A metal roof wind disaster vulnerability assessment method and system based on reinforcement learning

Abstract

The present invention provides a method and system for assessing the wind disaster vulnerability of metal roofs based on reinforcement learning, which belongs to the technical field of reinforcement learning. The method includes: obtaining the three-dimensional structure of the metal roof under different wind conditions and the damaged label for each position point in the three-dimensional structure of the roof based on the industrial Internet platform, and performing reinforcement learning to construct a vulnerability assessment model; obtaining the current three-dimensional structure of the target roof, and evaluating the current three-dimensional structure based on the vulnerability assessment model to obtain the current label set; performing quantitative analysis on the current label set to obtain a number of vulnerability features under different dimensions, and establishing a vulnerability matrix for the target roof to achieve vulnerability assessment. Human participation can be avoided, the standardization of damage analysis is guaranteed, and the accuracy of the assessment is effectively achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of reinforcement learning, and in particular to a method and system for assessing the wind damage vulnerability of a metal roof based on reinforcement learning. Background Art

[0002] The Industrial Internet platform is aimed at the digitalization, networking and intelligent needs of the manufacturing industry. It builds a service system based on massive data collection, aggregation and analysis, and supports the ubiquitous connection, elastic supply and efficient configuration of manufacturing resources. Among them, it is particularly important to obtain the structure and damage of metal roofs under wind conditions from the Industrial Internet platform to conduct vulnerability analysis.

[0003] Vulnerability analysis is carried out by manually measuring and analyzing the metal roof after wind damage. This process requires manual control of the damage to the metal layer after the wind disaster. There will be certain assessment errors in this process, resulting in inaccurate assessment results.

[0004] Therefore, the present invention proposes a metal roof wind disaster vulnerability assessment method and system based on reinforcement learning. Summary of the invention

[0005] The present invention provides a method and system for assessing the wind disaster vulnerability of metal roofs based on reinforcement learning, which is used to construct a vulnerability assessment model through reinforcement learning to intelligently analyze the damage of the roof structure and to quantitatively analyze the label set to construct a vulnerability matrix, thereby avoiding human participation, ensuring the standardization of damage analysis, and effectively achieving the accuracy of the assessment.

[0006] The present invention provides a method for assessing the wind damage vulnerability of metal roofs based on reinforcement learning, comprising:

[0007] Step 1: Based on the industrial Internet platform, the three-dimensional structure of the metal roof under different wind conditions and the damaged label of each position point in the three-dimensional structure of the roof are obtained, and reinforcement learning is performed to build a vulnerability assessment model;

[0008] Step 2: Obtain the current three-dimensional structure of the target roof, and evaluate the current three-dimensional structure based on the vulnerability assessment model to obtain a current label set;

[0009] Step 3: Quantitatively analyze the current label set to obtain several vulnerability features in different dimensions, and establish a vulnerability matrix for the target roof to achieve vulnerability assessment.

[0010] Preferably, reinforcement learning is performed to construct a vulnerability assessment model, including:

[0011] Obtaining an initial evaluation model after reinforcement learning, and selecting any two groups of damaged labels of the same metal roof under the same wind force condition, performing label averaging processing on the two groups of damaged labels to obtain evaluation output samples, and performing position averaging processing on the three-dimensional roof structures of the metal roofs corresponding to the two groups of damaged labels to obtain evaluation input samples;

[0012] Inputting the evaluation input sample into the initial evaluation model to obtain a comparison output sample, and comparing it with the evaluation output sample to determine a single evaluation accuracy of the initial evaluation model;

[0013] Determining the comprehensive evaluation accuracy of the initial evaluation model based on all single evaluation accuracies under different wind conditions;

[0014] When the comprehensive assessment accuracy is greater than a preset accuracy, the initial assessment model is regarded as a vulnerability assessment model.

[0015] Preferably, performing reinforcement learning to construct a vulnerability assessment model further includes:

[0016] When the comprehensive evaluation accuracy is not greater than the preset accuracy, The wind conditions to be trained are selected from all wind conditions, and the analysis logs consistent with each wind condition to be trained are retrieved from the learning logs of the reinforcement learning process of the initial evaluation model, wherein: is the preset accuracy; is the single assessment accuracy of the i-th wind condition; To comprehensively evaluate the accuracy; is the logarithmic function symbol; is a constant greater than 0;

[0017] Extracting a training factor set from the analysis log, and calculating a compliance coefficient of the training factor set;

[0018] At the same time, the standard roof structure before the wind condition to be trained and the first roof structure at different wind interference time points under the wind condition to be trained are obtained, and a position structure deformation matrix corresponding to the wind condition to be trained is constructed;

[0019] Adjusting abnormal elements in the position structure deformation matrix to obtain a new deformation matrix;

[0020] Determining eigenvectors of the new deformation matrix;

[0021] If the compliance coefficient is not greater than the preset coefficient, extracting the maximum value and the minimum value in each column vector from the new deformation matrix, and constructing a first vector based on all maximum values ​​and a second vector based on all minimum values;

[0022] Calculate the variance and mean of each column vector in the new deformation matrix;

[0023] Relying on the characteristic vector, the first vector, the second vector, the value variance and the value mean, construct a new sample corresponding to the wind force condition to be trained;

[0024] If the compliance coefficient is greater than a preset coefficient, two vectors are randomly extracted from the new deformation matrix, and combined with the eigenvector to construct a new sample corresponding to the wind condition to be trained;

[0025] The initial assessment model is continuously subjected to reinforcement learning based on all new samples obtained to obtain a vulnerability assessment model.

