A Method for Axial Force Servo Regulation of Foundation Pit Steel Supports Based on Digital Twin

Abstract

The present invention relates to a method for servo control of axial force of foundation pit steel supports based on digital twin, and the steps are as follows: S1. A foundation pit engineering simulation module constructs a foundation pit deformation prediction surrogate model with key soil parameters and servo steel support axial force as inputs; S2. An Internet of Things monitoring module collects foundation pit deformation data and servo steel support axial force data through an Internet of Things monitoring system; S3. A soil parameter update module uses the foundation pit deformation prediction surrogate model in S1 as a forward calculation model in probabilistic back analysis, combines with Bayes' theorem, and updates the posterior samples of soil parameters through data assimilation technology; S4. A deformation prediction and axial force control module predicts the foundation pit deformation in the next stage and proposes an axial force servo control scheme. This method is based on digital twin technology, deep learning algorithms and data assimilation technology, analyzes the influence of the axial force set value in the servo system on the foundation pit deformation, and provides a scientific and reasonable servo control scheme for the axial force of foundation pit steel supports.

Description

Technical Field

[0001] The present invention relates to the field of foundation pit engineering, and particularly to a method for servo control of axial force of steel supports for foundation pits based on digital twin. Background Technique

[0002] The steel support axial force servo system is a new technology for active control of foundation pit deformation. It reduces or even reverses foundation pit deformation by actively applying stress to the soil mass and generating forced deformation. The control effect of the steel support axial force servo system on soil deformation depends on its axial force setting value; in current engineering practice, this setting value mostly relies on the experience judgment of engineers, which is prone to control failure or over-conservatism. Numerical models can analyze the relationship between the steel support axial force setting value and foundation pit deformation to further optimize the axial force control scheme. As an important input parameter of the numerical model, soil parameters have a significant impact on the calculation results. However, due to the very limited site investigation data and the uncertainty of soil parameters themselves, there are often large differences between the predictions obtained from numerical models and actual monitoring, which affects the reliability of the servo steel support axial force control scheme. Therefore, it is urgent to develop an efficient surrogate model to reduce the computational cost required in the dynamic update process, integrate the real-time monitoring data during construction, dynamically update the soil parameters and their uncertainties, and reasonably propose a servo steel support axial force control scheme in a timely manner in the engineering scenario. Summary of the Invention

[0003] In view of the deficiencies in the background technique, the technical problem to be solved by the present invention is to provide a method for servo control of axial force of steel supports for foundation pits based on digital twin. This method analyzes the influence of the axial force setting value in the servo system on foundation pit deformation based on digital twin technology, deep learning algorithms, and data assimilation technology, provides a scientific and reasonable servo control scheme for the axial force of steel supports for foundation pits for engineers, and reduces engineering risks. The present invention significantly improves the intelligent level of the servo control of the axial force of steel supports for foundation pits and has broad engineering application prospects.

[0004] The present invention is achieved by adopting the following technical solutions: A method for servo control of axial force of steel supports for foundation pits based on digital twin, the steps are as follows:

[0005] S1. Foundation pit engineering simulation module,

[0006] Construct a foundation pit numerical model, determine the ranges of key soil parameters and axial force parameters of the foundation pit numerical model, and use a deep learning algorithm to construct a foundation pit deformation prediction surrogate model with key soil parameters and servo steel support axial force as inputs;

[0007] S2. Internet of Things monitoring module,

[0008] Collect foundation pit deformation data and servo steel support axial force data through the Internet of Things monitoring system;

[0009] S3. Soil parameter update module,

[0010] Take the foundation pit deformation prediction proxy model in S1 as the forward calculation model in probabilistic back analysis. Combine with Bayes' theorem, and through data assimilation technology, assimilate the foundation pit deformation data in S2 to dynamically update the key soil parameters and obtain the posterior samples of the key soil parameters;

[0011] S4. Deformation prediction and axial force control module,

[0012] Preset multiple sets of servo steel support axial force control schemes. Input the posterior samples of the key soil parameters in S3 and the axial force parameters of the servo steel support in the preset control schemes into the foundation pit deformation prediction proxy model in S1 to predict the foundation pit deformation in the next stage and propose a reasonable axial force servo control scheme.

