A variable cross-section composite pile reinforcement design method and system based on machine learning

Abstract

The application discloses a variable cross-section composite pile reinforcement design method and system based on machine learning, and the method comprises the following steps: acquiring multi-source data of the variable cross-section composite pile; performing feature extraction on the multi-source data to obtain key features and space-time features; based on the multi-source data, performing three-dimensional parameterized modeling and nonlinear finite element analysis to construct a variable cross-section shape and a stress nephogram of the pile body and obtain variable cross-section shape parameters; according to the key features, the space-time features and the variable cross-section shape parameters, performing multi-target prediction by using a preset ensemble learning model to obtain a structure safety coefficient, a material cost and construction feasibility; and based on the structure safety coefficient, the material cost and the construction feasibility, generating a steel reinforcement configuration scheme of the variable cross-section composite pile by using a mixed integer programming algorithm. The application can effectively improve the efficiency and accuracy of the variable cross-section composite pile reinforcement design.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method and system for designing reinforcement of variable cross-section composite piles based on machine learning. Background Technology

[0002] Variable cross-section composite piles are a type of pile foundation that combines different materials or cross-sectional dimensions. Through rational design and optimization, different pile materials or cross-sectional dimensions are used at different depths of the pile, thereby effectively improving the bearing capacity, deformation resistance, and adaptability to complex geological conditions of the pile foundation. This type of pile is suitable for infrastructure construction in soft soil foundations and complex geological environments, and can better cope with problems such as foundation settlement and lateral deformation, effectively improving the overall stability and safety of the pile foundation.

[0003] Most current variable cross-section composite pile reinforcement design systems have certain limitations. Traditional methods typically rely on empirical formulas and manual calculations, which cannot effectively handle complex multi-source data and changing engineering environments, leading to cumbersome design processes and susceptibility to human factors. In terms of optimization algorithms, most traditional systems cannot flexibly perform multi-objective optimization, usually focusing only on a single objective while neglecting other key factors such as construction feasibility. Existing variable cross-section composite pile reinforcement methods suffer from insufficient efficiency and accuracy. Summary of the Invention

[0004] To address the above technical problems, this invention provides a machine learning-based method and system for designing reinforcement of variable cross-section composite piles, which can effectively improve the efficiency and accuracy of reinforcement design for variable cross-section composite piles.

[0005] This invention provides a machine learning-based method for designing reinforcement of variable cross-section composite piles, including:

[0006] Acquire multi-source data of variable cross-section composite piles;

[0007] Feature extraction is performed on the multi-source data to obtain key features and spatiotemporal features;

[0008] Based on the multi-source data, three-dimensional parametric modeling and nonlinear finite element analysis are performed to construct the variable cross-sectional shape and stress cloud diagram of the pile body, and obtain the variable cross-sectional shape parameters.

[0009] Based on the key features, the spatiotemporal features, and the variable cross-section morphology parameters, a preset integrated learning model is used to perform multi-objective prediction to obtain the structural safety factor, material cost, and construction feasibility.

[0010] Based on the structural safety factor, the material cost, and the construction feasibility, a mixed integer programming algorithm is used to generate a reinforcement configuration scheme for the variable cross-section composite pile.

[0011] As an improvement to the above scheme, the base models of the ensemble learning model include: random forest model, gradient boosting decision tree model and deep residual network model.

[0012] As an improvement to the above scheme, the multi-source data includes: engineering geological parameters, load condition parameters, material performance parameters, cross-sectional parameters, and historical engineering data.

[0013] As an improvement to the above scheme, the acquisition of multi-source data of variable cross-section composite piles includes:

[0014] The engineering geological parameters of the variable cross-section composite pile were obtained through ground-penetrating radar detection, in-situ testing, and dynamic monitoring of groundwater level.

[0015] By designing load spectrum decomposition and simulating construction process loads, the load condition parameters of variable cross-section composite piles are obtained.

[0016] Material performance parameters of variable cross-section composite piles are obtained through concrete material testing and steel reinforcement performance testing.

[0017] The cross-sectional parameters of the variable cross-section composite pile are obtained by using variable cross-section geometric parameterization and cross-section stiffness parameter calculation.

[0018] Retrieve historical engineering data of variable cross-section composite piles from the engineering database.

[0019] As an improvement to the above scheme, the feature extraction of the multi-source data to obtain key features and spatiotemporal features includes:

[0020] The multi-source data is normalized to obtain normalized data;

[0021] Based on the temporal and spatial correlation of the normalized data, a spatiotemporal correlation matrix is ​​constructed to obtain spatiotemporal features.

[0022] Principal component analysis was used to reduce the dimensionality of the normalized data to obtain feature data.

[0023] Based on the feature data, the mutual information entropy algorithm is used to calculate the mutual information between each feature data and the target variable, and the feature data with the largest mutual information are selected to obtain the key features.

