Indoor fire prediction method and device based on generative adversarial network, and electronic equipment

Abstract

The invention provides an indoor fire prediction method and device based on a generative adversarial network and electronic equipment, and belongs to the technical field of fire prediction, and the method comprises the steps: obtaining a wall layout, a fire position and a fire development moment of a target building; inputting the wall layout, the fire position and the fire development moment of the target building into a feature processing layer of a pre-constructed generative adversarial network model, and performing feature processing on the wall layout, the fire position and the fire development moment by the feature processing layer to obtain a feature map of the target building output by the feature processing layer; the feature map comprises a wall layout feature, a fire position feature and a fire development moment feature; and inputting the feature map of the target building into a prediction layer of the generative adversarial network model to obtain an indoor temperature field and a smoke visibility distribution field, output by the prediction layer, of the target building at the fire development moment. According to the method, the generative adversarial network model is adopted to predict the indoor fire, so that the efficiency and the real-time performance of indoor fire prediction are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fire prediction, and particularly to an indoor fire prediction method, device and electronic device based on a generative adversarial network. Background Art

[0002] Indoor fire prediction is a key link in building fire protection design, safety assessment and disaster emergency management. How to efficiently predict indoor fires is a key problem to be solved urgently. However, the existing fire simulation methods based on Computational Fluid Dynamics (CFD) are time-consuming in calculation, highly dependent on professional knowledge, inefficient and poor in real-time performance. Summary of the Invention

[0003] The present invention provides an indoor fire prediction method, device and electronic device based on a generative adversarial network to solve the defects of low efficiency and poor real-time performance in the prior art when using the CFD-based fire simulation method to predict indoor fires.

[0004] The present invention provides an indoor fire prediction method based on a generative adversarial network, including: Obtaining the wall layout, fire starting position and fire development time of a target building; Inputting the wall layout, fire starting position and fire development time of the target building into a pre-constructed generative adversarial network model to obtain the indoor temperature field and smoke visibility distribution field of the target building at the fire development time output by the generative adversarial network model; Wherein, the generative adversarial network model is trained based on the sample wall layout, sample fire starting position and sample fire development time of a sample building, and the indoor temperature field label and smoke visibility distribution field label of the sample building at the sample fire development time.

[0005] In some embodiments, the generative adversarial network model includes a feature processing layer and a prediction layer; Correspondingly, the step of inputting the wall layout, fire starting position and fire development time of the target building into a pre-constructed generative adversarial network model to obtain the indoor temperature field and smoke visibility distribution field of the target building at the fire development time output by the generative adversarial network model includes: Inputting the wall layout, fire starting position and fire development time of the target building into the feature processing layer, and the feature processing layer performs feature processing on the wall layout, fire starting position and fire development time to obtain the feature map of the target building output by the feature processing layer; the feature map includes wall layout features, fire starting position features and fire development time features; Input the feature map of the target building into the prediction layer to obtain the indoor temperature field and the smoke visibility distribution field of the target building at the fire development moment output by the prediction layer.

[0006] In some embodiments, the process of determining the generative adversarial network model includes: Obtain the sample wall layouts, sample fire starting positions, and sample fire development moments of multiple sample buildings, determine the indoor temperature field labels and smoke visibility distribution field labels of the multiple sample buildings at the sample fire development moments, and construct a training data set and a test data set; Use the sample wall layout, sample fire starting position, and sample fire development moment of the first sample building in the training data set as training data, and use the indoor temperature field label and smoke visibility distribution field label of the first sample building at the sample fire development moment as training labels to train the initial generative adversarial network model. After training is completed, obtain the generative adversarial network model.

[0007] In some embodiments, determining the indoor temperature field labels and smoke visibility distribution field labels of the multiple sample buildings at the sample fire development moments includes: Determine the sample simulation sketch of each sample building, where the sample simulation sketch includes the sample wall layout; Based on the sample simulation sketch, construct the computational fluid dynamics (CFD) model of each sample building; Based on the CFD model, sample fire starting position, and sample fire development moment of each sample building, dynamically simulate the fire development process of each sample building to obtain the indoor temperature field labels and smoke visibility distribution field labels of each sample building at the sample fire development moment.

[0008] In some embodiments, determining the sample simulation sketch of each sample building includes: Obtain the sample design drawing of each sample building, and extract the key elements of the sample design drawing, where the key elements include walls and doors and windows; Process the key elements of the sample design drawing to obtain the sample simulation sketch of each sample building.

