Optimization design method and system of variable cross-section group pile

Abstract

The application discloses a variable cross-section group pile optimization design method and system, which is applied to the technical field of variable cross-section group piles and comprises the following steps: obtaining load data of each distribution position of a sample variable cross-section group pile; determining each parameter load data section; obtaining a load influence parameter set; obtaining a load mapping relationship based on the load influence parameter set and the load data; obtaining initial load prediction data based on the load mapping relationship and a to-be-optimized parameter data set of a target variable cross-section group pile; analyzing the to-be-optimized parameter data set and the load data, processing the initial load prediction data according to an analysis result, and obtaining a load prediction condition; and optimizing the to-be-optimized parameter data set based on the load prediction condition, and obtaining a target design parameter set of the target variable cross-section group pile. The variable cross-section group pile optimization design method and system can accurately optimize design parameters and improve the reliability of the design parameters.

Description

Technical Field

[0001] This invention relates to the field of variable cross-section pile group design technology, and in particular to an optimized design method and system for variable cross-section pile groups. Background Technology

[0002] Variable cross-section pile groups for transmission towers can adjust their cross-sectional dimensions at different depths to adapt to the special requirements of the tower's tall structure on the foundation's bearing capacity. A well-designed variable cross-section pile group scheme can accurately capture changes in the load transfer path at the variable cross-section, which is crucial for ensuring the safe operation of the transmission tower under multiple operating conditions.

[0003] Existing methods for optimizing the design parameters of variable cross-section pile groups for transmission towers are based on stress simulation analysis using design parameters and typical soil layer parameters, followed by processing of the analysis results to guide the optimization of design parameters. However, in reality, the actual stress of variable cross-section pile groups for transmission towers is affected by many factors, and the simulated stress analysis results cannot accurately reflect the actual stress situation, thus failing to effectively guide the optimization of design parameters for variable cross-section pile groups for transmission towers. Summary of the Invention

[0004] This invention provides an optimization design method and system for variable cross-section pile groups to solve the technical problem that when optimizing the design parameters of variable cross-section pile groups for transmission towers based on simulation stress analysis results, the analysis results cannot accurately reflect the actual stress situation, resulting in a lack of effective guidance for the optimization of design parameters. This invention aims to achieve accurate optimization of design parameters and improve the reliability of design parameters.

[0005] To address the aforementioned technical problems, this invention provides an optimized design method and system for variable cross-section pile groups, the method comprising: Obtain load data at various distribution locations of the sample variable cross-section pile group; Based on the pile geometry of the sample variable cross-section pile group, the distribution locations are processed to determine the initial parameter load data segments. Based on the load transfer characteristics of the transmission tower, the load data segments of each initial parameter are optimized to determine the load data segments of each parameter. Based on the load influence degree of each parameter in each of the aforementioned parameter load data segments, determine each load influence parameter and integrate them to obtain a load influence parameter set; Based on the load influence parameter set and the load data, the load mapping relationship is obtained; Based on the load mapping relationship and the obtained target variable cross-section pile group parameter dataset, the initial load prediction data is obtained. The dataset of parameters to be optimized and the load data are analyzed, and the analysis results are used to process the initial load prediction data to obtain the load prediction results. Based on the load prediction, the dataset of parameters to be optimized is optimized to obtain the target design parameter set of the target variable cross-section pile group.

[0006] Preferably, the process of processing each of the distribution locations based on the pile geometry of the sample variable cross-section pile group to determine each initial parameter load data segment includes: Analyze the geometric structure of the pile body to determine the rate of change of the pile body's outer contour; Based on the change rate of the outer contour of the pile body, each segment interval is determined; Obtain the parameter data corresponding to each of the segmented intervals; The distribution locations are divided based on the segmented intervals, and the corresponding segmented load data is extracted from the load data. Based on the parameter data and segmented load data corresponding to each segmented interval, each initial parameter load data segment is determined.

[0007] Preferably, the optimization of each initial parameter load data segment based on the load transfer characteristics of the transmission tower to determine each parameter load data segment includes: The design parameter data of the transmission tower are processed using finite element analysis technology to determine the load transfer characteristics of the transmission tower. The load transfer characteristics are analyzed to determine the load distribution. Based on the load distribution, the initial parameter load data segment is optimized to determine each parameter load data segment.

[0008] Preferably, the step of determining each load influence parameter based on the load influence degree of each parameter in each of the said parameter load data segments, and integrating them to obtain a load influence parameter set, includes: Extract each parameter from each of the aforementioned parameter load data segments; Analyze the changes in the segmented load data when each of the aforementioned parameters changes; Quantify the changes to obtain the degree of load influence for each parameter; Based on preset standards, the parameters corresponding to the degree of influence of each load are filtered to obtain the load influence parameters. By integrating all the load influence parameters, the load influence parameter set is obtained.

[0009] Preferably, obtaining the load mapping relationship based on the load influence parameter set and the load data includes: The load influence parameter set and the load data are input into a pre-built artificial intelligence model for processing to obtain the load mapping relationship.

[0010] Preferably, the step of analyzing the dataset of parameters to be optimized and the load data, and processing the initial load prediction data with the analysis results to obtain the load prediction results, includes: Based on the parameter data in the parameter load data segment and the load mapping relationship, the sample benchmark prediction value is obtained; Based on the sample baseline prediction values ​​and the load data, residual data are obtained; Extract the feature data from the parameter data; The feature data and the residual data are input into a pre-constructed multi-kernel learning model for training to obtain a load correction model. The dataset of parameters to be optimized is input into the load correction model for processing to obtain load correction data. The initial load prediction data is processed based on the load correction data to obtain the load prediction.

[0011] Preferably, the feature data extracted from the parameter data includes: The parameter data is divided to obtain geometric parameter data and soil layer parameter data; Extract the feature information from the geometric parameter data and the soil layer parameter data respectively; The feature data is obtained by integrating geometric feature information and soil layer feature information.

[0012] Preferably, the optimization of the parameter dataset to be optimized based on the load prediction to obtain the target design parameter set of the target variable cross-section pile group includes: Extract the design parameters from the design requirements; The predicted load is processed in conjunction with the design parameters to obtain the target to be optimized. The parameters to be optimized are determined from the dataset of parameters to be optimized based on the objective to be optimized. Using the target to be optimized as the optimization objective, an artificial intelligence algorithm is used to process the parameters to be optimized to obtain the target optimization parameters; Based on the target optimization parameters, the target design parameter set is obtained.

[0013] Preferably, obtaining the target design parameter set based on the target optimization parameters further includes: Based on the target design parameter set and the load mapping relationship, the target load condition is obtained; Analyze the target load and the target to be optimized, and adjust the parameters to be optimized based on the analysis results.

[0014] Another aspect of the present invention provides an optimized design system for variable cross-section pile groups, comprising: The acquisition module is used to acquire load data at various distribution locations of the sample variable cross-section pile group; The determination module is used to process each of the distribution locations based on the pile geometry of the sample variable cross-section pile group and determine each initial parameter load data segment. The optimization module is used to optimize each of the initial parameter load data segments based on the load transfer characteristics of the transmission tower, and to determine each parameter load data segment. The influence module is used to determine each load influence parameter based on the load influence degree of each parameter in each of the aforementioned load data segments, and integrate them to obtain a load influence parameter set. The mapping module is used to obtain the load mapping relationship based on the load influence parameter set and the load data; The initial module is used to obtain initial load prediction data based on the load mapping relationship and the obtained target variable cross-section pile group parameter dataset. The prediction module is used to analyze the dataset of parameters to be optimized and the load data, and to process the initial load prediction data with the analysis results to obtain the load prediction situation. The target module is used to optimize the dataset of parameters to be optimized based on the load prediction to obtain the target design parameter set of the target variable cross-section pile group.

[0015] Compared with the prior art, the beneficial effects of the present invention are at least one of the following: This invention acquires load data at various distribution locations of a sample variable cross-section pile group, processes these locations based on the pile geometry to determine initial parameter load data segments, optimizes these initial parameter load data segments based on the load transfer characteristics of the transmission tower, further selects load influence parameters based on the load influence degree of each parameter in each parameter load data segment, and integrates them into a load influence parameter set. Based on this, a load mapping relationship is established to predict the initial load of the target variable cross-section pile group's parameter dataset to be optimized. Furthermore, the initial load prediction data is processed by combining the analysis results of the parameter dataset to be optimized and the load data to obtain the load prediction situation. Finally, the target design parameter set is obtained by optimizing the parameter dataset to be optimized based on the load prediction situation. This invention abandons the traditional approach of solely relying on simulation stress analysis results to guide optimization. Instead, it starts from actual load data, extracts load influence parameters, and establishes a data-driven load mapping relationship. This allows design parameter optimization to be directly based on actual load predictions rather than distorted simulation stress results, effectively solving the technical problem of design parameter optimization failure caused by the inability of simulation stress analysis to reflect actual stress conditions. This invention improves the matching degree and reliability between optimized design parameters and actual bearing stress distribution, and significantly enhances the accuracy and engineering applicability of variable cross-section pile group design parameter optimization. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the optimized design method for variable cross-section pile groups in one embodiment of the present invention. Figure 2 This is a schematic diagram of a pile foundation parameter load data segment in one embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of the optimized design system for variable cross-section pile groups in one embodiment of the present invention; Figure label: Among them, 11. Acquisition module; 12. Determination module; 13. Optimization module; 14. Influence module; 15. Mapping module; 16. Initialization module; 17. Prediction module; 18. Target module. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0018] In the description of this invention, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," "third," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0019] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0020] In the description of this invention, it should be noted that, unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this specification is for the purpose of describing specific embodiments only and is not intended to limit the invention. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0021] Current methods for optimizing the design parameters of variable cross-section pile groups for transmission towers rely on simulation stress analysis results based on typical parameters to guide the optimization. However, because actual stress is affected by a variety of complex factors, simulation results are often not accurate enough and cannot effectively guide design optimization.

