Foundation pit deformation prediction method, system, equipment and medium

Abstract

The invention provides a foundation pit deformation prediction method, system and device and a medium, and the method comprises the steps: S1, obtaining the first engineering information of a to-be-predicted foundation pit, and carrying out the preprocessing, and obtaining the first preprocessing engineering information; s2, inputting the first preprocessing engineering information into a constructed generative adversarial network model to obtain predicted foundation pit deformation data; in a generator of the generative adversarial network model, numerical analysis simulation is combined, and physical and mechanical characteristics of the foundation pit engineering are introduced, so that relatively accurate deformation prediction under the condition of relatively small data volume can be realized, and the problems of small data volume, multiple influence factors, unstable influence factors and the like of the foundation pit engineering are solved; the method has the advantages that the method is simple and easy to implement, multi-factor input and complex non-linear relations can be better processed, the expression ability and generalization ability of the model are improved, the reliability and practicability of model prediction results are improved, the prediction precision is improved, powerful support is provided for safety construction and risk control of foundation pit engineering, and foundation pit engineering practice is better served.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of foundation pit engineering, and particularly relates to a foundation pit deformation prediction method, system, device and medium. BACKGROUND

[0002] In the foundation pit engineering, the foundation pit deformation prediction is crucial for ensuring the engineering safety and reducing the construction risk. The foundation pit deformation is comprehensively affected by various factors such as geological conditions, supporting structure form, surrounding load and the like, and there is a complex nonlinear relationship between these factors and the foundation pit deformation. Although the traditional calculation methods such as the elastic mechanics method and the finite element method have certain theoretical basis, they have limitations in dealing with complex actual working conditions, and the accuracy and real-time performance of the calculation results are difficult to meet the engineering requirements. SUMMARY

[0003] The application aims to provide a foundation pit deformation prediction method, system, device and medium, which can improve the foundation pit deformation prediction accuracy and better serve the foundation pit engineering practice.

[0004] The application is implemented by the following technical solutions. A foundation pit deformation prediction method comprises the following steps: S1, obtaining first engineering information of a to-be-predicted foundation pit and preprocessing to obtain first preprocessed engineering information; S2, inputting the first preprocessed engineering information into a constructed generative adversarial network model to obtain predicted foundation pit deformation data; The step of constructing the generative adversarial network model comprises: S21, designing the generative adversarial network model, wherein the generative adversarial network model comprises a generator and a discriminator; S22, obtaining second engineering information of a plurality of constructed foundation pits and real foundation pit deformation data, and preprocessing the obtained second engineering information to obtain second preprocessed engineering information corresponding to the constructed foundation pits; S23, for each constructed foundation pit, inputting the second preprocessed engineering information into the generator to generate predicted foundation pit deformation data, establishing a numerical analysis model according to the second engineering information, and performing simulation analysis on the foundation pit excavation to obtain deformation curves at different excavation stages; S24, judging whether the predicted foundation pit deformation data and the deformation curves are consistent, if yes, executing step S25, and if not, executing step S26; S25, outputting the predicted foundation pit deformation data; S26, adjusting the generative adversarial network model and then re-executing steps S23 and S24; S27, inputting the second pre-processing engineering information, the output predicted foundation pit deformation data and the real foundation pit deformation data into the discriminator to generate a probability that the predicted foundation pit deformation data is the real foundation pit deformation data; S28, calculating a first loss value of the generator and a second loss value of the discriminator, and updating the parameters according to the first loss value and the second loss value to optimize the generative adversarial network model; S29, repeating the step S23, the step S24, the step S27 and the step S28 until the first loss value and the second loss value reach a minimum, and obtaining the generative adversarial network model.

[0005] Further, after the step of outputting the predicted foundation pit deformation data, the method further comprises: judging whether the foundation pit deformation range is received, and if the foundation pit deformation range is received, judging whether the predicted foundation pit deformation data is within the foundation pit deformation range; if the predicted foundation pit deformation data is within the foundation pit deformation range, outputting the predicted foundation pit deformation data; if the predicted foundation pit deformation data is not within the foundation pit deformation range, adjusting the generative adversarial network model and re-executing the step S23 and the step S24.

[0006] Further, the generator comprises a multi-layer perception, the multi-layer perception is provided with multiple hidden layers, and the number of hidden layer neurons gradually decreases, and each hidden layer adopts an activation function for nonlinear transformation; The discriminator comprises a multi-layer perception, the multi-layer perception is provided with multiple hidden layers, and each hidden layer adopts an activation function for nonlinear transformation.

[0007] Further, the engineering information comprises soil layer data, foundation pit excavation data and support structure scheme, the soil layer data comprises the number of soil layers and the thickness, cohesion and friction angle corresponding to each soil layer, the foundation pit excavation data comprises excavation depth, load and the distance between the load and the foundation pit edge, and the support structure scheme comprises support data and cross support data, the support data comprises support type and support parameters, and the cross support data comprises the number of cross supports, the cross support type corresponding to each cross support and cross support parameters.

