Tunnel three-dimensional model construction and fire simulation method

Abstract

The application discloses a tunnel three-dimensional model construction and fire simulation method, comprising the following steps: firstly, randomly generating tunnel parameters such as tunnel line, side profile, cross section shape and total width of road surface. Then, combining the tunnel line and side profile to determine the position coordinates of each position point on the road surface center line of the tunnel according to a fixed step length. Next, according to the randomly generated cross section shape and the set tunnel wall thickness, the contour curves of the inner and outer walls of the tunnel at each position point are respectively generated. Finally, combining the contour curves, the total width of the road surface and the random step length, the tetrahedron corresponding to each position point is generated to form the tunnel three-dimensional model by splicing. In this way, a diversity of tunnel three-dimensional models can be constructed without manual modeling. Combined with randomly generated fire source parameters and ventilation parameters, the tunnel fire simulation can be carried out through a fire simulation software, the tunnel fire simulation data is obtained, and the simulation data required for training of a deep learning model can be generated in large quantities.

Description

Technical Field

[0001] This invention relates to the field of computer-aided design technology, specifically to a method for constructing a three-dimensional tunnel model and simulating fire. Background Technology

[0002] In fire evolution prediction, traditional computational fluid dynamics-based methods cannot meet the real-time requirements of fire evolution prediction. Existing deep learning models suffer from high costs, long processing times, and difficulties in data measurement for full-scale and scaled-down real fire experiments. Publicly available tunnel fire datasets suffer from problems such as small sample sizes, poor quality, and missing data. Furthermore, they are limited to single-structure tunnels, have poor versatility, do not support predictions under dynamic ventilation conditions, offer limited predictable data types with low resolution, and cannot quantify the uncertainty of prediction results.

[0003] While tunnel fire simulation can address the issues of high cost, long processing time, and difficulty in data measurement in real fire experiments, most current tunnel fire simulations are typically built by manually modeling and modifying simulation parameters, making it difficult to generate large quantities of simulation data needed for training deep learning models. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention proposes a method for constructing a 3D tunnel model and simulating fires, capable of generating large quantities of simulation data required for deep learning model training. The specific technical solution is as follows: In a first aspect, a method for constructing a three-dimensional model of a tunnel is provided. In a first implementable method of this first aspect, the method includes: The tunnel alignment is generated based on the randomly obtained tunnel length, design speed, and set change rate factor. The side profile is generated based on the tunnel length, design speed, fixed step size, and scale factor. And generate the cross-sectional shape based on the set curve start point, number of endpoints, angle threshold, generation step size, and road surface proportion threshold; The position coordinates of each point on the center line of the tunnel surface are determined according to the tunnel alignment and side profile, and are divided into fixed step sizes. Based on the cross-sectional shape and the set tunnel wall thickness, the contour curves of the inner and outer walls of the tunnel at each location point are determined according to the corresponding location coordinates. Each location point is generated according to its corresponding contour curve, total road width, and fixed step length. All tetrahedrons are then combined and spliced ​​together to form the tunnel three-dimensional model.

[0005] In conjunction with the first possible implementation of the first aspect, in the second possible implementation of the first aspect, randomly generating the tunnel alignment includes: The tunnel length and design speed are randomly obtained. The sequence length is determined based on the tunnel length and a fixed step size. The minimum radius of curvature is determined based on the design speed. The maximum curvature and the wandering boundary are determined based on the minimum curvature radius, and the initial curvature is randomly generated within the range of the wandering boundary. The standard deviation of the Gaussian increment is determined based on the maximum curvature, sequence length, and a set rate of change factor. Combining the initial curvature, walking boundary, sequence length, and Gaussian increment standard deviation, a curvature sequence characterizing the tunnel alignment is generated using the reflection Gaussian random walk method.

[0006] In conjunction with the first possible implementation of the first aspect, the third possible implementation of the first aspect, which randomly generates the tunnel alignment, further includes: The horizontal plane where the tunnel alignment is located is divided into grids according to a fixed step size, and each grid is numbered to determine the row number and column number corresponding to each position point on the tunnel alignment. Iterate through each location point on the tunnel alignment, count the number of location points with the same row number as the location point, and record it as the number of rows. Count the number of location points with the same column number as the location point, and record it as the number of columns. Compare the number of rows with the number of columns; In response to the fact that the number of rows does not exceed the number of columns, search for the first position node whose coordinate value exceeds the x and y coordinates of the current position point in the grid of the row to which the position point belongs, as well as in the grids of the two adjacent rows. In response to the number of rows exceeding the number of columns, search for the first position node whose coordinate value exceeds the ordinate of the current position point in the grid of the column to which the position point belongs, as well as in the grids of the two adjacent columns. Based on the coordinates corresponding to the location point and the location node, calculate the positional distance between the location point and the location node, and compare the positional distance with a fixed step size; In response to the location distance not exceeding a fixed step size, the tunnel alignment is regenerated randomly.

[0007] In conjunction with the first implementable method of the first aspect, in the fourth implementable method of the first aspect, the side profile is randomly generated, including: The tunnel length and design speed are randomly obtained. The sequence length is determined based on the tunnel length and fixed step size. The minimum radius of curvature of the vertical curve is determined based on the design speed. The curvature limit is determined based on the set minimum radius of curvature of the vertical curve, and the directional incremental standard deviation is determined by combining the fixed step size and the scale factor. The initial slope is randomly generated within the set slope limits; By combining the initial slope, slope limit, sequence length, and directed incremental standard deviation, a slope sequence representing the side profile is generated using the reflection Gaussian random walk method.

[0008] In conjunction with the fourth possible implementation of the first aspect, the fifth possible implementation of the first aspect uses a reflection Gaussian random walk method to generate the slope sequence, including: The generated random variables follow the following probability density function: ; in, Let be the probability density function of the standard distribution. For symbolic functions, Let be the standard deviation of the semi-normal distribution followed by the negative random variable. The standard deviation of the semi-normal distribution followed by a positive random variable.

