CFRP-steel interface ultimate bearing capacity prediction method based on improved XGBoost algorithm

Abstract

The invention discloses a CFRP-steel interface ultimate bearing capacity prediction method based on an improved XGBoost algorithm, and belongs to the technical field of steel structure externally attached CFRP reinforcement, and the method comprises the following steps: obtaining tensile shear test data of a CFRP-steel interface, and carrying out preprocessing to obtain a test data set; wherein the test data set comprises a test set and a training set; based on an improved XGBoost algorithm, an initial CFRP-steel interface ultimate bearing capacity prediction model is constructed, and a loss function is determined based on prior physical knowledge; based on grid search and a five-fold cross validation method, determining optimal hyper-parameters of the initial CFRP-steel interface ultimate bearing capacity prediction model according to the test set and the loss function, obtaining the CFRP-steel interface ultimate bearing capacity prediction model, and performing evaluation and feature importance analysis respectively; acquiring ultimate bearing capacity data to be predicted, and processing the ultimate bearing capacity data through the CFRP-steel interface ultimate bearing capacity prediction model completing evaluation and feature importance analysis to obtain a prediction result.

Description

Technical Field

[0001] The invention belongs to the technical field of CFRP strengthening for steel structures, and particularly relates to a method for predicting the ultimate bearing capacity of CFRP-steel interfaces based on an improved XGBoost algorithm. Background Art

[0002] Carbon fiber-reinforced polymer (CFRP) is a common structural strengthening material. Due to its many advantages such as light weight, high strength, fatigue and corrosion resistance, and ease of construction, it has important application potential in the strengthening and repair of steel structures. In the bonding strengthening system, the bonding performance of the CFRP-steel interface is the key to ensuring the strengthening effect. At present, the bonding performance of the CFRP-steel interface is mainly characterized by the ultimate bearing capacity of the interface. Many scholars have proposed different bond-slip models based on the tensile shear test results of the CFRP-steel interface for failure mechanism research and bearing capacity prediction. Although these previous bond-slip models had high accuracy on the limited test data at that time, their generalization ability for new data was poor. Moreover, there were many influencing factors in the models themselves, with a wide variety and large differences in prediction results, and the application was complex. With the development of artificial intelligence and machine learning, some scholars have also used data-driven methods and various machine learning methods to predict the ultimate bearing capacity of CFRP-steel lap specimens. However, due to the single type and small number of test samples, the trained models have poor generalization ability and low prediction accuracy. Summary of the Invention

[0003] In view of this, the invention provides a method for predicting the ultimate bearing capacity of CFRP-steel interfaces based on an improved XGBoost algorithm, which is used to solve the problems of poor generalization ability and low prediction accuracy of the models trained due to the single type and small number of test samples.

[0004] To achieve the above object, the invention provides a method for predicting the ultimate bearing capacity of CFRP-steel interfaces based on an improved XGBoost algorithm, including the following steps:

[0005] Obtain the tensile shear test data of the CFRP-steel interface and perform preprocessing to obtain the test data set; wherein, the test data set includes: a test set and a training set;

[0006] Based on the improved XGBoost algorithm, construct an initial prediction model for the ultimate bearing capacity of the CFRP-steel interface, and determine the loss function based on prior physical knowledge;

[0007] Based on the grid search and 5-fold cross-validation method, determine the optimal hyperparameters of the initial CFRP-steel interface ultimate bearing capacity prediction model according to the test set and the loss function, obtain the CFRP-steel interface ultimate bearing capacity prediction model, and conduct evaluation and feature importance analysis respectively;

[0008] Obtain the data of the ultimate bearing capacity to be predicted, and process it through the CFRP-steel interface ultimate bearing capacity prediction model that has completed the evaluation and feature importance analysis to obtain the prediction result.

