Intelligent flame shape generation method for fire vehicle on cable system bridge

Abstract

This invention relates to the field of bridge safety technology and discloses an intelligent method for generating flame patterns of vehicles on fires on cable-stayed bridges. The method includes the following steps: S1, constructing a simulation model of the fire vehicle and calculating and outputting flame pattern data of the fire vehicle under different operating parameters; S2, constructing a dataset from the flame pattern data output by the simulation model, experimental data, and operating parameters, preprocessing it, and then dividing it into training, validation, and test sets; S3, constructing a fire vehicle flame pattern prediction model based on a conditional generative adversarial network; S4, training the model generator using a weighted combined loss function that includes adversarial loss and physical constraints; S5, obtaining operating parameters of the vehicle fire scene using a semantic recognition model; S6, inputting the identified operating parameters into the fully trained prediction model, calculating and outputting flame pattern data. This invention achieves rapid, accurate, and intelligent generation of flame patterns for vehicles on bridge fires.

Description

Technical Field

[0001] This invention relates to the field of bridge safety technology, specifically to a method for intelligently generating the flame pattern of a fire vehicle on a cable-stayed bridge. Background Technology

[0002] In recent years, with the widespread application of long-span cable-stayed bridges in important projects such as those spanning seas and rivers, fire safety issues have become increasingly prominent. Bridge vehicle fires have occurred frequently in recent years; the high-temperature flames and hot smoke generated by burning vehicles pose a serious threat to critical load-bearing components such as the main cables, suspenders, stay cables, and bridge towers. The mechanical properties of steel are extremely sensitive to temperature; when the cable temperature exceeds its material critical value, its strength and stiffness will significantly degrade, potentially leading to overall structural failure and catastrophic consequences.

[0003] In the field of bridge fire emergency assessment, time is of the essence, and efficiency determines success or failure. When a vehicle fire occurs on a bridge, emergency commanders need to make critical decisions in a very short time, and these decisions depend on an accurate assessment of the fire's development, the most crucial of which is the assessment of the fire's impact on the bridge's cable-stayed structure.

[0004] Currently, emergency assessment of bridge fires primarily relies on fire dynamics simulation methods based on computational fluid dynamics (CFD), with FDS software being the mainstream tool in the industry. This method establishes a detailed physical model, sets boundary conditions such as fire source power, wind speed, and distance, and numerically solves the Navier-Stokes equations to obtain the flame morphology characteristics under fire scenarios. While this method is theoretically rigorous, it has the following significant limitations in practical engineering applications: First, the computational efficiency is low, failing to meet the timeliness requirements of emergency assessments. A typical FDS simulation of a bridge cable vehicle fire scenario, due to its large computational domain, numerous meshes, and the need to calculate to a stable combustion state, usually takes several hours or even days to complete a single calculation. However, in actual emergency command, the decision-making window is often only a few minutes to tens of minutes. This "hour-level" or even "day-level" computational delay is completely unable to meet the golden time requirement of "minute-level" emergency decision-making, potentially leading to missing the best opportunity for response.

[0005] Second, extracting flame morphology features is difficult. Traditional simulation methods require complex post-processing to convert raw data into intuitive flame morphology parameters, a cumbersome process dependent on the experience of specialists. Current technologies struggle to achieve rapid, automatic extraction of key parameters such as flame height and morphological features.

[0006] Third, the high technical threshold hinders its widespread adoption in engineering practice. Software like FDS is complex to operate, requiring users to have a strong background in fluid mechanics and combustion science to correctly set parameters, create meshes, and interpret results. This limits the technology's adoption and application in bridge operation and maintenance management units, as most on-site managers lack relevant professional training and struggle to conduct rapid and effective safety assessments in the face of sudden fires.

[0007] Fourth, existing intelligent technologies have significant shortcomings in flame pattern prediction. Although deep learning has shown potential in the field of scientific computing, a dedicated model for generating flame patterns in bridge fires is still lacking. Existing methods struggle to simultaneously guarantee prediction efficiency and physical plausibility, and there is a lack of end-to-end solutions that directly map operating parameters to flame patterns.

