire scene, various fire locations and scales (fire source heat release rate) are set, the grid size is reasonably divided, the boundary conditions are set, the temperature measurement points are arranged, and three-dimensional dynamic numerical simulation is carried out to obtain real-time temperature data;

[0008] (3) Collect fire temperature data of vertical shaft-type urban highway tunnels under different working conditions, establish a database, use Python programming software to shuffle and normalize the data, and divide it into training set, validation set and test set;

[0009] (4) Select the long short-term memory neural network (LSTM) prediction model, write the LSTM code, extract data from the database for training, and evaluate the accuracy of the model;

[0010] (5) As time goes by, the prediction results are continuously iterated until the optimal state estimation is obtained, so that the LSTM model can timely predict the location and scale of the fire source by transmitting back the temperature when a real fire occurs in the tunnel;

[0011] (6) On the basis of ensuring the prediction accuracy, optimize the layout of URTS temperature sensors and reduce the number of temperature sensors as much as possible.

[0012] Compared with the prior art, the present invention has the following beneficial effects:

[0013] (1) In reality, it is very difficult to obtain temperature data in real tunnel fires. Conducting physical experiments also faces challenges such as high cost, high risk, and difficulty in data acquisition. For example, in order to simulate a real fire, it is necessary to build a large tunnel model and equip it with a complex combustion system and sensors. This will cost a lot of money and time, making it difficult to carry out real tunnel fire experiments and obtain temperature change data at different locations under real fire scenarios. FDS numerical simulation provides an effective means to obtain URTS tunnel temperature data. Through FDS, various tunnel fire scenarios can be quickly simulated, and accurate temperature data can be obtained under different environmental conditions, at different time points, and at different locations.

[0014] (2) Traditional tunnel fire detection systems usually rely on image recognition technology to identify fires by taking images with cameras. However, when a tunnel fire occurs, thick smoke will create a low-light environment, which will seriously affect the accuracy of image recognition and even cause system failure. LSTM technology overcomes this defect. It can effectively deal with noise and occlusion problems in low-light environments and predict the location and scale of real tunnel fires by pre-learning the fire simulation cases provided in FDS. This provides a reference for the future progress in the field of tunnel fire prediction and provides new ideas and methods for future fire research. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Attached Figure 1 It is a flow chart of the present invention.

[0016] Attached Figure 2 This is the URTS physical model constructed in the numerical simulation software of the present invention.

[0017] Attached Figure 3 The tunnel temperature sensor is arranged in the numerical simulation software according to the present invention.

[0018] Attached Figure 4 This is a temperature-time diagram of the measuring points after a tunnel fire occurs according to the present invention.

[0019] Attached Figure 5 It is the R2_loc indicator on the LSTM validation set and training set of the present invention.

[0020] Attached Figure 6 It is the R2_hrr indicator on the LSTM validation set and training set of the present invention.

[0021] Attached Figure 7 This is the loss change curve on the LSTM training set and the validation set of the present invention. DETAILED DESCRIPTION

[0022] The present invention is further described below in conjunction with the accompanying drawings and embodiments. The embodiments of the present invention include but are not limited to the following embodiments.

[0023] The present invention provides a method for predicting the location and scale of a shaft-type urban highway tunnel fire based on temperature data, which is mainly used to determine the location and scale of a fire source when a real fire occurs in a URTS, thereby creating more favorable conditions for personnel evacuation and firefighting rescue.

[0024] Figure 1 This is a flow chart of a method for predicting the scale of a tunnel fire according to an embodiment of the present invention, which specifically includes the following steps:

[0025] (1) Select appropriate fire simulation software. In this case, FDS software is selected (a three-dimensional dynamic numerical simulation software specially used for fire simulation, which is widely recognized in the fire field). According to the actual tunnel size, build a full-scale physical model and set the driving area and shaft, such as Figure 2 The driving area is 160m long, 12m wide, 5.5m high, with three lanes for one-way traffic. Five groups of shafts are set up on the top, with a height of 6m, a width of 2m, and a spacing of 17m. Each group of shafts has four unit shafts, each unit shaft is 2m long, and adjacent shafts are separated by partitions. The top opening of the shaft is directly connected to the outside, and the airflow exchange between the top opening of the shaft and the inside and outside of the tunnel is considered.

