Method for predicting consequences of gas explosion in confined space based on dynamic parameters of gas leakage process

Abstract

The application discloses a kind of limited space combustion consequence prediction methods based on gas leakage process dynamic parameter, collect every working condition from gas leakage starts every Δt time period concentration, velocity segmented time series data;Form three-dimensional primary ignition source array, according to the result of numerical simulation of combustion and explosion to simplify ignition source array;Get the maximum temperature field, static pressure field and dynamic pressure field data set on typical cross section in combustion process;Concentration, velocity segmented time series data set is mapped to maximum temperature field, static pressure field and dynamic pressure field data set, obtain combustible gas leakage explosion combustion consequence prediction data set, preliminary establish combustion consequence prediction model and train;Collect gas leakage concentration, velocity time series data in limited space input combustion consequence prediction model, obtain the maximum value field of combustion process temperature, static pressure and dynamic pressure.The application realizes real-time or super real-time combustible gas leakage in limited space after combustion process in the maximum overpressure, dynamic pressure and temperature field automatic prediction, and prediction result is accurate.

Description

Technical Field

[0001] This invention relates to a method for predicting the consequences of combustion and explosion in confined spaces based on dynamic parameters of a gas leakage process. Background Technology

[0002] In confined spaces, especially in highly confined areas such as large buildings and tunnels, there is a risk of gas leaks when natural gas storage, transportation, or application equipment is present. These accidents are often accompanied by a high degree of uncertainty regarding leak source parameters and environmental conditions, making it difficult to accurately predict the consequences of potential explosions caused by gas leaks. Particularly in responding to gas leak accidents, the ability to reasonably predict gas explosion loads is crucial not only for ensuring the safe evacuation and rescue of personnel in confined spaces but also for providing critical information support for effectively handling gas safety incidents.

[0003] Currently, there are three methods for calculating gas explosion pressure in predicting the consequences of gas leaks and explosions in confined spaces: empirical formula methods, numerical simulation methods, and calculation methods based on intelligent algorithms. However, these methods are all based on predicting the consequences of explosions in homogeneous gas mixtures or empirical methods, and lack prediction capabilities that are more adaptable to the actual gas leak process and scenario conditions that may lead to gas explosion consequences. Therefore, the prediction accuracy is not high. Summary of the Invention

[0004] The purpose of this invention is to provide a method for predicting the consequences of confined space combustion and explosion based on dynamic parameters of the gas leakage process with high prediction accuracy.

[0005] The objective of this invention is achieved through the following technical measures: a method for predicting the consequences of confined space combustion and explosion based on dynamic parameters of a gas leakage process, characterized by comprising the following steps:

[0006] S1. Conduct numerical simulation design of combustible gas leakage and diffusion process in confined space. Deploy concentration and velocity sensors in confined space to conduct numerical simulation of gas leakage and diffusion. Collect segmented time-series data of concentration and velocity from each sensor at Δt intervals from the start of gas leakage in each working condition. Normalize the segmented time-series data of concentration and velocity to form a segmented time-series dataset of concentration and velocity.

[0007] S2. Ignition sources are set up along the length direction with intervals Δx, the width direction with intervals Δy, and the height direction with intervals Δz in the confined space to form a three-dimensional primary ignition source array. Different ignition source positions are set on the ignition source array. Using the combustible gas cloud obtained after the gas leakage time ΔT under the typical working condition in step S1, combustion and explosion numerical simulations are carried out under the combustible gas cloud conditions. The temperature, static pressure, and dynamic pressure curves generated by the numerical simulation results are compared, and the ignition source array is simplified based on the results.

[0008] S3. Using the numerical simulation of gas leakage diffusion under different working conditions in the confined space in step S1, the combustible cloud formed under different gas leakage durations is simulated. Different ignition source positions in the simplified ignition source array are set sequentially to carry out gas combustion and explosion numerical simulation under different leakage time conditions. The two-dimensional temperature field, static pressure field and dynamic pressure field time series data of typical cross sections corresponding to different ignition source positions during the gas combustion and explosion process are obtained. For each pixel in the time series data, the maximum value in the time series data of the entire combustion and explosion process is extracted to obtain the maximum temperature field, static pressure field and dynamic pressure field dataset of typical cross sections during the combustion and explosion process.

