A support structure deformation prediction and active control method and device based on computational learning and a computer storage medium

Abstract

This invention discloses a method, device, and computer storage medium for predicting and actively controlling the deformation of support structures based on computational learning. It relates to the fields of geotechnical engineering and underground structure safety control technology. This method, based on the theory of support structure displacement calculation and combined with a theoretical calculation model, achieves accurate prediction of the horizontal displacement of the support structure, completes structural displacement calculation and dangerous area identification, and outputs deformation curves and potential risk locations. It employs a Bayesian optimization framework for automatic parameter optimization, initializes the prior distribution and sets the objective function, uses Gaussian process modeling to iteratively calculate the acquisition function, and uses a surrogate model to quickly predict performance to update the model. Finally, it outputs the optimal parameter combination to guide construction, realizing an intelligent closed-loop design from theoretical calculation to engineering control, and solving the problems of insufficient deformation prediction accuracy and lagging risk control in complex strata.

Description

Technical Field

[0001] This invention relates to the field of geotechnical engineering and underground structure safety control technology, and more specifically, to a method and device for predicting and actively controlling the deformation of support structures based on computational learning, as well as a computer storage medium. Background Technology

[0002] With the continuous development of urban underground space and the increasing demand for high-rise building foundation construction, deep foundation pit engineering is widely used in urban construction. However, under conditions of uneven soil texture, the deformation and stress problems of deep foundation pit retaining structures are becoming increasingly prominent. The physical and mechanical properties of different soil layers vary significantly; for example, the upper layer may be soft clay or silty clay, while the lower layer may be silt or gravel. This heterogeneity causes the retaining structure to bear complex earth pressure distributions during excavation, leading to concentrated horizontal displacement, shifts in peak bending moments, and even structural cracking and overall instability.

[0003] Existing methods for foundation pit design and analysis are primarily based on the assumptions of homogeneous strata and Euler-Bernoulli beams, employing traditional Rankine or Coulomb earth pressure theories or utilizing finite element numerical simulations. These methods achieve good results under homogeneous strata conditions, but their prediction accuracy significantly decreases in heterogeneous strata. Traditional theories do not consider the impact of variations in stratum stiffness, interlayer interactions, and shear effects on the support structure, leading to significant deviations in calculation results. While numerical simulations can reflect complexity, their modeling, parameter selection, and computational costs are high, making them unsuitable for rapid application in the early stages of engineering projects. Furthermore, existing methods mostly focus on passive response and control, implementing reinforcement measures only after deformation exceeds limits through construction monitoring, which introduces delays and risks.

[0004] To address the complexity of heterogeneous soil layers, some scholars have proposed layered analysis or segmented elastic model methods, assigning different physical parameters to each layer to improve computational accuracy. However, these methods still face challenges in engineering practice, including cumbersome data input, low computational efficiency, and insufficient visualization of results. Especially in multi-layered soil conditions, the displacement curves of support structures exhibit nonlinear characteristics, making it difficult for traditional calculation tools to automatically identify hazardous areas, often requiring manual judgment, thus increasing human error and time costs.

[0005] On the other hand, safety control in deep foundation pit engineering is gradually developing towards intelligence and proactivity. In recent years, with the rapid advancement of computer technology, data processing, and intelligent algorithms, the trend of combining mechanical calculations with program algorithms has become increasingly apparent. Computer programs can process a large number of input parameters in a very short time, quickly complete model calculations and risk analyses, and provide a new technical path for predicting structural deformation under complex geological conditions. However, there are still relatively few computer-aided systems for active deformation control of support structures, and most research remains at the level of a single calculation module, failing to systematically integrate theoretical calculations with engineering reinforcement decision-making recommendations.

[0006] Therefore, there is an urgent need for a systematic technical solution that can combine independent theories and program algorithms to achieve automated input of formation parameters, rapid prediction of structural deformation, intelligent identification of potential dangerous locations, and proposal of engineering reinforcement strategies. Summary of the Invention

[0007] This invention aims to overcome at least one of the defects of the prior art and provides a method, device, and computer storage medium for predicting and actively controlling the deformation of support structures based on computational learning. This addresses the problems of insufficient deformation prediction accuracy, lagging risk control, and the failure to systematically integrate theoretical calculations and engineering reinforcement decision-making suggestions in complex geological formations.

