A foundation pit deformation prediction method fusing multi-physical models and monitoring data

Abstract

The present application relates to a kind of fusion multi-physical model and the foundation pit deformation prediction method of monitoring data, steps are as follows: S1.based on the soil parameter statistics of foundation pit deformation calculation physical model, then through calculation physical model, obtain corresponding deformation calculation response, construct dictionary model;S2. Observation matrix is gradually constructed, from S1 dictionary model, identify the atom with high correlation of observation matrix, determine important atom, and estimate the weight and its uncertainty of important atom using Bayesian inference method;S3. Based on the important atom determined in S2, using sparse dictionary learning, linear weighted combination is carried out in combination with the weight result estimated, and the approximate representation of foundation pit excavation deformation and subsequent construction stage are carried out deformation prediction;S4. Dynamic update subsequent deformation prediction result.The present application fuses physical model information and field monitoring data, can realize the accurate prediction and uncertainty quantification of foundation pit deformation, and provides technical support for foundation pit engineering safety control.

Description

Technical Field

[0001] This invention relates to the field of foundation pit engineering, and in particular to a method for predicting foundation pit deformation by integrating multi-physics models and monitoring data. Background Technology

[0002] Existing methods for improving excavation deformation prediction by integrating on-site monitoring data mainly include inverse analysis and machine learning methods. Inverse analysis methods perform forward calculations based on physical models, offering good physical interpretability and allowing for the prediction of deformation at each subsequent excavation stage, even the entire process, by combining acquired phased monitoring data. However, these methods typically rely on a single forward calculation model, making the prediction results highly dependent on the model's validity. When the chosen model fails to accurately represent the complex deformation mechanisms in actual engineering projects, satisfactory prediction results are often difficult to obtain. Furthermore, probabilistic inverse analysis methods often suffer from high computational costs and low efficiency when considering soil parameter uncertainties. In contrast, machine learning methods offer higher computational efficiency, but purely data-driven methods typically require high-quality and high-quantity monitoring data. Due to economic and technical limitations, deformation monitoring data in excavation projects is often incomplete and sparse in both time and space, making it difficult to obtain sufficient data to train prediction models for specific projects. Furthermore, existing machine learning-based excavation deformation prediction methods are typically only applicable to short-term predictions of the next few days, making it difficult to effectively predict deformation over multiple excavation stages that can span several weeks. Therefore, there is an urgent need to provide a foundation pit deformation prediction method that can consider multiple physical models and integrate on-site monitoring data, achieving accurate and efficient prediction of foundation pit deformation and overcoming the aforementioned shortcomings of existing technologies. Summary of the Invention

[0003] In view of the shortcomings of the prior art, the technical problem to be solved by the present invention is to provide a method for predicting foundation pit deformation by integrating multi-physics models and monitoring data. This method enables the prediction and dynamic updating of deformation responses at each subsequent construction stage during foundation pit excavation, improving the accuracy and reliability of foundation pit deformation prediction and providing technical support for the safety control of foundation pit engineering.

[0004] This invention is achieved by the following technical solution: a method for predicting foundation pit deformation by integrating multi-physics models and monitoring data, comprising the following steps: S1. Constructing a dictionary based on the deformation response of the foundation pit using a multi-physics model. Based on multiple foundation pit deformation calculation physical models, soil parameter statistics are determined, and then the corresponding deformation calculation response is obtained through the corresponding calculation physical model to construct a dictionary model. S2. Identify important atoms and estimate their weights. The observation matrix is ​​constructed in stages, atoms with high correlation to the observation matrix are identified from the S1 dictionary model, important atoms are determined, and the weights and uncertainties of important atoms are estimated by Bayesian inference method. S3. Prediction of excavation deformation in foundation pits Based on the important atoms identified in S2, sparse dictionary learning is used, and linear weighted combination is performed with the estimated weight results to approximate the deformation of the foundation pit excavation and predict the deformation in the subsequent construction stage. S4. Dynamic updates, Continuously acquire monitoring data for each new excavation stage, repeat S2 and S3 for each stage, and dynamically update the excavation deformation prediction results for subsequent construction stages.

