Rapid assessment method for explosion damage of corrugated steel concrete composite structure

Abstract

The invention provides a rapid assessment method for explosion damage of a corrugated steel concrete composite structure. The rapid assessment method comprises the following steps: constructing a damage assessment data set: dividing the data set into a training set and a test set through data cleaning and regularization processing; the data comprises characteristics and corresponding target variables, the characteristics comprise the explosive load, the proportional explosion distance, the concrete thickness, the corrugated steel plate thickness, the waveform of corrugated steel and the concrete strength, and the target variables comprise regression prediction values; training the built random forest model by adopting the training set: randomly selecting a new data set in a replacement manner each time, independently training each decision tree in the random forest model, and averaging prediction results of all the decision trees to obtain a final maximum displacement after explosion of the corrugated steel concrete composite structure and a final displacement regression prediction result; after the model is evaluated and optimized, the features are input into the trained random forest model to predict the maximum displacement and the final displacement after displacement explosion.

Description

Technical Field

[0001] The invention relates to an explosion assessment technology, in particular to a rapid assessment method for explosion damage of a corrugated steel-concrete composite structure. Background Art

[0002] At present, there is a lack of research on the dynamic response of corrugated steel-concrete (CSPC) composite structures under blast loads. Researchers tend to selectively focus on factors that have a greater impact on the structure, while ignoring several minor influencing factors. On this basis, the least squares curve fitting method is used to fit the experimental or simulation data to predict the empirical formula. Although this method can obtain a preliminary prediction result, the large error will limit the accuracy of the prediction result. Summary of the invention

[0003] The present invention provides a method for rapid assessment of explosion damage of a corrugated steel-concrete composite structure, comprising: Step S100, constructing a damage assessment data set: dividing the data set into a training set and a test set through data cleaning and regularization processing; the data includes features and corresponding target variables, wherein the features include charge amount, proportional explosion distance, concrete thickness, corrugated steel plate thickness, corrugated steel waveform and concrete strength, and the target variable includes the value of regression prediction; Step S200, using the training set to train the constructed random forest model: randomly selecting a new data set with replacement each time and independently training each decision tree in the random forest model, averaging the prediction results of all decision trees to obtain the maximum displacement and final displacement regression prediction results of the final corrugated steel-concrete composite structure after explosion; Step S300, after model evaluation and tuning, the features are input into the trained random forest model to predict the maximum displacement and final displacement after displacement explosion.

[0004] Furthermore, in step S200, the number of decision trees is determined M , feature selection methods, decision tree growth methods, and use the scikit-learn library to build a random forest model.

[0005] Further, in step S200, random sampling is performed from the training set with replacement. m A decision tree is constructed to build a new N m A dataset of samples D m , select the optimal features through the influence of the features in the training set on the uncertainty of the data set, and build a decision tree. The process is: Step S201, constructing a root node: placing all training data at the root node, selecting an optimal feature, and using the optimal feature as the segmentation criterion for the current node; Step S202, data segmentation: the training data set is segmented into subsets according to the selected features, each subset corresponds to a value of the feature; if these subsets can be correctly classified, then leaf nodes are constructed and these subsets are divided into corresponding leaf nodes; if there are still subsets that cannot be correctly classified, then the optimal features are reselected for these subsets, and they are further segmented to construct corresponding nodes; Step S203, recursively constructing a subtree: repeating data segmentation for each data subset on each leaf node to construct a subtree until the maximum tree body is met and then stopping the construction; Step S204, generating leaf nodes: generating leaf nodes and assigning classification or regression results.

[0006] Furthermore, the method for obtaining the optimal features is as follows: Step S211, define the data set as , dataset D m Random sampling from the features {charge amount, proportional explosion distance, concrete thickness, corrugated steel plate thickness, corrugated steel waveform, concrete strength}, Step S212, obtain the i Features Information entropy ,in For the i Features k The number of samples in each category, For the i The total number of categories of samples with features; Step S203, obtain the i Features The conditional entropy ,in N For the dataset D The total number of features; Step S204, obtain the i Features For the dataset D The degree of influence G ( D , D i )

[0007] Step S205, select The corresponding eigenvalue is taken as the optimal eigenvalue.

[0008] Further, in step S200, forM The prediction results of the decision trees are averaged to obtain the final regression prediction result. The calculation method is as follows:

[0009] in, Represents the prediction result of the mth decision tree.