[0009] As an embodiment of the present invention, obtain the tensile-shear test data of the CFRP-steel interface, and perform preprocessing to obtain the test data set, including:

[0010] Obtain the tensile-shear test data of the CFRP-steel interface;

[0011] Fill in the default values of some features in the test samples for the tensile-shear test data, and remove the outliers in the data set through the isolation forest algorithm to obtain the test data set.

[0012] As an embodiment of the present invention, determine the optimal hyperparameters of the initial CFRP-steel interface ultimate bearing capacity prediction model according to the test set, and obtain the CFRP-steel interface ultimate bearing capacity prediction model, including:

[0013] According to the grid search and the preset hyperparameter value range, perform an exhaustive combination of hyperparameters to obtain a hyperparameter combination set;

[0014] Perform 5-fold cross-validation on the elements in the hyperparameter combination set through the training set, and calculate the RMSE corresponding to the elements in the verified hyperparameter combination set to obtain the RMSE set;

[0015] Select the hyperparameter combination with the smallest RMSE in the RMSE set as the optimal hyperparameter, and use the model trained according to the optimal hyperparameter as the CFRP-steel interface ultimate bearing capacity prediction model.

[0016] As an embodiment of the present invention, determine the loss function based on prior physical knowledge, including:

[0017] The calculation formula of the loss function is as follows:

[0018]

[0019] In the formula, LOSS represents the loss function, is the data loss, and λ1 is the coefficient of the data loss, is the physical loss, and λ2 is the coefficient of the physical loss, is the model prediction value, is the test true value, It is the calculated value based on prior physical knowledge.

[0020] As an embodiment of the present invention, a test and evaluation is carried out on the prediction model of the ultimate bearing capacity of the CFRP-steel interface, including:

[0021] The test set is processed by the prediction model of the ultimate bearing capacity of the CFRP-steel interface to obtain the test results;

[0022] According to the test results, the root mean square error, average value, determination coefficient and coefficient of variation are calculated respectively, and the calculation formulas are as follows:

[0023]

[0024]

[0025] In the formula, RMSE is the root mean square error, AVG is the average value, COV is the coefficient of variation, R 2 is the determination coefficient, n is the number of test samples, is the model prediction value, is the true test value, is the average value of the true values.

[0026] As an embodiment of the present invention, a feature importance analysis is carried out on the prediction model of the ultimate bearing capacity of the CFRP-steel interface, including:

[0027] The SHAP method is used to perform an interpretability analysis on the test results, and the feature importance ranking is obtained from both the global and local dimensions to enhance the interpretability of the model.

[0028] The beneficial effects of the present invention are as follows: The present invention uses the grid search and 5-fold cross-validation methods to optimize the hyperparameters of the prediction model of the ultimate bearing capacity of the CFRP-steel interface based on the improved XGBoost algorithm, makes full use of the limited data set, and reduces the contingency introduced by unreasonable data division; at the same time, by introducing prior physical knowledge to customize the loss function of the XGBoost algorithm, the training of the model is jointly constrained, and the physical consistency and generalization of the model for predicting the ultimate bearing capacity are improved; it solves the problems of poor generalization and low prediction accuracy of the model trained due to the single type and small number of test samples.

[0029] Other advantages, objectives and features of the present invention will be described in the subsequent specification, and to some extent will be obvious to those skilled in the art, or those skilled in the art can obtain teachings from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following specification. Brief Description of the Drawings

[0030] To make the objectives, technical solutions and beneficial effects of the present invention clearer, the present invention provides the following drawings for illustration:

[0031] Figure 1 is the process schematic diagram of the present invention;

[0032] Figure 2 is the schematic diagram of the single-shear form of the tensile-shear test of the CFRP-steel lap joint of the present invention;

[0033] Figure 3 is the schematic diagram of the double-shear form of the tensile-shear test of the CFRP-steel lap joint of the present invention;

[0034] Figure 4 is the process schematic diagram of the present invention;

