Deep learning-based grouting slurry diffusion form characterization method and system

Abstract

The present application relates to the technical field of grouting simulation, in particular to a grouting slurry diffusion form characterization method and system based on deep learning, wherein the method comprises: obtaining diffusion form data of slurry under different geological conditions, and taking the preprocessed data as a training set; constructing a grouting diffusion characterization model, training the grouting diffusion characterization model based on the training set, and obtaining the trained grouting diffusion characterization model; the total loss function used in the training process is a weighted sum result of a data loss function, a physical constraint loss function, an edge value condition loss function, an initial value condition loss function and an actual experience loss function; inputting temperature and pressure data of a to-be-characterized grouting area into the trained grouting diffusion characterization model to obtain prediction values of grouting diffusion flow rate and grouting diffusion distance. The present application carries out model verification and assists in carrying out intelligent control of grouting according to the established model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of grouting simulation, in particular to a grouting slurry diffusion form characterization method and system based on deep learning. BACKGROUND

[0002] With the gradual increase of the scale and construction difficulty of tunnels and underground engineering, the geological environment through which the engineering construction area passes is becoming more and more complex, and more and more sudden disasters are encountered, often causing great property losses and casualties. The sudden gushing water disaster and the excavation collapse problem caused by broken rock mass seriously restrict the construction efficiency and quality of the project. As the most commonly used method for this type of disaster, grouting often uses slurry with rheological properties, which gradually undergoes solidification phase change with time, and can effectively plug sudden gushing water disasters and effectively reinforce broken rock mass. However, the complexity of the rock-soil medium of underground engineering makes the slurry diffusion process highly concealed, and its diffusion situation is difficult to be directly determined. Therefore, how to effectively characterize the form and parameter changes in the slurry diffusion process is very important for efficient treatment of grouting disasters.

[0003] Existing grouting diffusion is mostly grouting model test and numerical simulation research, but grouting test is difficult to effectively restore the grouting medium on site, and is often affected by monitoring instruments, making it difficult to obtain accurate and effective grouting process diffusion data. In terms of numerical simulation, targeted numerical calculation methods can achieve visual simulation of grouting diffusion, but existing grouting theories are difficult to consider simulation research under complex medium conditions, and the diffusion theory of slurry-water interaction is not perfect, and there is still a lack of efficient characterization method for grouting slurry diffusion process. SUMMARY

[0004] In order to solve the problems of the prior art, the present application provides a grouting slurry diffusion form characterization method and system based on deep learning. Grouting simulation test, field monitoring, numerical simulation data are collected, data preprocessing, necessary labeling and feature extraction are performed, and the processed data are divided into a training set and a validation set. A grouting diffusion characterization model is established, a neural network framework is constructed, a loss function is defined in the network considering data loss, physical constraint loss, boundary condition loss, initial value condition loss and actual experience loss, dynamic adjustment of each loss weight is set to accelerate the convergence of model training and improve the calculation efficiency, the model is verified and the established model is used to assist the intelligent control of grouting.

[0005] On the one hand, a grouting slurry diffusion form characterization method based on deep learning is provided, comprising:

[0006] Obtain the diffusion form data of the slurry under different geological conditions, preprocess the obtained data, and use the preprocessed data as a training set;

[0007] The grouting diffusion characterization model comprises a spatial convolution layer and a time convolution layer, the spatial convolution layer is used for extracting spatial features, and the time convolution layer is used for extracting time features.

[0008] The grouting diffusion characterization model is trained based on the training set to obtain a trained grouting diffusion characterization model, wherein a total loss function used in the training process is a weighted sum result of a data loss function, a physical constraint loss function, an edge value condition loss function, an initial value condition loss function and an actual experience loss function.

[0009] The temperature and pressure data of the grouting region to be characterized are input into the trained grouting diffusion characterization model to obtain predicted values of grouting diffusion flow rate and grouting diffusion distance.