[0009] S4. Map the concentration and velocity segmented time series datasets to the maximum temperature field, static pressure field, and dynamic pressure field datasets after the a×Δt time period to obtain the combustible gas leakage explosion and combustion consequence prediction dataset. A preliminary combustion consequence prediction model is established. When a=0, real-time prediction of combustion consequences is achieved, and when a>0, ultra-real-time prediction of gas combustion consequences is achieved.

[0010] S5. Use the combustible gas leakage explosion consequence prediction dataset to train the explosion consequence prediction model, and use the loss values ​​of the maximum temperature field, static pressure field and dynamic pressure field in the explosion consequence prediction process as the loss function in the training process to carry out deep learning training and obtain the trained explosion consequence prediction model.

[0011] S6. Collect gas leakage concentration and velocity time-series data in the confined space and input them into the combustion and explosion consequence prediction model. The combustion and explosion consequence prediction model calculates the maximum temperature, static pressure and dynamic pressure field of the combustion and explosion process based on the concentration and velocity time-series data.

[0012] This invention combines existing deep learning methods and utilizes the concentration and velocity data monitored during the leakage of combustible gas in a confined space to achieve real-time or ultra-real-time automatic prediction of the maximum overpressure, dynamic pressure, and temperature field during the combustion and explosion process after a combustible gas leak in a confined space. The prediction results are accurate.

[0013] In simplifying the ignition source array process, if the difference between the gas temperature, static pressure, and dynamic pressure in the calculation results of the combustion and explosion conditions under the ignition source position conditions within a certain spatial range is within a set range, the ignition sources between this ignition source position spacing are ignored, and an ignition source is selected within this position area to replace the calculation results.

[0014] The scope defined in this invention is less than 20%.

[0015] In step S1 of the present invention, numerical simulation or experimental methods are used to design numerical simulation working conditions for the leakage and diffusion process of combustible gas in a confined space.

[0016] In step S1 of the present invention, the concentration and velocity segmented time-series data are enhanced by increasing noise, changing the relative position of the sensor and the leakage source, or by data mirroring.

[0017] In step S2 of this invention, a typical operating condition is that each factor of the leakage source condition and the confined space environment condition takes at least two levels.

[0018] The typical cross section described in this invention is the central cross section of a confined space.

[0019] In step S2 of this invention, the temperature, static pressure, and dynamic pressure curves are data from the central section along the width of the confined space.

[0020] Compared with the prior art, the present invention has the following significant effects:

[0021] (1) High degree of automation: This invention can predict the parameters of the combustion and explosion after a gas explosion by using time-series data obtained from distributed gas concentration sensors and gas velocity sensors arranged in a confined space.

[0022] (2) High prediction accuracy: This invention uses deep learning methods to predict the consequences of possible combustion and explosion accidents through the dynamic parameters of a highly nonlinear gas leakage process, and achieves high prediction accuracy.

[0023] (3) High adaptability: The present invention can automatically detect samples based on the collected sample information through a central integrated control system. There is no time limit. As long as the relevant parameters are set, detection can be carried out either day or night. Attached Figure Description

[0024] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0025] Figure 1 This is a flowchart of the combustible gas leakage data processing of the present invention;

[0026] Figure 2 This is a flowchart of the ignition source array setup of the present invention;

[0027] Figure 3 This is a flowchart of the explosion consequence data processing and prediction process of the present invention;

[0028] Figure 4 This is a flowchart of deep learning for predicting the consequences of a fire and explosion, as exemplified by this invention.

[0029] Figure 5 This is a schematic diagram of the deep learning prediction model in an example of the present invention;

[0030] Figure 6This is a schematic diagram illustrating the prediction results of the combustion and explosion consequences in an example of the present invention. Detailed Implementation

[0031] The present invention will now be described in detail with reference to the embodiments and accompanying drawings to help those skilled in the art better understand the inventive concept of the present invention. However, the scope of protection of the claims of the present invention is not limited to the following embodiments. For those skilled in the art, all other embodiments obtained without creative effort without departing from the inventive concept of the present invention are within the scope of protection of the present invention.

