A method, medium and system for establishing a dynamic distribution model of a gas flow field in a mining space

Abstract

The application provides a mining space gas flow field dynamic distribution model establishing method, medium and system, belongs to mining space gas flow field technical field, belong to, including: first, regularly collect the air data of multiple point positions in the mining space, including wind speed, wind direction, temperature, humidity, air pressure and gas concentration and the like. For each collection time, an interpolation method is used to construct an air dense matrix, which is combined into an air time series matrix along the time sequence. The matrix is normalized, and a multivariate mixed decomposition is used to obtain a stable and fluctuation time series matrix. Then, Fourier transform is performed on the two matrices respectively to obtain the frequency domain and time domain characteristics of stability and fluctuation. Based on these characteristics, a multi-layer tensor network structure is constructed, and a tensor decomposition method is used to extract low-dimensional feature tensors. Finally, the feature tensors are input into a pre-designed neural network model, and a model for effectively describing the dynamic distribution characteristics of gas in the mining space is trained.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of gas flow field in mining space, and particularly relates to a method for establishing a dynamic distribution model of a gas flow field in a mining space, a medium and a system. BACKGROUND

[0002] Current coal mining operations have been widely applied to underground mining, surface mining, open-pit mining and other modes. In these mining operations, gas is a safety hazard that must be focused on prevention and control. Gas is a naturally occurring flammable gas in coal seams or coal measures, mainly composed of methane, with the characteristics of high flammability and explosiveness. In the mining process, if the gas cannot be effectively controlled, it is easy to cause serious gas explosion accidents, causing huge casualties and property losses.

[0003] In order to realize the safe control of gas, the current mining enterprises generally establish a gas monitoring and early warning system. By arranging various types of sensing devices in the mining space, the parameters such as gas concentration, wind speed and direction, temperature and humidity are continuously monitored, and a special gas early warning model is used to analyze the monitoring data, so as to timely discover abnormal gas conditions and remind the mining personnel to take emergency measures. This gas early warning mechanism based on real-time monitoring and model analysis improves the safety of mining operations to a certain extent.

[0004] However, the existing gas early warning model has some limitations: these models usually only focus on the trend of gas concentration, and lack effective description of the dynamic distribution characteristics of gas in the mining space, making it difficult to accurately reflect the characteristics of gas flow field in complex operation environment. Therefore, it is urgent to develop a new type of gas flow field dynamic distribution model, which can more accurately describe and predict the gas distribution characteristics in complex mining environment, and provide more reliable technical support for gas safety control. SUMMARY

[0005] Therefore, the present application provides a method for establishing a dynamic distribution model of a gas flow field in a mining space, a medium and a system, which can solve the technical problem that the prior art usually only focuses on the trend of gas concentration and lacks effective description of the dynamic distribution characteristics of gas in the mining space.

[0006] The present application is implemented as follows:

[0007] The first aspect of the present application provides a method for establishing a dynamic distribution model of a gas flow field in a mining space, comprising the following steps:

[0008] S10, collecting air data of a plurality of collection points in the mining space every preset time interval, including wind speed, wind direction, temperature, humidity, air pressure and gas concentration;

[0009] S20, for each collection time, air data of each collection point is used to establish an air matrix, and an interpolation method is used to interpolate and supplement the space between the collection points, and the air matrix is converted into an air dense matrix;

[0010] S30, the air dense matrix of each collection time is combined into an air time sequence matrix according to a time sequence, and each element of the air time sequence matrix is normalized to obtain an air normalized time sequence matrix;

[0011] S40, the air normalized time sequence matrix is decomposed by using multivariate mixed decomposition calculation to obtain a stable time sequence matrix and a fluctuation time sequence matrix;

[0012] S50, the stable time sequence matrix and the fluctuation time sequence matrix are respectively subjected to fast Fourier transform, and the original time domain information is retained to obtain a stable frequency domain feature matrix, a stable time domain feature matrix, a fluctuation frequency domain feature matrix and a fluctuation time domain feature matrix;

[0013] S60, a multi-layer tensor network structure is constructed, the stable frequency domain feature matrix, the stable time domain feature matrix, the fluctuation frequency domain feature matrix and the fluctuation time domain feature matrix are used as input tensors, and a tensor decomposition method is used for dimension reduction and feature extraction of the input tensors to obtain a low-dimensional representation of the feature tensors;

[0014] S70, the feature tensors are input into a pre-designed embedded multi-branch pyramid neural network for training to obtain a gas flow field dynamic distribution model of the mining space.

[0015] Specifically, the step S10 specifically includes: a plurality of collection points are arranged in the mining space, wind speed sensors, wind direction sensors, temperature sensors, humidity sensors, pressure sensors and gas concentration sensors are arranged, and a predetermined collection time interval is set, for example, collection is performed every 10 minutes. At each collection time, these sensors will automatically record the air data of the wind speed, wind direction, temperature, humidity, air pressure and gas concentration monitored at the current time. After a certain time of continuous collection, the air data of the plurality of collection points at different times in the mining space can be obtained, which lays a foundation for the subsequent step of establishing a gas flow field model.

[0016] The step S20 specifically includes: first, an air matrix is constructed by using the air data of each collection point at each collection time, and then a bilinear interpolation method is used to interpolate and supplement the air matrix to obtain a more dense air dense matrix. In this way, the distribution of each measurement parameter in the mining space at the time can be more comprehensively reflected.

[0017] The step S30 specifically comprises: stacking the air dense matrix of each collection time according to time sequence to form a three-dimensional air time sequence matrix.

[0018] The step S40 specifically comprises: decomposing the normalized time sequence matrix by using a multivariate mixed decomposition algorithm to decompose it into two relatively independent parts, one is a stable time sequence matrix reflecting the stable change trend, and the other is a fluctuation time sequence matrix describing the fluctuation change characteristics. This decomposition is conducive to more comprehensively depicting the dynamic distribution characteristics of the gas flow field.

[0019] The step S50 specifically comprises: respectively performing fast Fourier transform on the stable time sequence matrix and the fluctuation time sequence matrix to obtain their feature matrices in the frequency domain, while the original time domain information is also retained. In this way, the frequency domain characteristics of the flow field change can be reflected, and the time domain characteristics can also be retained, which is helpful for more comprehensively depicting the dynamic characteristics of the gas flow field.

[0020] The step S60 specifically comprises: constructing a multi-layer tensor network structure, splicing the aforementioned extracted stable frequency domain feature matrix, stable time domain feature matrix, fluctuation frequency domain feature matrix and fluctuation time domain feature matrix into a 4-order input tensor, and then performing dimension reduction and feature extraction on the tensor by using a Tucker decomposition algorithm to obtain a lower-dimensional feature tensor. The feature tensor will be used as the input of the subsequent neural network model for learning and predicting the dynamic distribution of the gas flow field.