[0022] One embodiment of the present invention provides an optimized design method for variable cross-section pile groups. For details, please refer to [link to relevant documentation]. Figure 1 , Figure 1 The diagram shown is a flowchart illustrating an optimized design method for a group of variable cross-section piles according to one embodiment of the present invention, including: S1. Obtain the load data at each distribution location of the sample variable cross-section pile group; S2. Based on the pile geometry of the sample variable cross-section pile group, process the distribution locations to determine the initial parameter load data segments. S3. Based on the load transfer characteristics of the transmission tower, optimize the load data segments of each initial parameter and determine the load data segments of each parameter. S4. Based on the load influence degree of each parameter in each load data segment, determine each load influence parameter and integrate them to obtain the load influence parameter set. S5. Based on the load influence parameter set and load data, obtain the load mapping relationship; S6. Based on the load mapping relationship and the obtained target variable cross-section pile group parameter dataset, the initial load prediction data is obtained. S7. Analyze the dataset of parameters to be optimized and the load data, and use the analysis results to process the initial load prediction data to obtain the load prediction results. S8. Based on the load prediction, optimize the dataset of parameters to be optimized to obtain the target design parameter set of the target variable cross-section pile group.

[0023] First, load data is obtained at various distribution locations of the sample variable cross-section pile group. The core of step S1 is to obtain corresponding load data, including axial force, bending moment, shear force, and pile side friction, from the known sample variable cross-section pile group (a pile foundation composed of multiple piles with varying cross-sectional dimensions along its length) at different spatial distribution locations, such as pile tops, variable cross-section points, different pile depths, and different pile positions. This provides a fundamental data source for subsequent parameter segmentation, optimization, and predictive modeling. Sample variable cross-section pile group: refers to a known case used for analysis or modeling, consisting of multiple piles with cross-sectional dimensions (such as diameter or width) varying along the length. Various distribution locations refer to key measuring points selected spatially within the pile group, including pile tops, variable cross-section junctions, piles at different depths, different pile positions (such as corner piles or center piles), and pile ends. Load data refers to the mechanical parameters acting on or transmitted to each distribution location, commonly including axial force (pressure or tension along the pile), bending moment (moment causing pile bending), lateral shear force, and pile side friction and pile end resistance.

[0024] Obtaining load data at various distribution locations requires non-destructive or minimal-destructive indirect testing methods. First, surface strain gauges or resistance strain gauges are placed on the ground surface or exposed sections of the pile side at the connection between the pile top and the pile cap, and at locations with variable cross-sections, such as by excavating test pits or utilizing existing testing holes. Simultaneously, fiber optic sensors or sliding micrometers are implanted into the sides of key pile locations, such as corner piles and edge piles, through drilling to obtain the strain distribution along the pile depth. Then, a load transfer monitoring system based on the actual operating conditions of the transmission tower, such as tower foot pressure sensors or inclinometers, records the magnitude and direction of the load transferred from the superstructure to the pile cap. This is further supported by embedding earth pressure cells or pore water pressure gauges in the soil around the pile, using inclined holes or shallow embedding to obtain changes in soil pressure along the pile side. Finally, using the finite element inverse analysis method, with the measured strain as the boundary condition, the axial force, bending moment, and pile side friction at each depth are calculated using the elastic modulus of the pile material and the geometric parameters of the variable cross-section, thus obtaining the load data for each distribution location under the conditions of actual use. The entire process requires strict control of drilling depth and location to avoid significant damage to the pile's bearing capacity, and repeated measurements are necessary to eliminate errors caused by fluctuations in operating loads. Obtaining sample load data provides reliable training samples for subsequent analysis and is the fundamental data source for the entire technical approach.

[0025] Secondly, based on the pile geometry of the sample variable cross-section pile group, the distribution locations are processed to determine the initial parameter load data segments. The pile geometry is analyzed to determine the rate of change of the pile's outer contour; based on this rate of change, each segment interval is determined; the parameter data corresponding to each segment interval is obtained; the distribution locations are divided based on the segment intervals, and the corresponding segmented load data is extracted from the load data; based on the parameter data and segmented load data corresponding to each segment interval, each initial parameter load data segment is determined.

[0026] The geometric structure of the variable cross-section pile group refers to the variation of the pile diameter or width along the depth. The rate of change of the pile's outer contour is the speed at which the cross-sectional dimensions change with length, used to identify key locations of abrupt or gradual changes in the cross-section. Segmented intervals are several continuous depth ranges artificially divided from the entire pile based on this rate of change. The parameter data corresponding to each segmented interval includes geometric and material parameters such as the average cross-sectional dimensions, pile length, and concrete strength within that segment. The distribution locations are the specific measuring points of the sensors embedded in step S1. The segmented load data are the measured values ​​of axial force, bending moment, etc., extracted from the original load data and falling within each segmented interval. Finally, the initial parameter load data segment is a data unit formed by associating the parameter data of each segmented interval with its corresponding load data, serving as the basic input for subsequent optimization analysis.

[0027] Collect design drawings or conduct on-site measurements for each pile in the sample variable cross-section pile group. Record the cross-sectional dimensions at different depths, with the pile top as the zero depth point. For example, measure the pile diameter or side length every 0.5 meters. For locations with abrupt changes in cross-section, the depth point must be marked separately. Organize these discrete measurement points into a table corresponding to depth and cross-sectional dimensions, and further fit them to a function that varies along the depth, such as a piecewise constant function to describe the stepped variable cross-section, or a linear function to describe the tapered pile.

[0028] The rate of change can be defined as the absolute value of the change in cross-sectional diameter between adjacent depths divided by the change in depth, or as the percentage change in cross-sectional area. An engineering threshold is set; for example, a significant change in geometry is considered to have occurred when the diameter change exceeds 0.1 meters per meter or the area change exceeds 15%. Starting from the top of the pile, a point-by-point scan is performed. Whenever the rate of change exceeds the set threshold, a segment point is set at that location. The continuous depth range between two adjacent segment points constitutes a segmented interval. In this way, the entire pile is divided into several intervals with relatively uniform internal geometry.

[0029] For each defined segment, a set of parameters representing the geometric and material properties of that segment is extracted, including but not limited to: average diameter, average cross-sectional area, moment of inertia, segment length, concrete elastic modulus, and reinforcement ratio. The elastic modulus can be obtained from the design value or measured on-site using methods such as a rebound hammer, while the reinforcement ratio is determined based on the construction drawings. These parameters are then organized into a vector form, with one parameter vector corresponding to each segment.

[0030] The parameters are mainly obtained through two methods: one is to read or calculate them directly from the design drawings, such as the starting and ending depths of the segments, the segment length, the design diameter at each depth, the design values ​​of the concrete strength grade and elastic modulus, and the reinforcement ratio; the other is to obtain them through on-site measurements when the drawings are missing or the actual state after construction and use needs to be considered, including using borehole imaging or laser scanning to measure the external dimensions of the pile at different depths, using a rebound hammer or ultrasonic rebound combined method to detect the elastic modulus and strength of the concrete, and using a rebar detector to scan and determine the actual reinforcement ratio.

[0031] Organize the distribution locations and their depth coordinates obtained from the sensor placement or measurements in step S1. Then, based on the specific depth of each location, determine which segment interval it falls into and label each location with its respective segment interval. For example, a location at a depth of 1.2 meters might belong to the first segment interval, while a location at a depth of 2.5 meters might belong to the second segment interval.

[0032] Based on the grouping results from the previous step, all load records belonging to the same segment interval are selected from the original load dataset obtained in step S1. These load data typically include axial force, bending moment, shear force, and pile side skin friction. If there are multiple measuring points within the same segment interval, all data are retained for more refined analysis. Finally, the segmented load data vector corresponding to each segment interval is obtained.

[0033] The parameter data for each segmented interval is matched one-to-one with the segmented load data for that interval to form a complete data unit. This data unit includes the depth range, geometric parameters, material parameters, and various load data measured within that interval. After organizing all segmented intervals in this format, a complete set of initial parameter load data segments is obtained, serving as the basic input for subsequent optimization analysis. Throughout the process, spreadsheet software or a database system is used to store and manage this data, and the processing parameters at each step are recorded to ensure that the entire process can be reproduced and traced back. Segmenting the load according to the pile geometry discretizes the continuous load, facilitating the establishment of local correspondences between parameters and loads.

[0034] Then, based on the load transfer characteristics of the transmission tower, the load data segments of each initial parameter are optimized to determine the load data segments of each parameter. Finite element analysis technology is used to process the design parameter data of the transmission tower to determine its load transfer characteristics; the load transfer characteristics are analyzed to determine the load distribution; and based on the load distribution, the initial parameter load data segments are optimized to determine the load data segments of each parameter.