[0009] In some embodiments, before dynamically simulating the fire development process of each sample building, it further includes: Set the initial parameters of the CFD model of each sample building, where the initial parameters include the key combustion position, fire source, ceiling height, and grid size.

[0010] In some embodiments, the initial generative adversarial network model includes an initial generator and an initial discriminator; Correspondingly, the training of the initial generative adversarial network model includes: Input the sample wall layout, sample fire ignition position, and sample fire development time of the first sample building in the training dataset into the initial generator to obtain the predicted results of the indoor temperature field and the predicted results of the smoke visibility distribution field of the first sample building output by the initial generator; Input the predicted results of the indoor temperature field and the smoke visibility distribution field of the first sample building, as well as the indoor temperature field label and the smoke visibility distribution field label corresponding to the first sample building into the initial discriminator to obtain the discrimination result output by the initial discriminator; Based on the discrimination result, iteratively optimize the parameters of the initial generator and the initial discriminator to obtain the generative adversarial network model.

[0011] In some embodiments, after obtaining the generative adversarial network model, it further includes: Input the sample wall layout, sample fire ignition position, and sample fire development time of the second sample building in the test dataset into the generative adversarial network model to obtain the predicted results of the indoor temperature field and the predicted results of the smoke visibility distribution field of the second sample building output by the generative adversarial network model; Compare the predicted results of the indoor temperature field and the smoke visibility distribution field of the second sample building with the indoor temperature field label and the smoke visibility distribution field label corresponding to the second sample building to obtain the evaluation result of the generative adversarial network model; Wherein, the evaluation result includes structural similarity SSIM, normalized root mean square error NRMSE, and intersection over union IoU.

[0012] The present invention also provides an indoor fire prediction device based on a generative adversarial network, including: An acquisition unit for acquiring the wall layout, fire ignition position, and fire development time of a target building; A prediction unit for inputting the wall layout, fire ignition position, and fire development time of the target building into a pre-constructed generative adversarial network model to obtain the indoor temperature field and the smoke visibility distribution field of the target building at the fire development time output by the generative adversarial network model; Wherein, the generative adversarial network model is trained based on the sample wall layout, sample fire ignition position, and sample fire development time of a sample building, as well as the indoor temperature field label and the smoke visibility distribution field label of the sample building at the sample fire development time.

[0013] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the indoor fire prediction method based on a generative adversarial network as described in any one of the above is implemented.

[0014] The indoor fire prediction method, device, and electronic device based on a generative adversarial network provided by the present invention improve the efficiency and real-time performance of indoor fire prediction by obtaining the wall layout, fire starting position, and fire development time of a target building; inputting the wall layout, fire starting position, and fire development time of the target building into a pre-constructed generative adversarial network model, and obtaining the indoor temperature field and smoke visibility distribution field of the target building at the fire development time output by the generative adversarial network model. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0016] Figure 1 It is a flowchart of the indoor fire prediction method based on a generative adversarial network provided by an embodiment of the present invention.

[0017] Figure 2 It is an architecture diagram of the generative adversarial network model provided by an embodiment of the present invention.

[0018] Figure 3 It is a flowchart of the determination process of the generative adversarial network model provided by an embodiment of the present invention.

[0019] Figure 4 It is an architecture diagram of the generator of the generative adversarial network model provided by an embodiment of the present invention.

[0020] Figure 5 It is an architecture diagram of the discriminator of the generative adversarial network model provided by an embodiment of the present invention.

[0021] Figure 6 It is a schematic diagram of the indoor temperature field prediction result provided by an embodiment of the present invention.

[0022] Figure 7 It is a schematic diagram of the smoke visibility distribution field prediction result provided by an embodiment of the present invention.

[0023] Figure 8 It is a curve graph of the change of indoor temperature and smoke visibility with time provided by an embodiment of the present invention.

[0024] Figure 9 It is a schematic structural diagram of an indoor fire prediction device based on a generative adversarial network provided by an embodiment of the present invention.

[0025] Figure 10 It is a schematic structural diagram of an electronic device provided by an embodiment of the present invention. Detailed implementation manners

[0026] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0027] The terms "first", "second", etc. in the present invention are used to distinguish similar objects, rather than to describe a specific order or sequence. It should be understood that such terms can be interchanged under appropriate circumstances so that the embodiments of the present invention can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first" and "second" are generally of the same type, and the number of objects is not limited. For example, the first object can be one or multiple.