[0010] In another aspect, a grouting slurry diffusion form characterization system based on deep learning is provided, comprising:

[0011] The acquisition module is configured to acquire diffusion form data of the slurry under different geological conditions, and pre-process the acquired data, and the pre-processed data is used as a training set.

[0012] The construction module is configured to construct a grouting diffusion characterization model, the grouting diffusion characterization model comprising a spatial convolution layer and a time convolution layer, the spatial convolution layer being used for extracting spatial features, and the time convolution layer being used for extracting time features.

[0013] The training module is configured to train the grouting diffusion characterization model based on the training set to obtain a trained grouting diffusion characterization model, wherein a total loss function used in the training process is a weighted sum result of a data loss function, a physical constraint loss function, an edge value condition loss function, an initial value condition loss function and an actual experience loss function.

[0014] The output module is configured to input the temperature and pressure data of the grouting region to be characterized into the trained grouting diffusion characterization model to obtain predicted values of grouting diffusion flow rate and grouting diffusion distance.

[0015] The above technical solution has the following advantages or beneficial effects:

[0016] The diffusion data of the slurry under different geological conditions are collected through laboratory tests, field monitoring and acquisition, numerical simulation of slurry injection is carried out as a supplement to the actual test data, the data are preprocessed, necessary labeling and feature extraction are carried out, and the processed data are divided into training set and verification set. A grouting diffusion representation model is established, a neural network framework including a spatial convolution layer and a time convolution layer is constructed, a nonlinear activation process is carried out, a loss function in the network is defined, which comprehensively considers data loss, physical constraint loss, boundary condition loss, initial value condition loss and actual experience loss, the influence of the comprehensive loss function and the physical constraint loss on the network training is considered, the dynamic adjustment of each loss weight is set, the neural network optimizes the specific target in different stages, the neural network weight is optimized by the Adam optimizer and the regularization processing is introduced, the adaptive time step dynamic adjustment is carried out, the convergence of model training is accelerated and the calculation efficiency is improved, the model is verified and the model is used to assist the intelligent control of grouting. BRIEF DESCRIPTION OF DRAWINGS

[0017] The accompanying drawings, which form a part of this specification, are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification. The embodiments of these drawings are set to explain the application, and do not constitute an improper limitation on the application.

[0018] Figure 1 The method flowchart of example one. DETAILED DESCRIPTION

[0019] It should be pointed out that the following detailed description is exemplary, and is intended to provide further explanation of the application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as generally understood by those skilled in the art to which the application belongs.

[0020] Example one, as shown in the figure, the embodiment provides a grouting slurry diffusion form representation method based on deep learning, which comprises: Figure 1

[0021] S101: obtaining the diffusion form data of the slurry under different geological conditions, preprocessing the obtained data, and taking the preprocessed data as the training set;

[0022] S102: constructing a grouting diffusion representation model, the grouting diffusion representation model comprising a spatial convolution layer and a time convolution layer, the spatial convolution layer being used for extracting spatial features; the time convolution layer being used for extracting time features;

[0023] S103: training the grouting diffusion representation model based on the training set, obtaining the trained grouting diffusion representation model; wherein the total loss function used in the training process is the weighted sum result of the data loss function, the physical constraint loss function, the boundary condition loss function, the initial value condition loss function and the actual experience loss function;​

[0024] S104: input the temperature and pressure data of the grouting area to be characterized into the trained grouting diffusion characterization model to obtain predicted values of grouting diffusion flow rate and grouting diffusion distance.

[0025] Further, the S101: obtaining grout diffusion morphology data under different geological conditions, the grout diffusion morphology data includes: two-dimensional grout diffusion image, three-dimensional grout diffusion image, grout diffusion pressure, grout flow rate and grout temperature change.

[0026] Further, the grout diffusion morphology data obtained under different geological conditions can be collected by laboratory test or collected on site.

[0027] Further, the obtained data is preprocessed, and the preprocessing includes grouting numerical simulation, data denoising, data standardization processing and data annotation processing.