[0035] The load transfer characteristics of a transmission tower guide the path and distribution of external forces such as conductor tension, wind load, and tower self-weight through the tower structure to the pile group. Finite element analysis (FEM) is a numerical calculation method that discretizes the tower into elements and solves for their internal forces and deformations. It requires inputting the cross-sectional dimensions, material properties, and boundary constraints of each tower member. The design parameters of the transmission tower include geometric and mechanical inputs such as tower height, tower type, member specifications, and connection stiffness. Load distribution refers to the magnitude and proportion of the pile top reaction force on each pile in the pile group, obtained after finite element calculation. Finally, each parameter load data segment refers to a data segment that better matches the actual stress behavior of the transmission tower by modifying the initial parameter load data segment's segmentation or load values ​​based on the load distribution (e.g., adjusting the boundaries of the segments originally divided according to the geometric rate of change to positions that better reflect the actual stress gradient, or adjusting the load values ​​borne by each segment according to the actual distribution ratio).

[0036] First, obtain the design parameters of the transmission tower from the design drawings, including the tower type, tower height, cross-sectional dimensions and material properties of each member, and the connection method between the tower feet and the foundation. Input this data into finite element software to create a three-dimensional tower system model. Apply typical load combinations to the model, including vertical gravity of the conductors and ground wires, lateral wind loads, longitudinal unbalanced tensions, and icing loads. After solving, extract the forces transmitted from the tower feet to the foundation in three directions: vertical force, horizontal shear force, and bending moment, as well as the distribution ratio of these forces among the tower feet. Obtain the load transfer characteristics of the transmission tower, specifically represented by the reaction time history or extreme value distribution table of each tower foot.

[0037] The reaction force at the base of the pile cap obtained in the first step is used as a boundary condition and applied to the pile cap. Based on the assumption of pile cap rigidity, the pile cap deformation is much smaller than the pile foundation deformation. The total load on the bottom surface of the pile cap is distributed to the top of each pile using static equilibrium equations. The distribution method typically employs the rigid pile cap method on elastic foundations: first, the centroid of the pile group is calculated; then, based on the vertical resultant force, horizontal resultant force, and bending moment acting on the pile cap, the vertical force, horizontal force, and bending moment at the top of each pile are calculated separately. For example, the vertical resultant force is linearly distributed according to the distance of each pile from the centroid, and the horizontal resultant force is usually assumed to be uniformly distributed. The distance from the centroid refers to the distance between the center point of each pile in the pile group and the centroid (geometric center) of the entire pile cap bottom surface. The bending moment is distributed according to the stiffness and position of each pile. Finally, the load vector at the top of each pile is obtained, including axial force, shear force, and bending moment. If the geometric parameters of piles at different locations in the pile group are different, a stiffness-based adjustment distribution is also required. The load distribution is obtained, that is, the specific load value at the top of each pile.

[0038] For each pile, based on its variable cross-section geometry and the constraints of the soil around the pile, the load transfer method (tz curve, py curve, qz curve, or finite beam element method) is used to progressively calculate the load at the pile top downwards along the depth, obtaining the axial force, bending moment, and shear force distribution at different depths. Layered parameters of the soil around the pile are collected, such as the thickness, unit weight, internal friction angle, cohesion, and compression modulus of each layer, and a pile-soil interaction model is established. After calculation, the load values ​​at each depth are extracted. At variable cross-section locations, due to abrupt changes in cross-sectional stiffness, bending moment and shear force may exhibit discontinuities or peak variations, requiring element refinement during calculation. The output of this step is the load data for each pile at each point along the depth, i.e., the load distribution curve along the pile shaft.

[0039] Establishing pile-soil interaction models is a mature technology in geotechnical engineering, mainly divided into three categories: The first category is the load transfer method, also known as the Tz curve method, Py curve method, or QZ curve method. This method simplifies the soil along the pile and at the pile tip as a series of independent nonlinear springs, describing the interaction by establishing empirical relationships between pile displacement and pile side resistance and pile tip resistance. For example, for horizontally loaded piles, the Py curve method, recommended in the design and analysis of offshore fixed platform structures, is currently the most commonly used method, describing the relationship between pile side horizontal resistance and displacement using a hyperbolic model; for vertically loaded piles, the Tz curve is used to describe the relationship between pile side friction and relative displacement. The second category is the finite element numerical simulation method, which uses commercial software such as ANSYS to establish a three-dimensional continuous medium model, discretizing both the pile and soil into solid elements, setting contact pairs such as the Coulomb friction model or bond damage model at the pile-soil interface, and using appropriate soil constitutive models such as the Mohr-Coulomb model, modified Cambridge model, or cyclic boundary surface model to describe the nonlinear, plastic, and cyclic weakening characteristics of the soil. The third category is analytical / semi-analytical methods, such as the elastic theory method, shear displacement method, or virtual soil pile method. These methods derive and solve the governing equations of pile-soil interaction through theoretical derivation. In practical engineering applications, a technical approach that primarily uses the load transfer method and secondarily uses the finite element method is typically adopted. First, the finite element method is used to perform detailed analysis on a small number of typical working conditions and calibrate the load transfer parameters. Then, the calibrated load transfer model is extended to the rapid calculation of pile groups or complex working conditions.

[0040] The obtained load distribution along the pile is compared segment by segment with the initial parameter load data segment obtained in step S2. The initial parameter load data segment is based on the division and assignment of intervals according to pure geometric structure, while the newly calculated load distribution reflects the actual mechanical transmission law. Check the following differences: First, there is the difference in the magnitude of the load values ​​themselves, such as whether the average axial force in the geometric segment is close to the mechanical calculation value; Second, whether the boundary positions of the segmented intervals are reasonable. For example, in mechanical calculations, the extreme point of bending moment appears at a certain depth, but the geometric segmentation does not set a boundary at that point. Third, in certain sections where geometric changes are not obvious, the load gradient may be very large, such as at the junction of soft and hard soil layers, which requires further subdivision.

[0041] Based on these comparison results, optimization schemes are formulated: adjusting the start and end depths of the segmented intervals, merging or splitting certain intervals, and correcting the representative load value of each interval, such as replacing the original simple average with the integral average value of the interval obtained by mechanical calculation.

[0042] According to the established optimization scheme, the initial parameter load data segments are modified. Optimized segment interval numbers are added to tables or databases, the depth range of each segment is modified, and the average geometric parameters within each segment interval are recalculated. Because the depth range has changed, the geometric parameters need to be extracted again from the original geometric data, and the original load data is replaced with the average or characteristic load values ​​within that depth range obtained from mechanical calculations. If the load changes drastically within a segment interval, such as near a variable cross-section, it can be further subdivided into two sub-intervals, each assigned a different load value. After all modifications are completed, a new dataset is formed, resulting in the parameter load data segments. Compared to before optimization, the parameter load data segments better reflect the actual mechanical behavior of the transmission tower load transmitted through the pile group, providing more reliable input for subsequent parameter influence analysis and predictive modeling. Optimizing the data segments in conjunction with the actual load transmission characteristics of the transmission tower makes the segmentation results more consistent with the real stress mechanism, improving data quality.

[0043] This invention also provides a schematic diagram of pile foundation parameter load data segments, such as... Figure 2 As shown, Figure 2The diagram illustrates a pile foundation parameter load data segment in one embodiment of the present invention. Structure 1 represents a constant cross-section segment where the pile diameter remains constant along the depth; the geometric parameters in the corresponding parameter load data segment are constant, and the load exhibits a uniform or linearly gradually varying distribution. Structure 2 represents a gradually varying cross-section segment where the pile diameter changes continuously with depth; it needs to be subdivided into multiple sub-intervals based on the rate of change, with the average geometric parameters taken within each sub-interval; the load exhibits a non-linear distribution. Structure 3 represents a sudden change cross-section segment where the diameter undergoes a step change at a certain depth; this abrupt change point serves as the segment boundary, with independent data segments established on both sides. Due to the abrupt change in stiffness, bending moment and shear force exhibit discontinuities or peak values ​​at this location. By decomposing the actual pile geometry into combinations of these three basic types, it is possible to divide the pile from the top downwards into several initial parameter load data segments with uniform internal geometry. Based on this, the initial data segments are optimized by considering the load transfer characteristics of the transmission tower: Finite element analysis is used to process the tower's design parameters (tower type, member cross-section, connection method, etc.) to determine the vertical, horizontal, and bending moment distributions transmitted from the tower base to the pile cap. Then, the rigid pile cap method is used to distribute these loads to the top of each pile, and the actual load distribution curve is calculated downwards along the pile. The calculated load distribution is compared segment by segment with the initial segments to check if the segment boundaries match the extreme bending moment points or load gradient changes in the mechanical calculations. Based on this, the start and end depths of the segment intervals are adjusted, and intervals are merged or split. The representative load values ​​of each segment are corrected using the integral average value obtained from the mechanical calculations, ultimately resulting in parametric load data segments that better match the actual stress behavior of the transmission tower. The formula below the figure further illustrates that this segmentation and optimization process is driven by both the pile body's external profile change rate and load transfer characteristics, ensuring that each data segment has a single internal geometry and a clear load distribution pattern.

[0044] Furthermore, based on the load influence degree of each parameter in each load data segment, the load influence parameters are determined and integrated to obtain a load influence parameter set. This involves extracting each parameter from each load data segment; analyzing the changes in segmented load data when each parameter changes; quantifying the changes to obtain the load influence degree of each parameter; filtering parameters corresponding to each load influence degree based on preset standards to obtain individual load influence parameters; and integrating all load influence parameters to obtain a load influence parameter set.

