A method for intelligent prediction and active control of foundation pit deformation

Abstract

This invention discloses an intelligent prediction and active control method for foundation pit deformation, comprising: collecting monitoring data during foundation pit construction and preprocessing it; constructing a prediction model for retaining structure deformation and surface deformation; inputting the axial force of the steel supports and the depth of water level outside the foundation pit on the current day, as well as the depth of water level inside the foundation pit and the excavation depth on the following day, into the prediction model to obtain the deformation of the retaining structure and the surface deformation outside the foundation pit on the following day; when the deformation of the retaining structure and the surface deformation outside the foundation pit are less than a threshold, maintaining the axial force of the steel supports and the depth of water level outside the foundation pit on the current day; if the deformation of the retaining structure and the surface deformation outside the foundation pit are greater than the threshold, adjusting the axial force of the steel supports and the depth of water level outside the foundation pit until they meet the requirements, and using the adjusted axial force of the steel supports and the depth of water level outside the foundation pit as the construction parameters for the following day to control foundation pit deformation in advance. This invention can control deformation before it occurs, effectively overcomes the lag, and improves engineering safety and decision-making efficiency.

Description

Technical Field

[0001] This invention relates to the field of underground engineering technology, specifically to a method for intelligent prediction and active control of foundation pit deformation. Background Technology

[0002] With the continuous acceleration of urbanization in my country, the development and utilization of underground space is becoming increasingly frequent, and the number and scale of deep foundation pit projects are also growing rapidly. The safety and stability of foundation pit projects are directly related to the safety of adjacent buildings, underground pipelines, and municipal facilities. Therefore, effective prediction and precise control of foundation pit deformation are key to ensuring the safety of the entire project construction.

[0003] Currently, the control of foundation pit deformation mainly relies on traditional monitoring methods. When monitoring data indicates that the deformation is close to or exceeds the warning threshold, engineers then take remedial measures. However, this approach has a significant time lag. There is a non-negligible time difference between the discovery of abnormal deformation, the implementation of control measures, and the effectiveness of those measures. During this period, the deformation may have progressed further, even leading to irreversible safety risks. Moreover, subsequent adjustments are costly and inefficient.

[0004] In recent years, artificial intelligence technologies such as machine learning and deep learning have provided new approaches to predicting foundation pit deformation. Some methods have achieved rapid predictions by establishing purely data-driven models that link monitoring data with deformation results. However, these "black box" models rely entirely on the scale and quality of historical data, lacking the embedding and constraint of the inherent physical and mechanical laws of foundation pit engineering. When construction conditions exceed the range of historical data, the model's predictions may deviate significantly from physical laws, leading to prediction failure, insufficient generalization ability and reliability, and difficulty in providing credible decision support in complex and ever-changing real-world engineering projects.

[0005] Therefore, there is an urgent need in this field for an intelligent control method for foundation pit deformation that can overcome the above-mentioned shortcomings, so as to provide a scientific basis for optimizing construction parameters and realizing refined intelligent control of foundation pit deformation. Summary of the Invention

[0006] The purpose of this invention is to provide an intelligent prediction and active control method for foundation pit deformation to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, this invention provides a method for intelligent prediction and active control of foundation pit deformation, comprising the following steps:

[0008] S1. Collect monitoring data during the construction of the foundation pit. The monitoring data includes the axial force of the steel support, the depth of water in the foundation pit, the depth of water in the foundation pit, the excavation depth of the foundation pit, the lateral deformation of the foundation pit retaining structure, and the deformation of the ground surface outside the foundation pit.

[0009] S2. Preprocess the collected data, including outlier removal, sliding window smoothing filtering, and normalization, to improve data quality and model training efficiency.

