Deep foundation pit stability deformation control method and device based on digital twinning

Abstract

The invention discloses a deep foundation pit stability deformation control method and device based on digital twinning. The method comprises the steps that monitoring time sequence data of a deep foundation pit are collected and obtained; constructing a deep foundation pit deformation control digital twin platform; constructing a CNN-LSTM composite neural network based on an Adam optimization algorithm through a foundation pit deformation prediction module, and performing model training / verification on the CNN-LSTM composite neural network through the data training set and the data verification set to obtain an optimized CNN-LSTM composite neural network; based on the optimized CNN-LSTM composite neural network, performing time sequence change prediction on the axial force of the steel support according to the data test set to obtain a prediction result; the steel support servo module realizes servo control on the axial force of the steel support according to a prediction result on the basis of a constructed servo control strategy; and the foundation pit BIM module is used for constructing a deep foundation pit BIM model based on a Revit modeling technology and interacting with the deep foundation pit BIM model in real time through a RevitAPI data interface according to the output of the steel support servo module so as to realize visualization of deformation control of the deep foundation pit. The problems that in the deep foundation pit construction process, due to the fact that complex geological conditions are involved, the surrounding environment and an existing building are affected to a certain degree, and deformation and stability of the deep foundation pit cannot be accurately controlled are solved.

Description

Technical Field

[0001] The present invention relates to the field of deformation control of deep foundation pit engineering, and in particular to a method and device for controlling deformation of deep foundation pit stability based on digital twin. Background Art

[0002] Deep foundation pit engineering refers to the excavation of deep foundation pits in urban construction in order to carry out projects such as underground space development, subway construction, and high-rise building foundations. With the acceleration of urbanization, the demand for urban underground space has gradually increased, and the scale and depth of deep foundation pit engineering have continued to increase. Deep foundation pit construction usually involves complex geological conditions, and the construction process may have a certain impact on the surrounding environment and existing buildings, so it is necessary to accurately control the deformation and stability of the foundation pit.

[0003] Deformation control of deep foundation pits is a key technology to ensure the stability of the soil around the foundation pits and prevent major accidents such as ground subsidence and damage to surrounding buildings. During the construction of the foundation pits, structures such as the foundation pit walls and floor plates will deform with changes in the soil, changes in groundwater, and the impact of the construction process. Especially in deep foundation pit projects, the deformation problem becomes more serious as the depth increases. Therefore, real-time deformation monitoring, analysis, and control of foundation pits are particularly important. Deformation control of deep foundation pits usually requires the use of a series of monitoring equipment, calculation models, and control methods to ensure the safety and stability of the foundation pits during construction.

[0004] However, in the field of deep foundation pit deformation control, there are the following problems:

[0005] (1) Monitoring data lacks systematicity and comprehensiveness:

[0006] At present, the traditional methods for monitoring foundation pit deformation mainly include:

[0007] Settlement monitoring: Use ground settlement meters, inclinometers, laser scanners and other equipment to monitor the settlement of the ground or structure around the foundation pit. Displacement monitoring: Obtain the deformation of the foundation pit by measuring the displacement of the point (such as settlement, horizontal displacement, etc.). Steel support force monitoring: Use sensors to monitor the axial force changes of the steel support and evaluate the stability of the support structure. However, the data provided by traditional methods are mostly static, single-point observation results, lacking real-time dynamic monitoring and comprehensive prediction of foundation pit deformation.

[0008] (2) Lack of dynamic prediction and optimization capabilities:

[0009] During the design and construction of foundation pit projects, the deformation of foundation pits is often estimated with the help of numerical simulation technology (such as finite element analysis). By constructing a mathematical model of the foundation pit and the surrounding soil and structure, combined with the load and soil properties during the construction process, the deformation trend is predicted. Although this method can provide a basis for design, due to the complexity and uncertainty of the simulation assumptions, it can often only provide approximate results and is difficult to adapt to dynamic changes during the construction process.

[0010] (3) Limitations and locality of monitoring equipment

[0011] With the development of sensor technology, monitoring data collection is gradually moving towards automation and real-time. A large amount of field data is collected through sensor networks (such as soil strain gauges, displacement gauges, etc.), and real-time evaluation and early warning of foundation pit deformation are carried out through data analysis and prediction models. This method improves the real-time and accuracy of monitoring to a certain extent, but it often lacks multi-factor and global coordination and optimization.

[0012] (4) Lack of intelligent feedback and closed-loop control mechanism

[0013] Although some monitoring systems can provide real-time data feedback, traditional deformation monitoring systems often lack intelligent feedback mechanisms with the construction process. For example, when the monitoring data reaches a certain threshold, the system may not be able to automatically adjust the construction plan or issue an early warning, lacking real-time and dynamic feedback control. Summary of the invention

[0014] The present invention provides a deep foundation pit stability deformation control method and device based on digital twin to overcome the above-mentioned technical problems.

[0015] In order to achieve the above object, the technical solution of the present invention is:

[0016] A deep foundation pit stability deformation control method based on digital twins specifically includes the following steps:

[0017] S1: Collect and obtain the monitoring time series data of deep foundation pit;

[0018] The monitoring time series data at least includes foundation pit surface settlement data, ground-connected wall horizontal displacement data and steel support axial force data;

[0019] The surface settlement data of foundation pit and horizontal displacement data of ground-connected wall are used as monitoring time series characteristic data;

[0020] The steel support axial force data is used as monitoring time series label data to obtain a sample data set;

[0021] S2: Divide the sample data set into a data training set, a data verification set and a data test set according to a preset ratio;

[0022] S3: Based on the sample data set, the deep foundation pit deformation control digital twin platform is constructed to realize the deep foundation pit stability deformation control;

[0023] The deep foundation pit deformation control digital twin platform includes a data input module, a foundation pit deformation prediction module, a steel support servo module, and a foundation pit BIM module, which are connected in sequence;

[0024] The data input module is used to transmit the monitoring time series data of the deep foundation pit to the foundation pit deformation prediction module;

[0025] The foundation pit deformation prediction module is used to construct a CNN-LSTM composite neural network based on the Adam optimization algorithm, and to train / verify the CNN-LSTM composite neural network through the data training set and the data verification set to obtain the optimized CNN-LSTM composite neural network;

[0026] Based on the optimized CNN-LSTM composite neural network, the time series change prediction of the steel support axial force is carried out according to the data test set to obtain the prediction results;

[0027] The steel support servo module is used to realize the servo control of the steel support axial force according to the prediction results based on the constructed servo control strategy;

[0028] The foundation pit BIM module is used to construct a deep foundation pit BIM model based on Revit modeling technology to control the deformation of the deep foundation pit through the steel support axial force servo system. According to the output of the steel support servo module, it interacts with the deep foundation pit BIM model in real time through the Revit API data interface to realize the visualization of the deformation control of the deep foundation pit.