[0008] The technical solution adopted in this invention is a method for predicting and actively controlling the deformation of support structures based on computational learning, comprising the following steps:

[0009] S101: Determine the corresponding soil layer parameters and support structure parameters for the area where the foundation pit is located; S102: Input of engineering information and soil layer parameters; S103: Structural displacement calculation based on an autonomous theoretical model; S104: Automatic identification and autonomous risk classification of hazardous areas; S105: Active control of support structure deformation and dynamic optimization using Bayesian optimization algorithm. The organic combination of these steps forms a complete prediction-identification-control closed-loop method.

[0010] In step S101, based on comprehensive engineering geological surveys, standard penetration tests, static cone penetration tests, and other in-situ tests... Through systematic indoor geotechnical tests, a complete geotechnical engineering parameter database was established. This database not only includes basic parameters such as unit weight, cohesion, internal friction angle, and deformation modulus of each soil layer, but also determines the spatial variability characteristics of these parameters through statistical analysis. Simultaneously, based on the construction drawings and material testing reports of the support structure, geometric parameters such as thickness variation range and embedment depth, as well as constitutive mechanical parameters considering material nonlinearity, were accurately obtained. These validated parameters form the basis for subsequent theoretical analysis, and intelligent segmentation of soil layers and optimized segmentation of the support structure were achieved through clustering algorithms. This provides a reliable data foundation for subsequent accurate calculations and guides the completion of engineering geological stratification of the soil and the calculation segmentation of the support structure.

[0011] Step S102 inputs the acquired engineering information and soil parameters into the system, focusing on soil parameters such as unit weight, cohesion, and internal friction angle, as well as key parameters such as the thickness, embedment depth, moment of inertia, and elastic modulus of the support structure. Based on the soil layers and the location of the internal supports, the system rationally segments the support structure, ensuring mechanical consistency under complex stress conditions by strictly meeting displacement compatibility, rotation continuity, moment equilibrium, and shear force transfer conditions.

[0012] Step S103 is the core innovation of this invention, achieving accurate calculation of structural displacement through the establishment of an independent theoretical model. Addressing the sensitivity of soil pressure distribution and the complex mechanical response of the support structure in soft-hard interbedded strata, the model introduces a layered soil pressure calculation formula considering soil displacement and a shear deflection angle effect based on Timoshenko beam theory. By substituting these two theoretical models into the mechanical equilibrium equations of the support structure, fourth-order variable-coefficient non-homogeneous differential equations for the displacement of the support structure in the areas above and below the excavation face are derived. These equations are solved using a high-order power series expansion method, obtaining a ten-term series expansion that meets the engineering error requirement of ≤5%. A closed-loop analytical solution for the horizontal deformation of the support structure considering soil pressure and bending-shear coupling effects of structural displacement is established. The calculation process is as follows: The structural displacement calculation based on the autonomous theoretical model takes into account the sensitivity of soil pressure distribution in soft-hard interactive strata and the complex mechanical response of the support structure. It introduces a layered soil pressure calculation formula that takes into account soil displacement, as well as the structural deflection angle effect under shear action. The formula for calculating earth pressure is: In the formula: This is the earth pressure at rest; This refers to the active earth pressure when the soil reaches its ultimate displacement. The coefficient of earth pressure at rest for the nth soil layer; The active earth pressure coefficient of the nth soil layer; n is the soil layer on the calculation surface; This represents the ultimate displacement value of the support structure when the soil is in a state of ultimate equilibrium. Let be the unit weight of the i-th soil layer; Let be the thickness of the i-th soil layer; The unit weight of the nth soil layer; y represents the cohesion of the nth soil layer; y represents the displacement of the soil layer due to the support structure. The structural deflection angle effect under shear load is mainly determined based on Timoshenko beam theory, and the specific calculation formula is as follows:

[0013]

[0014]

[0015] In the formula: Let be the bending moment of the i-th pile element micro-element; The shear force of the i-th pile element is given by the following formula: The width for calculating earth pressure is the pile spacing. For the longitudinal bending stiffness of the support piles; For the longitudinal shear stiffness of the support piles; The rotation angle of the support pile section; The horizontal displacement of the i-th support pile; Combining the two theoretical calculation formulas above, and substituting them into the mechanical equilibrium formula for the support structure, we can obtain the fourth-order variable-coefficient non-homogeneous differential equation for the displacement above the excavation surface of the support structure as follows:

[0016] In the formula: ; ; ; .