[0010] S3. Construct a retaining structure deformation prediction model and a surface deformation prediction model using preprocessed monitoring data; wherein: the retaining structure deformation prediction model uses the axial force of the steel support, the excavation depth of the foundation pit, the dewatering depth inside the foundation pit, the dewatering depth outside the foundation pit, and the depth coordinates of the retaining structure as inputs, and the lateral deformation of the foundation pit retaining structure as outputs, and uses the elastic foundation beam control equation as physical constraints; the surface deformation prediction model uses the output of the retaining structure deformation prediction model as inputs, the surface deformation outside the foundation pit as outputs, and uses the boundary element equation based on the Mindlin solution as physical constraints; both the retaining structure deformation prediction model and the surface deformation prediction model adopt the KAN neural network architecture;

[0011] S4. During construction, the axial force of the steel support on the day, the depth of dewatering outside the foundation pit, and the planned depth of dewatering inside the foundation pit and the excavation depth of the foundation pit on the next day are input into the trained prediction model to predict the lateral deformation of the foundation pit retaining structure and the deformation of the ground surface outside the foundation pit on the next day.

[0012] S5. Compare the predicted maximum lateral deformation of the foundation pit retaining structure and the maximum surface deformation outside the foundation pit with their preset safety thresholds. If both are less than their preset safety thresholds, the construction parameters for the day are determined to be safe and controllable, and step S6 is executed. If either is greater than or equal to its preset safety threshold, the construction parameters for the day are determined to be unsafe and uncontrollable, and step S7 is executed.

[0013] S6. Carry out excavation and dewatering according to the original plan, and determine the steel support axial force and dewatering depth outside the foundation pit on the same day as the construction parameters for the next day.

[0014] S7. Adjust the axial force of the steel support and / or the depth of the groundwater outside the foundation pit, and repeat the prediction steps until the predicted lateral deformation of the foundation pit retaining structure and the deformation of the ground surface outside the foundation pit meet the safety requirements. Then, determine the adjusted axial force of the steel support and the depth of the groundwater outside the foundation pit as the construction parameters for the next day, thereby achieving early control of the foundation pit deformation.

[0015] Furthermore, in step S3, the preprocessed data is organized into a sample set in chronological order. Each sample includes the axial force of the steel support, the excavation depth of the foundation pit, the dewatering depth inside the foundation pit, the dewatering depth outside the foundation pit, and the depth coordinates of the retaining structure (corresponding to the lateral deformation of the retaining structure) for that day, as well as the corresponding monitored lateral deformation of the retaining structure and the surface deformation outside the foundation pit.

[0016] Furthermore, in step S3, a deformation prediction model for the retaining structure and a surface deformation prediction model are constructed using monitoring data from the past five days.

[0017] Furthermore, the expression for the governing equation of the elastic foundation beam is:

[0018]

[0019] in, For the stiffness of the foundation pit retaining structure, This represents the lateral deformation of the foundation pit retaining structure. The depth below the Earth's surface. For steel support springs or earth springs at depth Stiffness at the point, For depth Earth pressure and water pressure at the location.

[0020] Furthermore, the total loss function of the enclosure structure deformation prediction model It consists of a weighted average of the data fitting loss term, the governing equation loss term, and the boundary condition loss term:

[0021]

[0022] in, This is the loss term between the predicted results of the enclosure structure deformation prediction model and the measured data. This is the loss term for the governing equation; This is the loss term for the boundary conditions; The weights of the loss terms between the prediction results of the enclosure structure deformation prediction model and the measured data; To control the weights of the loss term in the equation; The weights of the boundary condition loss term.

[0023] Furthermore, boundary condition constraints are set at the fixed end of the enclosure structure, and the boundary condition loss is defined as the sum of the squares of the horizontal displacement and the rotation angle at the fixed end.

[0024] Furthermore, the total loss function of the surface deformation prediction model It consists of a weighted average of the data fitting loss term and the physical relationship loss term.

[0025] Furthermore, in step S7, the axial force of the steel support is increased and / or the depth of dewatering outside the foundation pit is reduced, and the construction parameters are optimized through iterative prediction.

