Comprehensive energy station hydrogen blast consequence real-time prediction method and system

Abstract

The invention discloses a real-time prediction method and system for a hydrogen blast consequence of a comprehensive energy station, and belongs to the technical field of neural network model application, and the method comprises the steps: S1, constructing a 3D numerical simulation model of the comprehensive energy station; s2, constructing a multi-scene explosion data set of the hydrogenation device; the data set comprises a first data set for simulating diffusion under different environmental conditions after hydrogen leakage and determining combustible gas spatial distribution data; the second data set is used for simulating the explosion of the combustible gas to obtain maximum explosion overpressure data under different scenes; s3, preprocessing the data set, constructing a reference data set suitable for deep learning model training, and dividing the reference data set into a training set and a test set; s4, constructing a DNN-DeConv-Ex model, and carrying out training and testing by utilizing the training set and the testing set; and S5, carrying out lightweight conversion on the DNN-DeConv-Ex model, and carrying out efficient and rapid prediction by using a lightweight model.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of neural network model application, and particularly relates to a hydrogen explosion consequence real-time prediction method and system for a comprehensive energy station. BACKGROUND

[0002] With the deepening of low-carbon transformation of energy structure, a new type of comprehensive energy station integrating diversified energy supply services such as refueling, gas filling, hydrogen filling and charging and replacing has emerged. Although the comprehensive energy station has broad development prospects and can effectively support the promotion and utilization of clean energy, its complex operating environment also faces serious threats from a series of potential accidents such as fire, explosion and leakage of toxic and harmful substances. The safety challenges of the comprehensive energy station mainly lie in two aspects: (1) the facilities and functions in the station are highly integrated, and multiple energy media (such as gasoline, diesel, natural gas, hydrogen and electric energy) coexist and are frequently converted, so that the risk points are dense and mutually influenced, and the risk control difficulty is significantly increased; (2) the risk evolution mechanism in the station is complex, a single initial event may trigger a chain reaction, and the accident consequences are often extremely serious, causing great damage to personnel, property and the surrounding environment.

[0003] For quantitative risk analysis of the comprehensive energy station, especially for evaluating the most dangerous explosion accident scene, the traditional method highly depends on accurate modeling of accident consequences, and its main methods include explosion experiment and computational fluid dynamics (CFD) simulation. However, due to the high risk, high cost and strict equipment site restrictions of full-size explosion experiment, it is usually difficult to safely and effectively obtain complete explosion overpressure spatial distribution data under real scenarios. Although the computational fluid dynamics (CFD) simulation technology can obtain high-fidelity flow field and explosion consequence calculation results based on physical equations, it relies on a complex numerical iterative solution process, resulting in extremely high calculation cost and long time consumption, and therefore is completely unsuitable for real-time prediction operations such as accident emergency disposal or risk dynamic evaluation, and cannot real-time derive and predict the explosion accident consequences under the complex coupling of various influencing factors (such as leakage position, leakage rate, obstacle layout, environmental wind speed and direction, etc.).

[0004] In view of the high risk and great environmental damage of explosion accidents in integrated energy stations, and the inherent defects of traditional experiments and simulation methods, which cannot provide real-time risk prediction support for such facilities. Through preliminary research and analysis, it is found that different functional areas of integrated energy stations (such as hydrogen storage area, hydrogen filling area, natural gas compression area, etc.) all have the risk of gas leakage and may cause explosion. Among them, due to the characteristics of hydrogen gas such as easy leakage, wide explosion limit range, low ignition energy and fast burning speed, the potential consequences of explosion after regional leakage are particularly serious. Therefore, rapid and accurate prediction of the consequences of explosion after hydrogen leakage in specific areas has important practical significance for guiding emergency evacuation, delineating safety warning range, optimizing fire rescue plan and other emergency disposal decisions.

