Threaded cast-in-place pile uplift bearing capacity prediction method and system based on machine learning

Abstract

The invention relates to the technical field of cast-in-place pile bearing capacity prediction, in particular to a threaded cast-in-place pile uplift bearing capacity prediction method and system based on machine learning. The prediction method comprises the following steps: S1, acquiring test data; s2, constructing a machine learning model according to the test data; the machine learning model is optimized; and S3, predicting the uplift bearing capacity of the threaded cast-in-place pile by using the optimized model. According to the method, the machine learning algorithm is utilized to establish the threaded cast-in-place pile uplift bearing capacity prediction model, the pile body uplift bearing capacity prediction model can be utilized to rapidly predict the uplift bearing capacity of other pile bodies of the same type in the area on the basis of a small number of field tests, the inspection time of pile foundation engineering is effectively saved, and the project economic cost is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of prediction of bearing capacity of cast-in-place piles, and in particular to a method and system for predicting the pull-out bearing capacity of threaded cast-in-place piles based on machine learning. Background Art

[0002] Threaded cast-in-place pile (screw cast-in-place pile) is a variable cross-section special-shaped cast-in-place pile composed of a cylindrical pile body at the top and a threaded pile body at the bottom. The lengths of the upper and lower pile sections can be adjusted according to the soil conditions of the foundation, and the outer diameter of the lower threaded pile body is the same as the diameter of the upper cylindrical pile body. The advantages of this type of pile are: good environmental protection effect; reliable pile body quality; simplified construction process and high efficiency; wide application range, and can be used as the pile foundation of pile foundation.

[0003] The vertical pull-out bearing capacity of threaded cast-in-place piles is affected by many factors, such as the geometric parameters of the pile body, thread parameters, and the physical and mechanical parameters of the soil around the pile. It is of great significance to design a pile foundation that meets the vertical bearing capacity requirements of the project according to the actual situation. On-site static load tests are an indispensable inspection method for most pile foundation projects. However, due to the high time and economic costs of conducting on-site static load tests, it is difficult to conduct large-scale tests. Sampling is often used to test the bearing capacity of piles. Due to the uncertainty of the site environment and stratum conditions, a small number of on-site static load tests cannot determine the vertical bearing capacity of all piles in the project site.

[0004] The Chinese patent application with publication number CN117371279A discloses a method for verifying the stability and bearing capacity of pile foundation based on the dissolution of limestone at the end of the cast-in-place pile foundation in karst development areas, including the following steps: S1 exploration and sampling; S2 analysis of sample indoor tests to determine the dissolution performance and physical and mechanical parameters of limestone; S3 monitoring the groundwater level in combination with hydrogeological conditions; S4 using data for numerical simulation to evaluate the stability of limestone; S5 determining the structural parameters of the pile foundation and estimating the load based on the results, engineering design and specifications; S6 evaluating the bearing capacity of the pile foundation by setting up instruments to measure the change in pile side resistance and calculating the friction angle and bearing capacity; S7 setting up a static load test instrument to apply loads and monitor the response and deformation characteristics to evaluate the stability of the pile foundation; S8 analyzing the modified bearing capacity of the pile foundation by limestone dissolution on the interaction between piles and soil to determine the bearing capacity of the pile foundation. This technical solution integrates multiple data collection methods to improve the reliability of the verification results, and by analyzing the interaction between limestone dissolution and piles and soil, the bearing capacity of the pile foundation is more accurately evaluated.

[0005] The Chinese patent application with publication number CN118643713A discloses a numerical simulation method for the compressive ultimate bearing capacity test of post-grouting bored piles, which includes the following steps: step 0, selecting the test pile position and carrying out the test pile construction; step 1, generating the initial stress influence area of ​​the numerical specimen; step 2, setting the parameters of the numerical specimen model; step 3, meshing and setting the boundary conditions of the numerical specimen model; step 4, establishing the numerical simulation analysis working condition of the compressive ultimate bearing capacity test; step 5, determining the parameters of the pile top loading test; step 6, carrying out the numerical simulation test of the compressive ultimate bearing capacity test and obtaining the numerical simulation test results. The numerical simulation method of the technical solution has an error within 15%, has certain feasibility, can avoid the error and discreteness caused by human and environmental factors in the field experiment, and can provide a reference for similar field construction.