[0035] Figure 5 is the sample distribution diagram of the ultimate bearing capacity prediction training set of the present invention under the optimal loss weight coefficient;

[0036] Figure 6 is the sample distribution diagram of the ultimate bearing capacity prediction test set of the present invention under the optimal loss weight coefficient;

[0037] Figure 7 is the global feature importance ranking diagram of the ultimate bearing capacity prediction test set of the present invention;

[0038] Figure 8 is the importance analysis diagram of the first local feature of the ultimate bearing capacity prediction test set of the present invention;

[0039] Figure 9 is the importance analysis diagram of the second local feature of the ultimate bearing capacity prediction test set of the present invention;

[0040] Figure 10 is the importance analysis diagram of the third local feature of the ultimate bearing capacity prediction test set of the present invention;

[0041] Figure 11 is the importance analysis diagram of the fourth local feature of the ultimate bearing capacity prediction test set of the present invention;

[0042] Figure 12 is the importance analysis diagram of the fifth local feature of the ultimate bearing capacity prediction test set of the present invention;

[0043] Figure 13 is the importance analysis diagram of the sixth local feature of the ultimate bearing capacity prediction test set of the present invention. Detailed implementation manners

[0044] As Figures 1 to 4As shown, the present invention provides a method for predicting the ultimate bearing capacity of CFRP-steel interface based on an improved XGBoost algorithm, comprising the following steps:

[0045] S1: Obtain the tensile shear test data of the CFRP-steel interface and preprocess it to obtain the test data set; wherein, the test data set includes: a test set and a training set;

[0046] S2: Based on the improved XGBoost algorithm, construct an initial prediction model for the ultimate bearing capacity of the CFRP-steel interface, and determine the loss function based on prior physical knowledge;

[0047] S3: Based on the grid search and 5-fold cross-validation method, determine the optimal hyperparameters of the initial prediction model for the ultimate bearing capacity of the CFRP-steel interface according to the test set and the loss function, obtain the prediction model for the ultimate bearing capacity of the CFRP-steel interface, and conduct evaluation and feature importance analysis respectively;

[0048] S4: Obtain the data of the ultimate bearing capacity to be predicted, and process it through the prediction model for the ultimate bearing capacity of the CFRP-steel interface that has completed the evaluation and feature importance analysis to obtain the prediction result.

[0049] Working principle of the above technical solution: In the actual application process, first, collect the tensile shear test data of the CFRP-steel interface, and preprocess these data to obtain the test data set. Among them, the tensile shear test data of the CFRP-steel interface include: the tensile strength, elastic modulus, Poisson's ratio, and adhesive layer thickness of the structural adhesive; the elastic modulus, thickness, and width of the CFRP material; the elastic modulus, thickness, and width of the steel plate; the bonding length and ultimate bearing capacity, etc. Then, divide the test data set into a training set and a test set according to a preset ratio. Specifically, the preset ratio is preferably 8:2. Among them, the collected tensile shear test data of the CFRP-steel interface are only the tensile shear test data under monotonic loading, without considering the working conditions under cyclic loading and fatigue loading, and without considering the tests using CFRP with high elastic modulus, as well as the tests carried out under special test conditions such as marine environment, high temperature and high humidity, salt spray environment, steel plate corrosion, and freeze-thaw cycle. Then, based on the improved XGBoost algorithm, construct an initial prediction model for the ultimate bearing capacity of the CFRP-steel interface, and determine the loss function based on prior physical knowledge. The loss function is a custom loss function. Then, determine all hyperparameter combinations within the preset hyperparameter range through grid search, divide the test set by the 5-fold cross-validation method, and train the initial prediction model for the ultimate bearing capacity of the CFRP-steel interface in combination with the loss function. After training is completed, calculate the RMSE, select the hyperparameter combination with the smallest RMSE as the optimal hyperparameter, and use the model trained with the optimal hyperparameter as the prediction model for the ultimate bearing capacity of the CFRP-steel interface. Then, perform a test process on the test set through the prediction model for the ultimate bearing capacity of the CFRP-steel interface to obtain the test results, and combine the test results to conduct model evaluation and feature importance analysis on the prediction model for the ultimate bearing capacity of the CFRP-steel interface. Finally, obtain the data of the ultimate bearing capacity to be predicted that needs to be predicted, and process it through the prediction model for the ultimate bearing capacity of the CFRP-steel interface that has completed the evaluation and feature importance analysis to obtain the prediction results. Specifically, take the ultimate bearing capacity of the CFRP-steel interface as the output, and the material properties and specimen dimensions of the tensile shear test as the input (data of the ultimate bearing capacity to be predicted).

