A deep neural network-based indoor fire evolution prediction method

Abstract

The present disclosure relates to a deep neural network-based indoor fire evolution prediction method and device, electronic equipment and storage medium. The method comprises: constructing a fire evolution simulation based on FDS, setting physical model data and key parameters of the fire evolution model, and establishing simulation data and fitting data of a preset fire scene; taking the simulation data as input, generating a temperature field map based on a preset first deep neural network model and a preset second deep neural network model; when a fire occurs, receiving on-site information, and generating a predicted temperature field map based on the preset deep neural network model according to the on-site information to complete the prediction of the fire evolution. The present disclosure improves the real-time performance and accuracy of fire evolution prediction by designing a two-stage deep neural network to reconstruct the input vector with high precision and fitting the simulation data of the field model.

Description

Technical Field

[0001] This disclosure relates to the field of disaster prediction, and more specifically, to a method, apparatus, electronic device, and computer-readable storage medium for predicting the evolution of indoor fires based on deep neural networks. Background Technology

[0002] Fire is a frequent and extremely dangerous disaster in people's daily lives. As high-rise buildings become more and more densely packed in cities, building fires are occurring more and more frequently. However, the technology for precise fire response is still relatively lagging behind. For example, the ambiguity in locating the disaster point affects the effectiveness of precise response, and the lack of rapid perception of the concentration of toxic smoke in the fire area threatens people's lives. In order to conduct evacuation more scientifically and effectively and reduce personnel and property losses, we need to predict the simulated evolution of fires to assist in emergency rescue command and decision-making at the fire scene.

[0003] Generally, depending on the different fire phenomena being simulated, the mathematical models currently used mainly include field models, region models, network models, and hybrid models combining these three types. Field models divide the fire chamber into many grids and obtain the various state parameters during the fire process by solving continuity equations, etc. Field models have high computer performance requirements and long computation times. Region models typically divide each room into two regions, assuming that the physical quantities within each region are uniform, and then derive a set of ordinary differential governing equations based on the principles of mass and energy conservation and the ideal gas law. Network models treat each confined space of the building as a unit, applying mass balance equations, etc., to model changes in temperature and smoke concentration.

[0004] In existing technologies, a fire trend prediction method based on deep learning and fire monitoring videos converts fire images into grayscale matrices and uses a distributed LSTM ensemble prediction model to predict the grayscale values ​​of the corresponding columns in the next frame. However, this method, which converts fire images into grayscale matrices for calculation, obtains incomplete information and has low accuracy. A fire evolution simulation method and its integrated fire evacuation simulation method, based on the mesoscopic lattice Boltzmann method, integrates relevant theoretical methods to establish a two-dimensional fire simulation model tailored to the characteristics of fires in subway stations. This method requires solving the Navier-Stokes equations, resulting in poor timeliness. In the race against time for emergency rescue, FDS is difficult to play a rapid role, and modeling based on the Navier-Stokes equations is time-consuming, requires professional computational fluid dynamics knowledge, and places high demands on the professional knowledge and practical skills of fire emergency rescue personnel. The cumbersome operation steps and learning costs make FDS less than ideal for use in tense fire rescue work. The multi-data fusion intelligent forest fire identification system and method fuses multiple data to predict forest fires. However, this method requires the fusion of multiple data, such as infrared thermal imagers, digital cameras, ultrasonic weather stations, geographic information systems, and host computers, etc., which consumes huge resources. In contrast, we only need to use the FDS building plan database.

[0005] Therefore, one or more methods are needed to solve the above problems.

[0006] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0007] The purpose of this disclosure is to provide a method, apparatus, electronic device, and computer-readable storage medium for predicting the evolution of indoor fires based on deep neural networks, thereby overcoming, at least to some extent, one or more problems caused by the limitations and defects of related technologies.

[0008] According to one aspect of this disclosure, a method for predicting the evolution of indoor fires based on deep neural networks is provided, comprising:

[0009] Based on a preset fire scenario, a fire evolution simulation based on the fire dynamics simulation tool FDS is constructed, and the physical model data and key parameters of the fire evolution model are set. Based on the fire evolution simulation, simulation data and fitting data of the preset fire scenario are established.

[0010] Using the simulated data as input, a first temperature field map is generated based on a preset first deep neural network model, and using the first temperature field map as input, a second temperature field map is generated based on a preset second deep neural network model.