[0009] In conjunction with the first possible implementation of the first aspect, in the sixth possible implementation of the first aspect, the side profile is randomly generated, including: Using the same construction method, unidirectional slopes with increasing and decreasing gradients are constructed respectively. The construction methods include: The tunnel length and design speed are randomly obtained. The sequence length is determined based on the tunnel length and fixed step size. The minimum radius of curvature of the vertical curve is determined based on the design speed. The curvature limit is determined based on the set minimum radius of curvature of the vertical curve, and the directional incremental standard deviation is determined by combining the fixed step size and the scale factor. The initial slope is randomly generated within the set slope limits; Combining the initial slope, slope limit, sequence length, and directed increment standard deviation, a slope sequence characterizing the unidirectional slope is generated using the reflection Gaussian random walk method. The transition function is established based on the window limits, and the side profile is generated by combining the slope sequences representing increasing and decreasing slopes of the unidirectional slope.

[0010] In conjunction with the first implementable method of the first aspect, in the seventh implementable method of the first aspect, the cross-sectional shape is randomly generated, including: For symmetrical cross-sectional shapes: Based on the set curve start point, number of endpoints, angle threshold, generation step size, and road surface proportion threshold, the endpoints of each half-section are generated iteratively to form the corresponding half-section endpoint sequence. Connect two adjacent half-section endpoints in the half-section endpoint sequence to form a corresponding half-section; Copy the half-section and mirror the copied half-section to form the cross-sectional shape; For asymmetrical cross-sectional shapes, two half-sections are generated using the same generation method as for symmetrical cross-sectional shapes, and the two half-sections are then spliced ​​together to form the cross-sectional shape.

[0011] Secondly, a tunnel fire simulation method is provided, wherein in a first implementable manner of the second aspect, the method includes: A tunnel three-dimensional model is constructed using any of the first to eighth implementable methods described in the first aspect; The three-dimensional tunnel model is voxelized to generate a corresponding voxelized tunnel model. Based on the voxelized tunnel model, tunnel fire simulation is performed using randomly acquired fire source parameters and ventilation parameters to obtain corresponding tunnel fire simulation data.

[0012] In conjunction with the first implementable method of the second aspect, the corresponding voxelized tunneling model is generated in the second implementable method of the second aspect, including: The voxelized tunnel model is compressed to obtain a voxelized tunnel model.

[0013] Beneficial effects: The tunnel 3D model construction and fire simulation method of the present invention can automatically generate a tunnel 3D model composed of a series of tetrahedrons by randomly generating tunnel alignment, side profile, cross-sectional shape and total road width without manual modeling. Combined with randomly obtained fire source parameters and ventilation parameters, tunnel fire simulation is carried out through fire simulation software to obtain tunnel fire simulation data. In this way, simulation data required for deep learning model training can be generated in large quantities. Attached Figure Description

[0014] To more clearly illustrate the specific embodiments of the present invention, the accompanying drawings used in the specific embodiments will be briefly described below. In all the drawings, the elements or parts are not necessarily drawn to scale.

[0015] Figure 1 A flowchart of a method for constructing a three-dimensional tunnel model according to an embodiment of the present invention; Figure 2 A flowchart of a tunnel fire simulation method provided in an embodiment of the present invention; Figure 3 The minimum radius of curvature curves for tunnels with different superelevation profiles; Figure 4 This is a schematic diagram illustrating the generation of the tunnel alignment. Figure 4 (a) is the probability density distribution of curvature. Figure 4 (b) is the sequence of curvature changes with the number of steps; Figure 5The minimum radius of curvature fitting curves for concave vertical curves (orange) and convex vertical curves (blue) are shown. Figure 6 This is an example diagram showing the generation of a one-way slope. Figure 6 (a) is the probability density distribution of the slope. Figure 6 (b) is the sequence of slope changes with the number of steps. Figure 6 (c) shows the generated side profile; Figure 7 Example diagram of the generation of a herringbone slope. Figure 7 (a) shows the gradient and height sequence of the uphill section. Figure 7 (b) shows the gradient and height sequence of the downhill section. Figure 7 (c) is the side profile corresponding to the generated herringbone slope, where the two red dashed lines are between the inflection points of the herringbone slope; Figure 8 This is an example diagram showing the generation of a V-shaped slope. Figure 8 (a) shows the gradient and height sequence of the uphill section. Figure 8 (b) shows the gradient and height sequence of the downhill section. Figure 8 (c) is the side profile corresponding to the generated V-shaped slope, where the inflection point of the V-shaped slope is between the two red dashed lines; Figure 9 A schematic diagram of the generated symmetrical cross-sectional shape; Figure 10 A schematic diagram of the generated asymmetric cross-sectional shape; Figure 11 For HRR curves; Figure 12 This is a ventilation volume curve; Figure 13 This is a schematic diagram of a 3D model of the tunnel. Figure 14 This is a schematic diagram of tetrahedral voxelization; Figure 15 This is a schematic diagram of the three-dimensional model of the tunnel after voxelization. Figure 16 This is a schematic diagram of mesh generation for self-intersection verification provided in an embodiment of the present invention. Detailed Implementation

[0016] The embodiments of the technical solution of the present invention will now be described in detail with reference to the accompanying drawings. These embodiments are merely illustrative of the technical solution of the present invention and are therefore intended to limit the scope of protection of the present invention.

[0017] Example 1 like Figure 1 The flowchart shown illustrates a method for constructing a 3D tunnel model, which includes: Step 1: Generate the tunnel alignment based on the randomly obtained tunnel length, design speed, and set change rate factor; and generate the side profile based on the tunnel length, design speed, fixed step size, and scale factor. And the cross-sectional shape is generated based on the set curve start point, number of endpoints, angle threshold, generation step size, and road surface proportion threshold; Step 2: Determine the position coordinates of each point on the center line of the tunnel surface according to a fixed step length based on the tunnel alignment and side profile. Step 3: Based on the cross-sectional shape and the set tunnel wall thickness, determine the contour curves of the inner and outer walls of the tunnel at each location point according to the corresponding position coordinates. Step 4: Generate tetrahedrons corresponding to each location point based on the corresponding contour curve, total road width, and fixed step length, and combine all tetrahedrons to form the tunnel 3D model.

[0018] Specifically, to ensure the quality and effectiveness of the dataset and improve the learning performance of the fire evolution prediction model, the following steps are taken: First, tunnel parameters such as tunnel alignment, side profile, cross-sectional shape, and total road width can be randomly generated. Then, combining the tunnel alignment and side profile, the coordinates of various points along the tunnel road centerline, divided by a fixed step size, can be determined. Next, based on the randomly generated cross-sectional shape and the set tunnel wall thickness, the contour curves of the inner and outer walls of the tunnel at each location point can be determined. Finally, combining the contour curves of the inner and outer walls, the total road width, and the fixed step size, tetrahedrons corresponding to each location point can be generated. These tetrahedrons are then sequentially assembled according to the coordinates of the location points to construct the 3D tunnel model. In this way, a 3D tunnel model can be constructed without manual modeling. Finally, combining randomly acquired fire source and ventilation parameters, tunnel fire simulation can be performed using fire simulation software to obtain tunnel fire simulation data. This allows for the generation of large-scale simulation data required for training deep learning models.

