A method for predicting shock-prominent complex dynamic disasters based on energy chain evolution.

Abstract

This invention discloses a method for predicting combined shock-outburst dynamic disasters based on energy chain evolution, relating to the field of coal mine underground dynamic disaster prediction technology. This invention uses automated numerical simulation to solve for the energy of combined shock-outburst dynamic disasters in roadways under different surrounding rock mechanical parameters, gas parameters, and shock load parameters, and automatically constructs energy datasets for combined shock-outburst dynamic disasters under various working conditions. This invention proposes a CNN-LSTM-Attention energy prediction model for combined shock-outburst dynamic disasters in roadways with energy and spatial feature constraints. The CNN model extracts spatial features of the roadway, the LSTM model obtains the energy time series, and the Attention mechanism automatically assigns weights for a spatiotemporal dual-region attention mechanism. The synergistic fusion of these three elements improves the accuracy of energy prediction for combined shock-outburst dynamic disasters in roadways.

Description

Technical Field

[0001] This invention relates to the field of underground dynamic disaster prediction technology in coal mines, specifically to a method for predicting combined shock-outburst dynamic disasters based on energy chain evolution. Background Technology

[0002] As coal mining in my country shifts towards deeper levels, roadways encounter complex geological environments characterized by high ground stress, high gas pressure, and intense mining disturbances. Under the influence of disturbances such as roof fractures and strong vibrations, the accumulated elastic energy of coal and rock, as well as the internal energy of gas, can be suddenly released. This can lead to dynamic phenomena at the tunneling face, such as gas ejection and coal / rock ejection—a combined dynamic disaster of impact and outburst—threatening safe mine production. Therefore, predicting such combined dynamic disasters at tunneling faces is of great significance.

[0003] The combined dynamic disaster of impact-outburst involves a complex chain evolution process of energy storage, conversion, and release. Specifically, under mining disturbance, the stress state of gas-bearing coal and rock mass changes, exceeding its yield strength under high stress. A damage zone forms near the tunneling face, releasing elastic energy. An elastic energy storage and deformation zone forms further away from the tunneling face. The damage zone leads to increased coal permeability, enhancing gas permeability and thus forming a gas accumulation zone. Under the action of adsorption-desorption stress and gas osmotic pressure, cracks in the coal and rock mass propagate rapidly. At this point, the coal and rock mass at the tunneling face reaches a critical state, and under the action of impact disturbance, a combined dynamic disaster of impact-outburst occurs, accompanied by energy release.

[0004] Predicting the energy chain evolution of combined shock-outburst dynamic disasters and taking corresponding blocking measures are crucial for effectively preventing coal and gas outbursts. Currently, common methods for predicting combined dynamic disasters include on-site accident analysis, physical similarity simulation experiments, and numerical simulation techniques; however, each method has its limitations. On-site accident analysis is lagging and cannot accurately and comprehensively obtain disaster-causing parameters for prediction; physical similarity experiments suffer from scale effects, long cycles, and high costs; while numerical simulation techniques offer advantages in repeatability, timeliness, and accuracy, traditional data simulation methods, when predicting combined shock-outburst dynamic disasters at tunneling faces, suffer from a single calculation and analysis model, only simulating the influence of a single variable, failing to reveal the coupling laws of lithological parameters, gas parameters, and shock load parameters, and lacking sufficient intelligence to achieve intelligent prediction. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a shock-protrusion complex dynamic disaster prediction method based on energy chain evolution, in order to overcome the technical bottleneck of insufficient performance in data-driven shock-protrusion complex dynamic disaster prediction.

[0006] The technical solution of this invention is as follows:

[0007] On the one hand, this invention provides a method for predicting shock-outburst combined dynamic disasters based on energy chain evolution, comprising the following steps:

[0008] Construct a multi-factor roadway impact-prominent complex dynamic disaster energy dataset;

[0009] The energy dataset of multi-factor roadway impact-outburst composite dynamic disaster is preprocessed and divided into training set, validation set and test set according to a set ratio;

[0010] A CNN-LSTM-Attention model for predicting the combined dynamic disaster of roadway impact-outburst is constructed by fusing spatial attributes, gas attributes, and impact attributes. This model is used to obtain the predicted energy value and the predicted spatial coordinates of the zone cell based on the input target time, basic mechanical parameters of the roadway surrounding rock, gas parameters, impact load parameters, and spatial coordinates of the zone cell.

[0011] The training set and validation set were used to train the CNN-LSTM-Attention model for predicting the combined dynamic disaster of roadway impact-outburst, which integrates spatial attributes, gas attributes and impact attributes. The trained model was obtained.

[0012] The test set is input into the trained CNN-LSTM-Attention tunnel impact-outburst composite dynamic disaster energy prediction model, which integrates spatial attributes, gas attributes, and impact attributes, to make predictions and obtain the predicted energy value and the predicted spatial coordinates of the zone cell.

[0013] Furthermore, the construction of the multi-factor roadway impact-outburst composite dynamic disaster energy dataset specifically includes the following steps:

[0014] A1: Defines the parameter space for multiphysics parameters;

[0015] The multiphysics parameters include the mechanical parameters of the surrounding rock foundation of the tunnel, gas parameters, and impact load parameters.