[0017] The fourth-order variable-coefficient nonhomogeneous differential equation for the displacement below the excavation surface of the support structure is:

[0018] Where: h is the excavation depth of the foundation pit under the calculation condition (m); The width for calculating soil reaction force; , ; denoted as , where is the magnitude of the active zone earth pressure distribution below the excavation surface of the foundation pit; m is the proportional coefficient of the horizontal resistance coefficient of the foundation soil; and x is the depth at which the displacement of the supporting structure is calculated. The analytical solution for the displacement of the support structure above the excavation face is:

[0019] In the formula: by ensuring that the displacement, rotation, bending moment, and shear force of each section of the support structure are equal during the specific solution process, the equilibrium equation is established to obtain the following: ; The analytical solution for the displacement of the support structure below the excavation face is: In the formula: ; ; These are the initial conditions; Let be the coefficient of earth pressure at rest for the j-th soil layer; Let be the active earth pressure coefficient of the j-th soil layer; The coefficient of static earth pressure below the excavation surface of the foundation pit; Let J be the unit weight of the j-th soil layer; Let be the unit weight of the i-th soil layer; Let be the cohesion of the j-th soil layer; j is the sum of the number of soil layers above the excavation face and the number of soil layers below the excavation face. This refers to the magnitude of the deflection caused by the stress on the support structure below the excavation face.

[0020] Step S104 involves automatic identification and autonomous risk classification of hazardous areas. Based on the analytical solution of the horizontal displacement of the support structure obtained in step S103, the displacement curve of the support structure during the excavation of the foundation pit is calculated. The displacement of the support structure is automatically compared with the preset multi-level safety thresholds. The safety thresholds are set according to the deformation control requirements of each project and should meet the design requirements of the specifications. According to the comparison results, the support structure is divided into three risk levels: safe state, early warning state, and dangerous state. The system automatically identifies the sections whose displacement exceeds the dangerous threshold as dangerous areas. Based on this, the system generates a visual early warning map containing the risk level and spatial location, and provides a basis for the active control decision in step S105.

[0021] Step S104 constructs the displacement field of the support structure, incorporating spatiotemporal evolution characteristics. A multi-index fusion risk assessment system is established, intelligently matching the calculated displacement with dynamic safety thresholds considering construction stages and environmental factors. The risk assessment module employs a fuzzy comprehensive evaluation method, comprehensively considering multi-dimensional indicators such as displacement magnitude, rate of change, and spatial distribution characteristics to achieve a precise three-level classification of safe, warning, and hazardous states. The system's intelligent early warning engine can automatically identify areas of abnormal displacement and generate visualized early warning maps with high positioning accuracy and clear risk levels, providing a scientific basis and data support for subsequent proactive control decisions.

[0022] Step S105: Based on the real-time risk assessment results, the system initiates an intelligent reinforcement decision-making mechanism. Utilizing a Bayesian optimization strategy, it dynamically selects the most cost-effective measures and parameter combinations. Each selected measure-parameter combination is evaluated for effectiveness and cost using a fast surrogate model. The results serve as new observation data, and the displacement response of the support structure after applying additional measures is calculated in real-time using the Bayesian formula, quantifying the reinforcement effect. This method establishes a complete closed-loop control mechanism of perception-decision-execution-verification. Through multiple rounds of iterative calculations and scheme optimization, it gradually eliminates weak links in the support structure system until the displacement response at all key locations meets dynamic design indicators and construction control requirements, forming a continuously improving intelligent control process.

