Grouting diffusion prediction method, equipment and medium
Through the combination of COMSOL Multiphysics software and neural network model, the problems of high cost, low accuracy and narrow application range of grouting prediction technology are solved, and efficient and accurate grouting effect prediction and wide application are achieved.
Patent Information
- Application Number
- CN202411394839.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-08
- Publication Date
- 2024-12-27
AI Technical Summary
The existing grouting prediction technology has high cost, low accuracy and narrow application range, making it difficult to accurately evaluate the grouting and grouting quality of solid waste landfills.
The COMSOL Multiphysics software is used to perform numerical simulation of slurry diffusion, and combined with neural network models, a variety of neural network models are constructed for training and verification, and the optimal model is selected for grouting diffusion prediction.
It realizes efficient and accurate grouting effect prediction, reduces the cost of initial data acquisition, and expands the application scope of grouting prediction.
Smart Images
Figure CN119203771A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of grouting diffusion, and in particular to a grouting diffusion prediction method, equipment and medium. Background Art
[0002] In-situ grouting repair at the bottom of solid waste landfills is a highly concealed underground construction project. Due to limited detection methods, it is difficult for construction workers to accurately evaluate the repair effect and grouting quality. Therefore, researchers generally rely on theoretical physical model derivation to analyze the factors affecting grouting diffusion and conduct verification experiments. However, the simplification of theoretical physical models leads to poor diffusion prediction results, and verification experiments are time-consuming and laborious. Therefore, computational fluid dynamics (CFD) technology has begun to be used to simulate grouting to solve the fluid flow problem during the grouting process. However, as the number of factors considered in the grouting simulation process increases, more computing modules need to be introduced, such as Darcy's law, multiphase flow in porous media, fluid-solid coupling, and slurry-soil chemical reactions. The complexity and running time of the model have increased significantly, and the model accuracy requirements have continued to increase, which has exceeded the processing capabilities of non-professional computers, especially in the field of ultra-large grouting simulation.
[0003] With the rise of neural networks, methods for predicting grouting diffusion using neural networks have emerged in the prior art. Patent CN114626287A developed an artificial neural network to predict grouting construction parameters based on formation parameters, but the required data is of various types, requiring a large amount of actual engineering data support and limited scope of application. Patent CN113123330A uses neural networks to predict grouting diffusion effects based on a large number of construction parameters and formation parameters, but it is only applicable to specific working conditions and requires actual engineering support.
[0004] In order to solve the above problems, the present invention proposes a grouting diffusion prediction method based on neural network, and uses simulation data to perform neural network prediction of grouting diffusion. Summary of the invention
[0005] In order to solve the problems of high cost, low precision and narrow application range of existing grouting prediction technology, the present invention provides a grouting diffusion prediction method, which numerically simulates the slurry diffusion process through COMSOL Multiphysics software and combines neural network with COMSOL model, effectively solving the problems of high cost, low precision and narrow application range of existing grouting prediction technology.
[0006] In a first aspect, the present invention provides a grouting diffusion prediction method, comprising the following steps:
[0007] Construct a COMSOL grouting model;
[0008] Initializing the COMSOL grouting model and meshing the COMSOL grouting model;
[0009] Performing grouting simulation based on the COMSOL model to obtain a first slurry volume fraction, wherein the slurry volume fraction is a ratio of the volume of the slurry to the total volume of the slurry and the soil pores;
[0010] Based on the COMSOL grouting model, all slurry volume fractions within the grouting range are derived at a fixed acquisition frequency, and the slurry volume fractions are normalized to obtain a first data set, wherein the first data set includes grouting inlet pressure data, grouting inlet flow data, soil density data, time data t, and spatial position coordinate data (x, y), wherein x is the abscissa of the slurry in the COMSOL grouting model, and y is the ordinate of the slurry in the COMSOL grouting model;
[0011] Dividing the first data set into multiple training sets and multiple test sets;
[0012] Construct multiple neural network models, use training sets for training, and obtain multiple trained neural network models;
[0013] Inputting the test set into a plurality of the trained neural network models to obtain the slurry volume fraction corresponding to each of the trained neural network models;
[0014] Calculate the mean absolute error value and the root mean square error value between the slurry volume fraction corresponding to each of the trained neural network models and the first slurry volume fraction;
[0015] Obtaining an optimal neural network model based on the mean absolute error value and the root mean square error value;
[0016] Get the data set to be tested;
[0017] Inputting the test data set into the optimal neural network model to obtain a second slurry volume fraction;
[0018] The grouting diffusion condition is obtained based on the second slurry volume fraction.
