Data-driven column sheet shear stiffness artificial cognition method

Abstract

The invention relates to a data-driven column sheet shear stiffness artificial cognition method. The method comprises the following steps: data acquisition: acquiring data through a column sheet shear performance experiment; establishing a data set: selecting a column piece structure parameter as the input of the data set, taking experimental acquisition data as the output to establish the data set, and dividing a training set and a test set; data arrangement: standardizing the data in the data set; establishing a prediction algorithm: performing parameter tuning on the XGBoost model by using a grid search method, and establishing a prediction model; model explanation: using an SHAP value method, and cognizing main factors influencing the shear stiffness and an action mechanism thereof through a visualization means. The method solves the problem that effective technical measures are difficult to provide to improve the anti-seismic performance of the column piece structure due to insufficient cognition on the shearing behavior of the column piece in the anti-seismic design of the high-rise industrial goods shelf, enhances cognition on internal factors of the shearing rigidity of the column piece and an interaction mechanism of the internal factors, and can accurately and quickly obtain the shearing rigidity value of the column piece.

Description

Technical Field

[0001] The present invention relates to a warehousing and logistics equipment technology, and in particular to a data-driven artificial recognition method for the shear stiffness of a column. Background Art

[0002] The column piece is composed of two columns and several diagonal cross braces. It provides lateral stiffness and stability for high-rise shelves and cargo. Accurately calculating the shear stiffness of the column piece is of great significance for the overall anti-roll design of the shelf under earthquake vibration environment.

[0003] Because the shear forces on columns vary across the structural components, establishing a universal analytical model remains difficult. Mechanical testing is the primary method for determining the shear stiffness of columns, but our understanding of the internal factors and interactions that influence this stiffness remains limited. The industry is eager to explore new research methods to enhance our understanding of column shear behavior, enabling us to implement more effective technical measures to improve the seismic performance of structures. Summary of the Invention

[0004] In order to solve the problem that current technology lacks understanding of the shear behavior of column plates in the seismic design of high-rise industrial shelves, and it is difficult to provide effective technical measures to improve the seismic performance of column plate structures, a data-driven artificial cognition method for column plate shear stiffness is proposed to enhance the understanding of the internal factors of column plate shear stiffness and their interaction mechanism.

[0005] The technical solution of the present invention is:

[0006] A data-driven artificial recognition method for the shear stiffness of a column plate comprises the following steps:

[0007] Data acquisition: Data is collected through the column sheet shear performance test;

[0008] Dataset establishment: Select the column structure parameters as the input of the data set, and the experimental data collected as the output to establish the data set, and divide it into training set and test set;

[0009] Data collation: standardize the data in the dataset;

[0010] Prediction algorithm establishment: Use grid search method to tune the parameters of XGBoost model and establish prediction model;

[0011] Model explanation: Using the SHAP value method, the main factors affecting shear stiffness and their mechanisms are understood through visualization.

[0012] Furthermore, a shear performance test of the column sheet was conducted, which included two horizontally arranged columns; at least one horizontal brace was horizontally arranged between the two columns; and a diagonal brace was obliquely arranged between the two columns; both ends of the horizontal brace and the diagonal brace were bolted to the two columns respectively; the two columns were placed horizontally and their out-of-plane displacement was constrained by the steel frame; nylon pads were placed between the columns and the steel frame to allow longitudinal displacement of the columns; a gap was left between the columns and the top steel frame to allow the columns to twist without excessive out-of-plane displacement; one end of the first column was nailed to prevent out-of-plane displacement, horizontal displacement and rotation around the vertical longitudinal axis; one end of the second column applied a longitudinal force F along the centroid axis of the column; the other end of the second column had a displacement sensor to record the horizontal displacement of the second column.

[0013] Furthermore, the calculation formula of the column shear stiffness is:

[0014]

[0015] Wherein, d is the width of the column piece, which refers to the distance between the centroid axes of the two columns; h is the height of the column piece, and the shear stiffness S ti The dimension is kN.

[0016] Preferably, the structural parameters of the column piece include, for example, the cross-sectional dimensions of the columns, the cross-sectional dimensions of the diagonal braces, the configuration of the diagonal braces, the height and width of the column piece.

[0017] Preferably, the XGBoost model tuning parameters include maximum depth, learning rate, and error verification using 10-fold cross validation.

[0018] Preferably, the visual interpretation means of the SHAP value method include SHAP value contribution diagrams of individual samples, SHAP value summary diagrams, feature dependency diagrams, feature interaction diagrams, and feature importance ranking diagrams.