[0029] Furthermore, the CNN-LSTM composite neural network based on the Adam optimization algorithm constructed in S3 consists of

[0030] Data input layer, convolution layer, pooling layer, LSTM network layer, fully connected layer and output layer;

[0031] The data input layer is used to transmit the acquired monitoring time series data of the deep foundation pit to the convolution layer;

[0032] The convolution layer is used to extract the multi-dimensional local spatiotemporal data features of the foundation pit surface settlement and the horizontal displacement of the ground-connected wall from the monitoring time series data;

[0033] The pooling layer is used to perform dimension reduction / compression operations on local spatiotemporal data features to obtain optimized local spatiotemporal data features;

[0034] The LSTM network layer is used to further extract the time series relationship characteristics of the foundation pit surface settlement and the horizontal displacement of the ground-connected wall according to the optimized local spatiotemporal data characteristics, and obtain time series data;

[0035] The fully connected layer is used to perform feature fusion on the time series data to obtain fusion features;

[0036] The feature fusion step specifically includes performing a one-dimensional flattening operation on the time series data through the Flatten function to obtain a one-dimensional data vector; performing a linear transformation on the one-dimensional data vector through a set weight matrix and bias, and obtaining a weighted sum of the one-dimensional data vector after the linear transformation;

[0037] The output layer is used to predict and output the steel support axial force data according to the fusion features.

[0038] Furthermore, S3 uses the data training set to train the CNN-LSTM composite neural network model, which specifically includes the following steps:

[0039] S31: define the foundation pit surface settlement data and the ground-connected wall horizontal displacement data as input data, and define the steel support axial force data as output data;

[0040] The data training set is input into the CNN-LSTM composite neural network for model training to obtain the trained CNN-LSTM composite neural network;

[0041] S32: taking the mean square error function between the predicted steel support axial force data and the actual steel support axial force data as the model loss function;

[0042] The trained CNN-LSTM composite neural network is model verified through the data verification set, and whether the output of the trained CNN-LSTM composite neural network converges is determined based on the model loss function;

[0043] If it is confirmed that the output of the trained CNN-LSTM composite neural network converges, then the trained CNN-LSTM composite neural network is the optimized CNN-LSTM composite neural network;

[0044] Otherwise, based on the Adam optimization algorithm, the network parameters of the trained CNN-LSTM composite neural network are adaptively updated, and step S31 is iteratively executed repeatedly.

[0045] Furthermore, the network parameters of the trained CNN-LSTM composite neural network are adaptively updated based on the Adam optimization algorithm described in S32, specifically including the following steps:

[0046] S321: define the initialization first-order moment and second-order moment of the network parameter gradient;

[0047] S322: Iteratively update the initialized first-order moment and second-order moment, and obtain the updated first-order moment and second-order moment; and iteratively update and obtain the expression of the updated first-order moment and second-order moment:

[0048] m t =β1m t-1 +(1-β1)g t

[0049]

[0050] Where: m t With v t They represent the exponential moving average of the gradient and the square of the gradient, i.e., the first-order moment and the second-order moment, respectively; β1 and β2 represent m t With v t The exponential decay rate hyperparameter of g t represents the gradient of the network parameters at time step t; Represents the gradient of the model loss function J(θ) with respect to the network parameter θ; that is, for each network parameter θ t Calculate a vector formed by the partial derivatives of the loss function;

[0051] S323: Obtaining a deviation between the first-order moment and the second-order moment according to the updated first-order moment and second-order moment;

[0052] And the expression for obtaining the deviation between the first-order moment and the second-order moment is

[0053]

[0054] Where: and Represent the deviations of the first-order moment and the second-order moment respectively; and Respectively represent the time step t, corresponding to m t With v t The exponential decay rate hyperparameter of ;

[0055] S324: Adaptively update the network parameters corresponding to each time step t in the CNN-LSTM composite neural network according to the deviation between the first-order moment and the second-order moment;

[0056] And the expression for adaptive update of network parameters is

[0057]

[0058] Where: θ t-1 represents the network parameters corresponding to the time step t-1; θ t Represents the network parameters corresponding to the time step t; α represents the learning rate; ε represents a positive design parameter.

[0059] Furthermore, the servo control strategy constructed in S3 specifically includes the following steps:

[0060] S001: Preset steel support axial force threshold;

[0061] S002: Based on the optimized CNN-LSTM composite neural network, the time series change prediction of the steel support axial force is carried out according to the data test set to obtain the current prediction results;

[0062] Determine whether the current prediction result meets the preset steel support axial force threshold;

[0063] If the predicted result is greater than the preset steel support axial force threshold, the steel support axial force servo system will be used to unload the steel support axial force on the deep foundation pit;

[0064] If the predicted result is less than the preset steel support axial force threshold, the steel support axial force servo system is used to load the steel support axial force on the deep foundation pit;

[0065] If the predicted result is equal to the preset steel support axial force threshold, the steel support axial force for the deep foundation pit is maintained through the steel support axial force servo system.