[0008] Further, the pre-processing step comprises: combining the thickness, cohesion and friction angle corresponding to each soil layer into a tensor according to the soil layer data , and combining each tensor into a soil layer feature matrix in the order of the soil layers from top to bottom, wherein, is the thickness of the i-th soil layer, is the cohesion of the i-th soil layer, is the friction angle of the i-th soil layer; ​​​Based on the foundation pit excavation data, the foundation pit excavation data is processed into a foundation pit feature matrix. Among them, in the foundation pit feature matrix, The pre-defined excavation digital code, The depth of the excavation. The preset load is digitally encoded. The distance between the load and the edge of the foundation pit. For load; Based on the support data, obtain the preset support digital code of the support type, and combine the obtained support digital code and support parameters into a tensor. Based on the cross brace data, obtain the preset cross brace digital code of the cross brace type, and combine the cross brace digital code and cross brace parameters of each cross brace into a tensor. Furthermore, the steps for calculating the first loss value of the generator and the second loss value of the discriminator include: The generator's first loss value is calculated using a first loss function, and the discriminator's second loss value is calculated using a second loss function. The first loss function is as follows: ; The second loss function is as follows: ; In the formula, G is the generator, D is the discriminator, z is random noise, x_geo is the soil feature matrix, x_support is the support structure feature matrix, x_character is the foundation pit feature matrix, and x_deform is the actual foundation pit deformation data.

[0009] This invention also discloses a foundation pit deformation prediction system, comprising: The acquisition module is used to acquire the first engineering information of the foundation pit to be predicted and to preprocess it to obtain the first preprocessed engineering information. The input module is used to input the first preprocessed engineering information into the constructed generative adversarial network model to obtain the predicted foundation pit deformation data; The input module includes: The design submodule is used to design generative adversarial network (GAN) models, which include generators and discriminators. The acquisition submodule is used to acquire the second engineering information and actual foundation pit deformation data of multiple constructed foundation pits, and to preprocess the acquired second engineering information to obtain the second preprocessed engineering information corresponding to the constructed foundation pits. The first input submodule is used to input the second preprocessed engineering information into the generator for each constructed foundation pit, generate predicted foundation pit deformation data, establish a numerical analysis model based on the second engineering information, and perform simulation analysis on the foundation pit excavation to obtain deformation curves at different excavation stages. The first judgment submodule is used to determine whether the predicted foundation pit deformation data matches the deformation curve. If they match, the first output submodule is executed; if they do not match, the first adjustment submodule is executed. The first output submodule is used to output the predicted foundation pit deformation data; The first adjustment submodule is used to adjust the generative adversarial network model and then re-execute the first input submodule and the first judgment submodule. The second input submodule is used to input the second preprocessed engineering information, the output predicted foundation pit deformation data and the actual foundation pit deformation data into the discriminator to generate the probability that the predicted foundation pit deformation data is the actual foundation pit deformation data. The computation submodule is used to calculate the first loss value of the generator and the second loss value of the discriminator, and update the parameters based on the first and second loss values ​​to optimize the generative adversarial network model. The repeated submodules are used for the first input submodule, the first decision submodule, the second input submodule, and the calculation submodule until the first loss value and the second loss value are minimized, thus obtaining the generative adversarial network model.

[0010] The present invention also discloses an electronic device, which includes: processor; Memory is used to store executable computer programs; Among them, the steps of implementing the foundation pit deformation prediction method when the processor executes the computer program.

[0011] The present invention also discloses a computer-readable storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the steps of the foundation pit deformation prediction method.

[0012] Compared with existing technologies, the beneficial effects of this invention are as follows: It improves existing generative adversarial network (GAN) models by incorporating numerical analysis simulation into the generator and introducing the physical and mechanical properties of foundation pit engineering. This enables relatively accurate deformation prediction with limited data, solving problems such as limited data volume, numerous influencing factors, and unstable influencing factors in foundation pit engineering. The optimized GAN model can better handle multi-factor inputs and complex nonlinear relationships, improving its expressive and generalization capabilities. This further enhances the reliability and practicality of the GAN model's prediction results, increases prediction accuracy, and provides strong support for safe construction and risk control in foundation pit engineering, better serving foundation pit engineering practice. Attached Figure Description

[0013] Figure 1 This is a flowchart illustrating the steps of the foundation pit deformation prediction method of the present invention; Figure 2This is a flowchart illustrating the steps of constructing a generative adversarial network model in the foundation pit deformation prediction method of the present invention. Figure 3 This is a schematic diagram of the generator design in the foundation pit deformation prediction method of the present invention; Figure 4 This is a schematic diagram of the discriminator design in the foundation pit deformation prediction method of the present invention; Figure 5 This is a schematic diagram of the modules of the foundation pit deformation prediction system of the present invention; Figure 6 This is a hardware structure diagram of the electronic device of the present invention. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0015] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0016] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0017] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0018] In the description of this invention, it should be noted that the terms "upper," "lower," "inner," "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship in which the product of this invention is usually placed when in use. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention.

[0019] Please see Figure 1 , Figure 1 This is a flowchart illustrating the steps of the foundation pit deformation prediction method of the present invention. A foundation pit deformation prediction method includes the following steps: S1. Obtain the first engineering information of the foundation pit to be predicted, and perform preprocessing to obtain the first preprocessed engineering information; S2. Input the first preprocessed engineering information into the constructed generative adversarial network model to obtain the predicted foundation pit deformation data; In step S1 above, the engineering information of the foundation pit to be predicted is obtained as the first engineering information. This engineering information can include data such as the geological conditions of the foundation pit, the form of the support structure, and the surrounding loads. Specifically, the engineering information includes soil layer data, foundation pit excavation data, and support structure scheme. Soil layer data includes the number of soil layers and the corresponding thickness, cohesion, and friction angle of each layer. Foundation pit excavation data includes the excavation depth, load, and the distance between the load and the edge of the foundation pit. The support structure scheme includes support data and cross bracing data. Support data includes support type and support parameters. Cross bracing data includes the number of cross braces, the type of cross brace corresponding to each cross brace, and cross bracing parameters. To effectively integrate the soil layer data, foundation pit excavation data, and support structure scheme from the engineering information into the generative adversarial network (GAN) model, enabling the GAN model to fully learn the complex relationship between various factors and foundation pit deformation, the soil layer data, foundation pit excavation data, and support structure scheme are preprocessed, i.e., a comprehensive and systematic coding process is performed.