[0011] A preset first deep neural network model and a preset second deep neural network model are configured on a preset terminal to correspond to the second temperature field map of a preset fire scenario. When a fire occurs, the terminal receives on-site information and generates a predicted temperature field map based on the preset first deep neural network model and the preset second deep neural network model according to the on-site information, thereby completing the prediction of the fire evolution.

[0012] In one exemplary embodiment of this disclosure, the method further includes:

[0013] The preset fire scenarios are fire scenarios in factories, commercial complexes, high-rise buildings, office buildings / teaching buildings, and apartments / residential buildings;

[0014] Physical model data is collected based on the preset fire scenario. The physical model data includes room dimensions, door dimensions, window dimensions, location of combustibles, and fuel used for combustion.

[0015] The key parameters of the fire evolution model are analyzed based on the Monte Carlo method, and the key parameters of the fire evolution model are generated, including the heat release rate per unit area, temperature, growth coefficient, and simulation time.

[0016] In one exemplary embodiment of this disclosure, the method further includes:

[0017] Based on the fire evolution simulation of a preset fire scenario, fitted data corresponding to the preset fire scenario is generated, and the fitted data is divided into a training set and a validation set according to a preset ratio.

[0018] In one exemplary embodiment of this disclosure, the preset first deep neural network model in the method further includes:

[0019] Using the ReLU activation function and the squared error function as the loss function, the indoor geometry (length, width, and height), heat release rate of the fire source, location of the ventilation opening, and time of the preset fire scenario are used as input vectors. The system is connected to a three-layer fully connected neural network and a dropout layer to perform deconvolution and convolution operations, generating the first temperature field map output.

[0020] In one exemplary embodiment of this disclosure, the preset second deep neural network model in the method further includes an encoder, a hidden layer, and a decoder, wherein:

[0021] The preset second deep neural network model uses the ReLU activation function and the squared error function as the loss function.

[0022] The encoder is composed of a convolutional neural network. The convolutional layer of the convolutional neural network has a 3*3 kernel, padding of 1, and stride of 2. The encoder is used to compress the latitude of the first temperature field map to a first preset size.

[0023] The hidden layer is a four-layer fully connected neural network, which is used to compress the first temperature field map of the first preset size from two-dimensional data to one-dimensional data.

[0024] The decoder is composed of convolutional neural networks. The convolutional kernel of the convolutional layer of the convolutional neural network has a 3*3 kernel, padding of 1, and stride of 1. The decoder is used to insert pixels into the first temperature field map of the one-dimensional data to complete the image reconstruction with high accuracy and generate a second temperature field map.

[0025] In one exemplary embodiment of this disclosure, the method further includes:

[0026] The fitted data generated by the fire evolution simulation of the preset fire scenario is divided into training set, validation set and test set according to a preset ratio;

[0027] Based on the Adam optimizer, the preset first deep neural network model and the preset second deep neural network model are trained using the stochastic gradient descent method.

[0028] When the loss of the test set no longer decreases for a consecutive preset number of rounds, the training of the preset first deep neural network model and the preset second deep neural network model is completed.

[0029] In one exemplary embodiment of this disclosure, the method further includes:

[0030] At the time of the fire, the on-site information was temperature data collected by a throwable sensor.

[0031] In one aspect of this disclosure, a deep neural network-based indoor fire evolution prediction device is provided, comprising:

[0032] The data fitting module is used to construct a fire evolution simulation based on the fire dynamics simulation tool FDS based on a preset fire scenario, and to complete the setting of physical model data and key parameters of the fire evolution model. Based on the fire evolution simulation, it establishes simulation data and fitting data for the preset fire scenario.

[0033] The model training module is used to generate a first temperature field map based on a preset first deep neural network model, using the simulated data as input, and to generate a second temperature field map based on a preset second deep neural network model, using the first temperature field map as input.

[0034] The fire evolution prediction module is used to configure a preset first deep neural network model and a preset second deep neural network model corresponding to the second temperature field map of a preset fire scenario on a preset terminal. When a fire occurs, it receives on-site information and generates a predicted temperature field map based on the preset first deep neural network model and the preset second deep neural network model according to the on-site information, thereby completing the prediction of fire evolution.

[0035] In one aspect of this disclosure, an electronic device is provided, comprising:

[0036] Processor; and

[0037] A memory storing computer-readable instructions that, when executed by the processor, implement the method according to any one of the preceding claims.

[0038] In one aspect of this disclosure, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method according to any one of the preceding claims.