[0010] Compared with the prior art, the present invention has the following advantages: (1) In order to improve the accuracy and reliability of prediction, the present invention uses a machine learning method that can simultaneously consider multiple influencing factors on the basis of a large number of numerical simulation results to predict the mid-span displacement of a corrugated steel-concrete composite structure under different conditions such as explosive load forms, charge amounts, proportional explosion distances, concrete thickness, corrugated steel plate thickness, corrugated steel waveforms and concrete strength; (2) This prediction model can realize the prediction of the maximum and final mid-span displacement of any parameter composite structure under contact, shallow buried and close-range explosion loads. Combined with the corresponding structural damage standards, the damage degree and failure mode of the corrugated steel-concrete composite structure can be evaluated, which can provide a reference for the explosion protection design and optimization of structures such as corrugated steel culverts, tunnels and bridges.

[0011] The present invention will be further described below in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a schematic diagram of the method flow of the present invention.

[0013] Figure 2 Comparison diagram of the maximum and final displacements in the mid-span under the action of contact explosion between simulation and model prediction, where (a) is the comparison diagram of the maximum displacement in the mid-span, and (b) is the comparison diagram of the final displacement in the mid-span.

[0014] Figure 3 The figure shows the comparison of the maximum and final displacement values ​​of the mid-span under the action of shallow-buried and close-range explosion between simulation and model prediction, where (a) is the comparison of the maximum displacement of the mid-span, and (b) is the comparison of the final displacement of the mid-span.

[0015] Figure 4 Comparison diagram of the predicted results of the contact explosion test set working condition and the finite element simulation results, where (a) is the comparison diagram of the maximum displacement in the mid-span, and (b) is the comparison diagram of the final displacement in the mid-span.

[0016] Figure 5 The following is a comparison chart of the predicted results of the shallow buried close-range explosion test set and the finite element simulation results, where (a) is the comparison chart of the maximum displacement in the mid-span, and (b) is the comparison chart of the final displacement in the mid-span. DETAILED DESCRIPTION

[0017] Combination Figure 5A rapid assessment method for explosion damage of a corrugated steel-concrete composite structure comprises the following steps: Step S100, constructing a damage assessment data set: dividing the data set into a training set and a test set through data cleaning and regularization processing; the data includes features and corresponding target variables, wherein the features include charge amount, proportional explosion distance, concrete thickness, corrugated steel plate thickness and concrete strength, and the target variable includes the value of regression prediction; Step S200, using the training set to train the constructed random forest model: randomly selecting a new data set with replacement each time and independently training each decision tree in the random forest model, averaging the prediction results of all decision trees to obtain the maximum displacement and final displacement regression prediction results of the final corrugated steel-concrete composite structure after explosion; Step S300, after model evaluation and tuning, the charge amount, proportional explosion distance, concrete thickness, corrugated steel plate thickness, and concrete strength are input into the trained random forest model to predict the maximum displacement and final displacement after displacement explosion.

[0018] In step S100, the training set is used to train the model, and the test set is used to evaluate the performance of the model. This embodiment establishes several shallow buried and contact explosion conditions of different working conditions by selecting different features, and some of the conditions are used as training sets, and the other conditions are used as test sets.

[0019] In step S200, the number of decision trees is determined M , feature selection method, decision tree growth method and other parameters, and use the scikit-learn library to build a random forest model.

[0020] In step S200, random sampling is performed from the training set with replacement. m A decision tree is constructed to build a new N m A dataset of samples D m , select the optimal features through the influence of the features in the training set on the uncertainty of the data set, and build a decision tree. The process is: Step S201, constructing a root node: placing all training data at the root node, selecting an optimal feature, and using the optimal feature as the segmentation criterion for the current node; Step S202, data segmentation: the training data set is segmented into subsets according to the selected features, each subset corresponds to a value of the feature; if these subsets can be correctly classified, then leaf nodes are constructed and these subsets are divided into corresponding leaf nodes; if there are still subsets that cannot be correctly classified, then the optimal features are reselected for these subsets, and they are further segmented to construct corresponding nodes; Step S203, recursively constructing a subtree: repeating data segmentation for each data subset on each leaf node to construct a subtree until the maximum tree body is met and then stopping the construction; Step S204, generating leaf nodes: generating leaf nodes and assigning classification or regression results.

[0021] The method for obtaining the optimal features is as follows: Step S211, define the data set as , dataset D m Random sampling from the features {charge, proportional explosion distance, concrete thickness, corrugated steel plate thickness, concrete strength}, Step S212, obtain the i Features Information entropy ,in For the i Features k The number of samples in each category, For the i The total number of categories of samples with features; Step S203, obtain the i Features The conditional entropy ,in N For the dataset D The total number of features; Step S204, obtain the i Features For the dataset D The degree of influence G ( D , D i )

[0022] Step S205, select The corresponding eigenvalue is taken as the optimal eigenvalue.