[0026] Compared with the prior art, the present invention has the following beneficial effects:

[0027] (1) Integration of physical laws and data-driven approaches: By embedding core physical and mechanical equations such as elastic foundation beam theory and Mindlin solution into the neural network as constraints, the model not only learns the patterns of historical data, but also strictly follows physical principles, which significantly improves the rationality and generalization ability of the prediction results, and can still maintain reliable predictions, especially in working conditions where historical data is not covered.

[0028] (2) Advance prediction and active control: It changes the traditional passive control mode of "monitoring-alarm-response", and can predict the deformation trend one day in advance and optimize and adjust the construction parameters in advance to control the deformation before it occurs, effectively overcome the lag, and improve the safety of the project and decision-making efficiency.

[0029] (3) Model advancement and accuracy: The KAN network structure is adopted, and its flexible activation function design can better capture complex nonlinear relationships. Combined with adaptive loss weight balancing technology, the convergence speed and prediction accuracy of the model are further improved.

[0030] In addition to the objectives, features, and advantages described above, the present invention has other objectives, features, and advantages. The invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0031] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings:

[0032] Figure 1 This is a control flowchart of an intelligent prediction and active control method for foundation pit deformation according to the present invention;

[0033] Figure 2 This is a schematic diagram of the structure of a single-layer KAN neural network in this invention;

[0034] Figure 3 This is a schematic diagram of the structure of the enclosure structure deformation prediction model in this invention;

[0035] Figure 4 This is a schematic diagram of the surface deformation prediction model in this invention. Detailed Implementation

[0036] The present invention will now be described in detail with reference to the embodiments shown in the accompanying drawings. However, it should be noted that these embodiments are not intended to limit the present invention. Equivalent transformations or substitutions in function, method, or structure made by those skilled in the art based on these embodiments are all within the scope of protection of the present invention.

[0037] Please see Figures 1 to 3This embodiment provides a method for intelligent prediction and active control of foundation pit deformation, including the following steps:

[0038] S1. On day t of construction, monitoring data will be collected using monitoring equipment, including the axial force of the steel supports, the depth of dewatering inside and outside the foundation pit, the excavation depth of the foundation pit, the lateral deformation of the foundation pit retaining structure, and the deformation of the ground surface outside the foundation pit. Specifically, this includes:

[0039] Record the daily excavation elevation of the foundation pit to determine its excavation depth. Install dewatering wells on both the inner and outer sides of the foundation pit to measure the daily water level elevations inside and outside the pit. Install an axial force servo system on each steel support to collect axial force data at high frequency (e.g., once per hour) and record the daily average or characteristic values ​​(e.g., maximum and minimum values). Pre-embed inclinometer tubes within 1m of the inner or outer side of the retaining structure and use an inclinometer to measure the horizontal deformation of the retaining structure. Set up a measuring point every 0.5~1.0m along the depth direction and measure once daily to obtain the horizontal displacement curve of the retaining structure along the depth direction (from top to bottom of the wall). Establish settlement monitoring points on the soil surface outside the foundation pit and use a level to periodically measure the vertical displacement of the ground surface outside the foundation pit.

[0040] S2. The collected raw data must undergo a rigorous preprocessing process to ensure its usability for model training. Statistical methods (such as the Laida criterion) or empirical judgment based on physical laws are used to identify and remove erroneous data that significantly deviates from the normal range. For data that may contain noise, a moving average method is used for smoothing to eliminate random interference and preserve trend changes. The size of the moving window can be set according to the actual situation, such as a 3-point or 5-point moving average. Simultaneously, to eliminate the adverse effects of different physical dimensions and orders of magnitude on model training and accelerate model convergence, all data must be normalized to the [0, 1] interval. Specifically, this step uses the max-min normalization method:

[0041]

[0042] in, The result of data normalization; This is the original data; and These are the minimum and maximum values ​​of the feature in the original dataset, respectively. It should be noted that when the model is used for prediction, the new input data should also be determined using the training set. and Normalize.

[0043] The preprocessed data is organized chronologically to form a sample set. Each sample contains the axial force of the steel supports for that day. Excavation depth of the foundation pit Depth of water in the foundation pit Depth of water outside the foundation pit Depth coordinates of the enclosure structure As input features, the lateral deformation of the foundation pit retaining structure and the deformation of the ground surface outside the foundation pit, measured on the same day, are used as supervision labels.