[0016] The parameter space represents the range of values ​​for multiphysics parameters;

[0017] The basic mechanical parameters of the surrounding rock of the tunnel include Young's modulus, cohesion, and tensile strength.

[0018] The gas parameters include gas pressure, coal permeability, and coal porosity;

[0019] The impact load parameters include impact intensity, impact frequency, and impact time;

[0020] A2: Set up a main loop function to iterate through the combination of multi-physics parameters, namely Young's modulus, cohesion, tensile strength, gas pressure, coal permeability, coal porosity, impact strength, impact frequency and impact time in sequence.

[0021] A3: Based on the geological conditions, a grid model of the tunnel is established, and the multi-physics parameter combination is traversed based on the constructed main loop function to perform static solution on the grid model of the tunnel and obtain the equilibrium result.

[0022] A4: Based on the equilibrium results, clear the displacement of all nodes in the mesh model of the tunnel to zero, delete the tunnel cross-section area in the mesh model of the tunnel, set a permeable boundary on the surface of the tunnel, and solve the tunnel excavation problem by traversing the combination of multi-physics parameters based on the constructed main loop function to obtain the tunnel excavation results.

[0023] A5: Clear the displacement of all nodes in the mesh model of the tunnel to zero, call the tunnel excavation results to perform dynamic calculations, and obtain the dynamic calculation results;

[0024] A6: Extract the impact-prominent energy data from the dynamic calculation results and construct a multi-factor roadway impact-prominent composite dynamic disaster energy dataset.

[0025] Furthermore, A3 specifically includes the following steps:

[0026] A3.1: Establish the geometric model of the tunnel and perform mesh generation to obtain the mesh model of the tunnel;

[0027] A3.2: Install large deformation switches and seepage switches;

[0028] A3.3: Set the mechanical constitutive model to the Mohr-Coulomb constitutive model;

[0029] A3.4: Set the fluid constitutive model to an isotropic permeation constitutive model;

[0030] A3.5: Set the initial parameters for the mechanical constitutive model and the fluid constitutive model, including density, Young's modulus, Poisson's ratio, cohesion, friction angle, tensile strength, coal permeability, and coal porosity;

[0031] A3.6: Set the gravitational acceleration, set the normal velocity constraint around the mesh model of the tunnel, set the bottom as a fixed constraint, and apply a set amount of pressure to the overlying rock layer to simulate the pressure of the overlying rock layer;

[0032] A3.7: Set up the geostress environment, gas pressure and gravity field, perform initial stress equilibrium solution, and obtain the equilibrium result.

[0033] Furthermore, the A4 specifically includes the following steps:

[0034] A4.1: Set the function for weakening coal body mechanical parameters by gas pressure;

[0035] Specifically, in FLAC 3D In this paper, the gas pressure weakening coal body mechanical parameter function Fish operatorgas(Zone) is defined to dynamically update the bulk modulus of each Zone element in the roadway mesh model. This function first obtains the gas pressure of the current Zone element using Zone.pp() and the bulk modulus using Zone.prop(). Then, based on the functional relationship between gas pressure and bulk modulus obtained from pre-conducted experimental tests, the gas pressure weakening coal body mechanical damage equation is written: b new = b - pp gas / 1000, where b new Let b be the new bulk modulus, and pp be the initial bulk modulus. gas Given the gas pressure, the new bulk modulus b is recalculated based on the gas pressure coal body mechanical damage equation. new The new bulk modulus is assigned to the current Zone element using Zone.prop(). In addition, the gas pressure weakened coal body mechanical parameter function Fish operator gas(Zone) is set as a callback function by the Fish callback add command, so that it is executed once for all Zone elements in each simulation step.

[0036] A4.2: Enable fluid-structure interaction and perform roadway excavation calculations based on the weakened coal body mechanical parameter function under gas pressure to obtain roadway excavation results.

[0037] Furthermore, A5 specifically includes the following steps:

[0038] A5.1: Write a dynamic load impact function to simulate impact loads, including the impact intensity, impact frequency and impact time of the impact load;

[0039] In FLAC 3D In the code, two Fish functions, Setup and Wave, are defined to generate a dynamic waveform signal that changes over time as an impact load. The Setup function initializes the frequency and angular frequency of the waveform signal and records the initial time. The Wave function dynamically calculates the waveform value based on the current simulation time. When the simulation time is less than the set value, a waveform signal based on a cosine function is generated. When the simulation time exceeds the set value, the signal stops.

[0040] A5.2: Remove the normal velocity constraint and the bottom fixed constraint of the mesh model of the tunnel, and set viscous boundary conditions around the mesh model of the tunnel.

[0041] A5.3: Set the impact load application location, set the Rayleigh damping and center frequency of the roadway mesh model, set the dynamic load calculation time, and perform dynamic calculation based on the dynamic load impact function to obtain the dynamic calculation results.

[0042] Further, A6 specifically involves: setting a total simulation time period and dividing it into n interval time periods, with the start time of each interval time period called the simulation time and the end time called the target time. Simulation is performed at each simulation time, and impact-outburst energy data is extracted from the dynamic calculation results and the current target time is printed. This process is repeated n times.

[0043] The shock-outburst energy data is a time series, including the spatial coordinates of Zone cells at different target times. The basic mechanical parameters, gas parameters, impact load parameters, and energy of the surrounding rock of the tunnel.

