A method for group control of deformation of support structure for ultra-large foundation pit

Abstract

This invention discloses a method for group control of deformation of support structures in ultra-large foundation pits, belonging to the field of foundation pit support regulation technology. The method includes: deploying a distributed sensor network to collect deformation data of the retaining structure in real time; a data processing platform preprocessing and fusing multi-source data to construct a holographic state vector; using a digital twin model corrected by on-site data for inversion analysis to obtain the bending moment, shear force distribution, and theoretical target axial force of each support; using the minimum overall deformation of the foundation pit, the most uniform distribution of support axial force, and no exceedance of bearing capacity as multiple objectives, and the servo axial force adjustment amount of each support as the decision variable, employing multi-objective particle swarm optimization to solve for the optimal combination of adjustment amounts; using a distributed controller to drive the servo jacks at the ends of each support in parallel to perform axial force adjustment; and providing real-time feedback and triggering local closed-loop correction during the adjustment process. This invention achieves autonomous, dynamic, and real-time group control of the deformation of multiple support structures throughout the entire construction process of the foundation pit.

Description

Technical Field

[0001] This invention belongs to the field of foundation pit support control technology, specifically relating to a method for group control of deformation of ultra-large foundation pit support structures. Background Technology

[0002] As urban underground space development deepens, the number of ultra-large and ultra-deep foundation pit projects is increasing, often located adjacent to sensitive environments such as operating subways, important pipelines, and existing buildings. Under these engineering conditions, the external loads (including earth pressure and water pressure) borne by the foundation pit retaining structure and its internal support structure change dynamically throughout the entire foundation pit construction process, as earthwork excavation and dewatering activities progress. Precisely controlling the deformation of the foundation pit retaining structure is crucial to ensuring the safety of the foundation pit itself and the normal operation of surrounding sensitive facilities.

[0003] Existing technologies include design methods for controlling deep foundation pit deformation using a steel support axial force servo system. These methods obtain lateral deformation control values ​​for the retaining structure and axial force control thresholds for the steel supports through continuous medium finite element analysis. Actual deformation and axial force values ​​are then dynamically monitored using detection elements, and the support force is adjusted by an automatic control system. Other research has proposed precise control methods for micro-deformation in deep foundation pits, using active compensation of the servo support force from an active servo control device and passive compensation from grouting reinforcement of the soil within the pit to collaboratively control the foundation pit deformation. Furthermore, some studies have attempted to optimize the support axial force using optimization methods such as particle swarm optimization, while others have proposed multi-stage, multi-objective servo support optimization control methods based on the TabPFN model and a support axial force servo system.

[0004] However, existing technologies still have the following shortcomings: First, in terms of the coordinated control of multiple servo supports above and below a single support structure, existing technologies lack effective multi-support coordinated control strategies. Multi-support structures are simultaneously stressed at different depths in the foundation pit, and the axial force distribution among the supports is coupled. Adjusting the axial force of one support will cause a redistribution of the axial force in other supports, thus affecting the overall deformation of the retaining structure. Most existing methods treat the axial force adjustment of each support independently, without fully considering the coordinated stress and deformation relationship between supports. Although some studies have proposed a "dual control method"—a dual control method of axial force control and displacement control—this method is still a staged control mode and lacks a real-time coordinated mechanism among multiple supports.

[0005] Secondly, in the planar direction of the foundation pit, there is also a lack of systematic theoretical methods for the coordinated stress and adjustment among multiple support structures. The application scenarios of existing servo steel supports are limited to small pits with a width of no more than 20m, which cannot meet the increasingly complex requirements for coordinated deformation control of multiple supports in ultra-large and deep foundation pits. With the dynamic changes in construction conditions (excavation, dewatering, etc.), how to achieve coordinated stress and overall adjustment among multiple supports within the foundation pit, ensuring that both the overall and local deformations of the foundation pit are within ideal ranges, is a pressing technical problem that needs to be solved in the field of foundation pit engineering.

[0006] Third, existing methods for controlling foundation pit deformation have low levels of automation, with monitoring and early warning typically relying on manual labor, resulting in significant errors and untimely data processing. Traditional supports are passive load-bearing structures, and concrete supports are subject to relaxation effects caused by shrinkage and creep, leading to irreversible compression of 15-30 mm. Although some studies have applied digital twin technology to foundation pit deformation control, existing digital twin models are mostly open-loop prediction modes, failing to form a closed-loop linkage with the multi-objective real-time optimization of support axial forces.

[0007] Fourth, existing methods for controlling foundation pit deformation often employ simplistic objective functions for multi-objective optimization, primarily focusing on the maximum displacement of the retaining structure as the sole control objective. They fail to comprehensively consider multi-dimensional control objectives such as the overall deformation distribution of the foundation pit, the uniformity of axial force distribution across supports, and the safety margin of support bearing capacity. Furthermore, existing methods do not adequately account for the coupled impact of pit bottom rebound on the displacement of the retaining structure during excavation, and lack mechanisms to directly integrate deformation constraints of sensitive protected objects such as adjacent subway tunnels into the optimization algorithm's search process.