[0019] In some embodiments, the constructing of multiple neural network models and using the training set for training to obtain multiple trained neural network models includes:
[0020] Constructing an RBF neural network model, a first BP neural network model, a second BP neural network model and a third BP neural network model; wherein,
[0021] The radial basis function of the RBF neural network model uses a Gaussian function;
[0022] The first BP neural network model, the second BP neural network model and the third BP neural network model use an input layer, an output layer and a hidden layer as the structure of the model, the output layer uses a linear activation function, the hidden layer uses a Tanh activation function, and the number of nodes in the hidden layer is m is the number of neurons in the input layer, n is the number of neurons in the output layer, and a is a constant between 1 and 10;
[0023] The RBF neural network model, the first BP neural network model, the second BP neural network model and the third BP neural network model are trained using the training set to obtain a trained RBF neural network model, a trained first BP neural network model, a trained second BP neural network model and a trained third BP neural network model.
[0024] In some embodiments, the step of inputting the test set into a plurality of the trained neural network models to obtain a plurality of slurry volume fractions comprises:
[0025] Inputting the test set into the trained RBF neural network model to obtain the slurry volume fraction corresponding to the trained RBF neural network model;
[0026] Inputting the test set into the trained first BP neural network model to obtain the slurry volume fraction corresponding to the trained first BP neural network model;
[0027] Inputting the test set into the trained second BP neural network model to obtain the slurry volume fraction corresponding to the trained second BP neural network model;
[0028] The test set is input into the trained third BP neural network model to obtain the slurry volume fraction corresponding to the trained third BP neural network model.
[0029] In some embodiments, inputting the test set into the trained RBF neural network model to obtain the slurry volume fraction corresponding to the trained RBF neural network model includes:
[0030] The grouting inlet pressure data, grouting inlet flow data, soil density data, time data t and spatial position coordinate data (x, y) are input into the trained RBF neural network model to obtain c 1 (x, y, t), where c 1 (x, y, t) is the slurry volume fraction at the spatial position coordinate (x, y) at time t output by the trained RBF neural network model.
[0031] In some embodiments, inputting the test set into the trained first BP neural network model to obtain the slurry volume fraction corresponding to the trained first BP neural network model includes:
[0032] The grouting inlet pressure data, grouting inlet flow data, soil density data and time data t are input into the trained first BP neural network model to obtain c(t), wherein c(t) is the slurry volume fraction at all spatial positions at time t.
[0033] In some embodiments, inputting the test set into the trained second BP neural network model to obtain the slurry volume fraction corresponding to the trained second BP neural network model includes:
[0034] The grouting inlet pressure data, grouting inlet flow data, soil density data and spatial position coordinate data (x, y) are input into the trained second BP neural network model to obtain c(x, y), wherein c(x, y) is the slurry volume fraction at the spatial position coordinates (x, y) at all times.
[0035] In some embodiments, inputting the test set into the trained third BP neural network model to obtain the slurry volume fraction corresponding to the trained third BP neural network model includes:
[0036] The grouting inlet pressure data, grouting inlet flow data, soil density data, time data t and spatial position coordinate data (x, y) are input into the trained third BP neural network model to obtain c 2 (x, y, t), where c 2 (x, y, t) is the slurry volume fraction at the spatial coordinate position (x, y) at time t output by the trained third BP neural network model.
[0037] In some embodiments, the mean absolute error value and the root mean square error value between the slurry volume fraction corresponding to each of the trained neural network models and the first slurry volume fraction include: a first mean absolute error value and a first root mean square error value between the first slurry volume fraction and the slurry volume fraction corresponding to the trained RBF neural network model, a second mean absolute error value and a second root mean square error value between the first slurry volume fraction and the slurry volume fraction corresponding to the trained first BP neural network model, a third mean absolute error value and a third root mean square error value between the first slurry volume fraction and the slurry volume fraction corresponding to the trained second BP neural network model, and a fourth mean absolute error value and a fourth root mean square error value between the first slurry volume fraction and the slurry volume fraction corresponding to the trained third BP neural network model;
[0038] The optimal neural network model is obtained based on the multiple mean absolute error values and the multiple root mean square error values, comprising:
[0039] Compare the first mean absolute error value, the second mean absolute error value, the third mean absolute error value and the fourth mean absolute error value, and select the neural network model with the smallest mean absolute error value;
[0040] Comparing the first root mean square error value, the second root mean square error value, the third root mean square error value and the fourth root mean square error value, and selecting a neural network model with a minimum root mean square error value;
[0041] If the neural network model with the smallest mean absolute error value and the neural network model with the smallest root mean square error value are the same neural network model, then the neural network model is the optimal neural network model;
[0042] If the neural network model with the smallest mean absolute error value and the neural network model with the smallest root mean square error value are different neural network models, the neural network model with the smallest root mean square error value is selected as the optimal neural network model.