[0008] Therefore, developing a method to overcome the above-mentioned defects and achieve rapid, accurate, and intelligent generation of bridge fire flame patterns has become a pressing technical problem in this field. Summary of the Invention

[0009] The technical problem to be solved by this invention is that existing bridge fire emergency assessment technologies are unable to simultaneously guarantee prediction efficiency and physical rationality, and lack the ability to directly map operating parameters to flame patterns. This invention provides a simple, convenient, and fast method for generating intelligent flame patterns of fire vehicles on cable-stayed bridges that can quickly and automatically extract flame pattern parameters.

[0010] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A method for intelligently generating flame patterns of vehicles on fire on cable-stayed bridges includes the following steps: Step S1: Construct a simulation model of a fire vehicle on a cable-stayed bridge to calculate and output flame pattern data of the fire vehicle under different working conditions; the working conditions parameters are fire source intensity and bridge surface wind speed. Step S2: Construct a dataset from the flame morphology data, experimental data, and corresponding working condition parameters output by the simulation model in Step S1; preprocess the dataset and divide it into a training set, a validation set, and a test set in a ratio of 8:1:1. Step S3: Construct a fire vehicle flame pattern prediction model based on conditional generative adversarial network, wherein the model is a generator-discriminator dual-path system; Step S4: Using the training set, train the model generator with a weighted combined loss function that includes adversarial loss and physical constraints to obtain the flame pattern prediction model. Step S5: Use a semantic recognition model to obtain the working condition parameters of the vehicle fire scene on the bridge, wherein the working condition parameters are the fire source intensity and the wind speed on the bridge surface; Step S6: Input the working condition parameters identified in step S5 into the prediction model that has been trained in step S4, calculate and output flame morphology data.

[0011] As a further improvement of the present invention, in step S1, the simulation model simplifies the two-dimensional cable plane into a one-dimensional vertical bar, and the arbitrary position of the cable is represented by the height position of the vertical bar; the two-dimensional spatial coordinates are simplified into one-dimensional position; the vehicle is simplified into a cube, and a fire source is set on the surface.

[0012] As a further improvement of the present invention, in step S1, the operating parameters include fire source intensity and bridge surface wind speed; wherein, the fire source intensity is densely sampled in a uniform distribution in the range of 0 to 50MW, and sparsely sampled in a uniform distribution in the range of 50 to 200MW; the bridge surface wind speed is sampled in a uniform distribution in the range of 0 to 15m / s; the fire source intensity and bridge surface wind speed are arranged and combined to achieve full coverage of the sample operating conditions.

[0013] As a further improvement of the present invention, in step S2, the preprocessing includes: mapping the non-uniform spatial temperature field data output by the simulation model to a preset regular grid through cubic spline interpolation, and normalizing the temperature data.

[0014] As a further improvement of the present invention, in step S3, the generation-discrimination dual-path system of the flame morphology prediction model includes a multi-stream generator and a dual-stream structure discriminator; The generator comprises an input layer, three first fully connected hidden layers, and a first output layer. The input layer is a concatenated vector of a noise vector and normalized working condition parameters, used for data input and feature initialization. The three first fully connected hidden layers are connected sequentially, with each first fully connected hidden layer followed by a batch normalization layer and an activation function, used for feature extraction and nonlinear transformation at different levels, respectively. The first output layer includes a first fully connected layer, used to output predicted temperature values ​​at multiple standard altitude points. The discriminator is a two-stream network structure, including a conditional stream, a data stream, a concatenation layer, two second fully connected hidden layers, and a second output layer. The conditional stream includes a second fully connected layer for processing the operating condition vector and extracting conditional features. The data stream includes a third fully connected layer for processing the temperature curve vector and extracting data features. The concatenation layer concatenates the feature vectors output from the conditional stream and the data stream into a fused feature vector. The two second fully connected hidden layers are connected sequentially, with each second fully connected hidden layer followed by an activation function, used for feature fusion and discriminant analysis, respectively. The second output layer includes a fourth fully connected layer for outputting the discriminant probability value.