[0013] Further, in S1, the specific steps to construct the foundation pit deformation prediction proxy model are as follows:

[0014] S11. Foundation pit numerical model, conduct finite element numerical modeling on the foundation pit project equipped with a steel support axial force servo system;

[0015] S12. Combine engineering design data, literature data, and geological exploration reports to determine the value ranges of the key soil parameters and axial force parameters of the finite element numerical model. The axial force parameter is the axial force value of the servo steel support;

[0016] S13. Based on the Latin hypercube sampling method, sample the key soil parameters and axial force parameters to generate N groups of parameter combination samples;

[0017] S14. Through the forward calculation of the finite element numerical model in S11, obtain the foundation pit deformation response calculation results corresponding to each sample. Take the combination of the key soil parameters and axial force parameters as the input feature and the corresponding foundation pit deformation response as the output label to construct a foundation pit deformation prediction proxy model dataset containing N groups of mapping relationships, where N is an integer not less than 1000;

[0018] S15. Divide the dataset in S14 into a training set and a test set according to the ratio of 8:2. Use the training set to train the bidirectional long short-term memory network (BiLSTM) to construct the foundation pit deformation prediction proxy model;

[0019] S16. Verify through the test set, and use the mean square error (MSE) or root mean square error (RMSE) or mean absolute error (MAE) or coefficient of determination (R 2 ) index to evaluate the performance of the model.

[0020] Further, the foundation pit deformation data includes the lateral deformation monitoring data of multiple monitoring points, and the servo steel support axial force data is the axial force monitoring data of the steel support equipped with a servo system collected by the Internet of Things system.

[0021] Further, the steps of probabilistic back analysis in S3 are as follows:

[0022] S31. Determine the prior distribution and prior samples of the key soil parameters to be updated currently through the value range of the key soil parameters in S1 or the posterior samples of the key soil parameters obtained from the previous update.

[0023] S32. Based on the foundation pit lateral deformation monitoring data in S2, considering that the deformation monitoring points of each foundation pit are independent of each other, establish independent likelihood functions for the monitoring data obtained from each monitoring point, and establish a joint likelihood function among multiple monitoring points, where the joint likelihood function is the product of the likelihood functions of single monitoring points.

[0024] S33. Take the servo steel support axial force data in S2 as the axial force parameter value, and take the prior samples of the key soil parameters in S31 as the key soil parameter values, and input them into the foundation pit deformation prediction proxy model to obtain a set of foundation pit deformation prediction values based on the prior samples of the key soil parameters.

[0025] S34. Input the foundation pit deformation prediction value in S33 into the joint likelihood function formula in S32 to obtain the likelihood function value, and then combine with Bayes' theorem to assimilate the foundation pit deformation monitoring data in S2, and obtain the posterior samples of the key soil parameters through Markov chain Monte Carlo (MCMC) simulation sampling, and obtain the statistical characteristics of the posterior samples.

[0026] Further, the iterative optimization of the key soil parameters in S31 is realized by using the sequential Bayesian update framework of the foundation pit deformation prediction proxy model. When updating for the first time, the prior samples of the key soil parameters are initialized and sampled according to the uniform distribution or Gaussian distribution parameters based on the parameter value range of the key soil parameters determined in S1. When updating subsequently, the posterior samples of the key soil parameters obtained from the previous parameter update are used as the prior samples of the key soil parameters for the next parameter update.

[0027] Further, the specific steps of the deformation prediction and axial force regulation module in S4 are as follows:

[0028] S41. According to the range of the axial force parameters in S1, combined with the actual construction conditions of the foundation pit project, preset multiple groups of servo steel support axial force regulation schemes.

[0029] S42. Input the posterior samples of the soil parameters obtained in S3 and the servo axial force values of multiple preset servo steel support axial force regulation schemes in S41 into the foundation pit deformation prediction surrogate model in S1 to obtain the foundation pit deformation prediction for the next excavation stage under different servo steel support axial force regulation schemes.

[0030] S43. According to the foundation pit deformation control requirements, quantitatively evaluate the rationality of the servo steel support axial force regulation scheme by using the failure probability, and propose a reasonable servo regulation scheme for the foundation pit steel support axial force.

[0031] Further, the foundation pit deformation control requirements should include the allowable maximum foundation pit deformation value and the maximum value failure probability value.

[0032] Further, the failure probability represents the proportion of failure samples in the total samples. If the foundation pit deformation prediction of the sample is greater than the maximum foundation pit deformation value, the sample is recorded as a failure sample.