[0008] In summary, there is an urgent need for an autonomous dynamic group control method that can achieve coordinated stress and deformation of multiple support structures throughout the entire construction process of the foundation pit. Summary of the Invention

[0009] In view of this, the purpose of this invention is to provide a method for controlling the deformation of support structures in ultra-large foundation pits, which solves the technical problem in the prior art that multiple support structures cannot autonomously and dynamically coordinate the deformation during the entire construction process of ultra-large foundation pits. It overcomes the shortcomings of traditional methods, such as reliance on manual experience, phased open-loop adjustment, lagging control, and uneven stress among supports. It achieves real-time dynamic and precise group control of the deformation of the foundation pit retaining structure, while taking into account the rebound coupling effect of the pit bottom and the protection constraints of the adjacent sensitive environment.

[0010] To achieve the above objectives, the present invention provides the following technical solution: This invention discloses a method for controlling the deformation of a support structure for ultra-large foundation pits, comprising the following steps: S1. Deploy a distributed sensor network to collect real-time data on the displacement of the retaining structure, the axial force of the support, the water level, the earth pressure, and the deformation of the surrounding area, and transmit the data to the data processing platform. S2. The data processing platform preprocesses and fuses multi-source data to construct a holographic state vector that includes the displacement distribution field of the retaining structure, the axial force distribution matrix of the support, the spatiotemporal field of the water level, and the earth pressure vector. S3. Based on the holographic state vector, an inversion analysis is performed using a digital twin model corrected by field data to obtain the bending moment, shear force distribution of the retaining structure and the theoretical target axial force of each support. The digital twin model adopts an elastic foundation beam-finite element coupled model, uses the "m" method nonlinear soil spring to simulate the soil inside the pit, and uses active earth pressure considering spatiotemporal effects to simulate the load outside the pit. S4. Taking the minimum overall deformation of the foundation pit, the most uniform distribution of the axial force of the support, and the bearing capacity not exceeding the limit as multiple objectives, the adjustment amount of the servo axial force of each support is used as the decision variable, and the displacement threshold, the upper and lower limits of the bearing capacity, and the limit of the difference in the axial force of the support are used as constraints. The optimal combination of adjustment amounts is solved by multi-objective particle swarm optimization. S5. Generate instructions based on the optimal adjustment amount, and drive the servo jacks at each support end to perform axial force adjustment in parallel through the distributed controller; S6. During the adjustment process, the actual axial force and displacement response are fed back in real time. If the deviation from the theoretical expectation exceeds the threshold, a local closed-loop correction is triggered.

[0011] Furthermore, the construction and modification of the digital twin model includes: An initial finite element model is established before construction. After the first support is installed and before the first layer is excavated, the model parameters are calibrated using initial monitoring data. After each construction condition is completed, the key parameters of the model are dynamically updated using Kalman filtering or ensemble Kalman filtering, with the measured displacement field, axial force matrix, and earth pressure vector as constraints, so as to minimize the root mean square error between the model output and the measured data. The updated model is used to predict the displacement and axial force of the next condition, which serves as the initial reference for multi-objective optimization.

[0012] Furthermore, the specific implementation of the target particle swarm optimization includes: Initialize the particle swarm and particle positions. Represent the axial force adjustment for each support; construct a three-objective fitness function: In the formula, The depth of the foundation pit. For the depth of the foundation pit The weighting coefficient at the location, For the current adjustment combination, the enclosure structure at depth The predicted displacement at that location, For the target displacement, For the first The path supports the current axial force. To adjust the average axial force, For the first The upper limit of the bearing capacity of the road support; Set constraints: , , , in, For depth Allowable displacement value, For the first Lower limit of road support bearing capacity This represents the upper limit of the allowable difference in axial force between any two supports; The solution is obtained by iteratively solving a non-dominated sorting particle swarm with elite retention. The individual optimal and global optimal are updated with non-dominated level and crowding distance until convergence. A set of optimal solutions is selected from the Pareto front and output.

[0013] Furthermore, the displacement distribution field of the enclosure structure in the holographic state vector is reconstructed using three-dimensional kriging interpolation, and its semi-variogram function is: In the formula, and The first The and the first The three-dimensional spatial coordinates of each sampling point The Euclidean distance between two points. For the partial sill value, For range parameters, For nugget constant, For the Kronecker function, and The predicted points were determined by fitting weighted least squares data using on-site measured data. The displacement at point is solved by the following system of equations: In the formula, The total number of sampling points participating in the interpolation. For the first Kriging weight coefficients for each sampling point For Lagrange multipliers; Predicted points after solving The displacement is , For the first Measured displacement values ​​at each sampling point.

[0014] Furthermore, the strategy for the distributed controller to drive the jacks in parallel to perform axial force adjustment is as follows: based on the optimal adjustment combination, each support is grouped according to its spatial position in the foundation pit, and the same group is adjusted synchronously, while different groups are executed sequentially from the middle to the edge or from bottom to top; the adjustment of each support is divided into 3 to 5 levels of loading, and after each level of loading, it is paused for 1 to 3 minutes until the axial force stabilizes and is compared with the target value. If the deviation exceeds ±2%, it is finely adjusted and then proceeds to the next level; after all adjustments are completed, a global verification is initiated. If the deviation between the measured axial force and displacement and the theoretical value exceeds the preset threshold, a global readjustment is triggered, and optimization and adjustment are re-executed.