[0019] In this embodiment, optionally, the tunnel alignment is randomly generated, including: The tunnel length and design speed are randomly obtained. The sequence length is determined based on the tunnel length and a fixed step size. The minimum radius of curvature is determined based on the design speed. The maximum curvature and the wandering boundary are determined based on the minimum curvature radius, and the initial curvature is randomly generated within the range of the wandering boundary. The standard deviation of the Gaussian increment is determined based on the maximum curvature, sequence length, and a set rate of change factor. Combining the initial curvature, walking boundary, sequence length, and Gaussian increment standard deviation, a curvature sequence characterizing the tunnel alignment is generated using the reflection Gaussian random walk method.

[0020] Specifically, tunnel alignment refers to the projection curve of the tunnel centerline onto a horizontal plane. When a vehicle travels through a curved section of a tunnel, it requires a centripetal force towards the center of curvature of its travel trajectory. This force is primarily the frictional force exerted by the ground on the tires. Therefore, tunnel alignment can be characterized by a curvature sequence.

[0021] Specifically, firstly, tunnel lengths can be randomly selected from the tunnel length dataset, and design speeds can be randomly selected from the tunnel design speed dataset. Based on the tunnel lengths and the set fixed step size, the sequence length of the curvature sequence corresponding to the desired tunnel alignment can be calculated. The specific calculation formula is as follows: ; in, For sequence length, The length of the tunnel. For fixed step size.

[0022] Meanwhile, the minimum radius of curvature can be determined based on the design speed. Specifically, the centripetal force required by the vehicle is directly proportional to the square of the vehicle speed and inversely proportional to the radius of curvature of the travel trajectory. However, the friction force that the ground can provide to the tires is limited. Therefore, the minimum radius of curvature of the tunnel alignment usually increases with the increase of the design speed. The values ​​of the minimum radius of curvature for different superelevations and design speeds are shown in Table 1, where the unit of radius of curvature is meters (m).

[0023] It should be understood that superelevation refers to a unidirectional transverse slope of the road surface at curves, which is designed to counteract the centrifugal force generated when a car is driving on a curve, allowing the vehicle to pass through the curve safely and comfortably.

[0024] Table 1. Comparison of different superelevations, design speeds, and minimum radius of curvature. By fitting the data in Table 1, we can obtain the minimum radius of curvature curves corresponding to different superelevations, such as... Figure 3 As shown in the figure. These two minimum radius of curvature curves allow us to determine the minimum radius of curvature corresponding to the randomly selected design velocity.

[0025] Then, the maximum curvature can be determined based on the minimum radius of curvature, and the wandering boundary can be determined from the maximum curvature. The initial curvature is randomly generated within the interval of the wandering boundary according to a uniform distribution. The specific formula for calculating the maximum curvature is as follows: ; It is the minimum radius of curvature.

[0026] Then, the standard deviation of the Gaussian increment can be determined based on the maximum curvature, sequence length, and the set rate of change factor. The specific calculation formula is as follows: .

[0027] in, To generate the Gaussian increment standard deviation required for the tunnel alignment, This is the rate of change factor.

[0028] Finally, by combining the initial curvature, walking boundary, sequence length, and Gaussian increment standard deviation, a curvature sequence characterizing the tunnel alignment can be generated using the reflection Gaussian random walk method. The specific calculation formula is as follows: ; in, The first in the curvature sequence One curvature, Let be a Gaussian random variable, which follows a standard normal distribution. .like Figure 4 As shown, Figure 4 (a) is the probability density distribution of curvature. Figure 4 (b) is the sequence of curvature changes with the number of steps.

[0029] By using each curvature in the curvature sequence, the tangent angle at each location point in the tunnel alignment can be determined. Based on the tangent angle, the displacement corresponding to the location point can be calculated, thus determining the two-dimensional coordinates of that point. The specific calculation formula is as follows: ; ; .

[0030] in, For the first The tangent angles corresponding to each location point. For the first The number of positions before each position point For the first The displacement corresponding to each position point For the first The two-dimensional coordinates corresponding to each location point.

[0031] In this embodiment, optionally, the side profile is generated randomly, including: The tunnel length and design speed are randomly obtained. The sequence length is determined based on the tunnel length and fixed step size. The minimum radius of curvature of the vertical curve is determined based on the design speed. The curvature limit is determined based on the set minimum radius of curvature of the vertical curve, and the directional incremental standard deviation is determined by combining the fixed step size and the scale factor. The initial slope is randomly generated within the set slope limits; By combining the initial slope, slope limit, sequence length, and directed incremental standard deviation, a slope sequence representing the side profile is generated using the reflection Gaussian random walk method.

[0032] Specifically, the side profile refers to the vertical curve of the tunnel, which can be represented by the longitudinal slope at various points along the line. The longitudinal slope is the ratio of the elevation difference between two points on the same slope section to its horizontal distance, usually expressed as a percentage. Ordinary tunnels usually adopt a unidirectional slope, while long and extra-long tunnels with well-developed groundwater can adopt a herringbone slope. Subsea tunnels are constrained by the civil engineering structure and generally adopt a V-shaped slope.

[0033] When randomly generating a unidirectional slope, the sequence length of the slope sequence representing the side profile and the minimum radius of curvature of the vertical curve can be calculated first based on the randomly selected tunnel length and design vehicle speed. The calculation method for the sequence length of the slope sequence is the same as that for the sequence length of the tunnel alignment, and will not be repeated here.

[0034] In this embodiment, the relationship between the minimum radius of curvature of the vertical curve and the design speed is shown in Table 2 below, where the unit of radius of curvature is meters (m): Table 2. Comparison of the relationship between minimum radius of curvature of vertical curves and design speed By fitting the data in Table 2, we can obtain the minimum radius of curvature fitting curves for convex and concave vertical curves, as shown below. Figure 5 As shown. By fitting the corresponding curves and combining them with randomly selected design velocities, the minimum radius of curvature of the vertical curves corresponding to the two types of unidirectional slopes can be determined. For example, by fitting the minimum radius of curvature of the concave vertical curve and combining it with randomly selected design velocities, the minimum radius of curvature of the vertical curve corresponding to the concave unidirectional slope can be determined.