[0043] In a second aspect, the present invention proposes an electronic device, comprising a processor and a memory, wherein the memory stores a program executable on the processor, and when the program is executed by the processor, the grouting diffusion prediction method described in the first aspect is implemented.
[0044] In a third aspect, the present invention proposes a storage medium having a computer program stored thereon, and when the computer program is executed by at least one processor, the grouting diffusion prediction method described in the first aspect is implemented.
[0045] In order to solve the problems of high cost, low accuracy and narrow application range of existing grouting prediction technology, the present invention has the following advantages:
[0046] The grouting prediction method provided by the present invention can efficiently and accurately predict the grouting effect, intuitively observe the spatiotemporal distribution change of the volume fraction during the slurry diffusion process, and greatly reduce the cost of obtaining the initial data for grouting prediction. It can be widely used in various scenarios in the grouting field. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] Figure 1 A flow chart of a grouting diffusion prediction method is shown;
[0048] Figure 2 A schematic diagram of the meshing results of the COMSOL grouting model is shown;
[0049] Figure 3 The slurry volume fraction distribution thermogram corresponding to the first BP neural network model is shown;
[0050] Figure 4 The slurry volume fraction distribution thermogram corresponding to the second BP neural network model is shown;
[0051] Figure 5 The slurry volume fraction distribution heat map corresponding to the third BP neural network model is shown
[0052] Figure 6 The slurry volume fraction distribution thermogram corresponding to the RBF neural network model is shown;
[0053] Figure 7 The figure shows the thermodynamic diagram of the slurry volume fraction distribution obtained by simulating the COMSOL grouting model. DETAILED DESCRIPTION
[0054] The present disclosure will now be discussed with reference to several exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and thus implement the present disclosure, rather than implying any limitation on the scope of the present disclosure.
[0055] As used herein, the term "including" and its variants are to be interpreted as open-ended terms meaning "including but not limited to". The term "based on" is to be interpreted as "based at least in part on". The terms "one embodiment" and "an embodiment" are to be interpreted as "at least one embodiment". The term "another embodiment" is to be interpreted as "at least one other embodiment". The orientation or position relationship indicated by the terms "upper", "lower", "left", "right", "front", "back", "top", "bottom", "inner", "outer", "vertical", "horizontal", "lateral", "longitudinal" and the like is based on the orientation or position relationship shown in the accompanying drawings. These terms are mainly for better describing the present application and its embodiments, and are not used to limit the indicated devices, elements or components to have a specific orientation, or to be constructed and operated in a specific orientation. In addition, in addition to being used to indicate an orientation or position relationship, some of the above terms may also be used to indicate other meanings, such as the term "upper" may also be used to indicate a certain dependency or connection relationship in some cases. For those of ordinary skill in the art, the specific meanings of these terms in this application can be understood according to the specific circumstances. In addition, the terms "install", "set", "provided with", "connected" and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral structure; it can be a mechanical connection, or an electrical connection; it can be directly connected, or indirectly connected through an intermediate medium, or it can be an internal connection between two devices, elements or components. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood according to the specific circumstances. In addition, the terms "first", "second", etc. are mainly used to distinguish different devices, elements or components (the specific types and structures may be the same or different), and are not used to indicate or imply the relative importance and quantity of the indicated devices, elements or components. Unless otherwise specified, "plurality" means two or more.
[0056] This embodiment discloses a grouting diffusion prediction method. Figure 1 As shown, the following steps are included:
[0057] Construct a COMSOL grouting model;
[0058] Initializing the COMSOL grouting model and meshing the COMSOL grouting model;
[0059] Performing grouting simulation based on the COMSOL model to obtain a first slurry volume fraction, wherein the slurry volume fraction is a ratio of the volume of the slurry to the total volume of the slurry and the soil pores;
[0060] Based on the COMSOL grouting model, all slurry volume fractions within the grouting range are derived at a fixed acquisition frequency, and the slurry volume fractions are normalized to obtain a first data set, wherein the first data set includes grouting inlet pressure data, grouting inlet flow data, soil density data, time data t, and spatial position coordinate data (x, y), wherein x is the abscissa of the slurry in the COMSOL grouting model, and y is the ordinate of the slurry in the COMSOL grouting model;
[0061] Dividing the first data set into multiple training sets and multiple test sets;
[0062] Construct multiple neural network models, use training sets for training, and obtain multiple trained neural network models;
[0063] Inputting the test set into a plurality of the trained neural network models to obtain the slurry volume fraction corresponding to each of the trained neural network models;
[0064] Calculate the mean absolute error value and the root mean square error value between the slurry volume fraction corresponding to each of the trained neural network models and the first slurry volume fraction;
[0065] Obtaining an optimal neural network model based on the mean absolute error value and the root mean square error value;
[0066] Get the data set to be tested;
[0067] Inputting the test data set into the optimal neural network model to obtain a second slurry volume fraction;
[0068] The grouting diffusion condition is obtained based on the second slurry volume fraction.