[0045] The parameter load data segment comprises the geometric and material parameters and their corresponding load data within each segment interval obtained after optimization in step S3. Each parameter refers to one of these parameters, such as diameter. Segment load data refers to the actual measured load values ​​within that segment interval. Load variation refers to the amount or trend of change in the segment load data after a parameter change. Load influence degree refers to the magnitude of the parameter's influence on the load; a larger value indicates greater importance. Load influence parameters are those parameters deemed to have a significant impact on the load after screening. The load influence parameter set is a collection of these screened parameters, used for subsequent model simplification or focused optimization.

[0046] To analyze the influence of each parameter on the load, the parameter values ​​are changed, and the load changes accordingly. Two common methods are: First, the single-parameter perturbation method, which changes only one parameter at a time while keeping others constant. The change range is typically ±5%, ±10%, or ±20% of the original value; for example, calculating the load when the average diameter increases by 10% and decreases by 10%. Second, the dimensionless sensitivity analysis method, which calculates the partial derivatives after normalizing the parameters. This embodiment of the invention uses the single-parameter perturbation method. The choice of the change range needs to be determined based on the actual variation range of the parameters. For example, the elastic modulus of concrete may fluctuate by ±15% in actual engineering, while the cross-sectional dimensions may have a processing error of ±5%.

[0047] This invention provides detailed descriptions of two embodiments to analyze the changes in segmented load data when each parameter changes, as follows: Example 1 establishes a mapping relationship based on empirical formulas from pile foundation specifications. Geometric parameters for each segment, such as average diameter, segment length, pile tip cross-sectional area, and soil parameters, such as the standard values ​​of side skin friction and end resistance for each soil layer, are collected. Then, the formula for the vertical bearing capacity of a single pile in the "Technical Code for Building Pile Foundations" is used. The side skin friction of the variable cross-section pile is accumulated segment by segment, then the pile tip resistance is added, and finally, the result is divided by a safety factor to obtain the characteristic value of the bearing capacity. This calculation process is then written into a callable function, such as an Excel formula or a Python function. By inputting the diameter and length of any segment, the function can quickly output the corresponding vertical load value without running any simulation software.

[0048] Example 2 establishes a mapping relationship between finite element analysis (FEM) calculations and a response surface surrogate model. First, a parameterized variable cross-section pile-soil interaction model is established in Abaqus or ANSYS, defining key parameters such as segment diameter and elastic modulus as variables. Then, 50-100 parameter combinations are designed using Latin hypercube sampling. FEM calculations are run in batches via scripts, extracting segment load data (such as axial force and bending moment) for each parameter group. Finally, a second-order polynomial response surface or Gaussian process regression model is trained using this input-output data. After verifying the accuracy, the surrogate model is used as the mapping relationship. Subsequently, by inputting any new set of parameters, the surrogate model can return predicted load values ​​in milliseconds, avoiding repeated, time-consuming FEM calculations.

[0049] Then, for each parameter in the parameter list, follow these steps: First, record the original parameter value and the corresponding original load value; then, increase the parameter by one times the original value plus a factor of change, keeping all other parameters unchanged, and calculate the new load value using the two examples above; then, decrease the parameter by one times the original value minus a factor of change, again keeping other parameters unchanged, and calculate another new load value. Next, calculate the absolute value of the difference between the load value after increasing the parameter and the original load value, and the absolute value of the difference between the load value after decreasing the parameter and the original load value. Take the average of these two absolute values ​​as the average load change caused by the parameter. For example, for the parameter of average diameter, the original value is 1.0m, and the original axial force is 800kN; after increasing by 10%, the diameter is 1.1m, and the axial force becomes 850kN; after decreasing by 10%, the diameter is 0.9m, and the axial force becomes 750kN; then the average load change is (|850-800|+|750-800|) / 2=50kN.

[0050] The obtained average load variation is dimensionless to allow for comparison between different parameters. A commonly used quantification index is the sensitivity coefficient, which is the relative rate of change of load divided by the relative rate of change of parameter. In the above embodiment, the change in the original load value of 50kN divided by the original load value of 800kN is 0.0625, and the change in the parameter value of 0.1m divided by the original parameter value of 1.0m is 0.1. Therefore, the sensitivity coefficient is equal to 0.0625 / 0.1 = 0.625.

[0051] A pre-defined screening threshold is established, which can be determined based on engineering experience or statistical patterns. Commonly used standards include: parameters with an absolute sensitivity coefficient greater than 0.3 are considered high-impact parameters, those greater than 0.1 but less than or equal to 0.3 are considered medium-impact parameters, and those less than or equal to 0.1 are considered low-impact parameters. Alternatively, a cumulative contribution rate method can be used. All parameters' sensitivity coefficients are sorted from largest to smallest absolute value, and the cumulative contribution rate is calculated by dividing the sum of the absolute sensitivity values ​​of the current parameter by the sum of the absolute sensitivity values ​​of all parameters. The top few parameters with a cumulative contribution rate reaching 80% or 90% are selected as key parameters. The parameters retained after screening are recorded to obtain the load influence parameters.

[0052] All selected load influence parameters are integrated into a set, resulting in a load influence parameter set. This set can be a list of parameter names or a data structure containing more information, such as parameter names, sensitivity coefficients, recommended value ranges, and optimization priorities. In subsequent steps, only the parameters in this set need to be considered, while other parameters with less influence are ignored, thus significantly simplifying the complexity of the optimization problem. Simultaneously, the complete analysis process, including the original data, perturbation amplitude, sensitivity coefficients, and screening thresholds, is recorded in a separate document for re-screening when adjusting the thresholds later. Screening load influence parameters eliminates secondary variables, reduces model complexity, and focuses on key factors to improve the efficiency of subsequent modeling.

[0053] Next, based on the load influence parameter set and load data, the load mapping relationship is obtained. The load influence parameter set and load data are then input into a pre-built artificial intelligence model for processing to obtain the load mapping relationship. The load influence parameter set is the collection of key parameters that significantly affect load changes, selected in the previous step. The load data are the corresponding measured or calculated load values. The pre-built artificial intelligence model refers to a machine learning model with a pre-designed structure but not yet trained, such as a neural network, support vector machine, or random forest consisting of an input layer, hidden layers, and an output layer. The final load mapping relationship is a trained function or model that can quickly predict the corresponding load value based on any set of load influence parameters; it can replace time-consuming physical experiments or numerical simulations.

[0054] The two types of data obtained in the previous steps are paired and organized: one type is the parameter values ​​in the load influence parameter set, such as average diameter, concrete elastic modulus, reinforcement ratio, etc., which serve as input features of the model; the other type is the corresponding load data, such as axial force, bending moment, or shear force values, which serve as output labels of the model. Each set of parameters and its corresponding load data constitutes a sample. All samples are organized into a table, where rows represent different sample segments, different pile locations, or different loading conditions, and columns include each input parameter and output load. Missing or outlier values ​​are checked, and deletion or imputation is performed as necessary.

[0055] To avoid model overfitting and objectively assess its generalization ability, the dataset needs to be randomly divided into three subsets: a training set (typically 70% of the total data) for training the model, a validation set (typically 15% of the total data) for tuning model hyperparameters, and a test set (typically 15% of the total data) for final model performance evaluation. Stratified sampling can be used to ensure that samples within different load ranges are representative in each subset. In Python, random partitioning functions from machine learning libraries can be used, with a random seed set to ensure reproducible results.

[0056] For load mapping problems, this embodiment of the invention uses a feedforward neural network, capable of fitting arbitrarily complex nonlinear relationships. The number of nodes in the input layer equals the number of load-influencing parameters; for example, three parameters require three input nodes. One or two hidden layers can be set, each containing five to ten neurons, with a linear rectified function as the activation function. The number of nodes in the output layer equals the number of load types to be predicted; for example, predicting only axial force requires only one node. The output layer either does not use an activation function or uses a linear activation function. The model is constructed by adding fully connected layers sequentially within a deep learning framework.

[0057] The input parameters of the training set and the corresponding load data are fed into the model for training. For regression problems, the mean squared error is typically used as the loss function, which calculates the average of the squares of the differences between the predicted and actual loads. An optimizer, such as an adaptive moment estimator, is specified, which automatically adjusts the learning rate to accelerate convergence. The number of training epochs is set; during each training epoch, all training data is forward-propagated to calculate the loss, and then back-propagated to update the network weights. After each epoch, the performance of the current model is evaluated using a validation set. If the validation loss does not decrease for several consecutive epochs, training is stopped early to prevent overfitting. After training, the weights and biases within the model are determined.

[0058] The trained model is then evaluated using a pre-reserved test set. The input parameters from the test set are substituted into the model to obtain predicted load values, which are then compared with the actual load values ​​in the test set. Common evaluation metrics include: Mean Absolute Percentage Error (MAP), which is the average of the absolute values ​​of the prediction errors divided by the actual values; and the Coefficient of Determination (COD), which measures how much data variation the model explains. Generally, a MAP of less than 10% and a COD greater than 0.9 are considered sufficient for engineering applications. If the accuracy is insufficient, attempts can be made to increase the number of hidden layer neurons, increase the training data, or change the model type.