[0023] In step S200, for M The prediction results of the decision trees are averaged to obtain the final regression prediction result. The calculation method is as follows:

[0024] in, Represents the prediction result of the mth decision tree.

[0025] In step S300, the number of decision trees and the number of training samples are set.

[0026] Comparison between the predicted value and the actual value of the mid-span displacement of the corrugated steel-concrete composite structure plate under contact explosion Figure 2 As shown in the figure, the comparison between the predicted value and the actual value of the mid-span displacement of the corrugated steel-concrete composite structure arch under shallow explosion is shown in the figure. Figure 3 It can be seen that the prediction model implemented in this paper can well complete the maximum displacement and final displacement of the mid-span of the corrugated steel-concrete composite structure with arbitrary parameters under two working conditions.

[0027] The regression performance indicator determination coefficient (R-squared) is used to evaluate the accuracy and generalization ability of the prediction model. The determination coefficient characterizes the quality of model fitting through the change of data. It is determined by SSR (Sum of Squares of the Regression), that is, the sum of the squares of the difference between the predicted data and the original data mean, and SST (Total sum of squares), that is, the sum of the squares of the difference between the original data and the mean. The specific calculation formula is as follows:

[0028]

[0029]

[0030] in is the mean of the target true value. The range of the coefficient of determination is [0,1]. The closer the result is to 1, the better the variables in the equation are. y The stronger the explanatory power, the better the model fits the data. The prediction results show that the R-squared values ​​of the contact explosion and shallow buried explosion prediction models are 0.98 and 0.94 respectively, proving that the model has good prediction accuracy. In order to further verify the accuracy of the prediction model, a total of 4 contact explosion and 4 shallow buried close-range explosion numerical simulation test sets were carried out to verify the working conditions. The model prediction results and simulation results are shown in Figure 4 , Figure 5 , it can be seen that the model can make a good prediction of the deformation of the combined structure.

Claims

1. A method for rapid assessment of explosion damage of corrugated steel-concrete composite structures, characterized in that: include: Step S100, constructing a damage assessment data set: dividing the data set into a training set and a test set through data cleaning and regularization processing; It includes features and corresponding target variables, wherein the features include charge amount, proportional explosion distance, concrete thickness, corrugated steel plate thickness, corrugated steel waveform and concrete strength, and the target variable includes the value of regression prediction; Step S200, using the training set to train the constructed random forest model: randomly selecting a new data set with replacement each time and independently training each decision tree in the random forest model, averaging the prediction results of all decision trees to obtain the maximum displacement and final displacement regression prediction results of the final corrugated steel-concrete composite structure after explosion; Step S300, after model evaluation and tuning, the features are input into the trained random forest model to predict the maximum displacement and final displacement after displacement explosion.

2. The method according to claim 1, characterized in that In step S200, the number of decision trees is determined M , feature selection methods, decision tree growth methods, and use the scikit-learn library to build a random forest model.

3. The method according to claim 2, characterized in that In step S200, random sampling is performed from the training set with replacement. m A decision tree is constructed to build a new N m A dataset of samples D m , select the optimal features through the influence of the features in the training set on the uncertainty of the data set, and build a decision tree. The process is: Step S201, constructing a root node: placing all training data at the root node, selecting an optimal feature, and using the optimal feature as the segmentation criterion for the current node; Step S202, data segmentation: the training data set is segmented into subsets according to the selected features, each subset corresponds to a value of the feature; if these subsets can be correctly classified, then leaf nodes are constructed and these subsets are divided into corresponding leaf nodes; if there are still subsets that cannot be correctly classified, then the optimal features are reselected for these subsets, and they are further segmented to construct corresponding nodes; Step S203, recursively constructing a subtree: repeating data segmentation for each data subset on each leaf node to construct a subtree until the maximum tree body is met and then stopping the construction; Step S204, generating leaf nodes: generating leaf nodes and assigning classification or regression results.

4. The method according to claim 3, characterized in that: The method for obtaining the optimal features is as follows: Step S211, define the data set as , dataset D m Randomly sample from the features {charge amount, proportional explosion distance, concrete thickness, corrugated steel plate thickness, corrugated steel waveform, concrete strength}, Step S212, obtain the i Features Information entropy ,in For the i Features k The number of samples in each category, For the i The total number of categories of samples with features; Step S203, obtain the i Features The conditional entropy ,in N For the dataset D The total number of features; Step S204, obtain the i Features For the dataset D The degree of influence G ( D , D i ) ， Step S205, select The corresponding eigenvalue is taken as the optimal eigenvalue.

5. The method according to claim 4, characterized in that In step S200, for M The prediction results of the decision trees are averaged to obtain the final regression prediction result. The calculation method is as follows: ， in, Represents the prediction result of the mth decision tree.

Images