[0024] As an improvement to the above scheme, the formula for the time correlation is:

[0025]

[0026] in, The temporal correlation between the i-th feature and the j-th feature at time t. Let be the covariance between the i-th feature and the j-th feature at time t. , Let be the variances of the i-th and j-th features at time t, respectively.

[0027] The formula for the spatial correlation is:

[0028]

[0029] in, Let represent the spatial correlation between the i-th feature and the j-th feature at spatial location s. Let be the covariance between the i-th feature and the j-th feature at spatial location s. Let be the variances of the i-th and j-th features at spatial location s, respectively.

[0030] As an improvement to the above scheme, the training method of the ensemble learning model includes:

[0031] Based on spatiotemporal features, a two-stream neural network is constructed to process temporal and spatial features separately, and tensor product operation is used to perform deep interaction between temporal and spatial features to obtain interactive features;

[0032] The key features of the variable cross-section composite pile, the variable cross-section morphological parameters, and the interaction features are used as input features of the model to train the base model of the integrated learning model in parallel.

[0033] Calculate the feature importance of the input features, and determine the weight of each base model based on the feature importance;

[0034] Based on the base model and its weights, an initial ensemble learning model is obtained.

[0035] Based on the initial ensemble learning model, multi-objective optimization and Bayesian optimization are performed to obtain the final ensemble learning model; wherein, the objective function of multi-objective optimization is constructed based on the structural safety factor, material cost and construction feasibility.

[0036] As an improvement to the above scheme, the step of generating a reinforcement configuration scheme for variable cross-section composite piles based on the structural safety factor, the material cost, and the construction feasibility, using a mixed integer programming algorithm, includes:

[0037] Obtain flexural capacity, crack width, and reinforcement ratio;

[0038] Using the bending capacity, crack width, reinforcement ratio, structural safety factor, material cost, and construction feasibility as constraints, an integer programming model is constructed by representing the quantity and distribution of steel bars with binary variables and constructing an objective function.

[0039] The integer programming model is input into a hybrid integer programming solver to obtain the optimal reinforcement configuration scheme.

[0040] As an improvement to the above scheme, the integer programming model adjusts the total amount of steel reinforcement by minimizing the objective function, the formula of which is:

[0041]

[0042] in, Let J be the cross-sectional area of ​​the j-th type of steel reinforcement. It is a binary variable.

[0043] This invention also provides a machine learning-based variable cross-section composite pile reinforcement design system, comprising:

[0044] The data acquisition module is used to acquire multi-source data of variable cross-section composite piles;

[0045] The feature extraction module is used to extract features from the multi-source data to obtain key features and spatiotemporal features;

[0046] The variable cross-section processing module is used to perform three-dimensional parametric modeling and nonlinear finite element analysis based on the multi-source data, construct the variable cross-section shape and stress cloud diagram of the pile body, and obtain the variable cross-section shape parameters.

[0047] The model prediction module is used to perform multi-objective prediction based on the key features, the spatiotemporal features and the variable cross-section morphology parameters using a preset integrated learning model, so as to obtain the structural safety factor, material cost and construction feasibility.

[0048] The reinforcement optimization module is used to generate a reinforcement configuration scheme for the variable cross-section composite pile based on the structural safety factor, the material cost, and the construction feasibility, using a mixed integer programming algorithm.

[0049] Compared to existing technologies, the beneficial effects of the machine learning-based reinforcement design method and system for variable cross-section composite piles provided in this invention are as follows: By acquiring multi-source data of variable cross-section composite piles, various influencing factors are fully considered, improving the comprehensiveness of the reinforcement scheme; by extracting key features and spatiotemporal features from the multi-source data, the most discriminative features can be obtained for subsequent prediction and optimization, improving the reliability of the design; by performing three-dimensional parametric modeling and nonlinear finite element analysis based on the multi-source data, the variable cross-section morphology and stress cloud diagram of the pile body are constructed, and the variable cross-section morphology parameters are obtained, further improving the design... The design accuracy ensures the structural safety of pile foundations under different working conditions. By using a pre-set integrated learning model to perform multi-objective prediction based on key features, spatiotemporal characteristics, and variable cross-section morphology parameters, the structural safety factor, material cost, and construction feasibility are obtained. Then, based on the structural safety factor, material cost, and construction feasibility, a mixed integer programming algorithm is used to generate the optimal reinforcement configuration scheme. This can automatically adjust and optimize the reinforcement scheme under complex and ever-changing design requirements, satisfying multiple objectives of structural safety, cost control, and construction feasibility. This improves design efficiency, reduces human error, and enhances the accuracy of reinforcement design for variable cross-section composite piles. Attached Figure Description

[0050] Figure 1 This is a flowchart illustrating a machine learning-based reinforcement design method for variable cross-section composite piles provided in an embodiment of the present invention.