[0059] The trained and validated model constitutes the load mapping relationship. This model is saved as a file, for example, in Python as a file containing the network structure and weight parameters. When needed, a function to load the model is simply written, taking any set of load influence parameter values ​​as input. The model will then return the predicted load values ​​within milliseconds. This mapping relationship can replace finite element calculations or empirical formula lookups in subsequent steps, significantly improving the efficiency of optimization iterations. Simultaneously, all parameters of the model training, including the random seed for data partitioning, network structure, and training epochs, are recorded in a document. Establishing the load mapping relationship enables rapid prediction from parameters to loads, replacing time-consuming physical experiments or numerical simulations.

[0060] Furthermore, based on the load mapping relationship and the obtained dataset of parameters to be optimized for the target variable cross-section pile group, initial load prediction data is obtained. The load mapping relationship is a function obtained through training the artificial intelligence model in the previous stage, which can quickly output the corresponding load value according to the input parameters. The target variable cross-section pile group refers to the actual engineering pile foundation that needs to be designed or optimized. The dataset of parameters to be optimized refers to the design parameters of the target variable cross-section pile group, such as the average diameter of each segment and the elastic modulus of concrete. These parameters are the objects of subsequent optimization and adjustment. The initial load prediction data is the preliminary load value calculated by substituting the parameters to be optimized into the load mapping relationship. It represents the stress response predicted by the model under the current parameter combination and serves as the benchmark for subsequent analysis and optimization.

[0061] The load mapping relationship is a function obtained through training the artificial intelligence model in step S5, and is usually saved as a file. Based on the tools used previously, write code to load the model. For example, in Python, if a neural network model is used, the relevant function can be called to load the file containing the network structure and weight parameters; if a random forest model is used, the saved model object will be loaded. After successful loading, the model will reside in memory and provide a prediction function that accepts an input parameter vector and returns the predicted load values. Ensure that the model file path is correct and that the order of the input parameters is exactly the same as during training.

[0062] Extract values ​​from the dataset of parameters to be optimized from the first step, following the feature order during training. For example, if the order of input features to the model during training was average diameter, concrete elastic modulus, and reinforcement ratio, then these three values ​​need to be extracted sequentially from the dataset of parameters to be optimized, forming a list or array. If some parameters are not provided in the dataset of parameters to be optimized, for example, if they are fixed as constants, then these constant values ​​need to be added.

[0063] The generated input array is passed to the loaded load mapping relationship to perform a forward calculation. Internally, the model calculates layer by layer from the input layer to the output layer based on the trained weights and biases, ultimately outputting one or more load values. For example, the model might output a predicted axial force of 850 kN at the top of the target variable cross-section pile group. This output is the initial load prediction data. By using the mapping relationship to make a preliminary prediction of the target pile, an initial load estimate can be quickly obtained, providing a benchmark for subsequent corrections.

[0064] Preferably, the dataset of parameters to be optimized and the load data are analyzed, and the analysis results are used to process the initial load prediction data to obtain the load prediction. Based on the mapping relationship between parameter data and load in the parameter load data segment, the sample benchmark prediction value is obtained; based on the sample benchmark prediction value and load data, the residual data is obtained; the feature data of the parameter data is extracted; the feature data and residual data are input into a pre-constructed multi-kernel learning model for training to obtain the load correction model; the dataset of parameters to be optimized is input into the load correction model for processing to obtain the load correction data; based on the load correction data, the initial load prediction data is processed to obtain the load prediction. Further, the parameter data is divided to obtain geometric parameter data and soil layer parameter data; the feature information of the geometric parameter data and soil layer parameter data is extracted respectively; the geometric feature information and soil layer feature information are integrated to obtain feature data.

[0065] The parameter load data segment is the dataset obtained after optimization in step S3, containing the geometric and material parameters of each segment and their corresponding load data. The sample baseline predicted value is the predicted load obtained by substituting these parameter data into the load mapping relationship established in step S5. The residual data is the difference between the sample baseline predicted value and the measured or actual load data, representing the prediction error of the mapping relationship. The feature data of the parameter data are numerical values ​​extracted from the original parameters to describe the characteristics of the sample, including geometric features such as cross-sectional dimensions and moment of inertia, and soil layer features such as the thickness of each soil layer and compression modulus. These features are obtained through partitioning, i.e., extracted from geometric parameters and soil layer parameters respectively and then integrated. The multi-kernel learning model is a machine learning model that can combine multiple kernel functions, such as linear kernels, polynomial kernels, and radial basis kernels, to capture complex nonlinear relationships in the data, suitable for learning the correlation between residuals and features. The load correction model is a function obtained after training that can predict residual values ​​based on input features. The parameter dataset to be optimized contains the parameter values ​​to be optimized for the target variable cross-section pile group. Load correction data refers to the predicted residual values ​​obtained after inputting the dataset of parameters to be optimized into the load correction model. Initial load prediction data refers to the preliminary predicted loads obtained in step S6 through the load mapping relationship. The final load prediction result is obtained by adding or superimposing the initial load prediction data and the load correction data; it is closer to the actual stress state than the initial prediction.

[0066] The parameter data from the parameter load data segment obtained in step S3 are input group by group into the load mapping relationship model established in step S5. The model will output a predicted load value for each sample, thus obtaining the sample baseline predicted value. For example, for a sample pile, its actual axial force is 800kN, while the model predicts a baseline value of 780kN. This operation is performed on all samples to obtain a set of predicted value sequences.

[0067] The baseline predicted value for each sample is compared group by group with the actual load data (e.g., 800 kN) directly read from the parameter load data segment. The residual is calculated as: Residual = Actual Load Value - Baseline Predicted Value. If the actual value is 800 kN and the predicted value is 780 kN, the residual is +20 kN; conversely, if the predicted value is 820 kN, the residual is -20 kN. After calculating for all samples, a set of residual data is obtained, with each residual corresponding to one sample. These residuals represent the prediction error of the basic mapping model on that sample. The goal of subsequent steps is to build a model to predict these residuals.

[0068] The parameter data for each sample is extracted from the parametric load data segment and divided into two categories: one is geometric parameter data, including the average diameter, segment length, moment of inertia, and depth of the variable cross-section for each segment of the pile; the other is soil parameter data, including the thickness of each soil layer around the pile, standard values ​​of side skin friction, standard values ​​of end resistance, compression modulus, and internal friction angle. For the geometric parameter data, extracted features may include: the coefficient of variation of segment diameters, the rate of change of diameter at variable cross-sections, and the slenderness ratio of the pile; for the soil parameter data, extracted features may include: the weighted average thickness of each soil layer, the compression modulus of the bearing layer at the pile tip, and the proportion of soft soil layers. These features need to be quantified into numerical forms; for example, the coefficient of variation of diameter is equal to the standard deviation of diameter divided by the average diameter.

[0069] All features extracted from the geometric parameter data and all features extracted from the soil layer parameter data are combined to form a unified feature vector. For example, the feature vector of a sample might be: diameter coefficient of variation of 0.12, slenderness ratio of 30, and soft soil layer proportion of 0.25. Each sample corresponds to a feature vector, and the feature vectors of all samples constitute a feature matrix. Simultaneously, each sample corresponds to a calculated residual value, which serves as the prediction target. The feature matrix and residual vectors are paired to form a dataset used to train and refine the model.

[0070] Multi-kernel learning is a machine learning method that automatically combines multiple kernel functions. A multi-kernel learning framework is chosen, such as multi-kernel learning based on support vector regression or implementations based on generalized multi-kernel learning. The obtained feature data is used as input, and the residual data as output, and fed into the model for training. The model learns the non-linear mapping relationship between features and residuals, automatically determining the weights of each kernel function. After training, a load correction model is obtained, which can predict the corresponding residual value based on any set of input features. Cross-validation is used during training to adjust model parameters, such as the type of kernel function and regularization coefficients, to ensure prediction accuracy.

[0071] Geometric and soil layer features are extracted from the dataset of parameters to be optimized and integrated into a feature vector. This feature vector is then input into a trained load correction model, which outputs a numerical value, known as the load correction data. This correction data represents the potential deviation in the predicted value of the foundation load mapping relationship for the current target pile. For example, a correction data of -50kN means that the predicted foundation value may be 15kN higher.

[0072] Extract the initial load prediction data obtained in step S6, which is the load value directly predicted for the target pile using the foundation load mapping relationship. Add the initial load prediction data to the obtained corrected load data to obtain the final load prediction. For example, if the initial prediction value is 850kN and the corrected data is -15kN, the final prediction value is 835kN. This corrected value is more accurate than the initial prediction value because it compensates for the systematic bias of the foundation mapping relationship on similar feature samples through a residual learning model. The final output load prediction can be used as the basis for optimizing the design parameters of the target variable cross-section pile group in the subsequent step S8.

[0073] It should be noted that the load correction model in this embodiment of the invention directly adopts the standard multi-kernel learning model framework without any improvement or innovation to its algorithm structure or kernel function combination method. This model uses a weighted combination of multiple basis kernel functions, such as linear kernels, polynomial kernels, and radial basis kernels, to learn the nonlinear mapping relationship between feature data and residual data. Its construction and training processes follow the standard multi-kernel learning workflow, including kernel function selection, weight optimization, and cross-validation. Therefore, it is essentially a load correction model that directly applies existing multi-kernel learning techniques. By correcting the initial prediction through residual analysis and multi-kernel learning, mapping relationship errors are compensated, significantly improving the accuracy of load prediction.

