Method and device for predicting deformation of surrounding rock of vertical shaft based on fluid-solid coupling simulation

Abstract

The present invention discloses a method and device for predicting the deformation of vertical shaft surrounding rock using fluid-solid coupling linkage simulation. The method comprises the following steps: Step 1: Using water content changes to describe the effect of rainwater infiltration on surrounding rock deformation, constructing a seepage-surrounding rock deformation model, and constructing a scouring-surrounding rock deformation model, and then establishing a vertical shaft excavation model under the action of rainwater; Step 2: Performing multi-scale linkage simulation on the seepage-surrounding rock deformation model, the scouring-surrounding rock deformation model, and the vertical shaft excavation model to obtain surrounding rock deformation data; Step 3: Using the surrounding rock deformation data as a training set to train a TCN-LSTM neural network model to obtain a vertical shaft surrounding rock deformation prediction model under the action of rainwater; Step 4: Acquiring real-time surrounding rock deformation data, inputting it into the vertical shaft surrounding rock deformation prediction model, and outputting a prediction result. The present invention can improve the accuracy, efficiency, and applicability of vertical shaft surrounding rock deformation prediction, while reducing the implementation complexity and computational complexity.

Description

Technical Field

[0001] The present invention relates to the technical field of vertical shaft construction engineering, and in particular to a method and device for predicting deformation of vertical shaft surrounding rocks based on fluid-solid coupling linkage simulation. Background Art

[0002] Surrounding rock deformation refers to the displacement, deformation, and stress redistribution of the rock mass surrounding an engineering structure caused by excavation and mining, typically in the form of expansion, compression, shearing, and bending. Monitoring and analyzing surrounding rock deformation can promptly identify potential geological hazards, guide the optimization of construction plans, and improve the durability and reliability of engineering structures, thereby ensuring the smooth progress of engineering projects and minimizing economic losses.

[0003] During shaft excavation and construction, the deformation and stability of the surrounding rock are key indicators. Existing technologies for predicting shaft surrounding rock deformation rely heavily on empirical formulas and simplified models. These formulas primarily rely on engineers' experience and historical data, and are used to predict surrounding rock deformation. However, these formulas are typically based on specific geological conditions and engineering environments, and are only applicable to specific conditions. They are not applicable to all geological environments and have certain limitations and scopes of application. When geological conditions or the construction environment change, their applicability and prediction accuracy decrease significantly. Furthermore, empirical formulas fail to account for complex geological structures, groundwater influences, and unforeseen changes during construction, resulting in low actual prediction accuracy.

[0004] Among structural analysis methods, the finite element method (FEM) can handle continuous media for structural analysis. However, the FEM requires the establishment of a sophisticated numerical model, resulting in high computational complexity and a complex solution process. Furthermore, it is poorly adaptable to complex nonlinear and fracture problems. Therefore, it is primarily applicable to homogeneous, isotropic continuous media and is ineffective for heterogeneous and anisotropic materials. Deformation prediction for granular materials and massive rock, for example, falls under the category of non-continuous media, for which the FEM is not well suited, let alone complex geological conditions such as large deformation, fracture, and sliding. Furthermore, these existing prediction methods lack the ability to process and dynamically update field monitoring data in real time, making it difficult to promptly reflect the actual deformation of the surrounding rock mass. Summary of the Invention

[0005] The purpose of the present invention is to address the problem of shaft surrounding rock deformation prediction and propose a shaft surrounding rock deformation prediction method and device based on fluid-solid coupling linkage simulation. By constructing a permeation-surrounding rock deformation model, a scouring-surrounding rock deformation model, and a shaft excavation model under rainwater action to perform multi-scale linkage simulation, a large amount of actual surrounding rock deformation data is generated and then the TCN-LSTM neural network model is trained. This can efficiently obtain an accurate shaft surrounding rock deformation prediction model, improve the accuracy, efficiency, and applicability of shaft surrounding rock deformation prediction, and reduce the implementation complexity and computational complexity. In order to achieve this purpose, the present invention discloses a shaft surrounding rock deformation prediction method based on fluid-solid coupling linkage simulation, and the steps adopted are:

[0006] Step 1: Use water content changes to describe the effect of rainwater infiltration on surrounding rock deformation, construct a bidirectional fluid-solid coupling surrounding rock deformation time series model under rainwater infiltration, and form a seepage-surrounding rock deformation model; based on the bidirectional fluid-solid coupling surrounding rock deformation time series model, simulate the fluid-solid coupling under rainwater scouring, and construct a scouring-surrounding rock deformation model to describe the deformation of the surrounding rock under rainwater scouring over time; based on the seepage-surrounding rock deformation model, the scouring-surrounding rock deformation model, and the shaft excavation parameters, establish a shaft excavation model under rainwater action;

[0007] Step 2: Perform multi-scale linkage simulation on the seepage-surrounding rock deformation model, the scouring-surrounding rock deformation model, and the vertical shaft excavation model under the action of rainwater to obtain surrounding rock deformation data generated by actual vertical shaft excavation at different time scales;

[0008] Step 3: Using the surrounding rock deformation data obtained in step 2 as a training set to train a TCN-LSTM neural network model, a prediction model for the surrounding rock deformation of the shaft under the action of rainwater is obtained. The TCN (temporal convolutional network)-LSTM (long short-term memory network) neural network model is a combination model of TCN and LSTM.

[0009] Step 4: Acquire real-time surrounding rock deformation data, input it into the vertical shaft surrounding rock deformation prediction model under the action of rainwater, and output the real-time surrounding rock deformation prediction result.

[0010] As a preferred embodiment of the present invention, in step 1, constructing a bidirectional fluid-solid coupled surrounding rock deformation time series model under rainwater infiltration to form a infiltration-surrounding rock deformation model includes:

[0011] Step 1.1. Establish a multi-layer geological model consisting of multiple geological layers, including a soil layer, an ash layer, a surrounding rock layer, and a bedrock layer;

[0012] Step 1.2. Treat all geological layers in the multi-layer geological model as porous media and use the soil water retention curve model to calculate the change in water content of each geological layer under the action of rainwater infiltration.

[0013] Step 1.3. Define the porosity, permeability, and pore water pressure of each geological layer in the multi-layer geological model based on the calculated water content of each geological layer;

[0014] Step 1.4. Use the time series model between seepage and deformation to describe the deformation trend of the surrounding rock to form a surrounding rock deformation time series model;

[0015] Step 1.5. Based on the surrounding rock deformation time series model, bidirectional fluid-solid coupling under seepage is realized to obtain the bidirectional fluid-solid coupled surrounding rock deformation time series model, and the bidirectional fluid-solid coupled surrounding rock deformation time series model is used as the seepage-surrounding rock deformation model.

[0016] As a preferred embodiment of the present invention, the calculation of the water content change of each geological layer under the action of rainwater infiltration using the soil water retention curve model in step 1.2 includes:

[0017] Step 1.2.1. Calculate the matric suction of each geological layer:

[0018]

[0019] Among them, ψ i (h) is the matrix suction of the ith geological layer at depth h, ψ i,top and ψ i,bottom are the matrix suction at the top and bottom of the ith geological layer, h i,top and h i,bottom are the top and bottom depths of the i-th geological layer;

[0020] Step 1.2.2. Calculate the water content at any point in each geological layer using the soil water retention curve model based on the matrix suction of each geological layer:

[0021]

[0022] Among them, θ ωi (h) is the water content at any point in the ith geological layer at depth h, θ ri is the residual water content of the i-th geological layer, θ si is the saturated water content of the i-th geological layer, α i 、n i are the fitting parameters in the soil water retention curve model,

[0023] Step 1.2.3. Calculate the change in matrix suction caused by rainwater infiltration using the Richards equation based on the calculated water content of each geological layer:

[0024]

[0025] Among them, K(θ ωi (h)) represents the water content-dependent permeability coefficient, represents a divergence operator, and t is time. As a preferred embodiment of the present invention, in step 1.5, bidirectional fluid-solid coupling under osmotic action is achieved by adopting a computational fluid dynamics (CFD) method and a discrete element method (DEM), wherein the porosity, permeability, and pore water pressure of each geological layer defined are used as material parameters of the computational fluid dynamics method, the discrete element method is used to describe the mechanical behavior between particles in each geological layer, and bidirectional coupling is achieved through the interaction between fluid pressure and solid deformation.