[0026] (2) Preset multiple working conditions including: fire source power, fire location, ambient temperature, and ambient wind speed. In this case, the ambient temperature and ambient wind speed are fixed, and the longitudinal position and power of the fire source are changed. Each fire source position and power considers 10 parameters to form an event library containing 100 cases, corresponding to 10 fire locations and 10 fire source heat release rates set on the longitudinal ground of the tunnel. The fire locations are 8m, 24m, 40m, 56m, 72m, 88m, 104m, 120m, 136m, and 152m away from the tunnel entrance, respectively, and the fire source heat release rates are 2.5MW, 5MW, 10MW, 14MW, 20MW, 30MW, 40MW, 60MW, 80MW, and 100MW, respectively. Temperature sensors are arranged under the ceiling and on the side walls of the shaft, but this patent does not limit the number and type of temperature sensors. In this case, for example Figure 3 As shown in the figure, along the center of the tunnel in the driving area, along the horizontal longitudinal x-axis and the vertical z-axis, the temperature sensors are arranged at intervals of 1m, and a temperature sensor is arranged in the center of each shaft, with a total of 820 temperature measurement points. In this case, numerical simulation is carried out, the simulation time is 300s, and the temperature data is output every 1s. The temperature at different locations in the driving area changes with time as shown in the figure. Figure 4 shown.

[0027] (3) In this case, in the Python language environment, the pandas library is called to process the temperature data into the input format of the LSTM network. For each simulation condition, the time series are grouped according to 1-60s, 2-61s, ... 241-300s, totaling 241 groups of data, each group lasts 60s, and each second contains 820 temperature data, forming a two-dimensional matrix [60,820]. Under 100 simulation conditions, a total of 24100 groups of data are obtained, and a three-dimensional matrix [24100, 60, 820] is obtained. At the same time, each data is labeled (fire location and scale), and each data point corresponds to a fire location and a heat release rate information, forming a two-dimensional matrix [24100, 2]. Finally, the data is normalized and shuffled, and 60% is used as a training set, 20% as a validation set, and 20% as a test set.

[0028] (4) In the Python language environment, write the LSTM neural network prediction model and write the neural network software in Python. In this case, the LSTM neural network prediction module calls the tensorflow deep learning third-party library. The neural network sets two layers of LSTM and two layers of dropout, the dropout rate is set to 0.1, and finally a layer of Dense layer is set as the output layer. The activation function uses linear activation, the mean square error is used to calculate the loss value, and the RMSProp optimization algorithm is used. Custom R 2 As a neural network evaluation index, 90% is added to the training. The performance of the two parameters of fire location and scale is evaluated separately. In this case, a total of 500 trainings are conducted, and 800 data are captured each time. The number of trainings can be set according to actual needs. This patent does not limit the number of trainings and the number of captures. Figures 5-6 The performance evaluation results are shown, including R on the training set and the validation set. 2 and loss. Figure 7 , the loss value on the training set gradually decreases with the increase of training times. The relative error can be used to calculate the prediction accuracy of the model on the test set.

[0029] (5) In this case, the overall loss of the LSTM model, which was trained multiple times, decreased with the number of training cycles on the training data set. The training efficiency was high at the beginning of the training, and there was no significant change after 200 training cycles, and the model reached convergence. The calculation prediction accuracy of the fire location and HRR was 97% and 94% respectively, indicating that the model had a high degree of fit to the data and the prediction results were in good agreement with the actual data. When an actual fire occurs, the system will use the trained LSTM neural network to infer the location of the fire and the heat release rate based on the temperature measurement point data sent back.

[0030] (6) In this case, Python was used to process the established database, and the temperature sensor layout intervals were set to 1m, 5m, 10m, 20m and 40m. The sensors were arranged from the center of the tunnel to both ends, and the coverage was set to 100%, 80%, 60%, 40% and 20% respectively. The results show that when the sensor interval is between 1-20m and the coverage reaches more than 80% of the total length of the tunnel, the prediction accuracy of the fire source location and scale can be maintained above 90%. For the URTS tunnel in the case, 20m interval and 80% coverage are used as the optimal layout scheme. While ensuring the prediction accuracy, the number of sensors required in the tunnel is reduced, and the efficient configuration of sensors is achieved.

[0031] The method proposed in the present invention can timely and accurately identify fire characteristics and predict fire location and scale based solely on URTS temperature measurement point data, which is of vital importance in guiding fire rescue operations and reducing infrastructure losses.