[0006] Although the existing technologies (CN117371279A, CN118643713A) can test the bearing capacity of pile foundations, the pile types tested by the existing technologies are very different from those of threaded cast-in-place piles, and the existing technologies are mainly used to predict the compressive bearing capacity of pile foundations. Therefore, the existing technologies (CN117371279A, CN118643713A) cannot be used to predict the vertical pull-out bearing capacity of threaded cast-in-place piles. Summary of the invention

[0007] The purpose of the present invention is to overcome the shortcomings that the existing simulation test method for the compressive ultimate bearing capacity of cast-in-place piles cannot be used to predict the vertical pull-out bearing capacity of threaded cast-in-place piles, and the on-site static load test time and economic cost are long and difficult to carry out on a large scale, and to provide a method and system for predicting the pull-out bearing capacity of threaded cast-in-place piles based on machine learning.

[0008] In order to achieve the above-mentioned object of the invention, the present invention provides the following technical solutions:

[0009] The present invention provides a method for predicting the pull-out bearing capacity of a threaded bored pile based on machine learning, comprising the following steps: S1, acquiring test data; S2, constructing a machine learning model according to the test data; and optimizing the machine learning model; S3, predicting the pull-out bearing capacity of the threaded bored pile using the optimized model.

[0010] According to a preferred implementation manner, the test data at least includes: pull-out bearing capacity of threaded cast-in-place piles, length of threaded piles, length of threaded sections, pile body diameter, thread spacing, thread width, thread thickness, thread inclination, cohesion of soil around piles, and internal friction angle of soil around piles.

[0011] According to a preferred implementation, step S2 includes: S21, inputting the parameters of the training set into a machine learning model based on different algorithms to obtain several prediction results; S22, calculating the average result of several prediction results; S23, using the training set and the average result to optimize the training of a specific model to obtain a final training model; S24, inputting the parameters of the training set into the final training model to obtain a final training result; inputting the parameters of the test set into the final training model to obtain a final test result; and determining whether to return to step S23 based on the final training result and the final test result.

[0012] According to a preferred implementation, in step S23, the GWO algorithm is used to optimize the hyperparameters of the specific model.

[0013] According to a preferred implementation, in step S21, the parameters of the training set are input based on the first model and the second model to obtain the prediction result 1 of the first model and the prediction result 2 of the second model respectively; in step S22, the average result of the prediction result 1 and the prediction result 2 is calculated.

[0014] According to a preferred embodiment, the condition for determining whether to return to step S23 in step S24 is to determine whether the error between the final result of the training and the final result of the test exceeds a threshold. If so, return to step S23 to continue optimization. If not, the final training model has completed training and there is no need to return to step S23.

[0015] According to a preferred implementation, step S2 includes: constructing several machine learning models based on different algorithms; and using the test data to train and optimize the several machine learning models.

[0016] According to a preferred implementation manner, step S3 uses several optimized machine learning models to predict the pull-out bearing capacity of the threaded cast-in-place piles, obtains several prediction results, and performs weighted calculation on the several prediction results to obtain a unique prediction result.

[0017] The present invention also provides a threaded cast-in-place pile pull-out bearing capacity prediction system based on machine learning, comprising: a data acquisition unit for acquiring test data; a model training unit for constructing a machine learning model and optimizing the machine learning model using the test data; and a prediction unit for predicting the pull-out bearing capacity of the threaded cast-in-place pile using the optimized machine learning model.

[0018] The present invention also provides a method for predicting the pull-out bearing capacity of threaded bored piles based on machine learning, comprising the following steps: step one, obtaining test data; step two, constructing several machine learning models based on the test data; and optimizing the machine learning models; step three, using the several optimized machine learning models to predict the pull-out bearing capacity of the threaded bored piles, obtaining several prediction results, and performing weighted calculation on the several prediction results to obtain a unique prediction result.