[0074] Finally, based on the load prediction, the dataset of parameters to be optimized is optimized to obtain the target design parameter set for the target variable cross-section pile group. Design indices are extracted from the design requirements; the load prediction and design indices are processed to obtain the target to be optimized; the parameters to be optimized are determined from the dataset of parameters to be optimized based on the target; using the target as the optimization objective, an artificial intelligence algorithm is used to process the parameters to be optimized to obtain the target optimized parameters; based on the target optimized parameters, the target design parameter set is obtained. Based on the target design parameter set and the load mapping relationship, the target load situation is obtained; the target load situation and the target to be optimized are analyzed, and the parameters to be optimized are adjusted based on the analysis results.

[0075] The predicted load is the final predicted load value obtained in step S7 after residual correction. The dataset of parameters to be optimized is the current set of geometric and material parameters that need to be adjusted in the target variable cross-section pile group, such as segment diameter and elastic modulus. The design indices in the design requirements refer to the performance constraints specified in engineering specifications or project requirements, such as pile top settlement not exceeding 10mm and the characteristic value of single pile bearing capacity not less than 1000kN. The optimization objective is the objective function formed by comparing the predicted load with the design indices, for example, minimizing the difference between the predicted load and the design bearing capacity. Artificial intelligence algorithms refer to optimization algorithms used to search for the optimal parameter combination, such as genetic algorithms, particle swarm optimization algorithms, or Bayesian optimization. The target optimization parameters are the optimal parameter values ​​obtained after iterative optimization by the algorithm. The target design parameter set is the final design parameter scheme formed by integrating these optimal parameters. The target load is the predicted load value obtained by substituting the target design parameter set back into the load mapping relationship.

[0076] Clearly list all performance constraints that must be met from the project's design specifications, technical specifications, or owner requirements. For variable cross-section pile foundations under transmission towers, common design parameters include: the characteristic value of the vertical bearing capacity of a single pile not less than a certain limit, such as not less than 1200 kN; the settlement at the pile top not exceeding a certain limit, such as not exceeding 10 mm; the maximum bending moment of the pile not exceeding the flexural bearing capacity of the cross-section; and the maximum compressive stress of the pile concrete not exceeding the material strength design value. Compile these design parameters into a list, with each item including clear numerical limits and comparison directions of greater than or equal to, or less than or equal to. These parameters will be used in subsequent steps to construct the optimization objective function and constraints.

[0077] The predicted values ​​from the load forecast are compared item by item with the extracted design indices to form the optimization objectives. These objectives are typically one or more quantifiable objective functions. For example: the difference in bearing capacity deficiency = design bearing capacity (1200kN) minus predicted bearing capacity (950kN) = 250kN, and this difference needs to be minimized to below 0; the difference in settlement excess = predicted settlement (12mm) minus design settlement (10mm) = 2mm, and this difference also needs to be minimized to below 0. If multiple objectives exist simultaneously, a single objective function can be constructed using a weighted summation method, or a multi-objective optimization method can be employed. Ultimately, a clear optimization objective is obtained: to find a set of parameters that satisfies all design indices, or minimizes the weighted sum of the degrees of non-satisfaction.

[0078] The optimization algorithm selects truly adjustable parameters from the dataset of parameters to be optimized. For example, the elastic modulus of concrete may be fixed due to a fixed material supplier, but the segment diameter and reinforcement ratio can be optimized within a certain range. The value range of each parameter to be optimized is clearly defined; for example, the average diameter can vary continuously between 0.8m and 1.2m, and the reinforcement ratio can vary between 0.5% and 2.0%. These value ranges constitute the search space of the optimization algorithm.

[0079] Choose an artificial intelligence algorithm suitable for continuous parameter optimization. This embodiment of the invention uses a genetic algorithm or a particle swarm optimization algorithm. Specific operations are as follows: First, initialize a group of individuals, each representing a set of parameters to be optimized; then, substitute each individual into the load mapping relationship model, and if necessary, also into the load correction model, to calculate the corresponding load prediction; then, calculate the fitness value of each individual based on the objective function constructed in the second step; next, generate the next generation population through selection, crossover, mutation, etc.; repeat the iteration until the fitness value meets the requirements or reaches the preset maximum number of iterations. The final optimal individual is the target optimization parameter, for example, an average diameter of 1.05m and a reinforcement ratio of 1.2%. During the optimization process, design indicators need to be treated as constraints, which can be achieved using a penalty function method, i.e., imposing a large penalty value on individuals that do not meet the constraints.

[0080] It should be noted that the genetic algorithm and particle swarm optimization algorithm in the embodiments of this invention directly adopt the standard original algorithm framework without any improvement or innovation to their selection, crossover, mutation operators, or velocity update, position update mechanisms. Specifically, the genetic algorithm optimizes parameters according to classic binary or real number encoding, roulette wheel selection, single-point or uniform crossover, and basic mutation operations; the particle swarm optimization algorithm follows standard velocity inertia weight, individual cognition, and social learning factor update rules. Both algorithms are applied in their original form for iterative search of the parameters to be optimized, and their core computational flow is consistent with the classic versions described in the literature, without introducing new algorithm variants or hybrid strategies.

[0081] The obtained target optimization parameters are combined with those fixed parameters that remain unchanged during the optimization process, such as concrete strength grade and pile length, to form a complete target design parameter set. This parameter set contains all the necessary geometric and material parameters for the target variable cross-section pile group and can be directly used for construction drawing design. For example, the target design parameter set might include: total pile length 15m, segment one length 5m diameter 1.05m, segment two length 10m diameter 0.85m, concrete strength grade C35, reinforcement ratio 1.2%, etc. This parameter set is then organized into a standard format document or table as the final output of the optimization step.

[0082] To verify whether the target design parameter set obtained in step five truly meets the design requirements, it needs to be re-substituted into the load mapping relationship model established in step S5 to calculate the corresponding predicted load values, referred to as the target load condition. For example, the predicted single pile bearing capacity under this parameter combination is calculated to be 1220 kN, and the predicted pile top settlement is 9.5 mm. This step is equivalent to independently verifying the optimization results, ensuring that the optimization process does not produce unreliable results due to premature convergence of the algorithm or improper parameter settings.

[0083] The obtained target load conditions are compared and analyzed with the determined optimization targets, i.e., design indicators. If the target load conditions fully meet all design indicators, for example, bearing capacity ≥ 1220kN and settlement ≤ 10mm ≤ 9.5mm, then the optimization is successful, and the current target design parameter set is the final result. If some indicators are still not met, for example, the bearing capacity is met but the settlement is slightly exceeded at 10.2mm, then the reasons need to be analyzed and the optimization strategy adjusted. Adjustment methods include: appropriately widening or narrowing the value range of the parameters to be optimized, modifying the optimization algorithm parameters such as increasing the population size, increasing the mutation rate, or redistributing the weight coefficients in the weighted objective function. After adjustment, return to step four to re-iterate the optimization iteration until the target design parameter set that meets all design indicators is obtained. The process and data of each iteration should be recorded for traceability and reproduction. Based on high-precision load prediction, an intelligent algorithm is used to automatically optimize design parameters and efficiently obtain the optimal parameter set that meets the design requirements.

[0084] It is important to note that the effective execution of the aforementioned design parameter optimization method based on the load influence parameter set and load mapping relationship highly depends on the accurate acquisition of load data at various distribution locations of the variable cross-section pile group and the reasonable quantification of the influence of parameter loads. If there are deviations in the load data acquisition or the parameter selection fails to fully reflect the actual load transfer characteristics, the load mapping relationship constructed above will not be able to output reliable design parameter optimization results. Therefore, this embodiment, based on the explanation of the design parameter optimization logic driven by load data, further provides a machine learning analysis foundation to support the implementation of this optimization method. Preferably, it provides a method for analyzing the load transfer mechanism of variable cross-section pile groups based on a two-layer heterogeneous model architecture. Regarding the existing technical content such as conventional model training strategies, general kernel function selection, and basic regularization methods involved in this machine learning method, those skilled in the art can determine them according to the sample size and distribution characteristics of the actual engineering data, and they will not be specifically described in this embodiment of the invention. The following section will focus on describing the implementation process of the two-layer model that is specifically strongly associated with the construction of the load influence parameter set and the establishment of load mapping relationship. This includes key technical steps to ensure that multi-source heterogeneous data can be accurately mapped to load prediction results, such as automatic parameter space segmentation, local feature screening, and multi-kernel residual compensation, thereby ensuring the integrity and feasibility of the overall technical solution.

[0085] Another embodiment of this invention proposes a machine learning-based method for analyzing the load transfer mechanism of variable cross-section pile groups. The method employs a two-layer heterogeneous model architecture to address the problems of strong nonlinearity, complex parameter coupling, and low computational efficiency encountered by traditional analytical methods and single machine learning models in predicting the load of variable cross-section pile groups. The core idea of ​​this embodiment is as follows: The first layer uses an improved multivariate adaptive regression spline model to automatically segment the parameter space and performs local feature filtering within each segment interval to establish a benchmark prediction model with strong interpretability. The second layer constructs a residual prediction model based on multi-kernel learning to compensate for the high-order nonlinear relationships and complex coupling effects between piles, soil, and loads that the first-layer model failed to fit. The outputs of the two models are added together to obtain the final accurate predictions of the loads of the diagonal piles, side piles, and center piles.