[0015] Furthermore, the multi-objective optimization incorporates a coupling correction for the retaining wall displacement caused by the pit bottom rebound, whereby the retaining wall displacement is expressed as: The corrected first objective function is then: In the formula, To predict the displacement without considering the rebound coupling effect at the bottom of the pit, For the depth of the retaining structure to rebound at the bottom of the pit The coupling correction amount for the displacement. As a reference length constant, For depth The springback effect affects the transmission coefficient. The attenuation coefficient is affected by the rebound effect; The coefficient is a dimensionless coupling coefficient. This represents the current completed excavation depth. This refers to the total designed excavation depth of the foundation pit. This represents the amount of soil unloaded per unit area within the pit. This refers to the resilient modulus of the soil at the bottom of the pit.

[0016] Furthermore, a penalty term for proximity to subway tunnel protection is introduced into the velocity update of the multi-objective particle swarm optimization. The first particle The rate update formula for the next iteration is: In the formula, For the first The first particle The velocity vector of the next iteration This is the inertia weighting coefficient; , These are individual and social learning factors, respectively. , for Random numbers within a range; For the first The optimal position of each individual particle. To be the globally optimal position For the first The first particle The position vector of the next iteration. This is the penalty gain coefficient; To constrain the penalty function The gradient at the current particle position; The constraint penalty function is: In the formula, The number of adjacent subway tunnel sections that need to be protected. The first adjustment under the current combination Predicted additional displacement for each tunnel section, For the first The allowable additional displacement threshold for each section.

[0017] Furthermore, the local closed-loop correction also includes decoupling compensation for adjacent support disturbances: Pre-calibrate the disturbance transfer coefficient matrix It is an n×n square matrix with elements Indicates the first The effect of the change in unit axial force on the first The disturbance transmission coefficient of the support is calibrated by applying a unit pulse axial force to each support sequentially under the first stable working condition of the foundation pit and recording the steady-state response of the remaining supports to the axial force. When for the first The track support is fine-tuned and compensated. At that time, by Calculate the predicted additional disturbance values ​​for the remaining supports, where For the first An n-dimensional unit vector with each element being 1 and the rest being 0; For the front most affected by the disturbance Adjacent supports ,according to Calculate the reverse compensation amount; in, , In order to be with the first The set of adjacent support numbers for the road support. To dynamically compensate for the reduction factor; within a time window not exceeding 5 seconds, the distributed servo hydraulic actuator simultaneously executes the first... Fine-tuning compensation for the support and reverse compensation for adjacent supports.

[0018] The beneficial effects of this invention are as follows: 1. This invention establishes a fully automated closed-loop control chain of "perception-analysis-decision-execution-feedback" through real-time data acquisition via a distributed sensor network, dynamic updates of a digital twin model, real-time solving of a multi-objective particle swarm optimization algorithm, and rapid response of a servo hydraulic actuation system. This enables autonomous, dynamic, real-time group control of the deformation of the retaining structure throughout the entire construction process of the foundation pit. Unlike existing technologies that rely on manual experience and phased open-loop adjustments, this invention can automatically complete the entire process of state perception, optimization calculation, and axial force adjustment after each construction condition without human intervention. The timeliness and accuracy of the control are significantly superior to traditional methods.

[0019] 2. This invention incorporates the axial force adjustment of each support structure as a unified decision variable into a multi-objective optimization framework to ensure the uniformity of the axial force distribution. Axial force difference constraint between supports As optimization objectives and constraints, the axial force adjustment of each support is coordinated at the optimization level, overcoming the problem of unbalanced force caused by the "individual operation" in traditional methods. At the same time, through the adjacent support disturbance decoupling compensation strategy, this invention can simultaneously compensate for the disturbance effect of adjacent supports during local fine-tuning, further improving the accuracy of multi-support coordinated control.

[0020] 3. This invention introduces a pit bottom rebound coupling correction term, incorporating the coupling effect of pit bottom rebound caused by excavation unloading on the displacement of the retaining structure into the optimization objective function, thus achieving dynamic compensation for the pit bottom rebound effect. Furthermore, by introducing a penalty term for adjacent subway tunnel protection, sensitive environmental protection requirements are directly integrated into the particle swarm optimization algorithm's search process, achieving integrated optimization of pit deformation control and surrounding environmental protection. These two mechanisms enable this invention to automatically avoid axial force adjustment schemes that may damage the surrounding environment while meeting the pit's own safety requirements, making it particularly suitable for ultra-large deep pit projects in sensitive environmental conditions such as those adjacent to operating subways. Attached Figure Description

[0021] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the following figures are provided for illustration: Figure 1 This is a flowchart of an embodiment of the present invention. Detailed Implementation