[0044] Furthermore, the CNN-LSTM-Attention roadway impact-outburst composite dynamic disaster energy prediction model with spatial attribute-gas attribute-impact attribute fusion constrained by energy and spatial characteristics includes a CNN layer, an LSTM network, an attention mechanism module, and a fully connected layer;

[0045] The CNN layer is used to extract features from the input features of each Zone unit to obtain spatial features;

[0046] The LSTM network is used to extract energy temporal features based on the input spatial features;

[0047] The input to the attention mechanism module is spatial features and energy temporal features, which are used to calculate the attention mechanism weights for spatiotemporal dual-region fusion and to use them to weight the spatial features and energy temporal features to obtain the fusion vector.

[0048] The fully connected layer is used to map the fusion vector to obtain the predicted energy value and the predicted spatial coordinates of the Zone cell.

[0049] Furthermore, the spatial features are represented as follows:

[0050] ;

[0051] In the formula, Spatial coordinates are Spatial characteristics of Zone units, For activation function, , , The kernel size is the convolution kernel size. For bias terms; , and These are indexes in three dimensions. These are the weights of the 3D convolution kernel; Spatial coordinates are The input features of the Zone element include the spatial coordinates of the Zone element, the mechanical parameters of the roadway surrounding rock foundation, the gas parameters, and the impact load parameters.

[0052] Furthermore, the calculation method for the attention mechanism weights in the spatiotemporal dual-region fusion is as follows:

[0053] First, energy gradient change is introduced as an external feature, and the weights of the time series attention mechanism are calculated:

[0054] ;

[0055] In the formula, For the weights of the time series attention mechanism, For the softmax function, The query matrix is ​​obtained by transforming energy time series features. The key matrix is ​​obtained by energy time series feature transformation. The energy gradient variation matrix is... For learning parameters, For transpose;

[0056] Then calculate the spatial attention mechanism weights for the Zone units:

[0057] ;

[0058] In the formula, For the unit space attention mechanism weights, For the softmax function, The query matrix is ​​obtained by transforming spatial features; The key matrix obtained through transformation is used as input for spatial features. For learning parameters;

[0059] Next, based on the temporal series attention mechanism weights and the Zone unit spatial attention mechanism weights, the attention mechanism weights for spatiotemporal dual-region fusion are calculated. The calculation method is as follows:

[0060] ;

[0061] In the formula, Weights for the attention mechanism in spatiotemporal dual-region fusion; Assign coefficients to time and space weights.

[0062] Secondly, this application proposes an electronic device, comprising: one or more processors, and a memory for storing instructions, which, when executed by the one or more processors, cause the one or more processors to execute the shock-protrusion complex dynamic disaster prediction method based on energy chain evolution.

[0063] Thirdly, this application proposes a computer-readable storage medium storing executable instructions that, when executed, cause a processor to perform the shock-protrusion complex dynamic disaster prediction method based on energy chain evolution.

[0064] Fourthly, this application proposes a computer program product, including a computer program or instructions that, when executed by a processor, implement the aforementioned shock-protrusion complex dynamic disaster prediction method based on energy chain evolution.

[0065] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0066] This invention uses Python and FLAC 3D By combining software, automated numerical simulations were used to solve the energy of combined dynamic disasters of impact-outburst in roadways under different surrounding rock mechanical parameters, gas parameters, and impact load parameters. Data sets of combined dynamic disaster energy of impact-outburst under various working conditions were automatically constructed.

[0067] Addressing the issue of the coupled correlation between spatial attributes, temporal series, and energy characteristics in energy data related to multi-condition roadway impact-outburst combined dynamic disasters, this invention proposes a CNN-LSTM-Attention energy prediction model for roadway impact-outburst combined dynamic disasters with energy and spatial feature constraints. The CNN model extracts roadway spatial features, the LSTM model acquires energy temporal series data, and the Attention mechanism automatically assigns weights based on a spatiotemporal dual-region attention mechanism. This synergistic fusion of the three components improves the accuracy of energy prediction for roadway impact-outburst combined dynamic disasters. Attached Figure Description

[0068] Figure 1 This is a flowchart of a shock-outburst combined dynamic disaster prediction method based on energy chain evolution in an embodiment of the present invention;

[0069] Figure 2 This is a diagram showing the initial equilibrium simulation stress results in an embodiment of the present invention;

[0070] Figure 3This is a diagram showing the updated mechanical parameters of the surrounding rock weakened by gas pressure in an embodiment of the present invention.

[0071] Figure 4 This is a cloud map showing the gas pressure distribution after fluid-structure interaction excavation in an embodiment of the present invention.

[0072] Figure 5 This is a diagram showing the training results of the CNN-LSTM-Attention model in an embodiment of the present invention;

[0073] Among them, (a) is a curve comparison of the predicted energy value, the actual energy value and the sample sequence; (b) is a density map of the predicted energy value and the actual energy value.

[0074] Figure 6 This is a cloud map showing the energy evolution of a combined shock-outburst dynamic disaster at a time of 0.02s in an embodiment of the present invention.

[0075] Figure 7 This is a cloud map showing the energy evolution of a combined shock-outburst dynamic disaster at a time of 0.04s in an embodiment of the present invention.

[0076] Figure 8 This is a cloud map showing the energy evolution of a combined shock-outburst dynamic disaster at a time of 0.06s in an embodiment of the present invention.

