Gas explosion source intensity and position inversion method based on deep learning

Abstract

The invention belongs to the technical field of gas explosion disaster inversion prevention, and discloses a gas explosion source intensity and position inversion method based on deep learning, and the method comprises the steps: obtaining to-be-inverted gas explosion sensing data; inputting the to-be-inverted gas explosion sensing data into a gas explosion source inversion model for prediction inversion to obtain explosion source position data and explosion intensity data; wherein the gas explosion source inversion model comprises a one-dimensional inversion model and a two-dimensional inversion model, and the one-dimensional inversion model and the two-dimensional inversion model are both constructed based on a neural network; and carrying out visual display on the explosion source position data and the corresponding explosion intensity data. According to the technical scheme, accurate inversion of the position and the strength of the explosion source can be rapidly achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of gas explosion disaster inversion prevention, and in particular relates to a gas explosion source intensity and position inversion method based on deep learning. Background Art

[0002] Inversion research on the intensity and location of gas explosion sources is of great significance for understanding the accident mechanism, assessing the impact and guiding rescue work.

[0004] Traditional methods of locating gas explosion sources mainly rely on sensor networks, which acquire data by deploying a large number of sensors in mines. The main advantage of this method is that it can monitor the gas concentration and other related parameters in the mine in real time, thereby preventing the occurrence of gas explosions to a certain extent. However, this method also has obvious shortcomings. First, the deployment and maintenance costs of sensor networks are high, especially in large-scale mines, where the deployment and maintenance of sensors require a lot of manpower and material resources. Second, data processing is complex. The environmental conditions in mines are complex and changeable, and the data collected by sensors often contain a lot of noise and interference. How to effectively process this data has become a major problem.

[0005] In recent years, research based on inversion theory has received increasing attention. Researchers have tried to analyze the characteristics of gas explosions (such as sound, vibration, etc.) and use mathematical models and computer algorithms to infer the location and intensity of the explosion source from known response data. The core idea of ​​this inversion method is to regard the gas explosion as an inverse problem and infer the potential cause by observing the phenomenon. Specifically, researchers collect experimental data or simulation results, combine them with appropriate mathematical models, and use inverse problem theory and algorithms to solve them in order to achieve accurate inversion of the gas leakage source. This method has certain advantages over traditional methods, especially in theory, it can achieve higher inversion accuracy and faster calculation speed, but traditional inversion methods rely on complex physical models and a large amount of data processing, with high computational costs and long inversion times. Summary of the invention

[0006] The purpose of the present invention is to provide a gas explosion source intensity and position inversion method based on deep learning to solve the problems existing in the above-mentioned prior art.

[0007] To achieve the above object, the present invention provides a method for inverting the intensity and position of a gas explosion source based on deep learning, comprising:

[0008] Acquire gas explosion sensor data to be inverted;

[0009] Inputting the gas explosion sensor data to be inverted into the gas explosion source inversion model for predictive inversion to obtain explosion source location data and explosion intensity data; wherein the gas explosion source inversion model includes a one-dimensional inversion model and a two-dimensional inversion model, and both the one-dimensional inversion model and the two-dimensional inversion model are constructed based on a neural network;

[0010] The explosion source location data and the corresponding explosion intensity data are visualized.

[0011] Optionally, the gas explosion sensor data to be inverted includes sensor measurement point locations and explosion pressure wave data.

[0012] Optionally, before the gas explosion sensor data to be inverted is input into the gas explosion source inversion model, the method also includes performing data cleaning, normalization and denoising on the gas explosion sensor data to be inverted to obtain preprocessed gas explosion sensor data to be inverted and using the preprocessed data as input of the gas explosion source inversion model.

[0013] Optionally, the training process of the gas explosion source inversion model specifically includes:

[0014] Acquire training data, wherein the training data includes gas explosion sensor training data and corresponding explosion source data, wherein the explosion source data includes explosion source location data and explosion intensity data;

[0015] An initial gas explosion source inversion model is constructed, the training data is input into the initial gas explosion source inversion model for predictive inversion, and training is performed with the goal of minimizing the loss between the initial training results after predictive inversion and the explosion source data corresponding to the gas explosion sensor training data to obtain the gas explosion source inversion model.