[0022] like Figure 1 As shown, this invention discloses a method for controlling the deformation of a support structure for ultra-large foundation pits, comprising the following steps: S1. During the entire process of foundation pit excavation, a distributed sensor network is deployed on the foundation pit retaining structure and each support structure to collect in real time the lateral displacement data of the retaining structure, the actual axial force data of each support structure, the groundwater level data in the foundation pit, the soil pressure data at the ends of each support structure, and the deformation data of the surrounding environment of the foundation pit. The above multi-source heterogeneous data is then transmitted to the data processing platform through a wireless communication network. S2. The data processing platform preprocesses the received multi-source heterogeneous data, including data cleaning, outlier removal, time-series alignment, and multi-source data fusion, and constructs a holographic state vector that reflects the current stress and deformation state of the foundation pit retaining structure and support structure. The holographic state vector includes the continuous displacement distribution field of the retaining structure along the depth direction, the actual axial force distribution matrix of each support structure, the spatiotemporal variation field of the groundwater level, and the dynamic variation vector of the earth pressure at the support end. S3. Based on the holographic state vector, a digital twin model of the foundation pit retaining structure-support system, which is pre-constructed and corrected by on-site measured data, is used to perform a stress-deformation inversion analysis of the foundation pit under its current state. This analysis yields the actual bending moment distribution, shear force distribution, and theoretical target axial force values ​​of each support structure at each depth under the current external load. The digital twin model adopts an elastic foundation beam-finite element coupled model, in which the retaining structure is simulated using beam elements considering pile-soil interaction, each support structure is simulated using variable concentrated force elements, the soil inside the pit is simulated using a nonlinear soil spring based on the "m" method, and the earth pressure outside the pit is simulated using an active earth pressure model considering spatiotemporal effects. S4. The multi-objective optimization function is to minimize the overall deformation of the foundation pit, ensure the most uniform axial force distribution of each support structure, and ensure that the bearing capacity of each support structure does not exceed the limit. The servo axial force adjustment of each support structure is used as the decision variable. The displacement control threshold at each depth of the retaining structure, the upper and lower limits of the bearing capacity of each support structure, and the upper limit of the axial force difference between each support structure are used as constraints. The multi-objective particle swarm optimization algorithm is used to solve the optimal combination of servo axial force adjustment of each support structure under the current working condition. S5. Based on the optimal servo axial force adjustment combination, generate axial force adjustment commands for the servo hydraulic system corresponding to each support structure, and drive the servo hydraulic jacks at the ends of each support structure in parallel through the distributed controller to perform axial force adjustment actions. S6. During the axial force adjustment process, the actual axial force response data of each support structure and the displacement response data of the enclosure structure are fed back in real time through the distributed sensor network and compared with the theoretical expected response calculated in step S4. If the deviation between the actual response and the theoretical expected response exceeds the preset threshold, a local closed-loop correction mechanism is triggered to fine-tune and compensate the axial force of the corresponding support structure. S7. Repeat steps S2 to S6 until the excavation of the foundation pit is completed.

[0023] In this scheme, firstly, multi-source data such as retaining displacement and support axial force are collected in real time through a distributed sensor network. After preprocessing, the data is fused into a holographic state vector, realizing a "digital mirror" of the current foundation pit. Secondly, a mechanical inversion is performed using a digital twin model dynamically corrected by on-site data to accurately calculate the theoretical target axial force of the support system under the current working conditions. Then, with the three objectives of minimizing deformation, uniform axial force, and safe bearing capacity, the optimal axial force adjustment of each support is solved in real time using a multi-objective particle swarm optimization algorithm. Finally, the optimization command is sent to the distributed servo hydraulic system for execution, and local fine-tuning is performed based on real-time feedback during execution, forming a fully automatic cycle of "perception → analysis → decision → execution → correction". This achieves autonomous, real-time, and dynamic group control of the deformation of the retaining structure throughout the entire construction process of the foundation pit, fundamentally solving the problem of control lag caused by dynamic changes in external loads (soil pressure, water pressure).

[0024] In one embodiment of the present invention, the method for constructing and real-time correcting the digital twin model of the foundation pit retaining structure-support system in step S3 includes: S31. Before the excavation of the foundation pit, an initial finite element numerical model of the foundation pit retaining structure-support system is established based on the foundation pit design documents, geological survey report and surrounding environmental data. The model uses plane strain assumption or three-dimensional solid elements to simulate the interaction between the retaining structure, support structure and soil. S32, after the first support of the foundation pit is installed and before the first layer of earthwork is excavated, the displacement data of the retaining structure, the axial force data of the support structure and the earth pressure data of the initial state are obtained through the distributed sensor network. The initial state data is used as the initial boundary conditions of the model to perform the first calibration of the initial finite element numerical model and determine the soil parameters, contact parameters and structural stiffness parameters in the model. S33. After each construction condition of the foundation pit excavation is completed, the displacement distribution field of the retaining structure, the actual axial force distribution matrix of each support structure and the dynamic change vector of the earth pressure at the support end obtained by the distributed sensor network under that condition are used as new constraints. Kalman filtering or ensemble Kalman filtering algorithm is used to dynamically update the key parameters in the digital twin model so as to minimize the root mean square error between the model output and the actual monitoring data. S34. The updated digital twin model is used to predict the stress and deformation of the next construction condition. The predicted displacement values ​​of the retaining structure at each depth under the next condition and the predicted axial force values ​​of each support structure under the next condition are output as the initial reference values ​​for the multi-objective optimization calculation in step S4.

