Prediction method of shear wall earthquake damage mode based on PCA fusion with machine learning

Abstract

A shear wall earthquake damage mode prediction method based on PCA fusion with machine learning constructs a shear wall failure mode dataset by collecting test data. Principal component analysis is used to extract dataset feature engineering. The results are used as input to a machine learning classifier model for training and selection of the most accurate model. Finally, an optimized model with the best parameter combination is obtained by hyperparameter optimization of the model, achieving rapid and accurate identification of shear wall failure modes. This is the first attempt in the relevant research field to use PCA for feature extraction. By extracting high-dimensional features from the constructed data, the model training effect is effectively improved, and high-precision and rapid identification of shear wall failure modes is achieved. Compared with traditional identification methods, it has a significant accuracy advantage and shows great application prospects in related engineering fields. Because the feature engineering required for model training is established through PCA rather than directly based on dataset parameter combinations, the generalization of the model is also improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of structural damage identification in structural health monitoring, and in particular relates to a shear wall earthquake damage mode prediction method based on principal component analysis fused with machine learning. Background Art

[0002] Shear walls, as the primary lateral force-resisting components in multi-story buildings, demonstrate excellent lateral resistance and construction efficiency. However, they also exhibit complex failure mechanisms under cyclical loads. While the lateral resistance of shear walls has garnered significant attention during the structural design phase, regional risk assessments and post-earthquake building retrofits have also necessitated the development of rapid methods for identifying shear wall failure modes to guide pre-earthquake emergency response mechanisms and organize post-earthquake assessments and rescue efforts.

[0003] However, due to the lack of robust identification models based on experience and mechanical properties, traditional failure mode recognition methods based on numerical simulation and loading tests are no longer suitable for rapid identification. Recently, with the rise of machine learning algorithms and increased open access to data, failure mode recognition driven by machine learning has gained widespread attention and achieved significant research progress.

[0004] Essentially a multi-class recognition problem, predicting shear wall failure modes using machine learning classification models (classifiers) relies heavily on feature extraction from the constructed dataset. However, current research on feature engineering relies heavily on conceptually combining dataset parameters, selecting parameters directly as classifier input. This lack of effective extraction of high-dimensional features significantly limits the classifier's prediction accuracy, while also hindering generalization due to the varying parameter types used in different models. Summary of the Invention

[0005] The purpose of the present invention is to propose a method for predicting earthquake damage and failure modes of shear walls based on principal component analysis and fusion of machine learning. This method compiles and analyzes data based on 181 sets of shear wall test data, constructs a database of characteristic variables containing information about the reinforcement, size, and loading of the specimens, and proposes a feature extraction method based on principal component analysis (PCA). The extracted feature engineering is input into 9 typical classifier models for training, and then the optimal classifier model is selected. Finally, the model is hyperparameter optimized to obtain an optimized model with the best parameter combination. The present invention constructs a wall earthquake damage sensitive parameter data set by compiling and analyzing component test data, and extracts high-dimensional features from the data and inputs them into the classifier model for training, thereby achieving high-precision and rapid identification of earthquake damage and failure modes of shear walls. At the same time, since the establishment of the feature engineering required for model training is achieved through PCA rather than directly based on the combination of data set parameters, the generalization of the model is also improved.

[0006] As the first attempt to use PCA for feature extraction in the relevant research field, the present invention achieves rapid identification of shear wall failure modes. Compared with traditional identification methods, it has obvious accuracy advantages and shows great application prospects in related engineering fields.

[0007] To achieve the purpose of the present invention, the present invention is implemented through the following technical solutions:

[0008] A method for predicting shear wall earthquake damage patterns based on PCA fusion machine learning is proposed. The method is characterized by constructing a wall earthquake damage sensitive parameter dataset through compilation and analysis of component test data. PCA high-dimensional feature extraction is performed on the data and the data is input into a classifier model for training, achieving high-precision and rapid identification of shear wall earthquake damage patterns.