[0077] Figure 9 This is a cloud map showing the energy evolution of a combined shock-outburst dynamic disaster at a time of 0.08s in an embodiment of the present invention.

[0078] Figure 10 This is a cloud map showing the energy evolution of a combined shock-outburst dynamic disaster at a time of 0.1s in an embodiment of the present invention. Detailed Implementation

[0079] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0080] Example 1:

[0081] This embodiment takes the geological conditions of a mine roadway as an example and uses the shock-outburst composite dynamic disaster prediction method based on energy chain evolution proposed in this invention to realize the prediction of shock-outburst composite dynamic disasters in a mine roadway, providing technical guidance for the subsequent prevention and control of shock-outburst composite dynamic disasters.

[0082] In this embodiment, FLAC is implemented using Python. 3D Numerical simulation of combined dynamic disasters of roadway impact and outburst under multi-factor conditions using software. Automated implementation using FLAC. 3DSoftware statics, fluid-structure interaction, and dynamics simulations are used to automatically construct a numerical simulation sample set for tunnel impact-outburst energy. Subsequently, a CNN-LSTM-Attention neural network model is built to predict the combined dynamic energy of tunnel impact-outburst disasters. Figure 1 As shown, the specific steps include:

[0083] Step 1: Construct a multi-factor roadway impact-prominent composite dynamic disaster energy dataset;

[0084] Step 1.1: Determine the influencing factors of impact-prominent composite dynamic disasters and define the parameter space of multiphysics parameters;

[0085] The multiphysics parameters include the mechanical parameters of the surrounding rock foundation of the roadway, gas parameters, and impact load parameters; the parameter space represents the range of values ​​for the multiphysics parameters.

[0086] The basic mechanical parameters of the surrounding rock of the tunnel include Young's modulus, cohesion, and tensile strength.

[0087] The gas parameters include gas pressure, coal permeability, and coal porosity;

[0088] The impact load parameters include impact intensity, impact frequency, and impact time;

[0089] The parameter space list of cohesion in the mechanical parameters of the roadway surrounding rock foundation set in this embodiment is: cohesions = [2.4e6, 2.5e6, 2.6e6];

[0090] Set the parameter space list for gas pressure in the gas parameters as Pore_pressures = [27e4, 30e4, 33e4];

[0091] Set the parameter space list for impact intensity in the impact load parameters as Wave_amplitudes = [1e4, 1.5e4, 2e4];

[0092] Step 1.2: Set up a main loop function to iterate through the combination of multi-physics parameters, namely Young's modulus, cohesion, tensile strength, gas pressure, coal permeability, coal porosity, impact strength, impact frequency and impact time in sequence;

[0093] In this embodiment, the Python programming language is used to construct a three-level nested loop structure. The outer loop iterates through the parameter space list of cohesion, the middle loop iterates through the parameter space list of gas pressure, and the inner loop iterates through the parameter space list of impact intensity, ensuring that the program can automatically substitute each combination of multiphysics parameters in sequence for subsequent calculations.

[0094] Step 1.3: Based on the geological conditions, establish a grid model of the tunnel and, based on the constructed main loop function, traverse the combination of multi-physics parameters to perform static solution on the grid model of the tunnel and obtain the equilibrium result;

[0095] Step 1.3.1: Establish the geometric model of the tunnel and perform mesh generation to obtain the mesh model of the tunnel;

[0096] In this embodiment, a rectangular grid with dimensions of 50m × 20m × 50m is established as the grid model of the tunnel;

[0097] Step 1.3.2: Set the large deformation switch and seepage switch (to open or close);

[0098] Step 1.3.3: Set the mechanical constitutive model to the Mohr-Coulomb constitutive model and assign parameters as shown in Table 1;

[0099] Table 1. Parameters used in the Mohr-Coulomb constitutive model for mechanical constitutive design;

[0100]

[0101] Step 1.3.4: Set the fluid constitutive model to an isotropic permeable constitutive model, and assign parameters as shown in Table 2;

[0102] Table 2. Parameters used in the fluid constitutive isotropic constitutive model;

[0103]

[0104] Step 1.3.5: Set the initial parameters for the mechanical constitutive model and the fluid constitutive model, including density, Young's modulus, Poisson's ratio, cohesion, friction angle, tensile strength, coal permeability, and coal porosity;

[0105] Step 1.3.6: Set the gravitational acceleration, set the normal velocity constraint around the mesh model of the tunnel, set the bottom as a fixed constraint, and apply a pressure of 12.5MPa to the overlying rock layer to simulate the pressure of the overlying rock layer;

[0106] Step 1.3.7: Set up the geostress environment, gas pressure, and gravity field, and perform initial stress equilibrium solution. The equilibrium results are as follows: Figure 2 As shown;

[0107] Step 1.4: Based on the equilibrium results, clear the displacement of all nodes in the mesh model of the tunnel to zero, delete the tunnel cross-section area in the mesh model of the tunnel, set a permeable boundary on the surface of the tunnel, and solve the tunnel excavation problem by traversing the combination of multi-physics parameters based on the constructed main loop function to obtain the tunnel excavation results.

