Intelligent grouting calculation method and system based on neural network and numerical solution coupling

Abstract

The invention discloses an intelligent grouting calculation method and system based on coupling of a neural network and numerical solution. The method comprises the steps that a fluid calculation control equation needed by working condition calculation is determined; according to the method, boundary conditions, grouting speed, grouting pressure, a flowing water initial flow field, a gravity field, a slurry diffusion form, a temperature field, slurry viscosity, constant-volume specific heat capacity, heat conductivity and density are used as input data, and numerical simulation of an initial stage is carried out on the input data through a numerical simulation method; the parameters of the last time step obtained through numerical simulation in the initial stage serve as initial data and are input into the trained neural network model, and prediction results of the slurry speed, the grouting pressure, the phase fraction and the viscosity data are obtained; according to the method, numerical simulation of an initial stage is carried out on input data through a physical solver, physical fields such as pressure, speed and temperature and local details and global spatio-temporal evolution trends among the fields are captured through a neural network model, and efficient grouting numerical simulation is achieved.

Description

Technical Field

[0001] The invention relates to the technical field of geotechnical engineering, and in particular to a grouting intelligent calculation method and system based on coupling of neural network and numerical solution. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] With the rapid advancement of tunnel engineering technology, the geological conditions faced are becoming more and more complex. The water disaster mechanisms under these conditions are complex and changeable, which brings unprecedented challenges to the pre-control decision-making of tunnel construction. This complexity not only increases the safety risks during the construction process, but also may lead to uncertainty in project schedule and cost. In this context, grouting technology, as an effective method of stratum reinforcement and water disaster prevention, is particularly important. It prevents the occurrence of sudden water disasters and improves geological conditions by injecting specific slurry, thereby ensuring the smooth progress of construction.

[0004] In order to accurately control the grouting process and improve its efficiency, grouting numerical simulation technology came into being. This technology relies on advanced computer simulation and can simulate the flow and solidification process of grouting in the ground before construction, helping engineers optimize grouting schemes. However, traditional simulation methods such as statistical models and numerical simulations have a series of limitations. Statistical models lack the flexibility to adapt to complex geological conditions, and although numerical simulations have high accuracy, they are complex and time-consuming to calculate, especially when faced with large-scale or complex geological conditions. These limitations are more obvious.

[0005] In recent years, with the development of deep learning technology, improving the efficiency and accuracy of grouting simulation by combining machine learning methods with simulation data has become a research hotspot. However, the performance of current deep learning models depends largely on a large amount of high-quality training data, while the actual construction environment is complex and the geological conditions are changeable, making it difficult to obtain high-quality training data. In addition, current deep learning models are often regarded as "black boxes" and their decision-making process lacks transparency. Common deep learning models cannot achieve a deep physical understanding, making it difficult to achieve fast and accurate learning and prediction, and it is difficult to verify their prediction accuracy, which makes the prediction results somewhat blind and uncertain. Summary of the invention

[0006] In order to solve the above problems, the present invention proposes an intelligent calculation method and system for grouting based on the coupling of neural network and numerical solution. The physical solver is used to perform numerical simulation of the input data in the initial stage, and the neural network model is used to capture the physical fields such as pressure, velocity, temperature, as well as the local details and global spatiotemporal evolution trends between the fields, thereby realizing efficient grouting numerical simulation.

[0007] In some embodiments, the following technical solutions are adopted:

[0008] A grouting intelligent calculation method based on coupling of neural network and numerical solution, comprising:

[0009] Determine the fluid calculation control equation required for the calculation condition; the fluid calculation control equation includes a momentum control equation, and a damping term is added to the momentum control equation;

[0010] The boundary conditions, grouting velocity, grouting pressure, initial flow field of dynamic water, gravity field, slurry diffusion morphology, temperature field, slurry viscosity, constant volume specific heat capacity, thermal conductivity and density are used as input data. The input data is numerically simulated in the initial stage by numerical simulation method. The parameters of the last time step obtained by the initial stage numerical simulation are used as initial data and input into the trained neural network model to obtain the prediction results of slurry velocity, grouting pressure, phase fraction and viscosity data.

[0011] Among them, the loss function of the neural network model includes a data fitting loss term, a physical residual loss term and a boundary loss function term; the momentum control equation with the damping term added is used as the physical residual loss term.

