RC column anti-seismic performance prediction method based on self-adaptive optimization integrated model

Abstract

The invention provides an RC column anti-seismic performance prediction method based on a self-adaptive optimization integrated model. The method comprises the following steps: S1, collecting multi-dimensional input features and corresponding mechanical property data of an RC column; s2, preprocessing the collected data, and dividing a data set; s3, constructing an integrated model based on RF and BP, and training the data set by using the integrated model to obtain a preliminary prediction model; and S4, optimizing the hyper-parameters of the integrated model by using a hyper-parameter adaptive optimization algorithm to obtain a final prediction model, and predicting the anti-seismic performance of the RC column by using the final prediction model. According to the method, the RF and the BP are integrated in parallel, the advantages of the RF and the BP in the aspects of noise resistance and nonlinear relation fitting capacity are combined, meanwhile, the key hyper-parameters of the RF model and the BP model are dynamically adjusted through the adaptive optimization algorithm, and therefore the problems that an existing model is insufficient in prediction precision, low in hyper-parameter tuning efficiency and the like in the anti-seismic performance prediction process are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of structural earthquake resistance, and particularly relates to a method for predicting the seismic performance of RC columns based on an adaptive optimization integration model. Background Art

[0002] As a key component in a structure that bears vertical loads and lateral seismic forces, the seismic performance of an RC column is directly related to the overall behavior and safety of the structure during an earthquake. Under seismic action, damage to the RC column often leads to a decrease in structural stiffness and loss of load-bearing capacity, and may even cause the structure to collapse. Therefore, how to accurately evaluate the seismic performance of RC columns has become an important issue in the field of structural engineering.

[0003] With the gradual development of building structure design towards performance-based and intelligent directions, the seismic performance evaluation of RC columns has become particularly important. The existing methods for evaluating the seismic performance of RC columns mainly include two aspects: one is to evaluate the seismic capacity of the RC column, that is, its maximum capacity to withstand seismic loads; the other is to evaluate the deformation behavior and response of the RC column under seismic action. Comprehensive analysis of the performance indicators in these two aspects can provide a more comprehensive understanding of the performance and safety of RC columns during an earthquake.

[0004] Currently, the seismic performance evaluation of RC columns usually relies on mechanical models and empirical models. Mechanical models can relatively accurately describe the mechanical behavior of RC columns by establishing complex physical formulas and calculation methods. However, these models usually require a large number of assumptions and simplifications, and the calculation process is complex, with low calculation efficiency. Empirical models establish prediction formulas by fitting a large amount of experimental data. Although the calculation is simple, the accuracy is limited, and the applicability under different load conditions is poor. Therefore, there are still significant challenges in terms of accuracy and calculation efficiency for existing calculation methods.

[0005] In recent years, machine learning methods have been widely applied in the seismic performance prediction of RC columns. In particular, methods such as random forest (RF) and backpropagation neural network (BP) have shown great potential in the seismic performance prediction of RC columns due to their excellent non-linear fitting ability and strong generalization performance. However, these methods still face some challenges, mainly including: sensitivity to data noise and outliers, insufficient generalization ability under complex data features, and low efficiency of hyperparameter optimization. To overcome these challenges, ensemble learning methods (such as Bagging and Boosting) have been gradually proposed as effective means to improve prediction performance. Ensemble learning can improve the stability and prediction accuracy of the model by combining the advantages of multiple base models. However, existing ensemble learning methods still have deficiencies in modeling the complex non-linear behavior of RC columns and the efficiency of hyperparameter optimization. Therefore, how to efficiently combine the advantages of multiple models and improve the accuracy and stability of the seismic performance prediction of RC columns has become an important issue in current technical research. Summary of the Invention

[0006] The present invention proposes a method for predicting the seismic performance of RC columns based on an adaptive optimization ensemble model, which solves the problems of poor prediction accuracy and stability in the seismic performance prediction of RC columns by traditional ensemble learning methods.