[0069] In this embodiment, a grouting diffusion prediction method is proposed. First, a COMSOL grouting model is constructed using COMSOL Multiphysics simulation software, and a grouting simulation is performed after the model is initialized and meshed, and a first slurry volume fraction is obtained through the grouting simulation; in addition, during the grouting simulation, all slurry volume fractions within the grouting range are derived at a fixed acquisition frequency, and the derived slurry volume fraction data are normalized using the Mapminmax function in Matlab to obtain a first data set, and the first data set is divided into a training set and a test set. Secondly, multiple neural network models are constructed, and the training set in the first data set is used to train multiple neural network models to obtain multiple trained neural network models, and then the test set in the first data set is input into the trained neural network model to obtain the slurry volume fraction corresponding to each of the trained neural network models. Thirdly, the mean absolute error value and the root mean square error value between the slurry volume fraction corresponding to each of the trained neural network models and the first slurry volume fraction are calculated, and the optimal neural network model is selected from multiple trained neural network models based on the mean absolute error value and the root mean square error value. Finally, the data set to be tested is obtained, and the data set to be predicted is input into the optimal neural network model to obtain the second slurry volume fraction. The grouting diffusion situation can be understood based on the second slurry volume fraction.
[0070] Specifically, COMSOL Multiphysics is a multi-physics field coupled CFD simulation software, in which information such as flow rate and pressure are stored and calculated in the form of digital matrices, which can be used as a data set for neural network training. When constructing the COMSOL grouting model, according to the experimental results of the in-situ grouting repair model of solid waste landfill leakage, COMSOL Multiphysics was used to construct a two-dimensional multi-physics field simulation model of single-hole horizontal jet grouting. The model is circular, and the radius of the circle is determined according to the actual scene. The grouting inlet is located at the center of the circle. Among them, the in-situ grouting repair model experiment of leakage is a comprehensive experimental process, which aims to evaluate the effect of grouting repair technology and optimize the grouting process parameters by simulating actual leakage conditions.
[0071] Specifically, Figure 2 As shown in the figure, the model is initialized with inlet conditions and meshed, and the transient calculation method is used. When grouting simulation is performed, the multiphase flow module of fluid flow is used for calculation, and the coordinate parameters and time step are set. The physical field interface selects Euler-Euler model turbulence. Through meshing, the simulation accuracy can be improved, the computing resources can be optimized, the requirements of different physical fields can be adapted, and efficient simulation can be achieved.
[0072] Specifically, based on the COMSOL grouting model, all slurry liquid integral values within the grouting range are derived at a fixed acquisition frequency, and then the Mapminmax function is used in Matlab to normalize the data to obtain the first data set. Obtaining the data set through the COMSOL grouting model can effectively reduce the cost of data acquisition, while also ensuring the accuracy of the data.
[0073] Specifically, the first data set includes grouting inlet pressure p, grouting inlet flow q, soil density ρ, time t and spatial position coordinates (x, y), where x is the horizontal coordinate of the slurry in the COMSOL grouting model, and y is the vertical coordinate of the slurry in the COMSOL grouting model. Different input parameters are used in different neural network models.
[0074] Specifically, the purpose of constructing multiple neural network models is to select an optimal neural network model to make the prediction result of the present invention more accurate.
[0075] Specifically, when training multiple neural network models, different input parameters are selected in the training set of the first data set according to the differences in the neural network models to achieve a better comparison effect.
[0076] Specifically, the test set is input into the trained neural network model to obtain the slurry volume fraction corresponding to each of the trained neural network models, these slurry volume fractions are calculated with the first slurry volume fraction to obtain the mean absolute error value and the root mean square error value between the two, the smallest error value is selected, and the corresponding neural network model is the optimal neural network model, and then the test data set is input into the optimal neural network model for prediction to obtain the second slurry volume fraction.
[0077] Specifically, the slurry volume fraction is the ratio of the volume of the slurry to the total volume of the slurry and the soil pores. Through the slurry volume fraction, we can know whether there is slurry flowing through a certain spatial coordinate position at time t, and if so, what is the slurry concentration at the spatial coordinate position at time t. For example: if the predicted second slurry volume fraction is 0, it means that there is no slurry flowing through the spatial coordinate position at this moment; if the second slurry volume fraction is 0.5, it means that the volume of slurry in the area occupies 1 / 3 of the area. Through the slurry volume fractions at all times and all positions, we can get the slurry volume fraction distribution thermodynamic map, which can more intuitively observe the entire grouting process.