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] The present invention uses a machine learning algorithm to establish a prediction model for the pull-out bearing capacity of threaded cast-in-place piles. The pile pull-out bearing capacity prediction model can be used to quickly predict the pull-out bearing capacity of other similar piles in the area based on a small amount of field tests, effectively saving the inspection time of pile foundation projects and reducing the economic cost of the project. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a basic flow chart of the method for predicting the pull-out bearing capacity of a threaded cast-in-place pile of the present invention;

[0022] Figure 2 It is a flow chart of a method for predicting the pull-out bearing capacity of a threaded cast-in-place pile preferably based on machine learning according to the present invention;

[0023] Figure 3 It is a structural diagram of the GMDH algorithm in the present invention;

[0024] Figure 4 Schematic diagram of the neuron selection process of the GMDH model in the present invention;

[0025] Figure 5 It is a schematic diagram of the optimization principle of the GWO algorithm in the present invention;

[0026] Figure 6 A schematic diagram of optimization training of a machine learning model according to a preferred embodiment of the present invention;

[0027] Figure 7 The prediction results and errors of the pull-out bearing capacity of threaded cast-in-place piles using the GWO-GMDH model;

[0028] Figure 8 The prediction results and errors of the pull-out bearing capacity of threaded cast-in-place piles by the unoptimized GMDH model;

[0029] Fig. 9 The results and errors of the prediction of the pull-out bearing capacity of threaded cast-in-place piles using the GEP model are shown. DETAILED DESCRIPTION

[0030] The present invention is further described in detail below in conjunction with test examples and specific implementation methods. However, this should not be understood as the scope of the above subject matter of the present invention being limited to the following embodiments, and all technologies realized based on the content of the present invention belong to the scope of the present invention.

[0031] Unless otherwise specified, the expressions "first", "second", "third", etc. in the description of the specific embodiments of the present invention are merely used to distinguish the description of the same or similar components and should not be understood as emphasizing or implying the relative importance of specific components.

[0032] In addition, in the description of the embodiments of the present invention, "several", "plurality" and "a number" represent at least 2. It can be any number such as 2, 3, 4, 5, 6, 7, 8, 9, and even more than 9.

[0033] Example 1

[0034] This embodiment provides a method for predicting the pull-out bearing capacity of threaded cast-in-place piles based on machine learning. Figure 1 The method for predicting the pull-out bearing capacity of threaded bored piles based on machine learning includes the following steps: S1, obtaining test data; S2, building a machine learning model based on the test data; and optimizing the machine learning model; S3, using the optimized model to predict the pull-out bearing capacity of threaded bored piles.

[0035] This embodiment uses a machine learning algorithm to establish a prediction model for the pull-out bearing capacity of threaded cast-in-place piles. The pile pull-out bearing capacity prediction model can be used to quickly predict the pull-out bearing capacity of other similar piles in the area based on a small number of field tests, effectively saving the inspection time of pile foundation projects and reducing the economic cost of the project.

[0036] Preferably, the test data may include: pull-out bearing capacity of threaded cast-in-place piles, length of threaded piles, length of threaded sections, pile diameter, thread pitch, thread width, thread thickness, thread inclination, cohesion of soil around piles, and internal friction angle of soil around piles.

[0037] Example 2

[0038] This embodiment is a further improvement of embodiment 1, and the repeated contents are not repeated. In this embodiment, step S2 includes: S21, inputting the parameters of the training set into the machine learning model based on different algorithms to obtain several prediction results; S22, calculating the average result of several prediction results; S23, optimizing the specific model using the training set and the average result to obtain the final training model; S24, inputting the parameters of the training set into the final training model to obtain the final training result; inputting the parameters of the test set into the final training model to obtain the final test result; determining whether to return to step S23 according to the final training result and the final test result.

[0039] Preferably, in step S23, a GWO algorithm is used to optimize the hyperparameters of the specific model.

[0040] Preferably, in step S21, the parameters of the training set are input based on the first model and the second model to obtain the prediction result 1 of the first model and the prediction result 2 of the second model respectively; in step S22, the average result of the prediction result 1 and the prediction result 2 is calculated.

[0041] Preferably, the condition for determining whether to return to step S23 in step S24 is to determine whether the error between the final result of training and the final result of testing exceeds a threshold. If so, return to step S23 to continue optimization. If not, the final training model has completed training and there is no need to return to step S23.