[0033] Further, the statistical characteristics of the posterior samples of the soil parameters include the mean and variance.

[0034] Advantages of the present invention:

[0035] (1) Based on the bidirectional long short-term memory network (BiLSTM) in deep learning, the present invention constructs a surrogate model for foundation pit deformation prediction. This surrogate model can provide calculation results highly consistent with traditional numerical models, and at the same time significantly improve the calculation efficiency. At the same time, using this surrogate model as the forward calculation model in the Bayesian parameter update process can effectively reduce the calculation cost during Bayesian update and improve the feasibility of engineering applications.

[0036] (2) The present invention adopts the data assimilation technology based on Bayesian theory to dynamically update the soil parameters by assimilating the foundation pit deformation monitoring data, thereby improving the prediction accuracy of the surrogate model, further accurately evaluating the influence of the servo steel support axial force setting value on the foundation pit deformation, and thus proposing a scientific and reasonable servo regulation scheme for the foundation pit steel support axial force. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a schematic flow chart of a servo regulation method for the axial force of a foundation pit steel support based on digital twin;

[0038] Figure 2 It is a schematic diagram of a finite element numerical model of a servo regulation method for the axial force of a foundation pit steel support based on digital twin provided in an embodiment of the present invention;

[0039] Figure 3 It is a comparison chart of the calculation results of the surrogate model and the finite element numerical model of a servo regulation method for the axial force of a foundation pit steel support based on digital twin provided in an embodiment of the present invention;

[0040] Figure 4 The prior and posterior distribution diagrams of soil parameters for a soil pressure servo control method of foundation pit steel supports based on digital twin provided in the embodiments of the present invention;

[0041] Figure 5 The foundation pit deformation prediction diagram for a soil pressure servo control method of foundation pit steel supports based on digital twin provided in the embodiments of the present invention;

[0042] Figure 6 The influence result diagram of the set value of the axial force of a single - channel servo steel support for a soil pressure servo control method of foundation pit steel supports based on digital twin provided in the embodiments of the present invention;

[0043] Figure 7 The influence result diagram of the set value of the axial force of a multi - channel servo steel support for a soil pressure servo control method of foundation pit steel supports based on digital twin provided in the embodiments of the present invention. Detailed implementation manners

[0044] To further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following describes in detail the specific implementation manners, structures, features, and their effects of the present invention in combination with the accompanying drawings and preferred embodiments.

[0045] Referring to Figures 1 - 7 as shown, the present invention provides a soil pressure servo control method of foundation pit steel supports based on digital twin, and the steps are as follows:

[0046] S1. Foundation pit engineering simulation module.

[0047] Build a foundation pit numerical model, determine the value ranges of the key soil parameters and axial force parameters of the foundation pit numerical model, and build a foundation pit deformation prediction proxy model with the key soil parameters and the axial force of the servo steel support as inputs by using a deep learning algorithm (bidirectional long short - term memory network BiLSTM). The specific steps are as follows:

[0048] S11. Foundation pit numerical model, perform finite - element numerical modeling on the foundation pit project equipped with a steel support axial force servo system;

[0049] S12. Combine engineering design materials, literature materials, and geological exploration reports to determine the value ranges of the key soil parameters and axial force parameters of the finite - element numerical model. The key soil parameters are mainly the soil parameters that have a significant impact on the foundation pit deformation (that is, based on the actual foundation pit project, determine the soil parameters that significantly affect the foundation pit project), and the axial force parameter is the axial force value of the servo steel support;

[0050] S13. Based on the Latin hypercube sampling method, sample the key soil parameters and axial force parameters to generate N groups of parameter combination samples;

[0051] S14. Perform forward calculations using the finite element numerical model in S11 to obtain the calculated results of the foundation pit deformation responses corresponding to each sample. Using the combination of key soil parameters and axial force parameters as input features and the corresponding foundation pit deformation responses as output labels, construct a dataset for the foundation pit deformation prediction surrogate model that contains N groups of mapping relationships, where N is an integer not less than 1000;

[0052] S15. Divide the dataset in S14 into a training set and a test set according to the ratio of 8:2. Use the training set to train the bidirectional long short-term memory network (BiLSTM) to construct a foundation pit deformation prediction surrogate model;

[0053] S16. Verify through the test set, and use the mean square error (MSE) or root mean square error (RMSE) or mean absolute error (MAE) or coefficient of determination (R 2 ) index to evaluate the performance of the model.