[0028] It should be understood that grouting numerical simulation refers to simulation of the grouting process based on existing fluid mechanics numerical theory such as vof, double Euler method, etc. through existing calculation software.

[0029] Data denoising refers to denoising of the obtained image data by bilateral filtering.

[0030] Data standardization processing refers to uniting the data of diffusion pressure, grout flow rate and grout temperature change and performing decimal scaling standardization processing to reduce the data scale to a set range.

[0031] Data annotation processing refers to annotation and feature extraction of the set key points and abnormal points in the grouting process.

[0032] Further, the preprocessed data is used as a training set, and the training set includes diffusion pressure and grout flow rate of known grout temperature change in the grouting process.

[0033] It should be understood that the grout diffusion morphology data obtained under different geological conditions, including two-dimensional or three-dimensional diffusion image, pressure, flow rate and temperature change, is collected by laboratory test, field monitoring and grout diffusion data under the same working condition as the laboratory test and field monitoring geological conditions. Grouting numerical simulation is performed to supplement the actual test data. Data preprocessing includes cleaning data, removing noise, standardizing the diffusion morphology, necessary annotation and feature extraction, and determination of the coverage of the training data under enough initial conditions and boundary conditions. The data is divided into a training set and a validation set.

[0034] Further, S102: a grouting diffusion characterization model is constructed, the grouting diffusion characterization model comprises a spatial convolution layer and a time convolution layer, the spatial convolution layer is used for extracting spatial features; the time convolution layer is used for extracting time features; wherein the grouting diffusion characterization model is realized by using a three-dimensional convolution neural network; the network structure of the three-dimensional convolution neural network comprises, which are connected in sequence: an input layer, a spatial convolution layer, a time convolution layer, a full connection layer and an output layer.

[0035] Wherein based on the two-dimensional convolution of the original convolution neural network, the convolution operation in the time dimension is increased, such as increasing from the original two-dimensional 3*3 convolution to 3*3*3, wherein the 3*3*3 is decomposed into a spatial convolution layer of 1*3*3 and a time convolution layer of 3*1*1, and the time convolution layer of 3*1*1 is further decomposed into three convolution layers of 1*1, which correspond to learning the features of the previous time step, the next time step and the overlapping time steps of the previous and next time steps respectively.

[0036] Further, the working process of the spatial convolution layer comprises:

[0037] The value of the (x, y) position on the (β-1)th feature map of the (γ-1)th convolution layer Can be expressed as:

[0038]

[0039] Wherein is an activation function, b β,γ is the added bias, N is the total number of feature map inputs in the (β-1)th layer connected to the current feature map, is the value of the kernel at (φ, ψ), Φ and Ψ are the height and width of the convolution kernel respectively.

[0040] Further, the working process of the time convolution layer comprises:

[0041] Suppose a one-dimensional time sequence T is input, the convolution operation can be expressed as

[0042] O(t)=∑T(t,a)·W(a)

[0043] O(t) represents the output feature after convolution, T(t, a) represents the data in the time sequence, and W(a) represents the convolution weight value.

[0044] It should be understood that the grouting diffusion characterization model is established, the three-dimensional convolution neural network is constructed, the input is spatial position and time and initial boundary condition, the output is the diffusion state quantity of the grouting, including pressure, flow rate, temperature, and multiphase distribution of grouting water, the spatial convolution layer is used to extract spatial features, and the one-dimensional time convolution layer is used to extract time features. In the full connection layer, the output data in the convolution layer is nonlinearly transformed by introducing a nonlinear activation function.

[0045] Further, the S103: the total loss function is the weighted sum result of the data loss function, the physical constraint loss function, the boundary condition loss function, the initial value condition loss function and the actual experience loss function; wherein the formula expression of the total loss function is:

[0046] L t = α d (t s )L d + α p (t s )L p + α b (t s )L b + α i (t s )L i + α en (t s )L e

[0047] Wherein, L t is the total loss function, α d represents the weight of the data loss function, α p represents the weight of the physical constraint loss function, α b represents the weight of the boundary condition loss function, α i represents the weight of the initial value condition loss function, α e represents the weight of the actual experience loss function, t s represents the time step;

[0048] n represents different experience losses under different medium conditions. If there are multiple media acting together, the experience loss is recalculated by weighted average according to the distribution of the multiple media, which can be expressed as Wherein θ represents different media, γ θ represents the weight coefficient of each different medium.