[0032] like Figures 1-6 The diagram illustrates a method for predicting the consequences of confined space combustion and explosion based on dynamic parameters of a gas leakage process, comprising the following steps:

[0033] S1, see also Figure 1 Numerical simulation of the flammable gas leakage and diffusion process within a confined space was conducted. Concentration and velocity sensors were deployed within the confined space to simulate the gas leakage and diffusion. Time-series data of concentration and velocity were collected from each sensor at intervals of Δt from the start of the gas leakage for each scenario. These time-series data were then normalized to form a concentration and velocity time-series dataset. The details are as follows:

[0034] Numerical simulation or experimental methods are used to design the numerical simulation working conditions of combustible gas leakage and diffusion process in confined space. The design must take into account the leakage source conditions such as leakage source flow rate and orientation, as well as the environmental conditions of the confined space such as ambient wind, obstacles, and structure (various factors of the working condition). The gas leakage and diffusion working conditions are denoted as N groups.

[0035] Numerical simulations of gas leakage and diffusion were conducted, with the simulation duration denoted as ΔT for each operating condition. A series of gas concentration and gas velocity sensor groups were installed in a confined space. The number of concentration sensors is denoted as M, and each concentration sensor location is sequentially numbered (from 1 to M); the number of velocity sensors is denoted as Q, and each velocity sensor location is sequentially numbered (from 1 to Q). The arrangement of concentration and velocity sensors was based on the principles of ease of sensor placement within the confined space and uniform capture of flow field information during the gas leakage process within the confined space. The concentration and velocity parameters of the sensor groups in the confined space were saved during the numerical simulation at every Δt time interval from the start of the gas leakage (0~Δt, Δt~2Δt, 2Δt~3Δt, …, (n-1)Δt~nΔt during the leakage diffusion numerical simulation, where n ranges from (1, ΔT / Δt)). The concentration and velocity segmented time-series data from each sensor under each operating condition are sequentially saved as tensor data as input data for the prediction model. The rank of the velocity tensor data is (N×ΔT / Δt, Δt, M), and the rank of the velocity tensor data is (N×ΔT / Δt, Δt, Q). The first dimension of the concentration and data tensor data represents that the tensor data has N×ΔT / Δt samples, the second dimension represents that each sample contains Δt time series, and the third dimension represents that there are M or Q sensors. Simultaneously, methods such as adding noise, changing the relative position of the sensor group and the leakage source, or data mirroring can be considered to enhance the data and improve the subsequent data prediction performance.

[0036] The resulting concentration and velocity tensor datasets are then globally normalized. Commonly used methods such as standard normalization or maximum-minimum normalization can be considered.

[0037] S2, Participate Figure 2 Within a confined space, ignition sources are arranged at intervals of Δx along the length, Δy along the width, and Δz along the height, forming a three-dimensional primary ignition source array. Different ignition source positions are set on the array. Using the combustible gas cloud obtained after a gas leakage time ΔT under typical conditions in step S1, numerical simulations of gas combustion and explosion under these combustible gas cloud conditions are conducted. The temperature, static pressure, and dynamic pressure curves (maximum pressure, dynamic pressure, and temperature curves) generated from the numerical simulation results are compared. Based on the results, the ignition source array is simplified. The details are as follows:

[0038] A three-dimensional primary ignition source array is set up in a confined space, wherein a series of ignition sources are set up at intervals of Δx in the length direction, Δy in the width direction, and Δz in the height direction of the confined space to form a three-dimensional primary ignition source array.

