A dynamic correction method of geological model based on multi-source monitoring data inversion

Abstract

The present application relates to geotechnical engineering technical field, especially to a kind of geological model dynamic correction method based on inversion of multi-source monitoring data.The method first establishes initial collaborative analysis model, and uses intrinsic orthogonal decomposition to build the forecast working condition set under different support-reinforcement combination;With deformation limit, bearing capacity safety factor as constraint, with comprehensive cost and construction period as target, solve the pareto optimal solution set by multi-objective genetic algorithm, generate support-reinforcement scheme library.In construction, synchronously collect foundation pit monitoring data and existing building foundation dynamic response data, after pretreatment, as multi-source observation vector;Based on initial model, construct forward prediction operator, use set kalman filter to sequentially invert and update soil parameters, and the updated parameters are replaced back to model to predict the remaining construction stage structure response;According to the early warning threshold trigger early warning and recommend adjustment measures from scheme library, feedback to site and identification platform.

Description

Technical Field

[0001] This invention relates to the field of geotechnical engineering technology, and in particular to a method for dynamic correction of geological models based on multi-source monitoring data inversion. Background Technology

[0002] With the acceleration of urban renewal, deep foundation pit projects in old urban areas are often adjacent to dense and weak existing building clusters. Traditional design methods often separate foundation pit support from the foundation reinforcement of surrounding buildings into two independent links, resulting in serious problems such as insufficient safety redundancy of support structure, excessive differential settlement of existing buildings, and even wall cracking and collapse during the assessment.

[0003] Numerous appraisal cases in recent years have shown that over 60% of disputes related to foundation pits in old urban areas stem from a lack of collaborative design between the two parties. Therefore, from the perspective of appraisal technical services, it is imperative to reverse the situation of reactive appraisal and handling. Based on this, it is necessary to independently develop an intelligent system that integrates deformation control and foundation reinforcement collaborative design. This system would feed back appraisal data and on-site monitoring information to the design end in real time, enabling joint simulation, conflict detection, and scheme optimization for dynamic control of foundation pit excavation and proactive reinforcement of existing building foundations. Summary of the Invention

[0004] To overcome the above shortcomings, this invention provides a dynamic correction method for geological models based on multi-source monitoring data inversion. It aims to improve the problems in traditional foundation pit design methods, such as the disconnect between foundation pit support and existing building foundation reinforcement, and the inability to effectively utilize monitoring data during construction to dynamically correct geological models, thereby realizing the transformation from static design to dynamic closed-loop control.

[0005] This invention provides the following technical solution: a method for dynamic correction of a geological model based on multi-source monitoring data inversion, comprising: Based on the foundation pit deformation data and existing building foundation data, an initial collaborative analysis model was established, and initial soil parameters were defined. Based on the initial collaborative analysis model, an intrinsic orthogonal decomposition is used to construct a set of predicted working conditions under different foundation pit support schemes and different foundation working conditions of existing buildings. In the set of predicted working conditions, the Pareto optimal solution set is solved with constraints such as the foundation pit deformation limit, the safety factor of the existing building foundation bearing capacity, differential settlement and tilt rate, and the comprehensive cost and construction period as objective functions, and a support-reinforcement scheme library is generated. During the construction of the foundation pit, the first type of monitoring data and the second type of monitoring data were collected, and the data were preprocessed to generate multi-source observation vectors; Based on the initial collaborative analysis model, a forward prediction operator is constructed. Taking the error between the multi-source observation vector and the predicted value as the target, the initial soil parameters are sequentially inverted and updated. The updated parameters are then substituted back into the model to recalculate the structural response of the remaining construction stage. A graded early warning threshold is set. When the predicted structural response exceeds any graded early warning threshold, an early warning signal is automatically triggered. Based on the recommended adjustment measures in the support-reinforcement scheme library, the recommended scheme is fed back to the on-site construction and evaluation platform.

[0006] Preferably, the steps for establishing an initial three-dimensional finite element collaborative analysis model include: Establish a semantic mapping rule base for heterogeneous data to convert foundation pit deformation data and existing building foundation data into a unified format; Boolean operations and coupling algorithms are used to geometrically fuse the geological model and the building foundation model; At the interface of the merged model, the soil element mesh nodes and the foundation component mesh nodes are set as shared nodes to coordinate the merged model; Kriging interpolation and random field simulation methods were used to spatially estimate and fill the stratigraphic parameters in the sparse area of ​​the fused model, and an initial three-dimensional finite element collaborative analysis model was established.

[0007] Preferably, the step of constructing the predicted working condition set includes: The intrinsic orthogonal decomposition method is used to collect snapshots of the response field of the initial collaborative analysis model under different operating parameters, and a snapshot matrix is ​​generated. Singular value decomposition is performed on the snapshot matrix to extract the main modes and construct a reduced-order model; Based on the reduced-order model, the structural response under different foundation pit support schemes and different foundation conditions of existing buildings is calculated in batches to form a set of predicted conditions.

[0008] Preferably, the step of generating a support-reinforcement scheme library includes: A multi-objective optimization model is constructed with constraints such as the excavation pit deformation limit, the safety factor of the existing building foundation bearing capacity, differential settlement and tilt rate, and the comprehensive cost and construction period as objective functions. A multi-objective genetic algorithm is used to iteratively solve the multi-objective optimization model based on the predicted working condition set to obtain the Pareto optimal solution set; The Pareto optimal solution set is stored as a support-reinforcement scheme library.