[0021] The step S70 specifically comprises: designing an embedded multi-branch pyramid neural network structure, which comprises: an embedded equation set module, an equation set error term branch network, an adaptive resolution module, four resolution model branch networks and a fusion and summary network. The embedded equation set module calculates the basic distribution characteristics according to the physical characteristics of the gas flow field and the air data, the equation set error term branch network is used to compensate the calculation error of the module, the adaptive resolution module can dynamically adjust the network resolution according to the input data, the four resolution model branch networks respectively generate gas flow field distribution results of different resolutions, and the fusion and summary network integrates these results to output the final multi-resolution gas flow field dynamic distribution model. This network structure can fully utilize the physical characteristics of the gas flow field and the multi-scale space-time information, and greatly improve the expression ability of the model.

[0022] On the basis of the above technical scheme, the mining space gas flow field dynamic distribution model establishing method of the application can be further improved as follows:

[0023] The embedded multi-branch pyramid neural network comprises an embedded equation set module, an equation set error term branch network, an adaptive resolution module, a first resolution model branch network, a second resolution model branch network, a third resolution model branch network, a fourth resolution model branch network and a fusion and aggregation network.

[0024] Further, the embedded equation set module is used to calculate the basic distribution characteristics of the gas flow field according to the physical characteristics of the gas flow field and the air data, including a gas diffusion equation, a gas flow velocity field equation, a gas pressure field equation and a gas distribution equation.

[0025] The gas diffusion equation considers the concentration gradient and the convection effect, and is used to describe the diffusion process of gas in the mining space.

[0026] The gas flow velocity field equation is based on the Navier-Stokes equation, and is used to describe the motion characteristics of gas in the mining space.

[0027] The gas pressure field equation is based on the state equation and the mass conservation principle, and is used to describe the distribution and change of the gas pressure in the mining space.

[0028] The gas distribution equation considers the source and sink terms and the boundary conditions, and is used to describe the overall distribution characteristics of gas in the mining space.

[0029] Specifically, the embedded equation set specifically comprises:

[0030] a) gas diffusion equation:

[0031]

[0032] In the formula, c is the gas concentration (mol / m 3 ), t is the time (s), u is the flow velocity vector (m / s), D is the diffusion coefficient (m 2 / s), Q s is the source term (mol / (m 3 ·s)), and Q r is the sink term (mol / (m 3 ·s).

[0033] Parameter acquisition method:

[0034] c is obtained by measuring the gas concentration sensor;

[0035] u is obtained by measuring the wind speed sensor;

[0036] D is obtained by fitting experimental data, D=D0·(T / T0) 1.75• (p0 / p), where D0 is the diffusion coefficient at standard state, T is the temperature, T0 is the standard temperature, p is the pressure, p0 is the standard pressure;

[0037] Q s and Q r Obtained by experimental measurement or numerical simulation.

[0038] b) Gas velocity field equation:

[0039]

[0040] where ρ is the gas density (kg / m 3 ), u is the velocity vector (m / s), p is the pressure (Pa), μ is the dynamic viscosity (Pa·s), g is the gravity acceleration vector (m / s 2 ), and F is the external force (such as the effect of ventilation) (N / m 3 ).

[0041] Parameter acquisition method:

[0042] ρ is calculated by the gas state equation, ρ = pM / (RT), where M is the molar mass, R is the gas constant, and T is the temperature;

[0043] u is obtained by measuring the wind speed sensor;

[0044] p is obtained by measuring the pressure sensor;

[0045] μ is obtained by experimental measurement or table lookup;

[0046] g is a known constant;

[0047] F is obtained by experimental measurement or numerical simulation.

[0048] c) Gas pressure field equation:

[0049]

[0050] where p is the pressure (Pa), u is the velocity vector (m / s), γ is the specific heat ratio (dimensionless), k is the permeability (m 2 ), φ is the porosity (dimensionless), μ is the dynamic viscosity (Pa·s), and Q p is the pressure source term (Pa / s).

[0051] Parameter acquisition method:

[0052] p is obtained by measuring the pressure sensor;

[0053] u is obtained by measuring the wind speed sensor;

[0054] γ is obtained by experimental measurement or table lookup;

[0055] k and φ are determined by rock sample experiments;

[0056] μ is obtained by experiment measurement or table lookup;

[0057] Q p It is estimated by experiment measurement or numerical simulation.

[0058] d) Gas distribution equation:

[0059]

[0060] In the formula, c is the gas concentration (mol / m 3 ), t is the time (s), u is the flow velocity vector (m / s), d is the diffusion coefficient (m 2 / s), Q s is the source term (mol / (m 3 ·s)), Q r is the sink term (mol / (m 3 ·s)), λ is the gas attenuation coefficient (1 / s), K ij is the turbulent diffusion tensor (m 2 / s).

[0061] Parameter acquisition method:

[0062] The acquisition methods of c, u, D, Q s and Q r are the same as the gas diffusion equation;

[0063] λ is obtained by experiment measurement, λ = ln2 / t 1 / 2 , where t 1 / 2 is the gas concentration half-life;

[0064] K ij is calculated by the turbulent flow model, where C μ is an empirical constant, k is the turbulent kinetic energy, and ε is the turbulent dissipation rate.

[0065] These equations constitute a complete physical model for describing the dynamic distribution of gas flow field in mining space, covering gas diffusion, flow, pressure change and overall distribution characteristics.

[0066] Further, the equation group error term branch network is used to compensate the calculation error of the embedded equation group module, the input is the output of the embedded equation group module and the actual measurement data, the output is the error compensation term, and the structure is a multi-layer perception network.

[0067] Further, the adaptive resolution module is used to dynamically adjust the network resolution according to the characteristics of the input data, specifically by analyzing the spatial and temporal variation characteristics of the input data to determine the most suitable resolution, the input is a feature tensor, the output is a resolution selection signal, and the structure is an adaptive network obtained by replacing the convolution module of the ResNet network with a hollow convolution and deleting the fully connected layer.

[0068] Further, the first resolution model branch network is used to generate a gas flow field distribution of the first resolution, the input is a feature tensor and the output of the embedded equation set module, and the output is a first resolution gas flow field distribution, and the structure is a U-Net network;

[0069] The second resolution model branch network is used to generate a gas flow field distribution of the second resolution, the input is a feature tensor and a down-sampling result of the output of the first resolution model, and the output is a second resolution gas flow field distribution, and the structure is a simplified version of the U-Net network followed by a GFSR module, which increases the second resolution to twice the first resolution;

[0070] The third resolution model branch network is used to generate a gas flow field distribution of the third resolution, the input is a feature tensor and a down-sampling result of the output of the second resolution model, and the output is a third resolution gas flow field distribution, and the structure is a ResNet-18 network followed by two GFSR modules in series, which increases the third resolution to four times the first resolution;

[0071] The fourth resolution model branch network is used to generate a gas flow field distribution of the fourth resolution, the input is a feature tensor and a down-sampling result of the output of the third resolution model, and the output is a fourth resolution gas flow field distribution, and the structure is a DenseNet-121 network followed by three GFSR modules in series, which increases the fourth resolution to eight times the first resolution.