[0039] Using the formed three-dimensional primary ignition source array, different ignition source positions are set on the array. Using the combustible gas cloud obtained after a gas leakage duration ΔT under typical conditions in step S1 (at least two levels for each factor), numerical simulations of gas combustion and explosion under these combustible gas cloud conditions are conducted. Representative temperature, static pressure, and dynamic pressure curves generated from the numerical simulation results are compared. Based on the results, the ignition source array can be simplified. The simplification principle is: if the differences between the gas temperature, static pressure, and dynamic pressure in the combustion and explosion calculation results under ignition source position conditions within a certain spatial range are within an acceptable range (generally, an error within 20%), then the ignition sources between these positions are ignored, and one ignition source within this range is selected to replace the calculation results. The simplified ignition source array is named the simplified ignition source array, and the number of ignition sources in the simplified ignition source array is denoted as O. The ignition sources are then sorted sequentially.

[0040] S3. Using the numerical simulation of gas leakage diffusion under different working conditions in the confined space in step S1, the combustible gas cloud formed under different gas leakage durations is simulated. Different ignition source positions in the simplified ignition source array are set sequentially to conduct numerical simulations of gas combustion and explosion under different leakage time conditions. The two-dimensional temperature field, static pressure field, and dynamic pressure field time series data of typical cross sections (the center cross section of the confined space) corresponding to different ignition source positions during the gas combustion and explosion process are obtained. For each pixel in the time series data, the maximum value in the time series data of the entire combustion and explosion process is extracted to obtain the maximum temperature field, static pressure field, and dynamic pressure field dataset of typical cross sections during the combustion and explosion process; as detailed below:

[0041] Using the numerical simulation of gas leakage diffusion under different operating conditions in the confined space in step 1, combustible gas clouds formed under different gas leakage durations are simulated (i.e., combustible gas clouds formed at times Δt, 2Δt, 3Δt, ..., nΔt in the leakage diffusion numerical simulation, where the value of n ranges from (1, ΔT / Δt)). Different ignition source positions in the simplified ignition source array formed in step S2 are then sequentially set to conduct numerical simulations of gas combustion and explosion under different leakage durations. The time-series data of the two-dimensional temperature field, static pressure field, and dynamic pressure field on the corresponding typical cross-section during the gas combustion and explosion process are saved. The number of steps for the saved time-series data is Te. The number of pixels in the length direction of the two-dimensional temperature field, static pressure field, and dynamic pressure field is denoted as X, and the number of pixels in the width direction is denoted as Y. Since there are O ignition source positions corresponding to each leakage sample, there are a total of N×ΔT / Δt×O groups of combustion and explosion conditions. The time series data of temperature field, static pressure field and dynamic pressure field in the combustion and explosion consequences are all stored as tensors of rank (N×ΔT / Δt×O,Te,X,Y).

[0042] For each pixel in the time series data of the two-dimensional temperature field, static pressure field and dynamic pressure field on the typical cross section in the obtained combustion and explosion consequences, the maximum value in the time series data of the entire combustion and explosion process is extracted, and the maximum temperature field, static pressure field and dynamic pressure field dataset on the typical cross section in the combustion and explosion process is obtained. The rank of the dataset tensor is (N×ΔT / Δt×O, X, Y).

[0043] S4. Map the concentration and velocity segmented time-series datasets to the maximum temperature field, static pressure field, and dynamic pressure field datasets after the a×Δt time interval to obtain a dataset for predicting the consequences of combustible gas leakage explosions. A preliminary prediction model for the consequences of combustion and explosion is then established. Details are as follows:

[0044] The prediction process maps the input concentration and velocity datasets to a dataset of combustion and explosion consequences following a × Δt time interval. This mapping process maps the concentration and velocity time-series data for each (e × Δt, (e+1) × Δt) time interval within each leakage and diffusion condition to the maximum temperature, static pressure, and dynamic pressure fields generated by ignition and explosion at time ((e+1) × Δt + a × Δt). Real-time prediction of combustion and explosion consequences is achieved when a is 0, and ultra-real-time prediction of gas combustion and explosion consequences is achieved when a is greater than zero. The input data for the prediction model includes concentration data of rank (N × (ΔT / Δt-a), Δt, M) and velocity data of rank (N × (ΔT / Δt-a), Δt, Q). The output data of the prediction model consists of three sets of explosion overpressure, dynamic pressure, and temperature data of rank (N × (ΔT / Δt-a) × O, X, Y).