[0020] Furthermore, the preprocessing steps include: S11. Based on the soil layer data, combine the thickness, cohesion and friction angle of each soil layer into a tensor, and combine the tensors into a soil layer feature matrix in the order of soil layers from top to bottom. S12. Based on the foundation pit excavation data, process the foundation pit excavation data into a foundation pit feature matrix. Among them, in the foundation pit feature matrix, The pre-defined excavation digital code, The depth of the excavation. The preset load is digitally encoded. The distance between the load and the edge of the foundation pit. For load; S13. Based on the support data, obtain the preset support digital code of the support type, and combine the obtained support digital code and support parameters into a tensor. Based on the cross brace data, obtain the preset cross brace digital code of the cross brace type, and combine the cross brace digital code and cross brace parameters of each cross brace into a tensor. S14. Combine the tensors obtained from the support data with the tensors obtained from the cross brace data to form the support structure feature matrix.

[0021] In step S11 above, based on the soil layer data, the thickness, cohesion, and friction angle of each soil layer in the foundation pit are obtained. Assuming the foundation pit has n soil layers, then the thickness, cohesion, and friction angle of each soil layer are obtained from the nth soil layer. The tensor formed by combining the thickness, cohesion, and friction angle data corresponding to each soil layer is: ,in, For the first The thickness of the soil layers, For the first The cohesion of the soil layers For the first The friction angle of each soil layer. Then, combining the tensors of the n soil layers in top-to-bottom order to form the soil layer characteristic matrix is: In the soil layer characteristic matrix, For the first The thickness of the soil layers, For the first The cohesion of the soil layers For the first The friction angle of each soil layer For the first The thickness of the soil layers, For the first The cohesion of the soil layers For the first The friction angle of each soil layer For the first The thickness of the soil layers, For the first The cohesion of the soil layers For the first The friction angle of the soil layers.

[0022] Therefore, if the soil data shows three soil layers, the thickness of the first soil layer is... Cohesion Friction angle The thickness of the second soil layer Cohesion Friction angle The thickness of the third soil layer Cohesion Friction angle The soil feature matrix obtained from this soil layer data is: .

[0023] In step S12 above, the digital code is excavated. It can be set to 1, load digital encoding If we set it to 2, then if the foundation pit excavation data is a foundation pit with an excavation depth of 5m and a uniformly distributed load of 20kPa 3m from the pit edge, then the foundation pit feature matrix obtained from this foundation pit excavation data is: .

[0024] In step S13 above, the support type and its parameters (such as pit depth, pile length, pile diameter, anchor cable spacing, cross-sectional area of ​​internal supports, horizontal and vertical spacing of internal supports, etc.) are first encoded. Each support type is encoded as a tensor (matrix). The first element of the tensor represents the support type's numerical code, and the subsequent elements represent the specific parameters of the support structure. This represents a single-row cast-in-place pile. The support parameters for a single-row cast-in-place pile include pile length, pile diameter, and pile spacing. Represents the pile length. Represents the pile diameter. This represents the pile spacing, from which a support code is derived. ; This represents a diaphragm wall. The support parameters for a diaphragm wall include wall depth and wall thickness. Represents the wall depth. Represents wall thickness; Representing sheet piles, the support parameters for sheet piles include pile length, cross-sectional area, and moment of inertia of the sheet pile section. Represents the pile length. Represents the cross-sectional area of ​​the sheet pile. Represents the moment of inertia of the sheet pile section; This represents double-row cast-in-place piles. The support parameters for double-row cast-in-place piles include the length of the first row of piles, the diameter of the first row of piles, the spacing between the first row of piles, the length of the second row of piles, the diameter of the second row of piles, and the spacing between the second row of piles. Represents the length of the first row of piles. Represents the diameter of the first row of piles. Represents the spacing between the first row of piles. Represents the length of the second row of piles. Represents the diameter of the second row of piles. This represents the spacing between the second row of piles; For representative method piles, the support parameters include pile length, cement-soil pile diameter, cross-sectional area of ​​the steel section, and moment of inertia of the steel section. Represents the pile length. Represents the diameter of cement-soil piles. Representative steel cross-sectional area, Moment of inertia of representative steel section; The representative of the cement-soil mixing pile gravity type is the support parameters for the cement-soil mixing pile gravity type, which include pile length, replacement ratio, and wall thickness. Represents the pile length. Represents the replacement rate. Represents wall thickness. Represents support type. , , , , and The numbers 1, 2, 3, 4, 5, and 6 can be used. For example, in the case of a foundation pit supported by cast-in-place piles, the piles are 20m long, 0.8m in diameter, and spaced 1.0m apart. The obtained support digital code for the cast-in-place pile support is... Given a value of 1, and considering the support parameters of the cast-in-place pile support, the resulting tensor is: Other types of support can be added according to the above methods. The final matrix size depends on the number of support types used in a foundation pit and the number of parameters for each support type. If the number of parameters for a support type is insufficient, the remaining empty spaces are filled with 0. For example, if a foundation pit uses single-row cast-in-place piles, sheet piles, double-row cast-in-place piles, and construction method piles, the resulting matrix would be as follows: .