[0009] Preferably, the first type of monitoring data includes ground settlement around the pit, lateral displacement of the retaining wall, axial force of the support, and groundwater level, while the second type of monitoring data includes differential settlement of the existing building foundation, settlement rate, foundation inclination rate, and changes in wall crack width.

[0010] Preferably, the step of constructing a forward prediction operator based on the initial collaborative analysis model includes: Key soil parameters affecting structural response in the initial collaborative analysis model are extracted and used as input variables for the forward prediction operator. The response values ​​of the monitoring points corresponding to the multi-source observation vector in the initial collaborative analysis model are extracted and used as the output variables of the forward prediction operator. Establish a numerical mapping relationship between the input variables and the output variables to form a repeatedly callable calculation function.

[0011] Preferably, the step of sequentially inverting and updating the initial soil parameters includes: Based on the initial soil parameters, an initial soil parameter set containing multiple set members is constructed, and each set member is input into the forward prediction operator to calculate the corresponding set of structural response prediction values. Calculate the Kalman gain matrix based on the residual between the multi-source observation vector and the set of predicted structural responses; The initial soil parameter set is updated using the Kalman gain matrix to obtain the updated parameter set, and the statistical distribution of the updated parameter set is used as the parameter inversion result for the current construction stage.

[0012] Preferably, the step of recalculating the structural response for the remaining construction phases includes: Substitute the updated parameter distribution into the initial collaborative analysis model to update the corresponding soil parameters in the model; Finite element analysis was performed to obtain the predicted structural response values ​​for each monitoring point during the remaining construction phase.

[0013] Preferably, the graded early warning thresholds include a yellow warning threshold, an orange warning threshold, and a red warning threshold, corresponding to 70%, 85%, and 100% of the design limit, respectively; the adjustment measures include at least one of increasing the number of support channels, adjusting the excavation sequence, and reinforcing the existing building foundation with grouting.

[0014] The present invention has the following beneficial effects: 1. In this invention, by establishing a heterogeneous data semantic mapping and collaborative analysis model, the barriers between engineering survey, appraisal, design and monitoring data are broken down, realizing integrated collaborative modeling of foundation pit support and existing building foundation reinforcement, fundamentally changing the situation where the two are separated in traditional design.

[0015] 2. In this invention, the intrinsic orthogonal decomposition reduction technique and multi-objective genetic algorithm are used to efficiently generate a Pareto optimal solution library for hundreds of support-reinforcement combination conditions. Under the premise of ensuring safety constraints, the overall cost and construction period are significantly reduced, providing a scientific decision-making tool for engineering design.

[0016] 3. In this invention, by utilizing the foundation pit monitoring data and the dynamic response data of existing building foundations, the soil parameters are updated in real time through ensemble Kalman filtering, and the structural response of the remaining construction stages is re-predicted based on the modified model. This achieves a closed-loop dynamic control of design-construction-monitoring-correction-redesign, which greatly improves the safety and self-adaptive capability of foundation pit construction in old urban areas. Attached Figure Description

[0017] Figure 1 This is a flowchart of a dynamic correction method for geological models based on multi-source monitoring data inversion proposed in this invention; Figure 2 This is a flowchart illustrating the process of generating a support-reinforcement scheme library proposed in this invention; Figure 3 This is a flowchart of the soil parameter inversion process proposed in this invention. Detailed Implementation

[0018] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1 In a first embodiment of the present invention, the present invention provides a method for dynamic correction of geological models based on multi-source monitoring data inversion, such as... Figure 1 As shown, it includes: S1. Based on the foundation pit deformation data and existing building foundation data, establish an initial collaborative analysis model and define initial soil parameters; Preferably, the steps for establishing an initial three-dimensional finite element collaborative analysis model include: Establish a semantic mapping rule base for heterogeneous data to convert foundation pit deformation data and existing building foundation data into a unified format; Boolean operations and coupling algorithms are used to geometrically fuse the geological model and the building foundation model; At the interface of the merged model, the soil element mesh nodes and the foundation component mesh nodes are set as shared nodes to coordinate the merged model; Kriging interpolation and random field simulation methods were used to spatially estimate and fill the stratigraphic parameters in the sparse area of ​​the fused model, and an initial three-dimensional finite element collaborative analysis model was established.

[0020] Specifically, the data on foundation pit deformation comes from field drilling, in-situ testing, and indoor geotechnical tests, while the data on existing building assessment comes from on-site inspections, historical drawings, and settlement observation records. These data differ in coordinate system, format, and semantic expression.

[0021] To ensure data format consistency, a heterogeneous data semantic mapping rule base needs to be established. This rule base uses predefined mapping tables to map stratum names in borehole columnar sections to unified soil type codes, and foundation types in assessment reports to structural element types identifiable by finite element analysis. Each mapping rule includes source data fields, target data fields, and transformation functions. These transformation functions include coordinate transformation, unit conversion, and classification code mapping. After semantic mapping, all foundation pit deformation data and existing building assessment data are converted into a unified coordinate system, data format, and semantic labels.

[0022] Geological models are typically represented as a three-dimensional continuous medium, using irregular triangular meshes to describe stratigraphic interfaces and the distribution of soil and rock masses. Building foundation models are represented as discrete structural components, including isolated foundations, strip foundations, pile foundations, or raft foundations. To achieve the fusion of these two models, Boolean operations are used for geometric fusion. First, the closed surfaces of the building foundation model are extracted. Then, a difference operation is performed within the solid space of the geological model, removing the space occupied by the foundation from the geological mass to form a geological mass containing the foundation cavity, while preserving the independent representation of the foundation shape. After the Boolean operation, the geological mesh and the foundation mesh are paired at the contact interface to form a unified coupling interface. At this point, the fused model can establish the physical field transfer relationship between the two models.