[0026] Preferably, adjusting the abnormal elements in the position structure deformation matrix to obtain a new deformation matrix includes:

[0027] The position structure deformation matrix is ​​normalized to obtain a standard matrix, and the standard matrix is ​​input into the function analysis model to obtain the maximum residual corresponding to the discrete equation of each row vector;

[0028] Selecting the minimum residual and the second smallest residual from all the maximum residuals, and obtaining the difference between the second smallest residual and the minimum residual;

[0029] Locking the elements that do not meet the structural deformation consistency standard from each row vector of the standard matrix and treating them as abnormal elements;

[0030] Based on the magnitude relationship between the value of the abnormal element and the structural deformation consistency standard, and in combination with the difference, the value of the abnormal element is adjusted to obtain a new deformation matrix.

[0031] Preferably, based on the magnitude relationship between the value of the abnormal element and the structural deformation consistency standard, and in combination with the difference, the value of the abnormal element is adjusted, including:

[0032] When the value of the abnormal element is larger than the structural deformation consistency standard, the value of the abnormal element is adjusted to be smaller based on the difference;

[0033] ;

[0034] in, Indicates the result of the downward adjustment of the corresponding abnormal element; Indicates the value of the corresponding abnormal element; Indicates the difference; represents the standard value based on the structural deformation consistency standard; represents the maximum residual of the row vector corresponding to the abnormal element; n1 represents the number of elements in the row vector corresponding to the abnormal element that need to be adjusted to a smaller value; Represents the variance of the difference between the values ​​of all elements that need to be adjusted to a smaller value and the standard value in the row vector corresponding to the abnormal element;

[0035] When the value of the abnormal element is smaller than the structural deformation consistency standard, the value of the abnormal element is adjusted to be larger based on the difference;

[0036] ;

[0037] in, represents the result of the upward adjustment of the corresponding abnormal element; n2 represents the number of elements that need to be adjusted upward in the row vector where the corresponding abnormal element is located; It represents the variance of the difference between the standard value and the values ​​of all elements that need to be adjusted larger in the row vector corresponding to the abnormal element.

[0038] Preferably, the current label set is quantitatively analyzed to obtain several vulnerability features in different dimensions, including:

[0039] Determine the vulnerability assessment dimensions, and perform quantitative analysis on the current tag set according to each assessment dimension;

[0040] The quantitative analysis results are input into the feature analysis model of the corresponding dimension to obtain the vulnerability characteristics.

[0041] Preferably, establishing a vulnerability matrix for the target roof to implement vulnerability assessment includes:

[0042] Build a vulnerability matrix ,in, It is the vulnerability characteristic under the first assessment dimension; is the vulnerability characteristic under the mth assessment dimension;

[0043] Calculate the feature weight of each vulnerability feature separately;

[0044] ;

[0045] in, Indicates The weight of the vulnerability characteristics under each assessment dimension; Indicates all The maximum value in ; Indicates the total number of types of wind conditions; represents the vulnerability characteristics of the g2nd assessment dimension under the g1th wind condition Vulnerability characteristics under non-g2 assessment dimensions The correlation function of Represents the vulnerability characteristics under all wind conditions except for the g2th evaluation dimension Vulnerability characteristics under non-g2nd evaluation dimensions The variance of the remaining functions other than the correlation function;

[0046] Construct a weight matrix based on the feature weights ,in, is the weight of the vulnerability feature under the first assessment dimension; is the weight of the vulnerability feature under the mth assessment dimension;

[0047] Based on the vulnerability matrix And the weight matrix , calculate the single value for each feature and the overall value for all features, and construct a value vector;

[0048] The value vector is matched with the vector-evaluation comparison table to obtain an evaluation result.

[0049] The present invention provides a metal roof wind disaster vulnerability assessment system based on reinforcement learning, comprising:

[0050] A model building module is used to obtain the three-dimensional structure of the metal roof under different wind conditions and the damaged label of each position point in the three-dimensional structure of the roof based on the industrial Internet platform, and to perform reinforcement learning to build a vulnerability assessment model;

[0051] A structure acquisition module, used for acquiring the current three-dimensional structure of the target roof, and evaluating the current three-dimensional structure based on the vulnerability assessment model to obtain a current label set;

[0052] The vulnerability assessment module is used to quantitatively analyze the current label set to obtain a number of vulnerability features in different dimensions, and to establish a vulnerability matrix for the target roof to achieve vulnerability assessment.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] By building a vulnerability assessment model through reinforcement learning to intelligently analyze the damage of the roof structure and quantitatively analyze the label set to build a vulnerability matrix, human participation can be avoided, the standardization of damage analysis is ensured, and the accuracy of the assessment is effectively achieved.