[0005] Furthermore, the specific steps of S1 include: S11. Determine the computational physical model for the deformation of the foundation pit; S12. Using on-site survey data and prior knowledge, determine the statistical quantities of soil parameters required in the physical model for calculating foundation pit deformation; S13. Generate soil parameter samples based on the statistics of the soil parameters, and input the soil parameter samples into the physical model for foundation pit deformation calculation to obtain the corresponding deformation calculation response; S14. Calculate the response based on the deformation and construct a dictionary model.

[0006] Furthermore, the specific steps of S2 include: S21. Taking the excavation stage as the data collection stage, collect on-site foundation pit deformation monitoring data and construct the observation matrix stage by stage; S22. Using the orthogonal matching pursuit method, based on a set correlation threshold, identify atoms with high correlation to the observation matrix from the dictionary model of S1, and determine the identified atoms as important atoms; S23. Use Bayesian inference to estimate the weights and uncertainties of the important atoms identified in S22.

[0007] Furthermore, the physical model for calculating the foundation pit deformation in S11 includes two or more of the following: numerical model, analytical model, and empirical model constructed based on a specific project. The foundation pit deformation includes lateral displacement of the retaining structure and settlement of the ground surface behind the wall. Furthermore, the statistics of the soil parameters are determined by the assumed prior distribution. When the prior distribution is uniform, the statistics include the upper and lower limits of the parameter values; when the prior distribution is Gaussian, the statistics include the mean and standard deviation of the parameters.

[0008] Furthermore, the observation matrix constructed in S21 has a dimension of P×S, where P represents the number of on-site monitoring points and S represents the number of excavation stages to be considered. The observation matrix is ​​gradually expanded with the excavation stages, and the monitoring data of the current stage is added to the observation matrix of the previous stage to construct the observation matrix of the target stage.

[0009] Furthermore, in S22, the number of important atoms used to identify them is determined by setting a correlation threshold. When a new important atom is introduced, if the correlation between the atom with the highest correlation and the observation matrix constructed in the current stage is less than the set correlation threshold, it is considered that the newly added atom has no significant correlation with the observation matrix, and the introduction of the atom is rejected and the iteration stops.

[0010] Furthermore, the step in S23 where the Bayesian inference method estimates the weights of the identified important atoms and their uncertainties includes: 1) The residual between the observation matrix and the deformation prediction value follows a normal distribution, and the prior distribution of its accuracy parameter is a gamma distribution; 2) A hierarchical prior strategy is adopted to construct the prior distribution of important atom weights, wherein the important atom weights follow a normal distribution, the precision parameter of the normal distribution follows an inverse gamma distribution, and the scale parameter of the inverse gamma distribution follows a gamma distribution. 3) Construct a joint prior distribution considering residual uncertainty and weight uncertainty; 4) Construct a likelihood function based on the sparse Bayesian method, and obtain the joint posterior probability distribution expression by combining the joint prior distribution with the likelihood function; 5) Obtain posterior samples of important atom weights through Gibbs sampling, and determine the estimated values ​​of important atom weights and quantify their uncertainty based on the posterior samples.

[0011] Furthermore, the physical model for calculating the deformation of the foundation pit in S11 consists of at least two physical models; In S12, the uncertain soil parameters corresponding to each physical model are determined using field survey data and prior knowledge, forming the statistics of soil parameters. In S13, based on the range or probability distribution of the above soil parameters, N sets of soil parameter samples are generated using Latin hypercube or Monte Carlo sampling methods. The soil parameter samples include multiple sets of parameter samples corresponding to each physical model. Each set of soil parameter samples is input into the corresponding calculation model to obtain the calculation response results of the foundation pit lateral displacement and surface settlement of each physical model. In S14, the calculated response results corresponding to each physical model obtained in S13 are combined to construct a dictionary B={B i}

[0012] Furthermore, the step of employing the orthogonal matching pursuit method in S22 includes: 1) Set the initial residual to r (0) = d, Activation Atom Set And let the number of iterations be t = 0; 2) in the t In the nth iteration, for the th i Matrix A i After normalization, the normalized matrix A is obtained. i ′, and calculate its relationship with the current residual r. (t) correlation ; 3) Determine the atomic index with the highest relevance and set a coherence threshold. δ As a termination condition for iteration.