[0015] As a further improvement of the present invention, in step S4, the weighted combination loss function of the flame morphology prediction model generator... Losses due to confrontation and physical constraints Together they constitute, specifically as follows: (1), In equation (1), The adversarial loss of the generator, whose goal is to deceive the discriminator, is calculated as follows: (2), In equation (2), This is the physical constraint loss, used to ensure the flame shape conforms to the empirical formula, and its calculation formula is as follows: (3), (4), In equations (1), (2), (3), and (4), λ1 is the hyperparameter; z is the random noise vector; c is the condition vector; E represents the mathematical expectation; P Z (Z) represents the probability distribution of random noise; P data The distribution of real data; D flame (·) represents the output probability of the discriminator; G flame (·) represents the generator's output; H pred H represents the predicted flame height. empirical The results are calculated using empirical formulas based on the classic fire plume model; P represents the fire source power, in kW. χ is the heat release rate, kW; D is the diameter of the heat source, m; χ is the convective heat fraction, with a value of 0.7.

[0016] As a further improvement of the present invention, in step S4, the training employs a gradient descent-based optimization algorithm and its hyperparameter configuration strategy, and uses an alternating iterative update strategy of discriminator first and generator follow-up to train the network model; the determination coefficient R is used through the test set. 2 The model is evaluated using the mean absolute error (MAE) and the coefficient of determination (R²). 2 >0.9, Mean Absolute Error (MAE) <5×10 -3 .

[0017] As a further improvement of the present invention, during the training process, the training parameters are adjusted by monitoring the training effect through a validation set, wherein the hyperparameter learning rate is 2×10⁻⁶. -4 The minimum number of training iterations is 500 to ensure prediction accuracy, and the maximum is 1000 to prevent overfitting.

[0018] As a further improvement of the present invention, in step S5, the semantic recognition model is an intelligent model based on a predefined semantic template and a rule parser.

[0019] Compared with existing technologies, the intelligent generation method for flame patterns of fire vehicles on cable-stayed bridges of the present invention has the following significant advantages: 1. A qualitative improvement in the efficiency of flame pattern prediction: Traditional CFD-based numerical simulation methods require several hours to complete a single flame pattern simulation. This invention, through training a fully-fledged conditional generative adversarial network model, shortens the prediction time from "hours" to "milliseconds", achieving an efficiency improvement of several orders of magnitude and well meeting the timeliness requirements of emergency assessment in bridge engineering.

[0020] 2. Significantly lowers the technical threshold for flame morphology analysis: This invention integrates semantic recognition technology to achieve intelligent conversion from natural language descriptions to flame morphology parameters. Users only need to input simple operating parameters to generate professional flame morphology features, without relying on complex simulation software operations or professional combustion science knowledge, significantly improving the ease of use and accessibility of the technology.

[0021] 3. Ensuring the physical rationality and engineering reliability of flame pattern prediction: This invention innovatively introduces a physical constraint term into the loss function, using an empirical formula for flame height to ensure that the prediction results conform to the laws of fire dynamics. Through this dual guarantee mechanism of "data-driven + physical constraint," it effectively avoids the physically unreasonable results that may arise from purely data-driven methods, significantly improving the engineering credibility of the prediction results.

[0022] 4. Excellent generalization ability and engineering applicability: Through training samples covering a large number of operating conditions, the model can adapt to the needs of flame morphology prediction under different fire source power and wind speed conditions. The flame morphology features generated by this method are based on the statistical regularities of a large number of operating conditions, and have strong universality and engineering guidance value.

[0023] 5. Provides intuitive decision support for fire safety assessment: The flame morphology characteristics generated by this invention can be directly used to assess the risk of flames licking bridge cables and the impact of heat radiation, providing intuitive and quantitative technical basis for engineering decisions such as fire separation distance design and fire extinguishing system layout. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the intelligent generation process of the flame pattern of a fire vehicle in a specific embodiment of the present invention.