[0026] As a preferred embodiment of the present invention, in the process of constructing the scour-surrounding rock deformation model in step 1, computational fluid dynamics and discrete element method are used to perform fluid-solid coupling simulation under the action of rainwater scour, and the steps include:

[0027] When using computational fluid dynamics, the fluid domain of rainwater is defined, and corresponding boundary conditions and initial conditions are applied, as well as a turbulence model is selected to simulate the scouring effect of rainfall. When using the discrete element method, the solid particles are modeled using the discrete element method, and each particle unit is regarded as an independent rigid body.

[0028] Data exchange between computational fluid dynamics (CFD) and discrete element methods (DEM) enables bidirectional coupling of fluid dynamics and the kinematic interaction of solid particles. The fluid pressure forces from the CFD method are transferred to the solid particles in the J-OCTA software and DEM to simulate the erosion of soil and volcanic ash layers caused by rainwater erosion and to evaluate the stress and deformation of each material layer. The J-OCTA software is used to address the interface problems of fluid-solid interaction between different geological layers.

[0029] As a preferred embodiment of the present invention, in the process of constructing the scour-surrounding rock deformation model in step 1, processing the interface problem of fluid-solid coupling of different geological layers includes:

[0030] Computational fluid dynamics and the discrete element method were used to simulate the particle loss process in soil layers during rainwater infiltration and soil erosion. The multi-scale model of J-OCTA software was used to couple the movement of soil particles under rainwater erosion. The relationship between pore water pressure and soil deformation in the fluid-solid coupling was solved to simulate the deformation and stress changes of soil under erosion and seepage conditions.

[0031] The interaction between the movement of soil particles in the soil layer under load and the fluid is described by the following formula:

[0032]

[0033] Among them F i is the total force on soil particle i, is the collision force between particles i and j, and are the electrostatic force and van der Waals force of particles i and j, respectively, F i 流体 is the drag force and buoyancy of the fluid on particle i;

[0034] The numerical solver of J-OCTA software is used to solve the following relationship model between pore water pressure and soil deformation in fluid-solid coupling:

[0035]

[0036] Where σ′ is the effective stress, α is the Biot coefficient, p is the pore water pressure, ρ is the soil density, g is the acceleration of gravity, k is the permeability coefficient, and μ is the fluid viscosity;

[0037] The coarse-graining method in the J-OCTA software is used to describe the interaction between volcanic ash particles in the volcanic ash layer, and the built-in MPS (Moving Particle Semi-Implicit) fluid model is called to simulate the behavior of particles separating and depositing from the volcanic ash layer as rainwater erodes. By establishing a coarse-graining model for volcanic ash particles in the J-OCTA software, the fine particles of volcanic ash are processed by coarse-graining, and the multi-scale coupling function of J-OCTA is used to perform multi-scale coupling simulations with the phase field method to deal with the erosion problem of the volcanic ash layer under rainwater erosion. The interaction potential between volcanic ash particles is described as:

[0038]

[0039] Where U(r) is the potential energy between particles, k B is the Boltzmann constant, A is the Hamaker constant, σ is the particle diameter, r is the particle distance, T is the temperature, and n is the fitting parameter;

[0040] The phase field method is used to describe the gradual loss and migration of volcanic ash at the interface through the continuous phase change equation according to the following formula:

[0041]

[0042] where φ represents the spatial distribution of volcanic ash concentration, μ is the chemical potential, and M is the kinetic energy coefficient.

[0043] As a preferred embodiment of the present invention, the process of constructing the scour-surrounding rock deformation model in step 1 further includes: invoking the MPS fluid model and the phase field fracture model in the J-OCTA software for coupling, and performing multi-scale coupling calculation on the phase field fracture model and the fluid model through the J-OCTA software, so as to use the phase field fracture model to simulate the crack propagation process that may be generated in the surrounding rock layer under rain scouring according to the following formula:

[0044]

[0045] Among them, G c is the fracture energy, Φ is the crack extension, l is the crack width scale, and G is the fracture toughness of the material;

[0046] By calling the MPS fluid model in the J-OCTA software and combining it with Darcy's law, we can solve the fracture seepage in the bedrock layer:

[0047]

[0048] Where q is the flow velocity, k is the permeability, μ is the fluid viscosity, is the pressure gradient.

[0049] As a preferred embodiment of the present invention, the step of training the TCN-LSTM neural network model in step 3 includes:

[0050] Step 3.1. The time series data X = {x1, x2, ..., x t} is input to the TCN layer in the TCN-LSTM neural network model, where xt represents the displacement and deformation of the surrounding rock observed at time point t;

[0051] Step 3.2. The TCN layer uses convolution operation to extract the time series data X = {x1, x2, ..., x t} features, and obtain time series features;

[0052] Step 3.3. Use the output of the TCN layer as the input to the LSTM layer in the TCN-LSTM neural network model. The LSTM layer captures the long-term dependencies of the time series data.

[0053] Step 3.4. Convert the output of the LSTM layer into the predicted deformation and calculate the difference between the predicted deformation and the actual observation;

[0054] Step 3.5. Use the initial hyperparameters to train the TCN-LSTM neural network model according to steps 3.1 to 3.4, and use the Bayesian optimization algorithm to select the optimal hyperparameters.

[0055] Step 3.6. Use the optimal hyperparameters to retrain the TCN-LSTM neural network model according to steps 3.1 to 3.4 to obtain the final shaft surrounding rock deformation prediction model under the action of rainwater.

[0056] As a preferred embodiment of the present invention, selecting the optimal hyperparameters using the Bayesian optimization algorithm in step 3.5 includes:

[0057] Step 3.5.1. Initialize the Gaussian process model, select the kernel function of the Gaussian process and the initial hyperparameter Θ (0) ;

[0058] Step 3.5.2. Using the current hyperparameter Θ (i) Train the TCN-LSTM model, where i is the number of execution steps, and calculate the loss L on the training set (i) , update the Gaussian process model to include the new observation Θ (i) , L (i) ;

[0059] Step 3.5.3. Calculate the acquisition function EI:

[0060] EI(Θ)=φ(Θ)-μ(Θ)

[0061] Where φ and μ are the standard deviation and mean of the Gaussian process prediction, respectively;

[0062] Step 3.5.4. Select the next hyperparameter Θ based on the acquisition function EI (i+1) Repeat steps 3.5.1 to 3.5.4 until the optimal hyperparameter set Θ is found opt .

[0063] A device for predicting deformation of surrounding rocks of a vertical shaft based on linkage simulation includes a processor and a memory, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to perform the above method.

[0064] Compared with the existing technology, the advantages of the present invention are: by considering the influence of rainwater infiltration and rainwater scouring on the deformation of the shaft surrounding rock, the present invention respectively constructs an infiltration-surrounding rock deformation model considering the effect of rainwater infiltration and a scouring-surrounding rock deformation model considering the effect of rainwater scouring, and then constructs a shaft excavation model under the action of rainwater. It can fully consider the effect of rainwater infiltration and scouring on the surrounding rock deformation, effectively improve the accuracy and reliability of the shaft surrounding rock deformation prediction, and enhance the scope of application and flexibility of the prediction. At the same time, by combining the TCN-LSTM neural network model training to form a shaft surrounding rock deformation prediction model, it can also reduce the complexity of the prediction, improve the efficiency and real-time performance of the prediction, thereby realizing efficient, accurate and real-time prediction of the shaft surrounding rock deformation under the action of rainwater. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] The present invention will be described in more detail below based on embodiments and with reference to the accompanying drawings, wherein:

[0066] Figure 1 3 is a flow chart of the method for predicting deformation of surrounding rock of a vertical shaft based on fluid-solid coupling simulation in this embodiment.

[0067] Figure 2 Schematic diagram of the multi-layer geological layer model constructed in this embodiment.

[0068] Figure 3 This is a schematic diagram of the results obtained by using the coarse-graining method of J-OCTA software to characterize the mud material evolution process under the action of rainwater in a specific embodiment.

[0069] Figure 4 Schematic diagram of the structure of the TCN-LSTM neural network used in this embodiment. DETAILED DESCRIPTION

[0070] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments, but the scope of protection of the present invention is not limited thereby.