[0013] Beneficial effects of this invention: (1) The present invention adopts a sparse dictionary learning method. Compared with the probabilistic inverse analysis method containing physical information, it does not require repeated calls to the forward calculation model, which can effectively reduce the calculation cost and improve the calculation efficiency. At the same time, it can integrate multiple physical models and avoid prediction deviations caused by unreasonable selection of a single forward calculation model.

[0014] (2) This invention constructs a dictionary through a physical model. Compared with the pure data-driven machine learning method, it does not rely on a large amount of historical training data and can integrate the existing physical knowledge of foundation pit excavation. Thus, under the condition of limited monitoring data, it can effectively predict the deformation of foundation pit in all subsequent excavation stages and has good engineering application value.

[0015] (3) This invention estimates the weights of important atoms through Bayesian inference and quantifies their uncertainty, thereby characterizing the uncertainty in the prediction of foundation pit deformation and improving the reliability of the prediction results. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a method for predicting foundation pit deformation by integrating multi-physics models and monitoring data. Figure 2 This is a schematic diagram of the finite element numerical model provided in the embodiments of the present invention; Figure 3 This is a schematic diagram of the lateral displacement field monitoring data provided in the embodiments of the present invention; Figure 4 This is a schematic diagram of the on-site monitoring data of surface subsidence provided in the embodiments of the present invention; Figure 5 This is a schematic diagram of the lateral displacement prediction results at each stage provided in the embodiments of the present invention; Figure 6This is a schematic diagram of the surface subsidence prediction results at each stage provided in the embodiments of the present invention. Detailed Implementation

[0017] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structure, features and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0018] Reference Figures 1-6 As shown, this invention provides a method for predicting foundation pit deformation by integrating multi-physics models and monitoring data, the steps of which are as follows: S1. Constructing a dictionary of foundation pit deformation response based on a multi-physics model.

[0019] Soil parameter statistics are determined based on multiple computational physical models for foundation pit deformation. Then, the corresponding deformation calculation response is obtained through the corresponding computational physical model, and a dictionary model is constructed. The specific steps are as follows: S11. Determine the computational physical model for the foundation pit deformation. The computational physical model for foundation pit deformation includes two or more of the following: a numerical model, an analytical model, and an empirical model constructed based on a specific project. The foundation pit deformation includes lateral displacement of the retaining structure and settlement of the ground surface behind the wall.

[0020] S12. Using field survey data and prior knowledge, determine the uncertain soil parameters corresponding to each physical model, compose the statistics of the soil parameters, and determine the statistics of the soil parameters required in the physical model for calculating foundation pit deformation. The statistics of the soil parameters are determined by the assumed prior distribution. When the prior distribution is uniform, the statistics include the upper and lower limits of the parameter values; when the prior distribution is Gaussian, the statistics include the mean and standard deviation of the parameters.

[0021] S13. Based on the statistical values ​​of the soil parameters, generate soil parameter samples and input the soil parameter samples into the corresponding physical model for foundation pit deformation calculation to obtain the corresponding deformation calculation response. Specifically, based on the value range or probability distribution of the above soil parameters, use the Latin hypercube sampling method to generate N sets of soil parameter samples. The soil parameter samples include multiple sets of parameter samples corresponding to each physical model. Input each set of soil parameter samples into the corresponding calculation model to obtain the calculation response results of foundation pit lateral displacement and surface settlement for each physical model.

[0022] S14. Construct a dictionary model based on the deformation calculation response. Specifically, combine the calculation response results corresponding to each physical model obtained in S13 to construct a dictionary B = {B...} i}

[0023] S2. Identify important atoms and estimate their weights.