[0025] Figure 2 This is a schematic diagram of the generator structure in a specific embodiment of the present invention.

[0026] Figure 3 This is a schematic diagram of the discriminator structure in a specific embodiment of the present invention.

[0027] Figure 4These are comparison images of the flame shapes generated by the model and the actual flame shapes under three different working conditions in a specific embodiment of the present invention; wherein, Figure 4 Image (a) shows the actual flame morphology under the conditions of a fire source intensity of 75MW and a bridge surface wind speed of 3.3m / s. Figure 4 (b) in the figure shows the flame morphology generated by the model under the conditions of 75MW fire source intensity and 3.3m / s wind speed on the bridge surface. Figure 4 (c) in the diagram represents the actual flame morphology under the conditions of a fire source intensity of 200MW and a bridge surface wind speed of 1.5m / s. Figure 4 (d) in the diagram represents the flame morphology generated by the model under the conditions of a fire source intensity of 200MW and a bridge surface wind speed of 1.5m / s. Figure 4 (e) in the diagram represents the actual flame morphology under the conditions of a fire source intensity of 200MW and a bridge surface wind speed of 2.0m / s. Figure 4 (f) in the figure is the flame pattern generated by the model under the working conditions of 200MW fire source intensity and 2.0m / s wind speed on the bridge surface. Detailed Implementation

[0028] The present invention will be further described below with reference to the accompanying drawings and specific preferred embodiments, but this does not limit the scope of protection of the present invention.

[0029] In the description of this invention, it should be understood that the terms "side", "center", "longitudinal", "lateral", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", "axial", "radial", "circumferential", etc., indicating the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, are only for the convenience of describing this invention and simplifying the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention.

[0030] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more unless otherwise explicitly specified.

[0031] Example like Figure 1 As shown, the intelligent generation method for flame patterns of fire vehicles on cable-stayed bridges of the present invention includes the following steps: Step S1: Construct a simulation model of a fire vehicle on a cable-stayed bridge to calculate and output flame pattern data of the fire vehicle under different fire source intensities and different bridge surface wind speed parameters.

[0032] Step S2: Construct a dataset from the flame morphology data, experimental data, and corresponding working condition parameters output by the simulation model in Step S1; preprocess the dataset and divide it into a training set, a validation set, and a test set in an 8:1:1 ratio. The working condition parameters mainly refer to the fire source intensity and the wind speed on the bridge deck.

[0033] Step S3: Construct a fire vehicle flame pattern prediction model based on conditional generative adversarial network. This model is a generator-discriminator dual-path system.

[0034] Step S4: Using the training set, train the model generator with a weighted combined loss function that includes adversarial loss and physical constraints to obtain the flame pattern prediction model.

[0035] Step S5: Use a semantic recognition model to obtain the fire source intensity and bridge deck wind speed parameters for a vehicle fire scenario on a bridge. The semantic recognition model is a dedicated intelligent model based on predefined semantic templates and rule parsers. By guiding users to perform structured input, it automatically identifies key semantic units in the text and extracts them as operating parameters.

[0036] In this embodiment, the predefined semantic template defines a set of standardized input formats and keyword patterns to identify operating parameters (such as fire source intensity and bridge surface wind speed) and their units in the user's text. The rule parser is used to parse the natural language text input by the user, apply template matching rules and syntax analysis algorithms to extract the operating parameter values, and convert them into a standardized data format that the model can process.

[0037] Predefined semantic templates are a core component of semantic recognition models, and their functions include: (1) Standardized input guidance: Provide structured input examples (such as "fire source intensity: X megawatts", "wind speed: Y meters / second") to guide users to input text in a standardized format and reduce ambiguity.

[0038] (2) Keyword pattern matching: Built-in keyword library (such as "fire source", "power", "wind speed") and unit identifier (such as "MW", "m / s") related to operating parameters, and quickly locate the parameter position through pattern matching.