[0086] During the data preparation phase, this embodiment of the invention collects the geometric parameters, layout parameters, soil layer parameters, load parameters, and corresponding load data of the variable cross-section pile group in sequence according to the actual working conditions. The geometric parameters include pile diameter, pile length, and pile cross-sectional dimensions; the layout parameters include pile top coordinates, pile spacing, and pile arrangement; the soil layer parameters include soil layer depth and soil strength parameters; and the load parameters include load magnitude, load application location, and load direction. After data collection, all parameters are standardized in format and units, and fields are rearranged. Missing fields are filled using interpolation methods and missing markers are added. Outlier removal and noise reduction are performed on the load data. The filled parameters are then normalized and combined row-wise with the load data of corner piles, edge piles, and center piles under the corresponding working conditions to form a structured training sample set.

[0087] The first layer of model construction begins with a multivariate adaptive regression spline model. Geometric, layout, soil layer, and load parameters from each training sample are used as input, and the corresponding three-pile load data are used as output, imported sequentially into the modeling process. In the forward progressive generation phase, cutoff points are set at multiple quantile positions within the value range of each input parameter. Based on each cutoff point, corresponding piecewise linear spline basis functions are generated and added sequentially to the candidate basis function set. After reaching the preset maximum number of basis functions, the backward progressive pruning phase begins. Deletion experiments are performed on each basis function in the candidate set. The contribution of each basis function to the global fit is calculated according to the generalized cross-validation criterion. Basis functions with contributions lower than or equal to a threshold are deleted, while combinations with contributions higher than the threshold are retained. Based on the retained basis functions, the segmented interval boundaries of each input parameter within its numerical domain are determined, thus obtaining complete segmented interval information in the parameter space.

[0088] After obtaining the segmented intervals, this embodiment of the invention further performs local feature filtering on each segmented interval. Training samples are categorized according to the segmented interval to which each input parameter belongs, ensuring that each sample corresponds to a specific segmented interval. Within each segmented interval, the geometric, layout, soil layer, and load parameters of all samples within that interval are read to generate a candidate feature list. For each feature in the list, based on the multivariate adaptive regression spline model within the current segmented interval, the local fitting contribution of that feature to load prediction is calculated. Specifically, this is measured by the change in model prediction error after deleting all spline basis functions corresponding to that feature. Features with a local fitting contribution greater than a threshold are recorded as locally effective features and retained; features with a contribution less than or equal to the threshold are deleted from the candidate list, forming a local feature subset for that segmented interval. Then, based on this local feature subset, the spline basis functions are recombined to construct an improved multivariate adaptive regression spline model using only locally effective features, and the benchmark predicted values ​​for the loads of corner piles, edge piles, and center piles are output. This mechanism allows different feature combinations to be used for different segment intervals, effectively avoiding global feature redundancy and cross-interval interference, and improving the structural interpretability of the model and the accuracy of parameter selection.

[0089] After obtaining the baseline prediction value, this embodiment of the invention enters the residual and feature data processing stage. First, real load data is read from the training sample set. The difference between the real load and the baseline prediction value is calculated for each sample according to the pile number order, generating residual data for corner piles, edge piles, and center piles. These residual data are then combined row by row to form a residual dataset. Simultaneously, geometric and layout parameters from the training sample set are extracted, including pile diameter, pile length, pile cross-sectional dimensions, pile top coordinates, pile spacing, and pile layout. These parameters are read by column slices to form geometric feature data. Soil layer parameters are read in the same way as soil layer feature data. Load parameters are read as load feature data. At this point, the three types of feature data correspond one-to-one with the residual data, providing input for the second-layer multi-core learning model.

[0090] The core of the second-layer model is a multi-kernel learning residual model. Its goal is to design specialized kernel functions for the three types of features: geometry, soil layer, and load, and to independently learn the nonlinear mapping relationship between residuals and features within each segmented interval. First, based on the segmented interval information obtained from the first-layer model, the geometric, soil layer, and load features corresponding to each working condition are divided into their respective segmented intervals, forming sample sets for each segmented interval. For geometric features, a pile group topology graph is constructed using each pile as a node and the distance between piles as edges. Pile diameter, pile length, variable cross-section size parameters, and pile location coordinates are written into the node attributes, and pile spacing is written into the edge weights. The graph similarity is calculated for the pile group topology graphs of any two training samples, and this is used as the kernel value of the geometric kernel function to generate the geometric kernel matrix. For soil layer features, the soil layers are sorted by depth, generating a layered sequence composed of layer information and corresponding soil layer parameters. The sequence similarity is calculated for the layered sequences of any two samples, and this is used as the kernel value of the soil layer kernel function to obtain the soil layer kernel matrix. For load characteristics, they are represented as load triples containing load amplitude, point of application parameters, and direction parameters. The kernel values ​​are calculated for amplitude, position, and direction respectively, and the three are combined in product form as the kernel value of the load kernel function to obtain the load kernel matrix.

[0091] To balance global consistency with local variability, this embodiment of the invention sets global kernel parameters and interval kernel parameter offsets for the three types of kernel functions in each segmented interval. The two are added together to obtain the actual interval kernel parameters for that segmented interval. Within each segmented interval, the geometric kernel matrix, soil layer kernel matrix, and load kernel matrix are added together to form a comprehensive kernel matrix. Then, the residual vector of that segmented interval is obtained, and the kernel ridge regression coefficient solving function is called to solve for the residual learning coefficient vector of the current segmented interval based on the comprehensive kernel matrix and the residual vector. The entire model is trained by optimizing a composite objective function, which is obtained by summing three terms: the first term is the residual fitting term for each segment interval, i.e., the difference between the residual vector in each segment interval and the residual prediction value obtained from the comprehensive kernel matrix and the residual learning coefficient vector; the second term is the coefficient regularization term, calculated based on the residual learning coefficient vector of each segment interval and the comprehensive kernel matrix of that interval; the third term is the offset shrinkage regularization term, calculated by the squared L2 norm of the interval kernel parameter offsets of the geometric, soil layer, and load kernel functions for each segment interval, and obtained by summing them in each segment interval. This term is multiplied by a shrinkage coefficient and added to the objective function. Using this objective function as the optimization objective, the gradient descent method is used to jointly update the global kernel parameters of the three types of kernel functions and the offsets of the three types of interval kernel parameters in each segment interval. The update direction of the global kernel parameters is jointly determined by the contributions of the objective functions of multiple segmented intervals, converging to a shared parameter part that simultaneously reduces the objective function values ​​of multiple segmented intervals. The update of the interval kernel parameter offset is constrained by both the residual fitting term and the offset shrinkage regularization term: when the number of samples in a certain segmented interval is large, the driving effect of the residual fitting term is stronger, and the offset is more likely to take a non-zero value to adapt to the local characteristics of that interval; when the number of samples is small, the constraint of the offset shrinkage regularization term is relatively stronger, and the offset is compressed to near zero, making the kernel parameter of that interval closer to the global kernel parameter. The iteration is repeated until the objective function meets the convergence condition or reaches the preset iteration threshold, finally outputting a fully trained multi-kernel learning residual model.

[0092] In the prediction phase, for the new variable cross-section pile group condition, the same preprocessing as in the training phase is first performed to generate new geometric feature data, soil layer feature data, and load feature data, which are then arranged in the order of recording to form the prediction input dataset. This dataset is then input line by line into the improved multivariate adaptive regression spline model to calculate the baseline predicted load values ​​for corner piles, edge piles, and center piles. Simultaneously, the same geometric, soil layer, and load feature data are input line by line into the trained multi-kernel learning residual model to calculate the corresponding residual predicted values ​​for corner piles, edge piles, and center piles. Finally, the baseline predicted values ​​and residual predicted values ​​are added in the order of recording to obtain the final predicted load results for corner piles, edge piles, and center piles. If the multi-kernel learning residual model does not fully converge, the residual range can also be output as a reference for prediction uncertainty.

[0093] To comprehensively evaluate the performance advantages of the embodiments of the present invention in rapid prediction scenarios, a batch prediction comparison test was conducted between the model of the embodiments of the present invention and traditional analysis methods under the same computing environment. When performing centralized prediction of 100 different working conditions, the total computation time of the traditional method was approximately 12 to 21 seconds, while the total computation time of the model of the present invention was only 0.06 to 0.11 seconds, with an overall prediction speed improvement of approximately 150 to 200 times, demonstrating a significant computational efficiency advantage.

[0094] While ensuring extremely fast prediction speed, the model in this embodiment of the invention still maintains high prediction accuracy. Specifically, please refer to Table 1, which shows the load prediction results of the model in this embodiment of the invention, including the comparison results of the actual load and predicted load of some corner piles and edge piles in the same set of working conditions. The unit of load is kN (kilonewtons). It can be seen that even within the millisecond-level inference time, the prediction error of corner piles is mainly distributed between 3% and 5%, and the prediction error of edge piles is mainly distributed between 4% and 6%. There is no significant decrease in accuracy due to the substantial increase in calculation speed. This result shows that this invention effectively achieves a good balance between fast computation and high-precision prediction by introducing a piecewise spline mechanism to extract the main control features within different parameter ranges and combining a multi-core learning structure to perform nonlinear compensation for the residuals.