[0012] As an optional solution, it also includes: for the output result of the neural network model, setting a% of the total time step to perform numerical solution verification; establishing a three-stage step-by-step trust verification mechanism, specifically:

[0013] In the first stage, a fixed number of time step nodes are used to perform preliminary verification of numerical calculations. If the error of the verification result meets the requirements, the second stage is entered. Otherwise, the initial grouting condition parameters are adjusted and the grouting intelligent calculation is performed again.

[0014] In the second stage, trust in the model is enhanced by gradually reducing the verification frequency until the model output is fully trusted. After full trust, simulation prediction is carried out, and after the prediction is completed, the third stage is entered;

[0015] In the third stage, the prediction results of the last time step are verified by numerical calculation. If the error of the verification result meets the requirements, the grouting simulation result is output; otherwise, the initial grouting condition parameters are adjusted and the grouting intelligent calculation is performed again; where a is the set value.

[0016] As an optional solution, when performing numerical simulation of the input data in the initial stage, the time span of the initial stage is dynamically adjusted according to the actual complexity of the working conditions; specifically:

[0017]

[0018] χ=α v m+α τ τ+αS S;

[0019] k=α u u+α p p+α f f+α d d;

[0020] Among them, T0 is the basic time span, α, β, γ, δ, α v , α τ , α S , α u , α p , α f , α d are weight coefficients, is the inverse of the grouting rate, is the inverse of the grouting pressure, χ is the slurry selection parameter, k is the geological condition complexity coefficient, m is the slurry viscosity, τ is the curing time, S is the chemical stability coefficient, u is the stability of the formation mechanical properties, p is the formation permeability, f is the degree of crack development, and d is the disaster-prone structure coefficient.

[0021] As an optional solution, a damping term is added to the momentum control equation, specifically:

[0022]

[0023] Among them, ρ is density, p is pressure, and ρvv is the momentum flux tensor, that is, the rate of change of fluid momentum per unit volume; is the pressure gradient, which indicates the rate of change of pressure in the x, y, and z directions; is the velocity gradient, which indicates the velocity change rate of the fluid in each x, y, and z direction; μ(t,T) is the viscosity function characterized by time t and slurry temperature T, which can be obtained through experiments; g is the gravitational acceleration, and F st is the surface tension; γ is the control damping strength, is the damping term, where γ is the coefficient that controls the damping strength; is the gradient operator, and Ρv is the momentum density, that is, the momentum per unit volume.

[0024] As an optional solution, the loss function of the neural network model is specifically:

[0025] L=L data +λ physics ·||R|| 2 +λ boundary ·L boundary ;

[0026] Among them, L data is the data fitting loss term, λ physics ·||R||2 is the physical residual loss term, λ physics To control the hyperparameters affecting the physical residual, ||R|| 2 is the square of the residual between the output of the neural network model and the calculation result of the fluid calculation control equation; L boundary is the boundary loss function term, L boundary A hyperparameter that balances the weight of the boundary condition loss in the total loss.

[0027] As an optional solution, the data fitting loss term L data for:

[0028]

[0029] Among them, y pred,i is the predicted value of the neural network at the i-th data point, y obs,i is the corresponding actual observed value, and N is the total number of data points.

[0030] As an optional solution, the boundary loss function term L boundary for:

[0031] L boundary =∑ i∈boundary ||u pred,i -u boundary,i || 2 ;

[0032] Among them, u pred,i is the model prediction value at the boundary position, u boundary,i The value specified for the boundary condition.

[0033] In other embodiments, the following technical solutions are adopted:

[0034] A grouting intelligent computing system based on coupling of neural network and numerical solution, comprising:

[0035] A fluid calculation control equation building module is used to determine the fluid calculation control equation required for the calculation condition; the fluid calculation control equation includes a momentum control equation, and a damping term is added to the momentum control equation;

[0036] The grouting intelligent calculation module is used to take the boundary conditions, grouting speed, grouting pressure, dynamic water initial flow field, gravity field, slurry diffusion morphology, temperature field, slurry viscosity, constant volume specific heat capacity, thermal conductivity and density as input data, perform numerical simulation of the input data in the initial stage through the numerical simulation method, take the parameters of the last time step obtained by the numerical simulation in the initial stage as the initial data, input them into the trained neural network model, and obtain the prediction results of slurry speed, grouting pressure, phase fraction and viscosity data;

[0037] Among them, the loss function of the neural network model includes a data fitting loss term, a physical residual loss term and a boundary loss function term; the momentum control equation with the damping term added is used as the physical residual loss term.