[0044] S3. Based on the KAN neural network as the core architecture, the neural network is trained using a sample set formed by the past five days' data on steel support axial force, water depth inside and outside the foundation pit, excavation depth of the foundation pit, deformation of the retaining structure, and surface deformation outside the foundation pit. This is used to construct a prediction model for retaining structure deformation and a prediction model for surface deformation. The structure of a single-layer KAN neural network is as follows: Figure 2 shown. Specifically:

[0045] Unlike traditional multilayer perceptrons, the core idea of ​​the KAN neural network is to replace the fixed activation functions on traditional network nodes with learnable univariate functions. Specifically, KAN sets a learnable univariate activation function at each edge of the network (i.e., at each connection weight). The activation function It consists of two parts:

[0046]

[0047] in, The SiLU function is used as the basis function in this invention. The introduction of basis functions is equivalent to adding residual connections to the activation function, which helps with gradient flow and the training stability of deep networks, achieving a smooth mapping of the input. This is a spline curve function used to fit the complex nonlinear components that deviate from the linear part of the activation function; its expression is:

[0048]

[0049] in, The number of grid intervals used when defining the B-spline function; Let be the order of the piecewise polynomial of the spline function; For the first spline curve function; For the first One learnable coefficient.

[0050] Therefore, a single-layer KAN neural network can be viewed as a set of univariate activation functions. The matrix formed:

[0051]

[0052] in, To connect the input layer The node to the output layer Univariate activation functions for each node.

[0053] When considering When using a multi-layered KAN neural network, its overall mapping function The result of composing the functions at each level:

[0054]

[0055] in, The input vector for the model; Indicates the first The matrix set consisting of all activation functions in a multilayer KAN neural network. =1,2,3… Integers; This represents mapping multiplication.

[0056] In embodiments of the present invention, such as Figure 3 As shown, the input layer of the retaining structure deformation prediction model has five neurons: the axial force of the steel support, the excavation depth of the foundation pit, the dewatering depth inside the foundation pit, the dewatering depth outside the foundation pit, and the depth coordinates of the retaining structure. It should be noted that this model learns the deformation field across the entire retaining structure area at once, rather than the deformation at a single location. The output layer has one neuron, outputting the predicted horizontal displacement of the retaining structure at the corresponding depth coordinates. Additionally, 2-4 hidden layers can be set depending on the complexity of the problem, and the width (number of neurons) of each layer can be 32, 64, or 128, etc.

[0057] The mechanical behavior of the retaining structure under earth pressure and support force can be described by an elastic foundation beam model, and its governing equations are fourth-order differential equations:

[0058]

[0059] in, For the stiffness of the foundation pit retaining structure, This represents the lateral deformation of the foundation pit retaining structure. The depth below the Earth's surface. For steel support springs or earth springs at depth Stiffness at the point, For depth The soil pressure and water pressure at the location. Specifically, the stiffness of the steel support spring can be calculated based on the axial force and the deformation of the retaining structure; the stiffness of the soil spring can be determined empirically or used as a learnable parameter.

[0060] During training, the prediction results of the KAN neural network The prediction results must not only match the measured data (i.e., the lateral deformation of the retaining structure collected during construction), but also satisfy the aforementioned control equations as much as possible. To this end, this invention uses automatic differential technology to calculate the prediction results. right The fourth derivative And calculate the residuals of the governing equations of the surface deformation prediction model. :

[0061]

[0062] For the retaining structure of the foundation pit, the displacement and rotation of its fixed end are 0; therefore, its boundary condition loss is the predicted horizontal displacement and rotation of the fixed end.

[0063] Total loss function of the building envelope deformation prediction model It consists of three weighted components: the data fitting loss term, the governing equation loss term, and the boundary condition loss term.