[0066] A device for a deep foundation pit stability deformation control method based on digital twins, comprising a steel support axial force servo system, a data acquisition box, a box-type fixed inclinometer, a steel support structure, a static level, and a deep foundation pit deformation control digital twin platform;

[0067] The steel support shaft force servo system includes a computer control system, a numerical control pump station and a support head assembly component; the support head assembly component at least includes a hydraulic jack and a steel support shaft force meter;

[0068] The steel support structure is arranged transversely inside the deep foundation pit, and both ends of the steel support structure are abutted against both ends of the ground connection wall of the deep foundation pit;

[0069] The data acquisition box and CNC pump station are respectively set on the ground on both sides of the deep foundation pit;

[0070] The static level is set on the ground at one side of the deep foundation pit and connected to the data acquisition box to monitor the surface settlement data of the foundation pit;

[0071] Box-type fixed inclinometers are installed on the ground-connected walls on both sides of the deep foundation pit, and are distributed on both sides of the steel support and connected to the data acquisition box to monitor the horizontal displacement data of the ground-connected walls;

[0072] The support head assembly and the steel support axial force gauge are respectively arranged at both ends of the steel support structure;

[0073] The numerical control pump station is connected to the support head assembly and the computer control system respectively;

[0074] The computer control system is used to obtain a drive control signal based on a servo control strategy, and to drive a hydraulic jack to control the axial force of a steel support of a deep foundation pit by controlling a numerically controlled pump station based on the drive control signal;

[0075] The steel support axial force meter is used to detect and obtain the steel support axial force of deep foundation pit in real time;

[0076] The control signal is a signal for controlling the axial force of the steel support of the deep foundation pit obtained in real time by the steel support axial force meter and a preset steel support axial force threshold value, so as to control the numerical control pump station to drive the hydraulic jack to realize the control of the axial force of the steel support of the deep foundation pit;

[0077] The deep foundation pit deformation control digital twin platform is communicated with the steel support axial force servo system and the data acquisition box through preset wireless modules.

[0078] Beneficial effect: The present invention provides a deep foundation pit stability deformation control method and device based on digital twin, which realizes real-time dynamic monitoring and comprehensive prediction of foundation pit deformation by acquiring monitoring time series data of deep foundation pit, and realizes dynamic prediction of deep foundation pit deformation by constructing a deep foundation pit deformation control digital twin platform, and realizes servo control of steel support axial force according to dynamic prediction results based on the constructed servo control strategy through the steel support servo module; a deep foundation pit BIM model for controlling the deformation of the deep foundation pit through the steel support axial force servo system is constructed through the foundation pit BIM module, and according to the output of the steel support servo module, real-time interaction with the deep foundation pit BIM model is carried out through the RevitAPI data interface to realize visualization of deep foundation pit deformation control, so that while improving the deformation control accuracy of the deep foundation pit, intelligent interaction of the deep foundation pit deformation can also be realized, which greatly improves the intelligence level of deep foundation pit stability deformation control based on digital twin. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0080] Figure 1 It is a flow chart of the deep foundation pit stability deformation control method based on digital twin of the present invention;

[0081] Figure 2 Schematic diagram of a deep foundation pit stability deformation control device based on digital twin in this embodiment;

[0082] Figure 3 It is the core block diagram of the deep foundation pit stability deformation control method of the digital twin in this embodiment;

[0083] Figure 4 This is a flow chart of the data input module in this embodiment;

[0084] Figure 5 This is a flow chart of the foundation pit deformation prediction module in this embodiment;

[0085] Figure 6 This is a flow chart of the steel support servo module in this embodiment;

[0086] Figure 7 This is a flow chart of the foundation pit BIM module in this embodiment;

[0087] Figure 8 Schematic diagram of a recurrent neural network in this embodiment;

[0088] Fig. 9 Schematic diagram of the structure of the long short-term memory neural network LSTM in this embodiment.

[0089] In the figure: 1. Support head assembly components; 2. Steel support structure; 3. Box-type fixed inclinometer; 4. Steel support axial force gauge; 5. Static level; 6. Data acquisition box; 7. Deep foundation pit deformation control digital twin platform; 8. CNC pump station; 9. Wireless module; 10. Foundation pit ground-connected wall. DETAILED DESCRIPTION

[0090] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0091] This embodiment provides a deep foundation pit stability deformation control method based on digital twins, such as Figure 1 As shown, the specific steps include:

[0092] S1: Collect and obtain the monitoring time series data of deep foundation pit;

[0093] The monitoring time series data at least includes foundation pit surface settlement data, ground-connected wall horizontal displacement data and steel support axial force data;

[0094] The surface settlement data of foundation pit and horizontal displacement data of ground-connected wall are used as monitoring time series characteristic data;

[0095] The steel support axial force data is used as monitoring time series label data to obtain a sample data set;

[0096] Specifically, Figure 4 As shown, it also includes preprocessing of the monitoring time series data of the deep foundation pit:

[0097] And the preprocessing is to save the collected monitoring time series data of the deep foundation pit into a preset SQL server database, and call the monitoring time series data in the SQL server database through the data input module of the deep foundation pit deformation control digital twin platform, and the deep foundation pit deformation control digital twin platform can display data and charts, and can perform operations such as adding, deleting, checking, and modifying the monitoring data. The deep foundation pit deformation control digital twin platform is obtained through secondary development based on Revit modeling technology, and the adding, deleting, checking, and modifying operations of the monitoring data are realized through programming, and the programming method or implementation means for performing the operations such as adding, deleting, checking, and modifying the monitoring data are existing publicly known technical means, which will not be described in detail here;

[0098] S2: Divide the sample data set into a data training set, a data verification set and a data test set according to a preset ratio;

[0099] S3: Based on the sample data set, the deep foundation pit deformation control digital twin platform is constructed to realize the deep foundation pit stability deformation control;

[0100] The deep foundation pit deformation control digital twin platform includes a data input module, a foundation pit deformation prediction module, a steel support servo module, and a foundation pit BIM module, which are connected in sequence;

[0101] The data input module is used to transmit the monitoring time series data of the deep foundation pit to the foundation pit deformation prediction module;

[0102] The foundation pit deformation prediction module is used to construct a CNN-LSTM composite neural network based on the Adam optimization algorithm, and to train / verify the CNN-LSTM composite neural network through the data training set and the data verification set to obtain the optimized CNN-LSTM composite neural network;

[0103] In a specific embodiment, Figure 3 As shown in the figure, the constructed CNN-LSTM composite neural network based on the Adam optimization algorithm consists of a data input layer, a convolution layer, a pooling layer, an LSTM network layer, a fully connected layer, and an output layer connected in sequence;

[0104] The data input layer is used to transmit the acquired monitoring time series data of the deep foundation pit to the convolution layer;

[0105] The convolution layer is used to extract the multi-dimensional local spatiotemporal data features of the foundation pit surface settlement and the horizontal displacement of the ground-connected wall from the monitoring time series data;

[0106] The pooling layer is used to perform dimension reduction / compression operations on local spatiotemporal data features to obtain optimized local spatiotemporal data features;