[0055] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0056] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0058] Figure 1 It is a flow chart of a method for assessing the wind damage vulnerability of a metal roof based on reinforcement learning in an embodiment of the present invention;

[0059] Figure 2 The structure diagram of a metal roof wind disaster vulnerability assessment system based on reinforcement learning in an embodiment of the present invention. DETAILED DESCRIPTION

[0060] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0061] The present invention provides a method for assessing the wind damage vulnerability of metal roofs based on reinforcement learning. Figure 1 As shown, including:

[0062] Step 1: Based on the industrial Internet platform, the three-dimensional structure of the metal roof under different wind conditions and the damaged label of each position point in the three-dimensional structure of the roof are obtained, and reinforcement learning is performed to build a vulnerability assessment model;

[0063] Step 2: Obtain the current three-dimensional structure of the target roof, and evaluate the current three-dimensional structure based on the vulnerability assessment model to obtain a current label set;

[0064] Step 3: Quantitatively analyze the current label set to obtain several vulnerability features in different dimensions, and establish a vulnerability matrix for the target roof to achieve vulnerability assessment.

[0065] In this embodiment, the industrial Internet platform stores the three-dimensional roof structure of the metal roof under different wind conditions and the damaged label of each position point, wherein the three-dimensional roof structure is for the metal roof, and the metal roof refers to a roof form that uses metal plates as roofing materials and combines the structural layer and the waterproof layer into one, and the damaged label can refer to the damage of different position points based on wind loads, such as the height of the metal plate warping, the position offset of the plate, etc., and then the samples can be trained, that is, {input: the actual three-dimensional roof structure of the metal roof and the coordinates of each position point output: the damaged label of the corresponding position point}.

[0066] In this embodiment, the three-dimensional structure of the roof refers to the point cloud scanning of the metal roof after the metal roof is subjected to corresponding wind damage, and the vulnerability assessment model is obtained based on the training of the neural network model.

[0067] In this embodiment, the wind force condition refers to the combination of different wind speeds and wind directions, and ultimately acts on the metal roof to obtain a damaged label under the wind force condition.

[0068] In this embodiment, the current three-dimensional structure is obtained based on point cloud scanning of the metal roof that needs damage analysis, and the point cloud scanning technology belongs to the existing technology.

[0069] In this embodiment, the current label set is directly outputted by analyzing the current three-dimensional structure based on the vulnerability assessment model, and the set includes damaged labels of different position points of the current three-dimensional structure.

[0070] In this embodiment, the quantitative analysis is performed from three dimensions: the exposure level, sensitivity level, and resilience of the vulnerability to comprehensively determine the vulnerability.

[0071] In this embodiment, the vulnerability matrix is ​​constructed based on the vulnerability characteristics, and the vulnerability assessment is determined based on the vulnerability matrix and the corresponding weights.

[0072] The beneficial effects of the above technical solution are: by constructing a vulnerability assessment model through reinforcement learning to intelligently analyze the damage of the roof structure and to quantitatively analyze the label set to construct a vulnerability matrix, human participation can be avoided, the standardization of damage analysis is ensured, and the accuracy of the assessment is effectively achieved.

[0073] The present invention provides a method for assessing the wind damage vulnerability of metal roofs based on reinforcement learning, and performs reinforcement learning to construct a vulnerability assessment model, including:

[0074] Obtaining an initial evaluation model after reinforcement learning, and selecting any two groups of damaged labels of the same metal roof under the same wind force condition, performing label averaging processing on the two groups of damaged labels to obtain evaluation output samples, and performing position averaging processing on the three-dimensional roof structures of the metal roofs corresponding to the two groups of damaged labels to obtain evaluation input samples;

[0075] Inputting the evaluation input sample into the initial evaluation model to obtain a comparison output sample, and comparing it with the evaluation output sample to determine a single evaluation accuracy of the initial evaluation model;

[0076] Determining the comprehensive evaluation accuracy of the initial evaluation model based on all single evaluation accuracies under different wind conditions;

[0077] When the comprehensive assessment accuracy is greater than a preset accuracy, the initial assessment model is regarded as a vulnerability assessment model.

[0078] In this embodiment, any two groups of damaged labels are obtained in order to average the labels. For example, label 11 indicates that the plate is 3 cm high and label 21 indicates that the plate is 3.2 cm high. At this time, after averaging, the plate is 3.1 cm high. It should be noted that the metal layer can be in a state of no damage or in a state of damage. That is, samples under different wind conditions with different damage conditions are obtained as much as possible, and diverse samples are provided as much as possible to provide a training basis for building a model.