[0049]

[0050] Wherein, m can correspond to five loss functions d, p, b, i and e respectively, and there is a floating threshold for the weight of each loss function, which is represented as α m (t s ) ∈ [m1, m2], α m (t s ) represents the loss weight at each time step, L M represents the sum of each loss function value at each time step, and L M = L d + Lp +L b +L i +L e .

[0051] Further, the data loss function is denoted as:

[0052]

[0053] where L d is the data loss function, N d is the number of training samples, denotes the corresponding true value, y j denotes the predicted value of the model, and j denotes the jth training sample.

[0054] Further, the physical constraint loss function is denoted as:

[0055]

[0056] where L p is the physical constraint loss function, N p is the number of physical constraint sampling points, denotes the differential operation at the sampling point, f o (x o , t o ) denotes the source term of the physical equation. For example, for the continuity equation its is its f o (x o , t o ) is 0, so

[0057] its is (expanded form of the continuity equation), where x, y, z are physical independent variables, are the velocity components in the x, y, z directions at each sampling point.

[0058] which can be expanded as where t o denotes the time step of the sampling point, and f o (x o , t o ) is 0, so

[0059] The physical constraint sampling points are sampled at different spatial positions using the numerical solution of the equation, focusing on selecting regions with large changes in physical parameters, such as boundary layers, turbulent flow regions, etc.

[0060] Further, the boundary condition loss function is represented as:

[0061]

[0062] where L b represents the boundary condition loss function, N b represents the number of boundary points, represents the boundary condition, v(x k , t k ) represents the predicted value on the boundary; such as for a boundary temperature that fluctuates with time, subject to boundary conditions at x = L, then

[0063] Further, the initial condition loss function is represented as:

[0064]

[0065] where L i represents the initial condition loss function, N i represents the number of initial condition sampling points, represents the initial condition function, w(x l , 0) represents the predicted value on the initial condition; such as for an initial boundary temperature, subject to boundary conditions at t = 0, then

[0066] Further, the actual experience loss function is represented as:

[0067]

[0068] where L e is the actual experience loss, N e is the actual experience loss constraint sampling point number, represents the miscible loss measure; ε(x q , t q ) is the void fraction change, which is the difference between the predicted and actual porosity at the sampling point, μ(x q , t q ) is the slurry viscosity change; v q is the velocity change, β is the adjustment coefficient, r is the pressure coefficient, is the pressure gradient, is the gradient operator.

[0069]

[0070] where T is the total time, c inis the slurry injection concentration, c out (t) is the predicted slurry concentration over time.

[0071] μ(x q ,t q )=∫0 T |μ yc (t)-μ sj (t)| 2 dt

[0072] where μ yc (t) is the predicted slurry viscosity over time, and μ sj (t) is the actually acquired slurry viscosity over time.

[0073] Considering the importance of physical equation constraints, the total loss function error and the physical constraint loss error in model training should be within the threshold, and a comprehensive judgment should be made.

[0074] Dynamic adjustment of different loss term weights is performed to enable the neural network to optimize specific targets in different stages. According to the diffusion of slurry in different media, physical constraints are gradually added to avoid the problem of affecting the convergence time due to convergence difficulties of physical constraints.

[0075] The neural network weights are optimized by the Adam optimizer to minimize the loss function and train the model. Regularization is used to prevent network overfitting, and gradient clipping is introduced to avoid gradient explosion.

[0076] Adaptive time step dynamic adjustment is performed, and in the model training process, the error value and gradient change rate of each time step are calculated. If the error is large and changes dramatically, the time step is reduced to capture more accurate states; otherwise, the time step is appropriately increased to improve computational efficiency.