[0025] Similarly, the horizontal type and its parameters are encoded. The encoding method for the horizontal brace is the same as that for the support in step S13 above. In the tensor (matrix) obtained for a certain horizontal brace, the first element is the horizontal brace numerical code of the horizontal brace type, and the subsequent elements are the horizontal brace parameters of that type. For example... Represents a slope. This represents the slope ratio of the first layer. This represents the slope ratio of the second layer. This represents the slope ratio of the third layer. Represents internal concrete support. Represents the cross-sectional area of ​​the support. This represents the distance between the support structure and the ground. Represents the horizontal spacing of the supports; Represents steel support. Represents the cross-sectional area of ​​the support. This represents the distance between the support structure and the ground. Represents the horizontal spacing of the supports; Represents anchor cable, Represents the cross-sectional area of ​​the anchor cable. This represents the distance between the anchor cable and the ground. Represents the horizontal spacing of the anchor cables; Represents soil nails, Represents the cross-sectional area of ​​the soil nail. This represents the distance between the anchor cable and the ground. This represents the horizontal spacing of the anchor cables; other types follow the same pattern, using different cross brace numerical codes to represent different cross brace types, such as representing the support type. , , , and The numbers 0, 1, 2, 3, and 4 can be used. For example, if the cross bracing is made of concrete with a cross-sectional area of ​​0.8 m², a distance of 0 m from the ground, and a horizontal spacing of 10 m, obtain the numerical code of the concrete cross bracing. Given a value of 1, and considering the cross bracing parameters of the concrete support, the resulting tensor is: .

[0026] In step S14 above, the tensor obtained from the support data is used as the first row of the support structure feature matrix; the tensor obtained from the first horizontal brace is used as the second row; the tensor obtained from the second horizontal brace is used as the third row; the tensor obtained from the third horizontal brace is used as the fourth row, and so on, thus obtaining the support structure feature matrix corresponding to the support structure scheme. For example, the support structure scheme for the foundation pit uses cast-in-place piles with a length of 20m, a diameter of 0.8m, and a spacing of 1.0m. The first support uses concrete supports with a cross-sectional area of ​​0.8m². 2 The first support is 0m above the ground, with a horizontal spacing of 10m. The second support uses anchor cables with a cross-sectional area of ​​140mm². 2 The support structure is 5m above the ground and 1.5m horizontally. Therefore, the characteristic matrix of the support structure obtained from the example support structure scheme is: Thus, the soil layer feature matrix, the foundation pit feature matrix, and the support structure feature matrix are fused into preprocessed engineering information, forming an input feature matrix containing multiple influencing factors.

[0027] Please refer to the following: Figure 2 , Figure 2 This is a flowchart illustrating the steps of constructing a generative adversarial network (GAN) model in the foundation pit deformation prediction method of this invention. In step S2, a GAN model is pre-constructed. The first pre-processed engineering information is input into the constructed GAN model to obtain the predicted foundation pit deformation data. The foundation pit deformation data adopts the original data format. The deep horizontal displacement is a series of deformation values ​​that vary along the depth, which is an n x 2 matrix. The first column represents the depth position, and the second column represents the deformation value. The unit is set to mm, as follows: ,in, , ... and This is the deformation value.

[0028] Specifically, the steps for constructing a generative adversarial network model include: S21. Design a generative adversarial network (GAN) model, which includes a generator and a discriminator. S22. Obtain the second engineering information and actual foundation pit deformation data of multiple constructed foundation pits, and preprocess the obtained second engineering information to obtain the second preprocessed engineering information corresponding to the constructed foundation pits. S23. For each constructed foundation pit, input the second pre-processed engineering information into the generator to generate predicted foundation pit deformation data, establish a numerical analysis model based on the second engineering information, and conduct simulation analysis on the foundation pit excavation to obtain deformation curves at different excavation stages. S24. Determine whether the predicted foundation pit deformation data matches the deformation curve. If they match, proceed to step S25. If they do not match, proceed to step S26. S25. Output predicted foundation pit deformation data; S26. After adjusting the generative adversarial network model, repeat steps S23 and S24. S27. Input the second preprocessed engineering information, the output predicted foundation pit deformation data and the actual foundation pit deformation data into the discriminator to generate the probability that the predicted foundation pit deformation data is the actual foundation pit deformation data. S28. Calculate the first loss value of the generator and the second loss value of the discriminator, and update the parameters based on the first and second loss values ​​to optimize the generative adversarial network model; S29. Repeat steps S23, S24, S27 and S28 until the first loss value and the second loss value reach their minimum, thus obtaining the generative adversarial network model.

[0029] In step S21 above, the generator includes a multilayer perceptron (MLP). The MLP has multiple hidden layers, each containing several neurons, with the number of neurons in each hidden layer gradually decreasing. Each hidden layer uses an activation function for nonlinear transformation. The input layer is used to receive preprocessed engineering information and a concatenation of random noise vectors. After being processed by multiple hidden layers, the output layer finally outputs the predicted foundation pit deformation data.