[0024] The observation matrix is ​​constructed stage by stage. Atoms with high correlation to the observation matrix are identified from the S1 dictionary model to determine important atoms. Bayesian inference methods are then used to estimate the weights and uncertainties of these important atoms. Specific steps include: S21. Taking the excavation stage as the data collection stage, collect on-site foundation pit deformation monitoring data and construct an observation matrix stage by stage. The constructed observation matrix has a dimension of P×S, where P represents the number of on-site monitoring points and S represents the number of excavation stages considered. The observation matrix is ​​gradually expanded with each excavation stage, adding the monitoring data of the current stage to the observation matrix of the previous stage to construct the observation matrix for the target stage.

[0025] S22. Using the orthogonal matching pursuit method, based on a set correlation threshold, atoms with high correlation to the observation matrix are identified from the dictionary model of S1, and the identified atoms are determined as important atoms. Specific steps include: 1) Set the initial residual to r (0) = d, Activation Atom Set And let the number of iterations be t = 0.

[0026] 2) in the t In the nth iteration, for the th i Matrix A i After normalization, the normalized matrix A is obtained. i ′, and calculate its relationship with the current residual r. (t) correlation .

[0027] 3) Determine the atomic index with the highest relevance and set a relevance threshold. δ This serves as the termination condition for the iteration. When a new important atom is introduced, if the correlation between the atom with the highest correlation and the observation matrix constructed in the current stage is less than the set correlation threshold, then... δ If the newly added atom is not significantly correlated with the observation matrix (i.e., the remaining residual no longer contains information that is significantly related to the deformation characteristics of the foundation pit), the atom is rejected and the iteration stops.

[0028] S23. Use Bayesian inference to estimate the weights and uncertainties of the important atoms identified in S22. Specific steps include: 1) The residual between the observation matrix and the deformation prediction value follows a normal distribution, and the prior distribution of its accuracy parameter is a gamma distribution.

[0029] 2) A hierarchical prior strategy is adopted to construct the prior distribution of important atom weights, wherein the important atom weights follow a normal distribution, the precision parameter of the normal distribution follows an inverse gamma distribution, and the scale parameter of the inverse gamma distribution follows a gamma distribution.

[0030] 3) Construct a joint prior distribution based on the consideration of residual uncertainty and weight uncertainty.

[0031] 4) Construct a likelihood function based on the sparse Bayesian method, and obtain the joint posterior probability distribution expression by combining the joint prior distribution and the likelihood function.

[0032] 5) Obtain posterior samples of important atom weights through Gibbs sampling, and determine the estimated values ​​of important atom weights and quantify their uncertainty based on the posterior samples.

[0033] S3. Prediction of excavation deformation of foundation pit.

[0034] Based on the key atoms identified in S2, sparse dictionary learning is employed, combined with the estimated weights, to perform a linear weighted combination, providing an approximate representation of the excavation deformation of the foundation pit and predicting deformation in subsequent construction stages. Specifically, the posterior samples in S2... Substitute each element into the following formula to predict the foundation pit deformation data D, thus achieving probabilistic prediction of foundation pit deformation.

[0035]

[0036] S4. Dynamic updates.

[0037] As monitoring data from the new excavation phase is continuously acquired, observation matrices d for each phase are constructed sequentially. i S2 and S3 are executed repeatedly in stages, and the excavation deformation prediction results of subsequent construction stages are dynamically updated.

[0038] Example: Refer to Figures 2-6 As shown, based on a certain target foundation pit project, the probability prediction of foundation pit deformation is carried out using the above-mentioned foundation pit deformation prediction method that integrates multi-physics model and monitoring data.

[0039] S1. Construct the dictionary model. This includes the following: S11. Determine the physical model for deformation calculation based on the target foundation pit project. In this embodiment, two physical models are used: a finite element model and a modified elastic foundation beam model (analytical solution model of the physical model).