[0038] In other embodiments, the following technical solutions are adopted:

[0039] A terminal device comprises a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0040] In other embodiments, the following technical solutions are adopted:

[0041] A computer-readable storage medium stores a computer program / instruction, which implements the steps of the above method when executed by a processor.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] (1) The method of the present invention constructs a neural network framework for accurately capturing the fluid dynamics characteristics. By adding a damping term to the momentum control equation, the numerical stability and physical credibility of the model are enhanced, and the excessive oscillation and non-physical behavior of the system are effectively controlled. The control equation of the slurry flow is embedded in the loss function as a physical constraint after regularization, realizing the neural network's understanding of physical laws. Based on the EDMD algorithm, the time-space analytical flow field of the sampled data is subjected to nonlinear modal decomposition, and the instantaneous dynamic changes of the slurry on the time scale and the acquisition of structural and periodic dynamic characteristics on the spatial scale are realized. In the process of capturing the flow field characteristics, the high-dimensional dilemma caused by random noise or non-critical features in the non-overfitting data is weakened, thereby achieving data dimensionality reduction.

[0044] (2) The method of the present invention performs an initial stage of numerical simulation of the input data through numerical simulation of physical equations. The trained model captures the physical fields such as pressure, velocity, temperature, as well as the local details and global spatiotemporal evolution trends between the fields. It replaces the physical solver to complete all subsequent iterative processes, and further physical solution iteration and verification are performed based on the simulation prediction results, realizing the mutual feedback verification between the physical solution and the neural network, ensuring the accuracy of the simulation data and realizing efficient grouting numerical simulation.

[0045] (3) The method of the present invention can dynamically adjust the time span of the initial stage according to the actual complexity of the working conditions. By integrating the dynamic grouting data and geological conditions, a dynamically adjusted time span calculation formula is proposed. Intelligent simulation is performed based on the time span, which effectively ensures the result accuracy under complex working conditions and the calculation efficiency under simple working conditions, and is adaptable to different geological conditions and grouting schemes.

[0046] Other features and advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a schematic diagram of the intelligent calculation method for grouting based on the coupling of neural network and numerical solution in an embodiment of the present invention. DETAILED DESCRIPTION

[0048] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used in the present invention have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0049] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0050] Embodiment 1

[0051] In one or more embodiments, a grouting intelligent calculation method based on the coupling of neural network and numerical solution is disclosed, combined with Figure 1 , specifically including the following process:

[0052] (1) Determine the fluid calculation control equations required for the calculation conditions; the fluid calculation control equations mainly include the continuity equation, momentum control equation, phase fraction equation, time transfer equation, state equation and energy equation, etc.

[0053] If there is no thermal effect in the multiphase flow problem being studied, the continuity equation, momentum control equation, phase fraction equation and time transport equation are mainly used.

[0054] If a three-phase flow problem is studied and there are thermal effects, the continuity equation, momentum control equation, phase fraction equation, time transfer equation, state equation and energy equation are mainly used to non-dimensionalize the control equations. Basic units such as speed, time and pressure are selected to further define new dimensionless variables. The dimensionless variables are substituted into the original control equations to convert the variables and parameters in the control equations into unitless forms, reduce the impact of different dimensions, and check the completeness and compatibility of the equations.

[0055] In this embodiment, a damping term is added to the momentum control equation and directly added to the control equation to enhance the numerical stability and physical credibility of the model and effectively control the excessive oscillation and non-physical behavior of the system. For example, the original momentum equation is:

[0056]

[0057] Where ρ is density, p is pressure, μ is the viscosity function characterized by time t and slurry temperature T, which can be obtained through experiments, g is the gravitational acceleration, F st is the surface tension.

[0058] After adding the damping term, the equation becomes:

[0059]

[0060] Where γ is the control damping strength, is the time derivative of the velocity vector.