[0039] An exemplary embodiment of this disclosure discloses a method for predicting indoor fire evolution based on a deep neural network. The method includes: constructing a fire evolution simulation based on a Fire Data System (FDS), setting physical model data and key parameters of the fire evolution model, and establishing simulation data and fitting data for a preset fire scenario; using the simulation data as input, generating a temperature field map based on a preset first deep neural network model and a preset second deep neural network model; and receiving on-site information when a fire occurs, generating a predicted temperature field map based on the preset deep neural network model according to the on-site information, thereby completing the prediction of fire evolution. This disclosure improves the real-time performance and accuracy of fire evolution prediction by designing a two-stage deep neural network to reconstruct the input vector with high precision and fitting the simulation data of the field model.

[0040] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0041] The above and other features and advantages of this disclosure will become more apparent from the detailed description of exemplary embodiments thereof with reference to the accompanying drawings.

[0042] Figure 1 A flowchart of an indoor fire evolution prediction method based on a deep neural network according to an exemplary embodiment of the present disclosure is shown;

[0043] Figure 2A schematic diagram of a preset first deep neural network model of an indoor fire evolution prediction method based on a deep neural network according to an exemplary embodiment of the present disclosure is shown.

[0044] Figure 3 A schematic diagram of a preset second deep neural network model for an indoor fire evolution prediction method based on a deep neural network according to an exemplary embodiment of the present disclosure is shown.

[0045] Figure 4 A schematic block diagram of an indoor fire evolution prediction device based on a deep neural network according to an exemplary embodiment of the present disclosure is shown.

[0046] Figure 5 A block diagram of an electronic device according to an exemplary embodiment of the present disclosure is schematically shown; and

[0047] Figure 6 The illustration shows a schematic diagram of a computer-readable storage medium according to an exemplary embodiment of the present disclosure. Detailed Implementation

[0048] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the embodiments set forth herein; rather, they are provided so that this disclosure will be thorough and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted.

[0049] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a thorough understanding of embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced without one or more of the specific details described, or other methods, components, materials, apparatuses, steps, etc., can be employed. In other instances, well-known structures, methods, apparatuses, implementations, materials, or operations are not shown or described in detail to avoid obscuring various aspects of this disclosure.

[0050] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, or in one or more software-hardened modules, or in different network and / or processor devices and / or microcontroller devices.

[0051] In this example embodiment, a method for predicting the evolution of indoor fires based on deep neural networks is first provided; (Refer to...) Figure 1As shown, this method for predicting the evolution of indoor fires based on deep neural networks may include the following steps:

[0052] Step S110: Based on the preset fire scenario, construct a fire evolution simulation based on the fire dynamics simulation tool FDS, and complete the setting of physical model data and key parameters of the fire evolution model. Based on the fire evolution simulation, establish simulation data and fitting data of the preset fire scenario.

[0053] Step S120: Using the simulated data as input, a first temperature field map is generated based on a preset first deep neural network model, and using the first temperature field map as input, a second temperature field map is generated based on a preset second deep neural network model.

[0054] Step S130: Configure a preset first deep neural network model and a preset second deep neural network model corresponding to the second temperature field map of the preset fire scenario on a preset terminal. When a fire occurs, receive on-site information and generate a predicted temperature field map based on the preset first deep neural network model and the preset second deep neural network model according to the on-site information, so as to complete the prediction of the fire evolution.

[0055] An exemplary embodiment of this disclosure discloses a method for predicting indoor fire evolution based on a deep neural network. The method includes: constructing a fire evolution simulation based on a Fire Data System (FDS), setting physical model data and key parameters of the fire evolution model, and establishing simulation data and fitting data for a preset fire scenario; using the simulation data as input, generating a temperature field map based on a preset first deep neural network model and a preset second deep neural network model; and receiving on-site information when a fire occurs, generating a predicted temperature field map based on the preset deep neural network model according to the on-site information, thereby completing the prediction of fire evolution. This disclosure improves the real-time performance and accuracy of fire evolution prediction by designing a two-stage deep neural network to reconstruct the input vector with high precision and fitting the simulation data of the field model.

[0056] The following will further explain an indoor fire evolution prediction method based on deep neural networks in this example embodiment.

[0057] In step S110, a fire evolution simulation based on the fire dynamics simulation tool FDS can be constructed based on a preset fire scenario, and the physical model data and key parameters of the fire evolution model can be set. Simulation data and fitting data of the preset fire scenario can be established based on the fire evolution simulation.