[0025] This scheme defines a four-step construction method for the digital twin model: "initial modeling → initial calibration → dynamic updating for each work condition → future prediction". An initial finite element model is established before construction; after the first support is completed, initial calibration is performed using measured data to match the actual site boundaries; after each construction condition is completed, the measured displacement field and axial force matrix are immediately used as new constraints, and a Kalman filter algorithm is introduced to dynamically update key model parameters (such as soil subgrade coefficients) to minimize the error between the model output and the measured data; and the updated model is used to predict the next work condition in advance.

[0026] This ensured that the digital twin model remained highly synchronized with the physical entity throughout the months-long construction period, possessing the ability to "self-evolve." This mechanism of continuous correction as construction progresses significantly improves the accuracy of predicting the displacement of the retaining structure under subsequent working conditions, providing highly reliable initial reference values ​​for multi-objective optimization.

[0027] In one embodiment of the present invention, the specific implementation of the multi-objective particle swarm optimization includes: S41. Initialize the particle swarm and particle positions. This indicates the axial force adjustment amount for each support; where This refers to the total number of support structures within the foundation pit that participate in group control. Indicates the first The servo axial force adjustment of the track support structure, the velocity vector of each particle is initialized to zero vector, the particle swarm size is set to 50~200, and the maximum number of iterations is set to 100~500. S42. Construct a three-objective fitness function ,in, Indicates the building envelope along the depth The weighted integral of the squared deviation between the displacement distribution in the direction and the target displacement distribution. The bottom of the pit, For the top of the pit, The depth of the foundation pit. For the depth of the foundation pit The weighting coefficient at the location, For the current adjustment combination, the enclosure structure at depth The predicted displacement at that location, For depth The target displacement control value at the location; This represents the variance of the adjusted axial force value for each support structure, where... For the first The path supports the current axial force. To adjust the average axial force; This represents the maximum value of the ratio of the axial force to the upper limit of the bearing capacity of each support structure after adjustment. For the first The upper limit of the bearing capacity of the road support; S43. Set constraints: , , , in, For depth Allowable displacement value, For the first Lower limit of road support bearing capacity This represents the upper limit of the allowable difference in axial force between any two supports; S44. The non-dominated sorting particle swarm optimization with elite retention is used for iterative solution. In each iteration, the individual optimal position and the global optimal position are updated according to the non-dominated level and crowding distance of the particle. The velocity and position of each particle are updated according to the particle swarm velocity update formula and the position update formula until the maximum number of iterations is reached or the fitness function converges. S45. From the final Pareto optimal solution set, select a set of optimal solutions based on the priority weight vector of the current construction condition as the combined output of the servo axial force adjustment amount of each support structure.

[0028] In this scheme, the adjustment amount of each support axis force is encoded as a particle position vector X. A three-dimensional fitness function is constructed as follows: The global deviation between the actual displacement and the target displacement of the retaining structure is quantified by deep weighted integration. The uniformity of axial force distribution is evaluated by calculating the variance of the adjusted axial force of each support. The structural safety margin is ensured by monitoring the ratio of maximum axial force to the upper limit of bearing capacity. Simultaneously, hard constraints are set for the limits of enclosure displacement, the upper and lower limits of single-bracket bearing capacity, and the upper limit of the difference in axial force between any two supports. The algorithm uses non-dominated sorting and crowding distance strategies for iterative optimization, ultimately selecting the optimal solution from the Pareto optimal solution set.

[0029] This scheme elevates the control of foundation pit supports from "single-objective displacement control" to a three-dimensional collaborative optimization of "displacement-uniformity-safety margin." In particular, it introduces "axial force difference constraints between supports." The algorithm ensures that multiple supports are subjected to coordinated forces in space, thus preventing sudden increases in local deformation caused by overload or slack of individual supports.

[0030] In one embodiment of the present invention, in step S2, the method for constructing the continuous displacement distribution field of the retaining structure along the depth direction in the holographic state vector is as follows: based on the discrete lateral displacement data collected by each inclinometer in the distributed sensor network at different depths of the retaining structure, a method based on the plane coordinates of the foundation pit is adopted. and depth The three-dimensional spatial kriging interpolation method is used to reconstruct the continuous displacement distribution of the entire field of the enclosure structure. The semi-variogram of the three-dimensional spatial kriging interpolation method is... The stress path characteristics for unloading during foundation pit excavation are set as follows: In the formula, and The first The and the first The three-dimensional spatial coordinates of each sampling point The Euclidean distance between two points. For the partial sill value, The range parameter is a variable parameter, and is taken as 1.5 to 3.0 times the equivalent excavation radius of the foundation pit. For nugget constant, For the Kronecker function, and The determination was made by fitting the weighted least squares method with the field measured data. Based on the semi-variogram function, any predicted point in the three-dimensional kriging interpolation method The displacement at point is solved by the following system of equations: In the formula, The total number of sampling points participating in the interpolation. For the first Kriging weight coefficients for each sampling point Given Lagrange multipliers, predict the points after solving. The displacement is , For the first Measured displacement values ​​at each sampling point.