[0072] Further, the fusion summary network is used to integrate the gas flow field distribution results of different resolutions, the input is the gas flow field distribution results of four resolutions, and the output is the final multi-resolution gas flow field dynamic distribution model, and the structure is an attention mechanism fusion network.

[0073] Further, the pyramid structure of the embedded multi-branch pyramid neural network, specifically, the spatial resolution is gradually reduced from the first resolution to the fourth resolution to expand the receptive field, and then the spatial resolution is gradually increased through the GFSR algorithm while maintaining information transmission and fusion between different resolutions.

[0074] Wherein, each higher resolution network receives the output from the lower resolution, and performs super-resolution processing and feature fusion through GFSR, thereby realizing effective integration of multi-scale information.

[0075] the first resolution, specifically the original grid resolution of the mining space;

[0076] the second resolution, specifically 2 times the first resolution;

[0077] the third resolution, specifically 4 times the first resolution;

[0078] the fourth resolution, specifically 8 times the first resolution;

[0079] This improved structural design not only captures the changing characteristics of the gas flow field at different spatial scales, but also reconstructs high-resolution detail information through the GFSR algorithm, greatly improving the model's expression ability for complex mining space gas distribution. The physical constraint loss function of the GFSR algorithm ensures the preservation of the physical characteristics of the gas flow field during the super-resolution process, making the model output both high-resolution and consistent with the actual gas flow rules.

[0080] wherein GFSR represents a gas flow super-resolution module, and the structure thereof includes:

[0081] a) Generator network:

[0082] The generator network uses a residual dense block (RDB) as a basic unit, and the specific structure is as follows:

[0083]

[0084] wherein F RDB (x) is the output of the RDB, x is the input, is a residual learning function, defined as:

[0085]

[0086] In the formula, the generator network includes 5 convolutional layers, W1-W5 are the convolutional layer weights of the generator network, and σ is an activation function (such as ReLU).

[0087] b) Discriminator network:

[0088] The discriminator adopts a PatchGAN structure, and the output is an NxN matrix, each element of which represents the authenticity score of the corresponding patch:

[0089] D(x) = σ'(W' n · f n-1 (W' n-1 ·... f1(W'1· x'))));

[0090] Where D(x) is the discriminator output, representing the authenticity score of the input x′, x′ is the input of the discriminator network, and W′ i Here, f represents the convolutional layer weights of the discriminator network, f is the feature extraction function, σ′ is the sigmoid activation function, and n represents the number of convolutional layers in the discriminator network, which defaults to 3. In other words, the output of the discriminator network is an N×N matrix, where each element represents the realism score of the corresponding image patch; N defaults to 70.

[0091] c) Loss function:

[0092] The total loss function consists of three parts:

[0093] L total =αL percep +βL adv +γL phys ;

[0094] Among them, L percep To perceive loss, L adv To combat the losses, L phys The loss is the physical constraint loss, and α, β, and γ are the weighting coefficients.

[0095] The GFSR module executes the following steps:

[0096] Step 1: Input the low-resolution gas flow field data into the generator network.

[0097] Step 2: The generator outputs high-resolution gas flow field data.

[0098] Step 3: The discriminator evaluates the generated high-resolution data and the real high-resolution data respectively.

[0099] Step 4: Calculate the total loss function.

[0100] Step 5: Update the parameters of the generator and discriminator through backpropagation.

[0101] Step 6: Repeat steps 1-5 until the model converges.

[0102] The training dataset for the GFSR module is as follows:

[0103] The training dataset consists of low-resolution and high-resolution gas flow field data pairs. The acquisition method is as follows:

[0104] Step 1: Use high-precision sensors to intensively sample the mining space to obtain high-resolution gas flow field data.

[0105] Step 2: Downsample the high-resolution data to obtain the corresponding low-resolution data.

[0106] Step 3: Pairing high-resolution data and low-resolution data to form training samples.

[0107] The training process of the GFSR module is as follows:

[0108] Step 1: Initialize the generator and discriminator network parameters.

[0109] Step 2: Randomly sample a batch of samples from the training dataset.

[0110] Step 3: Input the low-resolution sample into the generator to obtain the generated high-resolution result.

[0111] Step 4: Calculate the loss function.

[0112] Step 5: Update the network parameters through backpropagation.

[0113] Step 6: Repeat steps 2-5 until the model converges or reaches a preset number of iterations.

[0114] The second aspect of the application provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and the program instructions are used to execute the above-mentioned mining space gas flow field dynamic distribution model establishment method when running in a computer.

[0115] The third aspect of the application provides a mining space gas flow field dynamic distribution model establishment system, which comprises the above-mentioned computer-readable storage medium.

[0116] Compared with the prior art, the mining space gas flow field dynamic distribution model establishment method, medium and system provided by the application have the following advantages:

[0117] Firstly, the method fully utilizes multi-source air monitoring data, including wind speed, wind direction, temperature, humidity, air pressure and gas concentration, to comprehensively depict the dynamic change characteristics of the gas flow field in the mining space, which can not only reflect the change trend of the gas concentration, but also describe the flow distribution of the gas in the space.

[0118] Secondly, the method uses mathematical analysis methods such as multivariate mixed decomposition and fast Fourier transform to decompose the original complex spatiotemporal gas flow field data into two relatively independent parts of stable change and fluctuation change, which can not only extract the long-term trend characteristics of the flow field, but also capture the short-term fluctuation characteristics, thereby greatly improving the expression ability of the model for the gas distribution law under complex mining environment.

[0119] Further, the method designs an embedded multi-branch pyramid neural network structure, which combines the physical model and the data-driven model. On the one hand, the network contains an equation module based on the physical law of the gas flow field, which can calculate the basic distribution characteristics of gas diffusion, flow, and pressure using prior knowledge; on the other hand, the network also has the ability of adaptive resolution adjustment and multi-resolution feature fusion, which can extract representative spatio-temporal features from the original monitoring data, and realize accurate modeling of complex gas flow field through end-to-end deep learning. This fusion innovation not only ensures the physical rationality of the model output, but also greatly improves the expression ability of the actual gas distribution law.

[0120] Finally, the GFSR super-resolution module in the method can reconstruct high-resolution detail information from low-resolution gas flow field data, which realizes fine description of the complex mining environment under the premise of ensuring physical constraints. This is not only conducive to better analyzing the variation characteristics of gas at different scales, but also provides more reliable data support for subsequent gas safety warning.