[0058] In this example embodiment, the method further includes:

[0059] The preset fire scenarios are fire scenarios in factories, commercial complexes, high-rise buildings, office buildings / teaching buildings, and apartments / residential buildings;

[0060] Physical model data is collected based on the preset fire scenario. The physical model data includes room dimensions, door dimensions, window dimensions, location of combustibles, and fuel used for combustion.

[0061] The key parameters of the fire evolution model are analyzed based on the Monte Carlo method, and the key parameters of the fire evolution model are generated, including the heat release rate per unit area, temperature, growth coefficient, and simulation time.

[0062] In this example embodiment, the method further includes:

[0063] Based on the fire evolution simulation of a preset fire scenario, fitted data corresponding to the preset fire scenario is generated, and the fitted data is divided into a training set and a validation set according to a preset ratio.

[0064] In this example embodiment, the first step is to construct a preset FDS scenario: This involves simulating smoke in indoor fire scenarios within buildings. Smoke diffusion varies across different building types. Therefore, we perform physical modeling for five specific indoor fire scenarios: factories, commercial complexes, high-rise buildings, office / educational buildings, and apartments / residential buildings. Based on these physical models, we set the input data for FDS, enabling targeted simulation of these five specific indoor fire scenarios. Next, we use the Monte Carlo method to analyze key parameters of the fire evolution model, primarily combustible material distribution, wind speed, temperature, and growth rate. These parameters influence the simulation results to varying degrees. For simplicity, we assume these parameters follow a normal distribution. The physical model data for the five fire scenarios (i.e., room dimensions, door dimensions, window dimensions, location of combustible material, and fuel type) and FDS parameters (heat release rate per unit area, temperature, growth coefficient, and simulation time) are input into the FDS software. The software then outputs the smoke concentration at a single location at different times and the temperature field map T of each point on the selected plane over time, which are then compared with the actual values.

[0065] In this example embodiment, the second step involves constructing the model input data: the deep learning model relies on the simulated data from the Fire Detection and Ranging (FDS). Therefore, the fitting results of the FDS for different fire scenarios determine the upper limit of the deep learning model. Taking single-room fire prediction as an example, 50 sets of FDS prediction data with different ventilation opening locations, different fire source locations, and different fire time periods are selected as training data. Each FDS prediction duration is 360 seconds, and the grid is divided into 0.03m x 0.03m x 0.03m sections. The obtained FDS fitting data is then augmented.

[0066] In step S120, the simulated data can be used as input to generate a first temperature field map based on a preset first deep neural network model, and the first temperature field map can be used as input to generate a second temperature field map based on a preset second deep neural network model.

[0067] In this example embodiment, the preset first deep neural network model in the method further includes:

[0068] Using the ReLU activation function and the squared error function as the loss function, the indoor geometry (length, width, and height), heat release rate of the fire source, location of the ventilation opening, and time of the preset fire scenario are used as input vectors. The system is connected to a three-layer fully connected neural network and a dropout layer to perform deconvolution and convolution operations, generating the first temperature field map output.

[0069] In the embodiments of this example, as Figure 2 As shown, the preset first deep neural network model aims to predict more detailed outputs using a small number of input parameters. The model has a total of 11 layers. First, the vector is input into a three-layer fully connected neural network to obtain a high-dimensional feature vector representation. A dropout layer is placed after the first fully connected layer to prevent overfitting during training. Then, the temperature field map output is reconstructed through three consecutive deconvolutional and convolutional layers. The purpose of the deconvolutional layer is to increase the receptive field, allowing subsequent convolutional kernels to learn more global information. The purpose of the convolutional layer is to extract different features from the input. The first convolutional layer may only extract some low-level features such as edges, lines, and corners; more layers of the network can iteratively extract more complex features from low-level features. The x-axis and y-axis resolution of the data are increased exponentially through convolutional operations, finally yielding the temperature field map. Each layer of the model uses the ReLU activation function, and the model is trained using stochastic gradient descent with a mean squared error function as the loss function.

[0070] In this example embodiment, step S121: the input vector is represented as v∈R 1*35 (Including basic indoor geometry (length, width, height), heat release rate of fire source, location of ventilation opening and time), the vector is passed through three layers of fully connected neural network, where the first fully connected layer is 256*1, the dropout layer is 256*1, the second fully connected layer is 512*1, and the last fully connected layer is 1024*1, finally obtaining a 1024*1 vector, and its dimension is adjusted to a high-dimensional feature vector of 32*32*1;

[0071] In the embodiment of this example, step S122: The 32*32*1 high-dimensional feature vector is input into three consecutive deconvolutional layers and convolutional layers, with the number of convolutional kernels being 48, 12, and 3 respectively. Each deconvolutional layer uses a 5*5 kernel with a stride of 2 and half-filled convolutional kernels. After each deconvolutional layer, a corresponding convolutional layer with a 5*5 kernel size, a stride of 1, and half-filled convolutional kernels is added to improve the quality of the generated image, ultimately resulting in a 256*256*3 feature vector.