[0051] Figure 2 This is a structural schematic diagram of a machine learning-based variable cross-section composite pile reinforcement design system provided in an embodiment of the present invention. Detailed Implementation

[0052] 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. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] Please see Figure 1 , Figure 1 This is a flowchart illustrating a machine learning-based reinforcement design method for variable cross-section composite piles according to an embodiment of the present invention. The machine learning-based reinforcement design method for variable cross-section composite piles includes:

[0054] S1: Obtain multi-source data of variable cross-section composite piles;

[0055] Specifically, based on geological surveys, load conditions, material properties, cross-sectional parameters, and historical data, multi-source data of variable cross-section composite piles are obtained. Through comprehensive data collection, the embodiments of the present invention can fully consider various influencing factors such as engineering geology, load conditions, and material properties, and achieve precise optimization of the reinforcement scheme of composite piles.

[0056] As one optional embodiment, the multi-source data includes: engineering geological parameters, load condition parameters, material performance parameters, cross-sectional parameters, and historical engineering data.

[0057] As one optional embodiment, the acquisition of multi-source data for variable cross-section composite piles includes:

[0058] The engineering geological parameters of the variable cross-section composite pile were obtained through ground-penetrating radar detection, in-situ testing, and dynamic monitoring of groundwater level.

[0059] By designing load spectrum decomposition and simulating construction process loads, the load condition parameters of variable cross-section composite piles are obtained.

[0060] Material performance parameters of variable cross-section composite piles are obtained through concrete material testing and steel reinforcement performance testing.

[0061] The cross-sectional parameters of the variable cross-section composite pile are obtained by using variable cross-section geometric parameterization and cross-section stiffness parameter calculation.

[0062] Retrieve historical engineering data of variable cross-section composite piles from the engineering database.

[0063] S2: Perform feature extraction on the multi-source data to obtain key features and spatiotemporal features;

[0064] As one optional embodiment, the feature extraction of the multi-source data to obtain key features and spatiotemporal features includes:

[0065] The multi-source data is normalized to obtain normalized data;

[0066] Based on the temporal and spatial correlation of the normalized data, a spatiotemporal correlation matrix is ​​constructed to obtain spatiotemporal features.

[0067] Principal component analysis was used to reduce the dimensionality of the normalized data to obtain feature data.

[0068] Based on the feature data, the mutual information entropy algorithm is used to calculate the mutual information between each feature data and the target variable, and the feature data with the largest mutual information are selected to obtain the key features.

[0069] Specifically, multi-source data is normalized using Min-Max normalization, and the normalization formula is as follows:

[0070]

[0071] in, Represents normalized data. , These are the minimum and maximum values ​​of the features in the dataset.

[0072] Furthermore, based on the normalized data, a temporal correlation matrix is ​​constructed according to the spatiotemporal correlation of the data. In multi-source data, the values ​​of each feature data at different time points form a time series, and the values ​​at different spatial locations form a spatial series. For the time series, the temporal correlation between feature data is calculated. For the spatial series, the spatial correlation between feature data is calculated. Then, based on the temporal and spatial correlations between feature data, a spatiotemporal correlation matrix is ​​constructed to obtain the temporal and spatial dependencies of each feature data.

[0073] As one optional embodiment, the formula for the time correlation is:

[0074]

[0075] in, The temporal correlation between the i-th feature and the j-th feature at time t. Let be the covariance between the i-th feature and the j-th feature at time t. , Let be the variances of the i-th and j-th features at time t, respectively.

[0076] The formula for the spatial correlation is:

[0077]

[0078] in, Let represent the spatial correlation between the i-th feature and the j-th feature at spatial location s. Let be the covariance between the i-th feature and the j-th feature at spatial location s. Let be the variances of the i-th and j-th features at spatial location s, respectively.

[0079] Furthermore, for the extraction of key features, firstly, the normalized data is mapped to a new coordinate system using principal component analysis to reduce the dimensionality of the data; then, the information content between different feature data is calculated based on the mutual information entropy algorithm, and the most discriminative features are selected as key features.

[0080] In the dimensionality reduction process of principal component analysis, the covariance matrix is ​​calculated on the normalized data matrix, and the covariance matrix is ​​decomposed into eigenvalues ​​to obtain eigenvalues ​​and eigenvectors. Then, the eigenvalues ​​are sorted from largest to smallest, and the principal components corresponding to the first k eigenvectors are selected. These principal components will form a new coordinate system. The original data is projected onto these principal components to obtain the dimensionality-reduced feature data.

[0081] Mutual information entropy is a way to measure the correlation and information content between two variables. First, each feature is represented by its entropy, as shown in the formula:

[0082]

[0083] in, Features entropy, Features The probability distribution, Features The possible values ​​of ;

[0084] The mutual information between features is calculated based on the entropy of each feature, using the following formula:

[0085]

[0086] in, for , The joint entropy;

[0087] Furthermore, by calculating the mutual information between each feature data point and the target variable, the feature data points with higher mutual information are selected as key features. The target variable can be obtained from step S3.

[0088] S3: Based on the multi-source data, perform three-dimensional parametric modeling and nonlinear finite element analysis to construct the variable cross-sectional shape and stress cloud diagram of the pile body, and obtain the variable cross-sectional shape parameters.