[0019] Furthermore, shear performance test: a longitudinal force F is applied along the centroid axis of the column at one end in 10 steps, starting from 0 and continuing until failure; a displacement sensor is installed at the other end of the first column to record the horizontal displacement of the first column and obtain the slope k of the load-deformation curve and its linear part. ti ; Finally, the shear stiffness value of the column piece is calculated using the shear stiffness calculation formula.

[0020] Further, the specific steps are as follows:

[0021] During the data collection phase, through the shear performance test of the column piece and in combination with the actual needs of the project, a variety of column pieces of different sizes and shapes were selected for shear performance tests to obtain their shear stiffness;

[0022] During the dataset establishment phase, based on the data obtained from the test, the structural parameters of the column pieces were selected as the input indicators of the database, and the output was the shear stiffness value of the column pieces obtained from the test. Through data standardization, the dimensional influence of the data was eliminated, and an engineering dataset of the shear stiffness of the column pieces was established.

[0023] During the model training phase, the XGBoost algorithm's hyperparameters will be tuned using a grid search algorithm, with key parameters selected for hyperparameter optimization. During the training process, the engineering dataset needs to be divided into a training set and a test set.

[0024] The objective function λ of the XGBoost algorithm (t) :

[0025]

[0026] in, represents the loss function, y i is the shear stiffness of the column obtained through the test, is the predicted value, Ω(f t ) is the regularization term;

[0027] The loss function is Expand the second-order Taylor formula at , integrate the formula, and remove the constant term in the formula to obtain the final objective function

[0028]

[0029] in, γ is the leaf node penalty factor, λ is the leaf weight penalty factor, T and ω j are the number of leaf nodes and the weight value of the j-th leaf respectively;

[0030] During the model interpretation phase, model training was used to obtain the final prediction model for the shear stiffness of the column. Using the SHAP value method, visualization was used to identify the main factors affecting shear stiffness and their mechanisms of action, providing differentiated guidance for improving the structural performance of the column used in racking.

[0031] The principles of the SHAP artificial cognition method are as follows:

[0032]

[0033] Among them, y i is the predicted value obtained by the XGBoost algorithm; x i represents feature j, where j ranges from 1 to p; Represents a constant value with no input, usually the average predicted value; Represents the SHAP value of feature j; input xi and x′ i By x i =h x (x′ i ) related;

[0034] The calculation formula of SHAP value is:

[0035]

[0036] Where T is a subset of features; f(T) is the prediction in feature T;

[0037] SHAP value f(x ij ) indicates the influence of the input feature on the overall result; when f(x ij )>0, indicating that the input feature improves the prediction result and plays a positive role; on the contrary, when f(x ij )<0, indicating that the input feature reduces the prediction result and has a negative effect.

[0038] The beneficial effects of the present invention are:

[0039] This method first provides a method for collecting experimental data on the shear performance of column panels, selects column panels of various sizes to obtain their shear stiffness, and establishes an engineering database; secondly, the grid search method is used to tune the parameters of the XGBoost algorithm (an efficient ensemble learning algorithm) to obtain the final prediction model with the best accuracy and generalization performance; finally, the SHAP artificial cognitive method (an interpretable method based on cooperative game theory) is used to understand the main factors affecting shear stiffness and their action mechanisms through visualization, providing differentiated guidance for improving the structural performance of shelf panels.

[0040] Based on the column piece test, the present invention establishes a data-driven artificial cognition method for the shear stiffness of the column piece based on XGBoost and SHAP. Without the need for real-time updating of the model, accurate and rapid acquisition of the shear stiffness value of the column piece can be achieved, while reducing the experimental cost, providing a new means to further understand the influence mechanism of various structural factors on shear stiffness. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a schematic diagram of the shear performance experiment of the column sheet of the present invention;

[0042] Figure 2 Schematic diagram of vertical restraint in the shear performance experiment of the column piece of the present invention;

[0043] Figure 3 It is a schematic diagram of a specific embodiment of the present invention;

[0044] Figure 4 This is a SHAP summary graph obtained based on an example of the present invention.

[0045] Figure ID:

[0046] 1. Displacement sensor; 2. First column; 3. First cross brace; 4. First diagonal brace; 5. Second cross brace; 6. Second diagonal brace; 7. Third cross brace; 8. Second column; 9. First steel frame; 10. First steel frame; 11. First nylon pad; 12. Second nylon pad. DETAILED DESCRIPTION

[0047] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0048] A data-driven artificial recognition method for the shear stiffness of a column plate comprises the following steps:

[0049] Data acquisition: Data is collected through the column sheet shear performance test;

[0050] Dataset establishment: Select the column structure parameters as the input of the data set, and the experimental data collected as the output to establish the data set, and divide it into training set and test set;

[0051] Data collation: standardize the data in the dataset;

[0052] Prediction algorithm establishment: Use grid search method to tune the parameters of XGBoost model and establish prediction model;

[0053] Model explanation: Using the SHAP value method, the main factors affecting shear stiffness and their mechanisms are understood through visualization.