[0072] In this example embodiment, step S123: Output a 256*256*3 temperature field map T l Calculate the loss function and backpropagate to update the model. The loss function is:

[0073]

[0074] Where k is the number of input vectors.

[0075] In this example embodiment, the preset second deep neural network model in the method further includes an encoder, a hidden layer, and a decoder, wherein:

[0076] The preset second deep neural network model uses the ReLU activation function and the squared error function as the loss function.

[0077] The encoder is composed of a convolutional neural network. The convolutional layer of the convolutional neural network has a 3*3 kernel, padding of 1, and stride of 2. The encoder is used to compress the latitude of the first temperature field map to a first preset size.

[0078] The hidden layer is a four-layer fully connected neural network, which is used to compress the first temperature field map of the first preset size from two-dimensional data to one-dimensional data.

[0079] The decoder is composed of convolutional neural networks. The convolutional kernel of the convolutional layer of the convolutional neural network has a 3*3 kernel, padding of 1, and stride of 1. The decoder is used to insert pixels into the first temperature field map of the one-dimensional data to complete the image reconstruction with high accuracy and generate a second temperature field map.

[0080] In the embodiments of this example, as Figure 3As shown, since the output at this point is a coarse output after inverse transformation, a large amount of detailed information in the temperature field map is lost, failing to meet the 256*256 spatiotemporal resolution requirement. Therefore, information recovery is performed using the pre-set second deep neural network model, approximating the true output of FDS through encoding and decoding networks. The model is derived from an autoencoder and consists of three parts: an encoder, hidden layers, and a decoder. The purpose is to reduce the dimensionality of the data, because the temperature field map obtained in the first stage is too high-dimensional. Therefore, we hope to encode it into a low-dimensional feature vector z = E(x) through the encoder E. The principle of encoding is to retain as much original information as possible. Therefore, we train a decoder D, hoping to reconstruct the original information through z, i.e., x ≈ D(E(x)). The optimization objective is as follows:

[0081]

[0082] Both encoder E and decoder D are composed of different convolutional neural network combinations, with a four-layer fully connected neural network as the hidden layer. Each convolutional neural network layer in encoder E is called a downsampling layer, including a convolutional layer with a 3x3 kernel, padding of 1, stride of 2, and LeakyReLU activation function. Its function is to compress the image dimensions to a specified size. In decoder D, convolutional upsampling is performed to enlarge the image size by inserting new pixels between the original image pixels, thereby increasing the image size. The upsampling layer uses bilinear interpolation, which utilizes four existing pixel values ​​near the interpolation point in the original image to jointly determine the pixel value of the interpolation point. This method achieves anti-aliasing and includes a convolutional layer with a 3x3 kernel, padding of 1, stride of 1, and LeakyReLU activation function to learn more information.

[0083] In this example embodiment, step S124: the input is the three-channel temperature field map Tl∈R obtained in the first stage. 256*256*3 The original field image is input into encoder E, and after three downsampling layers and three consecutive convolution operations, 256 images with a resolution of 32*32 are obtained. In order to reduce the model training parameters and improve the model efficiency, before reducing the two-dimensional image to a one-dimensional vector, we first use a 1*1 convolution kernel to convolve the obtained 256 images into a single two-dimensional image with a size of 32*32.

[0084] In this example embodiment, step S125: The two-dimensional image is then input into a 4-layer fully connected neural network to transform it into a 1*1024 one-dimensional vector.

[0085] In this example embodiment, step S126: Finally, the high-precision reconstruction of one-dimensional data into two-dimensional data is completed through the decoder's reverse operation. The 1*1024 one-dimensional vector is first transformed into a 32*32 two-dimensional single-channel image. Then, a 1*1 convolution kernel convolves the transformed single-channel image into 256 two-dimensional images of the same size (32*32). This is to ensure that the decoder recovers a three-channel color image. Finally, the 32*32*256 data is upsampled three times to recover a 256*256*3 three-channel temperature field map T. h .