[0023] At the interface of the fusion model, the displacement continuity between soil elements and foundation components is crucial for simulating soil-structure interactions. To ensure continuous displacement transfer, soil element mesh nodes and foundation component mesh nodes are set as shared nodes. Specifically, the foundation components are first meshed to generate foundation surface nodes; then, these surface nodes are used as constraint points to mesh the geological model, forcing the geological mesh to pass through existing nodes. Shared nodes automatically ensure continuous displacement between the soil and foundation at the interface, eliminating the need for additional contact condition definitions.

[0024] Due to the limited number of boreholes in engineering exploration, stratigraphic parameters exhibit significant spatial variability in sparsely drilled areas. To reasonably estimate stratigraphic parameters in data-sparse regions, kriging interpolation is employed for spatial estimation. Kriging estimates are presented below. Known points near the point to be estimated Corresponding parameter value Weighted average: ; in Here are the Kriging weights. Kriging interpolation alone cannot reflect the spatial variability of the parameters. To quantify this variability, a random field simulation method is introduced. This is implemented using a Karhunen-Loève expansion to generate a random field: ; In the formula The trend term obtained by Kriging estimation, and These are the eigenvalues ​​and eigenfunctions of the covariance matrix, respectively. The variables are independent standard normal random variables. Multiple parameter fields with equal probability are generated by truncating and retaining the first few main characteristic modes. By superimposing the mean estimated by Kriging with the fluctuation term of the random field simulation, a complete three-dimensional distribution of parameters such as elastic modulus, Poisson's ratio, internal friction angle, cohesion, and natural unit weight of each soil layer can be obtained in the data sparse region.

[0025] After the parameter field is filled, each element or geological region in the model has been assigned corresponding material parameters. To facilitate subsequent inversion updates, these parameters are summarized and defined as the initial soil parameter vector. This vector is a high-dimensional vector, and its components include the elastic modulus, Poisson's ratio, internal friction angle, cohesion, etc., of each stratigraphic unit. This initial soil parameter vector It is stored in the collaborative analysis model as the baseline state for inversion and updating in subsequent steps.

[0026] S2. Based on the initial collaborative analysis model, the intrinsic orthogonal decomposition is used to construct a set of predicted working conditions under different foundation pit support schemes and different foundation working conditions of existing buildings. Preferably, the step of constructing the predicted working condition set includes: The intrinsic orthogonal decomposition method is used to collect snapshots of the response field of the initial collaborative analysis model under different operating parameters, and a snapshot matrix is ​​generated. Singular value decomposition is performed on the snapshot matrix to extract the main modes and construct a reduced-order model; Based on the reduced-order model, the structural response under different foundation pit support schemes and different foundation conditions of existing buildings is calculated in batches to form a set of predicted conditions.

[0027] Specifically, after establishing the initial collaborative analysis model, it is necessary to quickly predict the structural response under different foundation pit support schemes and different foundation conditions of existing buildings, so as to provide sufficient sample data for subsequent multi-objective optimization. To this end, the intrinsic orthogonal decomposition method is used to construct a reduced-order model to achieve efficient calculation of batch conditions.

[0028] The intrinsic orthogonal decomposition method first requires obtaining the response fields of the original finite element model under several typical working conditions; these response fields are called snapshots. Assume the parameters of the foundation pit support scheme to be investigated include pile diameter, pile length, number of supports, etc., and the working condition parameters of the existing building foundation include natural foundation, pile foundation, composite foundation, etc. These parameters are then combined to generate... A typical working condition is defined. For each typical working condition, a complete finite element calculation is performed using the initial collaborative analysis model to extract key response fields from the model, such as the ground settlement field around the pit, the lateral displacement field of the retaining wall, and the settlement field of the existing building foundation. Each response field can be represented as a high-dimensional vector. ,in This represents the total number of degrees of freedom in the model. Arrange all snapshots by column to form a snapshot matrix. Each column of this matrix corresponds to the complete response field under a certain working condition.

[0029] To extract the main feature modes from the snapshot matrix, singular value decomposition is performed on the snapshot matrix: ; in Let be a left singular vector matrix, and its column vectors be... This refers to the intrinsic orthogonal decomposition mode; For a diagonal matrix, the elements on the diagonal are... These are singular values, reflecting the energy contribution of each mode; Let be the right singular vector matrix. The magnitude of the singular values ​​determines the importance of the corresponding mode. To construct a reduced-order model, retain the first... The main modes are required to ensure that the cumulative energy percentage reaches a preset threshold, typically requiring: ; The truncated left singular vector matrix is ​​denoted as Its column vectors span a low-dimensional subspace.

[0030] In a low-dimensional subspace, the response field under any operating condition can be approximated as a linear combination of truncated modes: ; in For the operating condition parameter vector, These are the corresponding modal coefficients. The governing equations of the original finite element model... By projecting the response field onto a lower-dimensional subspace, a reduced-order system can be obtained.