[0089] As one optional embodiment, step S3 specifically includes:

[0090] S31: Set key control sections according to the pile axis, define section control parameters, and construct a non-uniform rational B-spline (NURBS) model;

[0091] S32: Construct spatial topological relationships by establishing a Frenet frame for the pile axis and calculating the gradient tensor of the cross section;

[0092] S33: Based on the concrete and steel reinforcement parameter data from multi-source data, nonlinear finite element analysis is performed, and a stress invariant cloud map is constructed to obtain the variable cross-sectional shape parameters. The stress invariant cloud map is a graphical representation of the stress distribution of the pile cross-section obtained through finite element analysis, and the stress magnitude can be viewed through the color gradient of the cloud map.

[0093] Specifically, in step S31, the pile axis is fitted using multiple discrete points, or a continuous geometric curve is generated from the collected data. Then, based on the pile's geometric characteristics and construction design requirements, the positions of several key sections are determined, and the control parameters for defining these sections are obtained. Based on these parameters—namely, the pile axis control points and section control parameters (such as radius, shape factor, stiffness distribution, and weighting coefficients set based on the section surface shape)—a non-uniform rational B-spline surface model is constructed. The formula is:

[0094]

[0095] in, For points on the curved surface, Let them be two variables in the parameter space. , For B-spline basis functions, The weights of the control points, This refers to the location of the control point.

[0096] In step S32, the Frenet frame describes the coordinate system of the tangent, normal, and subnormal directions of the curve, parameterizes the pile axis, and obtains the tangent, normal, and subnormal T(s), N(s), and B(s) of the pile axis at position s; based on the axis position and cross-sectional shape, the deformation and gradual change of the cross-section on each cross-section are calculated using the following formula:

[0097]

[0098] Then, obtain the cross-sectional gradient tensor: Based on the Frenet frame and cross-sectional shape gradient, a local coordinate system is calculated at each key location to establish the topological relationship of the pile in space.

[0099] Furthermore, embodiments of the present invention also dynamically adjust the adaptive mesh based on the rate of change of cross-sectional curvature.

[0100] In step S33, during the finite element analysis, the geometric and material properties (concrete and steel reinforcement) of the pile body must first be modeled. It is assumed that the nonlinear behavior of the material is described by a bilinear stress-strain relationship or a more complex constitutive relation. Taking concrete and steel reinforcement as an example, assuming the stress-strain relationship of the concrete adopts the Mohr-Coulomb criterion or the Cesar-Castillo constitutive model, the problem is transformed into a set of algebraic equations by discretizing the geometry and material properties of the pile body using the finite element analysis method. The discrete equations can be expressed as: ,in, It is the stiffness matrix. It is a displacement vector. This is the external force vector. The deformation and displacement of the structure are calculated using this equation, and the stress cloud diagram can then display the stress situation of each part of the structure based on these displacements and deformations. By constructing the variable cross-sectional shape of the pile and the stress cloud diagram, variable cross-sectional shape parameters can be obtained, providing key input data for subsequent model prediction and reinforcement optimization, thus improving the accuracy and reliability of reinforcement design.

[0101] S4: Based on the key features, the spatiotemporal features and the variable cross-section morphology parameters, a preset integrated learning model is used to perform multi-objective prediction to obtain the structural safety factor, material cost and construction feasibility.

[0102] As one of the optional embodiments, the base models of the ensemble learning model include: a random forest model, a gradient boosting decision tree model, and a deep residual network model.

[0103] Specifically, the ensemble learning model is used for multi-objective prediction, including predicting stress distribution under different loads and estimating the feasibility of reinforcement schemes.

[0104] As one optional embodiment, the training method of the ensemble learning model includes:

[0105] Based on spatiotemporal features, a two-stream neural network is constructed to process temporal and spatial features separately, and tensor product operation is used to perform deep interaction between temporal and spatial features to obtain interactive features;

[0106] The key features of the variable cross-section composite pile, the variable cross-section morphological parameters, and the interaction features are used as input features of the model to train the base model of the integrated learning model in parallel.

[0107] Calculate the feature importance of the input features, and determine the weight of each base model based on the feature importance;

[0108] Based on the base model and its weights, an initial ensemble learning model is obtained.

[0109] Based on the initial ensemble learning model, multi-objective optimization and Bayesian optimization are performed to obtain the final ensemble learning model; wherein, the objective function of multi-objective optimization is constructed based on the structural safety factor, material cost and construction feasibility.

[0110] Specifically, based on the constructed spatiotemporal correlation matrix, a two-stream neural network is built in both spatial and temporal dimensions, and deep interaction of spatial-temporal features is achieved based on tensor product operations. By deeply interacting features in both spatial and temporal dimensions, the performance of the ensemble learning model can be enhanced, enabling each base model to make better predictions based on the spatiotemporal features of the data, thereby improving the accuracy of the overall model.