[0054] S2. Internet of Things monitoring module.

[0055] Real-time collect the foundation pit deformation data and the axial force data of the servo steel supports through the Internet of Things monitoring system. Specifically, in the foundation pit project, deploy an inclinometer sensor network along the diaphragm wall of the foundation pit, and deploy an axial force monitoring sensor network on the horizontal support structure of the foundation pit. Realize the real-time collection and transmission of deformation and axial force data through the Internet of Things monitoring platform. The foundation pit deformation data includes the foundation pit lateral deformation monitoring data of multiple monitoring points, and the servo steel support axial force data is the axial force monitoring data of the steel supports equipped with servo systems collected by the Internet of Things system.

[0056] S3. Soil parameter update module.

[0057] Use the foundation pit deformation prediction surrogate model in S1 as the forward calculation model in the probabilistic inverse analysis. Combine Bayes' theorem, and assimilate the foundation pit deformation data in S2 through data assimilation technology to dynamically update the key soil parameters and obtain the posterior samples of the key soil parameters. The specific steps are as follows:

[0058] S31. Determine the prior distribution and prior samples of the current updated key soil parameters through the value range of the key soil parameters in S1 or the posterior samples of the key soil parameters obtained from the previous update. The prior distribution can be denoted as p(θ), where θ represents the key soil parameters. Specifically, the iterative optimization of the key soil parameters is realized by the sequential Bayesian update framework of the foundation pit deformation prediction surrogate model. At the first update, the prior samples of the key soil parameters are initialized by sampling according to the uniform distribution or Gaussian distribution parameters, etc., based on the parameter value range of the key soil parameters determined in S1. In subsequent updates, the posterior samples of the key soil parameters obtained from the previous parameter update are used as the prior samples of the key soil parameters for the next parameter update. The specific joint prior distribution can be expressed by Equation (1):

[0059] (1)

[0060] In the formula, μ lnθj and σ lnθj respectively represent the mean and variance of the key soil parameters after logarithmization. θ j represents the j th key soil parameter, n represents the number of key soil parameters. Based on the above prior distribution, prior samples are generated.

[0061] S32. Based on the foundation pit lateral deformation monitoring data in S2, considering that each foundation pit deformation monitoring point is independent of each other, establish independent likelihood functions for the monitoring data obtained from each monitoring point, and establish a joint likelihood function among multiple monitoring points, where the joint likelihood function is the product of the likelihood functions of single monitoring points. The joint likelihood function can be denoted as p(d|θ). The specific likelihood function is shown in Equation (2).

[0062] (2)

[0063] In the formula, d is the foundation pit lateral deformation monitoring data in S2; θ is the key soil parameter; d i is the data of the i th monitoring point, g(θ) is the foundation pit deformation prediction based on the key soil parameter θ, σ ε is the standard deviation of the observation error (a constant), k represents the number of monitoring points.

[0064] S33. Take the axial force data of the servo steel support in S2 as the axial force parameter value, and take the prior sample of the key soil parameters in S31 as the key soil parameter value, and input them into the foundation pit deformation prediction proxy model to obtain a set of foundation pit deformation prediction values g(θ) based on the prior sample of the key soil parameters;

[0065] S34. Input the foundation pit deformation prediction value g(θ) in S33 into the joint likelihood function formula in S32 to obtain the likelihood function value, and then combine the Bayesian theorem, that is , to obtain the posterior sample. Assimilate the foundation pit deformation monitoring data in S2, and obtain the posterior sample of the key soil parameters (the posterior sample approximating the posterior distribution) through Markov chain Monte Carlo (MCMC) simulation sampling, and obtain the statistical characteristics of the posterior sample. The two statistical characteristics of the posterior sample of the soil parameters include the mean and variance.

[0066] S4. Deformation prediction and axial force regulation module.