[0108] Step 1.4.1: Set the function for weakening coal body mechanical parameters by gas pressure;

[0109] Specifically, in FLAC 3D In this simulation, the gas pressure weakened coal mechanics parameter function `Fish operatorgas(Zone)` is defined to dynamically update the bulk modulus of each Zone element in the roadway's mesh model. This function first obtains the gas pressure of the current Zone element using `Zone.pp()` and the bulk modulus using `Zone.prop()`. Then, based on the functional relationship between gas pressure and bulk modulus obtained from pre-conducted experimental tests, a gas pressure coal mechanics damage equation is written: `newb = b - gaspp / 1000`, where `newb` is the new bulk modulus, `b` is the initial bulk modulus, and `Gaspp` is the gas pressure. The new bulk modulus `newb` is recalculated according to the gas pressure coal mechanics damage equation and assigned to the current Zone element using `Zone.prop()`. Furthermore, using the `Fishcallback add` command, the gas pressure weakened coal mechanics parameter function `Fish operator gas(Zone)` is set as a callback function, executing once for all Zone elements in each simulation step, thus achieving the dynamic update function of the bulk modulus. The simulation results are as follows: Figure 3 As shown.

[0110] Step 1.4.2: Enable fluid-structure interaction and perform roadway excavation calculations based on the weakened coal body mechanical parameter function caused by gas pressure. The roadway excavation results are as follows: Figure 4 As shown;

[0111] Step 1.5: Clear the displacement of all nodes in the mesh model of the tunnel to zero, call the tunnel excavation results to perform dynamic calculations, and obtain the dynamic calculation results;

[0112] Step 1.5.1: Write a dynamic load impact function to simulate the impact load, including the impact intensity, impact frequency and impact time of the impact load;

[0113] In FLAC 3D In this code, two Fish functions, Setup and Wave, are defined to generate a dynamic waveform signal that changes over time as an impact load. The Setup function initializes the frequency and angular frequency of the waveform signal and records the initial time. The Wave function dynamically calculates the waveform value based on the current simulation time. When the simulation time is less than 0.02 seconds, a waveform signal based on a cosine function is generated, and the signal stops when it exceeds 0.02 seconds.

[0114] Step 1.5.2: Remove the normal velocity constraint and the bottom fixed constraint of the mesh model of the tunnel, and set viscous boundary conditions around the mesh model of the tunnel.

[0115] Step 1.5.3: Set the impact load application location, set the Rayleigh damping and center frequency of the roadway mesh model, set the dynamic load calculation time, and perform dynamic calculation based on the dynamic load impact function to obtain the dynamic calculation results;

[0116] Step 1.6: Extract the impact-outburst energy data from the dynamic calculation results and construct a multi-factor roadway impact-outburst composite dynamic disaster energy dataset;

[0117] Specifically, a total simulation time period is set and divided into n interval time periods. The start time of each interval time period is called the simulation time and the end time is called the target time. Simulation is performed at each simulation time. Impact-outburst energy data is extracted from the dynamic calculation results and the current target time is printed to ensure the visualization of the simulation progress. This process is repeated n times.

[0118] The shock-outburst energy data is a time series, including the spatial coordinates of Zone cells at different target times. The basic mechanical parameters, gas parameters, impact load parameters, and energy of the surrounding rock of the tunnel;

[0119] In this embodiment, at each target time point, the coordinates of the center position of each Zone cell in the mesh model of the tunnel are extracted. The maximum principal stress, intermediate principal stress, and minimum principal stress combination (σ1, σ2, σ3) of each zone element, the shear modulus of each zone element, the bulk modulus of each zone element, the Young's modulus and Poisson's ratio of each zone element, and then the energy of each zone element are calculated.

[0120] The methods for calculating Young's modulus and Poisson's ratio are as follows: , , nu is Poisson's ratio, bulk is bulk modulus, shear is shear modulus, E is Young's modulus;

[0121] The energy is calculated as follows: Ue represents energy, sigma1 represents the first principal stress (maximum principal stress), sigma2 represents the second principal stress (intermediate principal stress), and sigma3 represents the third principal stress (minimum principal stress).

[0122] The dataset list in this embodiment is shown below;

[0123] Table 1. List of datasets;

[0124]

[0125] Step 2: Preprocess the multi-factor roadway impact-outburst composite dynamic disaster energy dataset and divide it into training set, validation set and test set according to the set ratio;

[0126] This embodiment reads a multi-factor roadway impact-outburst composite dynamic disaster energy dataset, including target time, basic mechanical parameters of the roadway surrounding rock, gas parameters, impact load parameters, and spatial coordinates of the Zone cells. The data is preprocessed, and then standardized and normalized in sequence to achieve dimensionless distribution among various indicators, and the training set, validation set and test set are divided.

[0127] Step 3: Construct a CNN-LSTM-Attention energy prediction model for combined dynamic disasters of roadway impact-outburst, which integrates spatial attributes, gas attributes, and impact attributes. This model is used to predict the energy of such disasters based on the input target time, basic mechanical parameters of the roadway surrounding rock, gas parameters, impact load parameters, and the spatial coordinates of the Zone cells. This yields predicted energy values ​​and predicted spatial coordinates of the Zone cells;

[0128] The multi-factor roadway impact-outburst combined dynamic disaster energy dataset obtained in Step 1 possesses a coupling correlation among spatial attributes, time series, and energy characteristics. Existing prediction models suffer from insufficient extraction of spatiotemporal features, inadequate identification of the coupling relationship between energy and space, and poor model prediction capabilities. To address this, this invention proposes a CNN-LSTM-Attention roadway impact-outburst combined dynamic disaster energy prediction model with spatial attribute-gas attribute-impact attribute fusion constrained by energy and spatial characteristics.