[0005] To break through the limitations of traditional methods and achieve rapid prediction of explosion overpressure after hydrogen leakage, some researchers have proposed a new idea of using deep learning algorithms to build efficient proxy models. For example, Chinese patent CN116306377B proposes a deep learning proxy model based on long short-term memory network (LSTM), which is specifically used to predict the concentration and spatial distribution of hydrogen leakage in a confined space, and thus realizes high-precision and rapid prediction of hydrogen concentration field. However, it must be clearly pointed out that hydrogen leakage itself is not the final consequence of the accident, and pure hydrogen leakage (without reaching the lower limit of explosion or encountering ignition source) is difficult to cause direct harm to building structures and personnel. The accumulation of leaked hydrogen in the air, encountering a fire source or causing a violent explosion, is the key link that may lead to catastrophic losses in hydrogen leakage events. Therefore, single hydrogen leakage concentration prediction cannot meet the quantitative risk assessment of integrated energy stations; the dynamic distribution range and intensity of explosion shock wave overpressure in the station and surrounding environment after hydrogen leakage and explosion have extremely important reference value and decision support role for fire personnel to accurately assess the explosion power, scientifically develop fire fighting and rescue strategies, effectively protect their own safety, and minimize accident losses. SUMMARY

[0006] In view of the defects of the prior art, the present application provides a real-time prediction method and system for hydrogen combustion and explosion consequences in an integrated energy station, which can predict the hydrogen combustion and explosion consequences in an integrated energy station in real time.

[0007] The specific technical solutions adopted by the present application are as follows: The first invention of the present patent is to provide a real-time prediction method for hydrogen combustion and explosion consequences in an integrated energy station, wherein the integrated energy station comprises at least one hydrogenation device, and the method comprises: S1, constructing a 3D numerical simulation model of the integrated energy station; S2, based on the 3D numerical simulation model, a hydrogenation device multi-scenario explosion data set is constructed; the data set includes: A first data set simulates the diffusion of hydrogen leakage under different environmental conditions to determine the combustible gas spatial distribution data; A second data set simulates combustible gas explosion to obtain maximum explosion overpressure data under different scenarios; S3, preprocessing the data set, constructing a benchmark data set suitable for deep learning model training, and dividing the benchmark data set into training set and test set; S4, constructing a DNN-DeConv-Ex model, and training and testing using the training set and test set; S5, lightening conversion of the DNN-DeConv-Ex model, and prediction using the lightened model.

[0008] Preferably, the multi-scenario parameters include one or more of the position of the hydrogenation device, the leakage direction and quantity, the leakage aperture size and time, the wind direction and wind speed.

[0009] Preferably, in S3, the benchmark data set is D=[X,Y], represented as: ; Wherein, X represents the model input, m represents the number of multi-scenario parameters; Y n1×n2 represents the two-dimensional data characteristics of the model output, n is the row number; the benchmark data set D is randomly divided into training set and test set on the basis of normalization.

[0010] Preferably, the DNN-DeConv-Ex model includes a fully connected neural network and a deconvolution neural network, and the fully connected neural network and the deconvolution neural network are connected through a reshape layer, which converts a one-dimensional feature vector into a multi-dimensional feature vector.

[0011] Preferably, S5 includes using TFLite dynamic range quantization to compress the floating-point weight of the model into integer form int8, and converting the.h5 model into a smaller.tflite format model.

[0012] Preferably, it further includes S6, using PythonTkinter tool to construct a graphical user interface for TFLite lightened inference model, and realizing real-time visual prediction of gas leakage explosion maximum overpressure distribution.

[0013] The second invention of the patent is to provide a real-time prediction system for hydrogen explosion consequences of a comprehensive energy station, the comprehensive energy station including at least one hydrogenation device, and the system including: A 3D model module for constructing a 3D numerical simulation model of the comprehensive energy station; a data set module, based on the 3D numerical simulation model, constructs a hydrogenation device multi-scenario explosion data set; the data set includes: a first data set simulating diffusion under different environmental conditions after hydrogen leakage to determine combustible gas spatial distribution data; a second data set simulating combustible gas explosion to obtain maximum explosion overpressure data under different scenarios; a preprocessing module, preprocessing the data set, constructing a benchmark data set suitable for deep learning model training, and dividing the benchmark data set into a training set and a test set; a training and testing module, constructing a DNN-DeConv-Ex model, and training and testing using the training set and the test set; a lightweight module, performing lightweight conversion on the DNN-DeConv-Ex model, and using the lightweight model for prediction.