[0016] Optionally, the one-dimensional inversion model includes an input layer, a hidden layer and an output layer connected in sequence;

[0017] The input layer receives the gas explosion sensor data to be inverted as an input vector, obtains the time series characteristics of the input vector through the hidden layer, and the output layer inverts one-dimensional explosion source position data and one-dimensional explosion intensity data according to the time series characteristics.

[0018] Optionally, the two-dimensional inversion model includes a convolutional layer, a pooling layer and a fully connected layer connected in sequence;

[0019] The convolution layer extracts features of the gas explosion sensor data to be inverted to obtain a feature map; the pooling layer downsamples the feature map to obtain a downsampled feature map;

[0020] The outputs of the convolution layer and the pooling layer are feature mapped through the fully connected layer to output two-dimensional explosion source position data and two-dimensional explosion intensity data.

[0021] Optionally, the explosion source location data and the corresponding explosion intensity data are visualized, specifically including:

[0022] A user interaction interface is constructed based on the constructed gas explosion source inversion model and the corresponding model output results, and a visualization display of explosion source location data, explosion intensity data and inversion result analysis curves is performed based on the user interaction interface.

[0023] Optionally, the user interaction interface includes a storage sub-interface, a one-dimensional position inversion result sub-interface, a one-dimensional intensity inversion sub-interface, a two-dimensional position inversion result sub-interface, a two-dimensional intensity inversion sub-interface and an inversion result analysis curve sub-interface.

[0024] The technical effects of the present invention are:

[0025] The present invention can automatically extract features from the observation data of gas explosions through a neural network model, and deeply explore the interactive relationship between overpressure and position data and the intensity and source point of the explosion source, so as to quickly realize the accurate inversion of the explosion source position and intensity. At the same time, by writing a visual evaluation software, the inversion results can be viewed in real time for intuitive evaluation and verification. The present invention solves the problems that gas explosion accidents are difficult to predict and the explosion source intensity and position cannot be inverted. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0027] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0028] Figure 1 is a flow chart of the steps of inversion of the location and intensity of a gas explosion source in an embodiment of the present invention;

[0029] Figure 2 It is a basic principle diagram for realizing gas explosion source inversion in an embodiment of the present invention;

[0030] Figure 3 is a working condition diagram of a gas explosion simulation experiment in an embodiment of the present invention;

[0031] Figure 4 is a diagram of a recurrent neural network structure in an embodiment of the present invention;

[0032] Figure 5 is a one-dimensional intensity inversion model evaluation coefficient diagram in an embodiment of the present invention;

[0033] Figure 6 is a comparison diagram of the absolute error results of one-dimensional intensity and position inversion in an embodiment of the present invention;

[0034] Figure 7 This is a diagram showing an interface of an inversion effect visualization module in an embodiment of the present invention;

[0035] Figure 8 It is a flow chart for implementing the inversion of the intensity and position of the gas explosion source in an embodiment of the present invention. DETAILED DESCRIPTION

[0036] Various exemplary embodiments of the present invention will now be described in detail. This detailed description should not be considered as limiting the present invention, but should be understood as a more detailed description of certain aspects, features, and embodiments of the present invention.

[0037] It should be understood that the terms described in the present invention are only for describing special embodiments and are not intended to limit the present invention. In addition, for the numerical range in the present invention, it should be understood that each intermediate value between the upper and lower limits of the scope is also specifically disclosed. Each smaller range between the intermediate value in any stated value or stated range and any other stated value or intermediate value in the described range is also included in the present invention. The upper and lower limits of these smaller ranges can be independently included or excluded in the scope.