[0042] Example 3

[0043] This embodiment is a further improvement of Embodiment 1 and Embodiment 2, and the repeated contents are not repeated here. This embodiment provides a method for predicting the pull-out bearing capacity of threaded cast-in-place piles based on machine learning, comprising the following steps: Step 1, obtaining test data; Step 2, constructing several machine learning models according to the test data; and optimizing the machine learning models; Step 3, using the several machine learning models that have been optimized to predict the pull-out bearing capacity of threaded cast-in-place piles, obtaining several prediction results, and performing weighted calculation on the several prediction results to obtain a unique prediction result.

[0044] Preferably, the several machine learning models constructed in step 2 of this embodiment can be constructed based on different machine learning algorithms, or can be constructed based on different machine learning algorithms.

[0045] Preferably, this embodiment constructs several machine learning models to predict the pull-out bearing capacity of threaded cast-in-place piles, and then comprehensively evaluates the prediction results of different machine learning models to obtain a final and unique prediction result.

[0046] Preferably, the method for processing the multiple prediction results in step three can also adopt the average method, median method, mode method, truncated mean method, comprehensive scoring method and standardized scoring method, etc., to convert the multiple prediction results into a final and unique prediction result.

[0047] Example 4

[0048] This embodiment is a further improvement of embodiment 3, and the repeated contents are not repeated here. Figure 2In this embodiment, three machine learning models are constructed, namely the GMDH model, the GEP model and the XGBoost model, and the GMDH model, the GEP model and the XGBoost model are used to generate a machine learning integrated model, which is then used to predict the pull-out bearing capacity of threaded cast-in-place piles.

[0049] See also Figure 2 In this embodiment, the pull-out bearing capacity test data of the threaded cast-in-place pile is first obtained. Preferably, the test data may include: pull-out bearing capacity of the threaded cast-in-place pile, length of the threaded pile, length of the threaded section, diameter of the pile body, thread pitch, thread width, thread thickness, thread inclination, cohesion of the soil around the pile, internal friction angle of the soil around the pile, and other parameters.

[0050] Preferably, after obtaining the test data, the obtained test data is randomly divided into a training set and a test set. Preferably, the ratio of the training set to the test set can be 8:2.

[0051] Preferably, after obtaining the training set and the test set, the GMDH algorithm, the GEP algorithm and the XGBoost algorithm are respectively combined to construct the GMDH model, the GEP model and the XGBoost model, and the GWO algorithm is used to optimize the hyperparameters of the GMDH model, the GEP model and the XGBoost model.

[0052] See also Figure 3 and Figure 4 , the general implementation process of the GMDH algorithm is:

[0053] (1) Pair the neurons in the input layer with other neurons in the layer to generate neurons in the first hidden layer. New neurons will be generated in the next layer only when the relationship between each pair of neurons in the previous layer satisfies the Ivakhnenko polynomial. The Ivakhnenko polynomial can be expressed as:

[0054]

[0055] Among them, H k,m represents the mth neuron in the kth layer; H k-1,i and H k-1,j Represent the output values ​​of the i-th and j-th neurons in the (k-1) layer respectively; C1~C6 are polynomial coefficients.

[0056] (2) Neurons are selected based on the root mean square error (RMSE) of the output values ​​of each pair of neurons. Coupled neurons with RMSE values ​​higher than a preset threshold are eliminated, and coupled neurons that meet the requirements are retained.

[0057] (3) Select the neuron with the smallest RMSE value as the output value of the algorithm.

[0058] The general implementation process of the GEP algorithm is:

[0059] (1) Select an appropriate fitness function, such as root mean square error (RMSE);

[0060] (2) Define a series of function geometry, such as arithmetic operators, +, -, *, / , etc.; elementary mathematical functions, ln, sin, cos, exp, etc.; other functions, inv, max, min, etc.; Boolean operators, etc.; relational operators, <, >, =, ≠, ≤, ≥, etc.; conditional operators, if, or, and, etc.;

[0061] (3) Establishing the genetic structure, such as head-tail structure, number of genes, etc.;

[0062] (4) Select genetic manipulation;

[0063] (5) Select link function, etc.