[0079] In this embodiment, the evaluation input sample is: the three-dimensional structure of the roof after position averaging processing, and the evaluation output sample is: the damaged label after label averaging processing.

[0080] In this embodiment, the initial evaluation model refers to the model obtained after strengthening the learning of the neural network according to the samples obtained from the industrial Internet platform.

[0081] In this embodiment, the comparison output sample is an output result obtained based on the model, which is convenient for comparison with the evaluation output sample.

[0082] In this embodiment, the single evaluation accuracy=the label difference of the same position point determined based on the comparison output sample and the evaluation output sample×the sum of the weights of the corresponding position points.

[0083] In this embodiment, the comprehensive evaluation accuracy is the average value of all single evaluation accuracies.

[0084] In this embodiment, the preset accuracy value is 0.7.

[0085] The beneficial effect of the above technical solution is: by averaging any two groups of damaged labels under different wind conditions, it is convenient to obtain the corresponding evaluation input and output samples, and then compare the output samples, so as to facilitate a single evaluation and then realize a comprehensive evaluation, providing an accurate analysis basis for building a vulnerability assessment model.

[0086] The present invention provides a method for assessing the wind damage vulnerability of metal roofs based on reinforcement learning, wherein reinforcement learning is performed to construct a vulnerability assessment model, and the method further comprises:

[0087] When the comprehensive evaluation accuracy is not greater than the preset accuracy, The wind conditions to be trained are selected from all wind conditions, and the analysis logs consistent with each wind condition to be trained are retrieved from the learning logs of the reinforcement learning process of the initial evaluation model, wherein: is the preset accuracy; is the single assessment accuracy of the i-th wind condition; To comprehensively evaluate the accuracy; is the logarithmic function symbol; is a constant greater than 0;

[0088] Extracting a training factor set from the analysis log, and calculating a compliance coefficient of the training factor set;

[0089] At the same time, the standard roof structure before the wind condition to be trained and the first roof structure at different wind interference time points under the wind condition to be trained are obtained, and a position structure deformation matrix corresponding to the wind condition to be trained is constructed;

[0090] Adjusting abnormal elements in the position structure deformation matrix to obtain a new deformation matrix;

[0091] Determining eigenvectors of the new deformation matrix;

[0092] If the compliance coefficient is not greater than the preset coefficient, extracting the maximum value and the minimum value in each column vector from the new deformation matrix, and constructing a first vector based on all maximum values ​​and a second vector based on all minimum values;

[0093] Calculate the variance and mean of each column vector in the new deformation matrix;

[0094] Relying on the characteristic vector, the first vector, the second vector, the value variance and the value mean, construct a new sample corresponding to the wind force condition to be trained;

[0095] If the compliance coefficient is greater than a preset coefficient, two vectors are randomly extracted from the new deformation matrix, and combined with the eigenvector to construct a new sample corresponding to the wind condition to be trained;

[0096] The initial assessment model is continuously subjected to reinforcement learning based on all acquired new samples to obtain a vulnerability assessment model.

[0097] In this embodiment, the wind force condition to be trained is selected from the following conditions: The log of model learning will be generated in the process of reinforcement learning, which is automatically generated. Therefore, the analysis log of the wind condition can be directly retrieved later.

[0098] In this embodiment, the training factor set includes predefined training factors and values ​​for the factors.

[0099] That is, a series of parameter information will be generated in the analysis log. For example, the predefined training factors are: learning efficiency, the first number of the label learning parameter involved in the wind force condition to be trained, the second number of the label learning parameter not involved in the wind force condition to be trained, etc. At this time, the value corresponding to the learning efficiency is 80%, the value corresponding to the first number is 30, and the value corresponding to the second number is 20.

[0100] In this embodiment, ;

[0101] When the actual production value of the corresponding factor is greater than the standard value of the corresponding factor, the sub-coefficient of the corresponding factor is regarded as 1; otherwise, the sub-coefficient of the corresponding factor is regarded as (standard value of the corresponding factor - actual production value of the corresponding factor) / standard value of the corresponding factor.

[0102] At this time, the compliance coefficient is the average value of all sub-coefficients.

[0103] In this embodiment, The value of is 0.5.

[0104] In this embodiment, the first roof structure refers to a structure corresponding to a metal roof under the wind conditions to be trained.

[0105] In this embodiment,

[0106] .

[0107] In this embodiment, an abnormal element refers to an element that does not meet the structural deformation consistency standard. For example, under normal circumstances, if the warping height of position point 1 at time point 01 is 3 cm, the corresponding warping height of position point 2 at time point 01 should be between 1 cm and 2 cm. However, the warping height of position point 2 at time point 01 is 3.4 cm. At this time, 3.4 cm is regarded as an abnormal element.

[0108] In this embodiment, the eigenvector is obtained based on the eigenanalysis of the new deformation matrix, which belongs to the prior art.