[0107] The LSTM network layer is used to further extract the time series relationship characteristics of the foundation pit surface settlement and the horizontal displacement of the ground-connected wall according to the optimized local spatiotemporal data characteristics, and obtain time series data;

[0108] Specifically, the long short-term memory neural network (LSTM) is a variant of the recurrent neural network (RNN). The traditional RNN model has a relatively simple network structure, which only includes an input layer, a hidden layer, and an output layer. Figure 8 However, when the weights in the network (such as u and v) are set too large or too small, gradient explosion or gradient disappearance problems will occur, which will affect the prediction accuracy of the model and cause large deviations in the prediction results. Fig. 9 As shown in the figure, the principle of LSTM (Long Short-Term Memory) neural network is based on its unique gating mechanism. It manages the flow of information in time series through three key components: forget gate, input gate and output gate. The forget gate allows the network to forget past information, the input gate selectively adds new information to the memory unit based on the current input and previous state, and the output gate determines which information should be output to the next layer. This structural design enables LSTM to effectively process long sequence data and solves the problem of gradient disappearance or gradient explosion in traditional RNN;

[0109] In the LSTM neural network, each neuron can be regarded as a basic decision-making unit, which uses activation functions (such as Sigmoid or ReLU) to decide whether to let information pass. They determine the retention and update ratio of information. The number of neurons is an important hyperparameter that needs to be carefully adjusted to obtain the best performance. Generally speaking, the number of neurons can be calculated by the following formula:

[0110] A=C·(m+n)

[0111] Where: C is a constant; M is the dimension of the feature; N is the number of output categories or the dimension of the target value; in this embodiment, a set of steel support axial forces are predicted by inputting two sets of foundation pit ground-connected wall horizontal displacements and two sets of foundation pit surrounding surface settlements. For example, the input feature number is 4, and the dimension of a set of data is 1, M-4, N=1; in actual applications, the number of neurons can also be adjusted according to experience and experiments, and different numbers can be tried and the best configuration can be selected through methods such as cross-validation;

[0112] Among them, the convolutional neural network (CNN) is a deep learning model, similar to the multi-layer perceptron of artificial neural networks, and is particularly suitable for processing data with a grid-like structure. The core principle is to use convolutional layers, activation functions and pooling layers to extract and abstract the features of input data (the features include low-level features, intermediate features and high-level features: Low-level features (primary abstraction): CNN may capture the changing pattern of foundation pit settlement in the initial convolutional layer, such as faster settlement in a certain area, or the horizontal displacement of the ground-connected wall presents a certain regularity. For example, the convolutional layer may recognize some simple patterns, such as local settlement changes, displacement fluctuations, etc. Intermediate features (gradual abstraction): As the network goes deeper, CNN begins to combine multiple low-level features to form more meaningful intermediate features. For example, it may be possible to identify the correlation between a certain settlement pattern and the displacement of the ground-connected wall, and speculate that the increased settlement in some areas may be related to changes in the support system. High-level features (final abstraction): In deeper networks, CNN will integrate high-level features related to the axial force of steel supports from all low-level and intermediate features. For example, the complex relationship between the overall settlement trend of the foundation pit and the support axial force, or the nonlinear relationship between the overall change in the displacement of the ground-connected wall and the strength of the support axial force may be identified.). The convolution layer uses a set of learnable filters (or convolution kernels) to slide on the input data and perform local perception and feature extraction to obtain multi-dimensional local spatiotemporal data features; the pooling layer reduces the dimension of the feature map by downsampling while retaining important features and enhancing the translation invariance of the model. Finally, these features are sent to the fully connected layer for classification or regression tasks. The principle of CNN is as follows: Figure 3 shown.

[0113] The convolution layer is the core structure of CNN, and the selection of the convolution kernel size needs to be reasonably designed according to the task requirements and the size of the features to improve the performance and generalization ability of the network. In each convolution layer, the calculation formula for the feature matrix size extracted after convolution is as follows:

[0114] w=s·(w+2p-k)+1

[0115] Where: w represents the input matrix size; k represents the convolution kernel size; s represents the stride length; p represents the number of zero-padding layers;

[0116] The convolutional layer extracts features from the monitoring time series data of the deep foundation pit. These operations can identify and capture local patterns in the data. In this way, the convolutional layer can extract key local spatiotemporal data features from the multidimensional time series, not just spatial features. The key local spatiotemporal data features are specifically assumed to include multidimensional time series data: foundation pit settlement, ground-connected wall displacement, and steel support axial force, and they all change over time. The convolutional layer can extract the following spatiotemporal features:

[0117] Temporal features: The convolutional layer can extract local temporal change patterns from each time series, such as the changing trend of foundation pit settlement and the sudden change of ground-connected wall displacement.

[0118] Spatial features: The convolutional layer can learn the interdependencies between different measurement points (e.g., foundation pit settlement and ground-anchored wall displacement at different locations) and capture spatial patterns.

[0119] Spatiotemporal features: The convolutional layer can identify time-space interaction features, such as how the change in foundation pit settlement at a certain moment affects the displacement of the ground-connected wall at different locations, and then affects the change in the axial force of the steel support;

[0120] The fully connected layer is used to perform feature fusion on the time series data to obtain fusion features;

[0121] The feature fusion step specifically includes performing a one-dimensional flattening operation on the time series data through the Flatten function to obtain a one-dimensional data vector; performing a linear transformation on the one-dimensional data vector through a set weight matrix and bias, and obtaining a weighted sum of the one-dimensional data vector after the linear transformation;

[0122] The output layer is used to predict and output the steel support axial force data according to the fusion features;

[0123] In a specific embodiment, Figure 5 As shown in the figure, the CNN-LSTM composite neural network model is trained through the data training set, which specifically includes the following steps

[0124] S31: define the foundation pit surface settlement data and the ground-connected wall horizontal displacement data as input data, and define the steel support axial force data as output data;

[0125] The data training set is input into the CNN-LSTM composite neural network for model training to obtain the trained CNN-LSTM composite neural network;

[0126] S32: S32: The mean square error function between the predicted steel support axial force data and the actual steel support axial force data is used as the model loss function;

[0127] The trained CNN-LSTM composite neural network is model verified through the data verification set, and whether the output of the trained CNN-LSTM composite neural network converges is determined based on the model loss function;