[0078] In some embodiments, the constructing of multiple neural network models and using the training set for training to obtain multiple trained neural network models includes:
[0079] Constructing an RBF neural network model, a first BP neural network model, a second BP neural network model and a third BP neural network model; wherein,
[0080] The radial basis function of the RBF neural network model uses a Gaussian function;
[0081] The first BP neural network model, the second BP neural network model and the third BP neural network model use an input layer, an output layer and a hidden layer as the structure of the model, the output layer uses a linear activation function, the hidden layer uses a Tanh activation function, and the number of nodes in the hidden layer is m is the number of neurons in the input layer, n is the number of neurons in the output layer, and a is a constant between 1 and 10;
[0082] The RBF neural network model, the first BP neural network model, the second BP neural network model and the third BP neural network model are trained using the training set to obtain a trained RBF neural network model, a trained first BP neural network model, a trained second BP neural network model and a trained third BP neural network model.
[0083] In this embodiment, four neural network models are proposed, among which the first BP neural network model, the second BP neural network model and the third BP neural network model are different in that their input parameters and output results are inconsistent, and the number of hidden layer nodes is also inconsistent. The number of hidden layer nodes of the three is m is the number of neurons in the input layer, n is the number of neurons in the output layer, and a is a constant between 1 and 10. The output layers of the three use linear activation functions, and the hidden layers use Tanh activation functions.
[0084] Specifically, the BP neural network model, also known as the backpropagation neural network (Backpropagation Neural Network), is a feedforward artificial neural network model. Its core is to use the error backpropagation algorithm to adjust and optimize the network weights. It has strong nonlinear mapping and generalization capabilities, and can automatically extract the features of input data and perform pattern recognition through learning and training.
[0085] Specifically, the Tanh activation function (hyperbolic tangent function) is an important nonlinear activation function in deep learning. Its output range is between -1 and 1, and it has strong representation ability. And because its output has a zero-centered property, it helps to reduce the problem of gradient disappearance.
[0086] In this example, an RBF neural network model is also proposed. The RBF (Radial Basis Function) neural network model is a neural network based on radial basis function. It has outstanding performance in dealing with nonlinear problems, function approximation, pattern recognition and data classification. The RBF neural network realizes nonlinear mapping from input to output through radial basis function, can approximate any nonlinear function, and has strong nonlinear mapping ability.
[0087] Specifically, by using the training set to train the four neural network models respectively, four trained neural network models can be obtained, namely: a trained RBF neural network model, a trained first BP neural network model, a trained second BP neural network model and a trained third BP neural network model.
[0088] Specifically, during the training process, the Adam optimizer was used and the learning rate was set to 0.01. The training was completed after 200 Epochs reached convergence.
[0089] In some embodiments, the step of inputting the test set into a plurality of the trained neural network models to obtain a plurality of slurry volume fractions comprises:
[0090] Inputting the test set into the trained RBF neural network model to obtain the slurry volume fraction corresponding to the trained RBF neural network model;
[0091] Inputting the test set into the trained first BP neural network model to obtain the slurry volume fraction corresponding to the trained first BP neural network model;
[0092] Inputting the test set into the trained second BP neural network model to obtain the slurry volume fraction corresponding to the trained second BP neural network model;
[0093] The test set is input into the trained third BP neural network model to obtain the slurry volume fraction corresponding to the trained third BP neural network model.
[0094] In some embodiments, inputting the test set into the trained RBF neural network model to obtain the slurry volume fraction corresponding to the trained RBF neural network model includes:
[0095] The grouting inlet pressure data, grouting inlet flow data, soil density data, time data t and spatial position coordinate data (x, y) are input into the trained RBF neural network model to obtain c 1 (x, y, t), where c 1(x, y, t) is the slurry volume fraction at the spatial position coordinate (x, y) at time t output by the trained RBF neural network model.
[0096] In some embodiments, inputting the test set into the trained first BP neural network model to obtain the slurry volume fraction corresponding to the trained first BP neural network model includes:
[0097] The grouting inlet pressure data, grouting inlet flow data, soil density data and time data t are input into the trained first BP neural network model to obtain c(t), wherein c(t) is the slurry volume fraction at all spatial positions at time t.
[0098] In some embodiments, inputting the test set into the trained second BP neural network model to obtain the slurry volume fraction corresponding to the trained second BP neural network model includes:
[0099] The grouting inlet pressure data, grouting inlet flow data, soil density data and spatial position coordinate data (x, y) are input into the trained second BP neural network model to obtain c(x, y), wherein c(x, y) is the slurry volume fraction at the spatial position coordinates (x, y) at all times.