[0064] The basic principle of XGBoost algorithm is:

[0065] The model output is as follows:

[0066]

[0067] For the k-th iteration, the following objective function is used:

[0068]

[0069] By using the second-order Taylor expansion and removing the constant term, we can obtain:

[0070]

[0071] The tree is defined using leaf scores and index functions, i.e. So,

[0072] For a certain tree structure:

[0073]

[0074] Preferably, after obtaining the training set, the GMDH model established based on the GMDH algorithm, the GEP model established based on the GEP algorithm, and the XGBoost model established based on the XGBoost algorithm are trained using the training set respectively, and during the training process, the GWO algorithm is used to perform hyperparameter optimization on the GMDH model, the GEP model, and the XGBoost model.

[0075] The GWO algorithm searches for the best hyperparameters of the three models. The best performance parameters are selected to establish models for predicting the pull-out bearing capacity of threaded cast-in-place piles. Figure 5, the basic principle of GWO optimization is:

[0076] The GWO algorithm simulates the hierarchy mechanism and hunting behavior of gray wolf populations in nature. It simulates social hierarchy through four types of gray wolves (α, β, δ, ω), and simulates the hunting behavior of wolves through the process of tracking, encircling, chasing, and attacking prey, so as to achieve the purpose of optimizing search.

[0077] A population of gray wolves is randomly generated in the search space, and the solution with the best fitness value, the second best solution, and the third best solution in the population are regarded as α, β, and δ wolves respectively, and the remaining solutions are regarded as ω wolves. Then α, β, and δ wolves are responsible for leading, and ω wolves follow α, β, and δ wolves to complete hunting optimization by searching, surrounding, and attacking prey, and finally obtain the optimal solution.

[0078] Preferably, after the GMDH model, the GEP model and the XGBoost model are optimized, the three machine learning models are weighted to obtain a machine learning integrated model for predicting the pull-out bearing capacity of threaded cast-in-place piles. Preferably, the specific method of weighted calculation can be to take the average of the prediction results of each model.

[0079] Preferably, the process of predicting the pull-out bearing capacity of threaded cast-in-place piles by machine learning integrated model is:

[0080] Enter a sample data;

[0081] The GMDH model, GEP model and XGBoost model predict the pull-out bearing capacity of threaded cast-in-place piles on the sample data, and obtain the prediction results of the GMDH model, the GEP model and the XGBoost model;

[0082] The prediction results of the GMDH model, the GEP model and the XGBoost model are weighted to obtain the final and unique prediction result.

[0083] Example 5

[0084] This embodiment is a further improvement of Embodiment 4, and the repeated contents are not repeated here. In this embodiment, the machine learning model constructed in step S2 can be a machine learning model among the GMDH model, the GEP model and the XGBoost model, or can be other machine learning models that use the pull-out bearing capacity data of threaded cast-in-place piles for learning and training.

[0085] Preferably, in this embodiment, only one machine learning model may be constructed for training and optimization in step S2. Only one optimized machine learning model may be used to predict the pull-out bearing capacity of the threaded cast-in-place pile in step S3.

[0086] Preferably, in step S3, when the pull-out bearing capacity of the threaded cast-in-place pile is predicted using the optimized machine learning model, the prediction result can be obtained through one prediction or through multiple predictions. Preferably, the method of obtaining the prediction result through multiple predictions can include: average method, weighted average method, median method, mode method, truncated mean method, comprehensive scoring method and standardized scoring method.

[0087] Example 6

[0088] This embodiment is a further improvement of Embodiment 5, and the repeated contents are not repeated here. In this embodiment, step S2 includes:

[0089] S21. Input the parameters of the training set into the first model and the second model to obtain the prediction result 1 of the first model and the prediction result 2 of the second model.

[0090] S22, calculate the average result of prediction result 1 and prediction result 2.

[0091] S23, using the training set and the average result to optimize the specific model and obtain the final training model. Preferably, in step S23, the GWO algorithm is used to optimize the hyperparameters of the specific model.

[0092] S24, input the parameters of the training set into the final training model to obtain the final training result; input the parameters of the test set into the final training model to obtain the final test result; determine whether to return to step S23 according to the final training result and the final test result.