[0038] Unless otherwise indicated, all technical and scientific terms used herein have the same meanings as those of ordinary skill in the art generally understood by the present invention. Although the present invention describes only preferred methods, any method similar or equivalent to that described herein may also be used in the implementation or testing of the present invention. All documents mentioned in this specification are incorporated by reference to disclose and describe the methods related to the documents. In the event of a conflict with any incorporated document, the content of this specification shall prevail.

[0039] It will be apparent to those skilled in the art that various modifications and variations may be made to the specific embodiments of the present invention description without departing from the scope or spirit of the present invention. Other embodiments derived from the present invention description will be apparent to those skilled in the art. The present application description and examples are exemplary only.

[0040] The words “include,” “including,” “have,” “contain,” etc. used in this article are open-ended terms, meaning including but not limited to.

[0041] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0042] Embodiment 1

[0043] like Figure 1 - Figure 8 As shown, in this embodiment, a method for inversion of gas explosion source intensity and position based on deep learning is provided, including: obtaining gas explosion sensor data to be inverted; inputting the gas explosion sensor data to be inverted into a gas explosion source inversion model for predictive inversion to obtain explosion source position data and explosion intensity data; wherein the gas explosion source inversion model includes a one-dimensional inversion model and a two-dimensional inversion model, and both the one-dimensional inversion model and the two-dimensional inversion model are constructed based on neural networks; and visually displaying the explosion source position data and the corresponding explosion intensity data.

[0044] Research on the inversion of gas explosion source intensity and location is of great significance for preventing accidents, improving safety levels, improving gas prevention and control technology and emergency rescue. Through continuous technological innovation and multidisciplinary cooperation, it is expected that more complex inversion problems will be solved in the future, providing more reliable solutions for the monitoring and control of gas explosion risks. The continuous development of gas explosion source inversion technology will provide a solid technical guarantee for coal mine safety production and help achieve sustainable development of the coal mining industry.

[0045] The present embodiment relates to the technical field of inversion prevention of gas explosion disasters, and specifically to a gas explosion source intensity and position inversion system based on deep learning. It includes data acquisition, data preprocessing, feature extraction, model training, inversion calculation, adaptive optimization, and result output and evaluation. The data after the gas explosion is collected by using a variety of sensors arranged in the mine, and preprocessing operations such as denoising and normalization are performed. The key characteristic parameters such as explosion waveform, arrival time and energy are extracted from the preprocessed data for deep learning, and input into a one-dimensional recurrent neural network (RNN) model and a two-dimensional convolutional neural network (CNN) for model training. By adjusting the model parameters and structure, the present embodiment can adaptively perform real-time data mining and inversion of the intensity and position of the gas explosion source, and improve the inversion accuracy through adaptive optimization; by writing a visual evaluation software, the inversion results can be viewed in real time for intuitive evaluation and verification. The present embodiment solves the problems that gas explosion accidents are difficult to predict and the intensity and position of the explosion source cannot be inverted.

[0046] Neural networks, especially deep learning technology, with its powerful data processing and automatic feature extraction capabilities, make it possible to quickly and accurately invert the intensity and location of gas explosion sources. In the study of gas explosion source inversion, the neural network model has achieved accurate inversion of the explosion source location and intensity through training, providing important support for the prevention and control of gas explosion accidents. Compared with traditional methods, neural networks have shown obvious advantages in processing large amounts of data and accurately obtaining inversion models.

[0047] Specifically, through a large amount of data training, neural networks can learn the complex nonlinear relationships in the gas explosion process, thereby achieving higher inversion accuracy. By training the neural network model, features can be automatically extracted from the observed data of the gas explosion, thereby achieving accurate inversion of the explosion source location and intensity. These technologies not only improve the accuracy and efficiency of inversion, but also provide new ideas and methods for the prevention and control of gas explosion accidents.

[0048] This embodiment aims to solve the problems of the prior art and proposes a method for improving the accuracy of inversion of the intensity and position of a gas explosion source. The technical solution of this embodiment is as follows:

[0049] S1: Acquiring data source and preprocessing data. This embodiment obtains data from public data sets of the United States Bureau of Mines and Chongqing Coal Research Institute, including data sources including maximum overpressure information sets at different locations of different working condition measurement points, and performs preprocessing.