[0023] Another objective of this invention is to provide a computational learning-based device for predicting and actively controlling the deformation of support structures, comprising: The parameter determination module is used to determine the corresponding soil layer parameters and support structure parameters of the area where the foundation pit is located; The parameter input module is used for inputting engineering information and soil layer parameters; The structural displacement calculation module is used for structural displacement calculation based on the autonomous theoretical model. The autonomous risk classification module is used for automatic identification and autonomous risk classification of hazardous areas. The intelligent reinforcement design and iterative verification module is used to determine the reinforcement parameter combination of deep mixing piles for each high-risk location identified by the autonomous classification module using a Bayesian optimization algorithm. It automatically adjusts the pile design parameters to meet the displacement control specifications and starts the next intelligent optimization iteration. This cycle continues until the displacement of the support structure meets the design specifications.

[0024] This device is particularly suitable for displacement analysis, risk identification, and active reinforcement design of retaining structures in deep foundation pit projects with uneven soft and hard strata, thereby improving the safety and stability of foundation pit support structures.

[0025] Another object of the present invention is to provide a computer storage medium storing a computer program that implements the above-described method for predicting and actively controlling the deformation of support structures based on computational learning.

[0026] Compared with existing technologies, the beneficial effects of this invention are significantly improved. By establishing an accurate theoretical model and employing a high-order power series solution method, it achieves higher computational accuracy than traditional empirical formula methods and numerical simulation methods. Furthermore, this invention innovatively constructs a closed-loop control method encompassing calculation, identification, reinforcement, verification, construction, and feedback, realizing a fundamental shift from traditional passive support to modern active control, significantly improving the safety and reliability of foundation pit engineering. This method, through multi-source data fusion technology and iterative optimization mechanisms, demonstrates strong adaptability to complex geological conditions and can automatically adjust control strategies based on actual engineering responses, exhibiting excellent engineering applicability. In addition, the visualization early warning system based on analytical solutions and the multi-level risk classification method provide a scientific basis for engineering decision-making, effectively reducing engineering risks and management costs. This invention organically combines theoretical models with intelligent algorithms, promoting the development of support structure technology from empirical to theoretical, and from standardized to intelligent, providing a complete intelligent solution for deep foundation pit engineering. This method deeply integrates mechanical models with computer algorithms, realizing intelligent closed-loop design from theoretical calculation to engineering control. It significantly improves the safety, reliability, and economy of deep foundation pit support structures in uneven soft and hard strata, and has good prospects for engineering promotion and application. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the overall structure of the present invention.

[0028] Figure 2 This is a flowchart illustrating the specific process of experimentation, calculation, identification, reinforcement, and verification of this invention.

[0029] Figure 3 This is the mechanical theoretical model of the support structure above the excavation face in this invention.

[0030] Figure 4 This is the mechanical theoretical model of the support structure below the excavation face in this invention.

[0031] Figure 5 This is a schematic diagram of the point-like reinforcement measure of the present invention.

[0032] Figure 6 This is a flowchart illustrating the process and principle of the machine learning module of the present invention.

[0033] Figure 7 A block diagram illustrating the Bayesian optimization principle of this invention. Detailed Implementation

[0034] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the invention. To better illustrate the following embodiments, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions; it is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0035] Example 1 The technical solutions 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.

[0036] like Figure 1 As shown, the figure contains the following core components and working conditions: (1) The original underground support structure is located outside the foundation pit, usually a continuous underground wall or a pile retaining structure, which constitutes the main retaining and water-stopping system before the foundation pit is excavated. This structure serves as the reference carrier for stress and deformation in the figure, and its displacement is also the main object of system prediction and control; (2) Point-like deep mixing pile reinforcement: a precise and local reinforcement measure taken to deal with the high-risk or displacement-exceeding areas identified by the system. These mixing piles are not uniformly distributed, but are specifically arranged in key parts where the displacement is too large or the soil is weak after simulation calculation. The pile body is mixed with the original soil through cement and other curing agents to form a cement-soil composite with significantly improved strength and modulus, thereby forming an effective lateral constraint and reinforcement for the original support structure and controlling its further deformation; (3) Foundation pit excavation area: located inside the original support structure, indicating the space where earthwork excavation is being carried out or is planned. The unloading caused by excavation is the direct cause of the lateral displacement and deformation of the support structure. The diagram clearly shows the spatial relationship between reinforcement work and excavation activities; (4) Fiber optic displacement monitoring: a fiber optic sensing system vertically buried or attached along the original underground support structure. This monitoring method can realize continuous and high-precision displacement distribution measurement of the support structure along the depth direction, replacing the traditional discrete point inclinometer, providing full-section, real-time continuous displacement field training for subsequent machine learning models, and collecting and verifying data. It is the core of the "sensing" link in the closed-loop system.