[0093] Preferably, the condition for determining whether to return to step S23 in step S24 is to determine whether the error between the final result of training and the final result of testing exceeds a threshold. If so, return to step S23 to continue optimization. If not, the final training model has completed training and there is no need to return to step S23.

[0094] Preferably, the first model, the second model and the specific model are machine learning models that are used to learn and train the pull-out bearing capacity data of threaded cast-in-place piles according to different algorithms.

[0095] See also Figure 6 Preferably, the first model may be a GEP model, the second model may be a GMDH model, and the specific model may be an XGBoost model.

[0096] Step S2 may include:

[0097] S21. Input the parameters of the training set into the GEP model to obtain prediction result 1, and input the parameters of the training set into the GMDH model to obtain prediction result 2.

[0098] S22, calculate the average result of prediction result 1 and prediction result 2.

[0099] S23. Use the training set and the average result to train the XGBoost model, and use the GWO algorithm to optimize the hyperparameters of the XGBoost model to obtain the final training model.

[0100] S24, input the parameters of the training set into the final training model to obtain the final training result; input the parameters of the test set into the final training model to obtain the final test result; determine whether to return to step S23 according to the final training result and the final test result.

[0101] Preferably, the condition for determining whether to return to step S23 in step S24 is to determine whether the error between the final result of training and the final result of testing exceeds a threshold. If so, return to step S23 to continue optimization. If not, the final training model has completed training and there is no need to return to step S23.

[0102] Example 7

[0103] This embodiment is a further improvement of Embodiment 5, and the repeated contents are not repeated here. In this embodiment, the GMDH model is optimized by using the GWO algorithm, and a GWO-GMDH model is constructed to predict the pull-out bearing capacity of threaded cast-in-place piles.

[0104] The number of hidden layers, the maximum number of neurons in each layer, and the selection pressure in the GMDH algorithm are hyperparameters that must be considered when establishing a prediction model. Reasonable values ​​of hyperparameters can significantly reduce the complexity of the model and improve the accuracy of calculations. Since the GMDH model generally uses the least squares method to calculate the optimal value, the complexity of the model increases with the increase in the number of hidden layers, and each neuron contains less information. In order to fully combine the existing information, a large number of neuron coupling pairs are required for calculation, which greatly increases the calculation time and complexity of the model.

[0105] This embodiment uses the GWO algorithm to optimize the hyperparameter selection process of the GMDH model and determine the most appropriate hyperparameter values ​​for the GMDH model. The steps of using the GWO algorithm to optimize the GMDH model are as follows:

[0106] (1) Initialize the number and range of algorithm hyperparameters, number of iterations, and population size.

[0107] (2) Determine the α, β and δ wolves; specifically, determine the initial positions of the α, β and δ wolves.

[0108] (3) Calculate the fitness value; use the RMSE of the coupled neurons as the algorithm fitness function, calculate the fitness function value of each wolf after each iteration, then update the position of the wolf and recalculate the fitness function value.

[0109] (4) Determine the optimal solution based on the calculation result of the fitness function and define it as the global optimal solution. Then update the positions of the α, β and δ wolves, and determine whether to end the calculation based on whether the algorithm reaches the maximum number of iterations. The GMDH model locally searches for each possible hyperparameter value based on the updated position of the wolf and calculates the corresponding fitness function value. Finally, the hyperparameter corresponding to the optimal fitness function is selected as the hyperparameter of the GMDH model.

[0110] After optimization, the optimal hyperparameter values ​​of the GMDH algorithm are: the number of hidden layers is 7; the maximum number of neurons in each layer is 42; and the selection pressure is 0.53.

[0111] Preferably, this embodiment also constructs a GEP model based on the GEP algorithm to predict the pull-out bearing capacity of the threaded cast-in-place pile. The algorithm steps of the GEP algorithm are briefly described as follows:

[0112] (1) Select an appropriate fitness function, such as root mean square error (RMSE);

[0113] (2) Define a series of function geometry, such as arithmetic operators, +, -, *, / , etc.; elementary mathematical functions, ln, sin, cos, exp, etc.; other functions, inv, max, min, etc.; Boolean operators, etc.; relational operators, <, >, =, ≠, ≤, ≥, etc.; conditional operators, if, or, and, etc.;

[0114] (3) Establishing the genetic structure, such as head-tail structure, number of genes, etc.;

[0115] (4) Select genetic manipulation;

[0116] (5) Select link function, etc.