[0028] Figure 1 It is a schematic flow chart of an indoor fire prediction method based on a generative adversarial network provided by an embodiment of the present invention. As Figure 1 shown, an indoor fire prediction method based on a generative adversarial network is provided, including the following steps: step 110 and step 120. The flow steps of this method are only a possible implementation manner of the present invention.

[0029] Step 110: Obtain the wall layout, fire starting position and fire development time of the target building.

[0030] Optionally, obtain the design drawing of the target building, and process the design drawing of the target building to obtain the wall layout of the target building.

[0031] Optionally, the target building is a residential building, an office building, a shopping mall or other buildings.

[0032] Optionally, determine the fire starting position and fire development time through on-site investigation or according to surveillance videos.

[0033] Step 120: Input the wall layout, fire starting position and fire development time of the target building into a pre-constructed generative adversarial network model to obtain the indoor temperature field and smoke visibility distribution field of the target building at the fire development time output by the generative adversarial network model; Among them, the generative adversarial network model is trained based on the sample wall layout, sample fire ignition position, and sample fire development time of the sample building, as well as the indoor temperature field label and smoke visibility distribution field label of the sample building at the sample fire development time.

[0034] Among them, the indoor temperature field refers to the distribution state of the indoor temperature at each point of the target building slice over time; combustibles burn and release heat, forming a local high-temperature area. The hot gas rises, and the cold air sinks. During the fire growth stage, the indoor temperature rises sharply.

[0035] Among them, the smoke visibility distribution field refers to the distribution of the degree of light attenuation caused by fire smoke in the indoor space, reflecting the degree of limited vision during personnel evacuation; the influencing factors of the smoke visibility distribution field include smoke particle concentration, smoke layer height, light source conditions, etc.

[0036] Optionally, the fire development time is extended to form a time channel, which is merged with the spatial channel of the wall layout, so that the fire development time and the wall layout are in the same dimension.

[0037] Figure 2 This is the architecture diagram of the generative adversarial network model provided by the embodiment of the present invention. As Figure 2 shown, the generative adversarial network model includes an input module, a training module, and an output module. The input module is used to input the wall layout, fire ignition position, and fire development time of the target building. The output module is used to output the indoor temperature field and smoke visibility distribution field of the target building at the fire development time. The training module includes a generative network and a discriminative network. The generative network is used to predict the indoor temperature field and smoke visibility distribution field of the target building at the fire development time based on the wall layout, fire ignition position, and fire development time of the target building. The discriminative network is used to evaluate the authenticity of the data generated by the generative network to guide the generative network to generate results closer to the fire physical field.

[0038] In the embodiment of the present invention, by obtaining the wall layout, fire ignition position, and fire development time of the target building; inputting the wall layout, fire ignition position, and fire development time of the target building into the pre-constructed generative adversarial network model, the indoor temperature field and smoke visibility distribution field of the target building at the fire development time output by the generative adversarial network model are obtained, improving the efficiency and real-time performance of indoor fire prediction.

[0039] In some embodiments, the generative adversarial network model includes a feature processing layer and a prediction layer; Correspondingly, inputting the wall layout, fire ignition position, and fire development time of the target building into the pre-constructed generative adversarial network model to obtain the indoor temperature field and smoke visibility distribution field of the target building at the fire development time output by the generative adversarial network model includes: Input the wall layout, fire location, and fire development time of the target building into the feature processing layer. The feature processing layer processes the wall layout, fire location, and fire development time to obtain the feature map of the target building output by the feature processing layer. The feature map includes wall layout features, fire location features, and fire development time features. Input the feature map of the target building into the prediction layer to obtain the indoor temperature field and smoke visibility distribution field of the target building at the fire development time output by the prediction layer.

[0040] Optionally, extract features from the wall layout, fire location, and fire development time to obtain wall layout features, fire location features, and fire development time features.

[0041] Optionally, fuse the wall layout features, fire location features, and fire development time features to obtain the feature map of the target building.

[0042] Figure 3 It is a schematic flowchart of the determination process of the generative adversarial network model provided by the embodiments of the present invention. As Figure 3 shown, in some embodiments, the determination process of the generative adversarial network model includes: Obtain the sample wall layout, sample fire location, and sample fire development time of multiple sample buildings, determine the indoor temperature field labels and smoke visibility distribution field labels of multiple sample buildings at the sample fire development time, and construct a training data set and a test data set. Use the sample wall layout, sample fire location, and sample fire development time of the first sample building in the training data set as training data, and use the indoor temperature field labels and smoke visibility distribution field labels of the first sample building at the sample fire development time as training labels to train the initial generative adversarial network model. After training is completed, a generative adversarial network model is obtained.