[0121] In summary, the mining space gas flow field dynamic distribution model establishment method proposed by the present application makes innovative breakthroughs in mathematical analysis, deep learning modeling, and physical constraint fusion, and significantly improves the expression ability of the gas distribution law in complex mining environment compared with the prior art, solving the technical problem that the prior art usually only focuses on the change trend of gas concentration and lacks effective description of the dynamic distribution characteristics of gas in the mining space. BRIEF DESCRIPTION OF DRAWINGS

[0122] Figure 1 The flowchart of the method provided by the present application is shown in Figure 1.

[0123] Figure 2 is a multi-resolution gas distribution comparison chart;

[0124] Figure 3 is a gas concentration dynamic change prediction chart;

[0125] Figure 4 is a multi-parameter gas flow field analysis chart;

[0126] Figure 5 is a safety risk assessment heat map;

[0127] Figure 6 is a 3D scatter plot of gas concentration, temperature, and humidity. DETAILED DESCRIPTION

[0128] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be described clearly and completely with reference to the drawings in the embodiments of the present application.

[0129] As Figure 1 shown is a flow chart of a method for establishing a gas flow field dynamic distribution model of a mining space according to the first aspect of the present application, the method comprising the following steps:

[0130] S10, collecting air data of a plurality of collection points in the mining space every preset time interval, including wind speed, wind direction, temperature, humidity, air pressure and gas concentration;

[0131] S20, for each collection time, using the air data of each collection point to establish an air matrix, and using an interpolation method to interpolate and supplement the space between the collection points, and converting the air matrix into an air dense matrix;

[0132] S30, merging the air dense matrix of each collection time into an air time sequence matrix according to the time sequence, and normalizing each element of the air time sequence matrix to obtain an air normalized time sequence matrix;

[0133] S40, decomposing the air normalized time sequence matrix using multivariate mixed decomposition calculation to obtain a stable time sequence matrix and a fluctuation time sequence matrix;

[0134] S50, performing fast Fourier transform on the stable time sequence matrix and the fluctuation time sequence matrix respectively, and retaining the original time domain information to obtain a stable frequency domain feature matrix, a stable time domain feature matrix, a fluctuation frequency domain feature matrix and a fluctuation time domain feature matrix;

[0135] S60, constructing a multi-layer tensor network structure, taking the stable frequency domain feature matrix, the stable time domain feature matrix, the fluctuation frequency domain feature matrix and the fluctuation time domain feature matrix as input tensors; using a tensor decomposition method to reduce the dimension and extract features of the input tensors to obtain a low-dimensional representation of the feature tensors;

[0136] S70, inputting the feature tensors into a pre-designed embedded multi-branch pyramid neural network for training to obtain a gas flow field dynamic distribution model of the mining space.

[0137] The specific implementation of the above steps is described in detail as follows:

[0138] The specific implementation of step S10 is to periodically collect at preset collection points in the mining space using multiple sensor devices. First, a plurality of collection points are set in the mining space, and the positions of the collection points need to fully consider the uniformity of the spatial distribution to ensure that the gas flow field characteristics of the entire mining space are comprehensively and accurately collected. Second, wind speed sensors, wind direction sensors, temperature sensors, humidity sensors, pressure sensors, and gas concentration sensors are deployed at each collection point. These sensors can monitor the air flow characteristics and gas concentration at the collection point in real time. Third, a predetermined collection time interval is set, such as every 10 minutes. This time interval needs to be determined according to the actual gas flow field change speed of the mining site to ensure that the collection frequency can fully reflect the dynamic changes of the gas flow field. At each collection time, these sensors automatically record the data monitored at the current time, including wind speed, wind direction, temperature, humidity, air pressure, and gas concentration. After a certain period of continuous collection, the air data of the multiple collection points in the mining space at different times can be obtained. These data lay the foundation for the subsequent step of establishing a gas flow field model.

[0139] The specific implementation of step S20 is to first process and integrate the data at each collection time. For a collection time t, an air matrix A(t) is constructed using the air data of each collection point at that time. The number of rows of this air matrix is equal to the total number of collection points, and the number of columns is 6, corresponding to the wind speed, wind direction, temperature, humidity, air pressure, and gas concentration. That is, the element a i,j of the matrix A(t) represents the value of the jth parameter measured by the ith collection point at time t.

[0140] Because the distribution of collection points is often not dense enough to completely cover the entire mining space, an interpolation method is needed to fill and supplement the space between collection points. Common interpolation methods include Kriging interpolation, spline interpolation, etc. Here, the most common bilinear interpolation method is used. Specifically, the air matrix A(t) is interpolated and supplemented to obtain a more dense air dense matrix B(t). The number of rows and columns of this dense matrix is greatly increased, which can more comprehensively reflect the distribution of each measurement parameter in the mining space at that time.

[0141] The specific implementation of step S30 is to stack the air dense matrix B(t) at each collection time in time sequence to form a three-dimensional air time sequence matrix C. The first dimension of this time sequence matrix represents time, and the second and third dimensions correspond to spatial coordinates. That is, the element c i,j,k of the matrix C represents the value of a certain parameter measured at position (j, k) in the mining space at time i.

[0142] Due to the large difference in the dimension and numerical range between different parameters, in order to eliminate the influence of such difference on subsequent analysis, it is necessary to normalize each element of the time series matrix C. The specific normalization method can adopt minimum-maximum normalization or Z-score normalization, etc. The matrix obtained after normalization is denoted as D, which is called air normalized time series matrix.

[0143] The specific implementation of step S40 is to decompose the normalized time series matrix D by using a multilinear decomposition algorithm. This algorithm can decompose the three-dimensional matrix D into the product of two two-dimensional matrices and a diagonal matrix. Among them, the two two-dimensional matrices represent the eigenvectors of the time dimension and the space dimension respectively, and the diagonal matrix contains the singular values of these eigenvectors. Through this decomposition, the original complex time series matrix D can be decomposed into two relatively independent parts: one is a stable time series matrix E reflecting the stable change trend, and the other is a fluctuation time series matrix F describing the fluctuation change characteristics.

[0144] The specific implementation of step S50 is to perform Fast Fourier Transform (FFT) on the stable time series matrix E and the fluctuation time series matrix F respectively. FFT is a high-efficiency discrete Fourier transform algorithm that can convert time-domain signals into frequency-domain representation. By performing FFT transform on E and F, their characteristic matrices in the frequency domain can be obtained, denoted as G and H respectively. At the same time, in order to preserve the original time-domain information, it is also necessary to preserve the time-domain characteristic matrices of E and F, denoted as I and J.

[0145] In summary, the purpose of steps S40 and S50 is to extract two types of features from the gas flow field time series data: one type is the feature reflecting the long-term stable change trend, and the other type is the feature describing the short-term fluctuation change. By preserving both time-domain and frequency-domain features, the dynamic distribution characteristics of the gas flow field can be more comprehensively described.