[0037] like Figure 2 The diagram shown illustrates the overall workflow and logical block diagram of the intelligent support structure system based on computational learning and active control provided by this invention. This diagram systematically demonstrates the complete closed-loop process from data input, intelligent computation, risk identification to active control and iterative optimization. Specifically, the process in the diagram includes the following core stages and logical modules: Engineering Information and Soil Parameter Input: This module serves as the initialization and data-driven source for the method. It is responsible for inputting the geometric dimensions of the foundation pit, the physical and mechanical parameters of each soil layer, the design parameters of the support structure, and construction condition information, providing a complete data foundation for subsequent mechanical calculations and machine learning analysis.

[0038] S101 and S102: This stage is the initial stage of the entire method, providing basic engineering information for the method and detailed parameter data for subsequent refined models.

[0039] S103: Calculation of structural displacement based on the proprietary theoretical model: This is the core calculation step. The system calls upon the proprietary theoretical model for calculating the displacement of the support structure, and combines it with the input strata and structural parameters to calculate and analyze the initial displacement field, providing a basis for risk assessment.

[0040] S104: Automatic identification and risk classification of hazardous areas. This step is intelligent diagnosis. The method is based on the calculation results of S103. Through preset rules or integrated machine learning classification models, it automatically identifies high-risk sections with excessive displacement. Based on indicators such as displacement value and rate of change, it quantifies and classifies the risk, realizing the transformation from data to risk perception.

[0041] S105: Intelligent Reinforcement Design and Iterative Verification: This is the decision-making and optimization closed-loop process. The method automatically generates point reinforcement schemes, represented by deep mixing piles or cast-in-place piles, for the hazardous areas identified in S104, and recalculates the structural displacement after reinforcement. Through a design-calculation-judgment cycle, until the displacement of the support structure fully meets design specifications and safety requirements, an intelligent decision-making process from problem identification to solution closure is formed.

[0042] Figure 2 The flowchart clearly reveals the integrated technical path of this invention, which includes data-driven approaches, theoretical calculations, intelligent diagnosis, optimized control, and iterative verification. Each step is tightly connected through data flow and logical judgment, forming a complete solution for the safety control of deep foundation pit support that combines theoretical rigor with the adaptability of artificial intelligence.

[0043] Due to the complexity of internal force changes in the structural support system, this paper adopts a segmented approach to divide the support structure into several segments. A coordinate system is established for each segment, with the origin of each coordinate system from the top of the structure downwards being... ( =1,2,…), with coordinate axes aligned with the diagram. For the support structure above the excavation face, consider a micro-element for the i-th segment of the support structure along the x-direction. The force analysis of this micro-element is as follows: Figure 3 As shown. Figure 4 To calculate the deformation and stress of the support structure below the excavation surface of the foundation pit, the support structure below the excavation surface is considered as a Timoshenko beam placed vertically on the Winkler elastic foundation. Since the support structure above the excavation surface is subjected to stress and deformation, the support structure below the excavation surface has the initial condition of being subjected to bending moment and shear force at the excavation surface location.

[0044] like Figure 5 The diagram shown illustrates the implementation of the intelligent reinforcement method for support structures provided by this invention under conditions including special geological treatment and drilling processes. The diagram further illustrates the key procedures and monitoring methods before and after point reinforcement under complex geological conditions. Specifically, the diagram includes the following core components and working condition characteristics: Piling mud: Circulating mud used in the drilling process of point-grown deep cast-in-place piles to maintain the stability of the borehole wall and prevent collapse. Its hydrostatic pressure can effectively balance the soil pressure and water pressure outside the borehole, ensuring the quality of borehole formation and construction safety, and creating the necessary conditions for subsequent concrete pouring to form a reinforced pile.

[0045] Original underground support structure: refers to the underground continuous wall or support piles that have been constructed and have assumed the initial support function before the implementation of point-like deep cast-in-place pile reinforcement. This structure is the initial object for the system to perform displacement calculation and risk identification, and it is also the carrier and protection target of subsequent reinforcement measures.