[0035] The curvature limit can then be determined by the minimum radius of curvature of the vertical curve, as shown in the following formula: ; in, It represents the minimum radius of curvature of the vertical curve.

[0036] Then, calculate the directed incremental standard deviation based on the curvature limit. The specific calculation formula is as follows: ; in, is the scale factor.

[0037] Then, multiple random variables can be randomly generated. Since the minimum radius of curvature of the convex vertical curve is different from that of the concave vertical curve, the random variables are required not to follow a standard normal distribution, but rather to satisfy a new distribution. In this embodiment, the probability density function corresponding to this new distribution can be set as follows: ; in, Let be the probability density function of the standard distribution. For symbolic functions, Let be the standard deviation of the semi-normal distribution followed by the negative random variable. Let be the standard deviation of the semi-normal distribution followed by the positive random variable, and its specific calculation formula is as follows: ; ; in, Let be the minimum radius of curvature of the convex vertical curve. Let be the minimum radius of curvature of the concave vertical curve.

[0038] After generating random variables, the direction of the random variables can be randomly determined according to the probabilities of the two types of one-way slopes. Multiple directed random variables are then generated by combining the random variables with the directed increment standard deviation. The specific calculation formulas for the probabilities of the two types of one-way slopes are as follows: ; ; in, The probability is negative. This refers to the directed incremental standard deviation of the concave unidirectional slope calculated using the aforementioned directed incremental standard deviation algorithm. The probability is positive. This is the directed incremental standard deviation of the convex unidirectional slope calculated using the directed incremental standard deviation algorithm described above.

[0039] Positive random variable is The negative random variable is , It is a random variable.

[0040] Then you can set the slope limits. An initial slope is randomly generated. Finally, combining the initial slope, slope limits, sequence length, and directed incremental standard deviation, a reflection Gaussian random walk method is used to generate a slope sequence representing the side profile, such as... Figure 6 As shown, Figure 6 (a) is the probability density distribution of the slope. Figure 6 (b) is the sequence of slope changes with the number of steps. Figure 6(c) shows the generated side profile. The specific formula for calculating the slope is as follows: ; in, The first in the slope sequence A slope, For the first A directed random variable.

[0041] In this embodiment, the height of each location point can be obtained by accumulating the slopes in the slope sequence and combining this with a fixed step size. The position coordinates of each location point can then be determined by combining this with their corresponding two-dimensional coordinates. The specific formula for calculating the height is as follows: , The first in the slope sequence The number of slopes before each slope.

[0042] In this embodiment, optionally, the side profile is generated randomly, including: Using the same construction method, unidirectional slopes with increasing and decreasing gradients are constructed respectively. The construction methods include: The tunnel length and design speed are randomly obtained. The sequence length is determined based on the tunnel length and fixed step size. The minimum radius of curvature of the vertical curve is determined based on the design speed. The curvature limit is determined based on the set minimum radius of curvature of the vertical curve, and the directional incremental standard deviation is determined by combining the fixed step size and the scale factor. The initial slope is randomly generated within the set slope limits; Combining the initial slope, slope limit, sequence length, and directed increment standard deviation, a slope sequence characterizing the unidirectional slope is generated using the reflection Gaussian random walk method. The transition function is established based on the window limits, and the side profile is generated by combining the slope sequences representing increasing and decreasing slopes of the unidirectional slope.

[0043] Specifically, when constructing the side profiles corresponding to the herringbone and V-shaped slopes, the aforementioned method for constructing unidirectional slopes can be used first to construct unidirectional slopes with increasing and decreasing gradients, respectively. Then, a transition function is established based on the set window limits. In this embodiment, the transition function is: ; ; ; in, , For window boundaries, Center of the window For window size, , These represent the left and right positions of the transition window, respectively. For a herringbone slope, , , For V-shaped slopes, , , .

[0044] Finally, by combining the transition function characterizing the transitional line shape, and the slope sequences characterizing unidirectional slopes with increasing and decreasing slopes, the side profiles corresponding to the herringbone or V-shaped slopes are generated. The specific expression is: ; in, For transition function, The first in the slope sequence of a unidirectional slope with increasing gradient A slope value, The first in the slope sequence of a unidirectional slope with decreasing gradient A slope value.

[0045] Human-shaped slope Figure 7 As shown, Figure 7 (a) shows the gradient and height sequence of the uphill section. Figure 7 (b) shows the gradient and height sequence of the downhill section. Figure 7 (c) is the side profile corresponding to the generated V-shaped slope. (V-shaped slope, for example...) Figure 8 As shown, Figure 8 (a) shows the gradient and height sequence of the uphill section. Figure 8 (b) shows the gradient and height sequence of the downhill section. Figure 8 (c) is the side profile corresponding to the generated V-shaped slope.

[0046] In this embodiment, optionally, the cross-sectional shape is randomly generated, including: For symmetrical cross-sectional shapes: Based on the set curve start point, number of endpoints, angle threshold, generation step size, and road surface proportion threshold, the endpoints of each half-section are generated iteratively to form the corresponding half-section endpoint sequence. Connect two adjacent half-section endpoints in the half-section endpoint sequence to form a corresponding half-section; Copy the half-section and mirror the copied half-section to form the cross-sectional shape; For asymmetrical cross-sectional shapes, two half-sections are generated using the same generation method as for symmetrical cross-sectional shapes, and the two half-sections are then spliced ​​together to form the cross-sectional shape.

[0047] Specifically, the cross-sectional shape of a tunnel refers to the shape of the curved segment formed by the intersection of the tunnel's inner wall and a plane perpendicular to the tunnel's centerline. Tunnel cross-sectional shapes can be divided into symmetrical and asymmetrical types. For a symmetrical cross-section, a half-section can be generated first, then copied, and the copied half-section can be mirrored and stitched together to obtain a complete symmetrical cross-sectional shape, such as... Figure 9 As shown. For asymmetrical cross-sections, two half-sections can be generated directly first, and then one half-section can be mirrored and spliced ​​with the other half-section to obtain a complete cross-section, as shown. Figure 10 As shown.