[0086] In this example embodiment, step S127: Output a high-precision temperature field map T h The model uses the mean square error function as the loss function to calculate the predicted temperature field map T from the model output. h The difference between the temperature field plot T obtained from the FDS simulation and the actual temperature field plot T is represented by the following loss function:

[0087]

[0088] In this example embodiment, the method further includes:

[0089] The fitted data generated by the fire evolution simulation of the preset fire scenario is divided into training set, validation set and test set according to a preset ratio;

[0090] Based on the Adam optimizer, the preset first deep neural network model and the preset second deep neural network model are trained using the stochastic gradient descent method.

[0091] When the loss of the test set no longer decreases for a consecutive preset number of rounds, the training of the preset first deep neural network model and the preset second deep neural network model is completed.

[0092] In step S130, a preset first deep neural network model and a preset second deep neural network model corresponding to the second temperature field map of a preset fire scenario can be configured on a preset terminal. When a fire occurs, on-site information is received, and a predicted temperature field map is generated based on the preset first deep neural network model and the preset second deep neural network model according to the on-site information, thereby completing the prediction of the fire evolution.

[0093] In this example embodiment, the method further includes:

[0094] At the time of the fire, the on-site information was temperature data collected by a throwable sensor.

[0095] In this example embodiment, step S131, model training, involves dividing the dataset into a training set, a validation set, and a test set in a 7:2:1 ratio. The model is trained on a server using the Adam optimizer and stochastic gradient descent until the test set loss no longer decreases for five consecutive rounds. Training is then stopped, and the model with the minimum loss on the validation set is taken as the final model, saved to the file system, and deployed to the corresponding IoT central platform.

[0096] In this example embodiment, step S132 uses the following: when a fire occurs in real time, the relevant data collected on-site by various thrown devices (mainly sensors) is preprocessed and input into the model for inference and simulation in the background. After predicting the temperature field map, the results are presented in the form of a webpage to assist firefighters in making decisions.

[0097] In this example embodiment, the present invention only requires input of the initial fire source characteristics, and can quickly and directly output a temperature field map through a deep learning model. Firefighters do not need to understand complex computational fluid dynamics knowledge to perform tedious calculations, reducing their burden. It has strong real-time performance and can quickly provide a more reasonable rescue plan upon receiving a disaster report. The present invention directly outputs a temperature field map, which is equivalent to using deep learning to implicitly learn the trend prediction of temperature and smoke concentration, and obtain the flow field information of interest, providing a highly versatile method for predicting the evolution of indoor fires. The present invention improves the real-time performance and accuracy of prediction by designing a two-stage deep neural network to reconstruct the input vector with high precision and fitting the simulation data of the field model.

[0098] It should be noted that although the steps of the method in this disclosure are described in a specific order in the accompanying drawings, this does not require or imply that the steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additional or alternative steps may be omitted, multiple steps may be combined into one step, and / or a step may be broken down into multiple steps.

[0099] Furthermore, in this example embodiment, an indoor fire evolution prediction device based on a deep neural network is also provided. (Refer to...) Figure 4 As shown, the indoor fire evolution prediction device 400 based on a deep neural network may include: a data fitting module 410, a model training module 420, and a fire evolution prediction module 430. Wherein:

[0100] The data fitting module 410 is used to construct a fire evolution simulation based on the fire dynamics simulation tool FDS based on a preset fire scenario, and to complete the setting of physical model data and key parameters of the fire evolution model, and to establish simulation data and fitting data of the preset fire scenario based on the fire evolution simulation.

[0101] The model training module 420 is used to generate a first temperature field map based on a preset first deep neural network model, using the simulation data as input, and to generate a second temperature field map based on a preset second deep neural network model, using the first temperature field map as input.

[0102] The fire evolution prediction module 430 is used to configure a preset first deep neural network model and a preset second deep neural network model corresponding to the second temperature field map of a preset fire scenario on a preset terminal. When a fire occurs, it receives on-site information and generates a predicted temperature field map based on the preset first deep neural network model and the preset second deep neural network model according to the on-site information, thereby completing the prediction of the fire evolution.

[0103] The specific details of each of the above-mentioned indoor fire evolution prediction device modules based on deep neural networks have been described in detail in the corresponding indoor fire evolution prediction method based on deep neural networks, so they will not be repeated here.

[0104] It should be noted that although several modules or units of an indoor fire evolution prediction device 400 based on a deep neural network are mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0105] Furthermore, in an exemplary embodiment of this disclosure, an electronic device capable of implementing the above-described method is also provided.