[0128] If it is confirmed that the output of the trained CNN-LSTM composite neural network converges, then the trained CNN-LSTM composite neural network is the optimized CNN-LSTM composite neural network;

[0129] Otherwise, based on the Adam optimization algorithm, the network parameters of the trained CNN-LSTM composite neural network are adaptively updated, and step S31 is iteratively executed repeatedly;

[0130] Among them, the Adam optimization algorithm, as a stochastic optimization method with adaptive momentum, is often used for gradient descent to perform first-order gradient optimization on random objective functions. Compared with traditional stochastic gradient descent, it can adaptively adjust the learning rate of each parameter, and at the same time has the characteristics of momentum optimization, with faster convergence speed and better generalization performance when training models. Traditional stochastic gradient descent cannot change the learning rate during training, while the Adam optimizer maintains two dynamic variables: the first-order moment estimate of the gradient (that is, the average value of the gradient) and the second-order moment estimate (that is, the average value of the square of the gradient). By taking a weighted average of these two dynamic variables during iteration, different parameter corrections are obtained to achieve parameter updates;

[0131] In a specific embodiment, in S32, based on the Adam optimization algorithm, the network parameters of the trained CNN-LSTM composite neural network are adaptively updated, which specifically includes the following steps:

[0132] S321: define the initialization first-order moment and second-order moment of the network parameter gradient;

[0133] S322: Iteratively update the initialized first-order moment and second-order moment, and obtain the updated first-order moment and second-order moment; and iteratively update and obtain the expression of the updated first-order moment and second-order moment:

[0134] m t =β1m t-1 +(1-β1)g t

[0135]

[0136] Where: m t With v t They represent the exponential moving average of the gradient and the square of the gradient, i.e., the first-order moment and the second-order moment, respectively; β1 and β2 represent m t With v t The exponential decay rate hyperparameters of g are set to 0.9 and 0.999 respectively; t represents the gradient of the network parameters at time step t; Represents the gradient of the model loss function J(θ) with respect to the network parameter θ; that is, for each network parameter θ t Calculate a vector formed by the partial derivatives of the loss function;

[0137] S323: Since both the first-order moment and the second-order moment are initialized to 0, in order to offset the deviation, Adam corrects the moving average deviation by the following formula, that is, according to the updated first-order moment and second-order moment, obtains the deviation between the first-order moment and the second-order moment;

[0138] And the expression for obtaining the deviation between the first-order moment and the second-order moment is

[0139]

[0140] Where: and Represent the deviations of the first-order moment and the second-order moment respectively; and Respectively represent the time step t, corresponding to m t With v t The exponential decay rate hyperparameter of ;

[0141] S324: Adaptively update the network parameters corresponding to each time step t in the CNN-LSTM composite neural network according to the deviation between the first-order moment and the second-order moment;

[0142] And the expression for adaptive update of network parameters is

[0143]

[0144] Where: θ t-1 represents the network parameters corresponding to the time step t-1; θ t represents the network parameters corresponding to the time step t; α represents the learning rate and the default value is α=0.001, ε represents the positive design parameter and the default value is 10 -8 , used to prevent division by zero;

[0145] Based on the optimized CNN-LSTM composite neural network, the time series change prediction of the steel support axial force is carried out according to the data test set to obtain the prediction results;

[0146] Although the LSTM model in this embodiment performs well in processing time series data, its processing ability for spatial features is limited. At the same time, a single LSTM model is prone to gradient vanishing or explosion when training on long sequence data, resulting in slow convergence. By combining CNN with LSTM, this embodiment can effectively extract spatial features and reduce the burden of the LSTM layer, so that the model converges faster and has stronger stability when processing long time series data. Nevertheless, this combined model still has certain limitations. In order to solve this problem, this embodiment further introduces the Adam optimization algorithm to make the model perform better. The Adam optimization algorithm can dynamically adjust the learning rate according to the parameter changes of different layers through an adaptive learning rate mechanism, accelerate the convergence speed, and avoid the slow training or local optimal situation caused by a fixed learning rate in a single CNN-LSTM model. In addition, since the deep foundation pit deformation prediction data is complex and has spatiotemporal characteristics, the traditional deep model is prone to gradient oscillation or instability in a complex environment. By introducing the momentum mechanism of the Adam optimization algorithm, the network parameter update can be smoothed, the gradient oscillation of the CNN and LSTM layers can be reduced, and the overall stability of the model can be improved. When processing long time series data, Adam combines the first-order moment and the second-order moment to effectively alleviate the problem of gradient vanishing or exploding, making the CNN-LSTM compound neural network more accurate when facing complex, multi-dimensional spatiotemporal data. In addition, the Adam optimization algorithm reduces the workload of manual parameter adjustment and is more efficient than a single CNN-LSTM compound neural network. Especially in engineering applications, it can save time and resources and greatly improve the convenience and effect of actual operations.

[0147] like Figure 6 As shown, the steel support servo module is used to realize the servo control of the steel support axial force according to the prediction results based on the constructed servo control strategy;

[0148] Specifically, the constructed servo control strategy includes the following steps:

[0149] S001: Preset steel support axial force threshold;

[0150] S002: Based on the optimized CNN-LSTM composite neural network, the time series change prediction of the steel support axial force is carried out according to the data test set to obtain the current prediction results;

[0151] Determine whether the current prediction result meets the preset steel support axial force threshold;

[0152] If the predicted result is greater than the preset steel support axial force threshold, the steel support axial force servo system will be used to unload the steel support axial force on the deep foundation pit;

[0153] If the predicted result is less than the preset steel support axial force threshold, the steel support axial force servo system is used to load the steel support axial force on the deep foundation pit;

[0154] If the predicted result is equal to the preset steel support axial force threshold, the steel support axial force servo system is used to maintain the steel support axial force on the deep foundation pit;

[0155] The foundation pit BIM module is used to build a deep foundation pit BIM model based on Revit modeling technology to control the deformation of the deep foundation pit through the steel support axial force servo system, and interact with the deep foundation pit BIM model in real time through the Revit API data interface according to the output of the steel support servo module to realize the visualization of the deformation control of the deep foundation pit. Figure 7 As shown, in the process of unloading or loading the axial force of the steel support of the deep foundation pit, the axial force data loaded by the steel support servo system is also interacted with the deep foundation pit BIM model through the Revit API, and the deformation of the foundation pit model is observed in real time, so as to visualize the deformation of the foundation pit. After the steel support servo module is used in this embodiment, the final deformation of the foundation pit can be viewed in the BIM module, and the data in the whole process of deformation control are interacted with the deep foundation pit BIM model in real time to realize the visualization of the control process. In addition, an early warning system is also integrated in the data twin platform. The early warning system is used to provide automatic alarm prompts when the monitoring time series data of the deep foundation pit exceeds the preset early warning range, and set the threshold (axial force limit or displacement limit) as needed. The early warning system compares the data / predicted data obtained from the actual sensor with the set steel support axial force threshold, and sets the alarm level according to the difference between the data and the set steel support axial force threshold. The alarm level can be distinguished by color and displayed in color in the deep foundation pit BIM model. The implementation method or means of the alarm level setting and color display are existing well-known technical means, which will not be described in detail here.