[0067] Preset multiple sets of servo steel support axial force regulation schemes, input the posterior sample of the key soil parameters in S3 and the servo steel support axial force parameters in the preset regulation schemes into the S1 foundation pit deformation prediction proxy model, predict the foundation pit deformation in the next stage, and propose a reasonable axial force servo regulation scheme. The specific steps are as follows:

[0068] S41. According to the range of the axial force parameters in S1 and in combination with the actual construction conditions of the foundation pit project, preset multiple sets of servo steel support axial force regulation schemes;

[0069] S42. Input the posterior sample of the soil parameters obtained in S3 and the servo axial force values of the multiple sets of servo steel support axial force regulation schemes preset in S41 into the S1 foundation pit deformation prediction proxy model to obtain the foundation pit deformation prediction in the next excavation stage under different servo steel support axial force regulation schemes;

[0070] S43. According to the requirements of foundation pit deformation control, the rationality of the servo steel support axial force regulation scheme is quantitatively evaluated by the failure probability, and a reasonable foundation pit steel support axial force servo regulation scheme is proposed. Specifically, the foundation pit deformation control requirements include the allowable maximum foundation pit deformation value and the allowable maximum failure probability value. Among them, the allowable maximum foundation pit deformation value is set according to the specification requirements, and the maximum failure probability value can be determined according to engineering experience. The failure probability represents the proportion of failure samples in the total samples. If the predicted foundation pit deformation of a sample is greater than the maximum foundation pit deformation value, this sample is recorded as a failure sample. The maximum foundation pit deformation value can be the maximum lateral displacement value of the retaining structure or the maximum ground surface settlement value outside the pit. Since the axial force application value and the structural failure probability usually show a negative correlation, that is, the larger the axial force application value, the fewer the number of failure samples. Based on this, the determination of the axial force regulation scheme needs to comprehensively consider the balance relationship between the axial force application value and the failure probability. It is recommended to select the working condition with a failure probability value less than the allowable maximum failure probability value and close to this threshold as the optimal servo regulation scheme for the foundation pit steel support axial force. The failure probability values of the remaining non-optimal regulation schemes are also proposed for engineers' reference.

[0071] Example:

[0072] 1) Foundation pit engineering simulation module.

[0073] First, establish a finite element numerical model of the foundation pit project equipped with a steel support axial force servo system (as shown in Figure 2 ). The length and width of this model are both 70 m. The foundation pit excavation depth is 16.4 m, the excavation width is 10 m, and it is excavated in 6 times. The length of the diaphragm wall is 36 m, and a total of 5 horizontal supports are set. Among them, the first one is a concrete support, the second and fourth ones are ordinary prestressed steel supports, and the third and fifth ones are steel supports equipped with a servo system (hereinafter simply referred to as "servo steel supports"). The horizontal spacing of the steel supports is 3 m. The concrete support and the diaphragm wall both use C30 concrete. The diameter and thickness of the ordinary prestressed steel support are 609 mm and 16 mm respectively, and the diameter and thickness of the servo steel support are 800 mm and 16 mm respectively. The soil hardening constitutive model is used to simulate the soil behavior.

[0074] Then, based on the comprehensive engineering design data, literature data, and geological exploration report, determine the key soil parameters of the finite element numerical model as the internal friction angles of clay-1 and clay-2 φ and the reference tangent modulus of the consolidation test , and determine the φ of clay-1 φ -1) and the φ of clay-2 φ -2) with the value range of 9~40°, and the of clay-1 -1) and the (denoted as -2) The value ranges are 1.2 - 4.0 MPa and 3.0 - 6.0 MPa respectively. The value range of the axial force parameter is determined according to the design data and specifications, and is 0 - 1200 kN / m.

[0075] Secondly, based on the above value ranges, 2000 groups of parameter combination samples are generated through Latin hypercube sampling. N = 2000. In this embodiment, the foundation pit deformation is the lateral displacement of the diaphragm wall. These 2000 groups of samples are batch - input into the Figure 2 shown finite - element numerical model to obtain the corresponding calculated values of the diaphragm wall lateral displacement. It should be noted that since the deformation mode of the diaphragm wall in the first excavation stage is different from that in the remaining excavation stages, and its excavation depth is only 1 m, only the lateral displacement of the diaphragm wall in the excavation stages 2 - 5 is considered. Based on these 2000 groups of parameter samples and the corresponding calculated results of the diaphragm wall lateral displacement, a data set is constructed. The data set is randomly divided into a training set (including 1600 groups of data) and a test set (including 400 groups of data) in a ratio of 8:2;

[0076] Finally, based on the training set, the BiLSTM is trained to construct a foundation pit deformation prediction proxy model for a total of 5 stages from excavation stage 2 to excavation stage 6. The input parameters of the proxy model are the key soil parameter values and the servo steel support axial force values, and the number of input parameters changes according to the construction conditions; the number of key soil parameters is always 4 ( φ -1, φ -2, -1 and -2). The performance of the proxy model is tested with the test set and quantitatively evaluated through two indexes, R 2 and MAE. The evaluation results are as Figure 3 shown. It can be seen from Figure 3 that the calculated results of the proxy model are basically on the 45° diagonal line of the calculated results of the numerical model, and both R 2 and MAE are satisfactory, indicating that the established proxy model can replace the foundation pit numerical model as the forward calculation model in Bayesian updating.