[0045] A hybrid deep learning model for predicting the maximum overpressure, temperature, and dynamic pressure field of combustible gas leakage explosion is established using deep learning methods such as long short-term memory, convolutional neural networks, or their variants, based on time-series data of gas leakage concentration and velocity as input data.

[0046] S5. Train the explosion consequence prediction model using the combustible gas leakage explosion consequence prediction dataset, and use the loss values ​​of the maximum temperature field, static pressure field, and dynamic pressure field during the explosion consequence prediction process as the loss function during training to conduct deep learning training, thereby obtaining the trained explosion consequence prediction model; specifically as follows:

[0047] The loss values ​​of the maximum temperature field, static pressure field, and dynamic pressure field during the prediction of combustion and explosion consequences are used as the loss function in the training process. Deep learning training is carried out, and the trained combustion and explosion consequences prediction model is saved.

[0048] See steps S3 to S5 above. Figure 3 .

[0049] S6. Collect gas leakage concentration and velocity time-series data in the confined space and input them into the combustion and explosion consequence prediction model. The combustion and explosion consequence prediction model calculates the maximum temperature, static pressure and dynamic pressure field of the combustion and explosion process based on the concentration and velocity time-series data.

[0050] In practical applications, sensors are arranged in a confined space using the same concentration and velocity sensor locations as in step S1, and each sensor is numbered using the same sorting method as in step S1. The sensor group is run, and the time-series data of the concentration and velocity sensor group in the confined space is recorded in real time. The data is then sliced ​​according to the method for cutting the concentration and velocity sensor time-series data in step S1.

[0051] A trained combustion and explosion consequence prediction model is used. The processed time-series data of gas leakage concentration and velocity in the confined space are input into the model. The model calculates the maximum temperature, static pressure, and dynamic pressure fields of the combustion and explosion process based on the concentration and velocity time-series data.

[0052] Example

[0053] See Figures 4-6 Three-dimensional numerical simulations of combustible gas leakage under 17 leakage rates and 9 environmental wind speeds in a narrow confined space were conducted using numerical simulation algorithms, totaling 153 working conditions. The underground narrow confined space was 750m long, 15m wide, and 7.5m high. The simulation time for each working condition was 600s. During the numerical simulation, the time-series data of the distributed combustible gas concentration sensors and airflow velocity sensors on the confined space dome during the leakage and diffusion process were saved. Gas concentration and velocity sensors were evenly set with the same sensor spacing of 20m, with 37 concentration and velocity sensors respectively. The time step of the sensor data was 0.5s, and the sensors were numbered sequentially.

[0054] For 153 sets of 600-second time-series data on the concentration and velocity of combustible gas leakage during the diffusion process, data segmentation and processing were performed. The 600-second combustible gas leakage data was divided into 30 segments with 20-second time intervals. Furthermore, the concentration and velocity time-series data of each 20-second segment were normalized as a whole.

[0055] Two typical leakage rates and gas clouds formed after 600 seconds of leakage diffusion under two different ambient wind conditions were set up in a confined space. A three-dimensional initial ignition source array was used to numerically simulate the ignition of the uniformly distributed combustible gas cloud. The spacing of the ignition source array along the length of the confined space was 50m, the spacing along the height of the confined space was 1m, and the spacing along the width of the confined space was 1m. After data processing, the combustion and explosion consequences under different ignition source location conditions were obtained. The relative difference in explosion overpressure between all ignition source locations along the length of the confined space was greater than 20%. The relative differences between ignition source locations along the height and width of the confined space were both less than 10%. Therefore, the differences between ignition source locations along the width and height of the confined space were ignored. To simplify the entire ignition source array, only a series of ignition source locations were set along the length of the confined space, with a spacing of 50m between the ignition source locations, for a total of 15 ignition source locations.