[0050] S2: One-dimensional recurrent neural network model training. First, extract each group of maximum overpressure data in the public data set. Secondly, introduce the recurrent neural network model to invert the location and intensity of the gas explosion source by training the maximum overpressure at different locations. Then, optimize the model by combining parameter adjustment, automatic update of learning rate and other strategies to invert the overpressure data.

[0051] S3: Two-dimensional convolutional neural network model training. First, extract each group of different location information and maximum overpressure data in the public data set. Use the convolutional neural network to mine the characteristics of multidimensional data to train the maximum overpressure inversion model of gas explosion source location and intensity at different locations. Then, use the model optimization strategy to adjust the model weight parameters to invert the overpressure data.

[0052] S4: Visualization of inversion effect is achieved by obtaining the real working condition data value and the trained model inversion data value to generate the intensity and position inversion comparison curve.

[0053] The acquisition of data source and data preprocessing in step S1 are mainly divided into the following two steps:

[0054] S11: Obtain data. Based on the original data provided by the U.S. Bureau of Mines and Chongqing Coal Research Institute’s public data sets, obtain the maximum overpressure data at different locations of measuring points under different operating conditions.

[0055] S12: Preprocess the original data, clean and remove duplicate data, missing data and invalid data from the data source, count the ratio of complete data to missing data, and extract valid data according to a certain ratio and divide it into training set, validation set and test set.

[0056] The recurrent neural network training in step S2 above can be specifically divided into the following three steps:

[0057] S21: The basic structure of RNN includes input layer, hidden layer and output layer. The output of the hidden layer will be fed back to the hidden layer of the next time step to form a recurrent connection. In this way, the network can invert the current output based on the previous input and hidden state.

[0058] At each time step, the RNN receives an input vector and a hidden state vector. The input vector contains the input data of the current time step, and the hidden state vector contains the hidden state of the previous time step. The hidden state of the current time step is obtained by weighted summing the input vector and the hidden state vector and processing them through the activation function.

[0059] The calculation of the hidden state can be expressed using the basic formula of a recurrent neural network:

[0060]

[0061] Among them, x t is the input vector of the current time step, h t-1 is the hidden state of the previous time step, W {ih and W {hh} is the weight matrix, b h is the bias vector and f is the activation function.

[0062] After the hidden state is calculated, it can be used as the hidden state of the next time step, and can also be processed by the output layer for inversion or other tasks.

[0063] The calculation of the final output layer can be expressed using the following formula

[0064]

[0065]

[0066] Here, t is the linear transformation of the output layer at time step t, is the inversion output at time step t, W {hois the weight matrix from the hidden layer to the output layer, b o It is the bias term of the output layer, and the Linear function is used to output the probability distribution in the inversion problem.

[0067] S22: Training a neural network involves using a large amount of data to adjust the network parameters to minimize the loss function. This process is usually implemented using the back-propagation algorithm and the gradient descent algorithm.

[0068] The model loads the 16 training and 5 test data sets into memory, then creates a recurrent neural network with one input layer, 20 hidden layers, and one output layer. The neural network has two hidden processing layers, each of which has 20 nodes. The number of input and output nodes is determined by the data, but the number of hidden layers and the number of nodes in each are free parameters that must be determined by trial and error. Training a neural network means that the values ​​of these weights are constantly updated during training, and biases are calculated using training data with known correct input and output values, which define the behavior of the neural network. After training, the model is then tested on the test data to calculate the model's accuracy. Test accuracy is a rough measure of how well the model is expected to perform on new data that it has never seen before.

[0069] S23: Effectively extract data features, save the trained models as "yiweiqiangdu_model.pth" and "yiweiweizhi_model.pth", and then package them into the program code for verification and use with other data sets or simulation experimental data.