[0106] Those skilled in the art will understand that various aspects of the present invention can be implemented as systems, methods, or program products. Therefore, various aspects of the present invention can be specifically implemented as entirely hardware embodiments, entirely software embodiments (including firmware, microcode, etc.), or embodiments combining hardware and software aspects, collectively referred to herein as “circuit,” “module,” or “system.”

[0107] The following reference Figure 5 To describe an electronic device 500 according to such an embodiment of the present invention. Figure 5 The electronic device 500 shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0108] like Figure 5As shown, the electronic device 500 is manifested in the form of a general-purpose computing device. The components of the electronic device 500 may include, but are not limited to: at least one processing unit 510, at least one storage unit 520, a bus 530 connecting different system components (including storage unit 520 and processing unit 510), and a display unit 540.

[0109] The storage unit stores program code that can be executed by the processing unit 510, causing the processing unit 510 to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of the present invention. For example, the processing unit 510 can perform actions such as... Figure 1 Steps S110 to S130 are shown in the diagram.

[0110] Storage unit 520 may include a readable medium in the form of a volatile storage unit, such as random access memory (RAM) 5201 and / or cache memory 5202, and may further include a read-only memory (ROM) 5203.

[0111] Storage unit 520 may also include a program / utility 5204 having a set (at least one) program module 5203, such program module 5205 including but not limited to: operating system, one or more application programs, other program modules and program data, each or some combination of these examples may include an implementation of a network environment.

[0112] Bus 550 can represent one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the various bus structures.

[0113] Electronic device 500 can also communicate with one or more external devices 570 (e.g., keyboard, pointing device, Bluetooth device, etc.), and with one or more devices that enable a user to interact with electronic device 500, and / or with any device that enables electronic device 500 to communicate with one or more other computing devices (e.g., router, modem, etc.). This communication can be performed via input / output (I / O) interface 550. Furthermore, electronic device 500 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 560. As shown, network adapter 560 communicates with other modules of electronic device 500 via bus 550. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with electronic device 500, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0114] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, terminal device, or network device, etc.) to execute the methods according to the embodiments of this disclosure.

[0115] In exemplary embodiments of this disclosure, a computer-readable storage medium is also provided, on which a program product capable of implementing the methods described above is stored. In some possible embodiments, various aspects of the invention may also be implemented as a program product comprising program code that, when the program product is run on a terminal device, causes the terminal device to perform the steps of the various exemplary embodiments of the invention described in the "Exemplary Methods" section above.

[0116] refer to Figure 6 As shown, a program product 600 for implementing the above-described method according to an embodiment of the present invention is described. It may employ a portable compact disc read-only memory (CD-ROM) and include program code, and may run on a terminal device, such as a personal computer. However, the program product of the present invention is not limited thereto. In this document, the readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, apparatus, or device.

[0117] The program product may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0118] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of sending, propagating, or transmitting programs for use by or in conjunction with an instruction execution system, apparatus, or device.

[0119] The program code contained on the readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0120] Program code for performing the operations of this invention can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0121] Furthermore, the above figures are merely illustrative of the processes included in the method according to exemplary embodiments of the present invention, and are not intended to be limiting. It is readily understood that the processes shown in the above figures do not indicate or limit the temporal order of these processes. Additionally, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0122] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the claims.

[0123] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A method for predicting the evolution of indoor fires based on deep neural networks, characterized in that, The method includes: Based on a preset fire scenario, a fire evolution simulation based on the fire dynamics simulation tool FDS is constructed, and the physical model data and key parameters of the fire evolution model are set. Based on the fire evolution simulation, simulation data and fitting data of the preset fire scenario are established. Using the simulated data as input, a first temperature field map is generated based on a preset first deep neural network model, and using the first temperature field map as input, a second temperature field map is generated based on a preset second deep neural network model. A preset first deep neural network model and a preset second deep neural network model corresponding to the second temperature field map of a preset fire scenario are configured on a preset terminal. When a fire occurs, the terminal receives on-site information and generates a predicted temperature field map based on the preset first deep neural network model and the preset second deep neural network model according to the on-site information, thereby completing the prediction of the fire evolution. The preset first deep neural network model also includes: Using the ReLU activation function and the squared error function as the loss function, the indoor geometry of the preset fire scenario, the heat release rate of the fire source, the location and time of the ventilation opening are used as input vectors. The vectors are connected to a three-layer fully connected neural network and a dropout layer to perform deconvolution and convolution operations, generating a first temperature field map output. The indoor geometry is the length, width and height of the room. The preset second deep neural network model further includes an encoder, hidden layers, and a decoder, wherein: The preset second deep neural network model uses the ReLU activation function and the squared error function as the loss function. The encoder is composed of a convolutional neural network. The convolutional layer of the convolutional neural network has a 3*3 kernel, padding of 1, and stride of 2. The encoder is used to compress the latitude of the first temperature field map to a first preset size. The hidden layer is a four-layer fully connected neural network, which is used to compress the first temperature field map of the first preset size from two-dimensional data to one-dimensional data. The decoder is composed of convolutional neural networks. The convolutional kernel of the convolutional layer of the convolutional neural network has a 3*3 kernel, padding of 1, and stride of 1. The decoder is used to insert pixels into the first temperature field map of the one-dimensional data to complete the image reconstruction with high accuracy and generate a second temperature field map.