[0100] In some embodiments, inputting the test set into the trained third BP neural network model to obtain the slurry volume fraction corresponding to the trained third BP neural network model includes:
[0101] The grouting inlet pressure data, grouting inlet flow data, soil density data, time data t and spatial position coordinate data (x, y) are input into the trained third BP neural network model to obtain c 2 (x, y, t), where c 2 (x, y, t) is the slurry volume fraction at the spatial coordinate position (x, y) at time t output by the trained third BP neural network model.
[0102] In this embodiment, the test set is input into each trained neural network to obtain its corresponding slurry volume fraction.
[0103] Specifically, for the RBF neural network model, its input parameters include: grouting inlet pressure data p, grouting inlet flow data q, soil density data ρ, time data t and spatial position coordinate data (x, y) in the test set, and the output result is c 1 (x, y, t). 1(x, y, t) represents the slurry volume fraction at the spatial position coordinate (x, y) at time t output by the trained RBF neural network model, which is used for comparison with the first slurry volume fraction.
[0104] Specifically, for the first BP neural network model, its input parameters include: grouting inlet pressure data p, grouting inlet flow data q, soil density data ρ and time data t in the test set, and its output result is c(t). c(t) represents the slurry volume fraction at all spatial positions at time t, which is used for comparison with the first slurry volume fraction.
[0105] Specifically, for the second BP neural network model, its input parameters include: grouting inlet pressure data p, grouting inlet flow data q, soil density data ρ and spatial position coordinate data (x, y), and its output result is c(x, y). c(x, y) represents the slurry volume fraction at the spatial position coordinate (x, y) at all times, which is used for comparison with the first slurry volume fraction.
[0106] Specifically, for the third BP neural network model, its input parameters include: grouting inlet pressure data p, grouting inlet flow data q, soil density data ρ, time data t and spatial position coordinate data (x, y), and its output result is c 2 (x, y, t). 2 (x, y, t) represents the slurry volume fraction at the spatial coordinate position (x, y) at time t output by the trained third BP neural network model, which is used for comparison with the first slurry volume fraction.
[0107] Specifically, due to the different neural network models used, the input parameters and output results are also different, but a corresponding slurry volume fraction can be obtained through the results.
[0108] Specifically, comparing the slurry volume fractions predicted by multiple neural network models with the first slurry volume fraction is to obtain a better neural network model to ensure the accuracy of the final prediction result.
[0109] In some embodiments, the mean absolute error value and the root mean square error value between the slurry volume fraction corresponding to each of the trained neural network models and the first slurry volume fraction include: a first mean absolute error value and a first root mean square error value between the first slurry volume fraction and the slurry volume fraction corresponding to the trained RBF neural network model, a second mean absolute error value and a second root mean square error value between the first slurry volume fraction and the slurry volume fraction corresponding to the trained first BP neural network model, a third mean absolute error value and a third root mean square error value between the first slurry volume fraction and the slurry volume fraction corresponding to the trained second BP neural network model, and a fourth mean absolute error value and a fourth root mean square error value between the first slurry volume fraction and the slurry volume fraction corresponding to the trained third BP neural network model;
[0110] The optimal neural network model is obtained based on the multiple mean absolute error values and the multiple root mean square error values, comprising:
[0111] Compare the first mean absolute error value, the second mean absolute error value, the third mean absolute error value and the fourth mean absolute error value, and select the neural network model with the smallest mean absolute error value;
[0112] Comparing the first root mean square error value, the second root mean square error value, the third root mean square error value and the fourth root mean square error value, and selecting a neural network model with a minimum root mean square error value;
[0113] If the neural network model with the smallest mean absolute error value and the neural network model with the smallest root mean square error value are the same neural network model, then the neural network model is the optimal neural network model;
[0114] If the neural network model with the smallest mean absolute error value and the neural network model with the smallest root mean square error value are different neural network models, the neural network model with the smallest root mean square error value is selected as the optimal neural network model.
[0115] In this embodiment, the neural network model with the smallest error is compared based on the mean absolute error and the root mean square error between the slurry volume fraction corresponding to each trained neural network model and the first slurry volume fraction, that is, the optimal neural network model.
[0116] Specifically, when comparing, the mean absolute error value and the root mean square error value are compared separately, that is, the first mean absolute error value, the second mean absolute error value, the third mean absolute error value and the fourth mean absolute error value are compared, and the first root mean square error value, the second root mean square error value, the third root mean square error value and the fourth root mean square error value are compared. The neural network model with the smallest error value is selected as the optimal neural network model. For example: the first mean absolute error value, the second mean absolute error value, the third mean absolute error value and the fourth mean absolute error value are respectively: 4.74e-03, 2.56e-02, 1.72e-02 and 3.56e-03, and the third mean absolute error value can be obtained to be the smallest. For another example: the first root mean square error value, the second root mean square error value, the third root mean square error value and the fourth root mean square error value are respectively: 7.83e-03, 7.21e-02, 2.53e-02 and 6.12e-03, and the third root mean square error value can be obtained to be the smallest. Therefore, the third BP neural network model is finally selected as the optimal neural network model.