[0077] 2) Internet of Things monitoring module.

[0078] The Internet of Things monitoring system collects the servo steel support axial force monitoring data and the foundation pit deformation monitoring data from multiple monitoring points in real - time. In this embodiment, the foundation pit deformation monitoring data comes from 38 monitoring points at different depths of the same inclinometer tube.

[0079] 3) Soil parameter updating module.

[0080] According to the excavation sequence, first assimilate the deformation monitoring data of the second excavation stage and perform the first update on the key soil parameters.

[0081] Since the second excavation stage is the first update, the prior distribution of the soil parameters is determined based on their parameter value ranges. In this embodiment, the prior of the key soil parameters is determined to be a lognormal distribution, with the mean being the median of the range of each key soil parameter and the coefficient of variation being 30%; the joint prior distribution can be expressed by Equation (1):

[0082] (1)

[0083] In the formula, μ lnθj and σ lnθj respectively represent the mean and variance of the key soil parameters after logarithmization. θ j represents the j th key soil parameter. n represents the number of key soil parameters. In this embodiment n = 4. Based on the above prior distribution, prior samples are generated.

[0084] The likelihood function is shown in Equation (2).

[0085] (2)

[0086] In the formula, d is the monitoring data; θ is the key soil parameter; d i is the data of the i th monitoring point, and g(θ) is the foundation pit deformation prediction based on the key soil parameter θ. σ ε is the standard deviation of the observation error, which is taken as 3 mm in this embodiment; k represents the number of monitoring points. k = 38.

[0087] Therefore, based on the prior distribution of the soil parameters in Equation (1) (as shown by "prior" in Figure 4 ) and the servo axial force monitoring data of the second excavation stage, combined with the surrogate model of the second excavation stage, assimilate the lateral displacement monitoring data of the diaphragm wall in the second excavation stage. Through MCMC, obtain the posterior samples of the soil parameters and their statistical characteristics, as shown by "Stage 2" in Figure 4 . At this time, the soil parameters of the surrogate model in the third excavation stage have been updated to the parameter distribution shown by "Stage 2" in Figure 4 . Based on this soil parameter distribution, obtain the foundation pit deformation prediction of the third excavation stage as shown by "posterior" in "Stage 3" in Figure 5 ; "prior" in the figure represents based on the prior distribution of the soil parameters (i.e.,Figure 4 As can be seen from the figure, both the "posterior" and "prior" can basically cover the monitoring data, but the "posterior" is narrower and has lower uncertainty, indicating that the "posterior" is more accurate than the "prior", that is, updating the soil parameters of the proxy model can improve the accuracy of deformation prediction.

[0088] In the deformation prediction and axial force control module:

[0089] Since excavation stage 3 does not involve servo steel supports, deformation prediction and axial force adjustment are not required after the excavation stage 2 update is completed.

[0090] Repeat parameter update and deformation prediction and axial force adjustment:

[0091] After obtaining the foundation pit deformation monitoring of excavation stage 3, Figure 4 The distribution shown in "Stage 2" is used as the prior distribution of key soil parameters. Based on the proxy model of excavation stage 3, the deformation monitoring data of the foundation pit in excavation stage 3 is assimilated. The parameter update results are shown in Figure 4 As shown in "Phase 3" in Figure 5 In "Stage 4", it can be seen that the deformation posterior distribution (i.e., the calculation result obtained based on the updated proxy model) is more able to characterize the monitoring data than the prior, indicating that updating the soil parameters can improve the accuracy of deformation prediction. Since a servo steel support (the second support) needs to be erected after the excavation stage 3 is completed, it is necessary to study the effect of the axial force setting value on the foundation pit deformation and quantitatively evaluate it with the failure probability. In this embodiment, the maximum allowable underground continuous wall side displacement value is 0.4% of the excavation depth, and its specific value is Figure 5 The failure probability study results are shown in Figure 6 As shown in "Stage 4" in Figure 6, it can be seen that the failure probability decreases with the increase of the axial force setting value. Figure 6 As shown in the figure, engineers can select appropriate axial force setting values according to different failure probability requirements, thereby formulating corresponding steel support axial force servo control schemes. For example, if the maximum failure probability is required to be less than 0.5, it is reasonable to set the axial force value of the second support to 750 kN / m or above, and the axial force value of the second support set to 750 kN / m is selected as the optimal control scheme; the failure probabilities of other non-optimal control schemes can be calculated by Figure 6 The curve in is obtained for engineers' reference.