[0071] Considering that weather is an important factor in shaft construction, especially rainfall, it will have a significant impact on the deformation of the shaft surrounding rock. For example, consider the deformation of the surrounding rock under the action of rain:

[0072] (1) Rainwater infiltration

[0073] When rainwater seeps into the surrounding rock, it increases the pore water pressure within it. This increased water pressure acts on the surrounding rock, causing changes in the effective stress and a decrease in its mechanical strength, leading to deformation. Particularly in areas with well-developed joints and fissures, the seepage flow further widens these fissures, creating pathways for groundwater migration and thus affecting the integrity and stability of the surrounding rock. Horizontally, the seepage flow can cause the surrounding rock to converge horizontally; vertically, it can lead to vault settlement and floor uplift, both of which can affect the stability of the surrounding rock.

[0074] The effect of seepage on surrounding rock deformation is typically time-dependent. Over time, the seeping water continuously acts on the surrounding rock, causing gradual deformation and damage. This time dependence indicates that surrounding rock deformation and damage is a gradual process that requires long-term monitoring and assessment.

[0075] (2) Rainwater erosion

[0076] Rainwater erosion can damage the surface soil structure and reduce its resistance to erosion. This erosion can reduce soil stability, form mud, and increase the risk of landslides and erosion. The porosity and water content of the soil layer can increase, affecting the soil's mechanical properties. For volcanic ash layers, rainwater erosion can cause volcanic ash particles to mix with water, increasing the water's viscosity and density, affecting the water's flow characteristics and scouring capacity, and forming a muddy material. This material can flow down the slope, scouring and eroding the underlying surrounding rock layers, and changing the structure and porosity of the volcanic ash layer. While surrounding rock and bedrock layers may not directly form mud like soil layers due to their denser structure and higher strength, long-term rainwater erosion can cause surface weathering of these rock layers, forming geomorphic features such as potholes. Especially in areas with well-developed bedrock fissures, rainwater infiltration and erosion can cause damage and deformation of the surrounding rock.

[0077] As mentioned above, the rainfall effect is divided into rainwater infiltration and rainwater scouring. According to different rainwater effects, a shaft surrounding rock deformation prediction model is constructed, which can fully consider the influence of rainfall on the shaft surrounding rock deformation and improve the prediction accuracy.

[0078] The present invention considers the influence of rainwater infiltration and rainwater scouring on the deformation of the shaft surrounding rock, respectively constructs an infiltration-surrounding rock deformation model considering the effect of rainwater infiltration and an scouring-surrounding rock deformation model considering the effect of rainwater scouring, and then constructs a shaft excavation model under the action of rainwater. It can fully consider the effects of rainwater infiltration and scouring on the deformation of the surrounding rock, simulate the influence of rainwater infiltration and scouring on the deformation of the surrounding rock, thereby effectively improving the accuracy and reliability of the prediction of the shaft surrounding rock deformation, especially suitable for construction environments in rainy seasons, improving the scope of application and flexibility of the prediction, avoiding the limitations of the traditional empirical formula method, and at the same time combining TCN-LS The TM neural network model is trained to form a vertical shaft surrounding rock deformation prediction model. The model is used to realize real-time surrounding rock deformation prediction during the vertical shaft excavation process under the action of rainwater. It can also reduce the complexity of the prediction, improve the efficiency and real-time performance of the prediction. Through real-time processing and dynamic updating of data, it can timely reflect the actual situation of surrounding rock deformation, realize efficient, accurate and real-time prediction of vertical shaft surrounding rock deformation under the action of rainwater, overcome the shortcomings of traditional prediction methods in prediction accuracy, calculation efficiency and dynamic update capability, solve the surrounding rock stability problems that may be encountered in actual prediction projects, and ensure the safety and economy of the vertical shaft construction process under rainfall conditions.

[0079] like Figure 1 As shown, the method for predicting deformation of surrounding rock of a vertical shaft based on fluid-solid coupling simulation in this embodiment includes the following steps:

[0080] Step 1: Use the change in water content to describe the effect of rainwater infiltration on surrounding rock deformation, construct a two-way fluid-solid coupling surrounding rock deformation time series model under rainwater infiltration, and form a seepage-surrounding rock deformation model; based on the two-way fluid-solid coupling surrounding rock deformation time series model, simulate the fluid-solid coupling under rainwater scouring, and construct a scouring-surrounding rock deformation model to describe the deformation of the surrounding rock under rainwater scouring over time; according to the seepage-surrounding rock deformation model, the scouring-surrounding rock deformation model and the shaft excavation parameters, establish a shaft excavation model under the action of rainwater.

[0081] In this embodiment, in order to predict the deformation of surrounding rock under the action of seepage, a multi-layer geological model containing bedrock layers is first established, such as Figure 2 As shown in the figure, the change in water content is used to describe the infiltration of rainwater, and then the water content of different geological layers is analyzed. The following steps can be used to construct a time series model of surrounding rock deformation with bidirectional fluid-solid coupling under the action of rainwater infiltration and form a infiltration-surrounding rock deformation model:

[0082] Step 1.1. Establish a multi-layer geological model consisting of multiple geological layers, including soil layer, ash layer, surrounding rock layer and bedrock layer.

[0083] It should be noted that the surrounding rock layer in the multi-layer geological model refers to the rock surrounding a certain crustal material, which is different from the surrounding rock in the surrounding rock deformation prediction in this embodiment.

[0084] Step 1.2. Treat all geological layers in the multi-layer geological model as porous media and use the soil water retention curve model (VG model) to calculate the change in water content of each geological layer under the action of rainwater infiltration;

[0085] Step 1.3. Define the porosity, permeability, and pore water pressure of each geological layer in the multi-layer geological model based on the calculated water content of each geological layer;

[0086] Step 1.4. Use the time series model between seepage and deformation to describe the deformation trend of the surrounding rock to form a surrounding rock deformation time series model;

[0087] Step 1.5. Based on the surrounding rock deformation time series model, bidirectional fluid-solid coupling under seepage is realized to obtain a bidirectional fluid-solid coupled surrounding rock deformation time series model, and the bidirectional fluid-solid coupled surrounding rock deformation time series model is used as the seepage-surrounding rock deformation model.

[0088] The soil water retention curve model (VG model) can be used not only for soil layers, but also for other porous media that need to describe the water characteristic curve, such as rock, concrete, etc. In this embodiment, different geological layers are regarded as porous media, and the VG model is used to describe the relationship between matrix suction ψ and water content θ. w The relationship between them.

[0089] In this embodiment, the soil water retention curve model is used in step 1.2 to calculate the change in water content of each geological layer under the action of rainwater infiltration, which can be calculated using the following steps:

[0090] Step 1.2.1. Define the geological layer parameters of the geological layer, including the residual water content θ ri , saturated water content θ si and soil water retention curve model parameter α i and n i ;

[0091] Step 1.2.2. Calculate the matric suction of each geological layer:

[0092]

[0093] Among them, ψ i (h) is the matrix suction of the ith geological layer at depth h, ψ i,top and ψ i,bottom are the matrix suction at the top and bottom of the ith geological layer, h i,top and h i,bottom are the top and bottom depths of the i-th geological layer.

[0094] Step 1.2.3 Calculate the water content at any point in each geological layer using the soil water retention curve model based on the matrix suction of each geological layer:

[0095]

[0096] Among them, θ ωi (h) is the water content at any point in the ith geological layer at depth h, θ ri is the residual water content of the i-th geological layer, θ si is the saturated water content of the i-th geological layer, α i 、n i are the fitting parameters, The number of geological layers.

[0097] Step 1.2.4. Calculate the change in matrix suction caused by rainwater infiltration using the Richards equation based on the calculated water content of each geological layer:

[0098]

[0099] Among them, K(θ ωi (h)) represents the water content-dependent permeability coefficient, represents the divergence operator and t is the time.

[0100] Following the above steps, this embodiment constructs a universal model for calculating the matric suction of any geological layer i at depth h. This model not only describes the matric suction of a single layer, but also allows calculation of changes in matric suction across any number of layers. By referencing the top and bottom variables, it allows for a more intuitive description of how matric suction changes with depth, particularly when the thickness of the geological layers is uneven. Furthermore, based on this model for calculating matric suction across multiple layers, a water content calculation model is constructed, enabling precise calculation of changes in water content across multiple layers, and thus effectively determining the change in matric suction based on the water content of the matric suction.