[0070] The convolutional neural network training in step S3 above can be divided into the following three steps:

[0071] S31: The basic structure of CNN mainly includes convolution layer, pooling layer and fully connected layer. The convolution layer is one of the most important layers in CNN. It extracts the features of the image by convolving the input image with a set of convolution kernels. The pooling layer is responsible for downsampling the feature map output by the convolution layer. The fully connected layer is responsible for mapping the output of the convolution layer and the pooling layer to the final classification result. The typical CNN model is not just the three-layer structure mentioned above, but a multi-layer structure. For example, this software adopts a convolutional neural network model with a total of five layers, which is to first increase the dimension of the convolution layer data, reduce the dimension of the pooling layer, and then increase the dimension of the convolution layer data, reduce the dimension of the pooling layer, and finally output the inversion value of the fully connected layer.

[0072] S32: Same as S22.

[0073] S33: Effectively extract data features, save the trained models as "erweiqiangdu_model.pth" and "erweiweizhi_model.pth", and then package them into the program code for verification and use with other data sets or simulation experimental data.

[0074] The visualization of the inversion effect in the above step S4 is realized by the following steps:

[0075] S41: After the external data is inverted by the above four models, its inversion value can be temporarily stored in the local memory, and then the real gas explosion source intensity and position corresponding to the input external data are stored in the local memory at the same time. At this time, by comparing the two sets of corresponding data values ​​in the memory, a comparison curve chart of the gas explosion source position and intensity can be generated by calling the matplotlib function library.

[0076] In this embodiment, three key statistical indicators are used to evaluate the performance of the gas explosion source intensity inversion model based on convolutional neural network (CNN): mean square error (MSE), mean absolute error (MAE) and determination coefficient (R 2 These indicators together provide a comprehensive performance evaluation for this embodiment, which not only measures the accuracy of the model inversion, but also reflects the consistency between the model inversion results and the actual values.

[0077] The MSE value is 0.173, indicating the average error size of each inversion when the model inverts the intensity of the gas explosion source. A lower MSE value indicates that the model has a higher inversion accuracy. The MAE is 0.2021, which further verifies the accuracy and reliability of the model in practical applications. Compared with MSE, MAE is less sensitive to outliers and therefore provides another perspective on the inversion performance of the model. 2 The closer the value is to 1, the greater the variability explained by the model and the better the inversion effect. 2 The value is 0.9857, which indicates that the model is highly efficient in explaining the variability of gas explosion source intensity.

[0078] From the attached Figure 5It can be seen that the inversion of the location and intensity of the gas explosion source is accurate and reliable. Compared with the inversion of the location and intensity of the gas explosion source based on the position and overpressure data in previous research results, the explosion source location, explosion source intensity, and mean square error (MSE) have all dropped significantly, down to about 0.7. The values ​​of the mean square error and mean absolute error (MAE) are relatively close, and the coefficient of determination (R^2) score is close to 1, indicating that the model has a good fit. It can be concluded that this model shows good inversion accuracy and reliability in the task of inverting the intensity of the gas explosion source. These evaluation results not only prove the effectiveness of the proposed method, but also demonstrate the potential of the model in practical application scenarios.

[0079] This embodiment aims to solve the problem of inversion of gas explosion source intensity and position, comprehensively considers the maximum overpressure data and position data of the measuring point in the working condition, and deeply explores the interactive relationship between the overpressure and position data and the explosion source intensity and source point, so as to more accurately invert the gas explosion source intensity and position in the working condition. In order to make up for the shortcomings of the existing inversion method, this embodiment introduces a deep learning network model, and through adaptive adjustment of the network structure and parameter optimization, the loss function is close to zero, and good evaluation indicators and experimental results are achieved.

[0080] Figure 1 This is a flow chart of the inversion of the location and intensity of the gas explosion source in this embodiment, which specifically includes data acquisition, data preprocessing, feature extraction, model training, inversion calculation, adaptive optimization, and result output and evaluation.

[0081] Figure 2 This is the basic principle diagram of the inversion of gas explosion source in this embodiment. The one-dimensional inversion uses a recurrent neural network (RNN) to process the gas explosion source sequence data and predict the explosion location and intensity through time dependency. The two-dimensional inversion uses a convolutional neural network (CNN) to extract data features to achieve inversion prediction of the explosion location and intensity.