[0009] A method for predicting shear wall earthquake damage modes based on PCA fusion machine learning, characterized by comprising the following steps:

[0010] S1. Collect relevant data and construct a seismic damage data set S = {X1, X2, ..., X 10 ,Y}, where {X1,X2,...,X 10} are the sensitive characteristic parameters of the earthquake damage mode, which are r, f ck ,n,f y ,A s ,f yw ,ρ w ,f sw ,ρ sw and Vf / V, the specific meanings of the corresponding parameters are shown in Table 1. Y is the shear wall failure mode. According to the conventional Paulay-Priestley method, the failure modes are divided into three types according to the failure form of the specimen, namely, bending failure (F), bending-shear failure (FS) and shear failure (S).

[0011] Table 1. Definition of sensitive parameters of shear wall earthquake damage mode

[0012] parameter meaning 1 r Shear wall aspect ratio 2 <![CDATA[f ck ]]> Concrete compressive strength 3 n Axial pressure ratio 4 <![CDATA[f y ]]> Yield strength of longitudinal reinforcement in the restrained area 5 <![CDATA[A s ]]> Longitudinal reinforcement area in the restrained area 6 <![CDATA[f yw ]]> Web longitudinal reinforcement yield strength 7 <![CDATA[ρ w ]]> Web longitudinal reinforcement ratio 8 <![CDATA[f sw ]]> Web transverse reinforcement yield strength 9 <![CDATA[ρ sw ]]> Web transverse reinforcement ratio 10 <![CDATA[V f / V]]> Shear stress ratio

[0013] S2, for the characteristic parameters {X1, X2, ..., X 10}Perform PCA processing, and set the principal component dimension to 10, which is the same as the sensitive feature parameter. Convert the original parameters into projections in the principal component space. For example, for principal component PC1,

[0014] PC1=ω1X1+ω2X2+...+ω 10 X 10

[0015] Where w1 is the eigenvector of the covariance matrix ∑

[0016]

[0017] The principal component coefficient matrix W is extracted as the input for feature engineering, and N is the number of samples in the dataset.

[0018] S3. Randomly divide the extracted features into 70% training sets and the remaining 30% test sets for subsequent classifier model training and testing. To maximize the use of the constructed dataset, a cross-validation approach is used during model validation. This involves resampling the training dataset (at different sampling ratios and iterations) to define a validation set, eliminating the need for a separate validation set.

[0019] At the same time, before the model training input, the principal component coefficient matrix W of each sample is normalized to the interval [0,1], that is,

[0020]

[0021] Where W.min and W.max are the minimum and maximum values ​​of each row of the principal component coefficient matrix, respectively. std It is the normalized coefficient matrix used for model training input.

[0022] S4. Introducing 9 typical machine learning algorithms for classifier model training and testing, including Bayes, Nearest Neighbors, Logistic Regression, SVC, Decision Tree, Random Forest, AdaBoost, Gradient Boosting, and Perceptron Neural Network are used to build a mapping relationship between feature engineering and damage patterns.

[0023] S5. Test the prediction accuracy μ1 of different classifier models, perform model evaluation, and select the classifier model M0 with the highest prediction accuracy.

[0024]

[0025] S6. Optimize the hyperparameters of model M0 and use the grid search method to perform optimization calculations to obtain the optimized model M1 with the best parameter combination.

[0026] S7. Given the shear wall parameters of a failure mode to be predicted, extract the feature engineering according to the S2 method and input it into the model M1 for testing, and the corresponding predicted failure mode can be output.

[0027] Any matters not covered in the present invention are applicable to the prior art.

[0028] The beneficial effects of the present invention are as follows: by collecting test data to construct a shear wall failure mode dataset, principal component analysis is used to extract dataset feature engineering, the results are used as input to a machine learning classifier model for training and selecting the model with the highest accuracy, and finally, by performing hyperparameter optimization on the model, an optimized model with the best parameter combination is obtained, thereby achieving rapid and accurate identification of shear wall failure modes.