[0043] Optionally, construct a sample database covering various building wall layouts and fire scenarios.

[0044] Optionally, preprocess the sample wall layout, sample fire location, sample fire development time, as well as the indoor temperature field labels and smoke visibility distribution field labels to unify the data dimensions and scales and ensure the consistency of data in different scenarios.

[0045] Optionally, use CFD software to perform fire dynamics simulations to generate fire evolution data under different building wall layouts and fire locations. The fire evolution data includes spatio-temporal variation information of key physical quantities (such as temperature field and smoke visibility distribution field).

[0046] Among them, the CFD software includes the Fire Dynamics Simulator (FDS); FDS is used to simulate combustion, smoke diffusion, temperature distribution, and visibility changes in a fire scenario, and supports building fire protection design, safety assessment, and fire accident reconstruction.

[0047] It can be understood that by using the computational fluid dynamics CFD method for fire numerical simulation, large-scale and high-precision training datasets and test datasets can be quickly constructed.

[0048] Optionally, for the high-dimensional complex characteristics of the fire scenario, a generative adversarial network model GAN-pix2pix based on the pix2pix architecture is constructed. GAN-pix2pix is a deep learning model based on conditional generative adversarial networks, which can learn the complex mapping relationship between the input image and the output image and generate high-quality and structurally coherent target images.

[0049] In some embodiments, determining the indoor temperature field labels and smoke visibility distribution field labels of multiple sample buildings at the sample fire development moment includes: Determining the sample simulation sketch of each sample building, where the sample simulation sketch includes the sample wall layout; Based on the sample simulation sketch, constructing a computational fluid dynamics CFD model for each sample building; Based on the CFD model, sample ignition location, and sample fire development moment of each sample building, dynamically simulating the fire development process of each sample building to obtain the indoor temperature field labels and smoke visibility distribution field labels of each sample building at the sample fire development moment.

[0050] In some embodiments, determining the sample simulation sketch of each sample building includes: Obtaining the sample design drawing of each sample building and extracting the key elements of the sample design drawing, where the key elements include walls and doors and windows; Processing the key elements of the sample design drawing to obtain the sample simulation sketch of each sample building.

[0051] Optionally, when processing the key elements of the sample design drawing, different key elements are represented by different colors.

[0052] In some embodiments, before dynamically simulating the fire development process of each sample building, it further includes: Setting the initial parameters of the CFD model of each sample building, where the initial parameters include the key combustion location, fire source, ceiling height, and grid size.

[0053] Optionally, setting the boundary conditions of the CFD model of each sample building.

[0054] In some embodiments, the initial generative adversarial network model includes an initial generator and an initial discriminator; Correspondingly, training the initial generative adversarial network model includes: Inputting the sample wall layout, sample fire starting position, and sample fire development time of the first sample building in the training dataset into the initial generator to obtain the predicted results of the indoor temperature field and the predicted results of the smoke visibility distribution field of the first sample building output by the initial generator; Inputting the predicted results of the indoor temperature field and the smoke visibility distribution field of the first sample building, as well as the indoor temperature field label and the smoke visibility distribution field label corresponding to the first sample building, into the initial discriminator to obtain the discrimination result output by the initial discriminator; Based on the discrimination result, iteratively optimize the parameters of the initial generator and the initial discriminator to obtain the generative adversarial network model.

[0055] Optionally, adopt a strategy that combines supervised learning and adversarial training. Using CFD simulation data as the supervision signal, continuously optimize the adversarial game process of the initial generator and the initial discriminator, and calculate the loss function value. Comprehensively measure the error between the generated result and the target data according to the loss function value, and improve the generation accuracy of the model through iterative optimization.

[0056] In some embodiments, after obtaining the generative adversarial network model, it further includes: Inputting the sample wall layout, sample fire starting position, and sample fire development time of the second sample building in the test dataset into the generative adversarial network model to obtain the predicted results of the indoor temperature field and the predicted results of the smoke visibility distribution field of the second sample building output by the generative adversarial network model; Comparing the predicted results of the indoor temperature field and the smoke visibility distribution field of the second sample building with the indoor temperature field label and the smoke visibility distribution field label corresponding to the second sample building to obtain the evaluation result of the generative adversarial network model; Among them, the evaluation results include Structural Similarity Index Measure (SSIM), Normalized Root Mean Square Error (NRMSE), and Intersection over Union (IoU).