[0061] (2) The boundary conditions, grouting velocity, grouting pressure, initial flow field of dynamic water, gravity field, slurry diffusion morphology, temperature field, slurry viscosity, constant volume specific heat, thermal conductivity and density are taken as input data. The input data is numerically simulated in the initial stage through the numerical simulation method to obtain the calculation results of slurry velocity, grouting pressure, phase fraction and viscosity parameters. The parameters of the last time step obtained from the numerical simulation in the initial stage are taken as initial data and input into the trained neural network model to obtain the prediction results of slurry velocity, grouting pressure, phase fraction and viscosity data.

[0062] In this embodiment, the neural network model is constructed based on the nonlinear mode decomposition algorithm (EDMD).

[0063] The loss function of the neural network model is:

[0064] L=L data +λ physics ·||R|| 2 +λ boundary ·L boundary ;

[0065] Among them, L data is the data fitting loss term, λ physics ·||R|| 2 is the physical residual loss term, λ physics To control the hyperparameters affecting the physical residual, ||R|| 2 is the square of the residual between the output of the neural network model and the calculation result of the fluid calculation control equation; L boundary is the boundary loss function term, L boundary A hyperparameter that balances the weight of the boundary condition loss in the total loss.

[0066] In this embodiment, a damping term is added to the momentum control equation and directly added to the control equation to enhance the numerical stability and physical credibility of the model and effectively control the excessive oscillation and non-physical behavior of the system. For example, the original momentum equation is:

[0067]

[0068] Where ρ is density, p is pressure, μ is the viscosity function characterized by time t and slurry temperature T, which can be obtained through experiments, g is the gravitational acceleration, F st is the surface tension.

[0069] After adding the damping term, the equation becomes:

[0070]

[0071] Where γ is the control damping strength, is the time derivative of the velocity vector.

[0072] The modified governing equations are used directly to define the physical residual part in the loss function of the neural network:

[0073]

[0074] The loss function is constructed as:

[0075] L=L data +λ physics ·||R|| 2

[0076] Among them, L data is the data fitting loss term, λ physics ·||R|| 2 is the physical residual loss term, λ physics To control the hyperparameters affected by physical residuals, which include but are not limited to residuals of equations such as conservation of momentum, conservation of energy, and conservation of mass, ||R|| 2 It is the square of the residual between the output of the neural network model and the calculation result of the fluid calculation control equation.

[0077] Data fitting loss

[0078] where y pred,i is the predicted value of the neural network at the i-th data point, y obs,i is the corresponding actual observed value, and N is the total number of data points.

[0079] Physical residual loss

[0080] Among them, R j is the residual of the physical control equation calculated based on the output of the neural network at the jth evaluation point, M is the total number of evaluation points, and λ physics It is a hyperparameter used to adjust the weight of the physical residual in the total loss. Similarly, this residual includes but is not limited to the residuals of equations such as conservation of momentum, conservation of energy, and conservation of mass.

[0081] Construct the boundary loss function term L boundary :

[0082] L boundary =∑ i∈boundary ||u pred,i -u boundary,i || 2 ;

[0083] Among them, u pred,i is the model prediction value at the boundary position, u boundary,i The values ​​specified for boundary conditions, where boundary conditions are known physical quantities or constraints that must be satisfied at the boundaries of a system or simulation region.

[0084] Through the above steps, physical constraints are embedded in the loss function, which constrains the solutions in the computational domain and on the boundary to follow physical laws.

[0085] In this embodiment, a parameter feature database is constructed by randomly sampling a numerical simulation case library, wherein the numerical simulation case library stores field data such as pressure field, velocity field and phase fraction field of grouting at different time and position points.

[0086] Data is extracted according to the set time step to obtain the data values ​​of flow field characteristic data (pressure, velocity, etc.) and spatiotemporal characteristic data (time span, spatial position), and the outliers in the data set are checked and corrected. Based on the EDMD algorithm, the nonlinear modal decomposition of the time-space analytical flow field of the sampled data is realized.

[0087] First, the Gaussian radial basis function (RBF) is selected as the dictionary function of EDMD, and the expression is:

[0088] φ(x)=exp(-||xc|| 2 / (2σ 2 ));

[0089] Where c is the center of the kernel and σ is the standard deviation.