[0064]

[0065] in, This is the loss term between the predicted results of the enclosure structure deformation prediction model and the measured data. This is the loss term for the governing equation; This is the loss term for the boundary conditions; The weights of the loss terms between the prediction results of the enclosure structure deformation prediction model and the measured data; To control the weights of the loss term in the equation; These are the weights for the boundary condition loss term. Specifically:

[0066] Loss terms of the prediction results of the building envelope deformation prediction model and the measured data The root mean square error is used, and its expression is:

[0067]

[0068] in, The number of data points; No. The measured values ​​of each data point;

[0069] Loss term of the governing equation of the building envelope deformation prediction model The root mean square error is used, and its expression is:

[0070]

[0071] in, The number of configuration points randomly sampled within the computational domain for calculating the physical residuals;

[0072] Loss term of boundary conditions in the deformation prediction model of the building envelope Defined as:

[0073]

[0074] in, This represents the predicted horizontal displacement at the bottom of the fixed end of the foundation pit retaining structure.

[0075] In this embodiment of the invention, to prevent certain loss terms from dominating the training process due to large gradients, Set to 1, and dynamically adjust using an adaptive loss weight algorithm. and In each training cycle In the middle, the weights are updated as follows:

[0076]

[0077]

[0078] in, This is a hyperparameter, typically set to 0.9; Indicates loss item Relative to model parameters The gradient norm; This represents the maximum value of the gradient norm of the loss term between the model's predicted results and the measured data. For loss items The average value of the gradient norm. This ensures that the gradient magnitudes of each loss term are relatively balanced, thus ensuring stable training.

[0079] In an embodiment of the present invention, Figure 4 The surface deformation prediction model shown uses the horizontal displacement distribution along the depth of the entire retaining structure predicted by the retaining structure deformation prediction model as input. In the actual network, this can be discretized into displacement values ​​at a series of depth points. The output layer is the predicted vertical displacement value of the ground surface outside the foundation pit. This model also uses a KAN neural network structure.

[0080] The surface deformation caused by foundation pit excavation mainly originates from the lateral deformation of the retaining structure. This invention uses the boundary element method based on Mindlin solutions as physical constraints. The Mindlin solution gives the displacement at any point within a semi-infinite elastic body caused by a concentrated force. The retaining structure that generates lateral displacement is discretized into... Each segment is a tiny segment, and a distributed virtual force to be determined acts on each segment. The displacement generated by the virtual force on the retaining structure should be equal to the horizontal displacement of the retaining structure predicted by the retaining structure deformation prediction model. The vertical displacement generated by the virtual force at any location on the ground surface outside the foundation pit can be obtained through the Mindlin solution.

[0081] Specifically, based on the MINDLIN solution, for any point in the enclosure structure... Lateral displacement for:

[0082]

[0083]

[0084] in, The Poisson's ratio of the soil; The shear modulus of the soil; , , For virtual force The coordinates; , , For any point The coordinates; The total number of segments to be divided for the building envelope; For the first The length of the segment; , .

[0085] Using the horizontal displacement of the enclosure structure predicted by the enclosure structure deformation prediction model as the boundary condition, the matrix expression of the horizontal displacement and virtual force of the enclosure structure can be obtained:

[0086]

[0087]

[0088] in, For the first Virtual force For the building envelope The impact of location. , , The horizontal displacement of the retaining structure is predicted by the deformation prediction model. The virtual force can be calculated from the matrix. , , .

[0089] According to the MINDLIN solution, the virtual horizontal force acting on the enclosure structure Any point on the Earth's surface caused The vertical deformation is as follows:

[0090]

[0091]

[0092] in, , , any point on the Earth's surface The coordinates.

[0093] In neural network training, the predicted surface deformation results are required. Predicted horizontal displacement of the enclosure structure The above physical relationships are satisfied. The residuals of the governing equations of the surface deformation prediction model. Defined as:

[0094]

[0095] in, Based on the predicted horizontal displacement of the enclosure structure The theoretical value calculated using the boundary element method based on the Mindlin solution.