[0092] Similarly, the monitoring data of the second support is used as the axial force parameter value in the proxy model input parameters, and the monitoring data of excavation stage 4 is assimilated to update the soil parameters; the deformation of the subsequent excavation stage is predicted based on the updated soil parameters, and the servo axial force control scheme is analyzed. The probability distribution of the updated soil parameters is as follows: Figure 4as shown in "Phase 4" in; Deformation prediction is as Figure 5 as shown in "Phase 5" in; The research results of the axial force setting value are as Figure 6 as shown in "Phase 5" in.

[0093] Similarly, taking the monitoring data of the second strut as the axial force parameter in the input parameters of the surrogate model, combining with the monitoring data of Excavation Phase 5, assimilating the foundation pit deformation monitoring data of Phase 5, the posterior distribution of the updated soil parameters is as Figure 4 as shown in "Phase 5" in; Foundation pit deformation prediction is as Figure 5 as shown in "Phase 6" in. Since after the completion of Excavation Phase 5, another servo steel strut (the fourth strut) needs to be installed. At this time, there are two servo steel struts in total. Based on the surrogate model updated by assimilating the deformation monitoring data of Excavation Phase 5, study the influence of the axial force setting values of these two servo steel struts (the second and fourth struts) on the foundation pit deformation during Excavation Phase 6. The results are as Figure 7 shown. Based on the determined maximum failure probability requirement, the engineer can Figure 7 determine a suitable steel strut servo control scheme, and then achieve effective control of the foundation pit deformation.

[0094] The above is only a preferred embodiment of the present invention, and it is not intended to limit the present invention in any form. Although the present invention has been disclosed above with the preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can make some changes or modifications to the equivalent embodiments with equivalent changes within the scope of the technical solution of the present invention. However, as long as it does not depart from the content of the technical solution of the present invention, any brief modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. A method for servo control of the axial force of foundation pit steel supports based on digital twin, characterized in that: steps As follows: S1. Foundation pit engineering simulation module, Construct a numerical model of the foundation pit, determine the range of key soil parameters and axial force parameters of the foundation pit numerical model, and use a deep learning algorithm to construct a foundation pit deformation prediction proxy model with key soil parameters and the axial force of the servo steel support as inputs; S2. Internet of Things monitoring module, Collect foundation pit deformation data and servo steel support axial force data through the Internet of Things monitoring system; S3. Soil parameter update module, Take the foundation pit deformation prediction proxy model in S1 as the forward calculation model in the probabilistic inverse analysis. Combine Bayes' theorem, and assimilate the foundation pit deformation data in S2 through data assimilation technology to dynamically update the key soil parameters and obtain the posterior samples of the key soil parameters; S4. Deformation prediction and axial force regulation module, Preset multiple groups of servo steel support axial force regulation schemes. Input the posterior samples of the key soil parameters in S3 and the axial force parameters of the servo steel support in the preset regulation schemes into the foundation pit deformation prediction proxy model in S1 to predict the foundation pit deformation in the next stage and propose a reasonable axial force servo regulation scheme.

2. The method for servo control of the axial force of the foundation pit steel support based on digital twin according to claim 1, wherein: In S1, the specific steps for constructing the foundation pit deformation prediction proxy model are as follows: S11. Foundation pit numerical model, perform finite element numerical modeling on the foundation pit project equipped with a steel support axial force servo system; S12. Combine engineering design data, literature data, and geological exploration reports to determine the value ranges of the key soil parameters and axial force parameters of the finite element numerical model. The axial force parameter is the axial force value of the servo steel support; S13. Based on the Latin hypercube sampling method, sample the key soil parameters and axial force parameters to generate N groups of parameter combination samples; S14. Forward calculation is carried out through the finite element numerical model in S11 to obtain the calculation results of the foundation pit deformation response corresponding to each sample. Taking the combination of key soil parameters and axial force parameters as input features and the corresponding foundation pit deformation response as output labels, a foundation pit deformation prediction surrogate model dataset containing N sets of mapping relationships is constructed, where N is an integer not less than 1000; S15. Divide the dataset in S14 into a training set and a test set in a ratio of 8:2, and use the training set to train the bidirectional long short-term memory network (BiLSTM) to construct a foundation pit deformation prediction proxy model; S16. Verify through the test set and evaluate the performance of the model using the mean squared error (MSE) or root mean squared error (RMSE) or mean absolute error (MAE) or coefficient of determination (R 2 ) index.