[0117] Specifically, when the neural network model corresponding to the minimum value of the mean absolute error value and the minimum value of the root mean square error value are not the same neural network model, the neural network model corresponding to the minimum value of the root mean square error value is taken as the optimal neural network model.
[0118] Specifically, in practical applications, the slurry volume fraction corresponding to each neural network model can be used to obtain the respective slurry volume fraction distribution thermograms, and the optimal neural network model can be selected by observing the spatiotemporal distribution of the thermograms and the thermograms obtained by the first slurry volume fraction. Figure 7 As shown in the figure, the grouting process simulated by the COMSOL grouting model is shown. The slurry is ejected horizontally from the grouting inlet, and gradually spreads in a fan shape over time, extending in the ejection direction. Its temporal and spatial distribution is similar to that of classic jet grouting. The slurry volume fraction gradually decreases from the grouting port to the horizontal direction. By comparison Figures 3 to 6 From the slurry volume fraction heat maps of each neural network model shown, it can be seen that the slurry volume fraction distribution heat map of the third BP neural network model is closest to the heat map obtained by the first slurry volume fraction, and their spatiotemporal distribution is the most similar. Therefore, the third BP neural network model is the optimal neural network model.
[0119] In a second aspect, the present invention proposes an electronic device, comprising a processor and a memory, wherein the memory stores a program executable on the processor, and when the program is executed by the processor, the grouting diffusion prediction method described in the first aspect is implemented.
[0120] In a third aspect, the present invention proposes a storage medium having a computer program stored thereon, and when the computer program is executed by at least one processor, the grouting diffusion prediction method described in the first aspect is implemented.
[0121] In summary, the prediction method of the present invention can greatly reduce the cost of developing the original simulation program, shorten the research and development cycle, improve the accuracy and reliability of the prediction, and intuitively observe the spatiotemporal distribution change process of the slurry volume fraction during the slurry diffusion process.
[0122] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, the embodiments should be considered exemplary and non-restrictive in all respects, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims be included in the present invention.
[0123] In addition, it should be understood that although the present specification is described according to embodiments, not every embodiment contains only one independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation methods that those skilled in the art can understand.
Claims
1. A grouting diffusion prediction method, characterized in that: The steps include: Construct a COMSOL grouting model; Initializing the COMSOL grouting model and meshing the COMSOL grouting model; Performing grouting simulation based on the COMSOL model to obtain a first slurry volume fraction, wherein the slurry volume fraction is a ratio of the volume of the slurry to the total volume of the slurry and the soil pores; Based on the COMSOL grouting model, all slurry volume fractions within the grouting range are derived at a fixed acquisition frequency, and the slurry volume fractions are normalized to obtain a first data set, wherein the first data set includes grouting inlet pressure data, grouting inlet flow data, soil density data, time data t, and spatial position coordinate data (x, y), wherein x is the abscissa of the slurry in the COMSOL grouting model, and y is the ordinate of the slurry in the COMSOL grouting model; Dividing the first data set into multiple training sets and multiple test sets; Construct multiple neural network models, use training sets for training, and obtain multiple trained neural network models; Inputting the test set into a plurality of the trained neural network models to obtain the slurry volume fraction corresponding to each of the trained neural network models; Calculate the mean absolute error value and the root mean square error value between the slurry volume fraction corresponding to each of the trained neural network models and the first slurry volume fraction; Obtaining an optimal neural network model based on the mean absolute error value and the root mean square error value; Get the data set to be tested; Inputting the test data set into the optimal neural network model to obtain a second slurry volume fraction; The grouting diffusion condition is obtained based on the second slurry volume fraction.
2. The grouting diffusion prediction method according to claim 1, characterized in that: The multiple neural network models are constructed and trained using the training set to obtain multiple trained neural network models, including: Constructing an RBF neural network model, a first BP neural network model, a second BP neural network model and a third BP neural network model; wherein, The radial basis function of the RBF neural network model uses a Gaussian function; The first BP neural network model, the second BP neural network model and the third BP neural network model use an input layer, an output layer and a hidden layer as the structure of the model, the output layer uses a linear activation function, the hidden layer uses a Tanh activation function, and the number of nodes in the hidden layer is m is the number of neurons in the input layer, n is the number of neurons in the output layer, and a is a constant between 1 and 10; The RBF neural network model, the first BP neural network model, the second BP neural network model and the third BP neural network model are trained using the training set to obtain a trained RBF neural network model, a trained first BP neural network model, a trained second BP neural network model and a trained third BP neural network model.