[0129] The CNN-LSTM-Attention model for predicting the energy of combined dynamic disasters of roadway impact and outburst with spatial attribute-gas attribute-impact attribute fusion constrained by energy and spatial characteristics includes a CNN layer, an LSTM network, an attention mechanism module, and a fully connected layer.

[0130] The CNN layer is used to extract features from the input features of each Zone unit to obtain spatial features;

[0131]

[0132] In the formula, Spatial coordinates are Spatial characteristics of Zone units, For activation function, , , The kernel size is the convolution kernel size. For bias terms; , and These are indexes in three dimensions. These are the weights of the 3D convolution kernel; Spatial coordinates are The input features of the Zone element include the spatial coordinates of the Zone element, the mechanical parameters of the roadway surrounding rock foundation, the gas parameters, and the impact load parameters;

[0133] The LSTM network is used to extract energy temporal features based on the input spatial features;

[0134] In this embodiment, the number of LSTM neurons and the number of hidden layers are set to 80 to identify energy temporal features;

[0135] The input to the attention mechanism module is spatial features and energy temporal features, which are used to calculate the attention mechanism weights for spatiotemporal dual-region fusion and to use them to weight the spatial features and energy temporal features to obtain the fusion vector.

[0136] The fully connected layer is used to map the fusion vector to obtain the predicted energy value and the predicted spatial coordinates of the Zone cell;

[0137] The calculation method for the attention mechanism weights in the spatiotemporal dual-region fusion is as follows:

[0138] First, energy gradient change is introduced as an external feature, and the weights of the time series attention mechanism are calculated:

[0139]

[0140] In the formula, For the weights of the time series attention mechanism, For the softmax function, The query matrix is ​​obtained by transforming energy time series features. The key matrix is ​​obtained by energy time series feature transformation. The energy gradient variation matrix is... For learning parameters, For transpose;

[0141] Then calculate the spatial attention mechanism weights for the Zone units:

[0142]

[0143] In the formula, For the unit space attention mechanism weights, For the softmax function, The query matrix is ​​obtained by transforming spatial features; The key matrix obtained through transformation is used as input for spatial features. For learning parameters;

[0144] Next, based on the temporal series attention mechanism weights and the Zone unit spatial attention mechanism weights, the attention mechanism weights for spatiotemporal dual-region fusion are calculated. The calculation method is as follows:

[0145]

[0146] In the formula, Weights for the attention mechanism in spatiotemporal dual-region fusion; Assign coefficients to time and space weights;

[0147] In this embodiment, the head of the Attention mechanism is set to 1 and the key is set to 2, and the fusion is performed through a linear weighting method.

[0148] Step 4: Use the training set and validation set to train the CNN-LSTM-Attention roadway impact-outburst combined dynamic disaster energy prediction model that integrates spatial attributes, gas attributes and impact attributes, to obtain the trained CNN-LSTM-Attention roadway impact-outburst combined dynamic disaster energy prediction model that integrates spatial attributes, gas attributes and impact attributes.

[0149] During the model training phase, a physical constraint term is added to the weighted mean squared error to form the loss function:

[0150]

[0151] In the formula, MSE is the weighted mean square error; This is an energy constraint term, which is a constant, to avoid abnormal energy values. This is a spatial coordinate information constraint term, which is a constant to ensure the spatial smoothness of adjacent coordinates; and The weights are used to train the model, evaluate the model error, and the training results are as follows: Figure 5 As shown;

[0152] Step 5: Input the test set into the trained CNN-LSTM-Attention tunnel impact-outburst composite dynamic disaster energy prediction model that integrates spatial attributes, gas attributes, and impact attributes to make predictions and obtain the predicted energy value and the predicted spatial coordinates of the Zone cell.

[0153] In this embodiment, the dataset to be predicted is imported, including the basic mechanical parameters of the surrounding rock of the roadway, gas parameters, impact load parameters, and spatial coordinates (x, y, z) of the Zone cells. The predicted energy values ​​and the predicted spatial coordinates of the Zone cells are obtained, and the data is plotted as an energy cloud map, such as... Figures 6-10 As shown.

[0154] Example 2:

[0155] This embodiment proposes an electronic device, including: one or more processors, and a memory, wherein the memory is used to store instructions, and when the instructions are executed by the one or more processors, the one or more processors execute the shock-outburst combined dynamic disaster prediction method based on energy chain evolution.

[0156] The electronic device may be a mobile phone, computer, or tablet computer, etc., and includes a memory and a processor. The memory stores a computer program, which, when executed by the processor, implements the shock-outburst combined dynamic disaster prediction method based on energy chain evolution as described in the embodiments. It is understood that the electronic device may also include an input / output (I / O) interface and communication components.

[0157] The processor is used to execute all or part of the steps in the shock-outburst combined dynamic disaster prediction method based on energy chain evolution as described in the above embodiments. The memory is used to store various types of data, which may include, for example, instructions for any application or method in an electronic device, as well as application-related data.

[0158] The processor can be implemented as an Application Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), controller, microcontroller, microprocessor, or other electronic components, and is used to execute the shock-outburst composite dynamic disaster prediction method based on energy chain evolution described in the above embodiments.