[0031] Once the reduced-order model is constructed, it can be used to quickly calculate the structural response under a large number of different load combinations. First, generate the parameter set for all support schemes and foundation load combinations that need to be examined. ,in The total number of operating conditions can reach hundreds. For each operating condition parameter... The reduced-order model is called to calculate the corresponding structural response prediction value. This forms a set of predicted operating conditions: ; Each prediction result includes key response indicators such as the maximum surface settlement around the pit, the lateral displacement distribution of the retaining wall, the axial force of the support, and the differential settlement and tilt rate of the existing buildings under that working condition.

[0032] S3. In the set of predicted working conditions, with the constraints of the pit deformation limit, the safety factor of the bearing capacity of the existing building foundation, the differential settlement and the tilt rate, and with the comprehensive cost and construction period as the objective functions, solve the Pareto optimal solution set and generate a support-reinforcement scheme library. Preferably, the steps for generating the support-reinforcement scheme library are as follows: Figure 2 As shown, it includes: A multi-objective optimization model is constructed with constraints such as the excavation pit deformation limit, the safety factor of the existing building foundation bearing capacity, differential settlement and tilt rate, and the comprehensive cost and construction period as objective functions. A multi-objective genetic algorithm is used to iteratively solve the multi-objective optimization model based on the predicted working condition set to obtain the Pareto optimal solution set; The Pareto optimal solution set is stored as a support-reinforcement scheme library.

[0033] Specifically, after obtaining the set of predicted working conditions, it is necessary to select the support-reinforcement combination scheme with the optimal overall cost and construction period. To this end, a multi-objective genetic algorithm is adopted, with the pit deformation limit, the safety factor of the existing building foundation bearing capacity, differential settlement and tilt rate as constraints, and the overall cost and construction period as objective functions, to search for the Pareto optimal solution set.

[0034] First, define the decision variables, objective function, and constraints for the optimization problem. The decision variables include parameters for the foundation pit support scheme and parameters for the existing building foundation reinforcement scheme. Support scheme parameters include the diameter, length, spacing, number of supports, and cross-sectional dimensions of the retaining piles; reinforcement scheme parameters include the grouting depth, grouting range, anchor length, and spacing of the root piles. Let the decision variable vector be denoted as... ,in The number of variables.

[0035] There are two objective functions: overall cost and construction period. Overall cost... This includes support costs, reinforcement costs, and anticipated costs of risk management: ; in For the cost of support structure materials and construction, Direct costs for foundation reinforcement The latter is the cost of handling risks that may arise from exceeding the deformation limit (such as compensation to residents, structural repair, etc.), and its value can be estimated based on the statistical patterns of historical appraisal cases.

[0036] Construction period This includes the time for support construction, reinforcement construction, and mutual waiting time, all measured in days.

[0037] The constraints are divided into four categories. The first category is the deformation limit of the foundation pit, which requires the maximum settlement of the ground surface around the pit. Not exceeding the design allowable value Maximum lateral displacement of the retaining wall Not exceeding the allowable value The second category is the safety factor for the bearing capacity of existing building foundations, which requires... ,in For the load-bearing capacity safety factor, The values ​​are those required by regulations. The third category is differential settlement of existing buildings. Not exceeding the limit The fourth category is the tilt rate of existing buildings. Not exceeding the limit .

[0038] The above constraints can be uniformly expressed as inequality constraints. .

[0039] During the iterative process of a genetic algorithm, it is necessary to evaluate the objective function value and constraint satisfaction for each individual (i.e., the set of decision variable values). For any decision variable... First, based on the values, the corresponding combination of support and reinforcement schemes is determined. Then, several sample points closest to this combination are retrieved from the predicted working condition set, and the structural response under the current decision variables is calculated using interpolation methods, including... , , , Then calculate the overall cost. and construction period .

[0040] If an individual violates any constraint, a penalty function is applied during the optimization process to reduce its fitness.

[0041] A multi-objective genetic algorithm, such as a non-dominated sorting genetic algorithm, is used to iteratively solve the above multi-objective optimization problem. After the algorithm completes its iterations, all individuals on the first non-dominated frontier are obtained, which constitutes the Pareto optimal solution set. Each solution in this set represents a support-reinforcement combination scheme, characterized by the fact that no other scheme can simultaneously outperform it in terms of both overall cost and construction period objectives.

[0042] Each solution in the Pareto optimal solution set, along with its corresponding objective function value and constraint satisfaction, is stored as a support-reinforcement solution library. This library allows engineers to select appropriate support-reinforcement combinations based on specific project risk preferences and budgets. It can also be used by subsequent early warning steps to quickly recommend adjustment measures when an early warning is triggered.

[0043] S4. During the construction of the foundation pit, collect the first type of monitoring data and the second type of monitoring data, and perform data preprocessing to generate multi-source observation vectors; Preferably, the first type of monitoring data includes ground settlement around the pit, lateral displacement of the retaining wall, axial force of the support, and groundwater level, while the second type of monitoring data includes differential settlement of the existing building foundation, settlement rate, foundation inclination rate, and changes in wall crack width.

[0044] Specifically, the first category of monitoring data consists of routine monitoring items during the foundation pit construction process, including ground settlement around the pit, lateral displacement of the retaining wall, axial force of the supports, and groundwater level. Ground settlement around the pit is monitored periodically using a level or hydrostatic level by setting up settlement monitoring points around the foundation pit. Lateral displacement of the retaining wall is measured segment by segment along the depth direction using an inclinometer embedded in the wall. Axial force of the supports is measured by embedding axial force gauges or steel stress gauges at the ends of the steel supports or inside the concrete supports. Groundwater level is monitored in real time by setting up water level observation holes around the foundation pit, each equipped with a water level gauge.