2. The method as described in claim 1, characterized in that, The method further includes: The preset fire scenarios are fire scenarios in factories, commercial complexes, high-rise buildings, office buildings / teaching buildings, and apartments / residential buildings; Physical model data is collected based on the preset fire scenario. The physical model data includes room dimensions, door dimensions, window dimensions, location of combustibles, and fuel used for combustion. The key parameters of the fire evolution model are analyzed based on the Monte Carlo method, and the key parameters of the fire evolution model are generated, including the heat release rate per unit area, temperature, growth coefficient, and simulation time.

3. The method as described in claim 1, characterized in that, The method further includes: Based on the fire evolution simulation of a preset fire scenario, fitted data corresponding to the preset fire scenario is generated, and the fitted data is divided into a training set and a validation set according to a preset ratio.

4. The method as described in claim 1, characterized in that, The method further includes: The fitted data generated by the fire evolution simulation of the preset fire scenario is divided into training set, validation set and test set according to a preset ratio; Based on the Adam optimizer, the preset first deep neural network model and the preset second deep neural network model are trained using the stochastic gradient descent method. When the loss of the test set no longer decreases for a consecutive preset number of rounds, the training of the preset first deep neural network model and the preset second deep neural network model is completed.

5. The method as described in claim 1, characterized in that, The method further includes: At the time of the fire, the on-site information was temperature data collected by a throwable sensor.

6. An indoor fire evolution prediction device based on deep neural networks, characterized in that, The device includes: The data fitting module is used to construct a fire evolution simulation based on the fire dynamics simulation tool FDS based on a preset fire scenario, and to complete the setting of physical model data and key parameters of the fire evolution model. Based on the fire evolution simulation, it establishes simulation data and fitting data for the preset fire scenario. The model training module is used to generate a first temperature field map based on a preset first deep neural network model, using the simulated data as input, and to generate a second temperature field map based on a preset second deep neural network model, using the first temperature field map as input. The fire evolution prediction module is used to configure a preset first deep neural network model and a preset second deep neural network model corresponding to the second temperature field map of a preset fire scenario on a preset terminal. When a fire occurs, it receives on-site information and generates a predicted temperature field map based on the preset first deep neural network model and the preset second deep neural network model according to the on-site information, thereby completing the prediction of fire evolution. The preset first deep neural network model also includes: Using the ReLU activation function and the squared error function as the loss function, the indoor geometry, heat release rate of the fire source, ventilation location and time of the preset fire scenario are used as input vectors. A three-layer fully connected neural network and a dropout layer are connected to perform deconvolution and convolution operations to generate a first temperature field map output; the indoor geometry is the length, width and height of the room. The preset second deep neural network model further includes an encoder, hidden layers, and a decoder, wherein: The preset second deep neural network model uses the ReLU activation function and the squared error function as the loss function. The encoder is composed of a convolutional neural network. The convolutional layer of the convolutional neural network has a 3*3 kernel, padding of 1, and stride of 2. The encoder is used to compress the latitude of the first temperature field map to a first preset size. The hidden layer is a four-layer fully connected neural network, which is used to compress the first temperature field map of the first preset size from two-dimensional data to one-dimensional data. The decoder is composed of convolutional neural networks. The convolutional kernel of the convolutional layer of the convolutional neural network has a 3*3 kernel, padding of 1, and stride of 1. The decoder is used to insert pixels into the first temperature field map of the one-dimensional data to complete the image reconstruction with high accuracy and generate a second temperature field map.

Images