[0156] This embodiment acquires monitoring time series data of deep foundation pits to realize real-time dynamic monitoring and comprehensive prediction of foundation pit deformation. By constructing a digital twin platform for deep foundation pit deformation control, dynamic prediction of deep foundation pit deformation is realized based on a CNN-LSTM composite neural network based on the Adam optimization algorithm. Servo control of the steel support axial force is realized according to the dynamic prediction result based on the constructed servo control strategy through the steel support servo module. A deep foundation pit BIM model for controlling the deformation of the deep foundation pit through the steel support axial force servo system is constructed through the foundation pit BIM module. According to the output of the steel support servo module, real-time interaction with the deep foundation pit BIM model is carried out through the Revit API data interface to realize visualization of the deformation control of the deep foundation pit. This improves the accuracy of deep foundation pit deformation control and realizes intelligent interaction of deep foundation pit deformation, greatly improving the intelligence of deep foundation pit stability deformation control based on digital twins.

[0157] A deep foundation pit stability deformation control device based on digital twin, such as Figure 2 As shown, it includes a steel support axial force servo system, a data acquisition box 6, a box-type fixed inclinometer 3, a steel support structure 2, a static level 5 and a deep foundation pit deformation control digital twin platform 7;

[0158] The steel support shaft force servo system includes a computer control system, a numerical control pump station 8 and a support head assembly component 1; the support head assembly component 1 at least includes a hydraulic jack and a steel support shaft force meter 4;

[0159] In a specific embodiment, the steel support shaft force servo system mainly includes three parts:

[0160] The first part is the host, which is the computer control system. The computer control system consists of a program-controlled host and a display, and can transmit data and adjust the axial force with the on-site CNC pump station. The computer control system is integrated on the digital twin platform, which means that the steel support servo system can be controlled on the platform.

[0161] The second part is the CNC pump station, which at least includes the existing control cabinet and hydraulic pump, etc. The control cabinet is composed of a series of mechanical and electronic components, including PLC controller, variable frequency motor and wireless module, and the hydraulic pump includes hydraulic valve assembly, power supply assembly, AC contactor, cable and oil pipe interface, etc. The CNC pump station serves as an intermediate link to connect the program control host and the support head assembly component 1, and transmit information between the two to realize the measurement and control of the steel support axial force;

[0162] The third part is the support head assembly component 1, which is connected to the steel support structure 2 and installed at the set position of the deep foundation pit retaining structure. The support head assembly component 1 is connected to the CNC pump station through oil pipes and cables. The support head assembly component 1 includes at least a hydraulic jack, a steel support shaft force meter 4, and ultrasonic displacement sensors and other detection elements;

[0163] This embodiment also includes the steps of installing the steel support shaft force servo system:

[0164] S100: Install the steel purlin of the deep foundation pit. The steel purlin uses two I-beams and is supported by a triangular bracket on the same horizontal plane. Each section of the steel purlin is welded as a whole. The space between the steel purlin and the ground-connected wall 10 is filled with cement mortar to ensure that the steel purlin is stable under stress. While installing the steel purlin on site, the support head assembly component 1 is installed on the ground. The support head uses bolts of the same size and spacing as the steel support. One end of the support head is directly equipped with a steel plate, which is connected with high-strength bolts and is convenient and reliable.

[0165] S101: A crawler crane is used on site to hoist the connected steel support, i.e., the steel support structure 2, to the area where it needs to be installed, and slowly lowers it onto the steel purlin. At the same time, the steel support is manually adjusted to the preset position of the deep foundation pit;

[0166] S102: After the steel support structure 2 is installed, the CNC pump station is immediately connected to the support head assembly component 1 through the reserved interface, mainly connecting the oil pipeline and the pressure and displacement sensors. After the connection is completed, the support head assembly component 1 is loaded according to the set axial force value through the CNC pump station placed around the deep foundation pit. When the support axial force or displacement value reaches the designed value, the loading is stopped. At this time, the double mechanical locks on both sides of the jack are automatically locked. After the axial force threshold is set on the digital twin platform, the automatic axial force monitoring is turned on. The axial force result predicted by the prediction system is compared with the set threshold to ensure that the steel support axial force is always within the preset steel support axial force threshold, so as to meet the requirements of deep foundation pit deformation control;

[0167] S103: The support head assembly component 1 is equipped with a pressure sensor and an ultrasonic displacement sensor to monitor the axial force and displacement of the steel support; the digital twin platform is dynamically adjusted according to the set design axial force value and the foundation pit deformation prediction module. When the actual axial force of the steel support structure is detected to be lower than the preset steel support axial force threshold, the steel support axial force servo system will start to automatically compensate for the axial force; when the axial force value of the steel support is detected to be greater than the preset steel support axial force threshold, the steel support axial force servo system will automatically reduce the axial force to keep it within the design value range, so as to ensure that the deformation of the foundation pit will not be too large. At the same time, the steel support axial force servo system can also manually mechanically drive the hydraulic jack at the monitoring end to adjust the size of the steel support axial force value to adapt to the steel support axial force value under different working conditions;

[0168] The steel support structure 2 is transversely arranged inside the deep foundation pit, and the two ends of the steel support structure 2 abut against the two ends of the ground-connected wall 10 of the deep foundation pit;

[0169] The data acquisition box 6 and the numerical control pump station are respectively arranged on the ground at both sides of the deep foundation pit;

[0170] The static level 5 is set on the ground at one side of the deep foundation pit and is connected to the data acquisition box 6 to monitor the surface settlement data of the foundation pit;