[0040] like Figure 2 As shown, the first physical model is a finite element model. The maximum excavation depth of the foundation pit is 19.7 m, and the excavation width is 41.2 m. The excavation process is divided into 7 stages (of which... Figure 2(A finite element model was created for half of the excavation pit). To reduce the impact of boundary conditions on the calculation results, the distance from the diaphragm wall to the right boundary was set to approximately four times the maximum excavation depth. The retaining structure consists of a diaphragm wall with a thickness of 0.9 m and a depth of 35 m, five concrete floor slabs, and two steel supports. The site is divided into six layers from top to bottom, mainly consisting of alternating layers of clay and sand. Plate elements were used to simulate the diaphragm wall, with a Young's modulus of 21 MPa and a Poisson's ratio of 0.15. Point-to-point anchor elements were used to simulate both the concrete floor slabs and the steel supports. The Young's modulus of the concrete floor slabs was 21 MPa, and the Young's modulus of the steel supports was 210 MPa. Both had a Poisson's ratio of 0.15, and the concrete floor slab thickness was 0.15 m. The cross-sectional areas of the upper and lower steel supports were 0.012 m² and 0.0219 m², respectively, with horizontal spacing of 8.0 m and 3.4 m. The mechanical behavior of sand and clay was simulated using the Mohr-Coulomb model and the soil hardening model, respectively.

[0041] The second physical model is the analytical solution model, specifically the modified elastic foundation beam model.

[0042] First, the wall lateral displacement curve is obtained based on the elastic foundation beam theory. Then, combining the soil loss theory and the empirical relationship between wall lateral displacement and surface settlement, the surface settlement profile is calculated. The governing equations of the elastic foundation beam theory are as follows: (1) In the formula, EI The bending stiffness of the wall; y Indicates lateral displacement of the wall; z To calculate the location depth; cf ( z ) for depth z The soil spring constant at that location; b 0 represents the calculated width of the wall, which is usually taken as 1 m; e a ( z ) for depth z Active earth pressure at the location. Assume the earth spring constant. cf ( z The depth increases linearly with the depth below the excavation surface. (2) In the formula, h e Indicates the excavation depth at the target stage; k s The slope is the soil spring constant; z b Indicates the depth below the excavation surface. z b = z -h e The active earth pressure theory of Rankine was used to calculate... e a ( z ), (3) In the formula, k a The slope of the earth pressure above the excavation surface; q The excavation caused overloading of the surrounding surface. γ w This represents the weighted unit weight of the soil layer. Assuming the surface settlement profile follows a normal distribution curve, the distance from the excavation pit... x Surface subsidence at point m s ( x )for: (4) In the formula, s m This represents the maximum settlement value. x m It is the horizontal distance from the location of maximum settlement to the edge of the pit; r denoted as the standard deviation of the normal distribution curve. x m and r The value of is related to the maximum excavation depth of the foundation pit. H e related, x m = α 1 H e (5) r = α 2 H e (6) In the formula, α 1 and α 2 is an empirical coefficient. Based on the formation loss theory, the envelope area of ​​the settlement curve is assumed to be... A s Envelope area of ​​wall lateral displacement A d The following relationship must be satisfied: A s = α 3 A d (7) when A s , x m and rIt is known that the following formula can be used to calculate... s m : (8) S12. Determine the statistical quantities of the soil parameters required in the physical model for calculating the deformation of the foundation pit.

[0043] The three layers of clay in the finite element model These soil parameters are considered to have uncertainties. Based on comprehensive survey data, literature review, and engineering experience, the 95% confidence intervals (i.e., the 2.5th and 97.5th percentiles) for these three clay layers are determined to be 1.4–6.4 MPa, 1.4–8.4 MPa, and 5.0–10.0 MPa. This will be applied to the modified elastic foundation beam model. k s , k a , α 1. α 2 and α 3 is used as an uncertainty parameter. Among them, k s Obligation is based on a lower limit of 0 and an upper limit of 1×10. 4 kN / m 4 Uniform distribution; k a It follows a mean of 11.7 kN / m 4 A log-normal distribution with a coefficient of variation of 0.3; α 1. It follows a log-normal distribution with a mean of 0.5 and a coefficient of variation of 0.3; α 2. They follow a log-normal distribution with a mean of 0.75 and a coefficient of variation of 0.3. α 3 follows a log-normal distribution with a mean of 1 and a coefficient of variation of 0.3.

[0044] S13. Obtain the corresponding deformation calculation response.