[0057] Among them, SSIM is used to measure the similarity of the structure, brightness, and contrast between the prediction result and the true value, with a value range of [-1, 1]. The closer the value is to 1, the higher the similarity between the two; NRMSE is a normalized metric for measuring the error between the predicted value and the true value. The smaller the error, the better the prediction performance; IoU is used to measure the overlap degree between the predicted region and the true region, with a value range of [0, 1]. The larger the value, the higher the overlap degree between the prediction result and the true value.

[0058] Optionally, compare the influence of generators and discriminators with different network architectures on the prediction results, and conduct hyperparameter analysis to determine the optimal model configuration.

[0059] Table 1 is an evaluation table of generators and discriminators with different network architectures provided by the embodiments of the present invention. As shown in Table 1, the network types of the generators include ResNet and U-Net. The number of ResNet residual blocks is 9 or 6. The input size of U-Net-256 is 256×256 pixels, and the input size of U-Net-128 is 128×128 pixels; the network types of the discriminators include basic, n-layers, and pixel; the comprehensive evaluation results (i.e., structural similarity SSIM, normalized root mean square error NRMSE, and intersection over union IoU) of the network architecture combining U-Net-256 and pixel are the best.

[0060] Table 1 Evaluation Table of Generators and Discriminators with Different Network Architectures Among them, the basic discriminator is the simplest discriminator structure, usually composed of multiple fully connected layers or shallow convolutional networks. The n-layers discriminator is a discriminator based on a convolutional neural network, which gradually extracts features by stacking multiple convolutional layers. The pixel discriminator independently judges the authenticity of each pixel point and is usually implemented in combination with a fully convolutional network.

[0061] Figure 4 is the architecture diagram of the generator of the generative adversarial network model provided by the embodiments of the present invention. As Figure 4 shown, the network of the generator of the generative adversarial network model is U-Net-256. U-Net-256 includes an encoder and a decoder, and there is a skip connection between the encoder and the decoder; the encoder consists of 8 downsamplings, and each layer contains a convolutional block and a max pooling layer; the decoder gradually restores the resolution through upsampling and skip connections.

[0062] Figure 5 is the architecture diagram of the discriminator of the generative adversarial network model provided by the embodiments of the present invention. As Figure 5As shown, the network of the generator of the generative adversarial network model is pixel. Pixel independently judges the authenticity of each pixel of the input image and outputs a discrimination matrix with the same size as the input (such as 256×256×1). Each element represents the probability that the corresponding pixel is real data.

[0063] Optionally, in order to further analyze the prediction performance of the generative adversarial network model, a typical case is randomly selected from the test set, and the indoor temperature field and the smoke visibility distribution field at critical moments (100 s, 200 s, 300 s) are compared respectively. The results are as Figures 6 - 7 .

[0064] Figure 6 It is a schematic diagram of the indoor temperature field prediction result provided by the embodiment of the present invention. As Figure 6 shown, the indoor temperature field prediction result generated by the generative adversarial network model is highly consistent with the indoor temperature field simulation result obtained based on FDS (i.e., the true label of the indoor temperature field) in terms of the overall trend and local details. Especially, the temperature distribution near the fire source is similar to the convection-diffusion mode.

[0065] Figure 7 It is a schematic diagram of the smoke visibility distribution field prediction result provided by the embodiment of the present invention. The smoke visibility distribution field prediction result generated by the generative adversarial network model is basically consistent with the smoke visibility distribution field simulation result obtained based on FDS (i.e., the true label of the smoke visibility distribution field) in terms of the spatial distribution and concentration gradient, indicating that the generative adversarial network model can better learn the characteristics of fire smoke flow; in some complex scenarios (such as near obstacles), there is a slight smoothing effect on the smoke diffusion predicted by the generative adversarial network model, but the overall error is small.

[0066] Table 2 is the evaluation table of the generative adversarial network model provided by the embodiment of the present invention. As shown in Table 2, the mean value of the structural similarity SSIM is generally high. The mean value of SSIM for all test cases is above 0.93, and the highest value (0.9796) is for case 39, indicating that the generated image is highly consistent with the structural information of the real fire scene; the mean value of the normalized root mean square error NRMSE is low and the error is small. The mean value of NRMSE for all test cases is less than 0.02, and the lowest value (0.0081) is for case 39, indicating that the prediction error of the generative adversarial network model at the pixel level is small; the mean value of the intersection over union IoU is high and the spatial matching is good. The IoU for all test cases is above 0.89, and the highest value (0.9856) is for case 18, indicating that the generative adversarial network model can accurately predict the spatial distribution of fire smoke and temperature; generally speaking, the generative adversarial network model can accurately predict the fire dynamic process, and the indoor temperature field and the smoke visibility distribution field generated by it are highly consistent with the FDS calculation results.