[0111] A two-stream neural network processes the spatial and temporal flows in the spatiotemporal correlation matrix separately, and its outputs undergo deep interaction via tensor product, as shown in the formula:

[0112]

[0113] in, This represents the tensor product operation; deep interactions can be further fused using multi-layer neural networks.

[0114] Furthermore, a parallel computing framework is used to train multiple base models simultaneously: a random forest model, a gradient boosting decision tree model, and a deep residual network model, to improve computational efficiency. Specifically, the random forest model constructs multiple decision trees and uses a voting process; the training process involves randomly selecting samples and feature subsets to train each tree. The gradient boosting decision tree model adjusts the current model based on the residuals of the previous model during each training iteration. The deep residual network model addresses the vanishing gradient problem in deep network training by introducing a residual module; the output of the residual module is: ,in, It is the output of the residual block. It is input. These are network weights. By combining the advantages of multiple base models, the accuracy of predictions is effectively improved.

[0115] Furthermore, the contribution weight of each base model is adjusted according to the importance of each feature. Specifically, the feature importance of each feature can be calculated using information gain, Gini index, or gradient value. The machine learning in this application dynamically adjusts the weights of the base models to adjust their predictive ability on specific features through various mechanisms, including: calculating the Gini impurity reduction or information gain (entropy reduction) to quantify the contribution of features to prediction; evaluating the performance of each base model on a specific feature subset based on a reserved validation set and dynamically weighting them; using a two-stream neural network to extract cross-sectional morphology and load change features respectively, fusing them through tensor product, and adjusting their weights based on the performance of the base models on the fused spatiotemporal features.

[0116] Furthermore, assume that the outputs of the three base models are as follows: , , After determining the weights of each base model, the final prediction output of the ensemble learning model is:

[0117]

[0118] in, , , These are the weights of the three base models.

[0119] When a base model has a strong predictive ability for a specific feature, its weight is increased; conversely, its weight is decreased. By dynamically adjusting the contribution weights of each base model, the model can focus more on important features, thereby improving the overall accuracy of prediction.

[0120] Furthermore, an objective optimization function is constructed based on three aspects: structural safety factor, material cost, and construction feasibility. A crossover and mutation operator is designed to generate new population data, and new individuals are selected to replace old ones. Based on a Bayesian optimization framework, a Gaussian process function model is constructed, and iterative optimization is performed by optimizing the data acquisition function. Multi-objective optimization balances safety factor, cost, and feasibility, while Bayesian optimization tunes hyperparameters and network structure to obtain the final ensemble learning model. Specifically, when constructing the crossover and mutation operator, crossover and mutation operators from genetic algorithms are used to generate new population data. Crossover operations can employ single-point crossover, double-point crossover, or uniform crossover, while mutation operations can randomly change certain decision variables.

[0121] S5: Based on the structural safety factor, the material cost, and the construction feasibility, a mixed integer programming algorithm is used to generate a reinforcement configuration scheme for the variable cross-section composite pile.

[0122] As one optional embodiment, the step of generating a reinforcement configuration scheme for the variable cross-section composite pile based on the structural safety factor, the material cost, and the construction feasibility using a mixed integer programming algorithm includes:

[0123] Obtain flexural capacity, crack width, and reinforcement ratio;

[0124] Using the bending capacity, crack width, reinforcement ratio, structural safety factor, material cost, and construction feasibility as constraints, an integer programming model is constructed by representing the quantity and distribution of steel bars with binary variables and constructing an objective function.

[0125] The integer programming model is input into a hybrid integer programming solver to obtain the optimal reinforcement configuration scheme.

[0126] Specifically, the structural safety factor, material cost, and construction feasibility predicted by the integrated learning model are used as constraints. Simultaneously, a set of nonlinear constraint equations is established, incorporating national standards for flexural bearing capacity, crack width, and reinforcement ratio. The flexural bearing capacity, crack width, and reinforcement ratio are calculated based on the variable cross-sectional shape parameters obtained in step S3. The calculation formula is:

[0127]

[0128] in, The design compressive strength of concrete. This represents the cross-sectional area of ​​the reinforcing steel. The effective height of the component, This is the anchorage length of the reinforcing bar;

[0129] Crack width The calculation formula is usually based on crack control theory, and the formula is:

[0130]

[0131] in, For the stress of the steel reinforcement, For reinforcement ratio, It is a constant;

[0132] Minimum reinforcement ratio The calculation formula is:

[0133]

[0134] in, This represents the yield strength of the steel reinforcement.

[0135] Furthermore, the quantity and distribution of steel reinforcement in the reinforcement scheme are represented by binary variables, and an integer programming model is constructed. The integer programming model adjusts the total amount of steel reinforcement by minimizing the objective function, the formula of which is:

[0136]

[0137] in, Let J be the cross-sectional area of ​​the j-th type of steel reinforcement. It is a binary variable.