[0082] Figure 3 This is a gas explosion simulation experiment working condition diagram. The experimental pipeline is made up of short pipes of 2.5m in length each, so 2.5m is also the shortest length of the gas and other mixed gases, and the volume formed is 0.1m 3 And taking 2.5m as the basic unit, the explosion source intensity is indirectly changed by changing the number of pipelines, that is, the pipeline length (2.5m, 5m, 7.5m), so as to determine the inverted explosion source intensity data.

[0083] Figure 4This is a diagram of the structure of a recurrent neural network. A recurrent neural network (RNN) processes sequence data through a recurrently connected hidden layer, enabling it to remember previous input information. This structure enables the RNN to take into account previous inputs when processing the current input, thereby capturing temporal dependencies. The figure shows the relationship between the input layer, hidden layer, and output layer, highlighting the memory mechanism of the RNN when processing sequence data.

[0084] Figure 5 It is a one-dimensional intensity inversion model evaluation coefficient diagram, which shows the average error level of the model when inverting the intensity of gas explosion sources. A lower mean square error (MSE) value indicates that the model has a higher inversion accuracy. In contrast, the mean absolute error (MAE) is less sensitive to outliers, so it can be used to evaluate the inversion performance of the model from another perspective. The coefficient of determination (R 2 ) value is closer to 1, indicating that the variability explained by the model is greater and the inversion effect is better.

[0085] Figure 6 This is a comparison chart of the absolute error results of one-dimensional intensity and position inversion. It can be seen from the figure that the inversion accuracy of the position and intensity of the gas explosion source is high and the reliability is strong.

[0086] Figure 7 This is the interface display diagram of the inversion effect visualization module. Users can realize the inversion function by clicking the four buttons on the interface. Taking one-dimensional inversion as an example, when the data is ready, the user can click the "one-dimensional position inversion" button in the first row, and the system will output the inversion results of the corresponding data in the first row. Similarly, users can obtain the intensity inversion results by clicking the "one-dimensional intensity inversion" button. It should be noted that when the user re-enters the data or imports new data, the previous inversion results will not be saved when the button is clicked again. The inversion results of the second input data will overwrite the previous results. Therefore, if you need to save the output results, please manually copy and save them in the output box. If the input data contains known values ​​of the explosion source intensity and position, the user can enter these values ​​in the text box in the lower left corner, and then click the "generate image" button on the right. This operation will show the correspondence between the results of the built-in model inversion of the program and the true values, helping users to verify the pros and cons of the inversion results.

[0087] The inversion system of this embodiment includes: a data acquisition module, a data preprocessing module, a feature extraction module, a neural network model, an inversion calculation module and a result analysis module. The data acquisition module collects sensor data after the gas explosion. The data preprocessing module performs denoising and standardization on the collected data. The feature extraction module extracts characteristic parameters such as explosion waveform, arrival time and energy from the preprocessed data. The neural network model performs inversion calculation through the trained network model. The inversion calculation module calculates the intensity and position of the explosion source based on the neural network model. The result analysis module analyzes and evaluates the inversion result. The control terminal is used for system operation and result viewing. The data storage module is used to store historical data and model parameters, store collected sensor data, preprocessed data, characteristic parameters and inversion calculation results.

[0088] The data acquisition module includes multiple types of sensors, which respectively collect data such as pressure waves and measurement point positions during explosions. The data processing steps include data cleaning, normalization and denoising, using minimum-maximum normalization technology.

[0089] The neural network model includes a recurrent neural network (RNN) or a convolutional neural network (CNN), and the model parameters are determined by offline training. The neural network model is trained by the collected sensor data, and the training data includes experimental data and simulation data under various gas explosion scenarios. Grid search and cross-validation methods are used to adjust parameters, and early stopping method is used to avoid overfitting.

[0090] The result analysis module also includes data visualization tools to display the location and intensity of the explosion source obtained by inversion calculation, as well as the scope of the accident impact.