[0067] Table 2 Evaluation Table of Generative Adversarial Network Model Optionally, in order to further quantitatively analyze the prediction accuracy of the generative adversarial network model, a comparative analysis was conducted on the prediction results of the temperature and smoke visibility at the key monitoring positions (sensors) of 1 case in the test dataset, and curves of the temperature and smoke visibility at the corresponding positions changing with time were plotted, as Figure 8 shown

[0068] Figure 8 is the curve graph of the indoor temperature and smoke visibility changing with time provided by the embodiment of the present invention. As Figure 8 shown, the changing trends of the temperature and smoke visibility predicted by the generative adversarial network model are basically consistent with the simulation results of the benchmark simulation tool FDS. Specifically, the temperature curve highly coincides with the FDS results in both the heating process at the initial stage and the final stable stage, and the prediction error is small; the changing trend of the smoke visibility also accurately reflects the process of rapid increase and subsequent gradual decrease of the smoke visibility, and can capture the key dynamic characteristics

[0069] Next, the indoor fire prediction device based on the generative adversarial network provided by the embodiment of the present invention will be described. The indoor fire prediction device based on the generative adversarial network described below can be correspondingly referred to the indoor fire prediction method based on the generative adversarial network described above

[0070] Figure 9 is the structural schematic diagram of the indoor fire prediction device based on the generative adversarial network provided by the embodiment of the present invention. As Figure 9 shown, the indoor fire prediction device 900 based on the generative adversarial network includes an acquisition unit 910, configured to acquire the wall layout, fire starting position, and fire development time of the target building a prediction unit 920, configured to input the wall layout, fire starting position, and fire development time of the target building into a pre-constructed generative adversarial network model, and obtain the indoor temperature field and smoke visibility distribution field of the target building at the fire development time output by the generative adversarial network model wherein, the generative adversarial network model is trained based on the sample wall layout, sample fire starting position, and sample fire development time of the sample building, as well as the indoor temperature field label and smoke visibility distribution field label of the sample building at the sample fire development time

[0071] Optionally, the generative adversarial network model includes a feature processing layer and a prediction layer Correspondingly, input the wall layout, fire location, and fire development time of the target building into the pre-constructed generative adversarial network model, and obtain the indoor temperature field and smoke visibility distribution field of the target building at the fire development time output by the generative adversarial network model, including: Input the wall layout, fire location, and fire development time of the target building into the feature processing layer. The feature processing layer performs feature processing on the wall layout, fire location, and fire development time to obtain the feature map of the target building output by the feature processing layer; the feature map includes wall layout features, fire location features, and fire development time features. Input the feature map of the target building into the prediction layer to obtain the indoor temperature field and smoke visibility distribution field of the target building at the fire development time output by the prediction layer.

[0072] Optionally, the determination process of the generative adversarial network model includes: Obtain the sample wall layout, sample fire location, and sample fire development time of multiple sample buildings, determine the indoor temperature field labels and smoke visibility distribution field labels of multiple sample buildings at the sample fire development time, and construct a training data set and a test data set. Use the sample wall layout, sample fire location, and sample fire development time of the first sample building in the training data set as training data, and use the indoor temperature field label and smoke visibility distribution field label of the first sample building at the sample fire development time as training labels to train the initial generative adversarial network model. After training is completed, obtain the generative adversarial network model.

[0073] Optionally, determining the indoor temperature field labels and smoke visibility distribution field labels of multiple sample buildings at the sample fire development time includes: Determine the sample simulation sketch of each sample building, and the sample simulation sketch includes the sample wall layout. Based on the sample simulation sketch, construct the computational fluid dynamics (CFD) model of each sample building. Based on the CFD model, sample fire location, and sample fire development time of each sample building, dynamically simulate the fire development process of each sample building to obtain the indoor temperature field labels and smoke visibility distribution field labels of each sample building at the sample fire development time.

[0074] Optionally, determining the sample simulation sketch of each sample building includes: Obtain the sample design drawing of each sample building, extract the key elements of the sample design drawing, and the key elements include walls and doors and windows. Process the key elements of the sample design drawing to obtain the sample simulation sketch of each sample building.