[0014] Preferably, the multi-scenario parameters include one or more of the position of the hydrogenation device, the leakage direction and quantity, the leakage aperture size and time, the wind direction and wind speed.

[0015] Preferably, the DNN-DeConv-Ex model includes a fully connected neural network and a deconvolution neural network, and the fully connected neural network and the deconvolution neural network are connected through a reshape layer, which converts a one-dimensional feature vector into a multi-dimensional feature vector.

[0016] Preferably, it further includes a graphical user interface to realize real-time visual prediction of the maximum overpressure distribution of gas leakage explosion.

[0017] The third invention of the patent is to provide a computer program product, including a computer program, which, when executed by a processor, performs the above-mentioned real-time prediction method of hydrogen explosion consequences in a comprehensive energy station.

[0018] The fourth invention of the patent is to provide an information data processing terminal for realizing the above-mentioned real-time prediction method of hydrogen explosion consequences in a comprehensive energy station.

[0019] The fifth invention of the patent is to provide a computer-readable storage medium, including instructions, which, when executed on a computer, cause the computer to perform the above-mentioned real-time prediction method of hydrogen explosion consequences in a comprehensive energy station.

[0020] The advantages and positive effects of the present application are: By adopting the above technical solution, the present application has the following technical effects: The present application uses deep learning technology to directly map a small number of key input parameters such as wind direction, wind speed, and leakage diameter to leakage explosion pressure distribution results through an end-to-end training process, thereby effectively breaking through the high cost and low efficiency limitations of traditional experiments and numerical simulation in the analysis of comprehensive energy stations, and realizing millisecond-level real-time prediction capability. Experimental results show that the prediction efficiency of the present application is improved by more than 20-30 times compared with traditional methods, and the prediction error is stably controlled within ± 15%, which not only greatly shortens the response time, but also ensures the high reliability of the results, fully meeting the strict requirements of real-time and accuracy in the disposal of comprehensive energy station accidents. The present application further builds a user-friendly GUI interface through model lightweight design, uses efficient algorithm optimization and modular architecture, so that the achievement has cross-platform rapid deployment capability, supporting mainstream operating systems such as Windows and Linux. This design ingeniously converts complex scientific calculations into intuitive and interactive application systems, and users can visualize the results in real time through drag-and-drop operations or parameter adjustment, significantly improving the engineering practicability and promotion value of the method. The present application can be seamlessly applied in the operating environment of comprehensive energy stations, significantly reducing the technical use threshold by simplifying the operation process and reducing hardware requirements, and facilitating rapid promotion to more energy station scenarios such as gas stations, hydrogen energy stations and other diversified facilities. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 A flowchart of a preferred embodiment of the present application; Figure 2 Another flowchart of a preferred embodiment of the present application; Figure 3 A DNN-DeConv-Ex neural network model structure diagram in a preferred embodiment of the present application; Figure 4 A model verification process diagram in a preferred embodiment of the present application; Figure 5 A login interface diagram of a system in a preferred embodiment of the present application; Figure 6 An operation display interface diagram of a system in a preferred embodiment of the present application; Figure 7 A 3D numerical simulation model diagram in a preferred embodiment of the present application. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0023] Please refer to Figure 1 ; The first embodiment is a real-time prediction method for hydrogen explosion consequences of a comprehensive energy station, wherein the comprehensive energy station comprises at least one hydrogenation device, for example, the comprehensive energy station is a new type of comprehensive energy station integrating oil filling, gas filling, hydrogen filling and power supply services, and the method comprises the following steps: S1, constructing a 3D numerical simulation model of the comprehensive energy station; specifically: First, the specific structure layout, equipment type, equipment position and equipment parameters of the comprehensive energy station are obtained, and then a 3D numerical simulation model of the comprehensive energy station is constructed by using a CFD numerical simulation software, as shown in Figure 7 ; S2, constructing a hydrogenation device multi-scenario explosion data set based on the 3D numerical simulation model; the data set comprises: A first data set simulates the diffusion of hydrogen leakage under different environmental conditions to determine the combustible gas spatial distribution data; A second data set simulates combustible gas explosion to obtain maximum explosion overpressure data under different scenarios; The multi-scenario parameters in the present application include one or more of the position of the hydrogenation device, the leakage direction and quantity, the leakage aperture size and time, the wind direction and the wind speed.