[0007] The technical solution of the present invention is implemented as follows:

[0008] In the first aspect of the present invention, a method for predicting the seismic performance of RC columns based on an adaptive optimization ensemble model is provided, including the following steps:

[0009] S1, collecting multi-dimensional input features and corresponding mechanical property data of RC columns;

[0010] S2, preprocessing the collected data and dividing the data set;

[0011] S3, constructing an ensemble model based on RF and BP, training the data set using the ensemble model to obtain a preliminary prediction model;

[0012] S4, optimizing the hyperparameters of the ensemble model using a hyperparameter adaptive optimization algorithm to obtain a final prediction model, and predicting the seismic performance of RC columns using the final prediction model.

[0013] Specifically, in step S1, the multi-dimensional input features of the RC columns include geometric dimensions, material properties, reinforcement conditions, and loading conditions.

[0014] Furthermore, the following 10 input features are selected for the multi-dimensional input features X of the RC columns:

[0015] X = {b, d, λ, f c, f l , f ysv , ρ l , ρ sv , s / b, n c};

[0016] Among them, b and d respectively represent the dimensions of the cross-section perpendicular to and parallel to the direction of the reciprocating horizontal force, and λ represents the shear span ratio; f c , f l and f ysv respectively represent the axial compressive strength of concrete, the yield strength of longitudinal reinforcement, and the yield strength of stirrups; ρ l and ρ sv respectively represent the reinforcement ratio of longitudinal reinforcement and the volumetric reinforcement ratio of stirrups, s represents the spacing of stirrups; n c represents the axial compression ratio.

[0017] Specifically, in step S1, the mechanical property data of the RC column are obtained based on the pseudo-static test. The lateral load-bearing capacity and displacement ductility coefficient of the RC column are selected from the mechanical property data as the prediction targets of the prediction model. The lateral load-bearing capacity refers to the maximum load-bearing capacity that the RC column can withstand when resisting the lateral seismic force; the displacement ductility coefficient refers to the ability of the RC column to withstand plastic deformation without failure when resisting the lateral seismic force.

[0018] Specifically, in step S2, the preprocessing of the data includes normalizing the data with different dimensions.

[0019] Specifically, in step S2, the data set is divided into three parts: 70% of the data is divided into the training set for model training; 20% of the data is divided into the validation set for subsequent dynamic weight allocation and hyperparameter optimization; 10% of the data is divided into the test set for evaluating the accuracy and generalization ability of the model.

[0020] Specifically, step S3 specifically includes the following steps:

[0021] Train RF and BP on the training set respectively to obtain two prediction models f RF and f BP ;

[0022] Calculate the mean absolute error of the prediction results of the two prediction models on the validation set respectively:

[0023]

[0024]

[0025] Among them, respectively represent the prediction models f RF and fBP The mean absolute error of the prediction results on the validation set, where n is the total amount of data in the validation set; respectively represent the prediction model f RF and f BP The predicted values on the validation set; y v represents the true value corresponding to the input feature x v in the validation set;

[0026] Based on the mean absolute error, the weights of the two prediction models f RF and f BP are dynamically adjusted using the following formula:

[0027]

[0028]

[0029] Define a relative error threshold θ for determining whether to use a single model or an ensemble model for prediction. The judgment conditions are as follows:

[0030]

[0031] If the above conditions are met, select the prediction model with the smaller mean absolute error as the single model for prediction; if the above conditions are not met, use the ensemble model after fusing the two prediction models f RF and f BP for prediction;

[0032] The prediction result of the ensemble model on the test set is:

[0033]

[0034] where, respectively represent the predicted values of the two prediction models f RF and f BP on the test set.

[0035] Specifically, in step S4, the method for optimizing the hyperparameters of the ensemble model using the improved BKA algorithm includes the following steps:

[0036] In the BKA population initialization stage, a random uniform distribution strategy is used to generate the population so that the individuals are distributed throughout the search space. The initialization formula is as follows:

[0037] P i = BK lb，j + rand(BK ub，j - BK lb，j );

[0038] where, P iDenote the position of the \(i\)-th individual after initialization; BK lb，j and BK ub，j respectively represent the lower and upper bounds of the \(j\)-th dimension of the black-winged kite; rand is a random number between \([0, 1]\);

[0039] In the BKA attack phase, the Levy flight strategy and the adaptive \(t\)-distribution strategy are introduced to improve the position of the black-winged kite. The formula is as follows:

[0040]