[0077] The model is verified under different boundary conditions and data to ensure that its predicted results conform to physical laws. Small random disturbances are applied to the input data to verify the stability and robustness of the model's prediction results. At the same time, the performance of the model under different geological conditions and parameter changes is tested to ensure its applicability in a wide range of engineering scenarios.

[0078] The grouting diffusion characterization model predicts the diffusion state quantity of slurry in different media to assist in determining the optimal grouting position, pressure, and flow rate, and to monitor the grouting parameters in real time and use the grouting diffusion characterization model for prediction, dynamically adjusting the pressure and flow rate to achieve optimal grouting effect and realize intelligent grouting control.

[0079] Embodiment Two

[0080] The embodiment provides a grouting slurry diffusion form characterization system based on deep learning, which comprises:

[0081] An acquisition module is configured to acquire diffusion form data of the slurry under different geological conditions, pre-process the acquired data, and use the pre-processed data as a training set;

[0082] A construction module is configured to construct a grouting diffusion representation model, the grouting diffusion representation model including a spatial convolution layer and a time convolution layer, the spatial convolution layer being used to extract spatial features, and the time convolution layer being used to extract time features;

[0083] A training module is configured to train the grouting diffusion representation model based on the training set to obtain a trained grouting diffusion representation model, wherein a total loss function used in the training process is a weighted summation result of a data loss function, a physical constraint loss function, an edge value condition loss function, an initial value condition loss function, and an actual experience loss function;

[0084] An output module is configured to input temperature and pressure data of a to-be-represented grouting area into the trained grouting diffusion representation model to obtain predicted values of grouting diffusion flow rate and grouting diffusion distance.

[0085] The above merely describes preferred embodiments of the present application but should not be used to limit the present application. Various modifications and changes can be made by those skilled in the art based on the spirit and principles of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall fall within the protection scope of the present application.

Claims

1. A method for characterizing the diffusion pattern of grouting slurry based on deep learning, characterized in that, The method comprises the following steps: acquire the diffusion form data of the slurry under different geological conditions, pre-process the acquired data, and use the pre-processed data as a training set; construct a grouting diffusion representation model, the grouting diffusion representation model comprising a spatial convolution layer and a time convolution layer, the spatial convolution layer being configured to extract spatial features; the time convolution layer being configured to extract time features; The spatial convolutional layer's operation includes: the first... The first convolutional layer On a feature map, located at ( x, y The value at ) Represented as: ； wherein, is an activation function, is an added bias, N is the i-th feature map connected to the current feature map in the l-th layer, is the total number of feature map inputs in the l-th layer, is the value of the kernel at (i, j), , , , are the height and width of the convolution kernel, respectively; The working process of the time convolution layer includes: assuming that a one-dimensional time sequence T is input, and the convolution operation is represented as: ； represents an output feature after convolution, represents data in a time series, represents a convolution weight value; train the grouting diffusion representation model based on the training set to obtain a trained grouting diffusion representation model; wherein the total loss function used in the training process is a weighted sum of a data loss function, a physical constraint loss function, an edge value condition loss function, an initial value condition loss function, and an actual experience loss function; input the temperature and pressure data of the grouting area to be represented into the trained grouting diffusion representation model to obtain predicted values of the grouting diffusion flow rate and the grouting diffusion distance.

2. The deep learning-based grouting slurry diffusion pattern characterization method according to claim 1, characterized in that, construct a grouting diffusion representation model, the grouting diffusion representation model comprising a spatial convolution layer and a time convolution layer, the spatial convolution layer being configured to extract spatial features; the time convolution layer being configured to extract time features; wherein the grouting diffusion representation model is implemented by using a three-dimensional convolutional neural network; the network structure of the three-dimensional convolutional neural network comprises, in sequence, an input layer, a convolution layer, a spatial convolution layer, a time convolution layer, and an output layer.