[0024] The first data set and the second data set in the present application are obtained by simulation software.

[0025] For example, hydrogenation device explosion simulation under multi-factor coupling is carried out. This simulation is divided into two stages: The first stage is the diffusion simulation of hydrogen leakage under different environmental conditions to determine the combustible gas spatial distribution.

[0026] The second stage is to carry out explosion simulation based on combustible gas to obtain the maximum explosion overpressure under the corresponding scenario.

[0027] 400 explosion scenarios are simulated. As a demonstration case, the simulation scenarios carried out in this scheme include 2 hydrogenation devices, 2 leakage directions (+X, -Z), 4 leakage apertures (5mm, 10mm, 15mm, 25mm), 5 wind directions (0° east wind, 90° south wind, 180° west wind, 270° north wind, 315° perennial wind direction), and 5 wind speeds (0.2m / s, 1.0m / s, 2m / s, 4m / s, 6m / s). Due to the high specificity of the gas distribution scenario determined by the position of the hydrogenation device and the direction of gas leakage, the feature migration ability between different scenarios is low, therefore, this scheme plans to construct 4 mixed deep learning models to predict the potential accident consequences when hydrogen leakage occurs in different directions of the 2 hydrogenation devices.

[0028] S3, preprocessing the data set, constructing a benchmark data set suitable for deep learning model training, and dividing the benchmark data set into a training set and a test set; The benchmark data set is D=[X, Y], for example: D=[X 5 ,Y 76×71 ], D=[X, Y] is represented as: ; Wherein, X represents the model input, m represents the number of parameters of multiple scenes; Y n1×n2 represents the two-dimensional data characteristics of the model output, n is the row number; the benchmark data set D is randomly divided into a training set and a test set on the basis of normalization.

[0029] In this scheme, m=5, i.e. 5 scene parameters related to hydrogen explosion, such as leakage position, leakage direction, leakage aperture, wind direction and wind speed. Y n1×n2 represents the model output parameter, Y 76×71 is adopted in this scheme, i.e. the horizontal explosion maximum overpressure distribution at the vertical height Z=1.5m, the pressure value appears in the form of a two-dimensional matrix of 76x71. The benchmark data set is randomly divided into a training set and a test set according to the proportion. The benchmark data set D constructed in this scheme is randomly divided into a training set (80%) and a test set (20%) on the basis of normalization.

[0030] S4, constructing a DNN-DeConv-Ex model, and training and testing using the training set and the test set; This step is to construct a hybrid deep learning model Fully Connected Deconvolutional Neural Network (DNN-DeConv-Ex) representing the nonlinear relationship between scene parameters and explosion maximum overpressure distribution. In this scheme, the DNN-DeConv-Ex model is constructed using Python programming language based on the deep learning framework Tensorflow. The input layer of the model is the leakage scene parameter, and the output layer is the maximum overpressure distribution caused by hydrogen explosion. The DNN-DeConv-Ex model mainly includes a fully connected neural network (DNN) and a deconvolutional neural network (DeConv). The model results are shown in the attached Figure 3 .

[0031] Table 1 DNN-DeConv-Ex neural network configuration

[0032] wherein different DeConv models are selected according to the leakage location and direction. The model input is a 1-dimensional array, which contains 3 feature vectors (wind direction, wind speed, and leakage diameter). The input vector is sequentially mapped through two fully connected neural networks (DNN) for feature mapping, thereby expanding the environmental vector to a fully connected neural network (DNN) containing 76x71x128 node information, and the activation function is ReLU.