[0045] Based on the aforementioned range or probability distribution of soil parameters, 100 sets of soil parameter samples were generated using the Latin hypercube sampling method. 50 sets of samples corresponded to the finite element model, and 50 sets corresponded to the modified elastic foundation beam model. Each set of soil parameter samples was input into its corresponding calculation model to obtain the calculated response results of the foundation pit lateral displacement and surface settlement. The arrangement of the observation points and... Figure 3 and Figure 4 The actual monitoring points for the foundation pit project shown are consistent, with a total of 36 monitoring points (17 lateral displacement monitoring points and 19 surface settlement monitoring points); the selected calculation response stage is consistent with the excavation stage of the measured data, specifically stage 3 to stage 7.

[0046] S14. Construct a dictionary model.

[0047] Combining the 50 sets of calculated responses from the finite element model in S1.3 and the 50 sets of calculated responses from the modified elastic foundation beam, a dictionary B={B i}, i = 1, 2, …, 100; Dictionary atom B i The dimensions are 36×5.

[0048] S2. Identify important atoms and estimate their weights.

[0049] S21. Construct the observation matrix. For excavation stage 3, based on... Figure 3 and Figure 4 The measured data for this stage are used to construct an observation matrix d3, which has a dimension of 36×1, and the subscript number "3" indicates the target stage. Based on the monitoring point information corresponding to the observation matrix, from dictionary atom B... i Extract the data at the corresponding positions to obtain matrix A. 3,i Its dimensions are 36×1, and A3={A 3,i}, i = 1, 2, …, 100.

[0050] S22. Identify important atoms.

[0051] In this embodiment, δ Take 0.15. When the maximum correlation is greater than or equal to... δ At that time, the corresponding atom index is added to the set of activated atoms S. Based on the atom indices in set S, the corresponding important atoms are extracted from the unnormalized dictionary to form a sub-dictionary A. 3,s And update the sparse coefficients and residuals using least squares regression: (9) (10) In the formula, ω Represents the atomic weight vector; ω s This represents the weight vector for important atoms, where only the weights of important atoms are retained, and the weights of the remaining atoms are set to zero. The atom identification process described above is repeated until the termination condition is met.

[0052] S23. Estimate the weights and uncertainties of important atoms. Assume that the observation matrix and the deformation prediction satisfy... (11) in, This represents the important atomic weight vector that takes into account uncertainty; ε The residuals are represented by a mean of zero and a variance of 1 / 2. τ Normal distribution; I is the identity matrix.

[0053] Assuming accuracy parameters τ The prior distribution follows the shape parameter a 1. Rate parameter b 10 -4 The gamma distribution is then used. Next, a hierarchical prior strategy is employed to construct the prior distribution of important atom weights, where the important atom weight vector... Assuming atomic weights Follows a random variable with a mean of zero and a variance of 1 / α i The normal distribution α i Obey the scale parameter is λ / 2. Inverse gamma distribution with a shape parameter of 1. Parameter λ Let be a random variable whose prior distribution follows a shape parameter c and rate parameters d All are 10 -4 The gamma distribution. Therefore, a joint prior distribution can be constructed. Further combining this with the likelihood function yields the joint posterior probability distribution expression. Finally, sampling is used to obtain... N = 1000 groups The posterior samples are used to quantify the uncertainty of the weights of important atoms.

[0054] S3. Predict the probability of foundation pit deformation. The aforementioned... Substitute the posterior samples one by one into the following formula to predict the foundation pit deformation data D, thereby achieving the probability prediction of foundation pit deformation.

[0055] (12) S4. Dynamic updates.

[0056] As monitoring data from the new excavation phase is continuously acquired, observation matrices d for each phase are constructed sequentially. i The observation matrix d4 for stage 4 has a dimension of 36×2; the observation matrix d5 for stage 5 has a dimension of 36×3; and the observation matrix d6 for stage 6 has a dimension of 36×4. Since stage 7 is the final excavation stage and there are no subsequent excavation stages, it is not necessary to construct the observation matrix d7 for this stage.