[0075] Optionally, the indoor fire prediction device based on the generative adversarial network further includes: A configuration unit for setting initial parameters of the CFD model of each sample building, where the initial parameters include key combustion positions, fire sources, ceiling heights, and grid sizes.

[0076] Optionally, the initial generative adversarial network model includes an initial generator and an initial discriminator; Correspondingly, training the initial generative adversarial network model includes: Inputting the sample wall layout, sample ignition position, and sample fire development time of the first sample building in the training dataset into the initial generator to obtain the predicted results of the indoor temperature field and the predicted results of the smoke visibility distribution field of the first sample building output by the initial generator; Inputting the predicted results of the indoor temperature field and the predicted results of the smoke visibility distribution field of the first sample building, as well as the indoor temperature field label and the smoke visibility distribution field label corresponding to the first sample building, into the initial discriminator to obtain the discrimination result output by the initial discriminator; Based on the discrimination result, iteratively optimizing the parameters of the initial generator and the initial discriminator to obtain the generative adversarial network model.

[0077] Optionally, the indoor fire prediction device based on the generative adversarial network further includes: A testing unit for inputting the sample wall layout, sample ignition position, and sample fire development time of the second sample building in the test dataset into the generative adversarial network model to obtain the predicted results of the indoor temperature field and the predicted results of the smoke visibility distribution field of the second sample building output by the generative adversarial network model; An evaluation unit for comparing the predicted results of the indoor temperature field and the predicted results of the smoke visibility distribution field of the second sample building with the indoor temperature field label and the smoke visibility distribution field label corresponding to the second sample building to obtain the evaluation result of the generative adversarial network model; Wherein, the evaluation result includes structural similarity SSIM, normalized root mean square error NRMSE, and intersection over union IoU.

[0078] It should be noted here that the indoor fire prediction device based on the generative adversarial network provided in the embodiment of the present invention can implement all the method steps implemented by the embodiment of the indoor fire prediction method based on the generative adversarial network, and can achieve the same technical effects. The same parts and beneficial effects as those in the method embodiment in this embodiment will not be specifically described herein.

[0079] Figure 10 It is a schematic structural diagram of the electronic device provided in the embodiment of the present invention, as Figure 10As shown in the figure, the electronic device may include: a processor 1010, a communications interface 1020, a memory 1030, and a communication bus 1040. Among them, the processor 1010, the communications interface 1020, and the memory 1030 complete communication with each other through the communication bus 1040. The processor 1010 may call the logical instructions in the memory 1030 to execute an indoor fire prediction method based on a generative adversarial network. The method includes: obtaining the wall layout, the fire ignition position, and the fire development time of a target building; inputting the wall layout, the fire ignition position, and the fire development time of the target building into a feature processing layer of a pre-constructed generative adversarial network model, and the feature processing layer performs feature processing on the wall layout, the fire ignition position, and the fire development time to obtain a feature map of the target building output by the feature processing layer; the feature map includes wall layout features, fire ignition position features, and fire development time features; inputting the feature map of the target building into a prediction layer of the generative adversarial network model to obtain an indoor temperature field and a smoke visibility distribution field of the target building at the fire development time output by the prediction layer; wherein, the generative adversarial network model is trained based on the sample wall layout, the sample fire ignition position, and the sample fire development time of a sample building, as well as the indoor temperature field label and the smoke visibility distribution field label of the sample building at the sample fire development time.

[0080] In addition, when the logical instructions in the above-mentioned memory 1030 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs that can store program codes.

[0081] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative effort.

[0082] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0083] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments or equivalently replace some of the technical features. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An indoor fire prediction method based on a generative adversarial network, characterized in that Including: Obtaining the wall layout, fire starting position, and fire development time of the target building; Inputting the wall layout, fire starting position, and fire development time of the target building into the feature processing layer of a pre-constructed generative adversarial network model, and performing feature processing on the wall layout, fire starting position, and fire development time by the feature processing layer to obtain the feature map of the target building output by the feature processing layer; the feature map includes wall layout features, fire starting position features, and fire development time features; Inputting the feature map of the target building into the prediction layer of the generative adversarial network model to obtain the indoor temperature field and smoke visibility distribution field of the target building at the fire development time output by the prediction layer; Wherein, the generative adversarial network model is trained based on the sample wall layout, sample fire starting position, and sample fire development time of the sample building, as well as the indoor temperature field label and smoke visibility distribution field label of the sample building at the sample fire development time.