[0171] The box-type fixed inclinometers 3 are arranged on the ground-connected walls 10 on both sides of the deep foundation pit, and are distributed on both sides of the steel support, and are connected to the data acquisition box 6 to monitor the horizontal displacement data of the ground-connected walls 10;

[0172] The support head assembly component 1 and the steel support axial force gauge 4 are respectively arranged at both ends of the steel support structure 2;

[0173] The numerical control pump station is respectively connected with the support head assembly component 1 and the computer control system;

[0174] The computer control system is used to obtain a drive control signal based on a servo control strategy, and to drive a hydraulic jack to control the axial force of a steel support of a deep foundation pit by controlling a numerically controlled pump station based on the drive control signal;

[0175] The steel support axial force meter 4 is used to detect and obtain the steel support axial force of the deep foundation pit in real time;

[0176] The control signal is a signal for controlling the axial force of the steel support of the deep foundation pit obtained in real time by the steel support axial force meter 4 and a preset steel support axial force threshold value to control the numerical control pump station 8 to drive the hydraulic jack to achieve control of the axial force of the steel support of the deep foundation pit;

[0177] The deep foundation pit deformation control digital twin platform 7 is respectively connected to the steel support axial force servo system and the data acquisition box 6 through a preset wireless module 9.

[0178] The simulation test of this embodiment is as follows:

[0179] A subway station in Ningbo adopts the open excavation method. The deep foundation pit stability deformation control device based on digital twin is used for the construction monitoring of the second steel support at a certain place on site. The monitoring time series data of the deep foundation pit monitoring point are obtained through the device. The monitoring data of the surface settlement monitoring points SD9-2 and SD10-2, the monitoring data of the horizontal displacement monitoring points CX09 and CX10 of the ground-connected wall, and the monitoring data of the steel support axial force monitoring point Zg1-2 from March 12, 2023 to May 31, 2023 are selected, totaling 5 groups of data, each with 81 data. The collected raw data are de-noised, and the pre-processed data are integrated into a multivariate time series. The steel support axial force is selected as the prediction target and set as the dependent variable. The other variables include the monitoring data of each monitoring point and 4 variables as independent variables. The first 70 data of the multivariate time series are selected as input, and the last 11 steel support axial forces are compared with the predicted values ​​of the model to realize the monitoring of multivariate information of a monitoring point, which effectively guides the on-site construction. The monitoring results are shown in Table 1.

[0180] Table 1. Monitoring time series data of deep foundation pit

[0181]

[0182]

[0183] According to the monitoring time series data obtained by the deep foundation pit stability deformation control device based on digital twin, the steel support axial force is predicted in the later stage. 70 groups of monitoring data are selected as the time series training set to obtain the steel support axial force prediction model. 11 groups of monitoring data are selected as the test set of the time series. Table 2 shows the prediction results of the steel support axial force in the later stage. It can be seen from the data in Table 2 that the maximum prediction error is 2.82%, which is relatively small and can meet the accuracy requirements of the project, indicating that this embodiment has a relatively good reliability in predicting the steel support axial force of the monitoring time series data.

[0184] Table 2. Prediction results of steel support axial forces

[0185] Serial number Axial force monitoring value / KN Axial force prediction results / KN Relative error / % 70 807.0 820.22 1.61 71 818.7 833.97 1.83 72 688.7 670.63 2.62 73 599.2 610.75 1.89 74 759.1 738.29 2.82 75 822.2 811.80 1.28 76 823.4 813.05 1.26 77 860.7 867.89 0.83 78 1061.3 1078.91 1.63 79 1076.2 1083.48 0.67 80 661.7 674.15 1.85 81 736.8 756.94 2.66

[0186] The specific prediction results are shown in Table 2. The advantages of the CNN-LSTM composite neural network based on the Adam optimization algorithm for steel support axial force prediction are its powerful nonlinear processing capabilities, multi-source data fusion, automatic feature extraction, and adaptive learning capabilities. These characteristics enable the CNN-LSTM composite neural network to have higher prediction accuracy and robustness in the face of deep foundation pits under complex engineering conditions, and can effectively cope with the deformation of deep foundation pits under complex environments.

[0187] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A deep foundation pit stability deformation control method based on digital twin, characterized in that: The specific steps include: S1: Collect and obtain the monitoring time series data of deep foundation pit; The monitoring time series data at least includes foundation pit surface settlement data, ground-connected wall horizontal displacement data and steel support axial force data; The surface settlement data of foundation pit and horizontal displacement data of ground-connected wall are used as monitoring time series characteristic data; The steel support axial force data is used as monitoring time series label data to obtain a sample data set; S2: Divide the sample data set into a data training set, a data verification set and a data test set according to a preset ratio; S3: Based on the sample data set, the deep foundation pit deformation control digital twin platform is constructed to realize the deep foundation pit stability deformation control; The deep foundation pit deformation control digital twin platform includes a data input module, a foundation pit deformation prediction module, a steel support servo module, and a foundation pit BIM module, which are connected in sequence; The data input module is used to transmit the monitoring time series data of the deep foundation pit to the foundation pit deformation prediction module; The foundation pit deformation prediction module is used to construct a CNN-LSTM composite neural network based on the Adam optimization algorithm, and to train / verify the CNN-LSTM composite neural network through the data training set and the data verification set to obtain the optimized CNN-LSTM composite neural network; Based on the optimized CNN-LSTM composite neural network, the time series change prediction of the steel support axial force is carried out according to the data test set to obtain the prediction results; The steel support servo module is used to realize the servo control of the steel support axial force according to the prediction results based on the constructed servo control strategy; The foundation pit BIM module is used to construct a deep foundation pit BIM model based on Revit modeling technology to control the deformation of the deep foundation pit through the steel support axial force servo system. According to the output of the steel support servo module, it interacts with the deep foundation pit BIM model in real time through the Revit API data interface to realize the visualization of the deformation control of the deep foundation pit.