[0029] As the first attempt to utilize PCA for feature extraction in this research field, this method effectively improves model training by extracting high-dimensional features from the constructed data, achieving high-precision and rapid identification of shear wall failure modes. Compared to traditional identification methods, this method offers significant accuracy advantages and shows great potential for application in related engineering fields. Furthermore, because the feature engineering required for model training is established through PCA rather than directly based on dataset parameter combinations, the model's generalizability is also improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is the overall flow chart of the present invention;

[0031] Figure 2 This is a schematic diagram of the classification of common failure modes of shear walls;

[0032] Figure 3 Schematic diagram of principal component analysis;

[0033] Figure 4 Schematic diagram of the nine machine learning algorithms introduced;

[0034] Figure 5 This is the model prediction confusion matrix (test set) of one embodiment of the present invention.

[0035] Figure 6 This is the confusion matrix (test set) of the optimized model according to one embodiment of the present invention. DETAILED DESCRIPTION

[0036] The embodiments of the present invention are further described below in conjunction with the accompanying drawings, taking the experimental data of shear walls under reciprocating loads as an example for illustration.

[0037] The design concept of the present invention is: by collecting test data to construct a shear wall failure mode dataset, using principal component analysis to extract dataset feature engineering, the results are used as input to the machine learning classifier model for training, and finally the optimized model is obtained by hyperparameter optimization of the model to achieve rapid and accurate identification of shear wall failure modes.

[0038] Shear wall failure pattern recognition method based on principal component feature extraction and gradient boosting, such as Figure 1 As shown, the following steps are included:

[0039] S1. Collect relevant data and construct the shear wall specimen design, response parameters and corresponding failure mode under reciprocating load. The dataset constructed in this embodiment includes 181 sets of shear wall specimen data. Each set of specimens includes size information, reinforcement information and load information, a total of 10 parameters, namely r, f ck ,n,f y ,A s ,f yw ,ρ w ,f sw ,ρ sw and V f / V, the specific meanings of the corresponding parameters are shown in Table 1. According to the failure form of the specimens (crack development, longitudinal yield, etc.), they are classified into three conventional categories, namely bending failure (F), bending-shear failure (FS) and shear failure (S). Figure 2 Several common failure mode classification diagrams are given, among which (a) is bending failure, (b) is bending-shear failure, and (ce) is uniformly classified as shear failure.

[0040] S2. Perform PCA processing on the data set, and the result is used as feature engineering input to the subsequent classifier model. In order to retain as much data information as possible, the principal component dimension is set to 10, which is consistent with the dimension of the sensitive feature parameters of the data set. The PCA data processing principle is as follows Figure 3 .

[0041] S3. Randomly divide the extracted feature engineering into training and test sets, of which 70% is used for model training and 30% is used for model testing.

[0042] S4. Introducing 9 typical machine learning algorithms for classifier model training and testing, including Bayes (Naive Bayes), Nearest Neighbors (Nearest Neighbors), Logistic Regression (Logistic Regression), SVC (Support Vector Machine), Decision Tree (Decision Tree), Random Forest (Random Forest), AdaBoost, Gradient Boosting (Gradient Boosting) and Perceptron Neural Network (Perceptron Neural Network), such as Figure 4 . Thus, a mapping relationship between feature engineering and failure mode is constructed.

[0043] S5, test the prediction accuracy μ1 of different classifier models, perform model evaluation, and the model test confusion matrix is ​​as follows Figure 5 The models with the highest prediction accuracy were selected as random forest and gradient boosting models.

[0044] S6. Grid search was used to optimize the hyperparameters of the random forest model and the gradient boosting model. The grid parameter search results are shown in Table 2. The optimized model with the best parameter combination was obtained. The hyperparameter optimization results are shown in Table 3.