[0039] Through templated design, predefined semantic templates improve the accuracy and efficiency of parameter recognition.

[0040] The rule parser is the execution component of the semantic recognition model, and its functions include: (1) Text parsing and feature extraction: Use rule engines (such as regular expressions) or lightweight natural language processing (NLP) techniques to perform word segmentation, entity recognition and relation extraction on user input, and identify parameter values ​​and units.

[0041] (2) Rule application and verification: Based on the rules of the predefined semantic template (such as numerical format verification and unit conversion rules), the extracted parameters are logically verified to ensure the validity of the data.

[0042] (3) Data standardization output: Convert the extracted parameters into the normalized format required by the model (e.g., convert “10 MW” into floating point number 10.0) and output them to the flame pattern prediction model.

[0043] The rule parser enables automated conversion from unstructured text to structured data, lowering the barrier to entry for users.

[0044] Step S6: Input the working condition parameters identified in step S5 into the prediction model that has been trained in step S4, calculate and output flame morphology data, such as flame tilt angle and height.

[0045] To obtain the most realistic flame image data, this embodiment utilizes the "Real-time Hybrid Test Platform for Vehicle-Fire-Wind Force of Substructure of Long-Span Bridge" disclosed in Chinese Patent Application 2025111956407 to conduct fire resistance tests on full-scale or large-scale models of cables under controllable parameter wind, fire, and force coupling conditions, and directly collects flame morphology data.

[0046] In step S1 of this embodiment, the simulation model simplifies the two-dimensional cable plane into a one-dimensional vertical bar, and the arbitrary position of the cable is represented by the height of the vertical bar; the two-dimensional spatial coordinates are simplified into one-dimensional positions; the vehicle is simplified into a cube, with a fire source set on its surface. For different vehicle types, corresponding to different maximum heat release rates, suggested maximum grid sizes are given. Specifically, for vehicles with a maximum heat release rate of 0–5 MW, the maximum grid size is 20 cm; for vehicles with a maximum heat release rate of 5–10 MW, the maximum grid size is 25 cm; for vehicles with a maximum heat release rate of 10–15 MW, the maximum grid size is 30 cm; for vehicles with a maximum heat release rate of 15–20 MW, the maximum grid size is 35 cm; and for vehicles with a maximum heat release rate of 20–200 MW, the maximum grid size is 50 cm. A 10-minute simulation of the maximum heat release efficiency flame is used to equivalently replace a 1.5-hour simulation of the actual flame. It is understood that the determination and interval division of the maximum heat release rate can be obtained through statistics and experiments, and will not be elaborated further here.

[0047] Furthermore, the operating parameters include fire source intensity and bridge deck wind speed; where fire source intensity is in the range of 0–50 MW, it is sampled densely with a uniform distribution, and where fire source intensity is in the range of 50–200 MW, it is sampled sparsely with a uniform distribution; where bridge deck wind speed is in the range of 0–15 m / s, it is sampled with a uniform distribution; the fire source intensity and bridge deck wind speed are arranged and combined, for example, each sampled fire source intensity is combined with each sampled bridge deck wind speed, in order to achieve full coverage of the sample operating conditions.

[0048] In step S2 of this embodiment, the preprocessing includes: mapping the non-uniform spatial temperature field data output by the simulation model to a 51×51 regular grid using cubic spline interpolation, and normalizing the temperature data.

[0049] In step S3 of this embodiment, the flame morphology prediction model generation-discrimination dual-path system includes a multi-stream generator and a dual-stream structure discriminator; The generator consists of an input layer, three fully connected hidden layers (first fully connected layers), and a first output layer. The input layer is a concatenated vector of the received noise vector and normalized condition parameters, used for data input and feature initialization. The three fully connected hidden layers are connected sequentially, each followed by a batch normalization layer and an activation function, used for feature extraction and nonlinear transformation at different levels, respectively. The first output layer includes a fully connected layer, used to output predicted temperature values ​​at multiple standard altitude points.