2. The indoor fire prediction method based on a generative adversarial network according to claim 1, wherein The determination process of the generative adversarial network model includes: Obtaining the sample wall layout, sample fire starting position, and sample fire development time of multiple sample buildings, determining the indoor temperature field label and smoke visibility distribution field label of the multiple sample buildings at the sample fire development time, and constructing a training data set and a test data set; Using the sample wall layout, sample fire starting position, and sample fire development time of the first sample building in the training data set as training data, and using the indoor temperature field label and smoke visibility distribution field label of the first sample building at the sample fire development time as training labels to train an initial generative adversarial network model, and after training is completed, obtaining the generative adversarial network model.

3. The indoor fire prediction method based on a generative adversarial network according to claim 2, wherein The determination of the indoor temperature field label and smoke visibility distribution field label of the multiple sample buildings at the sample fire development time includes: Determining the sample simulation sketch of each sample building, and the sample simulation sketch includes the sample wall layout; Based on the sample simulation sketch, constructing a computational fluid dynamics (CFD) model for each sample building; Based on the CFD model, sample fire starting position, and sample fire development time of each sample building, dynamically simulating the fire development process of each sample building to obtain the indoor temperature field label and smoke visibility distribution field label of each sample building at the sample fire development time.

4. The indoor fire prediction method based on a generative adversarial network according to claim 3, characterized in that, The determination of the sample simulation sketch of each sample building includes: Obtaining the sample design drawing of each sample building, and extracting the key elements of the sample design drawing, and the key elements include walls and doors and windows; Processing the key elements of the sample design drawing to obtain the sample simulation sketch of each sample building.

5. The indoor fire prediction method based on a generative adversarial network according to claim 3, wherein, Before dynamically simulating the fire development process of each sample building, it further includes: Setting the initial parameters of the CFD model of each sample building, and the initial parameters include the key combustion position, fire source, ceiling height, and grid size.

6. The indoor fire prediction method based on a generative adversarial network according to claim 2, characterized in that The initial generative adversarial network model includes an initial generator and an initial discriminator; Correspondingly, the training of the initial generative adversarial network model includes: Input the sample wall layout, sample fire starting position, and sample fire development time of the first sample building in the training dataset into the initial generator, and obtain the predicted results of the indoor temperature field and the predicted results of the smoke visibility distribution field of the first sample building output by the initial generator; Input the predicted results of the indoor temperature field and the predicted results of the smoke visibility distribution field of the first sample building, as well as the indoor temperature field label and the smoke visibility distribution field label corresponding to the first sample building into the initial discriminator, and obtain the discrimination result output by the initial discriminator; Based on the discrimination result, iteratively optimize the parameters of the initial generator and the initial discriminator to obtain the generative adversarial network model.

7. The indoor fire prediction method based on a generative adversarial network according to claim 2, wherein, After obtaining the generative adversarial network model, it further includes: Input the sample wall layout, sample fire starting position, and sample fire development time of the second sample building in the test dataset into the generative adversarial network model, and obtain the predicted results of the indoor temperature field and the predicted results of the smoke visibility distribution field of the second sample building output by the generative adversarial network model; Compare the predicted results of the indoor temperature field and the predicted results of the smoke visibility distribution field of the second sample building with the indoor temperature field label and the smoke visibility distribution field label corresponding to the second sample building to obtain the evaluation result of the generative adversarial network model; Among them, the evaluation result includes the structural similarity SSIM, the normalized root mean square error NRMSE, and the intersection over union IoU.

8. An indoor fire prediction device based on a generative adversarial network, characterized in that, It includes: An acquisition unit for acquiring the wall layout, fire starting position, and fire development time of the target building; A prediction unit for inputting the wall layout, fire starting position, and fire development time of the target building into the feature processing layer of a pre-constructed generative adversarial network model, and the feature processing layer performs feature processing on the wall layout, fire starting position, and fire development time to obtain the feature map of the target building output by the feature processing layer; the feature map includes wall layout features, fire starting position features, and fire development time features; input the feature map of the target building into the prediction layer of the generative adversarial network model to obtain the indoor temperature field and the smoke visibility distribution field of the target building at the fire development time output by the prediction layer; Among them, the generative adversarial network model is trained based on the sample wall layout, sample fire starting position, and sample fire development time of the sample building, as well as the indoor temperature field label and the smoke visibility distribution field label of the sample building at the sample fire development time.

9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the indoor fire prediction method based on the generative adversarial network according to any one of claims 1 to 7.

Images