[0109] In this embodiment, the value of the preset coefficient is 0.6.

[0110] In this embodiment, the first vector={the maximum value of the elements in each column of the new deformation matrix}.

[0111] The second vector = {the minimum value of each column element in the new deformation matrix}.

[0112] In this embodiment, the value variance is calculated based on the variance formula, and the value mean is obtained by averaging the values ​​in the column vector.

[0113] In this embodiment, the new sample whose compliance coefficient is not greater than the preset coefficient:

[0114] First vector, second vector, eigenvector, , , and compared with the corresponding three-dimensional structural level to obtain a new sample, where each element in the vector is the damage condition of the corresponding position point.

[0115] In this embodiment, the new sample whose compliance coefficient is greater than the preset coefficient:

[0116] The two vectors and feature vectors are then extracted and compared with the corresponding three-dimensional structural levels to obtain new samples.

[0117] In this embodiment, the reinforcement learning of the initial evaluation model is to continue to train the model using the acquired new samples to improve the training accuracy.

[0118] The beneficial effects of the above technical solution are: analyzing the situation where the comprehensive evaluation accuracy is less than the preset accuracy, specifically analyzing the compliance coefficient of the factors by screening the logs of the wind conditions to be trained, providing a basis for the subsequent determination of new samples in two ways, and further locking and adjusting the abnormal elements in the matrix to achieve the fullness of the new sample construction, thereby ensuring the accuracy of the initial evaluation model after learning.

[0119] The present invention provides a method for assessing the wind damage vulnerability of metal roofs based on reinforcement learning, which adjusts the abnormal elements in the position structure deformation matrix to obtain a new deformation matrix, including:

[0120] The position structure deformation matrix is ​​normalized to obtain a standard matrix, and the standard matrix is ​​input into the function analysis model to obtain the maximum residual corresponding to the discrete equation of each row vector;

[0121] Selecting the minimum residual and the second smallest residual from all the maximum residuals, and obtaining the difference between the second smallest residual and the minimum residual;

[0122] Locking the elements that do not meet the structural deformation consistency standard from each row vector of the standard matrix and treating them as abnormal elements;

[0123] Based on the magnitude relationship between the value of the abnormal element and the structural deformation consistency standard, and in combination with the difference, the value of the abnormal element is adjusted to obtain a new deformation matrix.

[0124] Preferably, when the value of the abnormal element is larger than the structural deformation consistency standard, the value of the abnormal element is adjusted to be smaller based on the difference;

[0125] ;

[0126] in, Indicates the result of the downward adjustment of the corresponding abnormal element; Indicates the value of the corresponding abnormal element; Indicates the difference; represents the standard value based on the structural deformation consistency standard; represents the maximum residual of the row vector corresponding to the abnormal element; n1 represents the number of elements in the row vector corresponding to the abnormal element that need to be adjusted to a smaller value; Represents the variance of the difference between the values ​​of all elements that need to be adjusted to a smaller value and the standard value in the row vector corresponding to the abnormal element;

[0127] When the value of the abnormal element is smaller than the structural deformation consistency standard, the value of the abnormal element is adjusted to be larger based on the difference;

[0128] ;

[0129] in, represents the result of the upward adjustment of the corresponding abnormal element; n2 represents the number of elements that need to be adjusted upward in the row vector where the corresponding abnormal element is located; It represents the variance of the difference between the standard value and the values ​​of all elements that need to be adjusted larger in the row vector corresponding to the abnormal element.

[0130] In this embodiment, if the warping height of position point 1 at time point 01 is 3 cm, the warping height of the corresponding position point 2 at time point 01 should be between 1 cm and 2 cm. At this time, the warping height of the corresponding position point 2 is 3.4 cm, which is considered to be larger than the consistent standard based on the structural deformation. If the warping height of the corresponding position point 2 is 0.8 cm, it is considered to be smaller than the consistent standard based on the structural deformation.

[0131] In this embodiment, the purpose of the standardization process is to normalize the values ​​of the elements in the matrix, thereby facilitating the rationality of subsequent processing.

[0132] In this embodiment, the function analysis model is obtained by training the neural network model based on different standard row vectors and discrete residuals of row vectors as samples, so the maximum residual can be directly obtained.

[0133] In this embodiment, the discretization is based on the discretization of the corresponding row vector and the structural type variation consistency standard. The greater the degree of discretization, the larger the corresponding residual, and the discrete equation is determined after eliminating one element in the row vector respectively. At this time, there will be a residual. A maximum value is extracted from all the residuals involved in the row vector and is regarded as the maximum residual.

[0134] In this embodiment, each row vector corresponds to a maximum residual. At this time, a minimum and a second minimum residual are extracted from all the maximum residuals. At this time, the difference = the second minimum residual - the minimum residual.

[0135] In this embodiment, after the adjustment is performed, the numerical processing is performed again to eliminate the normalization of the numerical value after the standardization processing, and the value in the actual unit dimension is obtained.