[0091] The control terminal includes a user interface for system operation and result viewing, and a user can view inversion results and historical data through the interface.

[0092] System functions and application scenarios: The system can monitor the location and intensity of gas explosion sources in real time and provide early warning functions in application scenarios such as mine monitoring and factory safety management. The system can operate stably under different environmental conditions, such as high temperature and high humidity environments, and ensure system reliability through the selection of high temperature and high humidity resistant materials and protective design.

[0093] The above is only a preferred specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for inversion of gas explosion source intensity and location based on deep learning, It is characterized in that include: Acquire gas explosion sensor data to be inverted; Inputting the gas explosion sensor data to be inverted into the gas explosion source inversion model for predictive inversion to obtain explosion source location data and explosion intensity data; wherein the gas explosion source inversion model includes a one-dimensional inversion model and a two-dimensional inversion model, and both the one-dimensional inversion model and the two-dimensional inversion model are constructed based on a neural network; The explosion source location data and the corresponding explosion intensity data are visualized.

2. According to claim 1, a method for inverting the intensity and position of a gas explosion source based on deep learning, It is characterized in that The gas explosion sensor data to be inverted includes sensor measuring point positions and explosion pressure wave data.

3. According to the method for inversion of gas explosion source intensity and position based on deep learning in claim 1, It is characterized in that Before the gas explosion sensor data to be inverted is input into the gas explosion source inversion model, the method also includes performing data cleaning, normalization and denoising on the gas explosion sensor data to be inverted to obtain preprocessed gas explosion sensor data to be inverted and using the preprocessed data as input of the gas explosion source inversion model.

4. According to claim 1, a method for inverting the intensity and position of a gas explosion source based on deep learning, It is characterized in that The training process of the gas explosion source inversion model specifically includes: Acquire training data, wherein the training data includes gas explosion sensor training data and corresponding explosion source data, wherein the explosion source data includes explosion source location data and explosion intensity data; An initial gas explosion source inversion model is constructed, the training data is input into the initial gas explosion source inversion model for predictive inversion, and training is performed with the goal of minimizing the loss between the initial training results after predictive inversion and the explosion source data corresponding to the gas explosion sensor training data to obtain the gas explosion source inversion model.

5. According to claim 1, a method for inverting the intensity and position of a gas explosion source based on deep learning, It is characterized in that The one-dimensional inversion model includes an input layer, a hidden layer and an output layer connected in sequence; The input layer receives the gas explosion sensor data to be inverted as an input vector, obtains the time series characteristics of the input vector through the hidden layer, and the output layer inverts one-dimensional explosion source position data and one-dimensional explosion intensity data according to the time series characteristics.

6. A method for inverting the intensity and position of a gas explosion source based on deep learning according to claim 1, It is characterized in that The two-dimensional inversion model includes a convolutional layer, a pooling layer and a fully connected layer connected in sequence; The convolution layer extracts features of the gas explosion sensor data to be inverted to obtain a feature map; the pooling layer downsamples the feature map to obtain a downsampled feature map; The outputs of the convolution layer and the pooling layer are feature mapped through the fully connected layer to output two-dimensional explosion source position data and two-dimensional explosion intensity data.

7. According to claim 1, a method for inverting the intensity and position of a gas explosion source based on deep learning, It is characterized in that The explosion source location data and the corresponding explosion intensity data are visualized, specifically including: constructing a user interaction interface based on the constructed gas explosion source inversion model and the corresponding model output results, and visually displaying the explosion source location data, explosion intensity data and inversion result analysis curves based on the user interaction interface.

8. A method for inverting the intensity and position of a gas explosion source based on deep learning according to claim 7, It is characterized in that The user interaction interface includes a storage sub-interface, a one-dimensional position inversion result sub-interface, a one-dimensional intensity inversion sub-interface, a two-dimensional position inversion result sub-interface, a two-dimensional intensity inversion sub-interface and an inversion result analysis curve sub-interface.

Images