[0138] Furthermore, the integer programming model is solved using a mixed integer programming (MIP) solver. Based on the current solution, a new reinforcement configuration scheme is calculated, and the configuration is gradually adjusted to meet the constraints, ultimately obtaining a reinforcement configuration scheme that satisfies the design requirements. Then, based on the structural parameters in the obtained optimal reinforcement configuration scheme, these parameters are input into CAD to automatically generate vector graphics, resulting in corresponding construction drawings. Based on the reinforcement configuration scheme and finite element analysis results, a construction guidance document is output, realizing the automatic generation of reinforcement design and construction drawings, greatly improving design efficiency and reducing human error. This invention achieves efficient and accurate pile foundation reinforcement optimization, automatically generating the optimal reinforcement scheme, ensuring structural safety, cost control, and construction feasibility, and greatly improving design efficiency and reliability.

[0139] Compared to existing technologies, this invention, by acquiring multi-source data on variable cross-section composite piles, fully considers various influencing factors, improving the comprehensiveness of reinforcement schemes. By extracting key and spatiotemporal features from the multi-source data, the most discriminative features can be obtained for subsequent prediction and optimization, enhancing design reliability. Based on the multi-source data, three-dimensional parametric modeling and nonlinear finite element analysis are performed to construct the variable cross-section morphology and stress cloud diagram of the pile body, obtaining variable cross-section morphology parameters, further improving design accuracy and ensuring the structural safety of the pile foundation under different working conditions. By using a preset integrated learning model for multi-objective prediction based on key features, spatiotemporal features, and variable cross-section morphology parameters, structural safety factor, material cost, and construction feasibility are obtained. Then, based on the structural safety factor, material cost, and construction feasibility, a mixed integer programming algorithm is used to generate the optimal reinforcement configuration scheme. This can automatically adjust and optimize the reinforcement scheme under complex and variable design requirements, meeting multiple objectives of structural safety, cost control, and construction feasibility, improving design efficiency, reducing human error, and enhancing the accuracy of variable cross-section composite pile reinforcement design.

[0140] Accordingly, the present invention also provides a machine learning-based variable cross-section composite pile reinforcement design system, which can realize all the processes of the machine learning-based variable cross-section composite pile reinforcement design method in the above embodiments.

[0141] Please see Figure 2 , Figure 2 This is a schematic diagram of a machine learning-based variable cross-section composite pile reinforcement design system provided in an embodiment of the present invention. The machine learning-based variable cross-section composite pile reinforcement design system includes:

[0142] Data acquisition module 201 is used to acquire multi-source data of variable cross-section composite piles;

[0143] Feature extraction module 202 is used to extract features from the multi-source data to obtain key features and spatiotemporal features;

[0144] The variable cross-section processing module 203 is used to perform three-dimensional parametric modeling and nonlinear finite element analysis based on the multi-source data, construct the variable cross-section shape and stress cloud diagram of the pile body, and obtain the variable cross-section shape parameters.

[0145] The model prediction module 204 is used to perform multi-objective prediction based on the key features, the spatiotemporal features and the variable cross-section morphology parameters using a preset integrated learning model, so as to obtain the structural safety factor, material cost and construction feasibility.

[0146] The reinforcement optimization module 205 is used to generate a reinforcement configuration scheme for the variable cross-section composite pile based on the structural safety factor, the material cost and the construction feasibility, using a mixed integer programming algorithm.

[0147] Preferably, the base models of the ensemble learning model include: a random forest model, a gradient boosting decision tree model, and a deep residual network model.

[0148] Preferably, the multi-source data includes: engineering geological parameters, load condition parameters, material performance parameters, cross-sectional parameters, and historical engineering data.

[0149] Preferably, the acquisition of multi-source data for the variable cross-section composite pile includes:

[0150] The engineering geological parameters of the variable cross-section composite pile were obtained through ground-penetrating radar detection, in-situ testing, and dynamic monitoring of groundwater level.

[0151] By designing load spectrum decomposition and simulating construction process loads, the load condition parameters of variable cross-section composite piles are obtained.

[0152] Material performance parameters of variable cross-section composite piles are obtained through concrete material testing and steel reinforcement performance testing.

[0153] The cross-sectional parameters of the variable cross-section composite pile are obtained by using variable cross-section geometric parameterization and cross-section stiffness parameter calculation.

[0154] Retrieve historical engineering data of variable cross-section composite piles from the engineering database.

[0155] Preferably, the feature extraction module 202 is specifically used for:

[0156] The multi-source data is normalized to obtain normalized data;

[0157] Based on the temporal and spatial correlation of the normalized data, a spatiotemporal correlation matrix is ​​constructed to obtain spatiotemporal features.

[0158] Principal component analysis was used to reduce the dimensionality of the normalized data to obtain feature data.

[0159] Based on the feature data, the mutual information entropy algorithm is used to calculate the mutual information between each feature data and the target variable, and the feature data with the largest mutual information are selected to obtain the key features.

[0160] Preferably, the formula for the time correlation is:

[0161]

[0162] in, The temporal correlation between the i-th feature and the j-th feature at time t. Let be the covariance between the i-th feature and the j-th feature at time t. , Let be the variances of the i-th and j-th features at time t, respectively.