[0136] The beneficial effects of the above technical solution are: by standardizing the matrix, it is easy to ensure the uniformity of values ​​and the convenience of calculation, and by obtaining the maximum residual of each row vector respectively to determine the difference, and combining the size relationship to adjust the value of the abnormal element, thereby ensuring the accuracy of the matrix and providing a reliable foundation for model learning.

[0137] The present invention provides a method for assessing the wind damage vulnerability of metal roofs based on reinforcement learning, which quantitatively analyzes the current label set to obtain several vulnerability features in different dimensions, including:

[0138] Determine the vulnerability assessment dimensions, and perform quantitative analysis on the current tag set according to each assessment dimension;

[0139] The quantitative analysis results are input into the feature analysis model of the corresponding dimension to obtain the vulnerability characteristics.

[0140] In this embodiment, quantitative analysis refers to the analysis of the exposure degree, sensitivity degree, and resilience of the information of the labels involved, that is, the quantitative analysis results are obtained based on the label information and the exposure-sensitivity-recovery comparison table, and the quantitative analysis results under each dimension are {quantitative values ​​of each position point}, and the table contains the exposure degree, sensitivity degree, and resilience corresponding to different label information.

[0141] In this embodiment, the feature analysis model is obtained by training a neural network model based on samples of different quantitative analysis results and features matching the results, and vulnerability features can be directly obtained, such as exposure level A1, sensitivity level A2, and resilience A3.

[0142] The beneficial effect of the above technical solution is: by quantitatively analyzing the label set through the evaluation dimension, the corresponding features are obtained, providing a basis for vulnerability assessment.

[0143] The present invention provides a method for assessing the wind damage vulnerability of metal roofs based on reinforcement learning, establishing a vulnerability matrix for the target roof and implementing vulnerability assessment, including:

[0144] Build a vulnerability matrix ,in, It is the vulnerability characteristic under the first assessment dimension; is the vulnerability characteristic under the mth assessment dimension;

[0145] Calculate the feature weight of each vulnerability feature separately;

[0146] ;

[0147] in, Indicates The weight of the vulnerability characteristics under each assessment dimension; Indicates all The maximum value in ; Indicates the total number of types of wind conditions; represents the vulnerability characteristics of the g2nd assessment dimension under the g1th wind condition Vulnerability characteristics under non-g2nd evaluation dimensions The correlation function of Represents the vulnerability characteristics under all wind conditions except for the g2th evaluation dimension Vulnerability characteristics under non-g2nd evaluation dimensions The variance of the remaining functions other than the correlation function;

[0148] Construct a weight matrix based on the feature weights ,in, is the weight of the vulnerability feature under the first assessment dimension; is the weight of the vulnerability feature under the mth assessment dimension;

[0149] Based on the vulnerability matrix And the weight matrix , calculate the single value for each feature and the overall value for all features, and construct a value vector;

[0150] The value vector is matched with the vector-evaluation comparison table to obtain an evaluation result.

[0151] In this example, the single value is: .

[0152] The overall value is: .

[0153] Value vector = {overall value all single values}.

[0154] In this embodiment, the vector-assessment comparison table includes different value vectors and assessment results that match the vectors, which can be directly matched, and the assessment results may be, for example: the economic loss is approximately 100,000, the repair difficulty is three stars, etc.

[0155] The beneficial effect of the above technical solution is: by establishing a vulnerability matrix and calculating the characteristic weight of each vulnerability feature, it is easy to calculate the single value and the characteristic value, which facilitates the matching of the evaluation results and further ensures the accuracy of the evaluation.

[0156] The present invention provides a metal roof wind disaster vulnerability assessment system based on reinforcement learning, such as Figure 2 As shown, including:

[0157] A model building module is used to obtain the three-dimensional structure of the metal roof under different wind conditions and the damaged label of each position point in the three-dimensional structure of the roof based on the industrial Internet platform, and to perform reinforcement learning to build a vulnerability assessment model;

[0158] A structure acquisition module, used for acquiring the current three-dimensional structure of the target roof, and evaluating the current three-dimensional structure based on the vulnerability assessment model to obtain a current label set;

[0159] The vulnerability assessment module is used to quantitatively analyze the current label set to obtain a number of vulnerability features in different dimensions, and to establish a vulnerability matrix for the target roof to achieve vulnerability assessment.

[0160] The beneficial effects of the above technical solution are: by constructing a vulnerability assessment model through reinforcement learning to intelligently analyze the damage of the roof structure and to quantitatively analyze the label set to construct a vulnerability matrix, human participation can be avoided, the standardization of damage analysis is ensured, and the accuracy of the assessment is effectively achieved.