[0090] Enter each data point into the RBF dictionary, for each point x in the dataset j , using all defined RBFs to calculate its eigenvectors:

[0091] ψ j(x j )=[φ1(x j ),φ2(x j ),…,φ k (x j )];

[0092] in, c k is the kth center point.

[0093] Combine the RBF eigenvectors of all points into a feature matrix ψ(X):

[0094]

[0095] Perform singular value decomposition on ψ(X):

[0096] Ψ(X)=U∑V * ;

[0097] Among them, U is the left singular vector, Σ is the singular value, and Σ is the right singular vector.

[0098] Construct an approximation A of the Koopman operator:

[0099] A=ψ(X′)VΣ -1 U * ;

[0100] Among them, X′ is the data matrix of the next time step, U * is the conjugate transposed matrix of U, which describes the row space characteristics of the data matrix. The column vectors constitute the orthogonal basis of the data row space. V is the right singular vector matrix, and its column vectors constitute the orthogonal basis of the data column space.

[0101] Perform eigenvalue decomposition on A, use numerical linear algebra libraries (such as NumPy, MATLAB, etc.) to perform eigenvalue decomposition on Koopman operator A, extract eigenvalues ​​and corresponding eigenvectors, and for each eigenvalue, calculate the modulus |λ| of the eigenvalue. If |λ|>1, the corresponding mode is growing, and if it is less than 1, the mode is decaying, indicating that the fluid is dynamically unstable. The system is optimized by adjusting the damping coefficient. In the grouting process, the damping coefficient is mainly used to describe the resistance or dissipation effect of the slurry during the flow and diffusion process, especially when the slurry passes through complex pores or cracks, the damping coefficient determines the rate of slurry flow decay. The damping coefficient can be related to the viscosity of the fluid, the formation characteristics, and the interaction between the slurry and the medium.

[0102] The instantaneous dynamic changes of the slurry on the time scale and the acquisition of structural and periodic dynamic features on the spatial scale are realized. In the process of capturing the flow field characteristics, the high-dimensional dilemma caused by random noise or non-critical features in the non-overfitting data is eliminated, and the dimension reduction of the data is realized. The extracted data is converted into a format suitable for machine learning training, such as a CSV file, and the data is standardized and converted into data with a mean of 0 and a standard deviation of 1. Based on the above steps, the extracted data is divided into a training set, a test set, and a comparison set according to the time ratio, and the neural network model is trained.

[0103] Perform pre-training tests, train the same case with different time steps, adjust the network weights and bias parameters according to the pre-training test results, and continue training until the training is completed.

[0104] The input data is numerically simulated in the initial stage through the numerical simulation method. The time span of the initial stage is dynamically adjusted according to the actual complexity of the working conditions. The factors considered include: grouting flow rate, grouting pressure, grouting flow rate, slurry selection, and geological conditions.

[0105]

[0106] X=α v m+α τ τ+α S S;

[0107] k=α u u+α p p+α f f+α d d;

[0108] Among them, T0 is the basic time span, α, β, γ, δ, α v , α τ , α S , α u , α p , α f , α d are weight coefficients, is the inverse of the grouting rate, is the inverse of the grouting pressure, χ is the slurry selection parameter, k is the geological condition complexity coefficient, m is the slurry viscosity, τ is the curing time, S is the chemical stability coefficient, u is the stability of the formation mechanical properties, p is the formation permeability, f is the degree of crack development, and d is the disaster-prone structure coefficient.

[0109] After determining the time span of the initial stage, set the time of the final stage to be consistent with the time span of the initial stage. Set the initial conditions, and the time step is the time span of the initial stage. After completing the numerical simulation, input the parameters of the last time step of the initial stage as the initial data into the trained neural network model to obtain the prediction results of slurry velocity, grouting pressure, phase fraction and viscosity data.

[0110] For the output results of the neural network model, 10% of the total time steps are set for numerical solution verification. A three-stage step-by-step trust verification mechanism is established to ensure the reliability and accuracy of the model; the details are as follows:

[0111] Assuming the total time steps are 100, the first 10 time steps are used for validation.

[0112] In the first stage, a fixed number a1 of time step nodes is used for preliminary verification. If the error of the preliminary verification result is controlled within 5%, the second stage is entered.