[0050] Beneficial effects of the above technical solution: Through the above technical solution, the present invention optimizes the hyperparameters of the prediction model for the ultimate bearing capacity of the CFRP-steel interface based on the improved XGBoost algorithm by using the grid search and 5-fold cross-validation method, makes full use of the limited data set, and reduces the contingency introduced by unreasonable data division. At the same time, by introducing prior physical knowledge to customize the loss function of the XGBoost algorithm, jointly constrain the training of the model, and improve the physical consistency and generalization ability of the model to predict the ultimate bearing capacity. It solves the problems of poor generalization ability and low prediction accuracy of the model trained due to the single type and small sample size of the test samples.

[0051] In one embodiment, the tensile shear test data of the CFRP-steel interface is obtained and preprocessed to obtain a test data set, including:

[0052] S11: Obtaining tensile shear test data of CFRP-steel interface;

[0053] S12: Fill in the default values of certain features in the test samples for the tensile shear test data, and remove outliers in the data set through the isolation forest algorithm to obtain the test data set.

[0054] The working principle and beneficial effects of the above technical solution are as follows: the test samples in the data set are preprocessed, including filling the default values of certain features in the test samples and using the isolation forest algorithm to remove the outliers in the data set; specifically, when filling the default values of certain features in the test samples, due to the differences in different test designs, not all test data are recorded, among which the Poisson's ratio of the adhesive, the thickness of the adhesive layer and the elastic modulus of the steel may have default values, and the present invention estimates them as 0.30, 0.5 mm and 200 GPa respectively; specifically, when using the isolation forest algorithm to remove the outliers in the data set, the isolation forest algorithm constructs a series of isolation trees (Isolation Trees), each tree is a binary search tree, which is used to isolate sample points; outliers are isolated earlier and at a shallower level because they are more different from most samples; after the test data is input into the isolation forest model, the model gives the anomaly score of the corresponding data in the range of 0-1. When the anomaly score is less than a certain threshold, the data will be identified as an outlier; for example, the default values in the data set are filled, and the isolation forest algorithm is used for outlier processing, and samples that are obviously deviated from the average level (73 in total) are removed. After the removal, the data sample size is reduced from 567 to 494, and the ultimate bearing capacity of the CFRP-steel interface is used as the output, and the material properties and specimen size of the tensile shear test are used as the input; the material properties and specimen size of the tensile shear test are shown in Table 1;

[0055] Table 1 - Input and output characteristics of CFRP-steel interface failure mode prediction model

[0056] Feature Unit Meaning Type <![CDATA[f a > MPa Tensile strength of structural adhesive Input <![CDATA[E a > MPa Elastic modulus of structural adhesive Input <![CDATA[υ a > - Poisson's ratio of structural adhesive Input <![CDATA[t a > mm Thickness of adhesive layer Input <![CDATA[E p > GPa Elastic modulus of CFRP Input <![CDATA[t p > mm Thickness of CFRP Input <![CDATA[b p > mm Width of CFRP Input <![CDATA[E s > GPa Elastic modulus of steel Input <![CDATA[t s > mm Thickness of steel plate Input <![CDATA[b s > mm Width of steel plate Input L mm Bonding length Input Failure Mode - Failure mode Output