[0057] To further verify the effectiveness of the present invention, in this embodiment, existing technology is also used to predict the deformation in subsequent stages of the project. Specifically, the training data used in the existing technology consists of... Figure 2 The finite element model shown is generated. The predicted results for wall lateral displacement and ground settlement are as follows: Figure 5 and Figure 6As shown. Wherein, "Prediction of this invention" represents the 95% confidence interval of the prediction result obtained using the technology provided by this invention; "Prior technology" represents the 95% confidence interval of the prediction result obtained using prior technology.

[0058] Adopt the current stage (stage) i The observation matrix is ​​used for the next excavation stage (stage). i The wall lateral displacement of +1) is predicted, and the prediction results are as follows: Figure 5 As shown. By Figure 5 It can be seen that the deformation trend represented by the predicted wall lateral displacement interval obtained by the present invention is basically consistent with the measured value. As the excavation stage progresses, the observation information contained in the observation matrix gradually increases, and the coverage of the predicted interval with the measured value gradually improves. Compared with the technology of the present invention, the lateral displacement prediction interval obtained by the existing technology is wider, indicating that its prediction uncertainty is greater. This shows that the present invention can effectively improve the prediction accuracy of wall lateral displacement in the subsequent excavation stage.

[0059] Adopt the current stage (stage) i The observation matrix is ​​used for the next excavation stage (stage). i +1) is used to predict surface subsidence, and the prediction results are as follows: Figure 6 As shown. By Figure 6 It can be seen that the deformation trend represented by the predicted surface settlement interval obtained by the present invention is basically consistent with the measured values. After introducing the monitoring data from excavation stages 3 to 6, the predicted surface settlement result for stage 7 is highly consistent with the measured values. Compared with the technology of the present invention, the existing technology provides a wider settlement prediction interval, indicating greater prediction uncertainty. Furthermore, the existing technology cannot accurately predict the location of the maximum settlement value, especially for stage 7, where it fails to effectively identify the main impact range of settlement. Therefore, the present invention has better applicability and effectiveness in surface settlement prediction.

[0060] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for predicting foundation pit deformation by integrating multi-physics models and monitoring data, characterized by the following steps: as follows: S1. Constructing a dictionary based on the deformation response of the foundation pit using a multi-physics model. Based on multiple foundation pit deformation calculation physical models, soil parameter statistics are determined, and then the corresponding deformation calculation response is obtained through the corresponding calculation physical model to construct a dictionary model. S2. Identify important atoms and estimate their weights. The observation matrix is ​​constructed in stages. Atoms with high correlation to the observation matrix are identified from the S1 dictionary model to determine important atoms. Bayesian inference methods are then used to estimate the weights and uncertainties of important atoms, specifically including: S21. Taking the excavation stage as the data collection stage, collect on-site foundation pit deformation monitoring data and construct the observation matrix stage by stage; S22. Using the orthogonal matching pursuit method, based on the set correlation threshold, the atoms with high correlation to the observation matrix are identified from the dictionary model of S1, and the identified atoms are determined as important atoms. The number of important atoms used to identify them is determined by setting the correlation threshold. When a new important atom is introduced, if the correlation between the atom with the highest correlation and the observation matrix constructed in the current stage is less than the set correlation threshold, it is considered that the new important atom has no significant correlation with the observation matrix, the atom is rejected, and the iteration stops. S23. The weights and uncertainties of the important atoms identified in S22 are estimated using the Bayesian inference method. Specifically, this includes constructing a prior distribution of the weights of important atoms using a hierarchical prior strategy, obtaining posterior samples of the weights of important atoms through Gibbs sampling, and determining the estimated values ​​of the weights of important atoms and quantifying their uncertainties based on the posterior samples. S3. Prediction of excavation deformation in foundation pits. Based on the important atoms identified in S2, sparse dictionary learning is used, and linear weighted combination is performed with the estimated weight results to approximate the deformation of the foundation pit excavation and predict the deformation in the subsequent construction stage. S4. Dynamic updates, Continuously acquire monitoring data for each new excavation stage, repeat S2 and S3 for each stage, and dynamically update the excavation deformation prediction results for subsequent construction stages.