[0117] Preferably, the prediction results and error of the pull-out bearing capacity of threaded cast-in-place piles using the GWO-GMDH model are as follows: Figure 7 The prediction results and errors of the unoptimized GMDH model for the pull-out bearing capacity of threaded cast-in-place piles are shown in Figure 8 As shown in Figure 2, the hyperparameters of the unoptimized GMDH model are set as follows: the number of hidden layers is set to 5; the maximum number of neurons in each layer is set to 28, and the selection pressure is set to 0.6. The prediction results and errors of the GEP model for the pull-out bearing capacity of threaded cast-in-place piles are shown in Figure 2. Fig. 9 shown.

[0118] according to Figure 7 (a) It can be seen that the R of the GWO-GMDH model 2 The R of the model training set and test set is 0.977, COV, MAPE, RRSE, and MAE are 0.0942, 0.0440, 0.0705, and 0.0338 respectively. 2The values ​​are 0.978 and 0.976 respectively. The performance of the training set and the test set is close, and there is no overfitting or underfitting phenomenon. The relative error between the model prediction value and the actual value is plotted as an error distribution diagram, as shown in Figure 7 As shown in (b), 75.6% of the data points have a relative error within 5%; 91.2% of the data points have an error within 10%; and the relative error of all data points of the GWO-GMDH model is within 20%.

[0119] Depend on Figure 8 (a) It can be seen that the unoptimized GMDH model R 2 The R of the model training set and test set is 0.955, COV, MAPE, RRSE, and MAE are 0.1124, 0.0473, 0.0849, and 0.0398 respectively. 2 The relative errors between the model prediction value and the actual value are plotted as an error distribution diagram, as shown in Figure 2. Figure 8 As shown in (b), it can be found that 52.1% of the data points have a relative error within 5%, 80% of the data points have an error within 10%, 91.2% of the data points have a relative error less than 20%, and all data points have a relative error within 25%.

[0120] Depend on Fig. 9 (a) It can be seen that the GEP model R 2 The R of the model training set and test set is 0.975, COV, MAPE, RRSE, and MAE are 0.1669, 0.0836, 0.1241, and 0.0558 respectively. 2 The values ​​are 0.976 and 0.973 respectively. There is a slight difference between the performance of the model training set and the test set, and it can be considered that there is no overfitting or underfitting phenomenon. The relative error between the model prediction value and the actual value is plotted as an error distribution diagram, such as Fig. 9 As shown in (b), 48.8% of the data points have a relative error within 5%, 75.5% of the data points have an error within 10%, 91.1% of the data points have a relative error less than 20%, and all data points have a relative error within 30%.

[0121] The different prediction model evaluation indicators of the GWO-GMDH model, the unoptimized GMDH model, and the GEP model are compared, and the comparison is shown in Table 1 below.

[0122] Table 1 Comparison of GWO-GMDH with other machine learning models

[0123]

[0124] It can be found from Table 1 that the GWO-GMDH model is worse than the unoptimized GMDH model. Except for the determination coefficient, the GEP model is similar to the GWO-GMDH model, but the other evaluation indicators are not as good as the GWO-GMDH model and the unoptimized GMDH model. 2 , COV, MAPE, RRSE and MAE values ​​increased by 2.3%, 16.1%, 6.4%, 16.5% and 15%, respectively, indicating that the optimization of the GMDH model by the GWO algorithm helps to improve the accuracy of the prediction model.

[0125] Example 8

[0126] The present embodiment provides a threaded cast-in-place pile pull-out bearing capacity prediction system based on machine learning, and the threaded cast-in-place pile pull-out bearing capacity prediction system based on machine learning provided by the present embodiment can be used to implement the threaded cast-in-place pile pull-out bearing capacity prediction method based on machine learning involved in Embodiment 1, Embodiment 2, Embodiment 3, Embodiment 4, Embodiment 5, Embodiment 6 and Embodiment 7.