[0054] Furthermore, the shear performance test of the column piece includes two horizontally arranged columns; at least one horizontal brace is horizontally arranged between the two columns; and a diagonal brace is obliquely arranged between the two columns; both ends of the horizontal brace and the diagonal brace are respectively connected to the two columns by bolts; the two columns are placed horizontally and their out-of-plane displacement is constrained by the steel frame; nylon pads are placed between the columns and the steel frame to allow longitudinal displacement of the columns; a gap is left between the columns and the top steel frame to allow the columns to twist without excessive out-of-plane displacement. One end of the first column is nailed to prevent out-of-plane displacement, horizontal displacement and rotation around the vertical longitudinal axis; one end of the second column applies a longitudinal force F along the column centroid axis; the other end of the second column has a displacement sensor to record the horizontal displacement of the second column.

[0055] Furthermore, the calculation formula of the shear stiffness of the column piece is:

[0056]

[0057] Wherein, d is the width of the column piece, which refers to the distance between the centroid axes of the two columns; h is the height of the column piece, and the shear stiffness S ti The dimension is kN.

[0058] Furthermore, the structural parameters of the column piece include, for example, the cross-sectional dimensions of the columns, the cross-sectional dimensions of the diagonal braces, the configuration of the diagonal braces, the height and width of the column piece, etc.

[0059] Furthermore, XGBoost model tuning parameters include maximum depth, learning rate, and 10-fold cross validation is used for error verification.

[0060] The visual interpretation methods of the SHAP value method include the SHAP value contribution diagram of individual samples, the SHAP value summary diagram, the feature dependency diagram, the feature interaction influence diagram, and the feature importance ranking diagram.

[0061] like Figure 1 、 Figure 2 As shown, in the shear performance test of the column piece in the present invention, the column piece includes columns 2 and 8 and diagonal cross braces 3, 4, 5, 6, and 7; the columns and the diagonal cross braces are connected by bolts; the column piece is horizontally placed on steel frames 9 and 10 to restrict its out-of-plane displacement; nylon pads 11 and 12 are inserted between the columns and the steel frame to allow longitudinal displacement of the columns; a gap is left between the columns and the steel frame to allow the columns to tilt without excessive out-of-plane displacement; one end of the second column 8 is nailed to prevent out-of-plane displacement, horizontal displacement, and rotation around the vertical longitudinal axis;

[0062] Shear performance test: A longitudinal force F is applied along the centroid axis of the column at one end of the first column 2 in 10 steps, starting from 0 and continuing until failure. A displacement sensor 1 is provided at the other end of the first column 2 to record the horizontal displacement of the first column 2 and obtain the slope k of the load-deformation curve and its linear part. ti ; Finally, the shear stiffness value of the column piece is calculated using the shear stiffness calculation formula.

[0063] Figure 3 A data-driven implementation process for the artificial cognition method of the shear stiffness of the column piece is given. The specific steps are as follows:

[0064] During the data collection phase, through the shear performance test of the column pieces and combined with the actual needs of the project, column pieces of various sizes and shapes (diagonal bracing configurations) were selected for shear performance tests to obtain their shear stiffness.

[0065] During the dataset establishment phase, based on the data obtained from the test, column segment structural parameters, such as the cross-sectional dimensions of the columns, the cross-sectional dimensions of the diagonal braces, and the configuration of the diagonal braces, were selected as input indicators for the database, and the output was the shear stiffness value of the column segment obtained from the test. Through data standardization, the dimensional influence of the data was eliminated, and an engineering dataset of the shear stiffness of the column segment was established.

[0066] During the model training phase, the XGBoost algorithm's hyperparameters are tuned using a grid search algorithm. Key parameters, such as maximum depth and learning rate, are selected for hyperparameter optimization to improve model accuracy and generalization performance. During training, the project dataset is divided into a training set and a test set. Generally, 70% of the dataset is used as the training set for model parameter tuning and training the final prediction model, while 30% is used as the test set for evaluating the final prediction model's results.

[0067] The grid search algorithm is a method for optimizing model performance by iterating over given parameter combinations. Its principle can be described as dividing all possible parameter combinations into several grids, then iterating over all intersections in the grid to calculate the error. K-fold cross-validation is then used to find the global optimal solution with the minimum error, effectively reducing model variance and bias. For small sample datasets, k is typically set to 10 based on extensive experimental trials.