[0163] The formula for the spatial correlation is:

[0164]

[0165] in, Let represent the spatial correlation between the i-th feature and the j-th feature at spatial location s. Let be the covariance between the i-th feature and the j-th feature at spatial location s. Let be the variances of the i-th and j-th features at spatial location s, respectively.

[0166] Preferably, the machine learning-based variable cross-section composite pile reinforcement design system is also used for:

[0167] Based on spatiotemporal features, a two-stream neural network is constructed to process temporal and spatial features separately, and tensor product operation is used to perform deep interaction between temporal and spatial features to obtain interactive features;

[0168] The key features of the variable cross-section composite pile, the variable cross-section morphological parameters, and the interaction features are used as input features of the model to train the base model of the integrated learning model in parallel.

[0169] Calculate the feature importance of the input features, and determine the weight of each base model based on the feature importance;

[0170] Based on the base model and its weights, an initial ensemble learning model is obtained.

[0171] Based on the initial ensemble learning model, multi-objective optimization and Bayesian optimization are performed to obtain the final ensemble learning model; wherein, the objective function of multi-objective optimization is constructed based on the structural safety factor, material cost and construction feasibility.

[0172] Preferably, the reinforcement optimization module 205 is specifically used for:

[0173] Obtain flexural capacity, crack width, and reinforcement ratio;

[0174] Using the bending capacity, crack width, reinforcement ratio, structural safety factor, material cost, and construction feasibility as constraints, an integer programming model is constructed by representing the quantity and distribution of steel bars with binary variables and constructing an objective function.

[0175] The integer programming model is input into a hybrid integer programming solver to obtain the optimal reinforcement configuration scheme.

[0176] Preferably, the integer programming model adjusts the total amount of steel reinforcement by minimizing the objective function, the objective function formula being:

[0177]

[0178] in, Let J be the cross-sectional area of ​​the j-th type of steel reinforcement. It is a binary variable.

[0179] In specific implementation, the working principle, control process and technical effects of the machine learning-based variable cross-section composite pile reinforcement design system provided in this embodiment are the same as those of the machine learning-based variable cross-section composite pile reinforcement design method in the above embodiments, and will not be repeated here.

[0180] This invention provides a machine learning-based method and system for designing reinforcement in variable cross-section composite piles. Its advantages include: acquiring multi-source data on variable cross-section composite piles, fully considering various influencing factors, and improving the comprehensiveness of the reinforcement scheme; extracting key and spatiotemporal features from the multi-source data to obtain the most discriminative features for subsequent prediction and optimization, thus improving design reliability; performing three-dimensional parametric modeling and nonlinear finite element analysis based on the multi-source data to construct the variable cross-section morphology and stress cloud diagram of the pile body, obtaining variable cross-section morphology parameters, further improving design accuracy, and ensuring the structural safety of the pile foundation under different working conditions; using a preset ensemble learning model for multi-objective prediction based on key features, spatiotemporal features, and variable cross-section morphology parameters to obtain structural safety factor, material cost, and construction feasibility; and then, based on the structural safety factor, material cost, and construction feasibility, using a mixed integer programming algorithm to generate the optimal reinforcement configuration scheme. This can automatically adjust and optimize the reinforcement scheme under complex and variable design requirements, meeting multiple objectives of structural safety, cost control, and construction feasibility, improving design efficiency, reducing human error, and enhancing the accuracy of variable cross-section composite pile reinforcement design.

[0181] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A machine learning-based method for designing reinforcement of variable cross-section composite piles, characterized in that, include: Acquire multi-source data of variable cross-section composite piles; Feature extraction is performed on the multi-source data to obtain key features and spatiotemporal features; Based on the multi-source data, three-dimensional parametric modeling and nonlinear finite element analysis are performed to construct the variable cross-sectional shape and stress cloud diagram of the pile body, and obtain the variable cross-sectional shape parameters. Based on the key features, the spatiotemporal features, and the variable cross-section morphology parameters, a preset integrated learning model is used to perform multi-objective prediction to obtain the structural safety factor, material cost, and construction feasibility. Based on the structural safety factor, the material cost, and the construction feasibility, a mixed integer programming algorithm is used to generate a reinforcement configuration scheme for the variable cross-section composite pile. The base models of the ensemble learning model include: random forest model, gradient boosting decision tree model and deep residual network model; The training methods for the ensemble learning model include: Based on spatiotemporal features, a two-stream neural network is constructed to process temporal and spatial features separately, and tensor product operation is used to perform deep interaction between temporal and spatial features to obtain interactive features; The key features of the variable cross-section composite pile, the variable cross-section morphological parameters, and the interaction features are used as input features of the model to train the base model of the integrated learning model in parallel. Calculate the feature importance of the input features, and determine the weight of each base model based on the feature importance; Based on the base model and its weights, an initial ensemble learning model is obtained. Based on the initial ensemble learning model, multi-objective optimization and Bayesian optimization are performed to obtain the final ensemble learning model; wherein, the objective function of multi-objective optimization is constructed based on the structural safety factor, material cost and construction feasibility.