[0159] Example 3:

[0160] This embodiment proposes a computer-readable storage medium that stores executable instructions. When these instructions are executed, if they are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium.

[0161] The computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the shock-outburst composite dynamic disaster prediction method based on energy chain evolution described in the various embodiments of this application.

[0162] The aforementioned storage media include: flash memory, hard disks, multimedia cards, card-type memory (e.g., SD (Secure Digital Memory Card) or DX (Memory Data Register, MDR) memory), random access memory (RAM), static random-access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic storage, disks, optical discs, servers, APP (Application) app stores, and other media capable of storing program verification codes. These media store computer programs, which, when executed by a processor, can implement the various steps of the aforementioned shock-protrusion composite dynamic disaster prediction method based on energy chain evolution.

[0163] Example 4:

[0164] This embodiment proposes a computer program product, including a computer program or instructions, which, when executed by a processor, implements the aforementioned shock-protrusion complex dynamic disaster prediction method based on energy chain evolution.

[0165] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope defined by the present invention.

Claims

1. A method for predicting shock-outburst composite dynamic disasters based on energy chain evolution, characterized in that, Includes the following steps: Construct a multi-factor roadway impact-prominent complex dynamic disaster energy dataset; The energy dataset of multi-factor roadway impact-outburst composite dynamic disaster is preprocessed and divided into training set, validation set and test set according to a set ratio; A CNN-LSTM-Attention model for predicting the combined dynamic disaster of roadway impact-outburst is constructed by fusing spatial attributes, gas attributes, and impact attributes. This model is used to obtain the predicted energy value and the predicted spatial coordinates of the zone cell based on the input target time, basic mechanical parameters of the roadway surrounding rock, gas parameters, impact load parameters, and spatial coordinates of the zone cell. A CNN-LSTM-Attention model for predicting the energy of a complex dynamic disaster involving roadway impact and protrusion, which integrates spatial attributes, gas attributes, and impact attributes constrained by energy and spatial characteristics, consists of a CNN layer, an LSTM network, an attention mechanism module, and a fully connected layer. The CNN layer is used to extract features from the input features of each Zone unit to obtain spatial features; The LSTM network is used to extract energy temporal features based on the input spatial features; The input to the attention mechanism module is spatial features and energy temporal features, which are used to calculate the attention mechanism weights for spatiotemporal dual-region fusion and to use them to weight the spatial features and energy temporal features to obtain the fusion vector. The fully connected layer is used to map the fusion vector to obtain the predicted energy value and the predicted spatial coordinates of the Zone cell; The spatial features are represented as follows: ； In the formula, Spatial coordinates are Spatial characteristics of Zone units, For activation function, , , The kernel size is the convolution kernel size. For bias terms; , and These are indexes in three dimensions. These are the weights of the 3D convolution kernel. Spatial coordinates are The input features of the Zone element include the spatial coordinates of the Zone element, the mechanical parameters of the roadway surrounding rock foundation, the gas parameters, and the impact load parameters; The training set and validation set were used to train the CNN-LSTM-Attention model for predicting the combined dynamic disaster of roadway impact-outburst, which integrates spatial attributes, gas attributes and impact attributes. The trained model was obtained. The test set is input into the trained CNN-LSTM-Attention tunnel impact-outburst composite dynamic disaster energy prediction model, which integrates spatial attributes, gas attributes, and impact attributes, to make predictions and obtain the predicted energy value and the predicted spatial coordinates of the zone cell.

2. The method for predicting shock-outburst combined dynamic disasters based on energy chain evolution according to claim 1, characterized in that, The construction of a multi-factor roadway impact-outburst composite dynamic disaster energy dataset specifically includes the following steps: A1: Defines the parameter space for multiphysics parameters; The multiphysics parameters include the mechanical parameters of the surrounding rock foundation of the tunnel, gas parameters, and impact load parameters. The parameter space represents the range of values ​​for multiphysics parameters; The basic mechanical parameters of the surrounding rock of the tunnel include Young's modulus, cohesion, and tensile strength. The gas parameters include gas pressure, coal permeability, and coal porosity; The impact load parameters include impact intensity, impact frequency, and impact time; A2: Set up a main loop function to iterate through the combination of multi-physics parameters, namely Young's modulus, cohesion, tensile strength, gas pressure, coal permeability, coal porosity, impact strength, impact frequency and impact time in sequence. A3: Based on the geological conditions, a grid model of the tunnel is established, and the multi-physics parameter combination is traversed based on the constructed main loop function to perform static solution on the grid model of the tunnel and obtain the equilibrium result. A4: Based on the equilibrium results, clear the displacement of all nodes in the mesh model of the tunnel to zero, delete the tunnel cross-section area in the mesh model of the tunnel, set a permeable boundary on the surface of the tunnel, and solve the tunnel excavation problem by traversing the combination of multi-physics parameters based on the constructed main loop function to obtain the tunnel excavation results. A5: Clear the displacement of all nodes in the mesh model of the tunnel to zero, call the tunnel excavation results to perform dynamic calculations, and obtain the dynamic calculation results; A6: Extract the impact-prominent energy data from the dynamic calculation results and construct a multi-factor roadway impact-prominent composite dynamic disaster energy dataset.