[0161] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A metal roof wind damage vulnerability assessment method based on reinforcement learning, characterized in that: include: Step 1: Based on the industrial Internet platform, the three-dimensional structure of the metal roof under different wind conditions and the damaged label of each position point in the three-dimensional structure of the roof are obtained, and reinforcement learning is performed to build a vulnerability assessment model; Step 2: obtaining the current three-dimensional structure of the target roof, and evaluating the current three-dimensional structure based on the vulnerability assessment model to obtain a current label set; Step 3: Quantitatively analyze the current label set to obtain several vulnerability features in different dimensions, and establish a vulnerability matrix for the target roof to achieve vulnerability assessment; Among them, reinforcement learning is used to build a vulnerability assessment model, including: Obtaining an initial evaluation model after reinforcement learning, and selecting any two groups of damaged labels of the same metal roof under the same wind force condition, performing label averaging processing on the two groups of damaged labels to obtain evaluation output samples, and performing position averaging processing on the three-dimensional roof structures of the metal roofs corresponding to the two groups of damaged labels to obtain evaluation input samples; Inputting the evaluation input sample into the initial evaluation model to obtain a comparison output sample, and comparing it with the evaluation output sample to determine a single evaluation accuracy of the initial evaluation model; Determining the comprehensive evaluation accuracy of the initial evaluation model based on all single evaluation accuracies under different wind conditions; When the comprehensive assessment accuracy is greater than a preset accuracy, the initial assessment model is regarded as a vulnerability assessment model; In addition, reinforcement learning is used to build a vulnerability assessment model, which also includes: When the comprehensive evaluation accuracy is not greater than the preset accuracy, The wind conditions to be trained are selected from all wind conditions, and the analysis logs consistent with each wind condition to be trained are retrieved from the learning logs of the reinforcement learning process of the initial evaluation model, wherein: is the preset accuracy; is the single assessment accuracy of the i-th wind condition; To comprehensively evaluate the accuracy; is the logarithmic function symbol; is a constant greater than 0; Extracting a training factor set from the analysis log, and calculating a compliance coefficient of the training factor set; At the same time, the standard roof structure before the wind condition to be trained and the first roof structure at different wind interference time points under the wind condition to be trained are obtained, and a position structure deformation matrix corresponding to the wind condition to be trained is constructed; Adjusting abnormal elements in the position structure deformation matrix to obtain a new deformation matrix; Determining eigenvectors of the new deformation matrix; If the compliance coefficient is not greater than the preset coefficient, extracting the maximum value and the minimum value in each column vector from the new deformation matrix, and constructing a first vector based on all maximum values ​​and a second vector based on all minimum values; Calculate the variance and mean of each column vector in the new deformation matrix; Relying on the characteristic vector, the first vector, the second vector, the value variance and the value mean, construct a new sample corresponding to the wind force condition to be trained; If the compliance coefficient is greater than a preset coefficient, two vectors are randomly extracted from the new deformation matrix, and combined with the eigenvector to construct a new sample corresponding to the wind condition to be trained; The initial assessment model is continuously subjected to reinforcement learning based on all acquired new samples to obtain a vulnerability assessment model.

2. The method for assessing wind damage vulnerability of metal roofs based on reinforcement learning according to claim 1, characterized in that: The abnormal elements in the position structure deformation matrix are adjusted to obtain a new deformation matrix, including: The position structure deformation matrix is ​​normalized to obtain a standard matrix, and the standard matrix is ​​input into the function analysis model to obtain the maximum residual corresponding to the discrete equation of each row vector; Selecting the minimum residual and the second smallest residual from all the maximum residuals, and obtaining the difference between the second smallest residual and the minimum residual; Locking the elements that do not meet the structural deformation consistency standard from each row vector of the standard matrix and treating them as abnormal elements; Based on the magnitude relationship between the value of the abnormal element and the structural deformation consistency standard, and in combination with the difference, the value of the abnormal element is adjusted to obtain a new deformation matrix.

3. The method for assessing wind damage vulnerability of metal roofs based on reinforcement learning according to claim 2, characterized in that: Based on the magnitude relationship between the value of the abnormal element and the structural deformation consistency standard, and in combination with the difference, the value of the abnormal element is adjusted, including: When the value of the abnormal element is larger than the structural deformation consistency standard, the value of the abnormal element is adjusted to be smaller based on the difference; ； in, Indicates the result of the downward adjustment of the corresponding abnormal element; Indicates the value of the corresponding abnormal element; Indicates the difference; represents the standard value based on the structural deformation consistency standard; Represents the maximum residual of the row vector corresponding to the abnormal element; Indicates the number of elements that need to be adjusted down in the row vector corresponding to the abnormal element; Represents the variance of the difference between the values ​​of all elements that need to be adjusted to a smaller value and the standard value in the row vector corresponding to the abnormal element; When the value of the abnormal element is smaller than the structural deformation consistency standard, the value of the abnormal element is adjusted to be larger based on the difference; ； in, represents the result of the upward adjustment of the corresponding abnormal element; n2 represents the number of elements that need to be adjusted upward in the row vector where the corresponding abnormal element is located; It represents the variance of the difference between the standard value and the values ​​of all elements that need to be adjusted larger in the row vector corresponding to the abnormal element.