[0041] where respectively represent the positions of the \(i\)-th black-winged kite in the \(t\)-th and \((t + 1)\)-th iteration steps; \(\beta\) is a scaling factor used to adjust the step size; denotes element-wise operation; Levy(\(\lambda\)) represents a random variable of the Levy distribution, \(\lambda\) is a parameter of the Levy distribution, \(\lambda\in(1, 3]\); \(g'\) is the dynamic probability, \(r\) is \([0, 1]\); \(t(iter)\) represents the value calculated in the \(iter\)-th iteration;

[0042] The scaling factor \(\beta\) is defined as:

[0043]

[0044] where represents the position of the \(i\)-th black-winged kite under the current global optimal solution;

[0045] The calculation formula of the dynamic probability \(g'\) is as follows:

[0046]

[0047] where \(\omega\) 1 determines the upper limit of the dynamic selection probability, \(\omega\) 2 determines the change range of the dynamic probability; \(T\) is the maximum number of iterations, \(t\) is the current number of iterations;

[0048] In the BKA migration phase, the Cauchy mutation factor \(C(0, 1)\) is introduced to explore potential solutions in a larger range. The formula is as follows:

[0049]

[0050] where is the position of the leading individual in the current \(j\)-th dimension; \(F\) i is the fitness of the current individual; \(F\) ri is the fitness of a random individual; \(C(0, 1)\) is the Cauchy mutation factor; \(h = 2\cdot\sin(r+\pi / 2)\), \(r\) is \([0, 1]\);

[0051] Taking the mean absolute error on the validation set as the optimization objective function, determine the value ranges of all hyperparameters; use the improved BKA algorithm to iteratively adjust the hyperparameter combinations and gradually approach the optimal value of the objective function; each generated hyperparameter combination corresponds to one model training, and evaluate the mean absolute error of the model through the validation set, and record the hyperparameter combination with the smallest mean absolute error as the optimal hyperparameter combination.

[0052] In a second aspect of the present invention, there is provided an electronic device, including a memory and a processor. A computer program that can run on the processor is stored in the memory, and when the processor executes the computer program, the steps of the RC column seismic performance prediction method are implemented.

[0053] In a third aspect of the present invention, there is provided a computer-readable storage medium. The storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the RC column seismic performance prediction method are implemented.

[0054] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention assigns weights to the RF and BP models based on the dynamic weighting strategy of the mean absolute error, improving the prediction accuracy and generalization ability of the lateral load-bearing capacity and displacement ductility coefficient; at the same time, combining the perturbation strategy of the improved BKA optimization algorithm, dynamically selecting the Levy flight strategy and the adaptive t-distribution strategy, realizing the adaptive adjustment of hyperparameters, and effectively avoiding the limitations of traditional manual parameter tuning. On this basis, the accuracy and robustness of the RC column seismic performance prediction are significantly improved, and at the same time, the applicability of the RC column seismic performance evaluation under complex seismic load conditions is expanded. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings without creative efforts based on these drawings.

[0056] Figure 1 It is a schematic flow chart of an RC column seismic performance prediction method based on an adaptive optimization integration model of the present invention;

[0057] Figure 2 It is a schematic diagram of the training process of the integration model in the embodiment of the present invention;

[0058] Figure 3 It is a schematic diagram of the principle of the hyperparameter adaptive optimization algorithm in the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0059] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative work belong to the scope of protection of the present invention.

[0060] Referring to Figures 1 to 3 As shown, in the first aspect of the embodiments of the present invention, a method for predicting the seismic performance of RC columns based on an adaptive optimization integration model is provided, including the following steps:

[0061] S1. Collect multi-dimensional input features and corresponding mechanical property data of RC columns;

[0062] S2. Preprocess the collected data and divide the data set;

[0063] S3. Construct an integration model based on RF and BP, and use the integration model to train the data set to obtain a preliminary prediction model;

[0064] S4. Use the hyperparameter adaptive optimization algorithm to optimize the hyperparameters of the integration model to obtain a final prediction model, and use the final prediction model to predict the seismic performance of RC columns.

[0065] Specifically, in step S1, the multi-dimensional input features of the RC column include geometric dimensions, material properties, reinforcement conditions, and load conditions. The seismic performance of the RC column is comprehensively affected by various factors such as geometric dimensions, material properties, reinforcement conditions, and load conditions.