[0045] The second category of monitoring data consists of dynamic response data of existing building foundations, including differential settlement, settlement rate, foundation inclination, and changes in wall crack width. Differential settlement is measured by placing settlement monitoring points at key locations in the existing building foundation (such as corners and both sides of settlement joints), using a level or hydrostatic level to measure the absolute settlement at each point, and then calculating the differential settlement value between any two points. Settlement rate is the amount of settlement change per unit time, calculated from continuous observation data. Foundation inclination is obtained by measuring the ratio of the elevation difference between two points on the same axis of the foundation to the distance between the two points. Changes in wall crack width are monitored in real time by installing crack gauges or vibrating wire displacement gauges at existing cracks in the existing building walls.

[0046] After all monitoring data is collected, data preprocessing is required. Preprocessing includes three stages: data cleaning, normalization, and time synchronization. Data cleaning removes outliers caused by sensor malfunctions, signal interference, or human error in readings; outlier detection methods based on the median can be used. Normalization transforms monitoring data with different dimensions to the same order of magnitude; Z-score normalization can be used to eliminate the impact of dimensional differences on the inversion algorithm. Time synchronization ensures that monitoring data from different sensors and different acquisition frequencies are aligned on the time axis; linear interpolation is typically used to interpolate each monitoring data point to a unified time node.

[0047] S5. Based on the initial collaborative analysis model, construct a forward prediction operator, take the error between the multi-source observation vector and the predicted value as the target, perform sequential inversion update of the initial soil parameters, and substitute the updated parameters back into the model to recalculate the structural response of the remaining construction stage. Preferably, the step of constructing a forward prediction operator based on the initial collaborative analysis model includes: Key soil parameters affecting structural response in the initial collaborative analysis model are extracted and used as input variables for the forward prediction operator. The response values ​​of the monitoring points corresponding to the multi-source observation vector in the initial collaborative analysis model are extracted and used as the output variables of the forward prediction operator. Establish a numerical mapping relationship between the input variables and the output variables to form a repeatedly callable calculation function.

[0048] Specifically, such as Figure 3 As shown, before the sequential inversion update, a reusable forward prediction operator needs to be constructed to quickly calculate the structural response under given soil parameters. First, key soil parameters affecting the structural response in the initial co-analysis model are extracted as input variables for the forward prediction operator. These key soil parameters include the elastic modulus, Poisson's ratio, internal friction angle, cohesion, and foundation contact stiffness of each soil layer. These parameters directly control the mechanical behavior of the finite element model and are the main objects of the inversion update. The input variables are denoted as vectors. ,in This represents the total number of key parameters.

[0049] Secondly, the response quantities of the monitoring points corresponding to the multi-source observation vectors in the initial collaborative analysis model are extracted as output variables of the forward prediction operator. These response quantities include the surface settlement value around the pit, the lateral displacement values ​​of the retaining wall at various depths, the axial force value of the support, the differential settlement value of the existing building, the foundation inclination rate, and the change value of the wall crack width. The output variables are denoted as vectors. ,in The total number of responses at monitoring points should be consistent with the dimension of the multi-source observation vector.

[0050] Then, a numerical mapping relationship is established between input and output variables, forming a repeatedly invoked calculation function. This function encapsulates the complete finite element calculation process from given soil parameters to the response values ​​at monitoring points. In specific implementation, the finite element input file of the initial collaborative analysis model can be parameterized, with key soil parameters set as variable placeholders. A script program receives the input variables, automatically modifies the corresponding parameter values ​​in the input file, calls the finite element solver to perform the calculation, and extracts the response values ​​at the specified monitoring points as output from the calculation result file. This script program is encapsulated as a function, denoted as... .

[0051] Preferably, the step of sequentially inverting and updating the initial soil parameters includes: Based on the initial soil parameters, an initial soil parameter set containing multiple set members is constructed, and each set member is input into the forward prediction operator to calculate the corresponding set of structural response prediction values. Calculate the Kalman gain matrix based on the residual between the multi-source observation vector and the set of predicted structural responses; The initial soil parameter set is updated using the Kalman gain matrix to obtain the updated parameter set, and the statistical distribution of the updated parameter set is used as the parameter inversion result for the current construction stage.

[0052] After obtaining the multi-source observation vectors, the initial soil parameters are sequentially inverted and updated using the ensemble Kalman filter algorithm. This includes the following steps.

[0053] First, an initial soil parameter set containing multiple set members is constructed based on the initial soil parameters. The initial soil parameter vector is denoted as... To characterize the uncertainty of the parameters, in Add perturbations to generate There are set members, each member is denoted as . , The perturbation is generated based on the prior covariance matrix of the parameters, typically assuming that the parameters follow a log-normal or normal distribution. This yields the initial parameter set. .

[0054] Then, each set member is input into the forward prediction operator to calculate the corresponding set of structural response predictions. For each set member... Call the forward prediction operator The predicted response vector is calculated. The predicted responses of all set members constitute the predicted response set. Calculate the mean of the predicted response. .