4. The method for assessing wind damage vulnerability of metal roofs based on reinforcement learning according to claim 1, characterized in that: A quantitative analysis of the current label set is performed to obtain several vulnerability features in different dimensions, including: Determine the vulnerability assessment dimensions, and perform quantitative analysis on the current tag set according to each assessment dimension; The quantitative analysis results are input into the feature analysis model of the corresponding dimension to obtain the vulnerability characteristics.

5. The metal roof wind disaster vulnerability assessment method based on reinforcement learning according to claim 1, characterized in that: Establish a vulnerability matrix for the target roof to achieve vulnerability assessment, including: Build a vulnerability matrix ,in, It is the vulnerability characteristic under the first assessment dimension; is the vulnerability characteristic under the mth assessment dimension; Calculate the feature weight of each vulnerability feature separately; ； in, Indicates The weight of the vulnerability characteristics under each assessment dimension; Indicates all The maximum value in ; Indicates the total number of types of wind conditions; Indicates In the case of wind force Vulnerability characteristics under each assessment dimension With non Vulnerability characteristics under each assessment dimension The correlation function of Indicates all wind conditions except for Vulnerability characteristics under each assessment dimension With non Vulnerability characteristics under each assessment dimension The variance of the remaining functions other than the correlation function; Construct a weight matrix based on the feature weights ,in, is the weight of the vulnerability feature under the first assessment dimension; is the weight of the vulnerability feature under the mth assessment dimension; Based on the vulnerability matrix And the weight matrix , calculate the single value for each feature and the overall value for all features, and construct a value vector; The value vector is matched with the vector-evaluation comparison table to obtain an evaluation result.

6. A metal roof wind disaster vulnerability assessment system based on reinforcement learning, characterized in that: include: A model building module is used to obtain the three-dimensional structure of the metal roof under different wind conditions and the damaged label of each position point in the three-dimensional structure of the roof based on the industrial Internet platform, and to perform reinforcement learning to build a vulnerability assessment model; A structure acquisition module, used for acquiring the current three-dimensional structure of the target roof, and evaluating the current three-dimensional structure based on the vulnerability assessment model to obtain a current label set; A vulnerability assessment module is used to quantitatively analyze the current label set to obtain a number of vulnerability features in different dimensions, and to establish a vulnerability matrix for the target roof to achieve vulnerability assessment; Among them, reinforcement learning is used to build a vulnerability assessment model, including: Obtaining an initial evaluation model after reinforcement learning, and selecting any two groups of damaged labels of the same metal roof under the same wind force condition, performing label averaging processing on the two groups of damaged labels to obtain evaluation output samples, and performing position averaging processing on the three-dimensional roof structures of the metal roofs corresponding to the two groups of damaged labels to obtain evaluation input samples; Inputting the evaluation input sample into the initial evaluation model to obtain a comparison output sample, and comparing it with the evaluation output sample to determine a single evaluation accuracy of the initial evaluation model; Determining the comprehensive evaluation accuracy of the initial evaluation model based on all single evaluation accuracies under different wind conditions; When the comprehensive assessment accuracy is greater than a preset accuracy, the initial assessment model is regarded as a vulnerability assessment model; In addition, reinforcement learning is used to build a vulnerability assessment model, which also includes: When the comprehensive evaluation accuracy is not greater than the preset accuracy, The wind conditions to be trained are selected from all wind conditions, and the analysis logs consistent with each wind condition to be trained are retrieved from the learning logs of the reinforcement learning process of the initial evaluation model, wherein: is the preset accuracy; is the single assessment accuracy of the i-th wind condition; To comprehensively evaluate the accuracy; is the logarithmic function symbol; is a constant greater than 0; Extracting a training factor set from the analysis log, and calculating a compliance coefficient of the training factor set; At the same time, the standard roof structure before the wind condition to be trained and the first roof structure at different wind interference time points under the wind condition to be trained are obtained, and a position structure deformation matrix corresponding to the wind condition to be trained is constructed; Adjusting abnormal elements in the position structure deformation matrix to obtain a new deformation matrix; Determining eigenvectors of the new deformation matrix; If the compliance coefficient is not greater than the preset coefficient, extracting the maximum value and the minimum value in each column vector from the new deformation matrix, and constructing a first vector based on all maximum values ​​and a second vector based on all minimum values; Calculate the variance and mean of each column vector in the new deformation matrix; Relying on the characteristic vector, the first vector, the second vector, the value variance and the value mean, construct a new sample corresponding to the wind force condition to be trained; If the compliance coefficient is greater than a preset coefficient, two vectors are randomly extracted from the new deformation matrix, and combined with the eigenvector to construct a new sample corresponding to the wind condition to be trained; The initial assessment model is continuously subjected to reinforcement learning based on all acquired new samples to obtain a vulnerability assessment model.

Images