[0066] Furthermore, to ensure the scientificity and effectiveness of the input features, the selection of input feature parameters needs to consider both significant influence and easy acquisition. In this embodiment, based on relevant literature and experimental experience, the following 10 input features are selected for the multi-dimensional input features X of the RC column:

[0067] X = {b, d, λ, f c , f l , f ysv , ρ l , ρ sv , s / b, n c};

[0068] Among them, b and d respectively represent the dimensions of the cross-section perpendicular to the reciprocating horizontal force direction and parallel to the reciprocating horizontal force direction, and λ represents the shear span ratio; f c , f l and f ysv respectively represent the axial compressive strength of concrete, the yield strength of longitudinal reinforcement, and the yield strength of stirrups; ρ l and ρ svrespectively represent the longitudinal reinforcement ratio and the volumetric stirrup ratio, s represents the stirrup spacing; n c represents the axial compression ratio.

[0069] Specifically, in step S1, the mechanical property data of the RC column is obtained based on the pseudo-static test. The lateral load-carrying capacity and displacement ductility coefficient of the RC column are selected from the mechanical property data as the prediction targets of the prediction model. The lateral load-carrying capacity refers to the maximum load-carrying capacity that the RC column can withstand when resisting the lateral seismic force, which reflects the strength characteristics of the RC column under the seismic action and directly determines whether the RC column can maintain the stability and safety of the structure; the displacement ductility coefficient refers to the ability of the RC column to withstand plastic deformation without failure when resisting the lateral seismic force, which measures the deformation ability of the RC column under the seismic action and characterizes that the RC column can still maintain sufficient load-carrying capacity and structural integrity under the condition of experiencing large deformations.

[0070] Among them, the lateral load-carrying capacity V can be directly obtained from the pseudo-static test data, and the displacement ductility coefficient μ needs to be calculated according to the pseudo-static test data, μ = Δ u / Δ y , Δ y is the yield displacement determined by the energy equivalence method, and Δ u is the ultimate displacement corresponding to the component load-carrying capacity dropping to 0.8V.

[0071] Specifically, to ensure the effectiveness of the model and improve the prediction accuracy, in step S2, the method for preprocessing the data is as follows: First, by statistically analyzing the distribution of the input and output data, understand the size and distribution of the data set. Then, use the Spearman correlation analysis method to evaluate the non-linear relationship between different input features. Then, perform normalization processing on the data with different dimensions to eliminate the influence of dimension differences on model training, thereby improving the stability and convergence speed of training.

[0072] Specifically, in step S2, the data set is divided into three parts: 70% of the data is divided into the training set for model training; 20% of the data is divided into the validation set for subsequent dynamic weight allocation and hyperparameter optimization; 10% of the data is divided into the test set for evaluating the accuracy and generalization ability of the model.

[0073] Specifically, step S3 specifically includes the following steps:

[0074] Train RF (Random Forest) and BP (Back Propagation Neural Network) on the training set respectively to obtain two prediction models (base learners) f RF and f BP ;

[0075] Calculate the mean absolute error of the prediction results of the two prediction models on the validation set respectively:

[0076]

[0077]

[0078] where, respectively represent the mean absolute error of the prediction results of the prediction models f RF and f BP on the validation set, and n is the total amount of data in the validation set; respectively represent the predicted values of the prediction models f RF and f BP on the validation set; y v represents the true value corresponding to the input feature x v in the validation set;

[0079] Based on the mean absolute error, dynamically adjust the weights of the two prediction models f RF and f BP using the following formula:

[0080]

[0081] This way of weight allocation ensures that the model with a smaller mean absolute error contributes more to the final prediction result.

[0082] To avoid models with significantly different performances from participating in the integration, define a relative error threshold θ to determine whether to use a single model or an integrated model for prediction. The judgment conditions are as follows:

[0083]

[0084] If the above conditions are met, select the prediction model with a smaller mean absolute error as a single model for prediction, that is:

[0085] If then: Q BP = 1, Q RF = 0;

[0086] If then: Q BP = 0, Q RF = 1;

[0087] If the above conditions are not met, use the integrated model after fusing the two prediction models f RF and f BP for prediction. The prediction result of the integrated model on the test set is:

[0088]

[0089] Among them, respectively represent the predicted values of two prediction models f RF and f BP on the test set.