[0127] Preferably, the threaded cast-in-place pile pull-out bearing capacity prediction system based on machine learning includes: a data acquisition unit, a model training unit and a prediction unit.

[0128] A data acquisition unit is used to acquire test data. A model training unit is used to construct a machine learning model and optimize the machine learning model using the test data. A prediction unit is used to predict the pull-out bearing capacity of the threaded cast-in-place pile using the optimized machine learning model.

[0129] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting the pull-out bearing capacity of threaded cast-in-place piles based on machine learning, characterized in that: The following steps are involved: S1. Acquire test data; and divide the test data into a test set and a training set; S2. construct a machine learning model based on the test data; and optimize the machine learning model; S3. Use the optimized model to predict the pull-out bearing capacity of threaded cast-in-place piles.

2. The method for predicting the pull-out bearing capacity of threaded cast-in-place piles based on machine learning according to claim 1 is characterized in that: The test data at least include: the pull-out bearing capacity of the threaded bored pile, the length of the threaded pile, the length of the threaded section, the pile body diameter, the thread pitch, the thread width, the thread thickness, the thread inclination, the cohesion of the soil around the pile, and the internal friction angle of the soil around the pile.

3. The method for predicting the pull-out bearing capacity of threaded cast-in-place piles based on machine learning according to claim 2 is characterized in that: Step S2 includes: S21, inputting the parameters of the training set into a machine learning model based on different algorithms to obtain several prediction results; S22, calculating the average result of several prediction results; S23, optimizing and training a specific model using the training set and the average result to obtain a final training model; S24, input the parameters of the training set into the final training model to obtain the final training result; input the parameters of the test set into the final training model to obtain the final testing result; determine whether to return to step S23 according to the final training result and the final testing result.

4. The method for predicting the pull-out bearing capacity of threaded cast-in-place piles based on machine learning according to claim 3 is characterized in that: In step S23, the GWO algorithm is used to optimize the hyperparameters of the specific model.

5. The method for predicting the pull-out bearing capacity of threaded cast-in-place piles based on machine learning according to claim 3 is characterized in that: In step S21, the parameters of the training set are input based on the first model and the second model to obtain prediction result 1 of the first model and prediction result 2 of the second model respectively; In step S22, the average result of the prediction result 1 and the prediction result 2 is calculated.

6. A method for predicting the pull-out bearing capacity of threaded cast-in-place piles based on machine learning according to claim 4 or 5, characterized in that: The condition for determining whether to return to step S23 in step S24 is to determine whether the error between the training final result and the test final result exceeds a threshold; If it exceeds, return to step S23 to continue optimization; If not, the final training model has completed training and there is no need to return to step S23.

7. The method for predicting the pull-out bearing capacity of threaded cast-in-place piles based on machine learning according to claim 2, characterized in that: Step S2 includes: Build several machine learning models based on different algorithms; The experimental data are used to train and optimize several machine learning models.

8. The method for predicting the pull-out bearing capacity of threaded cast-in-place piles based on machine learning according to claim 7, characterized in that: Step S3 uses the optimized machine learning models to predict the pull-out bearing capacity of the threaded cast-in-place piles, obtains several prediction results, and performs weighted calculation on the several prediction results to obtain a unique prediction result.

9. A machine learning-based prediction system for the pull-out bearing capacity of threaded cast-in-place piles, characterized in that: include: A data acquisition unit, used for acquiring test data; A model training unit, used to construct a machine learning model and optimize the machine learning model using the test data; The prediction unit predicts the pull-out bearing capacity of the threaded cast-in-place pile by using the optimized machine learning model.

10. A method for predicting the pull-out bearing capacity of threaded cast-in-place piles based on machine learning, characterized in that: The following steps are involved: Step 1: Obtain test data; Step 2: construct several machine learning models based on the test data; and optimize the machine learning models; Step 3: Use several optimized machine learning models to predict the pull-out bearing capacity of threaded cast-in-place piles, obtain several prediction results, and perform weighted calculation on the several prediction results to obtain a unique prediction result.

Images