[0057] In one embodiment, the optimal hyperparameters of the initial CFRP-steel interface ultimate bearing capacity prediction model are determined according to the test set to obtain the CFRP-steel interface ultimate bearing capacity prediction model, including:

[0058] S31: exhaustively combine the hyperparameters according to the grid search and the preset hyperparameter value range to obtain a hyperparameter combination set;

[0059] S32: Use the training set to perform 5-fold cross-validation on the elements in the hyperparameter combination set, and calculate the RMSE corresponding to the elements in the hyperparameter combination set after validation to obtain the RMSE set;

[0060] S33: Select the hyperparameter combination with the smallest RMSE in the RMSE set as the optimal hyperparameters, and use the model trained according to the optimal hyperparameters as the CFRP-steel interface ultimate bearing capacity prediction model;

[0061] The working principle and beneficial effects of the above technical solution: In the process of determining the model, for the training set, use the methods of grid search and 5-fold cross-validation to optimize the hyperparameters of the model, and select the optimal hyperparameter combination according to the minimum root mean square error (RMSE) after 5-fold cross-validation for each combination; among them, the main hyperparameters involved include the maximum depth of the tree (max_depth), the feature sampling ratio of the tree (colsample_bytree), the subsampling ratio of the training samples (subsample), the number of weak learners (n_estimators), and the learning rate (learning_rate); preset the value ranges of the above hyperparameters and perform exhaustive combinations through grid search, and then perform 5-fold cross-validation on each hyperparameter combination on the training set to obtain the average RMSE of the XGBoost model predicting the bearing capacity under this combination, and select the hyperparameters with the smallest average RMSE among all combinations as the optimal hyperparameter combination; finally, select the hyperparameter combination with the smallest RMSE in the RMSE set as the optimal hyperparameters, and use the model trained according to the optimal hyperparameters as the CFRP-steel interface ultimate bearing capacity prediction model; among them, the final results of hyperparameter optimization are shown in Table 2:

[0062] Table 2 - Optimization Results Table of Hyperparameters for CFRP-Steel Interface Failure Mode and Ultimate Bearing Capacity Prediction Model

[0063]

[0064] In one embodiment, determine the loss function based on prior physical knowledge, including:

[0065] The calculation formula of the loss function is as follows:

[0066]

[0067] In the formula, LOSS represents the loss function, is the data loss, λ1 is the coefficient of the data loss, is the physical loss, λ2 is the coefficient of the physical loss, is the model prediction value, is the experimental true value, is the calculated value based on prior physical knowledge.

[0068] Working principle and beneficial effects of the above technical solution: During actual use, when determining the loss function based on prior physical knowledge, the prior physical knowledge takes into account the influence of the Xia model, Teng model, and system factors (system dependence of input features), and the calculation expression is as follows:

[0069]

[0070] G f = 0.5τ f δ f

[0071] τ f = 0.8f a

[0072]

[0073] In the formula, P u_for i.e., the ultimate bearing capacity calculated by prior physical knowledge, τ f is the peak shear stress; f a is the tensile strength of the structural adhesive, δ f is the maximum slip of the bond-slip model, G a and t a are the shear modulus and thickness of the structural adhesive respectively, L e is the effective bond length, f is the system factor, and the calculation formula of f is as follows:

[0074]

[0075] k is the bond length correction factor, and the calculation is as follows:

[0076]

[0077] Through the ultimate bearing capacity calculation method of prior physical knowledge, the ultimate bearing capacity P of prior physical knowledge of each sample is calculated u_for ; and the first term in the loss function is "data loss", and the second term is "physical loss"; and are the coefficients of the two parts respectively, and the values of and have a greater impact on the training effect of the model, determining the degree of constraint of prior physical knowledge on the model; specifically, the prior physical knowledge considering the influence of the Xia model, Teng model, and system factors (system dependence of input features) is added to the loss function of the XGBoost algorithm, and the CFRP-steel interface ultimate bearing capacity prediction model based on the improved XGBoost algorithm is obtained by training with this loss function; by introducing prior physical knowledge to customize the loss function of the XGBoost algorithm, the training of the model is jointly constrained by "data loss" and "physical loss", improving the physical consistency and generalization of the model's predicted ultimate bearing capacity.