[0068] The objective function λ of the XGBoost algorithm (t) :

[0069]

[0070] in, represents the loss function, y i is the shear stiffness of the column obtained through the test, is the predicted value, Ω(f t ) is the regularization term.

[0071] The loss function is Expand the second-order Taylor formula at , integrate the formula, and remove the constant term in the formula to obtain the final objective function

[0072]

[0073] in, γ is the leaf node penalty factor, λ is the leaf weight penalty factor, T and ω j are the number of leaf nodes and the weight value of the j-th leaf respectively.

[0074] During the model interpretation phase, model training resulted in a final prediction model for column shear stiffness. Using the SHAP value method, visualization was used to understand the primary factors influencing shear stiffness and their mechanisms, providing differentiated guidance for improving the structural performance of shelf columns.

[0075] The principles of the SHAP artificial cognition method are as follows:

[0076]

[0077] Among them, y i is the predicted value obtained by the XGBoost algorithm; x i represents feature j, where j ranges from 1 to p; Represents a constant value with no input, usually the average predicted value; Represents the SHAP value of feature j; input x i and x′ i By x i =h x (x′ i ) related.

[0078] The calculation formula of SHAP value is:

[0079]

[0080] Here, T is a subset of features; f(T) is the prediction within feature T.

[0081] SHAP value f(x ij ) indicates the influence of the input feature on the overall result; when f(x ij )>0, indicating that the input feature improves the prediction result and plays a positive role; on the contrary, when f(x ij )<0, indicating that the input feature reduces the prediction result and has a negative effect.

[0082] SHAP visualization methods include individual sample SHAP value contribution plots, SHAP value summary plots, feature dependency plots, feature interaction plots, and feature importance ranking plots. The SHAP value summary plot shows the impact of all input features on the output; the feature dependency plot shows the impact of changes in a single input feature on the output; the feature interaction plot shows the impact of changes in two input features on the output; and the feature importance ranking plot shows the average absolute value of the SHAP values ​​of each input feature and ranks them, which can be considered as the importance ranking of each input feature.

[0083] Figure 4The figure shows a SHAP summary graph generated by an optional embodiment of the present invention. The horizontal axis indicates how the SHAP value is affected by feature changes, the vertical axis indicates the feature type, and the color of the dot indicates the magnitude of the feature value. The feature ranking from top to bottom on the vertical axis also represents the importance ranking of all 14 input features in this example. The top three features, UP2 (the wall thickness of the column in the column plate), BM1 (the configuration of the diagonal braces), and UP2 (the cross-sectional dimensions of the column), indicate that the column cross-sectional area and the diagonal brace structure are the primary factors affecting the shear stiffness of the column plate. BM5 (the support wall thickness) and the column plate height (FH) rank next, indicating that they have some influence on shear stiffness and can be considered secondary factors. BM4 (cross-sectional dimensions of the diagonal braces), BM2 (diagonal brace flange length), UP3 (column open end dimensions), and d (column plate width) have limited influence on shear stiffness. The remaining five features (such as the FS column plate span) have relatively little influence and can be considered unimportant. Observing the changes in each row's eigenvalues ​​individually, we find that the larger the UP2 (pillar wall thickness), the higher the SHAP value, indicating that increasing column wall thickness increases shear stiffness. Similar patterns are observed for UP1 and BM5. By analyzing the SHAP summary graph, we identify the primary factors and patterns influencing the shear stiffness of the column in this example, achieving data-driven artificial cognition.

[0084] The above-described embodiment merely represents one embodiment of the present invention. While the description is relatively specific and detailed, it should not be construed as limiting the scope of the patent. It should be noted that a person skilled in the art would be able to make various modifications and improvements without departing from the spirit of the present invention, and these modifications and improvements fall within the scope of protection of the present invention. Therefore, the scope of protection of the patent for this invention shall be determined by the appended claims.

Claims

1. A data-driven artificial recognition method for the shear stiffness of column plates, characterized in that: The following steps are involved: Data acquisition: Data is collected through the column sheet shear performance test; Dataset establishment: Select the column structure parameters as the input of the data set, and the experimental data collected as the output to establish the data set, and divide it into training set and test set; Data collation: standardize the data in the dataset; Prediction algorithm establishment: Use grid search method to tune the parameters of XGBoost model and establish prediction model; Model explanation: Using the SHAP value method, the main factors affecting shear stiffness and their mechanisms are understood through visualization.