2. The machine learning-based reinforcement design method for variable cross-section composite piles as described in claim 1, characterized in that, The multi-source data includes: engineering geological parameters, load condition parameters, material performance parameters, cross-sectional parameters, and historical engineering data.

3. The machine learning-based reinforcement design method for variable cross-section composite piles as described in claim 2, characterized in that, The acquisition of multi-source data for variable cross-section composite piles includes: The engineering geological parameters of the variable cross-section composite pile were obtained through ground-penetrating radar detection, in-situ testing, and dynamic monitoring of groundwater level. By designing load spectrum decomposition and simulating construction process loads, the load condition parameters of variable cross-section composite piles are obtained. Material performance parameters of variable cross-section composite piles are obtained through concrete material testing and steel reinforcement performance testing. The cross-sectional parameters of the variable cross-section composite pile are obtained by using variable cross-section geometric parameterization and cross-section stiffness parameter calculation. Retrieve historical engineering data of variable cross-section composite piles from the engineering database.

4. The machine learning-based reinforcement design method for variable cross-section composite piles as described in claim 1, characterized in that, The step of extracting features from the multi-source data to obtain key features and spatiotemporal features includes: The multi-source data is normalized to obtain normalized data; Based on the temporal and spatial correlation of the normalized data, a spatiotemporal correlation matrix is ​​constructed to obtain spatiotemporal features. Principal component analysis was used to reduce the dimensionality of the normalized data to obtain feature data. Based on the feature data, the mutual information entropy algorithm is used to calculate the mutual information between each feature data and the target variable, and the feature data with the largest mutual information are selected to obtain the key features.

5. The machine learning-based reinforcement design method for variable cross-section composite piles as described in claim 4, characterized in that, The formula for the time correlation is: in, The temporal correlation between the i-th feature and the j-th feature at time t. Let be the covariance between the i-th feature and the j-th feature at time t. , Let be the variances of the i-th and j-th features at time t, respectively. The formula for the spatial correlation is: in, Let represent the spatial correlation between the i-th feature and the j-th feature at spatial location s. Let be the covariance between the i-th feature and the j-th feature at spatial location s. Let be the variances of the i-th and j-th features at spatial location s, respectively.

6. The machine learning-based reinforcement design method for variable cross-section composite piles as described in claim 1, characterized in that, Based on the structural safety factor, the material cost, and the construction feasibility, a mixed integer programming algorithm is used to generate a reinforcement configuration scheme for the variable cross-section composite pile, including: Obtain flexural capacity, crack width, and reinforcement ratio; Using the bending capacity, crack width, reinforcement ratio, structural safety factor, material cost, and construction feasibility as constraints, an integer programming model is constructed by representing the quantity and distribution of steel bars with binary variables and constructing an objective function. The integer programming model is input into a hybrid integer programming solver to obtain the optimal reinforcement configuration scheme.

7. The machine learning-based reinforcement design method for variable cross-section composite piles as described in claim 6, characterized in that, The integer programming model adjusts the total amount of steel reinforcement by minimizing the objective function, the formula of which is: in, Let J be the cross-sectional area of ​​the j-th type of steel reinforcement. It is a binary variable.

8. A machine learning-based variable cross-section composite pile reinforcement design system, characterized in that, include: The data acquisition module is used to acquire multi-source data of variable cross-section composite piles; The feature extraction module is used to extract features from the multi-source data to obtain key features and spatiotemporal features; The variable cross-section processing module is used to perform three-dimensional parametric modeling and nonlinear finite element analysis based on the multi-source data, construct the variable cross-section shape and stress cloud diagram of the pile body, and obtain the variable cross-section shape parameters. The model prediction module is used to perform multi-objective prediction based on the key features, the spatiotemporal features and the variable cross-section morphology parameters using a preset integrated learning model, so as to obtain the structural safety factor, material cost and construction feasibility. The reinforcement optimization module is used to generate a reinforcement configuration scheme for the variable cross-section composite pile based on the structural safety factor, the material cost, and the construction feasibility, using a mixed integer programming algorithm. The base models of the ensemble learning model include: random forest model, gradient boosting decision tree model and deep residual network model; The training methods for the ensemble learning model include: Based on spatiotemporal features, a two-stream neural network is constructed to process temporal and spatial features separately, and tensor product operation is used to perform deep interaction between temporal and spatial features to obtain interactive features; The key features of the variable cross-section composite pile, the variable cross-section morphological parameters, and the interaction features are used as input features of the model to train the base model of the integrated learning model in parallel. Calculate the feature importance of the input features, and determine the weight of each base model based on the feature importance; Based on the base model and its weights, an initial ensemble learning model is obtained. Based on the initial ensemble learning model, multi-objective optimization and Bayesian optimization are performed to obtain the final ensemble learning model; wherein, the objective function of multi-objective optimization is constructed based on the structural safety factor, material cost and construction feasibility.

Images