[0048] The specific method for generating a half-section includes: starting from a certain starting point and initial angle, continuously moving towards the inside of the tunnel with random step size and angle changing according to random rate until the termination condition is met, and the movement trajectory is the generated half-section curve.

[0049] Specifically, first, the number of target endpoints, angle thresholds, and road surface proportion thresholds for the half-section curve are set. Then, based on the set curve start point, initial angle, and farthest distance of the half-section curve, the coordinates of the next endpoint are calculated according to a random step size, as shown in the following formula: ; in, For the endpoint coordinates, Let be the coordinates of the previous endpoint. The step size is random. The angle of the upper endpoint.

[0050] After determining the coordinates of the current endpoint, the calculated x-coordinate of the current endpoint can be compared with the set farthest distance, and the maximum value can be selected to update the farthest distance corresponding to the local iteration, that is: .

[0051] Simultaneously, the angle corresponding to the current endpoint can be updated based on the angle of the previous endpoint. The specific update calculation formula is as follows: ; ; in, The random rate of change of angle, For the number of target endpoints, For angle threshold, The angle corresponding to the current endpoint.

[0052] Once the update is complete, the coordinates of the next endpoint can be iteratively calculated based on the coordinates and angle of the current endpoint and a randomly generated step size. This iteration is repeated until the angle corresponding to the endpoint is lower than an angle threshold, or the x-coordinate corresponding to the endpoint is lower than a threshold. .in, The threshold for road surface proportion. This represents the farthest distance corresponding to the current iteration.

[0053] After iteratively generating multiple endpoints, to maximize the diversity of the generated cross-sectional curve shapes, ensure the quality and effectiveness of the dataset, and improve the learning performance of the fire evolution prediction model, adjacent endpoints can be connected by straight line segments or quadratic Bezier curve segments to construct the half-section curve.

[0054] The connection of endpoints using quadratic Bezier curve segments specifically includes: First, determine the tangent direction at the endpoints. The tangent direction at the beginning endpoint is... The vector, the direction of the tangent at the tail endpoint is For a vector, the direction angle of the internal endpoint between the head endpoint and the tail endpoint is the average of the direction angles of the front and rear segments.

[0055] Then, calculate the intersection of adjacent tangents, and use the intersection of tangents for each pair of adjacent internal endpoints as Bezier control points. The control points at the beginning and end are directly set as the corresponding beginning and end endpoints.

[0056] Then, a random selection method is used to connect each segment with probability. The Bezier curve is selected, but straight lines are forced at the beginning and end. In this embodiment, .

[0057] Then, the straight line segments retain their two endpoints, and the Bezier segments are uniformly sampled using a quadratic formula to obtain a specified number of points. The specific calculation formula is as follows: .

[0058] in, This is the progress factor.

[0059] Finally, the sampling points of each segment are sequentially spliced ​​together, duplicate endpoints between adjacent segments are removed, and the type of each segment is recorded.

[0060] In this embodiment, the contour curves of the inner and outer walls of the tunnel at each location point can be generated based on the randomly generated cross-sectional shape and the set tunnel wall thickness. Specifically, the normalized cross-sectional shape is stretched to its actual size according to the net width and net height of each location point. The net width refers to the total width of the road surface, and the net height can be obtained by random sampling within the range of 4~5.5m. Then, the cross-section is rotated to the tunnel alignment, and the midpoint of the bottom of the cross-section is placed on the tunnel centerline, thus obtaining the contour curve of the inner wall at each location point. Extending lines from the midpoint of the bottom of the cross-section to each endpoint of the inner contour of the cross-section, and taking the points whose extension length is equal to the tunnel wall thickness, yields the contour curve of the outer wall at each location point.

[0061] Finally, based on the contour curves of the tunnel's inner and outer walls at each location point, and using a fixed step size, corresponding tetrahedrons are generated. All tetrahedrons are then combined and assembled to form the tunnel's 3D model. Specifically, first, each location point is traversed. For each location point, the adjacent endpoint pairs of the contour curves of the tunnel's inner and outer walls are traversed in a counter-clockwise order, totaling four points. These four points, along with the corresponding four points on the contour curves of the tunnel's inner and outer walls at the next location point, form a hexahedron. Finally, this hexahedron is decomposed into six tetrahedrons to generate the corresponding tetrahedron. The final assembled 3D tunnel model is shown below. Figure 13 As shown.

[0062] Example 2 Example 2 is largely the same as Example 1, with the main difference being that in this example, the random generation of the tunnel alignment further includes: The horizontal plane where the tunnel alignment is located is divided into grids according to a fixed step size, and each grid is numbered to determine the row number and column number corresponding to each position point on the tunnel alignment. Iterate through each location point on the tunnel alignment, count the number of location points with the same row number as the location point, and record it as the number of rows. Count the number of location points with the same column number as the location point, and record it as the number of columns. Compare the number of rows with the number of columns; In response to the fact that the number of rows does not exceed the number of columns, search for the first position node whose coordinate value exceeds the x and y coordinates of the current position point in the grid of the row to which the position point belongs, as well as in the grids of the two adjacent rows. In response to the number of rows exceeding the number of columns, search for the first position node whose coordinate value exceeds the ordinate of the current position point in the grid of the column to which the position point belongs, as well as in the grids of the two adjacent columns. Based on the coordinates corresponding to the location point and the location node, calculate the positional distance between the location point and the location node, and compare the positional distance with a fixed step size; In response to the location distance not exceeding a fixed step size, the tunnel alignment is regenerated randomly.

[0063] Specifically, when the tunnel length is large and the minimum radius of curvature is small, the randomly generated alignment has a high probability of intersecting with itself. Therefore, the intersection verification of the randomly generated tunnel alignment can be performed. If the generated tunnel alignments intersect, the tunnel alignments need to be randomly generated again until the generated tunnel alignments do not intersect.

[0064] When determining whether randomly generated tunnel alignments intersect, we can check if there are two non-adjacent points on the generated tunnel alignment, and the distance between these two points does not exceed a fixed step size. If such points exist, the generated tunnel alignment is considered to have the potential for self-intersection, and a new tunnel alignment needs to be generated.

[0065] Specifically, to improve the verification efficiency of tunnel alignment self-intersection and increase tunnel modeling speed, a spatial mesh and dimension selection strategy can be adopted. First, the horizontal plane containing the tunnel alignment can be meshed with a fixed step size, and each mesh can be assigned a row and column number. The row number of each point on the tunnel alignment is then determined based on the row and column number of the mesh in which the point is located. and column number .