[0030] Please refer to the following: Figure 3 Generator design such as Figure 3 As shown, specifically, the generator is designed as follows: The first step is to design the input layer. The dimensions of the soil feature matrix are [L1L2], the pit feature matrix is ​​[L3L4], the support structure feature matrix is ​​[L5L6], and the random noise is [L7L8]. First, filtering is performed to extract specific features from the input data. A filter (also called a convolution kernel) is a small weight matrix (such as 1×1, 2×2, 3×3, etc.) that slides (convolves) through the input data in a convolution operation, extracting local features through dot product operations. Each filter focuses on detecting a specific type of feature. By reasonably setting the size of the convolution kernel, the first two dimensions of information in different dimensions can be kept consistent. The convolution kernel sizes for soil layer information, foundation pit features, support information, and random noise are n1×n1 (32 convolution kernels), n2×n2 (16 convolution kernels), n3×n3 (32 convolution kernels), and n4×n4 (16 convolution kernels), respectively. However, due to the different characteristics of various types of data, different numbers of convolution kernels are used, resulting in different numbers of channels, namely n×n×32, n×n×16, n×n×32, and n×n×16.

[0031] The second step involves designing a feature fusion layer. Feature fusion merges tensors along the channel dimension, resulting in a tensor with a dimension of n×n×96 (i.e., 32+16+32+16=96). Then, 64 1×1 convolutional kernels are used to transform the tensor into n×n×64, which is then input into the ReLU activation function.

[0032] The third step uses 16 1×1 convolution kernels to transform the tensor into n×n×16, then uses one 1×1 convolution kernel to transform the tensor into n×n×1, and finally inputs it into the ReLU activation function to obtain the deformed prediction matrix.

[0033] It should be noted that the size, number of convolutional kernels, number of network layers, and type of activation function used in the above network structure can be adjusted according to the actual situation.

[0034] The discriminator comprises a multilayer perceptron (MLP) and a convolutional neural network (CNN) structure. The MLP has multiple hidden layers, each employing an activation function for non-linear transformation. The input layer receives either real foundation pit deformation data concatenated with the input feature vector, or predicted foundation pit deformation data generated by the generator concatenated with the input feature vector. The input features undergo non-linear transformation and combination through multiple hidden layers. The output layer uses a sigmoid activation function to output the probability that the data represents the real data.

[0035] Please refer to the following: Figure 4 Discriminator design such as Figure 4 As shown, the discriminator is specifically designed as follows: The first step is to design the input layer. Similar to the generator, the convolution kernel sizes for soil layer information, foundation pit features, support information, and deformation data are n1×n1 (32 convolution kernels), n2×n2 (16 convolution kernels), n3×n3 (32 convolution kernels), and n4×n4 (16 convolution kernels), respectively, resulting in four tensors: n×n×32, n×n×16, n×n×32, and n×n×16.

[0036] The second step involves designing a feature fusion layer. Feature fusion merges tensors along the channel dimension, resulting in a tensor with a dimension of n×n×96 (i.e., 32+16+32+16=96). Then, 64 1×1 convolutional kernels are used to transform the tensor into n×n×64, which is then input into the ReLU activation function.

[0037] The third step involves flattening the multidimensional tensor into a one-dimensional vector. For example, a 3×3×3 tensor flattened results in a vector of length 9. In this invention, the output after flattening is a vector of length n×n×64 obtained in the previous step, which is then input into the ReLU activation function.

[0038] The fourth step is to set up a fully connected layer, input the vector of length n×n×64 obtained above, output a vector of length 128 and input it into the Leaky ReLU function.

[0039] The fifth step is to set up a fully connected layer, input the vector of length 128 obtained above, output a vector of length 64 and input it into the Leaky ReLU function.

[0040] The sixth step is to set up a fully connected layer, input the vector of length 64 obtained above, output a vector of length 32 and input it into the Leaky ReLU function.

[0041] The seventh step involves setting up a fully connected layer. Input the vector of length 32 obtained above, and output a vector of length 1, which is then fed into the Sigmoid function. Since the output range of the Sigmoid activation function is [0,1], a probability value of [0,1] can be obtained. The closer the output value is to 1, the closer it is to the true value.

[0042] In step S22 above, engineering information and actual foundation pit deformation data of multiple constructed foundation pits are collected. The collected engineering information of the constructed foundations is used as the second engineering information. Then, the second engineering information is preprocessed to obtain the second preprocessed engineering information. The preprocessing process is the same as steps S11 to S14, and will not be repeated here. The second preprocessed engineering information and the actual foundation pit deformation data are combined into a training set.

[0043] In steps S23 to S26 above, since there is relatively little data on the excavated foundation pit, it is difficult to accurately predict the deformation. Therefore, in the existing generative adversarial network model, combined with the calculation of the finite metadata analysis model, the physical and mechanical properties of the foundation pit project are introduced, which can achieve relatively accurate deformation prediction under the condition of less data. Specifically, after the generator generates the predicted foundation pit deformation data, a numerical analysis model is established using finite element software based on the engineering data of the constructed foundation pit. The excavation at each stage is simulated and analyzed to obtain deformation curves for different excavation stages. The generator then incorporates a deformation morphology judgment obtained from the numerical analysis to determine whether the predicted foundation pit deformation data obtained by the generator matches the deformation curve obtained from the numerical analysis. That is, the predicted curve is plotted based on the predicted foundation pit deformation data, and the prediction curve matches the deformation curve. If they match, the predicted foundation pit deformation data is output. If they do not match, the parameters of the generative adversarial network model are readjusted, i.e., the calculated weights are adjusted. Then, steps S23 and S24 are executed using the adjusted generative adversarial network model until the predicted foundation pit deformation data matches the deformation curve. Finally, the predicted foundation pit deformation data is output.