[0101] In this embodiment, after making corrections and setting initial conditions according to actual conditions, the water content changes of each geological layer under the action of rainwater infiltration can be obtained based on the above process. At the same time, the water content change curve can be constructed using the VG model, and the inflow and discharge per unit time can be further obtained.

[0102] Changes in water content have a significant impact on surrounding rock deformation. For example, increased water content can reduce surrounding rock strength, increase pore water pressure, intensify seepage, alter stress states, induce displacement, and couple with groundwater. These effects can reduce surrounding rock stability and increase the risk of deformation and failure (especially in weak, fractured, or layered rock masses). In other words, water content affects seepage, which in turn affects surrounding rock deformation. This example uses water content to describe the effect of rainwater seepage on surrounding rock deformation, effectively quantifying the effect of rainwater seepage on surrounding rock deformation.

[0103] The discrete element method (DEM) can handle complex nonlinear and fracture problems, has high flexibility, and is suitable for non-continuous media. In this embodiment, the rock mass is regarded as consisting of a large number of particles or blocks, and the deformation is predicted by simulating the interaction between these particles, so that deformation prediction of granular materials and block rocks can be achieved, thereby handling problems with complex geological conditions such as large deformation, fracture, and sliding. In this embodiment, in step 1.5, bidirectional fluid and solid coupling under osmotic action is achieved by adopting the computational fluid dynamics method (CFD) and the discrete element method (DEM), wherein the porosity, permeability and pore water pressure of each geological layer defined are used as material parameters of the computational fluid dynamics method, and the discrete element method is used to describe the mechanical behavior between particles in each geological layer, and bidirectional coupling is performed through the interaction between fluid pressure and solid deformation.

[0104] In coupling, changes in water content affect the boundary conditions of water flow and the interactions between substances, thereby altering the stress transfer and forces between the fluid and the solid. This example constructs a fluid-solid coupling and surrounding rock deformation model based on the effect of water content on surrounding rock deformation. The fluid-solid coupling process causes changes in the stress state of the surrounding rock, which in turn alters the seepage path and pore water pressure. This example uses iterative calculations to update the surrounding rock deformation state at each time step and feeds it back into the fluid simulation to achieve fluid-solid coupling.

[0105] The VG model can help estimate the changes in pore water pressure under different water content states, and the pore water pressure in turn affects the effective stress distribution of the surrounding rock. In this embodiment, the change in water content is combined with fluid pressure and surrounding rock deformation, and the stress and strain state of the rock mass is adjusted by changing the pore water pressure of the geological layer. The water content of each layer obtained by the VG model can be used to define the porosity, permeability and pore water pressure of each layer, input into the CFD method as the material parameter, and then use DEM to describe the mechanical behavior between particles in each geological layer. Bidirectional coupling is achieved through the interaction between fluid pressure and solid deformation. Since the change in water content is related to time, the deformation trend of the surrounding rock can be described by a time series model between seepage and deformation, and then a time series of surrounding rock deformation can be constructed.

[0106] According to the above process, a bidirectional fluid-solid coupling surrounding rock deformation time series model under seepage can be constructed using water content to represent the seepage effect, that is, a seepage-surrounding rock deformation model can be obtained.

[0107] Based on the above-mentioned multi-layer geological model constructed in rainwater infiltration, considering that the geological layer is actually a three-dimensional entity composed of rock stacking and deposition, and considering the scouring effect on the scouring surface of the geological layer, this embodiment uses the end face of the above-mentioned multi-layer geological model constructed (i.e., the end face used to characterize the vertical shaft wall) as the scouring surface of rainwater scouring.

[0108] When constructing a scour-rock deformation model that accounts for rainwater scour, computational fluid dynamics (CFD) and the discrete element method (DEM) were used to simulate fluid-structure coupling under rainwater scour. To achieve more precise and realistic coupling, J-OCTA software was used to address interface issues within the fluid-structure coupling model, providing more accurate input for predicting surrounding rock deformation.

[0109] In this embodiment, the following steps may be used to construct the scour-surrounding rock deformation model:

[0110] When using computational fluid dynamics, the fluid domain of rainwater is defined, and corresponding boundary conditions and initial conditions are applied, as well as a turbulence model is selected to simulate the scouring effect of rainfall. When using the discrete element method, the solid particles are modeled using the discrete element method, and each particle unit is regarded as an independent rigid body, which interacts according to Newton's laws of motion and the contact mechanics model.

[0111] Data exchange between computational fluid dynamics (CFD) and discrete element methods (DEM) enables bidirectional coupling of fluid dynamics and the kinematic interaction of solid particles. The fluid pressure force in the CFD method is transmitted to the solid particles in the J-OCTA software and the DEM. In turn, the solid deformation affects the flow field to simulate the erosion of soil and volcanic ash layers caused by rainwater erosion and to evaluate the stress and deformation of each layer of material. The J-OCTA software is used to address the interface problems of fluid-solid coupling between different geological layers.

[0112] When simulating fluid-solid coupling phenomena in different geological layers under the action of rainwater erosion, the J-OCTA software can simulate interface problems across scales from micro to macro. The solid material simulation at the interface is very accurate and can handle problems such as fracture, deformation, and stress transfer of complex materials. It is suitable for handling fluid-solid interactions at different scales.

[0113] Under the action of scouring, fluids penetrate geological layers, causing particle deposition and other effects, affecting their deformation and stability. In this example, J-OCTA software was used to analyze the interface problems of fluid-solid coupling in different geological layers, focusing on the mechanical interaction and deformation characteristics of the fluid and solid (different geological layers) at the interface. The J-OCTA software method can accurately simulate the interface problems in fluid-solid coupling, especially for dealing with problems such as fracture, deformation, and stress transfer in complex materials, and can improve the accuracy and applicability of the model when dealing with fluid-solid coupling problems.

[0114] In this embodiment, the construction of the scour-surrounding rock deformation model to address the interface issues of fluid-solid coupling between different geological layers includes:

[0115] 1. Soil layer

[0116] This embodiment uses computational fluid dynamics methods and discrete element method (i.e., DEM+CFD method) to simulate the particle loss process when rainwater infiltrates and scours the soil in the soil layer. The movement behavior of soil particles under rainwater scour is characterized by a multi-scale model using J-OCTA software. The relationship model between pore water pressure and soil deformation in fluid-solid coupling is solved by the finite element method to simulate the deformation and stress changes of the soil under scouring and seepage conditions.

[0117] For example, the movement of soil particles in a soil layer under load and the interaction between them and the fluid can be described using the following equation:

[0118]

[0119] Among them, F i is the total force on soil particle i, is the collision force between particles i and j, and are the electrostatic force and van der Waals force of particles i and j, respectively, F i 流体 is the drag force and buoyancy of the fluid on particle i. The combined effect of these forces determines the motion behavior of soil particles under rainwater erosion, which can be coupled and simulated using the multi-scale model of J-OCTA software.

[0120] When using general coupling methods to deal with interface problems, usually only the collisions between particles and the effects of the fluid on the particles are considered. In this embodiment, the J-OCTA software coupling calculations also take into account electrostatic forces and van der Waals forces. A multi-scale approach is adopted to consider the forces acting on the particles, so that more accurate results can be obtained.

[0121] In this embodiment, when establishing a multi-scale model of the soil layer in the J-OCTA software, the geometric shape, size distribution, and macrostructure of the soil particles and the soil layer are first defined; then, the corresponding physical and chemical properties are defined for the soil particles and the fluid, including electrostatic forces, van der Waals forces, and interaction forces between particles, as well as the permeability and porosity of the soil. The simulation boundary conditions are set according to the actual situation, and the multi-scale coupling function of the J-OCTA software is used to couple the movement of soil particles with the fluid flow, so as to combine the kinetic equations at the particle level with the continuum equations of fluid flow.

[0122] Rainwater infiltration into the soil causes changes in pore water pressure, affecting the stress distribution in the soil layer. For example, the relationship between pore water pressure and soil deformation in fluid-structure interaction can be described using Biot's equation based on the finite element method (FEM):

[0123]

[0124] where σ′ is the effective stress, α is the Biot coefficient, p is the pore water pressure, ρ is the soil density, g is the acceleration of gravity, k is the permeability coefficient, and μ is the fluid viscosity.