[0033] ; Subsequently, the DNN is connected with the DeConv through a Reshape layer. The role of the Reshape layer is to convert the one-dimensional feature vector 76x71x128 into a three-dimensional feature vector (76, 71, 128). Wherein 76x71 corresponds to the spatial size of the output pressure matrix, and 128 represents the number of channels, and each spatial point contains rich feature information. The deconvolution neural network (DeConv) is reduced from 128 to 1 through 4 times of transpose convolution operation, so that the 76x71 corresponding feature vector is reconstructed into the maximum overpressure distribution of the comprehensive energy station hydrogenation facility explosion; ; wherein, w [ w 1, w 2] represent neural network weight values, which will be iteratively optimized in the model training process, and finally establish a nonlinear relationship between the environmental parameters and the maximum overpressure distribution. The model training process is controlled by the Loss function, and the correlation coefficient R 2 The training results are evaluated, and the formula is as follows: ; ; wherein is the true pressure value, is the model prediction value. The model prediction results and error analysis can be referred to the attached Figure 4 , in Figure 4 , the leakage location is No. 1 hydrogenation device, the leakage direction is +X, the leakage diameter is 20 mm, the wind direction is 90 degrees, and the wind speed is 0.2 m / s. The leftmost graph is the reference data, the middle graph is the model prediction data, and the rightmost graph is the absolute error.

[0034] S5, performing lightweight conversion on the DNN-DeConv-Ex model, and performing prediction using the lightweight model; This step is based on the above-mentioned hydrogenation device explosion consequence real-time prediction model to carry out lightweight conversion, thereby optimizing the actual deployment strategy of the model. The DNN-DeConv-Ex established in this scheme is built under the Keras framework of Tensorflow, so the TFLite dynamic range quantization method is adopted, and the.h5 model is converted into a smaller.tflite format model by compressing the floating point weight of the model into an integer form int8. Therefore, the model constructed only needs to combine a small amount of floating point operation during inference, thereby balancing the inference speed and accuracy on the basis of reducing the model size.

[0035] In the second embodiment, please refer to Figure 2 A hydrogen explosion consequence real-time prediction method for a comprehensive energy station, based on the first embodiment, further comprises: S6, a graphical user interface is constructed for a TFLite lightweight inference model by using a Python Tkinter tool, and real-time visualized prediction of maximum overpressure distribution of gas leakage explosion is realized.

[0036] In order to improve the efficiency of the hydrogenation device explosion consequence real-time prediction method at the application end, the graphical user interface (GUI) is constructed by using the Python Tkinter tool and the TFLite lightweight inference model in this scheme, and the visualized prediction of the maximum overpressure distribution of gas leakage explosion is realized. The user can complete the model prediction and generate the maximum explosion overpressure distribution map when hydrogen leaks by selecting the equipment and the leakage direction and inputting the scene parameters (wind direction, wind speed, leakage direction, diameter, etc.).

[0037] This scheme provides a reference for the layout of the GUI, including a parameter input area and a prediction image output area. The parameter input area includes: device selection, including device 1 and device 2; leakage direction selection, including +X and +Z; the prediction loading deep learning model type is determined through the above two types of selection. Environmental parameter input, including 0~30mm leakage diameter, 0~360° wind direction, and 0~6m / s wind speed. The “predict and draw” button is used to call the maximum overpressure prediction model constructed in this scheme; the state prompt label is used to display the model loading and prediction state. The prediction image output area is used to convert the two-dimensional array predicted by the model into a pressure distribution map through the Matplotlib module and update it in real time. The specific layout is referred to in the attached Figure 5 and Figure 6 .

[0038] Finally, the GUI script and the model file are packaged into an executable program.exe by PyInstaller in this scheme. This tool kit can realize the customization function of the prediction function by configuring different types of deep learning models, which is convenient for rapid deployment in actual scenes such as comprehensive energy stations and rapid evaluation of safety risks.

[0039] In summary, the present application carries out hydrogenation device explosion simulation under multi-factor coupling, and establishes a hydrogenation device multi-scenario explosion data set; The present application aims at the hydrogenation device of the integrated energy station, and systematically constructs the hydrogen explosion data set under the multi-factor coupling conditions including the leakage position, the leakage direction, the leakage aperture, the wind direction and the wind speed. The data set systematically shows the quantitative characterization of the explosion consequences under the complex layout in multiple scenarios, provides a scientific reference basis for the safety risk assessment of the integrated energy station, and provides benchmark data for the subsequent real-time prediction method of the explosion consequences of the hydrogenation device.