[0056] For the combustible gas distribution obtained at 20-second time intervals during the combustible gas leakage and diffusion process under each operating condition, ignition source positions in a simplified ignition source array were set to perform numerical simulations of the gas cloud for each time interval under each operating condition. The temperature field, static pressure field, and dynamic pressure field data during the explosion process were saved. The temperature field, static pressure field, and dynamic pressure field data for each combustion and explosion condition are time-series data of a representative slice during the explosion process. Here, the representative slice is selected as the cross-section at the center of the confined space width direction. The shape of the time-series data of the numerical simulation of ignition and combustion and explosion of the gas cloud corresponding to each leakage process through the simplified ignition source array is (15, 500, 50, 15), where the first axis 15 represents different ignition source positions in the simplified ignition source array, the second axis 500 is the number of time steps in the combustion and explosion process, the third axis 50 is the pixel point selected along the length direction of the representative slice, and the fourth axis 15 is the pixel point selected along the width direction of the representative slice.

[0057] The temperature field, overpressure field, and dynamic pressure field data under each set of combustion and explosion conditions are processed. For each set of time-series data of temperature field, overpressure field, and dynamic pressure field with shape (15, 500, 50, 15), the maximum value of temperature field, overpressure field, and dynamic pressure field at each pixel point on the central section during the combustion and explosion process is selected over 500 time steps to obtain the maximum value field of temperature, static pressure, and dynamic pressure field. The shape of the maximum value field data of temperature, static pressure, and dynamic pressure field for each combustion and explosion condition is (15, 50, 15).

[0058] Global data normalization was performed on the maximum values ​​of temperature, static pressure, and dynamic pressure fields obtained in all combustion and explosion conditions.

[0059] The concentration and velocity data during the leakage and diffusion process under different operating conditions are used as model input data and mapped to the maximum temperature, static pressure, and dynamic pressure field data of different ignition source locations at the end of the corresponding time period during the ignition and explosion process, with α set to 0. The input and output data are mirrored along the length of the confined space and combined with the original data to achieve data augmentation. Furthermore, noise is added to further enhance the data. This yields a dataset for predicting the consequences of flammable gas leakage explosions.

[0060] A model for predicting combustion and explosion consequences was established, using time-series data of concentration and velocity during the diffusion of flammable gas leaks as input data, and the maximum overpressure field, temperature field, and dynamic pressure field data of the flammable gas explosion process formed during ignition and explosion after the leak as output data. A deep learning-based model for predicting combustion and explosion consequences was developed using algorithms such as Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), and Recurrent Neural Networks (RNN). (See [link to relevant documentation]). Figure 5 .

[0061] The dataset for predicting the consequences of flammable gas leak explosions was divided into training, validation, and test sets, with proportions of 90%, 5%, and 5%, respectively. The training set data was then fed into a deep learning model for predicting the consequences of flammable gas leaks. During training, the average mean square error of the predicted static pressure, temperature, and dynamic pressure fields in the test set was used as the evaluation metric. Training was considered complete when the metric reached below 0.01, and the training model was saved. After training, the predicted and actual values ​​of the maximum temperature, static pressure, and dynamic pressure fields at the ignition source location (number 8) were compared. (See [reference needed]). Figure 6 .

[0062] In practical applications, sensors are arranged in a confined space using the same concentration and velocity sensor locations as in step S1, and each sensor is numbered using the same sorting method as in step S1. The sensor group is run, and the time-series data of the concentration and velocity sensor group in the confined space is recorded in real time. The data is then sliced ​​according to the method for cutting the concentration and velocity sensor time-series data in step S1.

[0063] A trained combustion and explosion consequence prediction model is used. The processed time-series data of gas leakage concentration and velocity in the confined space are input into the model. The model calculates the maximum temperature, static pressure, and dynamic pressure fields of the combustion and explosion process based on the concentration and velocity time-series data.