[0146] The specific implementation of step S60 is to construct a multi-layer tensor network structure, taking the aforementioned extracted feature matrices as input. This network structure includes the following modules:

[0147] 1. Feature input module: splice the stable frequency-domain feature matrix G, the stable time-domain feature matrix I, the fluctuation frequency-domain feature matrix H, and the fluctuation time-domain feature matrix J into a 4-order input tensor.

[0148] 2. Tensor decomposition module: A Tucker decomposition algorithm is used to reduce the dimensionality and extract features from the input tensor. Tucker decomposition can decompose a high-dimensional tensor into a core tensor and several low-dimensional feature matrices in the form of a product. Through this decomposition, a lower-dimensional feature tensor can be obtained, which contains the main features of the input tensor.

[0149] 3. Supervised learning module: The extracted feature tensor is input into a pre-designed neural network model for supervised learning training. The specific structure of this neural network model will be described in detail in step S70.

[0150] Through this multi-layer tensor network structure, representative features can be effectively extracted from the original spatio-temporal gas flow field data, providing important input for subsequent gas flow field model construction.

[0151] The specific implementation of step S70 is to design an embedded multi-branch pyramid neural network structure to realize the training and prediction of the gas flow field dynamic distribution model. This network structure includes the following key modules:

[0152] 1. Embedded equation set module: This module calculates the basic distribution characteristics of the gas flow field based on the physical characteristics of the gas flow field and the air data. Specifically, it includes the gas diffusion equation, gas flow velocity field equation, gas pressure field equation, and gas distribution equation. These physical equations can describe the diffusion, flow, pressure change, and overall distribution characteristics of gas in the mining space.

[0153] 2. Equation set error term branch network: This branch network compensates for the calculation errors of the embedded equation set module. It takes the output of the embedded equation set module and the actual measurement data as input and outputs an error compensation term. This network uses a multi-layer perceptron (MLP) structure to implement.

[0154] 3. Adaptive resolution module: This module can dynamically adjust the resolution of the network according to the characteristics of the input data. It analyzes the spatial and temporal variation characteristics of the input data and selects the most appropriate resolution to output. This module uses an improved ResNet network structure, replaces the original convolution module with a dilated convolution, and removes the fully connected layer.

[0155] 4. Multi-resolution model branch network: Four resolution model branch networks are designed. The first resolution model uses a U-Net network structure, the second resolution model connects a GFSR module based on a simplified U-Net, the third resolution model uses a ResNet-18 network with two cascaded GFSR modules, and the fourth resolution model uses a DenseNet-121 with three cascaded GFSR modules. These networks can generate gas flow field distribution results with different resolutions.

[0156] 5. Fusion aggregation network: the role of this network is to integrate the distribution results of gas flow field of different resolutions, and output the final multi-resolution gas flow field dynamic distribution model. It adopts an attention mechanism fusion network structure.

[0157] This embedded multi-branch pyramid neural network structure can fully utilize the physical characteristics and multi-scale spatio-temporal information of the gas flow field, and effectively capture the dynamic change law of the complex mining space gas flow field. The GFSR module can reconstruct the high-resolution gas flow field details according to the low-resolution input, further improving the expression ability of the model. The whole network structure embodies the fusion innovation from the physical model to the data-driven model.

[0158] The second aspect of the present application provides a computer readable storage medium, wherein the computer readable storage medium stores program instructions, and the program instructions are used to execute the above-mentioned mining space gas flow field dynamic distribution model establishment method when running in a computer.

[0159] The third aspect of the present application provides a mining space gas flow field dynamic distribution model establishment system, which comprises the above-mentioned computer readable storage medium.

[0160] In order to better understand and implement the present application, a specific embodiment 1 of the method of the present application is provided below, and the steps of the embodiment 1 are described in detail as follows: first, in step S10, a plurality of collection points in the mining space need to be deployed with various sensor devices for regularly collecting air data. Specifically, each collection point should be configured with a wind speed sensor, a wind direction sensor, a temperature sensor, a humidity sensor, a pressure sensor and a gas concentration sensor, etc. These sensors can measure and record the current time parameters such as wind speed u, wind direction θ, temperature T, humidity H, air pressure p and gas concentration c. A predetermined collection time interval Δt is set, such as collecting once every 10 minutes. After a certain time of continuous collection, the air data of multiple collection points in the mining space at different times can be obtained. These data can be expressed as:

[0161]

[0162] Wherein, x i and y i represent the spatial coordinates of the i-th collection point, t j represents the j-th collection time. Through these air data, the foundation for the subsequent step of gas flow field model establishment can be laid.

[0163] In step S20, the data of each collection time needs to be processed and integrated. For a collection time t j, the air data of each collection point at this moment is used to construct an air matrix A(t j ):

[0164]

[0165] where n represents the total number of collection points. Since the distribution of collection points is often not dense enough, an interpolation method needs to be used to fill and supplement the space between the collection points. Common interpolation methods include Kriging interpolation, spline interpolation, etc. Here, the most common bilinear interpolation method is used. Specifically, the air matrix A(t j ) is interpolated and supplemented to obtain a more dense air dense matrix B(t j ):

[0166]

[0167] Here, m represents the number of rows and columns of the dense matrix, which is much larger than the number n of collection points. In this way, the distribution of each measurement parameter in the mining space at this moment can be more comprehensively reflected.

[0168] In step S30, the air dense matrix B(t j ) at each collection moment needs to be stacked in time sequence to form a three-dimensional air time sequence matrix C:

[0169]

[0170] where k represents the total number of collection moments. Since the dimensions and numerical ranges of different parameters differ greatly, in order to eliminate the influence of such differences on subsequent analysis, each element of the time sequence matrix C needs to be normalized. Common normalization methods include minimum-maximum normalization and Z-score normalization, and here Z-score normalization is used:

[0171]

[0172] where x represents the normalized element value, x i,j,l represents the original element value, μ i,j,l and σ i,j,l represent the mean and standard deviation of the element, respectively. The matrix obtained after normalization is denoted as D, which is called the air normalized time sequence matrix.

[0173] In step S40, the normalized time sequence matrix D needs to be decomposed using a multilinear decomposition algorithm. This algorithm can decompose the three-dimensional matrix D into the product form of two two-dimensional matrices and a diagonal matrix:

[0174] D = G x 1 U x 2 V T ;

[0175] where G is a diagonal matrix containing the singular values of D; U and V are two orthogonal matrices representing the eigenvectors of the time and spatial dimensions, respectively. Through this decomposition, the original complex time-series matrix D can be decomposed into two relatively independent parts:

[0176] E = G x 1 U;

[0177] F = G x 2 V T ;

[0178] where E is a stable time-series matrix reflecting stable trends, and F is a fluctuating time-series matrix describing fluctuation characteristics. This decomposition is beneficial for more comprehensively characterizing the dynamic distribution characteristics of the gas flow field.