[0050] like Figure 2 As shown, in the multi-stream generator structure, the conditional coding stream consists of a 64-node and a 128-node double fully connected layer, used to receive normalized fire source power and bridge surface wind speed operating parameters and extract operating features; the noise coding stream consists of a 256-node and a 512-node double fully connected layer, used to receive a 100-dimensional random noise vector and extract noise features; the feature fusion layer concatenates the conditional features and noise features, passing them sequentially through a 1024-node and a 2048-node fully connected layer. Each fully connected layer is followed by a batch normalization layer and a LeakyReLU activation function, ultimately outputting a 2601-dimensional temperature field vector, which is then mapped to the [0,1] range using a Sigmoid activation function. In this embodiment, the operating parameter features (128-dimensional) and noise features (512-dimensional) are concatenated, gradually amplified to a 51×51=2601-dimensional output through three fully connected layers, and finally compressed to 0-1 using a Sigmoid function.

[0051] The discriminator is a two-stream network structure, comprising a conditional stream, a data stream, a concatenation layer, two second fully connected hidden layers, and a second output layer. The conditional stream includes a second fully connected layer that processes the operating condition vector and extracts conditional features. The data stream includes a third fully connected layer that processes the temperature curve vector and extracts data features. The concatenation layer concatenates the feature vectors output from the conditional stream and the data stream into a fused feature vector. The two second fully connected hidden layers are connected sequentially, each followed by an activation function, used for feature fusion and discriminant analysis, respectively. The second output layer includes a fourth fully connected layer that outputs the discriminant probability value.

[0052] like Figure 3 As shown, in the two-stream discriminator, the conditional encoding stream is a fully connected layer with 64 and 128 nodes, used to process the operating condition vector, ensuring that the discriminator also knows the current operating conditions. The data stream is a fully connected layer with 512 and 256 nodes, used to process the temperature field vector. Two fully connected layers compress the features, flattening the two-dimensional temperature field into a one-dimensional vector. The 128-dimensional feature vector output from the two-stream network structure is concatenated to 256 dimensions, and then passed sequentially through the 512-node and 256-node fully connected layers. Each fully connected layer is followed by a LeakyReLU activation function and a Dropout layer. Finally, a discriminant probability value is output through a Sigmoid activation function. The expression for the Sigmoid activation function is: , where x is the output value of the fully connected layer and e is a natural constant.

[0053] By splicing together the operating condition features (128 dimensions) and the temperature field features (256 dimensions), the final output is a discrimination probability value located in the [0,1] interval, which is used to characterize the probability that the sample is a real sample. Dropout is added in the middle to prevent overfitting.

[0054] In step S4 of this embodiment, the weighted combination loss function of the flame morphology prediction model generator... Losses due to confrontation and physical constraints Together they constitute, specifically as follows: (1), In equation (1), The adversarial loss of the generator, whose goal is to deceive the discriminator, is calculated as follows: (2), In equation (2), This is the physical constraint loss, used to ensure the flame shape conforms to the empirical formula, and its calculation formula is as follows: (3), (4), In equations (1), (2), (3), and (4), λ1 is the hyperparameter; z is the random noise vector; c is the condition vector; E represents the mathematical expectation; P Z (Z) represents the probability distribution of random noise; P data The distribution of real data; D flame (·) represents the output probability of the discriminator; G flame (·) represents the generator's output; H pred H represents the predicted flame height. empirical The results are calculated using empirical formulas based on the classic fire plume model; P represents the fire source power, in kW. χ is the heat release rate, kW; D is the diameter of the heat source, m; χ is the convective heat fraction, which is 0.7 in this embodiment.