[0096] Total loss function of surface deformation prediction model Relatively simple, it consists of a weighted average of two parts: a data fitting loss term and a physical relationship loss term. The data fitting loss term and the physical relationship loss term are defined identically to those in the building envelope deformation prediction model, using mean squared error. Similarly, an adaptive loss weighting algorithm is used to dynamically adjust the weights of the physical relationship loss term. Specifically:

[0097]

[0098] in, This is the loss term between the prediction results of the surface deformation prediction model and the measured data. This is the loss term in the governing equation of the surface deformation prediction model; The weights of the loss terms between the prediction results of the surface deformation prediction model and the measured data are used. This represents the weight of the loss term in the governing equation of the surface deformation prediction model.

[0099] Combination Figure 3 and Figure 4 In this invention, the retaining structure deformation prediction model and the surface deformation prediction model each use the constructed total loss function as the optimization objective, and calculate the total loss function of the corresponding model through the backpropagation algorithm. and For each learnable parameter in the neural network, the gradient is calculated, and optimization algorithms such as gradient descent are used to iteratively train and adjust the parameters of the entire model based on these gradient information, thereby continuously reducing the loss function value and making the model's prediction results approximate the observed data while satisfying the constraints of physical laws.

[0100] S4. Keeping the axial force of the steel support and the depth of the groundwater outside the pit constant on day t, determine the planned depth of the groundwater inside the pit and the excavation depth of the pit on the next day (day t+1) according to the construction plan, and input them into the constructed retaining structure deformation prediction model to obtain the predicted lateral displacement distribution of the retaining structure on day t+1. Then input the predicted lateral displacement distribution of the retaining structure on day t+1 into the surface deformation prediction model to obtain the predicted surface displacement distribution outside the pit on day t+1.

[0101] S5. Compare the predicted maximum lateral displacement of the retaining structure and the maximum surface settlement or deformation value of key points with the preset safety threshold. If both predicted values ​​are less than their corresponding preset safety thresholds, the current construction parameters are determined to be safe and controllable, and step S6 is executed. If any predicted value is greater than or equal to its corresponding preset safety threshold, the construction parameters for the day are determined to be unsafe and uncontrollable, and step S7 is executed.

[0102] S6. Carry out excavation and dewatering according to the original plan, and determine the steel support axial force and dewatering depth outside the foundation pit on the same day as the construction parameters for the next day.

[0103] S7. Adjust the axial force of the steel support and / or the depth of dewatering outside the foundation pit, mainly by increasing the axial force of the steel support and decreasing the depth of dewatering outside the foundation pit; and re-execute the prediction and judgment steps, that is, input the adjusted axial force of the steel support and the depth of dewatering outside the foundation pit with the planned excavation depth of the foundation pit and the depth of dewatering inside the foundation pit for the next day into the prediction model to obtain new predicted values ​​of retaining structure deformation and surface deformation, and compare them again with the preset safety threshold until a set of construction parameters that can make the predicted deformation value less than the preset safety threshold is found, and use the optimized construction parameters as the actual construction parameters for day t+1 to carry out construction, so as to realize active intelligent control of foundation pit deformation.