3. The deep learning-based grouting slurry diffusion pattern characterization method according to claim 1, characterized in that, The total loss function is a weighted sum of a data loss function, a physical constraint loss function, an edge value condition loss function, an initial value condition loss function, and an actual experience loss function; wherein the formula expression of the total loss function is: ； wherein ， is the total loss function, is the weight of the data loss function, is the weight of the physical constraint loss function, is the weight of the boundary condition loss function, is the weight of the initial condition loss function, is the weight of the actual experience loss function, is the time step; n represents different experience losses under different medium conditions, wherein represents different media, represents the weight coefficient of the action of each different medium; ; respectively correspond to five loss functions d, p, b, i, e , there is a floating threshold value for the weight of each loss function, represented as , represents the loss weight at each time step, represents the sum of the loss function values at each time step, .

4. The deep learning-based grouting slurry diffusion pattern characterization method according to claim 1 or 3, characterized in that, The data loss function is represented as: ； in, For data loss function, The number of training samples. This represents the corresponding actual value. This represents the model's predicted value. Indicates the first One training sample; The physical constraint loss function is represented as: ； wherein, is the physical constraint loss function, is the number of physical constraint sampling points, denotes a differential operation at the sampling points, denotes a source term of the physical equation.

5. The deep learning-based grouting slurry diffusion pattern characterization method according to claim 1 or 3, characterized in that, The edge value condition loss function is represented as: ； wherein, represents the edge value condition loss function, represents the number of boundary points, represents the boundary condition, represents the predicted value on the boundary.

6. The deep learning-based grouting slurry diffusion pattern characterization method according to claim 1 or 3, characterized in that, The initial value condition loss function is represented as: ； wherein, represents the initial condition loss function, represents the initial condition sampling point number, represents the initial condition function, represents the predicted value on the initial condition.

7. The deep learning-based grouting slurry diffusion pattern characterization method according to claim 1 or 3, characterized in that, The actual experience loss function is represented as: ； wherein, is the actual experience loss, is the actual experience loss constraint sample point number, represents the miscibility loss measure; is the void fraction change, which is the difference between the predicted and actual void fraction at the sample point, is the slurry viscosity change; is the velocity change, is the adjustment coefficient, is the pressure coefficient, is the pressure gradient, is the gradient operator; ； where T is the total time, is the slurry injection concentration, is the predicted slurry concentration over time; ； wherein, is the predicted viscosity of the slurry as a function of time, is the actually acquired viscosity of the slurry as a function of time.

8. A grouting slurry diffusion pattern characterization system based on deep learning, characterized in that, The method comprises the following steps: an acquisition module configured to acquire the diffusion form data of the slurry under different geological conditions, pre-process the acquired data, and use the pre-processed data as a training set; a construction module configured to construct a grouting diffusion representation model, the grouting diffusion representation model comprising a spatial convolution layer and a time convolution layer, the spatial convolution layer being configured to extract spatial features; the time convolution layer being configured to extract time features; The spatial convolution layer, a working process includes: The value located at (x, y) on the first feature map of the convolution layer is represented as: x, y The value located at (x, y) on the first feature map of the convolution layer is represented as:​​ ； in, It is an activation function. It's an added bias. N It is the first one connected to the current feature map. The total number of feature map inputs in the layer. Is the core in ( , The value at ) , These are the height and width of the convolution kernel, respectively. The working process of the time convolution layer includes: assuming that a one-dimensional time sequence T is input, and the convolution operation is represented as: ； represents an output feature after convolution, represents data in a time series, represents a convolution weight value; a training module configured to train the grouting diffusion representation model based on the training set to obtain a trained grouting diffusion representation model; wherein the total loss function used in the training process is a weighted sum of a data loss function, a physical constraint loss function, an edge value condition loss function, an initial value condition loss function, and an actual experience loss function; an output module configured to input the temperature and pressure data of the grouting area to be represented into the trained grouting diffusion representation model to obtain predicted values of the grouting diffusion flow rate and the grouting diffusion distance.

Images