[0031] In this scheme, a three-dimensional kriging space interpolation method is used for discrete sensor measurement point data. A semi-variogram function with an exponential model structure is defined. Among them, the range parameters A scale of 1.5 to 3.0 times the equivalent excavation radius of the foundation pit is used to reflect the spatial correlation of the excavation's impact. The optimal weighting coefficients for each sampling point are obtained by solving the Kriging equations. Then, the coordinates of any spatial point of the enclosure structure can be calculated. Displacement estimates at the location. By designing a semi-variogram that considers the influence range of the excavation space, a high-precision continuous displacement cloud map of the entire retaining structure can be reconstructed, providing a basis for... Objective function at full depth The continuous integral provides reliable data support, significantly improving the overall precision of deformation control.

[0032] In one embodiment of the present invention, the specific control strategy for the distributed controller to drive the servo hydraulic jacks at the ends of each support structure to perform axial force adjustment actions in step S5 includes: S51. Based on the optimal servo axial force adjustment combination of each support structure output in step S4, the adjustment amount is divided into several adjustment groups according to the spatial distribution of each support structure in the foundation pit. The support structures in the same group perform axial force adjustment actions synchronously, and the axial force adjustment actions of different groups are performed sequentially from the middle of the foundation pit to the edge of the foundation pit or from the bottom of the foundation pit to the top of the foundation pit. S52. For each support structure, the servo hydraulic jack at its end performs axial force adjustment in a graded loading manner, dividing the total adjustment into 3 to 5 levels. After each level of loading is completed, pause for 1 to 3 minutes until the axial force sensor reading of the support structure stabilizes. Then, read the actual axial force value and compare it with the target axial force value of the level. If the deviation exceeds ±2% of the target value of the level, make a fine adjustment correction before entering the next level of loading. S53. After the axial force adjustment of all support structures is completed, the global stability verification program is started. The actual axial force data and the displacement data of the retaining structure after the adjustment of all support structures are collected through the distributed sensor network. The deviation between the actual axial force value of each support structure and the theoretical target axial force value calculated in step S4, as well as the deviation between the actual displacement value and the theoretical expected displacement value at each depth of the retaining structure, are calculated. If any deviation exceeds the preset allowable deviation threshold, the global readjustment program is triggered, and steps S4 to S5 are executed again.

[0033] In this scheme, supports are grouped according to their spatial location (synchronous within the same group, with adjustments made between groups in a sequence from the center to the edge or from bottom to top) to prevent structural torsion. Secondly, when adjusting a single support, a "graded loading, step-by-step approximation" principle is adopted (3-5 levels, with a 1-3 minute pause between levels), allowing the axial force sensor readings to stabilize before proceeding to the next level. Finally, after all adjustments are completed, a global verification is initiated. If the deviation between the measured and theoretical values ​​exceeds the limit, global re-optimization and readjustment are automatically triggered.

[0034] This solution effectively suppresses hydraulic shock and additional dynamic deformation of the enclosure structure caused by concentrated sudden loads from servo jacks through a process strategy of "grouping sequentially" and "graded loading." Its global closed-loop mechanism of "execution-verification-re-execution" ensures that the optimal adjustment amount output by the algorithm can be accurately reproduced by the physical actuator, significantly improving the stability and reliability of servo control.

[0035] In one embodiment of the present invention, in step S4, the multi-objective optimization function further introduces a correction term for the coupling effect of the pit bottom rebound caused by the excavation unloading on the displacement of the retaining structure, wherein the retaining structure is along the depth Predicted displacement value at Calculated using the following formula: in, Without considering the rebound coupling effect at the bottom of the pit, the retaining structure calculated by the digital twin model at the depth The predicted displacement value at that location, For the depth of the retaining structure to rebound at the bottom of the pit The coupling correction for the displacement is calculated using the following formula: The corrected first objective function is then: In the formula, As a reference length constant, For depth The springback effect affects the transmission coefficient. The attenuation coefficient due to rebound effect was determined by inversion using on-site measured data; The coupling coefficient between the rebound displacement at the bottom of the pit and the lateral displacement of the retaining structure is determined by a digital twin model, with a value ranging from 0.1 to 0.5. This represents the current completed excavation depth. This refers to the total designed excavation depth of the foundation pit. This represents the amount of soil unloaded per unit area within the pit. This refers to the resilient modulus of the soil at the bottom of the pit.

[0036] This solution compensates for the coupling effect in advance in the optimization algorithm, enabling the dynamic adjustment of the support axial force to actively "pre-resist" the additional deformation caused by rebound, thus significantly improving the deformation control accuracy of the retaining structure within the full excavation depth range.