[0055] Next, the Kalman gain matrix is ​​calculated based on the residuals between the multi-source observation vectors and the set of predicted structural responses. The multi-source observation vectors are denoted as... Its dimensions and Same. The observation residual is Kalman gain matrix The calculation formula is: ; in The cross-covariance matrix is ​​the sum of the parameter vector and the predicted response vector. To predict the autocovariance matrix of the response vector, The noise covariance matrix is ​​used for observation. These covariance matrices are calculated from the statistical information of the set members. Then, the initial soil parameter set is updated using the Kalman gain matrix to obtain the updated parameter set. The update formula for each set member is: ; in The artificially added observational perturbation term follows a pattern with a mean of zero and a covariance of 0. The parameter set follows a normal distribution to ensure statistical diversity after the update. The updated set yields a new set of parameters. .

[0056] Finally, the statistical distribution of the updated parameter set is used as the parameter inversion result for the current construction stage. Typically, the mean of the updated parameter set is taken as the best estimate of the parameters. Simultaneously, the covariance matrix of the parameters is recorded to characterize the uncertainty. This inversion result reflects the latest understanding of soil parameters at the current construction stage, integrating prior information with real-time monitoring data.

[0057] Preferably, the step of recalculating the structural response for the remaining construction phases includes: Substitute the updated parameter distribution into the initial collaborative analysis model to update the corresponding soil parameters in the model; Finite element analysis was performed to obtain the predicted structural response values ​​for each monitoring point during the remaining construction phase.

[0058] Specifically, after completing the parameter inversion update, the updated parameters need to be substituted back into the model to recalculate the structural response in the subsequent construction stage for graded early warning judgment.

[0059] First, substitute the updated parameter distribution into the initial co-analysis model to update the corresponding soil parameters in the model. Then, take the mean of the updated parameter set. The parameters such as elastic modulus, Poisson's ratio, internal friction angle, and cohesion are assigned to the corresponding material property fields in the initial co-analysis model. For different stratigraphic regions in the model, updates are performed separately based on the index of the parameter vector.

[0060] Then, finite element analysis (FEM) calculations are performed to obtain the predicted structural response values ​​for each monitoring point in the remaining construction phase. Specifically, using the excavation depth of the foundation pit at the end of the current construction phase and the reinforcement status of the existing building as initial conditions, subsequent construction steps (such as excavation of the next floor or application of the next support) are activated, and the finite element solver is called for step-by-step calculations. After the calculations are completed, the predicted response values ​​for each monitoring point corresponding to the multi-source observation vectors are extracted, including ground settlement around the pit, lateral displacement of the retaining wall, axial force of the supports, differential settlement of the existing building, and tilt rate.

[0061] S6. Set graded early warning thresholds. When the predicted structural response exceeds any graded early warning threshold, an early warning signal is automatically triggered. Based on the recommended adjustment measures in the support-reinforcement scheme library, the recommended scheme is fed back to the on-site construction and evaluation platform.

[0062] Preferably, the graded early warning thresholds include a yellow warning threshold, an orange warning threshold, and a red warning threshold, corresponding to 70%, 85%, and 100% of the design limit, respectively; the adjustment measures include at least one of increasing the number of support channels, adjusting the excavation sequence, and reinforcing the existing building foundation with grouting.

[0063] Specifically, this plan sets three levels of warning thresholds. The yellow warning threshold corresponds to 70% of the design limit, the orange warning threshold corresponds to 85% of the design limit, and the red warning threshold corresponds to 100% of the design limit.

[0064] The predicted structural response values ​​for the remaining construction phases are compared with threshold values ​​at each level. A yellow alert is triggered when the predicted value exceeds the yellow threshold but does not reach the orange threshold, prompting enhanced monitoring. An orange alert is triggered when the predicted value exceeds the orange threshold but does not reach the red threshold, preparing to initiate adjustment measures. A red alert is triggered when the predicted value reaches or exceeds the red threshold, immediately initiating adjustment measures.

[0065] Adjustment measures include at least one of the following: increasing the number of support channels, adjusting the excavation sequence, or reinforcing the existing building foundation with additional grouting. The system retrieves a matching adjustment plan from the support-reinforcement scheme library based on the type of response exceeding limits, and feeds the recommended measures back to the on-site construction platform and the evaluation platform. After the construction team makes adjustments according to the recommended plan, it continues to collect subsequent monitoring data, re-executes the inversion update and remaining stage prediction, forming a closed-loop control system of design-construction-monitoring-correction-redesign.

[0066] Example 2 This embodiment uses a renovation project of an old residential community as an example to illustrate the specific implementation process of the method of the present invention in an actual project.

[0067] A residential community built in the 1980s plans to construct an underground parking garage. The excavation pit measures 50 meters by 40 meters and has a depth of 10 meters. The pit is located only 3.2 meters east of a six-story brick-concrete residential building. This building uses a strip foundation on natural ground with a foundation depth of 1.5 meters. An assessment report indicates that the building's walls have several micro-cracks approximately 0.2 millimeters wide, and due to disturbance from previous nearby construction, there has been a cumulative settlement of approximately 8 millimeters. The soil strata at the pit site, from top to bottom, are: a 2-meter-thick layer of miscellaneous fill, an 8-meter-thick layer of silty clay, a 12-meter-thick layer of silty clay, and a groundwater level at a depth of 1.5 meters. The proposed support scheme for the pit is bored piles with two layers of internal bracing. The piles have a diameter of 800 millimeters and a length of 18 meters.

[0068] Step 1: Establish the initial collaborative analysis model Collect engineering survey data for the project, including columnar sections of three boreholes, standard penetration test values, and geotechnical test results. Compile assessment data for the residential building, including foundation layout plan, superstructure load statistics, existing settlement observation records, and crack distribution maps.