[0179] In step S50, Fast Fourier Transform (FFT) needs to be performed on the stable time-series matrix E and the fluctuating time-series matrix F, respectively. FFT is a high-efficiency discrete Fourier transform algorithm that can convert time-domain signals into frequency-domain representations. Specifically, performing FFT on E and F can obtain their characteristic matrices in the frequency domain:

[0180]

[0181]

[0182] where denotes the FFT transform operator. At the same time, in order to preserve the original time-domain information, the time-domain characteristic matrices of E and F also need to be preserved:

[0183] I = E;

[0184] J = F;

[0185] By retaining both time-domain and frequency-domain characteristics, the dynamic distribution characteristics of the gas flow field can be more comprehensively characterized.

[0186] In step S60, a multi-layer tensor network structure needs to be constructed, taking the aforementioned extracted characteristic matrices G, I, H, and J as inputs. This network structure includes the following modules:

[0187] 1. Feature input module: concatenate the four characteristic matrices into a 4-order input tensor X:

[0188]

[0189] 2. Tensor decomposition module: Apply Tucker decomposition algorithm to input tensor X for dimension reduction and feature extraction. Tucker decomposition can decompose a high-dimensional tensor X into a core tensor and several low-dimensional feature matrices U, V, W in the form of product:

[0190]

[0191] Through this decomposition, a lower-dimensional feature tensor can be obtained, which contains the main features of the input tensor X.

[0192] 3. Supervised learning module: input the extracted feature tensor into a pre-designed neural network model for supervised learning training. The specific structure of this neural network model will be introduced in step S70.

[0193] In summary, through this multi-layer tensor network structure, representative features can be effectively extracted from the original spatio-temporal gas flow field data, providing important input for subsequent gas flow field model construction.

[0194] Finally, in step S70, an embedded multi-branch pyramid neural network structure needs to be designed to realize the training and prediction of the gas flow field dynamic distribution model. This network structure includes the following key modules:

[0195] 1. Embedded equation set module: This module calculates the basic distribution characteristics of the gas flow field according to the physical characteristics of the gas flow field and air data. Specifically, it includes the following physical equations:

[0196] Gas diffusion equation:

[0197]

[0198] Gas flow field equation:

[0199]

[0200] Gas pressure field equation:

[0201]

[0202] Gas distribution equation:

[0203]

[0204] These physical equations can describe the diffusion, flow, pressure change, and overall distribution characteristics of gas in the mining space.

[0205] 2. System of equations error term branch network: The role of this branch network is to compensate for the calculation error of the embedded system of equations module. It takes the output of the embedded system of equations module and the actual measurement data as input, and outputs an error compensation term. The network uses a multi-layer perception (MLP) structure to achieve this.

[0206] 3. Adaptive resolution module: This module can dynamically adjust the resolution of the network according to the characteristics of the input data. It analyzes the spatial and temporal variation characteristics of the input data and selects the most appropriate resolution as the output. The module uses an improved ResNet network structure, replaces the original convolution module with a hollow convolution, and deletes the fully connected layer.

[0207] 4. Multi-resolution model branch network: Four resolution model branch networks are designed here. The first resolution model uses a U-Net network structure, the second resolution model connects a GFSR module based on a simplified U-Net, the third resolution model uses a ResNet-18 network plus two GFSR modules, and the fourth resolution model uses a DenseNet-121 plus three GFSR modules. These networks can generate gas flow field distribution results of different resolutions.

[0208] 5. Fusion and aggregation network: The role of this network is to integrate the gas flow field distribution results of different resolutions and output the final multi-resolution gas flow field dynamic distribution model. It uses an attention mechanism fusion network structure.

[0209] This embedded multi-branch pyramid neural network structure can fully utilize the physical characteristics of the gas flow field and the multi-scale spatio-temporal information, effectively capturing the dynamic variation of the complex mining space gas flow field. The GFSR module can reconstruct the high-resolution gas flow field details from the low-resolution input, further improving the model's expression ability.

[0210] In summary, this embodiment 1 can construct a high-precision, high-resolution gas flow field dynamic distribution model through the specific implementation of the above 7 steps, providing important decision support for mining control and safety production.

[0211] Specifically, the principle of the present application is: make full use of multi-source air monitoring data, and accurately capture the dynamic variation characteristics of the gas flow field in the mining environment through mathematical analysis and deep learning combined modeling method. Specifically, this method mainly includes the following key technical links:

[0212] 1. Data acquisition and preprocessing

[0213] First, multiple monitoring points are set up within the mining space, and sensors for wind speed, wind direction, temperature, humidity, pressure, and gas concentration are deployed to collect data on various air parameters on a regular basis. These data lay the foundation for subsequent flow field characteristic analysis and model establishment. In order to eliminate the differences in dimensions and numerical ranges between different parameters, normalization processing is required for the collected data.

[0214] 2. Flow field characteristic analysis

[0215] For each collection time, a spatial matrix is constructed using the air data from each monitoring point at that time, and then an interpolation method is used to supplement the matrix to obtain a more dense air dense matrix. All the dense matrices at different times are stacked in time sequence to form a three-dimensional spatiotemporal gas flow field data matrix. Next, a multivariate mixed decomposition algorithm is used to decompose the matrix into a matrix reflecting the stable change trend and a matrix describing the fluctuation change characteristics. Meanwhile, fast Fourier transform is performed on these two matrices to extract their features in the frequency and time domains. These features depict the dynamic change law of the gas flow field at different time scales, laying the foundation for subsequent deep learning modeling.

[0216] 3. Deep learning modeling

[0217] Based on the aforementioned extracted spatiotemporal features, an embedded multi-branch pyramid neural network is designed. The network includes four key modules:

[0218] (1) Embedded equation set module: This module constructs a set of partial differential equations describing the basic distribution law of gas based on the physical characteristics of the gas flow field, such as gas diffusion, flow, pressure change, etc. These equations can calculate the preliminary distribution characteristics of the gas flow field using the measured air parameter data.

[0219] (2) Equation set error term branch network: This network serves to compensate for the calculation errors of the embedded equation set module, further improving the accuracy of the model. It takes the output of the equation set module and the actual measurement data as input and obtains an error compensation term through end-to-end learning.

[0220] (3) Adaptive resolution module: This module can dynamically adjust the resolution of the neural network according to the spatiotemporal variation characteristics of the input data. This is beneficial to capturing the characteristics of the gas flow field at different scales.

[0221] (4) Multi-resolution model branch network: This network contains four sub-models with different resolutions, which can generate gas flow field distribution prediction results with different spatial resolutions. These results are integrated through an attention mechanism fusion network to output the final multi-resolution gas flow field dynamic distribution model.

[0222] In addition, in the multi-resolution model branch network, a GFSR super-resolution module is also embedded. Based on the idea of generative adversarial network, this module can reconstruct the high-resolution gas distribution details from low-resolution flow field data, providing support for fine description in complex mining environment.