[0090] In order to improve the accuracy of the prediction model, it is necessary to optimize the hyperparameters of the RF and BP models. Based on the mean absolute error on the validation set, find the optimal hyperparameter combination corresponding to each model, and use the optimized model for integration. However, the model contains a large number of hyperparameters (such as the number of trees, the depth of the trees, the learning rate, the hidden layer structure, the number of neurons, etc.). It is very difficult to find the optimal hyperparameter combination corresponding to the model by using the method of manual tuning, and the efficiency is extremely low. Based on this, the present invention proposes an improved BKA adaptive hyperparameter optimization method to solve this problem.

[0091] Specifically, in step S4, the method for optimizing the hyperparameters of the integrated model by using the improved BKA algorithm includes the following steps:

[0092] In the BKA population initialization stage, a random uniform distribution strategy is adopted to generate the population, so that the individuals are distributed in the entire search space, thereby enhancing the global search ability of the algorithm. The initialization formula is as follows:

[0093] P i = BK lb，j + rand(BK ub，j - BK lb，j );

[0094] Among them, P i represents the position of the i-th individual after initialization; BK lb，j and BK ub，j respectively represent the lower bound and the upper bound of the black-winged kite in the j-th dimension; rand is a random number between [0,1]; this strategy ensures that the population is evenly distributed in the entire search space, thereby improving the global exploration efficiency and avoiding the search falling into local optimality.

[0095] In the BKA attack stage, the Levy flight strategy and the adaptive t-distribution strategy are introduced to improve the position of the black-winged kite, thereby enhancing the global search ability of the algorithm. The formula is as follows:

[0096]

[0097] Among them, respectively represent the positions of the i-th black-winged kite in the t-th and t + 1-th iteration steps; β is a scaling factor used to adjust the step size; denotes an element-wise operation; Levy(λ) represents a random variable following the Levy distribution, where λ is the parameter of the Levy distribution and λ ∈ (1, 3]; g′ is the dynamic probability, and r is a random number between [0, 1]; t(iter) represents the value calculated in the iter-th iteration;

[0098] The scaling factor β is defined as:

[0099]

[0100] where, represents the position of the i-th black-winged kite under the current global optimal solution;

[0101] To better control the conversion between the Levy flight strategy and the adaptive t-distribution strategy and improve the convergence speed of the algorithm, a dynamic probability mechanism g′ is introduced to regulate the use of the Levy flight and the adaptive t-distribution; the calculation formula of the dynamic probability g′ is as follows:

[0102]

[0103] where, ω 1 determines the upper limit of the dynamic selection probability, and ω 2 determines the change range of the dynamic probability; T is the maximum number of iterations, and t is the current number of iterations. The dynamic probability mechanism can enhance the global search ability in the initial stage of optimization and focus on local optimization in the later stage, improving the convergence speed of the algorithm.

[0104] In the BKA migration stage, a Cauchy mutation factor C(0, 1) is introduced to explore potential solutions in a larger range, and the formula is as follows:

[0105]

[0106] where, is the leading individual position of the current j-th dimension; F i is the fitness of the current individual; F ri is the fitness of a random individual; C(0, 1) is the Cauchy mutation factor used to provide a larger search range; h = 2·sin(r + π / 2), where r is a random number between [0, 1].

[0107] Taking the mean absolute error on the validation set as the optimization objective function, determine the value range of all hyperparameters; use the improved BKA algorithm to iteratively adjust the hyperparameter combination to gradually approach the optimal value of the objective function; each generated hyperparameter combination corresponds to a model training, and evaluate the mean absolute error of the model through the validation set; perform iterative training on the model until the maximum number of iterations is reached, and record the hyperparameter combination with the smallest mean absolute error as the optimal hyperparameter combination for use in the ensemble model.