[0040] The present application establishes a real-time evaluation method of explosion consequences under multi-factor coupling for the hydrogenation facility of the integrated energy station.

[0041] The present application realizes the nonlinear relationship mapping of scene parameters to explosion pressure distribution and realizes the second-level prediction by using the deep learning method. Therefore, the method breaks through the time-consuming limitation of the traditional CFD numerical simulation, can significantly improve the calculation efficiency while ensuring the prediction accuracy, and thus provides technical support for the quantitative risk analysis, real-time emergency response and other tasks of the hydrogen explosion of the integrated energy station.

[0042] The present application realizes the systematic deployment and engineering application of the prediction model.

[0043] The present application proposes a complete and targeted systematic deployment scheme for the completed deep learning model. The model is lightened through dynamic range quantization, and the TFLite format is used to support the operation of low-power terminals. The model is classified and embedded in the system background, and can be updated to meet the new needs of the risk consequence prediction of the integrated energy station; at the same time, combined with the interactive graphical user interface (GUI), the integrated application closed loop of parameter input, model calling, result prediction and real-time visualization display is realized.

[0044] The third embodiment is an integrated energy station hydrogen explosion consequence real-time prediction system, which is used to execute the method of the first embodiment, and includes: A 3D model module constructs a 3D numerical simulation model of the integrated energy station; A data set module constructs a hydrogenation device multi-scenario explosion data set based on the 3D numerical simulation model; the data set includes: A first data set simulates the diffusion of hydrogen leakage under different environmental conditions to determine the combustible gas spatial distribution data; A second data set simulates the explosion of combustible gas to obtain the maximum explosion overpressure data under different scenarios; A preprocessing module pre-processes the data set, constructs a benchmark data set suitable for deep learning model training, and divides the benchmark data set into a training set and a test set; The training and testing module constructs the DNN-DeConv-Ex model and trains and tests the model by using the training set and the test set. The lightweight module performs lightweight conversion on the DNN-DeConv-Ex model and performs prediction by using the lightweight model.

[0045] On the basis of the third embodiment, a graphical user interface is further included to realize real-time visual prediction of the maximum overpressure distribution of gas leakage explosion.

[0046] The fourth embodiment, on the basis of the third embodiment, further includes a graphical user interface to realize visual prediction of the maximum overpressure distribution of gas leakage explosion.

[0047] The fifth embodiment is a computer readable storage medium storing a computer program, which, when executed by a processor, implements the real-time prediction method of hydrogen combustion explosion consequences of the integrated energy station.

[0048] The sixth embodiment is a computer program product including a computer program, which, when executed by a processor, implements the real-time prediction method of hydrogen combustion explosion consequences of the integrated energy station.

[0049] In the above embodiments, all or part of them can be realized by software, hardware, firmware or any combination thereof. When all or part of them are realized in the form of a computer program product, the computer program product includes one or more computer instructions. When the computer program instructions are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable systems. The computer instructions can be stored in a computer readable storage medium or transferred from one computer readable storage medium to another, for example, the computer instructions can be transferred from one website, computer, server or data center to another website, computer, server or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL) or wireless (such as infrared, wireless, microwave, etc.)) mode. The computer readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. that includes one or more available media sets. The available media can be magnetic media (such as floppy disk, hard disk, magnetic tape), optical media (such as DVD), or semiconductor media (such as solid state disk (SSD)) and the like.

[0050] The above merely describes the preferred embodiments of the present application, and it should be pointed out that, for those skilled in the art, several improvements and refinements can be made without departing from the principles of the present application, and these improvements and refinements should also be considered as falling within the protection scope of the present application.