[0055] In step S4, the Adam optimizer and its hyperparameter configuration strategy are used for training. An alternating iterative update strategy, with the discriminator preceding the generator, is employed to train the network model. The determination coefficient R0 is then used on the test set. 2 The model is evaluated using the mean absolute error (MAE) and the coefficient of determination (R²). 2 >0.9, Mean Absolute Error (MAE) <5×10 -3 During training, the training parameters were adjusted by monitoring the training effect using the validation set, with the hyperparameter learning rate being 2×10⁻⁶. -4 The minimum number of training epochs is 500 to ensure prediction accuracy, and the maximum is 1000 to prevent overfitting. After successful validation, the trained model weight file is deployed to a local professional model library for use by the semantic recognition model.

[0056] like Figure 4 The image shows a comparison between the flame shapes generated by the model and the actual flame shapes under three different operating conditions. As can be seen from the image, the predicted flame shapes match the actual flame shapes quite well under all three conditions. Furthermore, the model's MAE (Magnitude of Effect) is 2.45 × 10⁻⁶ based on the test set. -3 R 2 =0.944, which meets the accuracy requirements. The single prediction time is within 3 seconds, which fully demonstrates that the intelligent flame pattern generation method of the present invention can accurately simulate the actual flame pattern of a fire vehicle on a bridge, which is of great significance for guiding bridge fire rescue.

[0057] In this embodiment, Conditional Generative Adversarial Network (cGAN) technology is employed to achieve end-to-end generation from engineering condition parameters to a probability map of flame temperature distribution. Designed for bridge fire protection scenarios, the system generates a corresponding 51×51 grid of two-dimensional time-averaged temperature distribution probability cloud map based on two key input parameters: heat release rate and wind speed (without requiring vehicle location). Since the simulated flame in FDS differs significantly from the actual flame, this embodiment uses spatial temperature field data to represent flame characteristics, improving the physical consistency of the generated results and making them closer to measured data in both morphological features and numerical levels. In this embodiment, by training a fully functional cGAN model, the prediction time is reduced from "hours" to "milliseconds," achieving an efficiency improvement of several orders of magnitude and effectively meeting the timeliness requirements of emergency assessment in bridge engineering.

[0058] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for intelligently generating the flame pattern of a vehicle in fire on a cable-stayed bridge, characterized in that, Includes the following steps: Step S1: Construct a simulation model of a fire vehicle on a cable-stayed bridge to calculate and output flame pattern data of the fire vehicle under different working conditions; the working conditions parameters are fire source intensity and bridge surface wind speed. Step S2: Construct a dataset from the flame morphology data, experimental data, and corresponding working condition parameters output by the simulation model in Step S1; preprocess the dataset and divide it into a training set, a validation set, and a test set in a ratio of 8:1:

1. Step S3: Construct a fire vehicle flame pattern prediction model based on conditional generative adversarial network, wherein the model is a generator-discriminator dual-path system; Step S4: Using the training set, train the model generator with a weighted combined loss function that includes adversarial loss and physical constraints to obtain the flame pattern prediction model. Step S5: Use a semantic recognition model to obtain the working condition parameters of the vehicle fire scene on the bridge, wherein the working condition parameters are the fire source intensity and the wind speed on the bridge surface; Step S6: Input the working condition parameters identified in step S5 into the prediction model that has been trained in step S4, calculate and output flame morphology data.

2. The intelligent generation method for flame patterns of fire-prone vehicles on cable-stayed bridges according to claim 1, characterized in that, In step S1, the simulation model simplifies the two-dimensional cable plane into a one-dimensional vertical bar, and the arbitrary position of the cable is represented by the height position of the vertical bar; the two-dimensional spatial coordinates are simplified into one-dimensional position. The vehicle is simplified into a cube, with a fire source placed on its surface.

3. The intelligent generation method for flame patterns of fire-prone vehicles on cable-stayed bridges according to claim 2, characterized in that, In step S1, the operating parameters include fire source intensity and bridge surface wind speed. The fire source intensity is densely sampled in a uniform distribution when it is in the range of 0 to 50 MW, and sparsely sampled in a uniform distribution when it is in the range of 50 to 200 MW. The bridge surface wind speed is sampled in a uniform distribution when it is in the range of 0 to 15 m / s. The fire source intensity and bridge surface wind speed are arranged and combined to achieve full coverage of the sample operating conditions.