[0104] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent prediction and active control of foundation pit deformation, characterized in that, Includes the following steps: S1. Collect monitoring data during the construction of the foundation pit. The monitoring data includes the axial force of the steel support, the depth of water in the foundation pit, the depth of water in the foundation pit, the excavation depth of the foundation pit, the lateral deformation of the foundation pit retaining structure, and the deformation of the ground surface outside the foundation pit. S2. Preprocess the collected data, including outlier removal, sliding window smoothing filtering, and normalization. S3. Construct a retaining structure deformation prediction model and a surface deformation prediction model using preprocessed monitoring data. The retaining structure deformation prediction model takes the axial force of the steel support, the excavation depth of the foundation pit, the dewatering depth inside the foundation pit, the dewatering depth outside the foundation pit, and the depth coordinates of the retaining structure as inputs, and the lateral deformation of the retaining structure as the output, with the elastic foundation beam control equation as the physical constraint. The surface deformation prediction model takes the output of the retaining structure deformation prediction model as input, the surface deformation outside the foundation pit as the output, and the boundary element equation based on the Mindlin solution as the physical constraint. Both the retaining structure deformation prediction model and the surface deformation prediction model adopt the KAN neural network architecture. S4. During construction, the axial force of the steel support on the day, the depth of dewatering outside the foundation pit, and the planned depth of dewatering inside the foundation pit and the excavation depth of the foundation pit on the next day are input into the trained prediction model to predict the lateral deformation of the foundation pit retaining structure and the deformation of the ground surface outside the foundation pit on the next day. S5. Compare the predicted maximum lateral deformation of the foundation pit retaining structure and the maximum surface deformation outside the foundation pit with their preset safety thresholds. If both are less than their preset safety thresholds, the construction parameters for the day are determined to be safe and controllable, and step S6 is executed. If either is greater than or equal to its preset safety threshold, the construction parameters for the day are determined to be unsafe and uncontrollable, and step S7 is executed. S6. Carry out excavation and dewatering according to the original plan, and determine the steel support axial force and dewatering depth outside the foundation pit on the same day as the construction parameters for the next day. S7. Adjust the axial force of the steel support and / or the depth of the groundwater outside the foundation pit, and repeat the prediction steps until the predicted lateral deformation of the foundation pit retaining structure and the deformation of the ground surface outside the foundation pit meet the safety requirements. Then, determine the adjusted axial force of the steel support and the depth of the groundwater outside the foundation pit as the construction parameters for the next day, thereby achieving early control of the foundation pit deformation.

2. The intelligent prediction and active control method for foundation pit deformation according to claim 1, characterized in that, In step S3, the preprocessed data is organized into a sample set in chronological order. Each sample contains the axial force of the steel support, the excavation depth of the foundation pit, the dewatering depth inside the foundation pit, the dewatering depth outside the foundation pit, and the depth coordinates of the retaining structure for that day, as well as the corresponding lateral deformation of the foundation pit retaining structure and the deformation of the ground surface outside the foundation pit.

3. The intelligent prediction and active control method for foundation pit deformation according to claim 1, characterized in that, In step S3, a deformation prediction model for the retaining structure and a surface deformation prediction model are constructed using monitoring data from the past five days.

4. The intelligent prediction and active control method for foundation pit deformation according to claim 1, characterized in that, The expression for the governing equation of the elastic foundation beam is: in, For the stiffness of the foundation pit retaining structure, This represents the lateral deformation of the foundation pit retaining structure. The depth below the Earth's surface. For steel support springs or earth springs at depth Stiffness at the point, For depth Earth pressure and water pressure at the location.

5. The intelligent prediction and active control method for foundation pit deformation according to claim 1, characterized in that, The total loss function of the enclosure structure deformation prediction model It consists of a weighted average of the data fitting loss term, the governing equation loss term, and the boundary condition loss term: in, This is the loss term between the predicted results of the enclosure structure deformation prediction model and the measured data. This is the loss term in the governing equation; This is the loss term for the boundary conditions; The weights of the loss terms between the prediction results of the enclosure structure deformation prediction model and the measured data; To control the weights of the loss term in the equation; The weights of the boundary condition loss term.

6. The intelligent prediction and active control method for foundation pit deformation according to claim 5, characterized in that, Boundary condition constraints are set at the fixed end of the enclosure structure, and the boundary condition loss is defined as the sum of the squares of the horizontal displacement and the rotation angle at the fixed end.

7. The intelligent prediction and active control method for foundation pit deformation according to claim 5, characterized in that, The total loss function of the surface deformation prediction model It consists of a weighted sum of the data fitting loss term and the control equation loss term.

8. The intelligent prediction and active control method for foundation pit deformation according to claim 1, characterized in that, In step S7, the axial force of the steel support is increased and / or the depth of dewatering outside the foundation pit is reduced, and the construction parameters are optimized through iterative prediction.

Images