[0032] The above embodiments are only preferred specific implementation methods of the present application, but the protection scope of the present application is not limited thereto. Any changes or modifications that have no substantive significance on the main design concept and spirit of the present invention, as long as the technical problems they solve are still consistent with the present invention, should be included in the protection scope of the present invention.

Claims

1. A method for predicting the location and scale of a shaft-type urban highway tunnel fire based on temperature data, comprising the following steps: (1) Select appropriate fire simulation software, build a full-scale physical model of an urban road tunnel with shafts (URTS), and select the driving area (height, width, length, number of lanes) and shafts (length, width, height, spacing, number) according to the actual tunnel size; (2) Combined with the actual fire scene, various fire locations and scales (fire source heat release rate) are set, the grid size is reasonably divided, the boundary conditions are set, the temperature measurement points are arranged, and three-dimensional dynamic numerical simulation is carried out to obtain real-time temperature data; (3) Collect fire temperature data of shaft-type urban highway tunnels under different fire modes, establish a database, use programming software to shuffle and normalize the data, and divide it into training set, validation set and test set; (4) Select the long short-term memory neural network (LSTM) prediction model, write the LSTM code, extract data from the database for training, and evaluate the accuracy of the model; (5) As time goes by, the prediction results are continuously iterated until the optimal state estimation is obtained, so that the LSTM model can timely predict the location and scale of the fire source by transmitting back the temperature when a real fire occurs in the tunnel; (6) On the basis of ensuring the prediction accuracy, optimize the layout of URTS temperature sensors and reduce the number of temperature sensors as much as possible.

2. A method for predicting the location and scale of a shaft-type urban highway tunnel fire based on temperature data according to claim 1, characterized in that: In the step (1): According to the actual shaft-type urban highway tunnel required, select the driving area size, including: height, width, length, number of lanes, select the shaft size, including: length, width, height, spacing, number, and the selected fire simulation software can be Fluent or Fire Dynamic Simulator (FDS).

3. The method for predicting the location and scale of a shaft-type urban highway tunnel fire based on temperature data according to claim 1 is characterized in that: In the step (2): The actual fire scenario is a fire in a five-seater car, a minibus or a large bus. The fire location is random, the grid size depends on the heat release rate of the fire source, the boundary conditions include ambient wind speed and ambient temperature, and the temperature measurement points are located below the ceiling of the central longitudinal axis of the tunnel and on the inner walls of each shaft.

4. The method for predicting the location and scale of a shaft-type urban highway tunnel fire based on temperature data according to claim 1 is characterized in that: In the step (3): Different fire modes include fire location, fire scale, ambient wind speed, and ambient temperature. The programming software is Python. In the programming environment, the pandas special package is called to disrupt the data, and the MinMaxScaler and other functions are called to normalize the data. The proportions of the training set, validation set, and test set can be different.

5. The method for predicting the location and scale of a shaft-type urban highway tunnel fire based on temperature data according to claim 1 is characterized in that: In the step (4): Writing LSTM code includes calling the deep learning library tensorflow, setting the number of neural network layers, the number of neurons, the activation function, the loss function, and the evaluation index R 2 , where: the number of neural network layers can be two; the activation function can be Sigmoid, Tanh or ReLU; the loss value loss can be calculated using the mean square error; R 2 Determine whether the iteration converges. The value range is [0, 1]. The closer the value is to 1, the better the training effect. The average relative error is used to give the prediction accuracy on the test set. The calculation formula is: Where Vtrue i is the true value of the i-th sample, Vpre i is the predicted value of the i-th sample, and n is the total number of samples in the test set.

6. The method for predicting the location and scale of a shaft-type urban highway tunnel fire based on temperature data according to claim 1 is characterized in that: In the step (5): The best state estimation is the R on the validation set obtained by training the LSTM model multiple times. 2 When the preset value is reached and convergence is achieved, when an actual fire occurs, the prediction system will use the trained LSTM to infer the location of the fire and the heat release rate based on the temperature data sent back.

7. The method for predicting the location and scale of a shaft-type urban highway tunnel fire based on temperature data according to claim 1 is characterized in that: In the step (6): Based on the existing full range of sensor layouts, the number of sensors is reduced (including reducing the intervals and coverage), and Python is used to extract corresponding data from the full range of fire databases. A data set is established, and the relative error values ​​of each solution are compared through the trained LSTM model. The accuracy of each solution is predicted, and the optimal sensor layout solution is identified. This is to reduce the number of sensors as much as possible while ensuring the prediction accuracy and achieve optimal configuration.

Images