3. The method for predicting shock-outburst combined dynamic disasters based on energy chain evolution according to claim 2, characterized in that, The A3 specifically includes the following steps: A3.1: Establish the geometric model of the tunnel and perform mesh generation to obtain the mesh model of the tunnel; A3.2: Install large deformation switches and seepage switches; A3.3: Set the mechanical constitutive model to the Mohr-Coulomb constitutive model; A3.4: Set the fluid constitutive model to an isotropic permeation constitutive model; A3.5: Set the initial parameters for the mechanical constitutive model and the fluid constitutive model, including density, Young's modulus, Poisson's ratio, cohesion, friction angle, tensile strength, coal permeability, and coal porosity; A3.6: Set the gravitational acceleration, set the normal velocity constraint around the mesh model of the tunnel, set the bottom as a fixed constraint, and apply a set amount of pressure to the overlying rock layer to simulate the pressure of the overlying rock layer; A3.7: Set up the geostress environment, gas pressure and gravity field, perform initial stress equilibrium solution and obtain equilibrium results.

4. The method for predicting shock-outburst combined dynamic disasters based on energy chain evolution according to claim 2, characterized in that, The A4 specifically includes the following steps: A4.1: Set the function for weakening coal body mechanical parameters by gas pressure; Specifically, in FLAC 3D In this paper, the gas pressure weakening coal body mechanical parameter function Fish operator gas(Zone) is defined to dynamically update the bulk modulus of each Zone element in the roadway mesh model. This function first obtains the gas pressure of the current Zone element using Zone.pp() and the bulk modulus using Zone.prop(). Then, based on the functional relationship between gas pressure and bulk modulus obtained from pre-conducted experimental tests, the gas pressure coal body mechanical damage equation is written: b new = b - pp gas / 1000, where b new Let b be the new bulk modulus, and pp be the initial bulk modulus. gas Given the gas pressure, the new bulk modulus b is recalculated based on the gas pressure coal body mechanical damage equation. new The new bulk modulus is assigned to the current Zone element using Zone.prop(). In addition, the gas pressure weakened coal body mechanical parameter function Fish operator gas(Zone) is set as a callback function by the Fish callback add command, so that it is executed once for all Zone elements in each simulation step. A4.2: Enable fluid-structure interaction and perform roadway excavation calculations based on the weakened coal body mechanical parameter function under gas pressure to obtain roadway excavation results.

5. The method for predicting shock-outburst combined dynamic disasters based on energy chain evolution according to claim 2, characterized in that, The A5 specifically includes the following steps: A5.1: Write a dynamic load impact function to simulate impact loads, including the impact intensity, impact frequency and impact time of the impact load; In FLAC 3D In the code, two Fish functions, Setup and Wave, are defined to generate a dynamic waveform signal that changes over time as an impact load. The Setup function initializes the frequency and angular frequency of the waveform signal and records the initial time. The Wave function dynamically calculates the waveform value based on the current simulation time. When the simulation time is less than the set value, a waveform signal based on the cosine function is generated. When the simulation time exceeds the set value, the signal stops. A5.2: Remove the normal velocity constraint and the bottom fixed constraint of the mesh model of the tunnel, and set viscous boundary conditions around the mesh model of the tunnel. A5.3: Set the impact load application location, set the Rayleigh damping and center frequency of the roadway mesh model, set the dynamic load calculation time, and perform dynamic calculation based on the dynamic load impact function to obtain the dynamic calculation results.

6. The method for predicting shock-outburst combined dynamic disasters based on energy chain evolution according to claim 2, characterized in that, Specifically, A6 involves setting a total simulation time period and dividing it into n interval time periods. The start time of each interval time period is called the simulation time, and the end time is called the target time. Simulation is performed at each simulation time, and the impact-outburst energy data is extracted from the dynamic calculation results and the current target time is printed. This process is repeated n times. The shock-outburst energy data is a time series, including the spatial coordinates of Zone cells at different target times. The basic mechanical parameters, gas parameters, impact load parameters, and energy of the surrounding rock of the tunnel.

7. The method for predicting shock-outburst combined dynamic disasters based on energy chain evolution according to claim 1, characterized in that, The calculation method for the attention mechanism weights in the spatiotemporal dual-region fusion is as follows: First, energy gradient change is introduced as an external feature, and the weights of the time series attention mechanism are calculated: ； In the formula, For the weights of the time series attention mechanism, For the softmax function, The query matrix is ​​obtained by transforming energy time series features. The key matrix is ​​obtained by energy time series feature transformation. The energy gradient variation matrix is... For learning parameters, For transpose; Then calculate the spatial attention mechanism weights for the Zone units: ； In the formula, For the unit space attention mechanism weights, For the softmax function, The query matrix is ​​obtained by transforming spatial features; The key matrix obtained through transformation is used as input for spatial features. For learning parameters; Next, based on the temporal series attention mechanism weights and the Zone unit spatial attention mechanism weights, the attention mechanism weights for spatiotemporal dual-region fusion are calculated. The calculation method is as follows: ； In the formula, Weights for the attention mechanism in spatiotemporal dual-region fusion; Assign coefficients to time and space weights.

8. A computer program product, characterized in that, Includes a computer program or instructions that, when executed by a processor, implement the shock-protrusion complex dynamic disaster prediction method based on energy chain evolution as described in any one of claims 1-7.

Images