[0044] Furthermore, after step S25, which outputs the predicted foundation pit deformation data, the method further includes: S251. Determine whether the excavation pit deformation range has been received. If the excavation pit deformation range has been received, determine whether the predicted excavation pit deformation data is within the excavation pit deformation range. S252. If the predicted foundation pit deformation data is within the foundation pit deformation range, then output the predicted foundation pit deformation data. S253. If the predicted foundation pit deformation data is not within the range of foundation pit deformation, adjust the generative adversarial network model and re-execute steps S23 and S24.

[0045] In steps S251 to S253 above, an expert module is added to the generative adversarial network (GAN) model, allowing users to input deformation limits based on experience (from relevant standards, similar engineering cases, or experienced engineers). Therefore, when the deformation range of the foundation pit is received, it is determined whether the predicted foundation pit deformation data is within the deformation range. If it exceeds the deformation range, the parameters of the GAN model are readjusted, and then steps S23 and S24 are executed using the adjusted GAN model until the predicted foundation pit deformation data matches the deformation curve and meets the deformation range requirements. Only then is the predicted foundation pit deformation data output. This further enables relatively accurate deformation prediction with a smaller amount of data.

[0046] In steps S27 to S29 above, the second preprocessed engineering information, the output predicted foundation pit deformation data, and the actual foundation pit deformation data are input into the discriminator to generate the probability that the predicted foundation pit deformation data is the actual foundation pit deformation data. The closer the output probability is to 1, the closer the predicted foundation pit deformation data is to the actual deformation data. Then, the first loss function is used to calculate the first loss value of the generator, and the second loss function is used to calculate the second loss value of the discriminator. The first loss function is as follows: ; The second loss function is as follows: ; In the formula, G is the generator, D is the discriminator, z is random noise, x_geo is the soil feature matrix, x_support is the support structure feature matrix, x_character is the foundation pit feature matrix, and x_deform is the actual foundation pit deformation data.

[0047] The generator optimizes its parameters based on the adversarial loss fed back by the discriminator to improve the accuracy of the predicted foundation pit deformation data; the discriminator is updated at the same time to improve its ability to distinguish between the predicted foundation pit deformation data and the actual foundation pit deformation data, so as to minimize the first loss value and the second loss value, thus obtaining the generative adversarial network model.

[0048] Please see Figure 3 , Figure 3 This is a schematic diagram of the foundation pit deformation prediction system of the present invention. Corresponding to the aforementioned embodiments of the foundation pit deformation prediction method of the present invention, the present invention also provides a foundation pit deformation prediction system, including: Module 1 is used to acquire the first engineering information of the foundation pit to be predicted and to preprocess it to obtain the first preprocessed engineering information. Input module 2 is used to input the first preprocessed engineering information into the constructed generative adversarial network model to obtain the predicted foundation pit deformation data; Input module 2 includes: The design submodule is used to design generative adversarial network (GAN) models, which include generators and discriminators. The acquisition submodule is used to acquire the second engineering information and actual foundation pit deformation data of multiple constructed foundation pits, and to preprocess the acquired second engineering information to obtain the second preprocessed engineering information corresponding to the constructed foundation pits. The first input submodule is used to input the second preprocessed engineering information into the generator for each constructed foundation pit, generate predicted foundation pit deformation data, establish a numerical analysis model based on the second engineering information, and perform simulation analysis on the foundation pit excavation to obtain deformation curves at different excavation stages. The first judgment submodule is used to determine whether the predicted foundation pit deformation data matches the deformation curve. If they match, the first output submodule is executed; if they do not match, the first adjustment submodule is executed. The first output submodule is used to output the predicted foundation pit deformation data; The first adjustment submodule is used to adjust the generative adversarial network model and then re-execute the first input submodule and the first judgment submodule. The second input submodule is used to input the second preprocessed engineering information, the output predicted foundation pit deformation data and the actual foundation pit deformation data into the discriminator to generate the probability that the predicted foundation pit deformation data is the actual foundation pit deformation data. The computation submodule is used to calculate the first loss value of the generator and the second loss value of the discriminator, and update the parameters based on the first and second loss values ​​to optimize the generative adversarial network model. The repeated submodules are used for the first input submodule, the first decision submodule, the second input submodule, and the calculation submodule until the first loss value and the second loss value are minimized, thus obtaining the generative adversarial network model.

[0049] Furthermore, input module 2 also includes: The second judgment submodule is used to determine whether the excavation pit deformation range has been received. If the excavation pit deformation range has been received, it is then determined whether the predicted excavation pit deformation data is within the excavation pit deformation range. The second output submodule is used to output the predicted foundation pit deformation data if the second judgment submodule determines that it is true. The second adjustment submodule is used to adjust the generative adversarial network model and re-execute the first input submodule and the first judgment submodule if the predicted foundation pit deformation data is not within the foundation pit deformation range.