[0108] In this embodiment, the performance of the integrated model is quantitatively evaluated by the coefficient of determination (R 2 2), root mean square error (RMSE), and mean absolute error (MAE), and is compared with existing methods. The comparison results are shown in Table 1 below:

[0109] Table 1 Comparison of Model Prediction Performance

[0110]

[0111] The above Table 1 shows the prediction performance evaluation indicators of different machine learning models, where AO-RF-BP is the integrated model proposed in the present invention. The present invention conducts a comparative analysis with SVM, RF, XGBoost, BP, and CNN models. It can be seen from Table 1 that compared with other models, the AO-RF-BP integrated model has the best prediction performance. Specifically, the R 2 2 of the lateral load resistance and displacement ductility coefficient predicted by the integrated model of the present invention has been significantly improved, and both RMSE and MAE have been significantly improved.

[0112] In summary, a method for predicting the seismic performance of RC columns based on an adaptive optimization integrated model provided by the present invention parallelly integrates the random forest (RF) and the backpropagation neural network (BP), combines the advantages of both in anti-noise performance and non-linear relationship fitting ability, and realizes the dynamic fusion of the prediction results of the two models based on the dynamic weighting strategy of the mean absolute error (MAE). At the same time, an adaptive optimization algorithm is used to dynamically adjust the key hyperparameters of the RF and BP models, avoiding the inefficiency and limitations of traditional manual hyperparameter tuning. Compared with traditional single machine learning models, this method has significant advantages in terms of prediction accuracy and stability, providing an innovative and reliable solution for the efficient evaluation of the seismic performance of RC columns. The prediction method of the present invention solves the problems existing in the existing models in seismic performance prediction, such as insufficient prediction accuracy and low hyperparameter tuning efficiency, realizes the efficient prediction of the lateral load resistance and displacement ductility coefficient of RC columns, and significantly improves the prediction accuracy, generalization ability, and robustness of the model. This method not only provides an efficient and reliable solution for the safety assessment of RC columns under complex seismic load conditions, but also has important reference value and engineering application significance for the research and practice in the fields of seismic design and structural engineering.

[0113] In the second aspect of the embodiments of the present invention, an electronic device is provided, including a memory and a processor. The memory stores a computer program that can run on the processor, and when the processor executes the computer program, the steps of the method for predicting the seismic performance of RC columns are implemented.

[0114] In a third aspect of the embodiments of the present invention, a computer-readable storage medium is provided. The storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the RC column seismic performance prediction method are implemented.

[0115] The foregoing are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for predicting the seismic performance of RC columns based on an adaptive optimization integrated model, characterized in that: The following steps are involved: S1, collect multi-dimensional input features and corresponding mechanical property data of RC columns; S2, preprocess the collected data and divide the data set; S3, build an integrated model based on RF and BP, use the integrated model to train the data set, and obtain a preliminary prediction model; S4, using the hyperparameter adaptive optimization algorithm to optimize the hyperparameters of the integrated model to obtain the final prediction model, and using the final prediction model to predict the seismic performance of the RC column.

2. The method for predicting the seismic performance of RC columns based on the adaptive optimization integrated model according to claim 1, characterized in that: In step S1, the multi-dimensional input features of the RC column include geometric dimensions, material properties, reinforcement conditions and load conditions.

3. The method for predicting the seismic performance of RC columns based on the adaptive optimization integrated model as claimed in claim 2, characterized in that: The multi-dimensional input feature X of the RC column selects the following 10 input features: X={b,d,λ,f c ,f l ,f ysv ,r l ,r sv ,s / b,n c }; Where b and d represent the dimensions of the cross section in the direction perpendicular to and parallel to the reciprocating horizontal force, respectively, and λ represents the shear span ratio; f c 、f l and f ysv They represent the axial compressive strength of concrete, the yield strength of longitudinal reinforcement and the yield strength of stirrups respectively; ρ l and ρ sv They represent the reinforcement ratio of longitudinal reinforcement and the volume reinforcement ratio of stirrups, s represents the spacing of stirrups; n c Indicates the axial pressure ratio.

4. The method for predicting the seismic performance of RC columns based on the adaptive optimization integrated model according to claim 1, characterized in that: In step S1, the mechanical property data of the RC column is obtained based on a pseudo-static test, and the lateral displacement bearing capacity and displacement ductility coefficient of the RC column are selected from the mechanical property data as prediction targets of the prediction model. The lateral displacement bearing capacity refers to the maximum bearing capacity that the RC column can withstand when resisting lateral seismic force; the displacement ductility coefficient refers to the ability of the RC column to withstand plastic deformation without being destroyed when resisting lateral seismic force.