[0078] In one embodiment, testing and evaluating the prediction model for the ultimate bearing capacity of the CFRP-steel interface includes:

[0079] Processing the test set through the prediction model for the ultimate bearing capacity of the CFRP-steel interface to obtain test results;

[0080] Calculating the root mean square error, average value, determination coefficient, and coefficient of variation respectively according to the test results. The calculation formulas are as follows:

[0081]

[0082]

[0083] In the formula, RMSE is the root mean square error, AVG is the average value, COV is the coefficient of variation, R 2 is the determination coefficient, n is the number of test samples, is the model prediction value, is the true test value, is the average value of the true values;

[0084] The working principle and beneficial effects of the above technical solution: In the actual use process, use the final prediction model for the ultimate bearing capacity of the CFRP-steel interface to predict the test set to obtain test results. The weight coefficients corresponding to the two parts of the error are obtained when the root mean square error (RMSE) in the test is the smallest, and the evaluation indicators: root mean square error (RMSE), average value (AVG), determination coefficient (R2), and coefficient of variation (COV) are used to evaluate the test results;

[0085] For the embodiment, after considering the "physical loss" term, the optimal hyperparameter combination obtained through trial calculation is: λ1 = 0.95, λ2 = 0.05. At this time, the performance indicators and sample distributions of the model training set and test set are as Figure 5 and Figure 6 shown; at this time, the AVG of the test set = 1.000, R2 = 0.944, and the RMSE and COV are 4.829 and 0.122 respectively, while directly calculating the AVG of the test set samples using the Xia and Teng formula is 1.111, R2 = 0.398, and the RMSE and COV are 15.902 and 0.531 respectively; therefore, the model of the present invention has higher accuracy, smaller discreteness, stronger stability and generalization ability in predicting the ultimate bearing capacity of the CFRP-steel interface after appropriately considering prior physical knowledge.

[0086] In one embodiment, performing feature importance analysis on the prediction model for the ultimate bearing capacity of the CFRP-steel interface includes:

[0087] S36: Use the SHAP method to perform interpretability analysis on the test results, obtain the feature importance ranking from both global and local dimensions, and enhance the interpretability of the model;

[0088] The working principle and beneficial effects of the above technical solution: In the actual use process, when using SHAP, each feature of each sample has a certain contribution to the output result of the sample, and the size of the contribution can be quantitatively reflected by the SHAP value of the feature; in the local feature importance analysis, the influence of a certain fixed feature on the output can be analyzed according to the change of the SHAP value of this feature in different samples; while in the global feature importance analysis, the SHAP values of a certain fixed feature in all samples are averaged and sorted to obtain the global feature importance; specifically, Figures 7 to 13 They are respectively the global feature importance ranking diagram and partial local feature importance analysis diagram of the ultimate bearing capacity prediction test set in the embodiment. It can be seen that for the test set of the embodiment, the importance of the structural viscoelastic modulus and the bonding length is the highest, the width of CFRP, the thickness of CFRP, and the width of steel decrease in turn and the SHAP values are relatively close, and the SHAP values of other features are lower; the local analysis obtains the change trend of each feature value and the SHAP value, and then analyzes the contribution of each feature to the bearing capacity prediction; the interpretability analysis results based on the SHAP method show that the prediction results are consistent with the prior physical knowledge, have good physical consistency, and increase the interpretability and transparency of the model.

[0089] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solution of the present invention and not to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made in form and details without departing from the scope defined by the claims of the present invention.