2. The data-driven artificial recognition method for the shear stiffness of column pieces according to claim 1 is characterized in that: The shear performance test of the column sheet includes two horizontally arranged columns; at least one horizontal brace is horizontally arranged between the two columns; and a diagonal brace is obliquely arranged between the two columns; both ends of the horizontal brace and the diagonal brace are respectively connected to the two columns with bolts; the two columns are placed horizontally and their out-of-plane displacement is constrained by the steel frame; nylon pads are placed between the columns and the steel frame to allow longitudinal displacement of the columns; a gap is left between the columns and the top steel frame to allow the columns to twist without excessive out-of-plane displacement; one end of the first column is nailed to prevent out-of-plane displacement, horizontal displacement and rotation around the vertical longitudinal axis; one end of the second column applies a longitudinal force F along the column centroid axis; the other end of the second column has a displacement sensor to record the horizontal displacement of the second column.

3. The data-driven artificial recognition method for column shear stiffness according to claim 1, characterized in that: The calculation formula of the column shear stiffness is: Wherein, d is the width of the column piece, which refers to the distance between the centroid axes of the two columns; h is the height of the column piece, and the shear stiffness S ti The dimension is kN.

4. The data-driven artificial recognition method for column shear stiffness according to claim 1, characterized in that: The structural parameters of the column piece include, for example, the cross-sectional dimensions of the columns, the cross-sectional dimensions of the diagonal braces, the configuration of the diagonal braces, the height and width of the column piece.

5. The data-driven artificial recognition method for column shear stiffness according to claim 1 is characterized in that: The XGBoost model tuning parameters include maximum depth, learning rate, and 10-fold cross validation for error verification.

6. The data-driven artificial recognition method for the shear stiffness of column pieces according to claim 1 is characterized in that: The visual interpretation methods of the SHAP value method include the SHAP value contribution diagram of individual samples, SHAP value summary diagram, feature dependency diagram, feature interaction influence diagram, and feature importance ranking diagram.

7. The data-driven artificial recognition method for the shear stiffness of column pieces according to claim 2, characterized in that: Shear performance test: A longitudinal force F is applied along the centroid axis of the column at one end in 10 steps, starting from 0 and continuing until failure. A displacement sensor is installed at the other end of the first column to record the horizontal displacement of the first column and obtain the slope k of the load-deformation curve and its linear part. ti ; Finally, the shear stiffness value of the column piece is calculated using the shear stiffness calculation formula.

8. The data-driven artificial recognition method for column shear stiffness according to claim 1, characterized in that: The specific steps are as follows: During the data collection phase, through the shear performance test of the column piece and in combination with the actual needs of the project, a variety of column pieces of different sizes and shapes were selected for shear performance tests to obtain their shear stiffness; During the dataset establishment phase, based on the data obtained from the test, the structural parameters of the column pieces were selected as the input indicators of the database, and the output was the shear stiffness value of the column pieces obtained from the test. Through data standardization, the dimensional influence of the data was eliminated, and an engineering dataset of the shear stiffness of the column pieces was established. During the model training phase, the XGBoost algorithm's hyperparameters will be tuned using a grid search algorithm, with more critical parameters selected for hyperparameter optimization. During the training process, the engineering data set needs to be divided into a training set and a test set; The objective function λ of the XGBoost algorithm (t) : in, represents the loss function, y i is the shear stiffness of the column obtained through the test, is the predicted value, Ω(f t ) is the regularization term; The loss function is Expand the second-order Taylor formula at , integrate the formula, and remove the constant term in the formula to obtain the final objective function in, γ is the leaf node penalty factor, λ is the leaf weight penalty factor, T and ω j are the number of leaf nodes and the weight value of the j-th leaf respectively; During the model interpretation phase, model training was used to obtain the final prediction model for the shear stiffness of the column. Using the SHAP value method, visualization was used to identify the main factors affecting shear stiffness and their mechanisms of action, providing differentiated guidance for improving the structural performance of the column used in racking. The principles of the SHAP artificial cognition method are as follows: Among them, y i is the predicted value obtained by the XGBoost algorithm; x i represents feature j, where j ranges from 1 to p; Represents a constant value with no input, usually the average predicted value; Represents the SHAP value of feature j; input x i and x′ i By x i =h x (x′ i ) related; The calculation formula of SHAP value is: Where T is a subset of features; f(T) is the prediction in feature T; SHAP value f(x ij ) indicates the influence of the input feature on the overall result; when f(x ij )>0, indicating that the input feature improves the prediction result and plays a positive role; on the contrary, when f(x ij )<0, indicating that the input feature reduces the prediction result and has a negative effect.

Images