[0095] Table 1 Load prediction results of the model in the embodiment of the present invention Another embodiment of the present invention provides an optimized design system for variable cross-section pile groups. For details, please refer to [link to relevant documentation]. Figure 3 , Figure 3 The diagram shown illustrates the structure of an optimized design system for a group of variable cross-section piles according to one embodiment of the present invention, comprising: The acquisition module 11 is used to acquire load data at various distribution locations of the sample variable cross-section pile group; Module 12 is used to process the distribution locations based on the pile geometry of the sample variable cross-section pile group and determine the initial parameter load data segments. Optimization module 13 is used to optimize the load data segments of each initial parameter based on the load transfer characteristics of the transmission tower, and to determine the load data segments of each parameter. The influence module 14 is used to determine each load influence parameter based on the load influence degree of each parameter in each load data segment, and integrate them to obtain a load influence parameter set. Mapping module 15 is used to obtain load mapping relationships based on the load influence parameter set and load data; Initial module 16 is used to obtain initial load prediction data based on the load mapping relationship and the obtained target variable cross-section pile group parameter dataset; Prediction module 17 is used to analyze the dataset of parameters to be optimized and the load data, and to process the initial load prediction data with the analysis results to obtain the load prediction situation. Target module 18 is used to optimize the dataset of parameters to be optimized based on the load prediction situation, so as to obtain the target design parameter set of the target variable cross section pile group.

[0096] Accordingly, embodiments of the present invention provide a computer-readable storage medium, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform steps in the optimized design method for variable cross-section pile groups as described in the above embodiments, for example... Figure 1 Steps S1 to S8 as described above.

[0097] This invention departs from the traditional approach of simulating stress analysis of design parameters and processing the results. Instead, it directly acquires load data from various locations of a sample variable cross-section pile group. Initial parameter load data segments are defined based on the pile geometry, and then optimized using the actual load transfer characteristics of the transmission tower to determine the final parameter load data segments. Furthermore, a load influence parameter set is selected and integrated based on the load influence degree of each parameter. On this basis, this invention constructs a load mapping relationship between this load influence parameter set and the actual load data. Based on this, it predicts the parameter dataset to be optimized for the target variable cross-section pile group. Simultaneously, it uses analysis of the parameter dataset to be optimized and the load data to correct the prediction results. Finally, based on the corrected load prediction, it directly outputs the target design parameter set. This invention, by analyzing the load influence parameter set and the actual load data to construct a load mapping relationship, fundamentally avoids the technical bottleneck of inaccurate simulation stress analysis results. It significantly improves the matching accuracy between optimized design parameters and actual working conditions, eliminates the safety risks of transmission tower tilting or collapse caused by inaccurate design parameters, and makes the design parameter optimization process more in line with the actual load transfer mechanism of the project, thereby improving the reliability and engineering practicality of the optimization results.

[0098] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. An optimized design method for a group of piles with variable cross-sections, characterized in that, include: Obtain load data at various distribution locations of the sample variable cross-section pile group; Based on the pile geometry of the sample variable cross-section pile group, the distribution locations are processed to determine the initial parameter load data segments. Based on the load transfer characteristics of the transmission tower, the load data segments of each initial parameter are optimized to determine the load data segments of each parameter. Based on the load influence degree of each parameter in each of the aforementioned parameter load data segments, determine each load influence parameter and integrate them to obtain a load influence parameter set; Based on the load influence parameter set and the load data, the load mapping relationship is obtained; Based on the load mapping relationship and the obtained target variable cross-section pile group parameter dataset, the initial load prediction data is obtained. The dataset of parameters to be optimized and the load data are analyzed, and the analysis results are used to process the initial load prediction data to obtain the load prediction results. Based on the load prediction, the dataset of parameters to be optimized is optimized to obtain the target design parameter set of the target variable cross-section pile group; The pile geometry based on the sample variable cross-section pile group is used to process each of the distribution locations to determine each initial parameter load data segment, including: Analyze the geometric structure of the pile body to determine the rate of change of the pile body's outer contour; Based on the change rate of the outer contour of the pile body, each segment interval is determined; Obtain the parameter data corresponding to each of the segmented intervals; The distribution locations are divided based on the segmented intervals, and the corresponding segmented load data is extracted from the load data. Based on the parameter data and segmented load data corresponding to each segmented interval, each initial parameter load data segment is determined; The load influence parameters are determined based on the load influence degree of each parameter in each of the aforementioned load data segments, and then integrated to obtain a load influence parameter set, including: Extract each parameter from each of the aforementioned parameter load data segments; Analyze the changes in the segmented load data when each of the aforementioned parameters changes; Quantify the changes to obtain the degree of load influence for each parameter; Based on preset standards, the parameters corresponding to the degree of influence of each load are filtered to obtain the load influence parameters. By integrating all the load influence parameters, the load influence parameter set is obtained.

2. The optimized design method for variable cross-section pile groups as described in claim 1, characterized in that, The load transfer characteristics based on the transmission tower are used to optimize each of the initial parameter load data segments, and each parameter load data segment is determined, including: The design parameter data of the transmission tower are processed using finite element analysis technology to determine the load transfer characteristics of the transmission tower. The load transfer characteristics are analyzed to determine the load distribution. Based on the load distribution, the initial parameter load data segment is optimized to determine each parameter load data segment.

3. The optimized design method for variable cross-section pile groups as described in claim 1, characterized in that, The process of obtaining the load mapping relationship based on the load influence parameter set and the load data includes: The load influence parameter set and the load data are input into a pre-built artificial intelligence model for processing to obtain the load mapping relationship.

4. The optimized design method for variable cross-section pile groups as described in claim 1, characterized in that, The step of analyzing the dataset of parameters to be optimized and the load data, and processing the initial load prediction data with the analysis results to obtain the load prediction results, includes: Based on the parameter data in the parameter load data segment and the load mapping relationship, the sample benchmark prediction value is obtained; Based on the sample baseline prediction values ​​and the load data, residual data are obtained; Extract the feature data from the parameter data; The feature data and the residual data are input into a pre-constructed multi-kernel learning model for training to obtain a load correction model. The dataset of parameters to be optimized is input into the load correction model for processing to obtain load correction data. The initial load prediction data is processed based on the load correction data to obtain the load prediction.

5. The optimized design method for variable cross-section pile groups as described in claim 4, characterized in that, The feature data extracted from the parameter data includes: The parameter data is divided to obtain geometric parameter data and soil layer parameter data; Extract the feature information from the geometric parameter data and the soil layer parameter data respectively; The feature data is obtained by integrating geometric feature information and soil layer feature information.

6. The optimized design method for variable cross-section pile groups as described in claim 1, characterized in that, The optimization of the parameter dataset based on the load prediction results to obtain the target design parameter set for the target variable cross-section pile group includes: Extract the design parameters from the design requirements; The predicted load is processed in conjunction with the design parameters to obtain the target to be optimized. The parameters to be optimized are determined from the dataset of parameters to be optimized based on the objective to be optimized. Using the target to be optimized as the optimization objective, an artificial intelligence algorithm is used to process the parameters to be optimized to obtain the target optimization parameters; Based on the target optimization parameters, the target design parameter set is obtained.

7. The optimized design method for variable cross-section pile groups as described in claim 6, characterized in that, The process of obtaining the target design parameter set based on the target optimization parameters further includes: Based on the target design parameter set and the load mapping relationship, the target load condition is obtained; Analyze the target load and the target to be optimized, and adjust the parameters to be optimized based on the analysis results.

8. An optimized design system for variable cross-section pile groups, characterized in that, include: The acquisition module is used to acquire load data at various distribution locations of the sample variable cross-section pile group; The determination module is used to process each of the distribution locations based on the pile geometry of the sample variable cross-section pile group and determine each initial parameter load data segment. The optimization module is used to optimize each of the initial parameter load data segments based on the load transfer characteristics of the transmission tower, and to determine each parameter load data segment. The influence module is used to determine each load influence parameter based on the load influence degree of each parameter in each of the aforementioned load data segments, and integrate them to obtain a load influence parameter set. The mapping module is used to obtain the load mapping relationship based on the load influence parameter set and the load data; The initial module is used to obtain initial load prediction data based on the load mapping relationship and the obtained target variable cross-section pile group parameter dataset. The prediction module is used to analyze the dataset of parameters to be optimized and the load data, and to process the initial load prediction data with the analysis results to obtain the load prediction situation. The target module is used to optimize the parameter dataset to be optimized based on the load prediction to obtain the target design parameter set of the target variable cross-section pile group; The pile geometry based on the sample variable cross-section pile group is used to process each of the distribution locations to determine each initial parameter load data segment, including: Analyze the geometric structure of the pile body to determine the rate of change of the pile body's outer contour; Based on the change rate of the outer contour of the pile body, each segment interval is determined; Obtain the parameter data corresponding to each of the segmented intervals; The distribution locations are divided based on the segmented intervals, and the corresponding segmented load data is extracted from the load data. Based on the parameter data and segmented load data corresponding to each segmented interval, each initial parameter load data segment is determined; The load influence parameters are determined based on the load influence degree of each parameter in each of the aforementioned load data segments, and then integrated to obtain a load influence parameter set, including: Extract each parameter from each of the aforementioned parameter load data segments; Analyze the changes in the segmented load data when each of the aforementioned parameters changes; Quantify the changes to obtain the degree of load influence for each parameter; Based on preset standards, the parameters corresponding to the degree of influence of each load are filtered to obtain the load influence parameters. By integrating all the load influence parameters, the load influence parameter set is obtained.

Images