[0125] J-OCTA software can more comprehensively consider the forces acting on particles from a multi-scale perspective. In this embodiment, by using the numerical solver of J-OCTA software to solve equations (4), (5), and (6), more accurate results can be obtained than with traditional coupling methods. J-OCTA can create phase separation structure data as voxel grids of STL data, LS-DYNA, etc., to provide support for finite elements. Therefore, J-OCTA can directly call finite element software such as LS-DYNA for simulation, thereby solving (5) and (6) through the solver. In this way, combining J-OCTA can help quickly and accurately analyze and predict the mechanical behavior of materials in fluid-solid coupling problems.

[0126] This embodiment introduces J-OCTA to provide a more comprehensive description of the forces acting on particles and uses J-OCTA to perform multi-scale simulations. As a result, when performing coupled simulations, J-OCTA can handle simulations at the nanometer, micrometer, and even millimeter levels, while the finite element method can only be used to handle macro-scale structural analysis. Therefore, through this coupling method, multi-scale simulations from micro to macro can be achieved, providing a more comprehensive understanding of material design and performance prediction, thereby improving the accuracy of predictions.

[0127] This embodiment uses J-OCTA software to solve the above equations using the finite element method, which can effectively simulate the deformation and stress changes of soil under scouring and seepage conditions.

[0128] 2. Volcanic ash layer

[0129] Because volcanic ash layers are fine and loose, they are prone to erosion and particle migration under rainwater erosion. This example uses J-OCTA software to combine coarse-graining and finite element methods to address the dynamic behavior of volcanic ash layers under rainwater erosion while preserving the key physical properties of the material.

[0130] Fine volcanic ash particles are processed using a coarse-grained model to reduce computational complexity. This example uses a coarse-grained approach to describe the interactions of volcanic ash particles within a volcanic ash layer. A fluid model is used to simulate the separation and deposition of particles from the volcanic ash layer as they are washed away by rainwater. The phase field method is also used to address the erosion of the volcanic ash layer under rainwater.

[0131] In this example, the interaction between volcanic ash particles in the volcanic ash layer is described by using J-OCTA software in combination with a coarse-graining method. A coarse-graining model of volcanic ash particles is established in the J-OCTA software. The fine particles of volcanic ash are coarse-grained, and the interaction between particles can be analyzed at the molecular level. This not only reduces the amount of calculation, but also preserves the key physical properties between particles.

[0132] For example, the interaction potential between volcanic ash particles is described as:

[0133]

[0134] Where U(r) is the potential energy between particles, k B is the Boltzmann constant, A is the Hamaker constant, σ is the particle diameter, r is the distance between particles, T is the temperature, and n is the fitting parameter (the initial value can be 1.1). This potential function can simulate the adsorption, separation and aggregation processes of particles.

[0135] In J-OCTA, the fluid model includes parameters such as velocity, pressure, density, and viscosity, and can be a fluid model based on fluid dynamics. Furthermore, J-OCTA supports the automatic calculation of coarse-grained potential energy and can map the coarse-grained model to a molecular dynamics (MD) or finite element (FEM) model, which facilitates the construction of the fluid model. This embodiment directly calls the fluid model in J-OCTA to simulate the behavior of particles separating and depositing from the volcanic ash layer as they are washed away by rainwater. In multi-scale simulations, the fluid model can not only be combined with the coarse-grained method model, but also directly call the finite element software for more complex fluid-solid coupling simulations.

[0136] In a specific embodiment of the present invention, the coarse-grained method of J-OCTA software is used to characterize the mud material evolution process under the action of rainwater. The results are as follows: Figure 3 (a) to (d) show this. The results show that the key properties of various materials are preserved during the evolution process. Furthermore, as an input parameter for fluid-structure interaction, the fluid is not a fixed substance, but rather a variable fluid with real-time changes in response to the coupling effect. This can be accurately characterized using coarse-graining methods.

[0137] When using the phase field method to deal with the erosion of volcanic ash layers under rainwater erosion, the gradual loss and migration of volcanic ash at the interface can be described by continuous phase change equations, for example:

[0138]

[0139] Here, φ is the phase field variable representing the spatial distribution of volcanic ash concentration, μ is the chemical potential, and M is the kinetic energy coefficient. This equation can describe the dynamic behavior of particles at interfaces, particularly the erosion of volcanic ash by rain.

[0140] This example uses the multiscale coupling capability of J-OCTA to perform multiscale coupled simulations of the phase field method, combining the coarse-grained model with the fluid model, the finite element model, and the phase field method. This integrates physical processes at different scales into a unified simulation framework, solves interface problems using J-OCTA's solver, and analyzes the erosion and particle migration behavior of the volcanic ash layer under rainwater erosion.

[0141] 3. Surrounding rock layer

[0142] The surrounding rock layer may generate crack expansion under the erosion of rainwater, which can be simulated by the phase field fracture model. In this embodiment, the phase field fracture model is used to simulate the crack expansion process of the surrounding rock layer under the erosion of rainwater according to the following formula, which can be expressed as:

[0143]

[0144] Among them, G c is the fracture energy, Φ is a phase field variable representing crack expansion, l is the crack width scale, and G is the fracture toughness of the material. This model can simulate the crack growth process at the interface between different geological layers under the action of water erosion and stress, and specifically simulate the destructive behavior of materials at the interface.

[0145] In this embodiment, a geometric model of the surrounding rock layer (including size, shape, and material properties) is established in J-OCTA, the fluid model is coupled with the phase-field fracture model, and the phase-field fracture model and the fluid model are calculated through multi-scale coupling of J-OCTA. Boundary conditions and loads are set according to actual conditions, and a numerical solver is used to solve the coupled simulation.

[0146] 4. Bedrock layer

[0147] In this embodiment, the fracture seepage in the bedrock layer is described by Darcy's law combined with the finite element method, which can be expressed as:

[0148]

[0149] Where q is the seepage velocity, k is the permeability, μ is the fluid viscosity, and ▽p is the pressure gradient. This equation can be combined with the stress distribution of the bedrock to simulate the stress concentration and deformation of the bedrock layer under the action of fracture water pressure.

[0150] In this embodiment, the MPS fluid model is directly called in J-OCTA and combined with Darcy's law to solve the seepage state of rainwater in the cracks.

[0151] This embodiment addresses the interface issues of fluid-solid coupling in different geological layers in the manner described above, enabling real-time capture of the impact of water pressure changes on the mechanical behavior of each stratum, thereby more accurately analyzing the stress redistribution and deformation of soil and rock strata. It can also assess the impact of moisture on stratum stability, such as soil softening, surrounding rock fragmentation, and bedrock stress transmission, thereby helping to predict potential landslide, collapse, or fracture risks. Analyzing the four geological layers in the manner described above enables more accurate simulation of the interaction between permeability, water pressure, and stratum deformation, particularly at complex interfaces where processing accuracy is higher, enabling the model to provide more accurate surrounding rock deformation time series data input.

[0152] This embodiment uses J-OCTA software to accurately simulate material properties in interface problems, including adhesion, strength, permeability, friction coefficient, and elastic-plastic behavior at the interface. Combined with multiple scale models such as molecular dynamics, coarse-grained models, finite element method, and phase field method, it can analyze the mechanical response and physical properties of different geological layers or materials at the interface, such as porosity, pore water pressure, and interface deformation and stress transfer. It can effectively solve fluid-solid coupling problems, especially fluid-solid coupling problems in stress and deformation prediction in the interface area.

[0153] According to the above process, a fluid-solid coupling surrounding rock deformation model can be constructed to deal with interface problems based on multiple scale methods under rainfall scouring conditions, forming a scouring-surrounding rock deformation time series model. Through multi-scale coupling, the deformation of the surrounding rock under scouring conditions over time can be obtained.

[0154] Step 2: Perform multi-scale linkage simulation on the seepage-surrounding rock deformation model, the scouring-surrounding rock deformation model, and the shaft excavation model to obtain the surrounding rock deformation data generated by the actual shaft excavation at different time scales.