3. The grouting diffusion prediction method according to claim 2, characterized in that: The step of inputting the test set into the plurality of trained neural network models to obtain a plurality of slurry volume fractions comprises: Inputting the test set into the trained RBF neural network model to obtain the slurry volume fraction corresponding to the trained RBF neural network model; Inputting the test set into the trained first BP neural network model to obtain the slurry volume fraction corresponding to the trained first BP neural network model; Inputting the test set into the trained second BP neural network model to obtain the slurry volume fraction corresponding to the trained second BP neural network model; The test set is input into the trained third BP neural network model to obtain the slurry volume fraction corresponding to the trained third BP neural network model.
4. The grouting diffusion prediction method according to claim 3, characterized in that: The step of inputting the test set into the trained RBF neural network model to obtain the slurry volume fraction corresponding to the trained RBF neural network model includes: The grouting inlet pressure data, grouting inlet flow data, soil density data, time data t and spatial position coordinate data (x, y) are input into the trained RBF neural network model to obtain c1(x, y, t), wherein c1(x, y, t) is the slurry volume fraction at the spatial position coordinate (x, y) at time t output by the trained RBF neural network model.
5. The grouting diffusion prediction method according to claim 3, characterized in that: The step of inputting the test set into the trained first BP neural network model to obtain the slurry volume fraction corresponding to the trained first BP neural network model includes: The grouting inlet pressure data, grouting inlet flow data, soil density data and time data t are input into the trained first BP neural network model to obtain c(t), wherein c(t) is the slurry volume fraction at all spatial positions at time t.
6. The grouting diffusion prediction method according to claim 3, characterized in that: The step of inputting the test set into the trained second BP neural network model to obtain the slurry volume fraction corresponding to the trained second BP neural network model includes: The grouting inlet pressure data, grouting inlet flow data, soil density data and spatial position coordinate data (x, y) are input into the trained second BP neural network model to obtain c(x, y), wherein c(x, y) is the slurry volume fraction at the spatial position coordinates (x, y) at all times.
7. The grouting diffusion prediction method according to claim 3, characterized in that: The step of inputting the test set into the trained third BP neural network model to obtain the slurry volume fraction corresponding to the trained third BP neural network model includes: The grouting inlet pressure data, grouting inlet flow data, soil density data, time data t and spatial position coordinate data (x, y) are input into the trained third BP neural network model to obtain c2(x, y, t), wherein c2(x, y, t) is the slurry volume fraction at the spatial coordinate position (x, y) at time t output by the trained third BP neural network model.
8. The grouting diffusion prediction method according to claim 3, characterized in that: The mean absolute error value and the root mean square error value between the slurry volume fraction corresponding to each of the trained neural network models and the first slurry volume fraction include: a first mean absolute error value and a first root mean square error value between the first slurry volume fraction and the slurry volume fraction corresponding to the trained RBF neural network model, a second mean absolute error value and a second root mean square error value between the first slurry volume fraction and the slurry volume fraction corresponding to the trained first BP neural network model, a third mean absolute error value and a third root mean square error value between the first slurry volume fraction and the slurry volume fraction corresponding to the trained second BP neural network model, and a fourth mean absolute error value and a fourth root mean square error value between the first slurry volume fraction and the slurry volume fraction corresponding to the trained third BP neural network model; The optimal neural network model is obtained based on the multiple mean absolute error values and the multiple root mean square error values, comprising: Compare the first mean absolute error value, the second mean absolute error value, the third mean absolute error value and the fourth mean absolute error value, and select the neural network model with the smallest mean absolute error value; Comparing the first root mean square error value, the second root mean square error value, the third root mean square error value and the fourth root mean square error value, and selecting a neural network model with a minimum root mean square error value; If the neural network model with the smallest mean absolute error value and the neural network model with the smallest root mean square error value are the same neural network model, then the neural network model is the optimal neural network model; If the neural network model with the smallest mean absolute error value and the neural network model with the smallest root mean square error value are different neural network models, the neural network model with the smallest root mean square error value is selected as the optimal neural network model.
9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores a program that can be run on the processor, and when the program is executed by the processor, the grouting diffusion prediction method according to any one of claims 1 to 8 is implemented.
10. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by at least one processor, the grouting diffusion prediction method according to any one of claims 1 to 8 is implemented.
Citation Information
Cited By
Grouting simulation and pre-control decision-making method and system based on cross-scale data combination
CN119962436A