[0066] Then, we can iterate through each point on the tunnel alignment to determine the number of points with the same row number as the current point. That is, the number of rows, and the number of position points with the same column number as the current position. This refers to the number of columns. It also refers to the number of location points. With the number of location points Compare them.

[0067] if Then, the current position point and all position points with the same row number as the current position point are combined into a search sequence. and sequentially in the search sequence From the three adjacent position points , and Select the node whose coordinates exceed the x-coordinate of the currently traversed node, and then traverse to both sides of the selected node to find other nodes. Calculate the distance between the selected node and the other nodes, using the following formula: ; in, The selected location point has two-dimensional coordinates. These are the two-dimensional coordinates of other points in the search sequence.

[0068] The calculated distance is compared with a fixed step size. If the distance does not exceed the fixed step size, the randomly generated alignment may have self-intersections, and the tunnel alignment needs to be regenerated. Otherwise, there are no self-intersections, and the process can continue to traverse the next point in the tunnel alignment.

[0069] if Then, the current position point and all positions with the same column number as the current position point are combined into a search sequence. and according to the search sequence Using the same method, determine whether randomly generated tunnel alignments may be self-intersecting.

[0070] Example 3 Example 3 is largely the same as Example 2, with the main difference being that, in this example, to further improve the verification efficiency of tunnel alignment self-intersection and increase tunnel modeling speed, a hash table and a balanced binary tree can be used for searching.

[0071] Specifically, such as Figure 16 As shown, the blue line represents the tunnel shape, and each point represents a location point. For the generated tunnel shape, the plane it lies on is divided into a planar grid, and each grid is assigned row numbers 0, 1, 2...6 and column numbers 0, 1, 2...6. The corresponding location point coordinate sequence is as follows: ; The coordinates of each location point are: To conveniently describe the coordinates of a location point in a certain dimension, we can... Record as location point In dimensions The coordinates on, , Represents the x-dimensional dimension. This represents the y-dimension.

[0072] First, initialize two empty hash tables. For each hash table, the key is the grid cell index, and the value is the number of points inserted in that row (column) and the two adjacent rows (columns). Hash tables allow for the quick retrieval of the number of local points in a certain dimension of a planar grid.

[0073] Then, initialize two empty hash tables. For each hash table, the key is a grid cell index, and the value is a balanced binary tree. Each node in the tree stores a tuple. .in, Location point In dimensions The coordinate values ​​on the graph. A balanced binary tree uses... As the primary keyword, Maintaining a set of dots for the second key involves the following operations: insert : Put the binary pair Inserting a balanced binary tree middle; search in the lower bound Returns pairs of elements in the tree that are greater than or equal to two. If the smallest node does not exist, return null.

[0074] front-wheel drive For a balanced binary tree Given a node in the inorder traversal, return the node that is the direct predecessor of the node in the inorder traversal.

[0075] successor For a balanced binary tree Given a node in the inorder traversal, return its direct successor node in the inorder traversal sense.

[0076] Next, iterate through each location point in the coordinate sequence and perform the following process for each location point: First, determine the search dimensions. ,in, For the location point in dimension The grid cell index, i.e., the row and column number of the grid cell. , This is the step size parameter. It involves selecting a dimension with a smaller number of local points to reduce computational overhead.

[0077] Then use a hash table to obtain the search dimensions. Current grid cell index to be searched Corresponding balanced binary tree Then in a balanced binary tree Perform a lower bound search to obtain the node. If the node If not empty, perform the following operation: Initialize two scan nodes, the left scan node Right scan node Then the following process is executed repeatedly until... , All empty: if If not empty, then record The point index stored in is Calculate the location point To the location point The distance between them is calculated using the following formula: ; if If the step size parameter is not exceeded, the current index is returned immediately. .

[0078] if If the value is greater than the step size parameter, the left scan node is moved to its direct predecessor node. If there is no direct predecessor node, then set the left scan node to null.

[0079] if If not empty, then record The point index stored in is Calculate the location point To the location point The distance between them is calculated using the following formula: ; if If the step size parameter is not exceeded, the current index is returned immediately. .

[0080] if If the value is greater than the step size parameter, then the right scan node is moved to its direct successor node. If there is no direct successor node, then set the right scan node to null.

[0081] Then, determine the index. Does it exceed 1? If it exceeds 1, then it's a circular dimension. The following processes will be executed respectively: First calculate the position point In dimensions Grid cell index ; Then loop the grid offset. Perform the following operations respectively: Update Dimensions and grid cell index Corresponding point count: ; Location point Insert Dimension and grid cell index In the corresponding balanced binary tree, execute .

[0082] This process is repeated until all positions have been traversed. If a certain index exists... satisfy If the randomly generated tunnel alignment is self-intersecting, the tunnel alignment needs to be regenerated.

[0083] Example 4 like Figure 2 The flowchart shown illustrates a tunnel fire simulation method, which includes: Step S1: Using the above-described method for constructing a three-dimensional tunnel model, a three-dimensional tunnel model is constructed. Step S2: Perform voxelization on the three-dimensional tunnel model to generate a corresponding voxelized tunnel model; Step S3: Based on the voxelized tunnel model, and combined with randomly acquired fire source parameters and ventilation parameters, perform tunnel fire simulation to obtain corresponding tunnel fire simulation data.

[0084] Specifically, firstly, the tunnel 3D model can be constructed using the tunnel 3D model construction method described in Example 1 or Example 2. Then, the randomly generated tunnel 3D model can be voxelized, converting it into a voxelized mesh model, i.e., a voxelized tunnel model, so that the fire simulation software can recognize the tunnel model. Finally, the voxelized tunnel model can be imported into the fire simulation software, along with randomly generated fire source and ventilation parameters. Material parameters of the tunnel walls and observation point locations, such as thermal conductivity and specific heat, can be set. After setting these parameters, tunnel fire simulation can be performed using the fire simulation software to obtain the corresponding tunnel fire simulation data. Simulation data required for deep learning model training can be generated in large quantities without manual modeling and parameter adjustment.

[0085] In this embodiment, converting the 3D tunnel model into a voxel-based mesh model includes: first, traversing each tetrahedron in the 3D tunnel model to determine the mesh points inside the tetrahedron, obtaining the corresponding set of mesh points. Then, merging the set of mesh points corresponding to each tetrahedron, thus converting the 3D tunnel model into a voxel-based mesh model.