3. A method for servo control of the axial force of foundation pit steel supports based on digital twins according to claim 1 or 2, characterized in that: The foundation pit deformation data includes the foundation pit lateral deformation monitoring data of multiple monitoring points, and the servo steel support axial force data is the axial force monitoring data of the steel support equipped with a servo system collected by the Internet of Things system.

4. A method for servo control of the axial force of the foundation pit steel support based on digital twin according to claim 3, characterized in that: The steps of the probabilistic inverse analysis in S3 are as follows: S31. Determine the prior distribution and prior samples of the key soil parameters to be updated currently through the value range of the key soil parameters in S1 or the posterior samples of the key soil parameters obtained from the previous update; S32. Based on the foundation pit lateral deformation monitoring data in S2, considering that each foundation pit deformation monitoring point is independent of each other, establish independent likelihood functions for the monitoring data obtained from each monitoring point, and establish a joint likelihood function between multiple monitoring points. The joint likelihood function is the product of the likelihood functions of single monitoring points; S33. Input the servo steel support axial force data in S2 as the axial force parameter value and the prior samples of the key soil parameters in S31 as the key soil parameter values into the foundation pit deformation prediction proxy model to obtain a set of foundation pit deformation prediction values based on the prior samples of the key soil parameters; S34. Input the foundation pit deformation prediction value in S33 into the joint likelihood function formula in S32 to obtain the likelihood function value. Then, combine Bayes' theorem, assimilate the foundation pit deformation monitoring data in S2, and obtain the posterior samples of the key soil parameters through Markov chain Monte Carlo (MCMC) simulation sampling, and obtain the statistical characteristics of the posterior samples.

5. A method for servo control of the axial force of foundation pit steel supports based on digital twin according to claim 4, characterized in that: The iterative optimization of the key soil parameters in S31 is realized by the sequential Bayesian update framework using the foundation pit deformation prediction surrogate model. At the first update, the prior samples of the key soil parameters are initialized and sampled according to the parameter value ranges of the key soil parameters determined in S1, using a uniform distribution or a Gaussian distribution parameter distribution. In subsequent updates, the posterior samples of the key soil parameters obtained from the previous parameter update are used as the prior samples of the key soil parameters for the next parameter update.

6. A method for servo control of the axial force of foundation pit steel supports based on digital twin according to claim 1, characterized in that: S4 The specific steps of the deformation prediction and axial force control module are as follows: S41. According to the range of the axial force parameters in S1 and combining with the actual construction conditions of the foundation pit project, multiple groups of servo steel support axial force control schemes are preset. S42. The posterior samples of the soil parameters obtained in S3 and the servo axial force values of the multiple groups of servo steel support axial force control schemes preset in S41 are input into the foundation pit deformation prediction surrogate model in S1 to obtain the foundation pit deformation prediction for the next excavation stage under different servo steel support axial force control schemes. S43. According to the foundation pit deformation control requirements, the rationality of the servo steel support axial force control scheme is quantitatively evaluated using the failure probability, and a reasonable foundation pit steel support axial force servo control scheme is proposed.

7. A method for servo control of the axial force of the foundation pit steel support based on digital twin according to claim 6, characterized in that: The foundation pit deformation control requirements include the allowable maximum foundation pit deformation value and the maximum value failure probability value.

8. A method for servo regulation of axial force of foundation pit steel supports based on digital twin according to claim 7, characterized in that: The failure probability represents the proportion of failure samples in the total samples. If the foundation pit deformation prediction of the sample is greater than the maximum foundation pit deformation value, the sample is recorded as a failure sample.

9. A method for servo control of the axial force of foundation pit steel supports based on digital twin according to claim 4, characterized in that: The statistical characteristics of the posterior samples of the soil parameters include the mean and variance.

Images