2. The method for predicting foundation pit deformation by integrating multi-physics models and monitoring data according to claim 1, characterized in that: The specific steps of S1 include: S11. Determine the computational physical model for the deformation of the foundation pit; S12. Using on-site survey data and prior knowledge, determine the statistical quantities of soil parameters required in the physical model for calculating foundation pit deformation; S13. Generate soil parameter samples based on the statistics of the soil parameters, and input the soil parameter samples into the physical model for foundation pit deformation calculation to obtain the corresponding deformation calculation response; S14. Calculate the response based on the deformation and construct a dictionary model.

3. The method for predicting foundation pit deformation by integrating multi-physics models and monitoring data according to claim 2, characterized in that: The physical model for calculating the foundation pit deformation in S11 includes two or more of the following: numerical model, analytical model, and empirical model based on engineering construction. The foundation pit deformation includes lateral displacement of the retaining structure and settlement of the ground surface behind the wall.

4. A method for predicting foundation pit deformation by integrating multi-physics models and monitoring data according to claim 2 or 3, characterized in that: The statistics of the soil parameters are determined by the assumed prior distribution. When the prior distribution is uniform, the statistics include the upper and lower limits of the parameter values; when the prior distribution is Gaussian, the statistics include the mean and standard deviation of the parameters.

5. The method for predicting foundation pit deformation by integrating multi-physics models and monitoring data according to claim 1, characterized in that: The observation matrix d constructed in S21 has a dimension of P×S, where P represents the number of on-site monitoring points and S represents the number of excavation stages to be considered. The observation matrix is ​​gradually expanded with the excavation stage, and the monitoring data of the current stage is added to the observation matrix of the previous stage to construct the observation matrix of the target stage.

6. The method for predicting foundation pit deformation by integrating multi-physics models and monitoring data according to claim 1, characterized in that: The steps in S23 for estimating the weights and uncertainties of the identified important atoms using the Bayesian inference method include: (1) The residuals between the observation matrix and the deformation predictions follow a normal distribution, and the prior distribution of its accuracy parameter is a gamma distribution. ； (2) A hierarchical prior strategy is adopted to construct the prior distribution of important atom weights, wherein the important atom weights follow a normal distribution, the precision parameter of the normal distribution follows an inverse gamma distribution, and the scale parameter of the inverse gamma distribution follows a gamma distribution. (3) Construct a joint prior distribution based on considering residual uncertainty and weight uncertainty; (4) Construct the likelihood function based on the sparse Bayesian method, and obtain the joint posterior probability distribution expression by combining the joint prior distribution and the likelihood function; (5) Obtain posterior samples of important atom weights through Gibbs sampling, and determine the estimated values ​​of important atom weights and quantify their uncertainty based on the posterior samples.

7. A method for predicting foundation pit deformation by integrating multi-physics models and monitoring data according to claim 2 or 3, characterized in that: The physical model for calculating the deformation of the foundation pit in S11 consists of at least two physical models; In S12, the uncertain soil parameters corresponding to each physical model are determined using field survey data and prior knowledge, forming the statistics of soil parameters. In S13, based on the range or probability distribution of the above soil parameters, N sets of soil parameter samples are generated using Latin hypercube or Monte Carlo sampling methods. The soil parameter samples include multiple sets of parameter samples corresponding to each physical model. Each set of soil parameter samples is input into the corresponding calculation model to obtain the calculation response results of the foundation pit lateral displacement and surface settlement of each physical model. In S14, the calculated response results corresponding to each physical model obtained in S13 are combined to construct a dictionary B={B i } 8. The method for predicting foundation pit deformation by integrating multi-physics models and monitoring data according to claim 1, characterized in that: The steps in S22 using the orthogonal matching pursuit method include: (1) Set the initial residual to r (0) = d, Activation Atom Set And let the number of iterations be t = 0; (2) in the t In the nth iteration, for the th i Matrix A i After normalization, the normalized matrix A is obtained. i ′, and calculate its relationship with the current residual r. (t) correlation ; (3) Determine the atomic index with the highest relevance and set the relevance threshold. δ As a termination condition for iteration.

Images