[0050] Furthermore, the engineering information includes soil layer data, foundation pit excavation data, and support structure scheme. Soil layer data includes the number of soil layers and the corresponding thickness, cohesion, and friction angle of each soil layer. Foundation pit excavation data includes the excavation depth, load, and distance between the load and the edge of the foundation pit. Support structure scheme includes support data and cross bracing data. Support data includes support type and support parameters. Cross bracing data includes the number of cross braces, the cross bracing type corresponding to each cross brace, and cross bracing parameters. Acquisition module 1 includes: a first combination submodule, used to combine the thickness, cohesion, and friction angle of each soil layer into a tensor based on the soil layer data. And according to the soil layers from top to bottom, the tensors are combined into a soil layer feature matrix, where, For the first The thickness of the soil layers, For the first The cohesion of the soil layers For the first The friction angle of each soil layer; The processing submodule is used to process the foundation pit excavation data into a foundation pit feature matrix based on the foundation pit excavation data. Among them, in the foundation pit feature matrix, The pre-defined excavation digital code, The depth of the excavation. The preset load is digitally encoded. The distance between the load and the edge of the foundation pit. For load; The second combination submodule is used to obtain the preset support digital code of the support type according to the support data, and combine the obtained support digital code and support parameters into a tensor. It is also used to obtain the preset cross brace digital code of the cross brace type according to the cross brace data, and combine the cross brace digital code and cross brace parameters of each cross brace into a tensor.

[0051] Furthermore, the computation submodule includes: The calculation unit is used to calculate the first loss value of the generator using a first loss function and the second loss value of the discriminator using a second loss function. The first loss function is as follows: ; The second loss function is as follows: ; In the formula, G is the generator, D is the discriminator, z is random noise, x_geo is the soil feature matrix, x_support is the support structure feature matrix, x_character is the foundation pit feature matrix, and x_deform is the actual foundation pit deformation data.

[0052] The implementation process of the functions and roles of each module, submodule and unit in the above system is detailed in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0053] For the system embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The system embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units.

[0054] Corresponding to the aforementioned embodiments of the foundation pit deformation prediction method, the present invention also provides an electronic device, which may include: a processor; a memory for storing an executable computer program; wherein, when the processor executes the computer program, it implements the foundation pit deformation prediction method in any of the aforementioned method embodiments.

[0055] The embodiments of the foundation pit deformation prediction and system provided in this invention can all be applied to electronic devices. Taking software implementation as an example, as a logical device, it is formed by the processor of the electronic device loading the corresponding computer program instructions from non-volatile memory into memory and running them. From a hardware perspective, such as... Figure 4 As shown, except Figure 4 In addition to the processor, memory, network interface, and non-volatile memory shown, the electronic device may also include other hardware, such as a camera module; or, depending on the actual function of the electronic device, it may also include other hardware, which will not be elaborated further.

[0056] Corresponding to the aforementioned embodiments of the foundation pit deformation prediction method, this embodiment of the invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the foundation pit deformation prediction method in any of the aforementioned method embodiments.

[0057] Embodiments of the present invention may take the form of a computer program product implemented on one or more storage media containing program code (including but not limited to disk storage, CD-ROM, optical storage, etc.). The computer-readable storage medium may include: permanent or non-permanent removable or non-removable media. The information storage function of the computer-readable storage medium can be implemented by any feasible method or technology. The information may be computer-readable instructions, data structures, program models, or other data.

[0058] In addition, computer-readable storage media include, but are not limited to: phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or other non-transfer media that can be used to store information accessible by a computing device.

[0059] Compared with existing technologies, the beneficial effects of this invention are as follows: It improves existing generative adversarial network (GAN) models by incorporating numerical analysis simulation into the generator and introducing the physical and mechanical properties of foundation pit engineering. This enables relatively accurate deformation prediction with limited data, solving problems such as limited data volume, numerous influencing factors, and unstable influencing factors in foundation pit engineering. The optimized GAN model can better handle multi-factor inputs and complex nonlinear relationships, improving its expressive and generalization capabilities. This further enhances the reliability and practicality of the GAN model's prediction results, increases prediction accuracy, and provides strong support for safe construction and risk control in foundation pit engineering, better serving foundation pit engineering practice.

[0060] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Therefore, any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for predicting foundation pit deformation, characterized in that, Includes the following steps: S1. Obtain the first engineering information of the foundation pit to be predicted, and perform preprocessing to obtain the first preprocessed engineering information; S2. Input the first preprocessed engineering information into the constructed generative adversarial network model to obtain the predicted foundation pit deformation data; The steps involved in constructing a generative adversarial network model include: S21. Design the generative adversarial network model, which includes a generator and a discriminator; S22. Obtain the second engineering information and actual foundation pit deformation data of multiple constructed foundation pits, and preprocess the obtained second engineering information to obtain the second preprocessed engineering information corresponding to the constructed foundation pits. S23. For each constructed foundation pit, the second pre-processed engineering information is input into the generator to generate predicted foundation pit deformation data. A numerical analysis model is established based on the second engineering information, and the foundation pit excavation is simulated and analyzed to obtain deformation curves at different excavation stages. S24. Determine whether the predicted foundation pit deformation data matches the deformation curve. If they match, proceed to step S25. If they do not match, proceed to step S26. S25. Output the predicted foundation pit deformation data; S26. After adjusting the generative adversarial network model, re-execute steps S23 and S24; S27. Input the second preprocessed engineering information, the output predicted foundation pit deformation data and the actual foundation pit deformation data into the discriminator to generate the probability that the predicted foundation pit deformation data is the actual foundation pit deformation data. S28. Calculate the first loss value of the generator and the second loss value of the discriminator, and update the parameters based on the first and second loss values ​​to optimize the generative adversarial network model; S29. Repeat steps S23, S24, S27 and S28 until the first loss value and the second loss value reach their minimum, thus obtaining the generative adversarial network model.