5. The method for predicting the seismic performance of RC columns based on the adaptive optimization integrated model according to claim 1, characterized in that: In step S2, preprocessing the data includes normalizing the data with different dimensions.

6. The method for predicting the seismic performance of RC columns based on the adaptive optimization integrated model according to claim 1, characterized in that: In step S2, the data set is divided into three parts: 70% of the data is divided into a training set for model training; 20% of the data is divided into a validation set for subsequent dynamic weight allocation and hyperparameter optimization; 10% of the data is divided into a test set to evaluate the accuracy and generalization ability of the model.

7. The method for predicting the seismic performance of RC columns based on the adaptive optimization integrated model according to claim 1, characterized in that: Step S3 specifically includes the following steps: Train RF and BP on the training set respectively to obtain two prediction models f RF and f BP ; Calculate the mean absolute error of the prediction results of the two prediction models on the validation set respectively: in, and They represent the prediction model f RF and f BP The mean absolute error of the prediction results on the validation set, where n is the total amount of data in the validation set; and They represent the prediction model f RF and f BP Prediction value on the validation set; y v Represents the input feature x in the validation set v The corresponding true value; Based on the mean absolute error, the two prediction models f are dynamically adjusted using the following formula RF and f BP Weight: Define the relative error threshold θ to determine whether to use a single model or an integrated model for prediction. The judgment conditions are as follows: If the above conditions are met, the prediction model with the smaller mean absolute error is selected as the single model for prediction; if the above conditions are not met, two prediction models f are used RF and f BP The fused integrated model is used for prediction; Prediction results of the integrated model on the test set for: in, and Represents two prediction models f RF and f BP Predicted values ​​on the test set.

8. The method for predicting the seismic performance of RC columns based on the adaptive optimization integrated model according to claim 1, characterized in that: In step S4, the method for optimizing the hyperparameters of the integrated model using the improved BKA algorithm includes the following steps: In the BKA population initialization stage, a random uniform distribution strategy is used to generate the population so that individuals are distributed in the entire search space. The initialization formula is as follows: P i =BK lb，j +rand(UK ub，j -BK lb，j ); Among them, P i represents the position of the i-th individual after initialization; BK lb，j and BK ub，j They represent the lower and upper bounds of the j-th dimension black-winged kite respectively; rand is a random number between [0,1]; In the BKA attack phase, the Levy flight strategy and adaptive t distribution strategy are introduced to improve the position of the black kite. The formula is as follows: in, and denote the position of the i-th black-winged kite in the t-th and t+1-th iteration steps respectively; β is a scaling factor used to adjust the step size; represents an element-by-element operation; Levy(λ) represents a random variable of Levy distribution, λ is the parameter of Levy distribution, λ∈(1,3]; g′ is the dynamic probability, r is [0,1]; t(iter) represents the value calculated in the iterth iteration; The scaling factor β is defined as: in, represents the position of the i-th black-winged kite in the current global optimal solution; The calculation formula of dynamic probability g′ is as follows: Among them, ω1 determines the upper limit of the dynamic selection probability, ω2 determines the range of change of the dynamic probability; T is the maximum number of iterations, and t is the current number of iterations; Introducing the Cauchy mutation factor C(0,1) in the BKA migration phase to explore potential solutions in a larger range, the formula is as follows: in, is the leading individual position of the current j-th dimension; F i is the fitness of the current individual; F ri is the fitness of a random individual; C(0,1) is the Cauchy mutation factor; h = 2 sin(r+π / 2), r is [0,1]; The mean absolute error on the validation set is used as the optimization objective function to determine the value range of all hyperparameters. The improved BKA algorithm is used to iteratively adjust the hyperparameter combination to gradually approach the optimal value of the objective function. Each generated hyperparameter combination corresponds to a model training, and the mean absolute error of the model is evaluated through the validation set. The hyperparameter combination with the smallest mean absolute error is recorded as the optimal hyperparameter combination.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, characterized in that: When the processor executes the computer program, the steps of the method for predicting the seismic performance of an RC column as claimed in any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method for predicting the seismic performance of an RC column as claimed in any one of claims 1 to 8 are implemented.

Images