[0037] In one embodiment of the invention, a penalty term for adjacent subway tunnel protection is introduced into the velocity update of the multi-objective particle swarm optimization. The first particle The rate update formula for the next iteration is: In the formula, For the first The first particle Velocity vector of the next iteration, dimensions and axial force adjustment same, This is the inertia weighting coefficient, with a value ranging from 0.4 to 0.9; , These are individual and social learning factors, with values ​​ranging from 1.5 to 2.5. , for Random numbers within a range; For the first The optimal position of each individual particle. To be the globally optimal position For the first The first particle The position vector of the next iteration. The penalty gain coefficient, , The dimensionless penalty coefficient, The value ranges from 0.01 to 0.1. The reference force is the average of the upper limits of the bearing capacity of each support. To constrain the penalty function The gradient at the current particle position; the constraint penalty function is: In the formula, The number of adjacent subway tunnel sections that need to be protected. The first adjustment under the current combination The predicted additional displacement of each tunnel section is calculated using the tunnel-soil-excavation coupled sub-model in the digital twin model. For the first The allowable additional displacement threshold for each section.

[0038] In this scheme, when the axial force adjustment combination detected by the particle search causes displacement of the adjacent tunnel... Exceeding the time limit, It increases sharply. Its gradient... This generates a "repulsive force," forcibly dragging the particle swarm out of the solution space that could lead to tunnel damage. This scheme internalizes the protection standards for the sensitive environment (operating subway) into a dynamic constraint within the optimization algorithm, rather than a reactive post-implementation verification. This mechanism ensures that the particle swarm iteration always searches for the optimal solution within the dual feasible domains of "excavation safety" and "environmental micro-disturbances," mitigating the risk of adjacent tunnel structural safety accidents caused by improper axial force control from the decision-making source.

[0039] In one embodiment of the present invention, the local closed-loop correction further includes adjacent support disturbance decoupling compensation: pre-calibrating the disturbance transfer coefficient matrix. It is an n×n square matrix with elements Indicates the first The effect of the change in unit axial force on the first The disturbance transmission coefficient of the support is calibrated by applying a unit pulse axial force to each support sequentially under the first stable working condition of the foundation pit and recording the steady-state response of the remaining supports to the axial force. When for the first The track support is fine-tuned and compensated. At that time, by Calculate the predicted additional disturbance values ​​for the remaining supports, where For the first An n-dimensional unit vector with each element being 1 and the rest being 0; For the front most affected by the disturbance Adjacent supports ,according to Calculate the reverse compensation amount, where, , In order to be with the first The set of adjacent support numbers for the road support. The dynamic compensation reduction factor ranges from 0.6 to 0.9; within a time window not exceeding 5 seconds, the distributed servo hydraulic actuator simultaneously executes the first... Fine-tuning compensation for the support and reverse compensation for adjacent supports.

[0040] In this scheme, by using a pre-calibrated transfer matrix and an instant decoupling compensation strategy, the interference of single-point fine-tuning on the global support chain is minimized, ensuring the overall stability of the support system and the balance of axial force distribution during dynamic adjustment, so that "group control" can truly achieve a precise and collaborative industrial control level.

[0041] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made to it in form and detail without departing from the scope defined by the claims of the present invention.

Claims

1. A method for controlling the deformation of a support structure for an ultra-large foundation pit, characterized in that, Includes the following steps: S1. Deploy a distributed sensor network to collect real-time data on the displacement of the retaining structure, the axial force of the support, the water level, the earth pressure, and the deformation of the surrounding area, and transmit the data to the data processing platform. S2. The data processing platform preprocesses and fuses multi-source data to construct a holographic state vector that includes the displacement distribution field of the retaining structure, the axial force distribution matrix of the support, the spatiotemporal field of the water level, and the earth pressure vector. S3. Based on the holographic state vector, an inversion analysis is performed using a digital twin model corrected by field data to obtain the bending moment, shear force distribution of the retaining structure and the theoretical target axial force of each support. The digital twin model adopts an elastic foundation beam-finite element coupled model, uses the "m" method nonlinear soil spring to simulate the soil inside the pit, and uses active earth pressure considering spatiotemporal effects to simulate the load outside the pit. S4. Taking the minimum overall deformation of the foundation pit, the most uniform distribution of the axial force of the support, and the bearing capacity not exceeding the limit as multiple objectives, the adjustment amount of the servo axial force of each support is used as the decision variable, and the displacement threshold, the upper and lower limits of the bearing capacity, and the limit of the difference in the axial force of the support are used as constraints. The optimal combination of adjustment amounts is solved by multi-objective particle swarm optimization. S5. Generate instructions based on the optimal adjustment amount, and drive the servo jacks at each support end to perform axial force adjustment in parallel through the distributed controller; S6. During the adjustment process, the actual axial force and displacement response are fed back in real time. If the deviation from the theoretical expectation exceeds the threshold, a local closed-loop correction is triggered.

2. The method for controlling the deformation of a large foundation pit support structure according to claim 1, characterized in that: The construction and modification of the digital twin model includes: An initial finite element model is established before construction. After the first support is installed and before the first layer is excavated, the model parameters are calibrated using initial monitoring data. After each construction condition is completed, the key parameters of the model are dynamically updated using Kalman filtering or ensemble Kalman filtering, with the measured displacement field, axial force matrix, and earth pressure vector as constraints, so as to minimize the root mean square error between the model output and the measured data. The updated model is used to predict the displacement and axial force of the next condition, which serves as the initial reference for multi-objective optimization.