[0086] like Figure 14 As shown, when determining the grid points inside a tetrahedron, we can first determine the basic triangular region formed by the edges of the tetrahedron in the top view. Specifically, the basic triangular region is obtained by the projection of the starting face onto the xy plane: when there are 3 starting faces, we directly take the projected triangles of these three starting faces as the basic triangular region.

[0087] To distinguish whether a triangular face is a starting or ending face, first determine the z-components and relative positions of the normal vectors of the four triangular faces of the tetrahedron. Then, determine whether a triangular face is a starting or ending face based on two combined conditions: 1) The z-component of the normal vector of the triangular face (upward or downward). 2) The orientation of the face relative to the tetrahedron (outward or inward). If a triangular face is both upward and outward, or downward and inward, it is defined as a starting face. If it is both downward and outward, or upward and inward, it is defined as an ending face.

[0088] When there are two starting faces, first find the common edge of the two starting faces and the two remaining vertices. Then, use the common edge and the two remaining vertices to divide the projections of the two starting faces onto the xy plane into four basic triangular regions.

[0089] After determining the basic triangular region, the region can be scanned row by row and grid by grid to determine the lowest and highest points where the perpendicular line intersects the tetrahedron. The perpendicular line is a straight line passing through a grid point in the xy-plane and perpendicular to the xy-plane. For a series of voxel grid points on this line, the lowest point is obtained by intersecting the starting face of the tetrahedron, and the highest point is obtained by intersecting the ending face. All grid points between these two points lie inside the tetrahedron.

[0090] Specifically, based on the coordinates of the grid points, the z-values ​​of the grid points are calculated using the plane equations for the points entering the face, and these are taken as the lowest points. The z-values ​​of the grid points are then calculated using the plane equations for the points leaving the face, and these are taken as the highest points. If the z-value of the highest point is less than the z-value of the lowest point due to face swapping, the highest and lowest points are swapped. Finally, between the highest and lowest points, all integer z-coordinates are generated with a set grid step size; these coordinates represent the internal grid points on the perpendicular line.

[0091] The plane equation for the surface is as follows: The equation of the plane that leaves the surface is as follows: ; in, , , , These are the equation coefficients corresponding to the surface into which the equation is entered. , , Let be the three-dimensional coordinates of the normal vector of the entering face. Substituting these coordinates into the equation of the entering face, and combining them with the coordinates of any vertex in the entering face, the coefficients of the equation can be determined. . , , , These are the equation coefficients corresponding to the departure surface. , , Let be the three-dimensional coordinates of the normal vector to the surface. Substituting these coordinates into the equation for the surface, and combining them with the coordinates of any vertex on the surface, the coefficients of the equation can be determined. .

[0092] In this embodiment, FDS (Fire Dynamics Simulator), a CFD software specifically developed for fire simulation, can be used. Compared with general CFD software such as Fluent, FDS has richer and more practical fire simulation functions, and its fire simulation results are usually closer to the actual situation.

[0093] In this embodiment, optionally, generating the corresponding voxelized tunnel model includes: The voxelized tunnel model is compressed to obtain a voxelized tunnel model.

[0094] Specifically, in the voxelized tunnel model, each grid point represents a unit cube at that location, and the collection of these unit cubes is already a gridded model that FDS can recognize and use.

[0095] However, the number of unit cubes is usually enormous, significantly increasing the size of the model file, simulation input file, and the number of elements that FDS needs to calculate, thus reducing simulation efficiency. Therefore, these unit cubes can be merged into a series of cuboids to reduce the total number of cubes.

[0096] Considering the limited height of the tunnel model in local areas, it can be sliced ​​vertically to obtain a set of unit cubes on each horizontal plane, which can be viewed as multiple orthogonal polygons with holes on that plane. Existing minimum segmentation algorithms for orthogonal polygons can be used to divide the orthogonal polygons into several non-overlapping rectangles. These rectangles are then reconstructed into cuboids with a height of 1 for that layer, thus compressing a series of unit cubes into fewer cuboids, resulting in a compressed voxelized tunnel model, such as... Figure 15 As shown.

[0097] In this embodiment, the fire source parameters input into the simulation software include the fire source location and the heat release rate curve. The fire source location refers to the distance from the fire source to the tunnel entrance, representing a proportion of the tunnel's total length. The heat release rate (HRR) is the rate at which energy is released during a fire, a key factor in predicting its evolution and assessing its hazards. In fire growth models describing HRR over time, a fire model can be used, where the initial growth stage of the fire is approximately proportional to the square of the effective burning time. After a period of burning, the HRR tends to a constant value and no longer increases; this constant value is called the maximum fire source power. The fire model can be expressed as: ; in, The growth coefficient, Burning time The combustion boundary time, This represents the maximum fire source power.

[0098] When the combustible material is a passenger car, bus, or heavy truck, the range of values ​​for the growth coefficient, combustion threshold time, and maximum ignition source power is shown in Table 3.

[0099] Table 3. Parameter ranges for different combustibles In the simulation dataset, passenger cars, buses, and heavy trucks each account for 33.3% of the total sample as combustible materials. We can use a continuous uniform distribution to sample based on the corresponding value intervals in Table 3 and calculate the combustion threshold time. The specific calculation formula is as follows: ; The heat release rate curve can be generated by superimposing Gaussian noise onto a fire model, i.e.: ; in, The generated heat release rate curve is Gaussian noise, as shown in the figure. Figure 11 As shown.

[0100] In this embodiment, the ventilation parameters input into the simulation software include fan location and ventilation volume curve. The ventilation locations, based on tunnel engineering practice, are typically set approximately 100m from the tunnel entrances at both ends and in the middle of the tunnel. The locations of the three fans are as follows: ; ; ; in, The total length of the tunnel, This represents the normal distribution of tunnel ventilation fan locations, where the first number is the mean of the distribution, the second number is the variance, and the third and fourth numbers are the upper and lower limits of the cutoff.