[0046] Fiber optic displacement monitoring: A distributed fiber optic sensing system, vertically deployed along the initial support structure, is used to monitor the distribution of its lateral displacement along the depth in real time, continuously, and with high precision. This monitoring data constitutes the core input for machine learning models to make predictions and verify effects, realizing visualization and digital monitoring of the entire construction process.

[0047] Point-grouted deep cast-in-place piles: These are locally reinforced piles installed in high-risk areas based on the system's intelligent optimization results. Unlike mixing piles, cast-in-place piles use a process of drilling holes and then pouring concrete to form a high-strength concrete pile body. They are suitable for areas with higher reinforcement strength requirements or more complex soil conditions.

[0048] Figure 5 The invention reveals the adaptability of its technical solution to complex geological formations and specific processes: it strengthens the structure by using point-like deep-filled piles in areas with alternating soft and hard geological formations, combined with wall-supporting mud stabilization technology; at the same time, it relies on fiber optic monitoring data to track and provide feedback on the displacement of the initial support structure throughout the entire process, ensuring the accuracy and safety of the reinforcement construction, and demonstrating the refined construction of geological identification, process adaptation, precise reinforcement, and full-process monitoring.

[0049] like Figure 6 The diagram shown is a closed-loop flowchart of intelligent reinforcement decision-making and implementation based on Bayesian algorithms and machine learning in this invention. The diagram clearly illustrates the dynamic cycle of the entire process from risk input, intelligent decision-making, scheme verification to construction feedback.

[0050] The process begins with the input of the risk area and initial displacement. Based on the preliminary identification results, the system uses the target area to be reinforced and its deformation state as the starting point for optimization. It then enters the core Bayesian optimization decision module. This module first initializes the prior probability distribution based on existing knowledge, and then, through calculating the acquisition function, quantitatively weighs the exploration of the unknown parameter space against utilizing known efficient solutions, thereby selecting the current optimal candidate reinforcement scheme. This scheme is then output to the theoretical model accuracy verification stage, where simulation calculations and safety checks are performed using the support structure mechanical model proposed in this invention. The results are then judged: if they meet the specifications, the output is an optimized reinforcement design that can guide construction; if they do not meet the specifications, the scheme and result data are immediately fed back into the learning loop for redesign and trial calculations until the specifications are met.

[0051] Through actual construction and real-time monitoring data collection, the system obtains accurate response information from the support structure and provides feedback data for continuous model optimization. This field data is used to update the probabilistic model and machine learning surrogate model in Bayesian optimization, achieving knowledge evolution from prior to posterior, thus making the system more adaptable and accurate in subsequent decisions. The entire process forms a reinforcement learning closed loop characterized by intelligent decision-making, precise verification, and data-driven evolution. This not only significantly improves the scientific and economical nature of reinforcement scheme formulation but also endows the system with the ability to continuously improve itself through repeated practice.

[0052] like Figure 7As shown, this paper describes the active decision-making control closed loop based on Bayesian optimization in an intelligent support structure system. The process begins with initialization and modeling: a multi-objective function is defined, with displacement, cost, and schedule weighted as the core; the parameter space of the support measures is delineated; and a Gaussian process prior model is constructed using historical data or initial sampling, transforming the complex support response black-box function into a probabilistically predictable surrogate model. The core Bayesian optimization loop then unfolds: the system calculates the posterior distribution of unknown parameter points based on the current dataset, mainly including the predicted mean and variance, and then uses the expected improvement function to quantify the potential improvement value of each candidate scheme; by maximizing this function, the optimal parameter combination that best balances exploration and utilization is intelligently selected; subsequently, the rapid surrogate model performs instantaneous effect, cost, and schedule predictions for this combination, and uses the evaluation results as new samples to update the dataset, thereby continuously correcting the prediction accuracy of the surrogate model. This iterative loop continues until convergence conditions are met, such as the expected improvement approaching zero, reaching the maximum number of iterations, or the objective function value falling below a preset threshold. At this point, the system outputs the globally optimal or satisfactory reinforcement scheme and drives the actuator to implement it. The actual support response after the implementation of the solution is monitored in real time and fed back to the system to update the agent model again, forming a complete adaptive closed loop of perception-decision-execution-verification. The core advantage of this process is that it transforms the traditional empirical and static support decision-making into an intelligent process that can continuously learn, dynamically optimize, and quantify uncertainty through data-driven Bayesian learning, ultimately achieving the best overall balance between economy and efficiency while ensuring safety.