3. The method for controlling the deformation of a large foundation pit support structure according to claim 1, characterized in that, The specific implementation of multi-objective particle swarm optimization includes: Initialize the particle swarm and particle positions. Represent the axial force adjustment for each support; construct a three-objective fitness function: In the formula, The depth of the foundation pit. For the depth of the foundation pit The weighting coefficient at the location, For the current adjustment combination, the enclosure structure at depth The predicted displacement at that location, For the target displacement, For the first The path supports the current axial force. To adjust the average axial force, For the first The upper limit of the bearing capacity of the road support; Set constraints: , , , in, For depth Allowable displacement value, For the first Lower limit of road support bearing capacity This represents the upper limit of the allowable difference in axial force between any two supports; The solution is obtained by iteratively solving a non-dominated sorting particle swarm with elite retention. The individual optimal and global optimal are updated with non-dominated level and crowding distance until convergence. A set of optimal solutions is selected from the Pareto front and output.

4. The method for controlling the deformation of a large foundation pit support structure according to claim 1, characterized in that, The displacement distribution field of the enclosure structure in the holographic state vector is reconstructed using three-dimensional kriging interpolation, and its semi-variogram function is: In the formula, and The first The and the first The three-dimensional spatial coordinates of each sampling point The Euclidean distance between two points. For the partial sill value, For range parameters, For nugget constant, For the Kronecker function, and The predicted points were determined by fitting weighted least squares data using on-site measured data. The displacement at point is solved by the following system of equations: In the formula, The total number of sampling points participating in the interpolation. For the first Kriging weight coefficients for each sampling point For Lagrange multipliers; Predicted points after solving The displacement is , For the first Measured displacement values ​​at each sampling point.

5. The method for controlling the deformation of a large foundation pit support structure according to claim 1, characterized in that, The strategy for the distributed controller to drive the jacks in parallel to adjust the axial force is as follows: based on the optimal adjustment combination, each support is grouped according to its spatial position in the foundation pit. Supports in the same group are adjusted synchronously, while supports in different groups are adjusted sequentially from the middle to the edge or from bottom to top. The adjustment of each support is divided into 3 to 5 levels of loading. After each level of loading, the load is paused for 1 to 3 minutes until the axial force stabilizes and is compared with the target value. If the deviation exceeds ±2%, a fine adjustment is made before proceeding to the next level. After all adjustments are completed, a global verification is initiated. If the deviation between the measured axial force and displacement and the theoretical value exceeds a preset threshold, a global readjustment is triggered, and optimization and adjustment are re-executed.

6. The method for controlling the deformation of a large foundation pit support structure according to claim 1, characterized in that, The multi-objective optimization introduces a coupling correction for the retaining wall displacement based on the bottom rebound, and the retaining wall displacement is expressed as: The corrected first objective function is then: In the formula, To predict the displacement without considering the rebound coupling effect at the bottom of the pit, For the depth of the retaining structure to rebound at the bottom of the pit The coupling correction amount for the displacement. As a reference length constant, For depth The springback effect affects the transmission coefficient. The attenuation coefficient is affected by the rebound effect; The coefficient is a dimensionless coupling coefficient. This represents the current completed excavation depth. This refers to the total designed excavation depth of the foundation pit. This represents the amount of soil unloaded per unit area within the pit. This refers to the resilient modulus of the soil at the bottom of the pit.

7. The method for controlling the deformation of a large foundation pit support structure according to claim 1, characterized in that, The multi-objective particle swarm optimization incorporates a penalty term for proximity to subway tunnel protection in its velocity update. The first particle The rate update formula for the next iteration is: In the formula, For the first The first particle The velocity vector of the next iteration This is the inertia weighting coefficient; , These are individual and social learning factors, respectively. , for Random numbers within a range; For the first The optimal position of each individual particle. To be the globally optimal position For the first The first particle The position vector of the next iteration. This is the penalty gain coefficient; To constrain the penalty function The gradient at the current particle position; The constraint penalty function is: In the formula, The number of adjacent subway tunnel sections that need to be protected. The first adjustment under the current combination Predicted additional displacement for each tunnel section, For the first The allowable additional displacement threshold for each section.

8. The method for controlling the deformation of a large foundation pit support structure according to claim 1, characterized in that, The local closed-loop correction also includes decoupling compensation for adjacent support disturbances: Pre-calibrate the disturbance transfer coefficient matrix It is an n×n square matrix with elements Indicates the first The effect of the change in unit axial force on the first The disturbance transmission coefficient of the support is calibrated by applying a unit pulse axial force to each support sequentially under the first stable working condition of the foundation pit and recording the steady-state response of the remaining supports to the axial force. When for the first The track support is fine-tuned and compensated. At that time, by Calculate the predicted additional disturbance values ​​for the remaining supports, where For the first An n-dimensional unit vector with each element being 1 and the rest being 0; For the front most affected by the disturbance Adjacent supports ,according to Calculate the reverse compensation amount; in, , In order to be with the first The set of adjacent support numbers for the road support. To dynamically compensate for the reduction factor; within a time window not exceeding 5 seconds, the distributed servo hydraulic actuator simultaneously executes the first... Fine-tuning compensation for the support and reverse compensation for adjacent supports.

Images