2. The method for predicting foundation pit deformation according to claim 1, characterized in that, After the step of outputting the predicted foundation pit deformation data, the method further includes: Determine whether the excavation pit deformation range has been received; if the excavation pit deformation range has been received, determine whether the predicted excavation pit deformation data is within the excavation pit deformation range. If the predicted excavation pit deformation data is within the excavation pit deformation range, then the predicted excavation pit deformation data is output. If the predicted foundation pit deformation data is not within the range of foundation pit deformation, then adjust the generative adversarial network model and re-execute steps S23 and S24.

3. The method for predicting foundation pit deformation according to claim 1, characterized in that, The generator includes a multilayer perceptron, which has multiple hidden layers with the number of neurons in each hidden layer decreasing gradually. Each hidden layer is transformed nonlinearly using an activation function. The discriminator includes a multilayer perceptron, which has multiple hidden layers, each of which is nonlinearly varied using an activation function.

4. The method for predicting foundation pit deformation according to claim 1, characterized in that, The engineering information includes soil layer data, foundation pit excavation data, and support structure scheme. The soil layer data includes the number of soil layers and the corresponding thickness, cohesion, and friction angle of each soil layer. The foundation pit excavation data includes the excavation depth, load, and distance between the load and the edge of the foundation pit. The support structure scheme includes support data and cross brace data. The support data includes the support type and support parameters. The cross brace data includes the number of cross braces, the cross brace type corresponding to each cross brace, and the cross brace parameters.

5. The method for predicting foundation pit deformation according to claim 4, characterized in that, The preprocessing steps include: Based on the soil layer data, the thickness, cohesion and friction angle of each soil layer are combined into a tensor, and the tensors are combined into a soil layer feature matrix in the order of soil layers from top to bottom. Based on the excavation data of the foundation pit, the excavation data is processed into a foundation pit feature matrix. Among them, in the foundation pit feature matrix, The pre-defined excavation digital code, The depth of the excavation. The preset load is digitally encoded. The distance between the load and the edge of the foundation pit. For load; Based on the support data, a preset support digital code for the support type is obtained, and the obtained support digital code and support parameters are combined into a tensor. Based on the cross brace data, a preset cross brace digital code for the cross brace type is obtained, and the cross brace digital codes and cross brace parameters of each cross brace are combined into a tensor. The tensors obtained from the support data are combined with the tensors obtained from the cross brace data to form the support structure feature matrix.

6. The method for predicting foundation pit deformation according to claim 5, characterized in that, The steps of calculating the first loss value of the generator and the second loss value of the discriminator include: The generator's first loss value is calculated using a first loss function, and the discriminator's second loss value is calculated using a second loss function. The first loss function is as follows: ； The second loss function is as follows: ； In the formula, G is the generator, D is the discriminator, z is random noise, x_geo is the soil feature matrix, x_support is the support structure feature matrix, x_character is the foundation pit feature matrix, and x_deform is the actual foundation pit deformation data.

7. A foundation pit deformation prediction system, characterized in that, include: The acquisition module is used to acquire the first engineering information of the foundation pit to be predicted and to preprocess it to obtain the first preprocessed engineering information. The input module is used to input the first preprocessed engineering information into the constructed generative adversarial network model to obtain the predicted foundation pit deformation data; The input module includes: The design submodule is used to design the generative adversarial network model, which includes a generator and a discriminator. The acquisition submodule is used to acquire the second engineering information and actual foundation pit deformation data of multiple constructed foundation pits, and to preprocess the acquired second engineering information to obtain the second preprocessed engineering information corresponding to the constructed foundation pits. The first input submodule is used to input the second preprocessed engineering information into the generator for each constructed foundation pit, generate predicted foundation pit deformation data, establish a numerical analysis model based on the second engineering information, and perform simulation analysis on the foundation pit excavation to obtain deformation curves at different excavation stages. The first judgment submodule is used to determine whether the predicted foundation pit deformation data matches the deformation curve. If they match, the first output submodule is executed; if they do not match, the first adjustment submodule is executed. The first output submodule is used to output the predicted foundation pit deformation data; The first adjustment submodule is used to adjust the generative adversarial network model and then re-execute the first input submodule and the first judgment submodule. The second input submodule is used to input the second preprocessed engineering information, the output predicted foundation pit deformation data and the actual foundation pit deformation data into the discriminator to generate the probability that the predicted foundation pit deformation data is the actual foundation pit deformation data. The computation submodule is used to calculate the first loss value of the generator and the second loss value of the discriminator, and update the parameters based on the first and second loss values ​​to optimize the generative adversarial network model. The repeated submodules are used for the first input submodule, the first decision submodule, the second input submodule, and the calculation submodule until the first loss value and the second loss value are minimized, thus obtaining the generative adversarial network model.

8. An electronic device, characterized in that, include: processor; Memory is used to store executable computer programs; Wherein, when the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images