[0155] The shaft excavation model in this embodiment is a time-series model of surrounding rock deformation constructed using the discrete element method. The aforementioned infiltration-surrounding rock deformation model and scour-surrounding rock deformation model are both obtained using multi-scale coupling using J-OCTA software. J-OCTA can create phase separation structure data as STL data, voxel grids such as LS-DYNA, and provide support for finite element analysis. Furthermore, finite element software can be directly called from J-OCTA. J-OCTA also has the ability to couple with the discrete element method and can be applied with discrete element software such as EDEM. Based on this, in this embodiment, the infiltration-surrounding rock deformation model, the scour-surrounding rock deformation model, and the shaft excavation model are simulated in a coordinated manner using J-OCTA's multi-scale coupling software, allowing for data exchange between the three.

[0156] In this embodiment, the constructed seepage-surrounding rock deformation time series model and scouring-surrounding rock deformation time series model are combined with the on-site construction conditions of the shaft and the surrounding rock grade to determine the necessary parameters for shaft excavation, and establish a shaft excavation model under the action of rainwater. The shaft excavation model under the action of rainwater is the model obtained by linking the seepage-surrounding rock deformation model, the scouring-surrounding rock deformation model, and the shaft excavation model through J-OCTA.

[0157] Then, the central axis of the shaft was used as the cutting surface of the model. Considering the displacement deformation of the surrounding rock in the direction of the shaft section under the action of rainwater, the constructed infiltration-surrounding rock deformation time series model and scour-surrounding rock deformation time series model were used with the shaft excavation model for linkage simulation. After multiple simulations using the multi-scale numerical simulation method, a large amount of surrounding rock deformation data was generated by the actual shaft excavation under the action of rainwater.

[0158] The above-mentioned multi-scale numerical simulation method is the coupling method pointed out by J-OCTA, namely the coupling method of coarse-grained model, fluid model, phase field fracture model, finite element and discrete element.

[0159] This embodiment uses linked fluid-solid coupling simulation to accurately capture the impact of fluid flow on formation deformation, especially the transition and interaction at the interface; the use of multi-scale simulation methods can also more comprehensively capture the complexity of surrounding rock deformation under different geological conditions, provide more accurate simulation results, and flexibly handle physical processes at different scales from micro to macro, enhancing the applicability and accuracy of the model.

[0160] Step 3: Use the surrounding rock deformation data obtained in step 2 as a training set to train the TCN-LSTM neural network model to obtain a vertical shaft surrounding rock deformation prediction model under the action of rainwater. The TCN-LSTM neural network model is a combination model of TCN and LSTM.

[0161] In this embodiment, the surrounding rock deformation data within a set time range obtained by a multi-scale numerical simulation method is used as a learning sample of the algorithm to train a pre-established TCN-LSTM neural network model. After learning and training, a vertical shaft surrounding rock deformation prediction model under the action of rainwater is constructed. The prediction range is set according to the accuracy of the vertical shaft surrounding rock deformation prediction model, and the surrounding rock deformation prediction value can be automatically output to determine the deformation value of the specified section of the vertical shaft at various positions under rainwater conditions.

[0162] In this embodiment, a mathematical model of the surrounding rock deformation time series curve is established using a TCN-LSTM neural network. Specifically, the relationship between excavation time t and surrounding rock deformation y is established. This allows for the prediction of surrounding rock deformation in vertical shafts subjected to rainwater, based on the TCN-LSTM neural network. The TCN can identify local patterns and trends in time series, while the LSTM network can process data at each time step, capturing the dynamic characteristics of the time series. A large amount of surrounding rock deformation data obtained through multi-scale simulation methods is input. This surrounding rock deformation data includes surrounding rock displacements at different time points. The displacement series is time-dependent, and the time series data can be converted into features suitable for input, including the time point itself or time-related features such as historical deformation, deformation rate, and deformation acceleration.

[0163] This embodiment uses a combined model of a temporal convolutional network (TCN) and a long short-term memory network (LSTM) called TCN-LSTM to train a prediction model for shaft surrounding rock deformation under the action of rainwater. This can combine the TCN's ability to recognize local patterns in time series data and the LSTM's ability to capture long-term dependencies, effectively improving the model's prediction accuracy for surrounding rock deformation time series data.

[0164] In order to eliminate the influence of different dimensions and obtain better network training results, multiple sets of training samples are normalized and the sample values are converted to the (0,1) interval. The normalization process is as follows:

[0165]

[0166] The processed data is divided into a training set and a test set. The training set is used to build the model, and the test set is used to verify the predictive ability of the model.

[0167] In this embodiment, the TCN part effectively processes time series data through causal convolution operations, and the LSTM part can memorize important historical information. The TCN and LSTM are combined to form a TCN-LSTM neural network model. Figure 4As shown in the figure, in the TCN-LSTM neural network used in this embodiment, the first layer is the input layer, which receives time series data. The second layer, the TCN, extracts features from the input time series data. The TCN uses convolution operations to identify local patterns and trends in the time series. Different x represents different dimensions of the input data or data at different time steps. The feature dimension refers to the number of features after processing by the TCN and then feeding into the LSTM network. The third layer of the LSTM network consists of multiple LSTM cells. Each LSTM cell contains a forget gate, an input gate, and an output gate. The forget gate determines which information is forgotten or retained from the cell state. The input gate controls the addition of new information, including which part of the cell state is updated. The output gate determines the output of the LSTM cell, which is the final prediction or state. The fourth layer, the output layer, is the final result of the network. The fully connected layer converts the LSTM output into the final prediction result, mapping high-dimensional features to the target dimension.

[0168] In this embodiment, the following steps can be used to train the TCN-LSTM neural network model:

[0169] Step 3.1. The time series data X = {x1, x2, ..., x t} is input to the TCN layer in the TCN-LSTM neural network model, where xt represents the displacement and deformation of the surrounding rock observed at time point t;

[0170] Step 3.2. The TCN layer uses convolution operation to extract the time series data X = {x1, x2, ..., x t} features to obtain time series features.

[0171] Step 3.3. Use the output of the TCN layer as the input of the LSTM layer in the TCN-LSTM neural network model. The LSTM layer captures the long-term dependencies of the time series data.

[0172] In the TCN layer, for a single time point t, the output of each causal convolutional layer l is:

[0173]

[0174] in, is the output of the lth layer causal convolution at time point t, is the weight of the l-th convolution kernel at time delay τ, b (l) is the bias term of the lth layer, f (l) is the activation function of layer l.

[0175] In this embodiment, when the TCN layer includes multiple convolutional layers, the output Z of each layer t (l)It will be used as the input of the next layer, and the outputs of all layers can be combined into the final feature vector:

[0176]

[0177] Where M is the total number of TCN convolutional layers, Z t is the feature vector output by the TCN layer at time point t.

[0178] For the entire time series, the TCN layer generates a feature vector Z for each time point t t , forming a feature sequence:

[0179] Z={Z1,Z2,…Z T} (13)

[0180] Where T is the length of the time series.

[0181] The output of the TCN layer is used as the input of the LSTM layer, and the LSTM layer is used to capture the long-term dependencies of time series data, where the forget gate can be expressed as:

[0182] f t =σ(W f ·[Z t-1 ,W h ·h t-1 ]+b f ) (14)

[0183] Among them, W f is the weight matrix of the forget gate, W h is the weight associated with the hidden state at the previous time step, h t-1 is the hidden state of the previous time step, b f is the bias term, and σ is the sigmoid activation function.

[0184] The input gate can be expressed as:

[0185] i t =σ(W i ·[Z t ,W h ·h t-1 ]+b i ) (15)

[0186]

[0187] Among them, i t is the activation value of the input gate, C t is the candidate cell state, W i and W C is the weight matrix of the input gate and candidate unit state, b i and bC is the corresponding bias term.

[0188] The cell state can be updated as follows:

[0189]

[0190] Among them, C t is the cell state at the current time step, C t-1 is the cell state at the previous time step.

[0191] The output gate can be expressed as:

[0192] o t =σ(W o ·[Z t ,W h ,h t-1 ]+b o ) (18)

[0193] Among them, t is the activation value of the output gate, W o is the weight matrix of the output gate, b o is the bias term.