4. The intelligent generation method for flame patterns of fire-prone vehicles on cable-stayed bridges according to claim 1, characterized in that, In step S2, the preprocessing includes: mapping the non-uniform spatial temperature field data output by the simulation model to a preset regular grid using cubic spline interpolation, and normalizing the temperature data.

5. The intelligent generation method for flame patterns of vehicles on fires on cable-stayed bridges according to any one of claims 1 to 4, characterized in that, In step S3, the generation-discrimination dual-path system of the flame morphology prediction model includes a multi-stream generator and a dual-stream structure discriminator. The generator comprises an input layer, three first fully connected hidden layers, and a first output layer. The input layer is a concatenated vector of a noise vector and normalized working condition parameters, used for data input and feature initialization. The three first fully connected hidden layers are connected sequentially, with each first fully connected hidden layer followed by a batch normalization layer and an activation function, used for feature extraction and nonlinear transformation at different levels, respectively. The first output layer includes a first fully connected layer, used to output predicted temperature values ​​at multiple standard altitude points. The discriminator is a two-stream network structure, including a conditional stream, a data stream, a splicing layer, two second fully connected hidden layers, and a second output layer; the conditional stream includes a second fully connected layer for processing operating condition vectors and extracting conditional features; the data stream includes a third fully connected layer for processing temperature curve vectors and extracting data features. The concatenation layer concatenates the feature vectors output from the conditional stream and the data stream into a fused feature vector; the two second fully connected hidden layers are connected sequentially, and each second fully connected hidden layer is followed by an activation function, which is used for feature fusion and discriminant analysis, respectively; the second output layer includes a fourth fully connected layer, which is used to output the discriminant probability value.

6. The intelligent generation method for flame patterns of fire-prone vehicles on cable-stayed bridges according to claim 5, characterized in that, In step S4, the weighted combination loss function of the flame morphology prediction model Adversarial loss from generator and physical constraints Together they constitute, specifically as follows: （1）， In equation (1), The adversarial loss of the generator, whose goal is to deceive the discriminator, is calculated as follows: （2）， In equation (2), This is the physical constraint loss, used to ensure the flame shape conforms to the empirical formula, and its calculation formula is as follows: （3）， （4）， In equations (1), (2), (3), and (4), λ1 is the hyperparameter; z is the random noise vector; c is the condition vector; E represents the mathematical expectation; P Z (Z) represents the probability distribution of random noise; P data The distribution of real data; D flame (·) represents the output probability of the discriminator; G flame (·) represents the generator's output; H pred H represents the predicted flame height. empirical The results are calculated using empirical formulas based on the classic fire plume model; P represents the fire source power, in kW. χ is the heat release rate, kW; D is the diameter of the heat source, m; χ is the convective heat fraction, with a value of 0.

7.

7. The intelligent generation method for flame patterns of fire-prone vehicles on cable-stayed bridges according to claim 6, characterized in that, In step S4, the training employs a gradient descent-based optimization algorithm and its hyperparameter configuration strategy, using an alternating iterative update strategy of discriminator-first and generator-following-up to train the network model; the network model is trained using the test set with the coefficient of determination R. 2 The model is evaluated using the mean absolute error (MAE) and the coefficient of determination (R²). 2 >0.9, Mean Absolute Error (MAE) <5×10 -3 .

8. The intelligent generation method for flame patterns of fire-prone vehicles on cable-stayed bridges according to claim 7, characterized in that, During training, the training parameters were adjusted by monitoring the training effect using the validation set, with the hyperparameter learning rate being 2×10⁻⁶. -4 The minimum number of training rounds is 500, and the maximum is 1000.

9. The intelligent generation method for flame patterns of vehicles on fires on cable-stayed bridges according to any one of claims 1 to 4, characterized in that, In step S5, the semantic recognition model is an intelligent model based on predefined semantic templates and rule parsers.

Images