[0223] In summary, the mining space gas flow field dynamic distribution model establishment method proposed in the present application fully combines mathematical analysis and deep learning two modeling ideas. On the one hand, it uses multi-source air monitoring data, and through mathematical analysis methods such as spatio-temporal feature extraction, it comprehensively describes the dynamic change characteristics of the gas flow field; on the other hand, it designs an embedded multi-branch neural network architecture, which combines physical models and data-driven models, and can effectively learn and express the actual distribution law of gas in complex mining environment.

[0224] In order to better understand and implement the present application, the following provides an embodiment 2 of a specific application scenario of the present application: a coal mining enterprise carries out gas flow field dynamic distribution monitoring and modeling in a certain mining operation area, aiming to improve the gas safety control capability. The mining area is about 2 square kilometers, with complex terrain, including underground caverns, surface open pits and other different mining methods. In order to fully understand the spatio-temporal variation characteristics of the gas flow field in the area, the enterprise has taken the following specific measures:

[0225] 1. Multi-point air data acquisition

[0226] In the mining area, the enterprise has arranged 50 acquisition points, which are evenly distributed in the underground and surface. Each acquisition point is equipped with wind speed sensor, wind direction sensor, temperature sensor, humidity sensor, pressure sensor and gas concentration sensor and other monitoring equipment. These sensors can record the parameters monitored by themselves in real time, including wind speed u, wind direction θ, temperature T, humidity H, air pressure p and gas concentration c, etc. As shown in the following figure, it is a 3D scatter plot of gas concentration, temperature and humidity, which can reveal the potential relationship between the three parameters, and help to understand the influence of environmental factors on gas concentration. Figure 6

[0227] In order to fully reflect the dynamic change characteristics of the gas flow field, the enterprise sets a 10-minute acquisition time interval Δt, that is, every 10 minutes, the data acquisition and recording of each acquisition point are carried out. After 3 months of continuous monitoring, the enterprise has accumulated 12,960 time points (3 months 30 days 24 hours * 6 intervals) of air parameter data in the mining area.

[0228] 2. Air data preprocessing

[0229] For the obtained original monitoring data, first of all, it is necessary to carry out pretreatment to eliminate the differences in dimension and numerical range between different parameters. The specific steps are as follows: ​

[0230] (1) Constructing the air matrix: For a certain collection time t j , an air matrix A(t j ) is constructed using the monitoring data of each collection point at that time, with a size of 50x6:

[0231]

[0232] (2) Interpolation and supplementation: Due to the insufficient density of the collection points, a bilinear interpolation method is used to supplement the air matrix, obtaining a more dense air dense matrix B(t j ) with a size of 200x200.

[0233] (3) Time sequence matrix construction: The air dense matrices B(t j ) at all times are stacked in time sequence to form a 3D gas flow field time sequence matrix C with a size of 12,960x200x200.

[0234] (4) Data normalization: Each element of the time sequence matrix C is subjected to Z-score normalization processing to eliminate the dimensional differences between different parameters, obtaining the normalized air time sequence matrix D.

[0235] Through the above preprocessing steps, the enterprise obtains a gas flow field dynamic data containing 12,960 time points and a spatial resolution of 200x200, laying a foundation for subsequent feature analysis and model establishment.

[0236] 3. Feature extraction of gas flow field

[0237] For the normalized time sequence matrix D after preprocessing, the enterprise uses a multivariate mixed decomposition algorithm to decompose it, aiming to extract the stable and fluctuating features of the gas flow field. The specific steps are as follows:

[0238] (1) Multivariate mixed decomposition: The 3D matrix D is decomposed into two 2D matrices E and F, and a diagonal matrix G, i.e. D = G x E x F T . Among them, E reflects the stable time characteristics of the gas flow field, F reflects the stable spatial characteristics of the gas flow field, and G contains the singular values of these characteristics.

[0239] (2) Fast Fourier transform: Fast Fourier transform (FFT) is performed on E and F respectively to obtain their feature matrices G and H in the frequency domain. At the same time, the feature matrices i and J of E and F in the time domain are also retained.

[0240] Through the above feature extraction steps, the enterprise obtains four feature matrices: stable frequency domain feature G, stable time domain feature i, fluctuation frequency domain feature H, and fluctuation time domain feature J. These features comprehensively depict the dynamic change law of the gas flow field in the mining area in the time and space dimensions, including long-term trends and short-term fluctuation characteristics, providing important input for subsequent neural network modeling.

[0241] 4. Neural network model training

[0242] Based on the aforementioned extracted dynamic features of the gas flow field, the enterprise designs an embedded multi-branch pyramid neural network structure for training and predicting the gas flow field distribution in the mining area. The network mainly consists of the following modules:

[0243] (1) Embedded equation set module: According to the physical law of the gas flow field, this module constructs the following partial differential equations describing the basic distribution characteristics of the gas:

[0244] Gas diffusion equation:

[0245]

[0246] Gas flow velocity field equation:

[0247]

[0248] Gas pressure field equation:

[0249]

[0250] Gas distribution equation:

[0251]

[0252] This module uses the collected measured data, such as u, θ, T, H, p, and c, to calculate these physical equations and obtain the preliminary distribution characteristics of the gas flow field.

[0253] (2) Equation set error term branch network: This branch network takes the output of the embedded equation set module and the actual measurement data as input, and through end-to-end learning, outputs a term for compensating the calculation error of the equation set.

[0254] (3) Adaptive resolution module: This module can automatically determine the optimal resolution of the neural network according to the spatiotemporal variation characteristics of the input data. The specific method is to use an improved ResNet network structure, replace the original convolution module with a dilated convolution, and delete the fully connected layer.

[0255] (4) Multi-resolution model branch network: Here, four sub-model networks with different resolutions are designed:

[0256] First resolution model: U-Net network structure is adopted, and the output is a gas flow field distribution of 200x200;

[0257] Second resolution model: GFSR module is connected to the simplified U-Net, and the output is a flow field distribution of 400x400;

[0258] Third resolution model: ResNet-18 network is adopted, and two GFSR modules are connected, and the output is a flow field distribution of 800x800;

[0259] Fourth resolution model: DenseNet-121 is used, and three GFSR modules are connected, and the output is a flow field distribution of 1600x1600.

[0260] (5) Fusion and aggregation network: The function of this network is to integrate the gas flow field distribution results of different resolutions, and output the final multi-resolution gas flow field dynamic model.

[0261] In the training process, the enterprise first initializes the network parameters of each module, then randomly extracts training samples from the aforementioned feature matrices G, I, H and J, and inputs them into the neural network for end-to-end supervised learning. Through repeated iteration optimization, a deep learning model capable of accurately predicting the gas flow field dynamic distribution of the mining area is finally obtained.