[0101] Ventilation volume refers to the volume of gas passing through a tunnel cross-section per unit time. A ventilation volume curve is a sequence of changes in ventilation volume of a single fan over time. To generate a ventilation volume curve that is as meaningful as possible for controlling fire smoke, considering that the critical wind speed in tunnel fire scenarios is mostly between 2 m / s and 4 m / s, with an average critical wind speed of approximately 3 m / s, we can first calculate the average ventilation volume required for a single fan to generate the critical wind speed at the fire source. The specific calculation formula is as follows: ; in, This represents the average critical wind speed. Let be the cross-sectional area of ​​the tunnel at the fire source, and let be the index parameter uniformly and randomly sampled within the range of -1 to 0. Finally, the ventilation volume is generated using the reflection Gaussian random walk method, and the ventilation volume curve can be represented as: ; in, This represents a reflection Gaussian random walk. For time step, To reflect the standard deviation of the step size in a Gaussian random walk, For minimum ventilation speed, For maximum ventilation speed, The exponential parameter is calculated using the following formula: ; ; .

[0102] The generated ventilation volume curve is as follows Figure 12 As shown.

[0103] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A method for constructing a three-dimensional model of a tunnel, characterized in that, include: The tunnel alignment is generated based on the randomly obtained tunnel length, design speed, and set change rate factor. The side profile is generated based on the tunnel length, design speed, fixed step size, and scale factor. And generate the cross-sectional shape based on the set curve start point, number of endpoints, angle threshold, generation step size, and road surface proportion threshold; The position coordinates of each point on the center line of the tunnel surface are determined according to the tunnel alignment and side profile, and are divided into fixed step sizes. Based on the cross-sectional shape and the set tunnel wall thickness, the contour curves of the inner and outer walls of the tunnel at each location point are determined according to the corresponding location coordinates. Based on the corresponding contour curve, the total road width and the fixed step length, generate tetrahedrons corresponding to each location point, and combine all tetrahedrons to form the tunnel three-dimensional model; Randomly generating the tunnel alignment includes: The tunnel length and design speed are randomly obtained. The sequence length is determined based on the tunnel length and a fixed step size. The minimum radius of curvature is determined based on the design speed. The maximum curvature and the wandering boundary are determined based on the minimum curvature radius, and the initial curvature is randomly generated within the range of the wandering boundary. The standard deviation of the Gaussian increment is determined based on the maximum curvature, sequence length, and a set rate of change factor. Combining the initial curvature, walking boundary, sequence length, and Gaussian increment standard deviation, a curvature sequence characterizing the tunnel alignment is generated using the reflection Gaussian random walk method; The side profile is generated randomly, including: The tunnel length and design speed are randomly obtained. The sequence length is determined based on the tunnel length and fixed step size. The minimum radius of curvature of the vertical curve is determined based on the design speed. The curvature limit is determined based on the set minimum radius of curvature of the vertical curve, and the directional incremental standard deviation is determined by combining the fixed step size and the scale factor. The initial slope is randomly generated within the set slope limits; Combining the initial slope, slope limit, sequence length, and directed incremental standard deviation, a slope sequence characterizing the side profile is generated using the reflective Gaussian random walk method. The slope sequence is generated using the reflection Gaussian random walk method, including: The generated random variables follow the following probability density function: ； in, Let be the probability density function of the standard distribution. For symbolic functions, Let be the standard deviation of the semi-normal distribution followed by the negative random variable. Let be the standard deviation of the semi-normal distribution followed by the positive random variable; For symmetrical cross-sectional shapes, the cross-sectional shapes are randomly generated, including: Based on the set curve start point, number of endpoints, angle threshold, generation step size, and road surface proportion threshold, the endpoints of each half-section are generated iteratively to form the corresponding half-section endpoint sequence. Connect two adjacent half-section endpoints in the half-section endpoint sequence to form a corresponding half-section; Copy the half-section and mirror the copied half-section to form the cross-sectional shape; For asymmetric cross-sectional shapes, the cross-sectional shapes are randomly generated, including: Two half-sections are generated using the same generation method as the symmetrical cross-section shape, and the two half-sections are spliced ​​together to form the cross-section shape.

2. The method for constructing a three-dimensional tunnel model according to claim 1, characterized in that, Randomly generating the tunnel alignment also includes: The horizontal plane where the tunnel alignment is located is divided into grids according to a fixed step size, and each grid is numbered to determine the row number and column number corresponding to each position point on the tunnel alignment. Iterate through each location point on the tunnel alignment, count the number of location points with the same row number as the location point, and record it as the number of rows; and count the number of location points with the same column number as the location point, and record it as the number of columns. Compare the number of rows with the number of columns; In response to the fact that the number of rows does not exceed the number of columns, search for the first position node whose coordinate value exceeds the x and y coordinates of the current position point in the grid of the row to which the position point belongs, as well as in the grids of the two adjacent rows. In response to the number of rows exceeding the number of columns, search for the first position node whose coordinate value exceeds the ordinate of the current position point in the grid of the column to which the position point belongs, as well as in the grids of the two adjacent columns. Based on the coordinates corresponding to the location point and the location node, calculate the positional distance between the location point and the location node, and compare the positional distance with a fixed step size; In response to the location distance not exceeding a fixed step size, the tunnel alignment is regenerated randomly.

3. The method for constructing a three-dimensional tunnel model according to claim 1, characterized in that, The side profile is generated randomly, including: Using the same construction method, unidirectional slopes with increasing and decreasing gradients are constructed respectively. The construction methods include: The tunnel length and design speed are randomly obtained. The sequence length is determined based on the tunnel length and fixed step size. The minimum radius of curvature of the vertical curve is determined based on the design speed. The curvature limit is determined based on the set minimum radius of curvature of the vertical curve, and the directional incremental standard deviation is determined by combining the fixed step size and the scale factor. The initial slope is randomly generated within the set slope limits; Combining the initial slope, slope limit, sequence length, and directed increment standard deviation, a slope sequence characterizing the unidirectional slope is generated using the reflection Gaussian random walk method. The transition function is established based on the window limits, and the side profile is generated by combining the slope sequences representing increasing and decreasing slopes of the unidirectional slope.

4. A method for simulating tunnel fires, characterized in that, include: A tunnel three-dimensional model is constructed using the tunnel three-dimensional model construction method as described in any one of claims 1-3; The three-dimensional tunnel model is voxelized to generate a corresponding voxelized tunnel model. Based on the voxelized tunnel model, tunnel fire simulation is performed using randomly acquired fire source parameters and ventilation parameters to obtain corresponding tunnel fire simulation data.

5. The tunnel fire simulation method according to claim 4, characterized in that, Generate the corresponding voxelized tunnel model, including: The voxelized tunnel model is compressed to obtain the voxelized tunnel model.

Images