[0194] The hidden state at the current time step can be expressed as:

[0195] h t =o t *tanh(C t ) (19)

[0196] Among them, h t is the hidden state at the current time step, and tanh is the activation function.

[0197] In this embodiment, the TCN layer generates a feature vector Z for each time point t. t , forming a feature sequence Z as the input of LSTM, which is not input layer by layer, but describing the data as a feature sequence as input.

[0198] Step 3.4. Convert the output of the LSTM layer into the predicted deformation and calculate the difference between the predicted deformation and the actual observation.

[0199] When converting the output of the LSTM layer into a predicted deformation, a fully connected layer can be used to predict the deformation at the current time step, for example:

[0200]

[0201] in, is the predicted deformation at the current time step, W is the weight matrix, b is the bias term, and f is the nonlinear activation function.

[0202] Define the loss function to measure the difference between the actual observation value y and the predicted value For example, the mean square error (MSE) can be used as the loss function:

[0203]

[0204] Where N is the number of samples; y t is the actual deformation at the tth time step; is the deformation predicted by the model at the tth time step.

[0205] Step 3.5. Use the initial hyperparameters to train the TCN-LSTM neural network model according to steps 3.1 to 3.4, and use the Bayesian optimization algorithm to select the optimal hyperparameters.

[0206] Step 3.6. Use the optimal hyperparameters to retrain the TCN-LSTM neural network model according to steps 3.1 to 3.4 to obtain the final shaft surrounding rock deformation prediction model under the action of rainwater.

[0207] In this embodiment, the Bayesian optimization algorithm in step 3.5 can be used to select the optimal hyperparameters using the following steps:

[0208] Step 3.5.1. Initialize the Gaussian process model, select the kernel function of the Gaussian process and the initial hyperparameter Θ (0) .

[0209] Define the hyperparameter space Θ, including the learning rate, parameters of the TCN and LSTM layers, etc. Use Bayesian optimization to select the optimal hyperparameters, initialize the Gaussian process model, and select the kernel function of the Gaussian process. In this paper, the radial basis function (RBF) is selected:

[0210]

[0211] Where x and x′ are two points in the input space; ||xx′|| represents the Euclidean distance between the two points; l is the length parameter of the sum function; k(x,x′) is the output of the kernel function, which represents the similarity between the two output points.

[0212] Step 3.5.2. Using the current hyperparameter Θ (i) Train the TCN-LSTM model, where i is the number of execution steps, and calculate the loss L on the training set (i) , update the Gaussian process model to include the new observation Θ (i) , L (i) ;

[0213] Step 3.5.3. Select expected improvement (EI) as the acquisition function and calculate the acquisition function EI:

[0214] EI(Θ)=φ(Θ)-μ(Θ) (22)

[0215] Where φ and μ are the standard deviation and mean of the Gaussian process prediction, respectively;

[0216] Step 3.5.4. Select the next hyperparameter Θ based on the acquisition function EI (i+1) Repeat steps 3.5.1 to 3.5.4 until the optimal hyperparameter set Θ is found opt .

[0217] This embodiment adopts the Bayesian optimization algorithm to select the optimal hyperparameters, uses the radial basis function (RBF) as the kernel function of the Gaussian process, and constructs a performance probability model of the hyperparameters to guide the search process, thereby avoiding invalid attempts and finding the optimal hyperparameter combination within a smaller number of iterations. Moreover, since the surrounding rock deformation data changes over time, by utilizing the adaptive learning ability of the Bayesian optimization algorithm, the model can also quickly adapt to new data inputs, significantly enhancing the prediction ability for new samples and ensuring the continuous optimization and accuracy of model predictions.

[0218] Use the optimal hyperparameter Θ opt Training the TCN-LSTM model:

[0219] Θ * =argmin Θ L (23)

[0220] Where L is the loss function, which is the mean square error.

[0221] Finally, the test set is used to evaluate the model performance and calculate the prediction error. new Make a prediction, that is:

[0222]

[0223] The test set data usually includes deformation data at continuous time points to evaluate the model's ability to capture time series changes. At the same time, the model parameters are adjusted through continuously updated training data to adapt to changes in deformation data over time. The prediction model of this embodiment is trained and verified through a large amount of surrounding rock deformation data to ensure its effectiveness and reliability in actual engineering. At the same time, the model parameters are adjusted through continuously updated training data, and the model can adapt to changes in deformation data over time. The model obtained after the above processing is denormalized to obtain the predicted data of surrounding rock deformation, and finally a trained rainwater-effect vertical shaft surrounding rock prediction model based on the TCN-LSTM neural network is obtained, which can be used to predict new surrounding rock deformation data under rainfall conditions.

[0224] Step 4: Obtain real-time surrounding rock deformation data, input it into the vertical shaft surrounding rock deformation prediction model under the action of rainwater, and output the real-time surrounding rock deformation prediction result.

[0225] This embodiment constructs a seepage-surrounding rock deformation time series model and a scouring-surrounding rock deformation time series model based on the different effects of rainwater, and then performs a linkage simulation with the shaft excavation. It can accurately capture the impact of fluid flow on formation deformation and effectively generate a large amount of surrounding rock deformation time series data generated by actual shaft excavation. This time series data is then used to train the TCN-LSTM neural network model. By combining the local pattern recognition capability of the temporal convolutional network and the long-term dependency capture capability and local pattern of the long-short-term memory network, the prediction accuracy of the shaft surrounding rock deformation time series data can be effectively improved. At the same time, it can also achieve rapid learning, strong adaptability, intelligent optimization of hyperparameters and real-time data processing, effectively improving the prediction efficiency and adaptability of the model, while reducing computing costs and enhancing the practicality and reliability of the model in actual engineering.

[0226] While the present invention has been described with reference to preferred embodiments, various modifications may be made and equivalent components may be substituted without departing from the scope of the present invention. In particular, the various technical features described in the various embodiments may be combined in any manner, provided no structural conflicts exist. The present invention is not limited to the specific embodiments disclosed herein, but encompasses all technical solutions within the scope of the claims.

Claims

1. A method for predicting deformation of surrounding rock of a vertical shaft based on fluid-solid coupling simulation, characterized in that the steps include: Step 1: Use water content changes to describe the effect of rainwater infiltration on surrounding rock deformation, construct a bidirectional fluid-solid coupling surrounding rock deformation time series model under rainwater infiltration, and form a seepage-surrounding rock deformation model; based on the bidirectional fluid-solid coupling surrounding rock deformation time series model, simulate the fluid-solid coupling under rainwater scouring, and construct a scouring-surrounding rock deformation model to describe the deformation of the surrounding rock under rainwater scouring over time; based on the seepage-surrounding rock deformation model, the scouring-surrounding rock deformation model, and the shaft excavation parameters, establish a shaft excavation model under rainwater action; Step 2: Perform multi-scale linkage simulation on the seepage-surrounding rock deformation model, the scouring-surrounding rock deformation model, and the vertical shaft excavation model under the action of rainwater to obtain surrounding rock deformation data generated by actual vertical shaft excavation at different time scales; Step 3: Using the surrounding rock deformation data obtained in step 2 as a training set to train a TCN-LSTM neural network model, a prediction model for the surrounding rock deformation of the shaft under the action of rainwater is obtained. The TCN-LSTM neural network model is a combination model of TCN and LSTM. Step 4: obtaining real-time surrounding rock deformation data, inputting the data into the shaft surrounding rock deformation prediction model under the action of rainwater, and outputting the real-time surrounding rock deformation prediction result; In the process of constructing the scour-surrounding rock deformation model in step 1, the interface problem of fluid-solid coupling between different geological layers is handled as follows: Computational fluid dynamics and the discrete element method were used to simulate the particle loss process in soil layers during rainwater infiltration and soil erosion. The multi-scale model of J-OCTA software was used to couple the movement of soil particles under rainwater erosion. The relationship between pore water pressure and soil deformation in the fluid-solid coupling was solved to simulate the deformation and stress changes of soil under erosion and seepage conditions. The coarse-graining method in the J-OCTA software is used to describe the interaction between volcanic ash particles in the volcanic ash layer, and the built-in MPS fluid model is called to simulate the behavior of particles separating and depositing from the volcanic ash layer as rainwater erodes. By establishing a coarse-graining model of volcanic ash particles in the J-OCTA software, the fine particles of volcanic ash are coarse-grained, and using the multi-scale coupling function of J-OCTA, the phase field method is used for multi-scale coupling simulation to deal with the erosion problem of the volcanic ash layer under rainwater erosion, and to describe the interaction potential between volcanic ash particles.