Claims

1. A method for real-time prediction of the consequences of hydrogen combustion and explosion in an integrated energy station, wherein the integrated energy station includes at least one hydrogen refueling device, characterized in that, The method includes: S1. Construct a 3D numerical simulation model of the integrated energy station; S2. Based on the 3D numerical simulation model, construct a multi-scenario explosion dataset for hydrogen refueling devices; the dataset includes: The first dataset simulates the diffusion of hydrogen under different environmental conditions after a leak, determining the spatial distribution data of combustible gases. The second dataset simulates the explosion of combustible gases and obtains the maximum explosion overpressure data under different scenarios. S3. Preprocess the dataset to construct a benchmark dataset suitable for training deep learning models, and divide the benchmark dataset into a training set and a test set; S4. Construct a DNN-DeConv-Ex model and train and test it using the training and test sets. S5. Perform a lightweight transformation on the DNN-DeConv-Ex model and use the lightweight model for prediction.

2. The method for real-time prediction of hydrogen combustion and explosion consequences in an integrated energy station according to claim 1, characterized in that, The parameters for multiple scenarios include: the location of the hydrogenation unit, the direction and number of leaks, the size and timing of leak holes, and one or more of the following: wind direction and wind speed.

3. The method for real-time prediction of hydrogen combustion and explosion consequences in an integrated energy station according to claim 1, characterized in that, In S3, the baseline dataset is D=[X,Y], represented as: ； Where X represents the model input, m represents the number of parameters across multiple scenarios; Y n1×n2 The two-dimensional data features represent the output of the model, where n is the row and column number; the benchmark dataset D is randomly divided into training and test sets after normalization.

4. The method for real-time prediction of hydrogen combustion and explosion consequences in a comprehensive energy station according to claim 1, characterized in that, The DNN-DeConv-Ex model includes a fully connected neural network and a deconvolutional neural network, which are connected by a reshape layer that converts a one-dimensional feature vector into a multi-dimensional feature vector.

5. The method for real-time prediction of hydrogen combustion and explosion consequences in an integrated energy station according to claim 1, characterized in that, S5 includes the use of TFLite dynamic range quantization to compress the floating-point weights of the model into integer form int8, converting the .h5 model into a smaller .tflite format model.

6. The method for real-time prediction of hydrogen combustion and explosion consequences in an integrated energy station according to claim 5, characterized in that, It also includes S6, and uses the Python Tkinter tool to build a graphical user interface for the TFLite lightweight inference model, enabling real-time visualization and prediction of the maximum overpressure distribution in gas leak explosions.

7. A real-time prediction system for the consequences of hydrogen combustion and explosion in an integrated energy station, wherein the integrated energy station includes at least one hydrogen refueling device, characterized in that, The system includes: The 3D model module is used to construct a 3D numerical simulation model of the integrated energy station. The dataset module, based on the 3D numerical simulation model, constructs a multi-scenario explosion dataset for hydrogen refueling units; the dataset includes: The first dataset simulates the diffusion of hydrogen under different environmental conditions after a leak, determining the spatial distribution data of combustible gases. The second dataset simulates the explosion of combustible gases and obtains the maximum explosion overpressure data under different scenarios. The preprocessing module preprocesses the dataset, constructs a benchmark dataset suitable for training deep learning models, and divides the benchmark dataset into training and testing sets. The training and testing module constructs a DNN-DeConv-Ex model and uses the training and testing sets for training and testing. The lightweight module performs a lightweight transformation on the DNN-DeConv-Ex model and uses the lightweight model for prediction.

8. The real-time prediction system for hydrogen combustion and explosion consequences in an integrated energy station according to claim 7, characterized in that, The parameters for multiple scenarios include: the location of the hydrogenation unit, the direction and number of leaks, the size and timing of leak holes, and one or more of the following: wind direction and wind speed.

9. The real-time prediction system for hydrogen combustion and explosion consequences in an integrated energy station according to claim 7, characterized in that, The DNN-DeConv-Ex model includes a fully connected neural network and a deconvolutional neural network, which are connected by a reshape layer that converts a one-dimensional feature vector into a multi-dimensional feature vector.

10. The real-time prediction system for hydrogen combustion and explosion consequences in an integrated energy station according to claim 7, characterized in that, It also includes a graphical user interface to enable real-time visualization and prediction of the maximum overpressure distribution in gas leak explosions.

Images