[0069] A heterogeneous data semantic mapping rule base was established, unifying the encoding of borehole strata names, identifying the foundation type as a strip foundation, and recording crack width as the initial damage state. The strip foundation model of the residential building was embedded into the 3D geological model using Boolean operations, with shared nodes set at the interface between the foundation and the soil. Using the Kriging interpolation method, with the soil parameters of the three boreholes as known points, the elastic modulus, Poisson's ratio, internal friction angle, and cohesion of each soil layer throughout the site were estimated. In sparse borehole regions, Karhunen-Loève expansion was used to generate random fields to reflect the spatial variability of the parameters. The above parameters were summarized and defined as an initial soil parameter vector, which included elastic moduli of 4.2 MPa for the fill layer and 2.8 MPa for the silty clay layer.

[0070] Step 2: Construct a set of predicted working conditions and generate a solution library Support scheme variables: Pile diameter can be selected as 700 mm, 800 mm, or 900 mm; pile length can be selected as 16 m, 18 m, or 20 m; number of supports can be selected as one or two. Reinforcement scheme variables: For residential building foundations, options include unreinforced, grouting reinforced (depth 2 m or 3 m), and root pile reinforced (spacing 1.5 m or 2.0 m). All variable combinations result in 162 possible working conditions.

[0071] For each working condition, the initial collaborative analysis model was invoked for finite element analysis. Response fields such as ground settlement around the pit, wall lateral displacement, and differential foundation settlement were extracted as snapshots to construct a snapshot matrix. Singular value decomposition was performed on the snapshot matrix, retaining the top 20 principal modes with a cumulative energy share of 99%, and a reduced-order model was constructed.

[0072] With constraints including the excavation pit deformation limit (maximum surface settlement ≤ 30 mm, maximum lateral displacement of the wall ≤ 0.2%H, i.e., 20 mm), the safety factor of the existing building foundation bearing capacity ≥ 2.5, differential settlement ≤ 10 mm, and tilt rate ≤ 0.003, and with the comprehensive cost (support cost + reinforcement cost + expected risk management cost) and construction period as the objective functions, the NSGA-II algorithm was used to perform multi-objective optimization on the above 162 working conditions. The population size was set to 100, and the number of generations was set to 200. After the algorithm converged, a Pareto optimal solution set was obtained, containing 12 non-dominated solutions. Each solution and its corresponding objective function value were stored as a support-reinforcement solution library. Among them, the optimal compromise solution is: 800 mm diameter cast-in-place piles, 18 m long piles, two supports, combined with grouting reinforcement of the existing building foundation to a depth of 2.5 m, with a comprehensive cost of 1.85 million yuan and a construction period of 45 days.

[0073] Step 3: Construction Period Monitoring Data Collection and Inversion After the foundation pit construction begins, monitoring points will be set up according to the following plan: one surface settlement point every 20 meters around the pit, for a total of 12 points; one settlement point at each of the four corners and the midpoint of the long side of the residential building foundation, for a total of 6 points; three inclinometer tubes will be pre-embedded in the retaining wall; axial force gauges will be installed at the ends of the supports; and four crack gauges will be installed at the cracks in the existing building walls. All monitoring data will be uploaded in real time through an automated data acquisition system.

[0074] After the second layer of support was installed, multi-source observation vectors for the current stage were collected, including: the maximum settlement value of the ground around the pit was 18 mm, the maximum lateral displacement value of the wall was 12 mm, the differential settlement value of the residential building foundation was 5 mm, the settlement rate was 1.2 mm / day, and the increase in wall crack width was 0.05 mm.

[0075] A forward prediction operator was constructed based on the initial collaborative analysis model. An initial parameter set containing 50 members was generated, with each member having a normal perturbation conforming to the prior covariance added to the initial parameters. Each member was input into the forward prediction operator to calculate the predicted response set. Based on the residuals between the observed vector and the predicted response set, the Kalman gain matrix was calculated to update the parameter set. The statistical mean of the updated parameter set was used as the inversion result for the current construction stage. The inversion results showed that the elastic modulus of the silty clay layer was corrected from 2.8 MPa to 2.4 MPa, and the internal friction angle was corrected from 12 degrees to 10.5 degrees, indicating that the soil layer was actually softer than the exploration results suggested.

[0076] Step 4: Prediction and Early Warning of Remaining Construction Stages Substituting the updated parameters back into the initial collaborative analysis model, the structural response for the remaining construction phases (application of the third support and excavation to the bottom of the foundation pit) was recalculated. The prediction results show that the final settlement of the ground around the pit will reach 28 mm, close to the red warning threshold of 30 mm (100%); the final lateral displacement of the wall will reach 18 mm, which will not exceed the threshold; and the final differential settlement of the residential building foundation will reach 9 mm, exceeding the orange warning threshold (85% but not 100%).

[0077] The system automatically triggers an orange alert signal and pushes the alert information to the on-site construction platform and the assessment platform. At the same time, it searches the support-reinforcement scheme database for adjustment measures that are closest to the current scheme and can effectively control differential settlement. It recommends grouting reinforcement of the existing building foundation, increasing the grouting depth to 3.5 meters and controlling the grouting pressure at 0.5 MPa.