2. The method for predicting deformation of surrounding rock of a vertical shaft based on fluid-solid coupling simulation according to claim 1 is characterized in that: In step 1, a bidirectional fluid-solid coupled surrounding rock deformation time series model under rainwater infiltration is constructed to form an infiltration-surrounding rock deformation model including: Step 1.

1. Establish a multi-layer geological model consisting of multiple geological layers, including a soil layer, a volcanic ash layer, a surrounding rock layer, and a bedrock layer; Step 1.

2. Treat all geological layers in the multi-layer geological model as porous media and use the soil water retention curve model to calculate the change in water content of each geological layer under the action of rainwater infiltration. Step 1.

3. Define the porosity, permeability, and pore water pressure of each geological layer in the multi-layer geological model based on the calculated water content of each geological layer; Step 1.

4. Use the time series model between seepage and deformation to describe the deformation trend of the surrounding rock to form a surrounding rock deformation time series model; Step 1.

5. Based on the surrounding rock deformation time series model, bidirectional fluid-solid coupling under seepage is realized to obtain the bidirectional fluid-solid coupled surrounding rock deformation time series model, and the bidirectional fluid-solid coupled surrounding rock deformation time series model is used as the seepage-surrounding rock deformation model.

3. The method for predicting deformation of surrounding rock of a vertical shaft based on fluid-solid coupling simulation according to claim 2 is characterized in that: The calculation of the water content change of each geological layer under the action of rainwater infiltration using the soil water retention curve model in step 1.2 includes: Step 1.2.

1. Calculate the matric suction of each geological layer: in, It means the depth is No. The matrix suction of the geological layer, and They are The matrix suction at the top and bottom of each geological layer, and It is The depth of the top and bottom of each geological layer; Step 1.2.

2. Calculate the water content at any point in each geological layer using the soil water retention curve model based on the matrix suction of each geological layer: in, The depth is No. The water content at any point in a geological layer, For the The residual water content of the geological layer, For the The saturated water content of a geological layer, 、 are the fitting parameters in the soil water retention curve model, ; Step 1.2.

3. Calculate the change in matrix suction caused by rainwater infiltration using the Richards equation based on the calculated water content of each geological layer: in, represents the water content-dependent permeability coefficient, represents the divergence operator, It's time.

4. The method for predicting deformation of surrounding rock of a vertical shaft based on fluid-solid coupling simulation according to claim 2 is characterized in that: In step 1.5, bidirectional fluid-solid coupling under osmosis is achieved by adopting computational fluid dynamics methods and discrete element methods, wherein the porosity, permeability and pore water pressure of each geological layer are defined as material parameters of the computational fluid dynamics method, and the discrete element method is used to describe the mechanical behavior between particles in each geological layer, and bidirectional coupling is achieved through the interaction between fluid pressure and solid deformation.

5. The method for predicting deformation of surrounding rock of a vertical shaft based on fluid-solid coupling simulation according to claim 1 is characterized in that: In the process of constructing the scouring-surrounding rock deformation model in step 1, computational fluid dynamics and discrete element method are used to perform fluid-solid coupling simulation under the action of rainwater scouring, and the steps include: When using computational fluid dynamics, the fluid domain of rainwater is defined, and corresponding boundary conditions and initial conditions are applied, as well as a turbulence model is selected to simulate the scouring effect of rainfall. When using the discrete element method, the solid particles are modeled using the discrete element method, and each particle unit is regarded as an independent rigid body. Data exchange between computational fluid dynamics (CFD) and discrete element methods (DEM) enables bidirectional coupling of fluid dynamics and the kinematic interaction of solid particles. The fluid pressure forces from the CFD method are transferred to the solid particles in the J-OCTA software and DEM to simulate the erosion of soil and volcanic ash layers caused by rainwater erosion and to evaluate the stress and deformation of each material layer. The J-OCTA software is used to address the interface problems of fluid-solid interaction between different geological layers.

6. The method for predicting deformation of surrounding rock of a vertical shaft based on fluid-solid coupling simulation according to claim 5 is characterized in that: In the process of constructing the scour-surrounding rock deformation model in step 1, in dealing with the interface problem of fluid-solid coupling in different geological layers, the interaction between the movement of soil particles in the soil layer under stress and the fluid is described using the following formula: in Soil particles The total force, It is a particle and The collision force between and Particles and electrostatic force and van der Waals force, The fluid to particle Drag and buoyancy; The numerical solver of J-OCTA software is used to solve the following relationship model between pore water pressure and soil deformation in fluid-solid coupling: in is the effective stress, is the Biot coefficient, is the pore water pressure, is the soil density, is the acceleration due to gravity, is the permeability coefficient, is the fluid viscosity; The phase field method is used to describe the gradual loss and migration of volcanic ash at the interface through the continuous phase change equation according to the following formula: in, represents the spatial distribution of volcanic ash concentration, is the chemical potential, is the kinetic energy coefficient.

7. The method for predicting deformation of surrounding rock of a vertical shaft based on fluid-solid coupling simulation according to claim 6 is characterized in that: The process of constructing the scour-surrounding rock deformation model in step 1 further includes: invoking the fluid model and the phase field fracture model in the J-OCTA software for coupling, and performing multi-scale coupling calculations on the phase field fracture model and the fluid model through the J-OCTA software, so as to use the phase field fracture model to simulate the crack propagation process that may occur in the surrounding rock layer under rain scour according to the following formula: in, is the fracture energy, For the expansion of cracks, is the crack width scale, is the fracture toughness of the material; By calling the MPS fluid model in the J-OCTA software and combining it with Darcy's law, we can solve the fracture seepage in the bedrock layer: in, is the seepage velocity, is the permeability, is the fluid viscosity, is the pressure gradient.

8. The method for predicting deformation of surrounding rock of a vertical shaft based on fluid-solid coupling simulation according to any one of claims 1 to 7, characterized in that: The steps for training the TCN-LSTM neural network model in step 3 include: Step 3.

1. Time series data consisting of surrounding rock deformation data at different time scales Input to the TCN layer in the TCN-LSTM neural network model, where Indicates at a point in time Observed displacement and deformation of surrounding rock; Step 3.

2. Extract time series data using convolution operations by TCN layer The characteristics of the time series are obtained; Step 3.

3. Use the output of the TCN layer as the input to the LSTM layer in the TCN-LSTM neural network model. The LSTM layer captures the long-term dependencies of the time series data. Step 3.

4. Convert the output of the LSTM layer into the predicted deformation and calculate the difference between the predicted deformation and the actual observation; Step 3.

5. Use the initial hyperparameters to train the TCN-LSTM neural network model according to steps 3.1 to 3.4, and use the Bayesian optimization algorithm to select the optimal hyperparameters. Step 3.

6. Use the optimal hyperparameters to retrain the TCN-LSTM neural network model according to steps 3.1 to 3.4 to obtain the final shaft surrounding rock deformation prediction model under the action of rainwater.

9. The method for predicting deformation of surrounding rock of a vertical shaft based on fluid-solid coupling simulation according to claim 8, characterized in that: The Bayesian optimization algorithm used in step 3.5 to select the optimal hyperparameters includes: Step 3.5.

1. Initialize the Gaussian process model, select the kernel function and initial hyperparameters of the Gaussian process ; Step 3.5.

2. Using Current Hyperparameters Train the TCN-LSTM model, where is the number of execution steps and calculates the loss on the training set , update the Gaussian process model to include the new observation , ; Step 3.5.

3. Calculate the acquisition function : in and are the standard deviation and mean of the Gaussian process prediction, respectively; Step 3.5.

4. According to the acquisition function Choose hyperparameters for the next step Repeat steps 3.5.1 to 3.5.4 until the optimal set of hyperparameters is found. .

10. A device for predicting deformation of surrounding rock of a vertical shaft based on fluid-solid coupling simulation, comprising a processor and a memory, wherein the memory is used to store a computer program, characterized in that: The processor is configured to execute the computer program to perform the method according to any one of claims 1 to 9.

Images