[0113] In the second stage, the trust in the model is enhanced by gradually reducing the verification frequency until the model output is fully trusted. Specifically, if the error between the model output and the numerical calculation is less than 5% within the set time step a2, the model output is considered to be credible and enters a fully trusted state.

[0114] In the state of full trust, no numerical verification is performed and subsequent simulation prediction is carried out directly. After the prediction is completed, the third stage is entered to conduct strict numerical verification of the prediction results of the last a3 time steps. If the error of the verification result is still within 5%, the grouting simulation result can be output; if it is inconsistent, the initial grouting conditions are re-adjusted and calculated.

[0115] The sum of the time steps of the three stages is 10% of the total time step, that is: a1+a2+a3=10.

[0116] Of course, for the specific error range requirement, this embodiment is set to 5%. In other implementations, it can also be set according to actual needs.

[0117] The numerical simulation results and neural network simulation results are collated, and the neural network parameters are tuned and optimized according to the verification results and actual engineering excavation findings.

[0118] Embodiment 2

[0119] In one or more embodiments, a grouting intelligent computing system based on coupling of neural network and numerical solution is disclosed, characterized in that it includes:

[0120] A fluid calculation control equation building module is used to determine the fluid calculation control equation required for the calculation condition; the fluid calculation control equation includes a momentum control equation, and a damping term is added to the momentum control equation;

[0121] The grouting intelligent calculation module is used to take the boundary conditions, grouting speed, grouting pressure, dynamic water initial flow field, gravity field, slurry diffusion morphology, temperature field, slurry viscosity, constant volume specific heat capacity, thermal conductivity and density as input data, perform numerical simulation of the input data in the initial stage through the numerical simulation method, take the parameters of the last time step obtained by the numerical simulation in the initial stage as the initial data, input them into the trained neural network model, and obtain the prediction results of slurry speed, grouting pressure, phase fraction and viscosity data;

[0122] Among them, the loss function of the neural network model includes a data fitting loss term, a physical residual loss term and a boundary loss function term; the momentum control equation with the damping term added is used as the physical residual loss term.

[0123] The specific implementation of the above modules is the same as that in Example 1 and will not be described in detail.

[0124] Embodiment 3

[0125] In one or more embodiments, a terminal device is disclosed, including a server, the server including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the grouting intelligent calculation method based on the coupling of neural network and numerical solution in embodiment 1 when executing the program. For the sake of brevity, the specific process is not described in detail.

[0126] It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0127] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0128] In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in a processor or an instruction in the form of software.

[0129] Embodiment 4

[0130] In one or more embodiments, a computer-readable storage medium is disclosed, in which a plurality of instructions are stored, wherein the instructions are suitable for being loaded by a processor of a terminal device and executing the grouting intelligent calculation method based on coupling of neural network and numerical solution described in Example 1.

[0131] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A grouting intelligent calculation method based on the coupling of neural network and numerical solution, characterized in that: include: Determine the fluid calculation control equations required for the calculation conditions; The fluid calculation control equation includes a momentum control equation, and a damping term is added to the momentum control equation; The boundary conditions, grouting velocity, grouting pressure, initial flow field of dynamic water, gravity field, slurry diffusion morphology, temperature field, slurry viscosity, constant volume specific heat capacity, thermal conductivity and density are used as input data. The input data is numerically simulated in the initial stage by numerical simulation method. The parameters of the last time step obtained by the initial stage numerical simulation are used as initial data and input into the trained neural network model to obtain the prediction results of slurry velocity, grouting pressure, phase fraction and viscosity data. Among them, the loss function of the neural network model includes a data fitting loss term, a physical residual loss term and a boundary loss function term; the momentum control equation with the damping term added is used as the physical residual loss term.

2. A grouting intelligent calculation method based on coupling of neural network and numerical solution as claimed in claim 1, characterized in that: It also includes: for the output results of the neural network model, setting a% of the total time steps for numerical solution verification; establishing a three-stage step-by-step trust verification mechanism, specifically: In the first stage, a fixed number of time step nodes are used to perform preliminary verification of numerical calculations. If the error of the verification result meets the requirements, the second stage is entered. Otherwise, the initial grouting condition parameters are adjusted and the grouting intelligent calculation is performed again. In the second stage, trust in the model is enhanced by gradually reducing the verification frequency until the model output is fully trusted. After full trust, simulation prediction is carried out, and after the prediction is completed, the third stage is entered; In the third stage, the prediction results of the last time step are verified by numerical calculation. If the error of the verification result meets the requirements, the grouting simulation result is output; otherwise, the initial grouting condition parameters are adjusted and the grouting intelligent calculation is performed again; where a is the set value.