[0262] 5. Model application effect

[0263] The enterprise applies the trained neural network model to the actual production of the mining area, and establishes it, which has achieved remarkable results:

[0264] (1) Fine description of gas distribution: The model can output multi-resolution gas flow field distribution results, from 200x200 to 1600x1600, to meet the fine description needs in different application scenarios. For example, for underground cavern areas, a higher resolution (such as 800x800) model output can be used to fully reflect the distribution characteristics of gas in complex terrain; for surface open-pit areas, a lower resolution (such as 400x400) model can be used to balance the needs of calculation efficiency and accuracy. As shown in the following figure, it is a comparison chart of multi-resolution gas distribution, which shows the gas distribution under different resolutions from 200x200 to 1600x1600. Figure 2

[0265] (2) Prediction of gas dynamic changes: The model not only can describe the gas distribution at a certain time, but also can predict the dynamic changes of gas flow field in the future (such as 1 hour later) based on historical data. This provides more timely safety warning for mining personnel, and gains valuable reaction time for adjusting operation scheme and taking emergency measures.​Figure 3 Fig. 6 shows a gas concentration dynamic change prediction chart, which displays the actual value and predicted value of gas concentration within 24 hours, as well as the confidence interval of the prediction

[0266] (3) Auxiliary decision support: The output of the model includes not only the gas concentration distribution, but also multiple important parameters such as wind speed and wind direction. This is conducive to the mining management personnel to comprehensively analyze the formation mechanism of the gas flow field, scientifically formulate the ventilation system reconstruction scheme, and improve the pertinence and effectiveness of gas control. For example Figure 4 Fig. 7 shows a multi-parameter gas flow field analysis chart, which simultaneously displays the changes of gas concentration, wind speed and wind direction over time.

[0267] (4) Safety risk reduction: Through the accurate prediction and analysis of the dynamic distribution of the gas flow field, the enterprise can discover the possible high-gas accumulation area in advance, take targeted monitoring and prevention and control measures, and effectively reduce the risk of gas accidents. In the three-month application practice, no gas accidents occurred in the mining area, and the safety production level was significantly improved. For example Figure 5 Fig. 8 shows a safety risk assessment heat map, which displays the safety risk distribution of the mining area and marks the high-risk area; at the same time, Figure 5 Fig. 9 is a detailed view of Figure 2

[0268] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.​

Claims

1. A method for establishing a dynamic distribution model of a gas flow field in a mining space, characterized in that, The method comprises the following steps: S10, collecting air data of multiple collection points in the mining space every preset time interval, including wind speed, wind direction, temperature, humidity, air pressure and gas concentration; S20, for each collection time, establishing an air matrix using the air data of each collection point, and supplementing the space between the collection points by interpolation to convert the air matrix into an air dense matrix; S30, merging the air dense matrix of each collection time into an air time sequence matrix according to the time sequence, and normalizing each element of the air time sequence matrix to obtain an air normalized time sequence matrix; S40, decomposing the air normalized time sequence matrix by multivariate mixed decomposition calculation to obtain a stable time sequence matrix and a fluctuation time sequence matrix; S50, respectively performing fast Fourier transform on the stable time sequence matrix and the fluctuation time sequence matrix, and retaining the original time domain information to obtain a stable frequency domain feature matrix, a stable time domain feature matrix, a fluctuation frequency domain feature matrix and a fluctuation time domain feature matrix; S60, constructing a multi-layer tensor network structure, and taking the stable frequency domain feature matrix, the stable time domain feature matrix, the fluctuation frequency domain feature matrix and the fluctuation time domain feature matrix as input tensors; using a tensor decomposition method to reduce the dimension and extract features of the input tensors to obtain a low-dimensional representation of the feature tensors; S70, inputting the feature tensors into a pre-designed embedded multi-branch pyramid neural network for training to obtain a gas flow field dynamic distribution model of the mining space.

2. The method according to claim 1, wherein, The embedded multi-branch pyramid neural network comprises an embedded equation set module, an equation set error term branch network, an adaptive resolution module, a first resolution model branch network, a second resolution model branch network, a third resolution model branch network, a fourth resolution model branch network and a fusion and aggregation network.

3. The method according to claim 2, wherein, The embedded equation set module is used to calculate the basic distribution characteristics of the gas flow field according to the physical characteristics of the gas flow field and the air data, including a gas diffusion equation, a gas flow velocity field equation, a gas pressure field equation and a gas distribution equation.

4. The method according to claim 3, wherein, The equation set error term branch network is used to compensate the calculation error of the embedded equation set module, the input is the output of the embedded equation set module and the actual measurement data, the output is an error compensation term, and the structure is a multi-layer perception network.

5. The method according to claim 4, wherein, The adaptive resolution module is used to dynamically adjust the network resolution according to the characteristics of the input data, specifically by analyzing the spatial and temporal variation characteristics of the input data to determine the most suitable resolution level, the input is the feature tensor, the output is the resolution selection signal, and the structure is an adaptive network obtained by replacing the convolution module of the ResNet network with a hollow convolution and deleting the fully connected layer.

6. The method according to claim 5, wherein, The first resolution model branch network is used to generate a gas flow field distribution of the first resolution, the input is the feature tensor and the output of the embedded equation set module, the output is the first resolution gas flow field distribution, and the structure is a U-Net network. The second resolution model branch network is used for generating a second resolution gas flow field distribution, the input is a feature tensor and a down-sampling result of the first resolution model output, the output is a second resolution gas flow field distribution, and the structure is a simplified U-Net network followed by a GFSR module, and the GFSR module increases the second resolution to 2 times of the first resolution; The third resolution model branch network is used for generating a third resolution gas flow field distribution, the input is a feature tensor and a down-sampling result of the second resolution model output, the output is a third resolution gas flow field distribution, and the structure is a ResNet-18 network followed by two GFSR modules in cascade, and the GFSR modules increase the third resolution to 4 times of the first resolution; The fourth resolution model branch network is used for generating a fourth resolution gas flow field distribution, the input is a feature tensor and a down-sampling result of the third resolution model output, the output is a fourth resolution gas flow field distribution, and the structure is a DenseNet-121 network followed by three GFSR modules in cascade, and the GFSR modules increase the fourth resolution to 8 times of the first resolution.

7. The method according to claim 6, wherein, The fusion summary network is used for integrating different resolution gas flow field distribution results, the input is four resolution levels of gas flow field distribution results, the output is a final multi-resolution gas flow field dynamic distribution model, and the structure is an attention mechanism fusion network.

8. A computer-readable storage medium, characterized in that, The computer readable storage medium stores program instructions, and the program instructions are used for executing the mining space gas flow field dynamic distribution model establishment method in any one of claims 1-7 when running in the computer.

9. A system for establishing a dynamic distribution model of a gas flow field in a mining space, characterized in that The computer readable storage medium comprises the computer readable storage medium in claim 8. The computer readable storage medium comprises the computer readable storage medium in claim 8.

Images