[0053] Example 2 This embodiment provides a device for predicting and actively controlling the deformation of support structures based on computational learning, including: The parameter determination module is used to determine the corresponding soil layer parameters and support structure parameters of the area where the foundation pit is located; The parameter input module is used for inputting engineering information and soil layer parameters; The structural displacement calculation module is used for structural displacement calculation based on the autonomous theoretical model. The autonomous risk classification module is used for automatic identification and autonomous risk classification of hazardous areas. The intelligent reinforcement design and iterative verification module is used to determine the reinforcement parameter combination of deep mixing piles for each high-risk location identified by the autonomous classification module using a Bayesian optimization algorithm. It automatically adjusts the pile design parameters to meet the displacement control specifications and starts the next intelligent optimization iteration. This cycle continues until the displacement of the support structure meets the design specifications.

[0054] Example 3 This embodiment provides a computer storage medium storing a computer program that implements the above-described method for predicting and actively controlling the deformation of support structures based on computational learning.

[0055] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the technical solution of the present invention, and are not intended to limit the specific implementation of the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the claims of the present invention should be included within the protection scope of the claims of the present invention.

Claims

1. A method for predicting and actively controlling the deformation of support structures based on computational learning, characterized in that, Includes the following steps: S101: Determine the corresponding soil layer parameters and support structure parameters for the area where the foundation pit is located; S102: Input of engineering information and soil layer parameters; S103: Model-based structural displacement calculation; S104: Automatic identification and autonomous risk classification of hazardous areas; S105: For each high-risk location identified in S104, a Bayesian optimization algorithm is used to determine the combination of reinforcement parameters for the deep mixing piles, automatically adjust the pile design parameters to meet the displacement control specifications, and start the next intelligent optimization iteration. This process is repeated until the displacement of the support structure meets the design specifications.

2. The method for predicting and actively controlling the deformation of support structures based on computational learning according to claim 1, characterized in that, Step S101 includes: Behind the support structure are several layers of soil. The unit weight, cohesion, internal friction angle and deformation modulus parameters of each soil layer are determined. Based on the design data of the support structure, its thickness, penetration depth, moment of inertia of section and elastic modulus of material are determined. Combining the above parameters, the engineering geological stratification of the soil and the calculation segmentation of the support structure are completed.

3. The method for predicting and actively controlling the deformation of support structures based on computational learning according to claim 1, characterized in that, Step S102 includes: The engineering information and soil layer parameters input include the unit weight, cohesion, internal friction angle, and deformation modulus of each soil layer, as well as the thickness, embedment depth, moment of inertia, and elastic modulus of the support structure. The soil layers and the location of the internal supports of the support structure are used as dividing points to segment the support structure. The displacement curves of the support structure are calculated for each segment. Each segment is spliced ​​together using four parameters: displacement, rotation angle, bending moment, and shear force, to form a continuous support structure.

4. The method for predicting and actively controlling the deformation of support structures based on computational learning according to claim 1, characterized in that, Step S103 includes: The model-based structural displacement calculation introduces the layered earth pressure calculation formula for soil displacement, and determines the bending moment, shear force, and rotation angle of the support pile section through the calculation formula corresponding to the structural deflection angle effect under shear action. Combining the earth pressure calculation formula and the calculation formula corresponding to the structural deflection angle effect under shear action, and substituting them into the mechanical equilibrium formula of the support structure, a fourth-order variable coefficient non-homogeneous differential equation for the displacement above and below the excavation surface of the support structure is obtained. The fourth-order variable coefficient non-homogeneous differential equation for the displacement above and below the excavation surface of the support structure is expanded by a high-order power series to obtain a ten-term series expansion that meets the engineering requirement of error ≤5%. A closed analytical solution is established for the earth pressure considering structural displacement and the horizontal deformation of the support structure under bending-shear coupling in non-uniform strata, thus obtaining the analytical solution for the displacement of the support structure above and below the excavation surface.