2. A deep foundation pit stability deformation control method based on digital twin according to claim 1, characterized in that: The CNN-LSTM composite neural network based on the Adam optimization algorithm built in S3 consists of the following network components: Data input layer, convolution layer, pooling layer, LSTM network layer, fully connected layer and output layer; The data input layer is used to transmit the acquired monitoring time series data of the deep foundation pit to the convolution layer; The convolution layer is used to extract the multi-dimensional local spatiotemporal data features of the foundation pit surface settlement and the horizontal displacement of the ground-connected wall from the monitoring time series data; The pooling layer is used to perform dimension reduction / compression operations on local spatiotemporal data features to obtain optimized local spatiotemporal data features; The LSTM network layer is used to further extract the time series relationship characteristics of the foundation pit surface settlement and the horizontal displacement of the ground-connected wall according to the optimized local spatiotemporal data characteristics, and obtain time series data; The fully connected layer is used to perform feature fusion on the time series data to obtain fusion features; The feature fusion step specifically includes performing a one-dimensional flattening operation on the time series data through the Flatten function to obtain a one-dimensional data vector; performing a linear transformation on the one-dimensional data vector through a set weight matrix and bias, and obtaining a weighted sum of the one-dimensional data vector after the linear transformation; The output layer is used to predict and output the steel support axial force data according to the fusion features.

3. The method for controlling the stability deformation of a deep foundation pit based on digital twin according to claim 1 is characterized in that: In S3, the CNN-LSTM composite neural network model is trained through the data training set, which includes the following steps: S31: define the foundation pit surface settlement data and the ground-connected wall horizontal displacement data as input data, and define the steel support axial force data as output data; The data training set is input into the CNN-LSTM composite neural network for model training to obtain the trained CNN-LSTM composite neural network; S32: taking the mean square error function between the predicted steel support axial force data and the actual steel support axial force data as the model loss function; The trained CNN-LSTM composite neural network is model verified through the data verification set, and whether the output of the trained CNN-LSTM composite neural network converges is determined based on the model loss function; If it is confirmed that the output of the trained CNN-LSTM composite neural network converges, then the trained CNN-LSTM composite neural network is the optimized CNN-LSTM composite neural network; Otherwise, based on the Adam optimization algorithm, the network parameters of the trained CNN-LSTM composite neural network are adaptively updated, and step S31 is iteratively executed repeatedly.

4. A method for controlling deformation of deep foundation pit stability based on digital twin according to claim 3, characterized in that: The method of adaptively updating the network parameters of the trained CNN-LSTM composite neural network based on the Adam optimization algorithm described in S32 specifically includes the following steps: S321: define the initialization first-order moment and second-order moment of the network parameter gradient; S322: Iteratively update the initialized first-order moment and second-order moment, and obtain updated first-order moment and second-order moment; And the expressions for iteratively updating and obtaining the updated first-order moment and second-order moment are: m t =β1m t-1 +(1-β11)g t Where: m t With v t They represent the exponential moving average of the gradient and the square of the gradient, i.e., the first-order moment and the second-order moment, respectively; β1 and β2 represent m t With v t The exponential decay rate hyperparameter of g t represents the gradient of the network parameters at time step t; Represents the gradient of the model loss function J(θ) with respect to the network parameter θ; that is, for each network parameter θ t Calculate a vector formed by the partial derivatives of the loss function; S323: Obtaining a deviation between the first-order moment and the second-order moment according to the updated first-order moment and second-order moment; And the expression for obtaining the deviation between the first-order moment and the second-order moment is Where: and Represent the deviations of the first-order moment and the second-order moment respectively; and Respectively represent the time step t, corresponding to m t With v t The exponential decay rate hyperparameter of ; S324: Adaptively update the network parameters corresponding to each time step t in the CNN-LSTM composite neural network according to the deviation between the first-order moment and the second-order moment; And the expression for adaptive update of network parameters is Where: θ t-1 represents the network parameters corresponding to the time step t-1; θ t Represents the network parameters corresponding to the time step t; α represents the learning rate; ε represents a positive design parameter.

5. The method for controlling the stability deformation of a deep foundation pit based on digital twin according to claim 4 is characterized in that: The servo control strategy constructed in S3 specifically includes the following steps S001: Preset steel support axial force threshold; S002: Based on the optimized CNN-LSTM composite neural network, the time series change prediction of the steel support axial force is carried out according to the data test set to obtain the current prediction results; Determine whether the current prediction result meets the preset steel support axial force threshold; If the predicted result is greater than the preset steel support axial force threshold, the steel support axial force servo system will unload the steel support axial force on the deep foundation pit; If the predicted result is less than the preset steel support axial force threshold, the steel support axial force servo system is used to load the steel support axial force on the deep foundation pit; If the predicted result is equal to the preset steel support axial force threshold, the steel support axial force for the deep foundation pit is maintained through the steel support axial force servo system.

6. A device based on the deep foundation pit stability deformation control method based on digital twins according to any one of claims 1 to 5, characterized in that: It includes a steel support axial force servo system, a data acquisition box, a box-type fixed inclinometer, a steel support structure, a static level, and a deep foundation pit deformation control digital twin platform; The steel support shaft force servo system includes a computer control system, a numerical control pump station and a support head assembly component; the support head assembly component at least includes a hydraulic jack and a steel support shaft force meter; The steel support structure is arranged transversely inside the deep foundation pit, and both ends of the steel support structure are abutted against both ends of the ground connection wall of the deep foundation pit; The data acquisition box and CNC pump station are respectively set on the ground on both sides of the deep foundation pit; The static level is set on the ground at one side of the deep foundation pit and connected to the data acquisition box to monitor the surface settlement data of the foundation pit; Box-type fixed inclinometers are installed on the ground-connected walls on both sides of the deep foundation pit, and are distributed on both sides of the steel support and connected to the data acquisition box to monitor the horizontal displacement data of the ground-connected walls; The support head assembly and the steel support axial force gauge are respectively arranged at both ends of the steel support structure; The numerical control pump station is connected to the support head assembly and the computer control system respectively; The computer control system is used to obtain a drive control signal based on a servo control strategy, and to drive a hydraulic jack to control the axial force of a steel support of a deep foundation pit by controlling a numerically controlled pump station based on the drive control signal; The steel support axial force meter is used to detect and obtain the steel support axial force of deep foundation pit in real time; The control signal is a signal for controlling the axial force of the steel support of the deep foundation pit obtained in real time by the steel support axial force meter and a preset steel support axial force threshold value, so as to control the numerical control pump station to drive the hydraulic jack to realize the control of the axial force of the steel support of the deep foundation pit; The deep foundation pit deformation control digital twin platform is communicated with the steel support axial force servo system and the data acquisition box through preset wireless modules.

Images