Claims

1. A method for predicting the consequences of confined space combustion and explosion based on dynamic parameters of a gas leakage process, characterized in that... Includes the following steps: S1. Conduct numerical simulation design of combustible gas leakage and diffusion process in confined space. Deploy concentration and velocity sensors in confined space to conduct numerical simulation of gas leakage and diffusion. Collect segmented time-series data of concentration and velocity from each sensor at Δt intervals from the start of gas leakage in each working condition. Normalize the segmented time-series data of concentration and velocity to form a segmented time-series dataset of concentration and velocity. S2. Ignition sources are set up along the length direction with intervals Δx, the width direction with intervals Δy, and the height direction with intervals Δz in the confined space to form a three-dimensional primary ignition source array. Different ignition source positions are set on the ignition source array. Using the combustible cloud obtained after the gas leakage time ΔT under the typical working condition in step S1, numerical simulations of gas combustion and explosion under the combustible cloud conditions are carried out respectively. The temperature, static pressure, and dynamic pressure curves generated by the numerical simulation results are compared, and the ignition source array is simplified based on the results. S3. Using the numerical simulation of gas leakage diffusion under different working conditions in the confined space in step S1, the combustible cloud formed under different gas leakage durations is simulated. Different ignition source positions in the simplified ignition source array are set sequentially to carry out numerical simulation of gas combustion and explosion under different leakage time conditions. The two-dimensional temperature field, static pressure field and dynamic pressure field time series data of typical cross sections corresponding to different ignition source positions during the gas combustion and explosion process are obtained. For each pixel in the time series data, the maximum value in the time series data of the entire combustion and explosion process is extracted to obtain the maximum temperature field, static pressure field and dynamic pressure field dataset of typical cross sections during the combustion and explosion process. S4. Map the concentration and velocity segmented time series datasets to the maximum temperature field, static pressure field, and dynamic pressure field datasets after the a×Δt time period to obtain the combustible gas leakage explosion and combustion consequence prediction dataset. A preliminary combustion consequence prediction model is established. When a=0, real-time prediction of combustion consequences is achieved, and when a>0, ultra-real-time prediction of gas combustion consequences is achieved. S5. Use the combustible gas leakage explosion consequence prediction dataset to train the explosion consequence prediction model, and use the loss values ​​of the maximum temperature field, static pressure field and dynamic pressure field in the explosion consequence prediction process as the loss function in the training process to carry out deep learning training and obtain the trained explosion consequence prediction model. S6. Collect gas leakage concentration and velocity time-series data in the confined space and input them into the combustion and explosion consequence prediction model. The combustion and explosion consequence prediction model calculates the maximum temperature, static pressure and dynamic pressure field of the combustion and explosion process based on the concentration and velocity time-series data.

2. The method for predicting the consequences of confined space combustion and explosion based on dynamic parameters of gas leakage process according to claim 1, characterized in that: In the process of simplifying the ignition source array, if the difference between the gas temperature, static pressure and dynamic pressure in the calculation results of the combustion and explosion conditions under the ignition source position conditions in a certain spatial range is within the set range, the ignition sources between this ignition source position spacing are ignored, and an ignition source is selected in this position area to replace the calculation results.

3. The method for predicting the consequences of confined space combustion and explosion based on dynamic parameters of gas leakage process according to claim 2, characterized in that: The set range is less than 20%.

4. The method for predicting the consequences of confined space combustion and explosion based on dynamic parameters of gas leakage process according to claim 3, characterized in that: In step S1, numerical simulation or experimental methods are used to design the numerical simulation working conditions of the flammable gas leakage and diffusion process in a confined space.

5. The method for predicting the consequences of confined space combustion and explosion based on dynamic parameters of gas leakage process according to claim 4, characterized in that: In step S1, the concentration and velocity segmented time-series data are enhanced by increasing noise, changing the relative position of the sensor and the leakage source, or by data mirroring.

6. The method for predicting the consequences of confined space combustion and explosion based on dynamic parameters of gas leakage process according to claim 5, characterized in that: In step S2, a typical operating condition is that each factor of the leakage source condition and the confined space environment condition takes at least two levels.

7. The method for predicting the consequences of confined space combustion and explosion based on dynamic parameters of gas leakage process according to claim 6, characterized in that: The typical cross-section is the central cross-section of the confined space.

8. The method for predicting the consequences of confined space combustion and explosion based on dynamic parameters of gas leakage process according to claim 7, characterized in that: In step S2, the temperature, static pressure, and dynamic pressure curves are data from the central section along the width of the confined space.

Images