3. The grouting intelligent calculation method based on the coupling of neural network and numerical solution as claimed in claim 1 is characterized in that: When the input data is numerically simulated in the initial stage, the time span of the initial stage is dynamically adjusted according to the actual complexity of the working conditions; specifically: X=a v m+a τ t+a S S; k=a u u+a p p+a f f+a d d; Among them, T0 is the basic time span, α, β, γ, δ, α v , α τ , α S , α u , α p , α f , α d are weight coefficients, is the inverse of the grouting rate, is the inverse of the grouting pressure, χ is the slurry selection parameter, k is the geological condition complexity coefficient, m is the slurry viscosity, τ is the curing time, S is the chemical stability coefficient, u is the stability of the formation mechanical properties, p is the formation permeability, f is the degree of crack development, and d is the disaster-prone structure coefficient.

4. The grouting intelligent calculation method based on coupling of neural network and numerical solution as claimed in claim 1, characterized in that: Add the damping term to the momentum control equation, specifically: Among them, ρ is density, p is pressure, and ρvv is the momentum flux tensor, that is, the rate of change of fluid momentum per unit volume; is the pressure gradient, which indicates the rate of change of pressure in the x, y, and z directions; is the velocity gradient, which indicates the velocity change rate of the fluid in each x, y, and z direction; μ(t,T) is the viscosity function characterized by time t and slurry temperature T, which can be obtained through experiments; g is the gravitational acceleration, and F st is the surface tension; γ is the control damping strength, is the damping term, where γ is the coefficient that controls the damping strength; is the gradient operator, and Ρv is the momentum density, that is, the momentum per unit volume.

5. The grouting intelligent calculation method based on coupling of neural network and numerical solution as claimed in claim 1, characterized in that: The loss function of the neural network model is specifically: L=L data +λ physics ·||R|| 2 +λ boundary ·L boundary ; Among them, L data is the data fitting loss term, λ physics ·||R|| 2 is the physical residual loss term, λ physics To control the hyperparameters affecting the physical residual, ||R|| 2 is the square of the residual between the output of the neural network model and the calculation result of the fluid calculation control equation; L boundary is the boundary loss function term, L boundary A hyperparameter that balances the weight of the boundary condition loss in the total loss.

6. The grouting intelligent calculation method based on coupling of neural network and numerical solution as claimed in claim 5, characterized in that: The data fitting loss term L data for: Among them, y pred,i is the predicted value of the neural network at the i-th data point, y obs,i is the corresponding actual observed value, and N is the total number of data points.

7. The grouting intelligent calculation method based on coupling of neural network and numerical solution as claimed in claim 5, characterized in that: The boundary loss function term L boundary for: L boundary =∑ i∈boundary ∥and pred,i -and boundary,i ∥ 2 ; Among them, u pred,i is the model prediction value at the boundary position, u boundary,i The value specified for the boundary condition.

8. A grouting intelligent computing system based on the coupling of neural network and numerical solution, characterized in that: include: Fluid calculation control equation building module, used to determine the fluid calculation control equation required for calculation conditions; The fluid calculation control equation includes a momentum control equation, and a damping term is added to the momentum control equation; The grouting intelligent calculation module is used to take the boundary conditions, grouting speed, grouting pressure, dynamic water initial flow field, gravity field, slurry diffusion morphology, temperature field, slurry viscosity, constant volume specific heat capacity, thermal conductivity and density as input data, perform numerical simulation of the input data in the initial stage through the numerical simulation method, take the parameters of the last time step obtained by the numerical simulation in the initial stage as the initial data, input them into the trained neural network model, and obtain the prediction results of slurry speed, grouting pressure, phase fraction and viscosity data; Among them, the loss function of the neural network model includes a data fitting loss term, a physical residual loss term and a boundary loss function term; the momentum control equation with the damping term added is used as the physical residual loss term.

9. A terminal device, comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images