5. The method for predicting and actively controlling the deformation of support structures based on computational learning according to claim 4, characterized in that, The fourth-order variable-coefficient nonhomogeneous differential equation for the displacement above the excavation surface of the support structure is: In the formula: ; ; ; ; The width for calculating earth pressure is the pile spacing. For the longitudinal bending stiffness of the support piles; For the longitudinal shear stiffness of the support piles; Let be the horizontal displacement of the i-th support pile; x is the depth at which the displacement of the support structure is calculated. The fourth-order variable-coefficient nonhomogeneous differential equation for the displacement below the excavation surface of the support structure is: Where: h is the excavation depth of the foundation pit under the calculation condition (m); The width for calculating soil reaction force; , ; represents the magnitude of the active zone earth pressure distribution below the excavation surface of the foundation pit; m is the proportional coefficient of the horizontal resistance coefficient of the foundation soil.

6. The method for predicting and actively controlling the deformation of support structures based on computational learning according to claim 4, characterized in that, The analytical solution for the displacement of the support structure above the excavation face is: In the formula: by ensuring that the displacement, rotation, bending moment, and shear force of each section of the support structure are equal during the specific solution process, the equilibrium equation is established to obtain the following: ; The analytical solution for the displacement of the support structure below the excavation face is: In the formula: ; ; These are the initial conditions; Let be the coefficient of earth pressure at rest for the j-th soil layer; Let be the active earth pressure coefficient of the j-th soil layer; The coefficient of static earth pressure below the excavation surface of the foundation pit; Let J be the unit weight of the j-th soil layer; Let be the unit weight of the i-th soil layer; Let be the cohesion of the j-th soil layer; j is the sum of the number of soil layers above the excavation face and the number of soil layers below the excavation face. This refers to the magnitude of the deflection caused by the stress on the support structure below the excavation face.

7. The method for predicting and actively controlling the deformation of support structures based on computational learning according to claim 1, characterized in that, Step S104 includes: The automatic identification of dangerous areas and autonomous risk classification are based on the displacement curve of the support structure during the excavation of the foundation pit calculated in step S103. The displacement of the support structure is automatically compared with the preset multi-level safety thresholds. According to the comparison results, the support structure is divided into three risk levels: safe state, early warning state and dangerous state. The sections with displacement exceeding the dangerous threshold are automatically identified as dangerous sections. A visual early warning map containing risk level and spatial location is generated, and a basis is provided for the active control decision in step S105.

8. The method for predicting and actively controlling the deformation of support structures based on computational learning according to claim 1, characterized in that, Step S105 includes: Based on step S104, for the identified warning and danger zones, the prior probability distribution is initialized, a multi-objective function is set, and the parameter space is defined; initial samples are collected and a Gaussian process prior model is constructed; an iterative loop is entered, the posterior distribution and the expected improvement function are calculated, and the next evaluation point is selected by optimizing the acquisition function; a surrogate model is used to predict performance and update the dataset; the loop terminates based on three judgment criteria: the optimization benefit is lower than a set threshold, the maximum number of iterations is reached, or the objective function value meets the specification requirements, and finally, the reinforcement parameters that balance safety and economy are automatically optimized and output.

9. A device for predicting and actively controlling the deformation of a support structure based on computational learning, characterized in that, include: The parameter determination module is used to determine the corresponding soil layer parameters and support structure parameters of the area where the foundation pit is located; The parameter input module is used for inputting engineering information and soil layer parameters; The structural displacement calculation module is used for structural displacement calculation based on the autonomous theoretical model. The autonomous risk classification module is used for automatic identification and autonomous risk classification of hazardous areas. The intelligent reinforcement design and iterative verification module is used to determine the reinforcement parameter combination of deep mixing piles for each high-risk location identified by the autonomous classification module using a Bayesian optimization algorithm. It automatically adjusts the pile design parameters to meet the displacement control specifications and starts the next intelligent optimization iteration. This cycle continues until the displacement of the support structure meets the design specifications.

10. A computer storage medium, characterized in that, The computer program stores the method for predicting and actively controlling the deformation of the support structure based on computational learning as described in any one of claims 1 to 8.

Images