Claims

1. A prediction method for the ultimate bearing capacity of CFRP-steel interface based on an improved XGBoost algorithm, comprising the following steps: Obtain the tensile shear test data of the CFRP-steel interface and perform preprocessing to obtain a test data set; wherein, the test data set includes: a test set and a training set; Based on the improved XGBoost algorithm, construct an initial prediction model for the ultimate bearing capacity of the CFRP-steel interface and determine the loss function based on prior physical knowledge; Based on the grid search and 5-fold cross-validation method, determine the optimal hyperparameters of the initial prediction model for the ultimate bearing capacity of the CFRP-steel interface according to the test set and the loss function, obtain the prediction model for the ultimate bearing capacity of the CFRP-steel interface and conduct evaluation and feature importance analysis respectively; Obtain the data of the ultimate bearing capacity to be predicted and process it through the prediction model for the ultimate bearing capacity of the CFRP-steel interface that has completed the evaluation and feature importance analysis to obtain the prediction result.

2. The prediction method for the ultimate bearing capacity of the CFRP-steel interface based on the improved XGBoost algorithm according to claim 1, wherein Obtain the tensile shear test data of the CFRP-steel interface and perform preprocessing to obtain a test data set, including: Obtain the tensile shear test data of the CFRP-steel interface; Fill in the missing values of some features in the test samples of the tensile shear test data, and remove the outliers in the data set through the isolation forest algorithm to obtain the test data set.

3. The prediction method for the ultimate bearing capacity of the CFRP-steel interface based on the improved XGBoost algorithm according to claim 1, wherein Determine the optimal hyperparameters of the initial prediction model for the ultimate bearing capacity of the CFRP-steel interface according to the test set to obtain the prediction model for the ultimate bearing capacity of the CFRP-steel interface, including: Exhaustively combine the hyperparameters according to the grid search and the preset hyperparameter value range to obtain a set of hyperparameter combinations; Perform 5-fold cross-validation on the elements in the set of hyperparameter combinations through the training set, and calculate the RMSE corresponding to the elements in the verified set of hyperparameter combinations to obtain an RMSE set; Select the hyperparameter combination with the smallest RMSE in the RMSE set as the optimal hyperparameter, and use the model trained according to the optimal hyperparameter as the prediction model for the ultimate bearing capacity of the CFRP-steel interface.

4. The prediction method for the ultimate bearing capacity of the CFRP-steel interface based on the improved XGBoost algorithm according to claim 1, wherein Determine the loss function based on prior physical knowledge, including: The calculation formula of the loss function is as follows: where LOSS represents the loss function, is the data loss, and λ1 is the coefficient of the data loss, is the physical loss, and λ2 is the coefficient of the physical loss, is the model prediction value, is the true experimental value, is the calculated value based on prior physical knowledge.

5. The prediction method for the ultimate bearing capacity of the CFRP-steel interface based on the improved XGBoost algorithm according to claim 1, wherein Test and evaluate the prediction model for the ultimate bearing capacity of the CFRP-steel interface, including: Process the test set through the prediction model for the ultimate bearing capacity of the CFRP-steel interface to obtain the test result; Calculate the root mean square error, average value, determination coefficient and coefficient of variation respectively according to the test result, and the calculation formulas are as follows: Wherein, RMSE is the root mean square error, AVG is the average value, COV is the coefficient of variation, R 2 is the coefficient of determination, n is the number of test samples, is the model predicted value, is the true experimental value, is the average value of the true values.

6. A prediction method for the ultimate bearing capacity of CFRP-steel interface based on the improved XGBoost algorithm according to claim 1, characterized in that conduct feature importance analysis on the prediction model of the ultimate bearing capacity of the CFRP-steel interface, including: using the SHAP method to perform interpretability analysis on the test results, obtaining the feature importance ranking from both global and local dimensions, and enhancing the interpretability of the model.

Images