[0045] Table 2. Model hyperparameter optimization parameter search interval

[0046]

[0047]

[0048] Table 3. Best parameter combinations for model hyperparameter optimization

[0049]

[0050] S7. Test the optimized model with the test set. The model test confusion matrix is ​​as follows: Figure 6The calculated prediction accuracy of the gradient boosting optimization model, μ² = 0.98, was obtained, which is the final prediction model. Given the shear wall parameters for a specific failure mode, the feature engineering method described above was extracted and input into the final model (the optimized gradient boosting model) for testing, which then output the predicted failure mode.

[0051] The above embodiments describe the basic principles, main features, and technical approaches of the present invention, and also demonstrate the feasibility and effectiveness of the present invention in achieving structural damage identification. It should be emphasized that the embodiments described in the present invention are illustrative rather than restrictive, and are not intended to limit the scope of the present invention. Any changes made by those skilled in the art based on the technical solutions of the present invention, whether direct or indirect, are also included in the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for predicting shear wall earthquake damage and failure modes based on PCA fusion machine learning, characterized by: By compiling and analyzing component test data, a dataset of wall seismic damage-sensitive parameters was constructed. PCA high-dimensional feature extraction was performed on the data and input into a classifier model for training, enabling high-precision and rapid identification of shear wall seismic damage patterns. S1. Collect relevant data and construct a seismic damage dataset for shear wall specimen design, response parameters, and corresponding failure modes under cyclic loads. ,in are the sensitive characteristic parameters of the earthquake damage mode, which are r , f ck , n , f y , A s , f yw , ρ w , f sw , ρ sw and V f / V , the specific meanings of the corresponding parameters are shown in Table 1; Y For the shear wall failure mode, the conventional Paulay-Priestley method was used according to the failure form of the specimen; Table 1. Definition of sensitive parameters of shear wall earthquake damage mode S2, characteristic parameters PCA processing was performed, and the principal component dimension was set to 10, which is the same as the sensitive feature parameter; Convert the original parameters into projections in the principal component space. For principal component PC1, in w 1 is the eigenvector of the covariance matrix ∑ The principal component coefficient matrix Extracted as feature engineering as input, N is the number of samples in the dataset; S3, randomly divide the extracted feature engineering for subsequent classifier model training and testing; At the same time, before the model training input, the principal component coefficient matrix of each sample W Normalize to the interval [0,1], that is Where, W.min and W.max are the minimum and maximum values ​​of each row of the principal component coefficient matrix, W std is the normalized coefficient matrix used for model training input; S4. Introducing nine typical machine learning algorithms for classifier model training and testing, including Naïve Bayes, Nearest Neighbors, Logistic Regression, SVC (Support Vector Machine), Decision Tree, Random Forest, AdaBoost, GradientBoosting, and Perceptron Neural Network, to establish a mapping relationship between feature engineering and damage patterns; S5, test the prediction accuracy μ1 of different classifier models, perform model evaluation, and select the classifier model M0 with the highest prediction accuracy; S6. Optimize the hyperparameters of model M0 by using a grid search method to perform optimization calculations to obtain an optimized model M1 with the best parameter combination; S7. Given the shear wall parameters of a failure mode to be predicted, extract the feature engineering according to the S2 method and input it into the model M1 for testing, and the corresponding predicted failure mode can be output.

2. The prediction method according to claim 1, wherein: The step S1: classifying the failure modes into three types, namely, bending failure (F), bending-shear failure (FS) and shear failure (S).

3. The prediction method according to claim 1, wherein: In step S3, 70% of the data is used as a training set, and the remaining 30% is used as a test set.

4. The prediction method according to claim 1, wherein: Step S3: In order to maximize the use of the constructed dataset, a cross-validation method is adopted in the model validation process, that is, a validation set is defined by resampling the training dataset, thereby eliminating the need to separately divide the validation set.

5. The prediction method according to claim 4, wherein: Different sampling ratios and different numbers of iteration steps.

Images