[0078] The construction team implemented supplementary grouting according to the recommended plan. After grouting was completed, subsequent monitoring data was collected, and the inversion update and prediction cycle was restarted. After two iterations, the final predicted settlement value stabilized at 26 mm, and the differential settlement stabilized at 7 mm, both within the warning threshold. The foundation pit was successfully completed, and no new structural damage occurred to the residential building.

[0079] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A dynamic correction method of a geological model based on inversion of multi-source monitoring data, characterized in that, include: Based on the foundation pit deformation data and existing building foundation data, an initial collaborative analysis model was established, and initial soil parameters were defined. Based on the initial collaborative analysis model, an intrinsic orthogonal decomposition is used to construct a set of predicted working conditions under different foundation pit support schemes and different foundation working conditions of existing buildings. In the set of predicted working conditions, the Pareto optimal solution set is solved with constraints such as the foundation pit deformation limit, the safety factor of the existing building foundation bearing capacity, differential settlement and tilt rate, and the comprehensive cost and construction period as objective functions, and a support-reinforcement scheme library is generated. During the construction of the foundation pit, the first type of monitoring data and the second type of monitoring data were collected, and the data were preprocessed to generate multi-source observation vectors; Based on the initial collaborative analysis model, a forward prediction operator is constructed. Taking the error between the multi-source observation vector and the predicted value as the target, the initial soil parameters are sequentially inverted and updated. The updated parameters are then substituted back into the model to recalculate the structural response of the remaining construction stage. A graded early warning threshold is set. When the predicted structural response exceeds any graded early warning threshold, an early warning signal is automatically triggered. Based on the recommended adjustment measures in the support-reinforcement scheme library, the recommended scheme is fed back to the on-site construction and evaluation platform.

2. The method according to claim 1, characterized in that, The steps to establish an initial three-dimensional finite element co-analysis model include: Establish a semantic mapping rule base for heterogeneous data to convert foundation pit deformation data and existing building foundation data into a unified format; Boolean operations and coupling algorithms are used to geometrically fuse the geological model and the building foundation model; At the interface of the merged model, the soil element mesh nodes and the foundation component mesh nodes are set as shared nodes to coordinate the merged model; Kriging interpolation and random field simulation methods were used to spatially estimate and fill the stratigraphic parameters in the sparse area of ​​the fused model, and an initial three-dimensional finite element collaborative analysis model was established.

3. The method of claim 1, wherein, The steps for constructing a set of predicted working conditions include: The intrinsic orthogonal decomposition method is used to collect snapshots of the response field of the initial collaborative analysis model under different operating parameters, and a snapshot matrix is ​​generated. Singular value decomposition is performed on the snapshot matrix to extract the main modes and construct a reduced-order model; Based on the reduced-order model, the structural response under different foundation pit support schemes and different foundation conditions of existing buildings is calculated in batches to form a set of predicted conditions.

4. The method of claim 1, wherein, The steps to generate a support-reinforcement scheme library include: A multi-objective optimization model is constructed with constraints such as the excavation pit deformation limit, the safety factor of the existing building foundation bearing capacity, differential settlement and tilt rate, and the comprehensive cost and construction period as objective functions. A multi-objective genetic algorithm is used to iteratively solve the multi-objective optimization model based on the predicted working condition set to obtain the Pareto optimal solution set; The Pareto optimal solution set is stored as a support-reinforcement scheme library.

5. The method of claim 1, wherein, The first type of monitoring data includes ground settlement around the pit, lateral displacement of the retaining wall, axial force of the support, and groundwater level. The second type of monitoring data includes differential settlement of existing building foundations, settlement rate, foundation inclination rate, and changes in wall crack width.

6. The method for dynamic correction of a geological model based on multi-source monitoring data inversion according to claim 1, characterized in that, The steps for constructing the forward prediction operator based on the initial collaborative analysis model include: Key soil parameters affecting structural response in the initial collaborative analysis model are extracted and used as input variables for the forward prediction operator. The response values ​​of the monitoring points corresponding to the multi-source observation vector in the initial collaborative analysis model are extracted and used as the output variables of the forward prediction operator. Establish a numerical mapping relationship between the input variables and the output variables to form a repeatedly callable calculation function.

7. The method for dynamic correction of a geological model based on multi-source monitoring data inversion according to claim 6, characterized in that, The steps for sequential inversion updating of initial soil parameters include: Based on the initial soil parameters, an initial soil parameter set containing multiple set members is constructed, and each set member is input into the forward prediction operator to calculate the corresponding set of structural response prediction values. Calculate the Kalman gain matrix based on the residual between the multi-source observation vector and the set of predicted structural responses; The initial soil parameter set is updated using the Kalman gain matrix to obtain the updated parameter set, and the statistical distribution of the updated parameter set is used as the parameter inversion result for the current construction stage.

8. The method for dynamic correction of a geological model based on multi-source monitoring data inversion according to claim 7, characterized in that, The steps for recalculating the structural response for the remaining construction phases include: Substitute the updated parameter distribution into the initial collaborative analysis model to update the corresponding soil parameters in the model; Finite element analysis was performed to obtain the predicted structural response values ​​for each monitoring point during the remaining construction phase.

9. The method for dynamic correction of a geological model based on multi-source monitoring data inversion according to claim 1, characterized in that, The graded early warning thresholds include a yellow warning threshold, an orange warning threshold, and a red warning threshold, which correspond to 70%, 85%, and 100% of the design limit, respectively; the adjustment measures include at least one of increasing the number of support channels, adjusting the excavation sequence, and reinforcing the foundation of existing buildings with grouting.

Images