A training method, device and medium for a structural risk calculation model

Abstract

The present invention provides a training method, device and medium for a structural risk calculation model, which belongs to the technical field of bridge structure monitoring, establishes a training sample data set; constructs a change rate matrix of the training sample; defines variables, and reconstructs a multivariate phase space; then extracts features; divides based on different monitoring categories to obtain corresponding sets; spatially interpolates the degree of change of the change rate to obtain a risk space; divides the degree of change of the monitoring data change rate to obtain a risk space of the same category of monitoring data change rate; superimposes the risk space of each monitoring category, and outputs the risk level space after the spatial risk of all monitoring categories is superimposed. The present invention analyzes the risk situation of the three-dimensional space of the bridge structure in real time, so that the potential risk hazards in the bridge structure can be predicted and prevented in advance, thereby improving the service life of the bridge structure.

Description

Technical Field

[0001] The invention belongs to the technical field of bridge structure monitoring, and in particular relates to a training method, equipment and medium for a structure risk calculation model. Background Art

[0002] Bridge engineering has now entered the stage of large-scale repair and maintenance from the era of large-scale infrastructure. Deploying and installing various types of monitoring sensor equipment on important structural nodes of bridges and using Internet of Things technology to perceive the operation status of bridges in real time are crucial to ensuring the safe operation of urban lifelines. Scientific observation of the long-term performance of bridges during their service and real-time perception and analysis of safety risk situations are the hot spots and focuses of current industry research.

[0003] In the prior art, the analysis of the risk status of bridge safety operation mainly includes the following two aspects: (1) Based on finite element analysis, when an abnormality occurs at a certain monitoring point, the structural system of the bridge is calculated to conduct a detailed analysis of its stress state; (2) Based on a semi-quantitative risk analysis model, such as setting the classification and weight of the risk assessment index system through hierarchical analysis method, risk matrix, expert scoring, etc., to achieve an overall analysis of the risk status of bridge structure safety.

[0004] However, evaluating the risk of safe operation of bridges through finite element analysis requires the use of professional analysis software and the establishment of a large finite element model, which often cannot perform real-time spatial interpolation of the rate of change of bridge monitoring data, nor can it construct a three-dimensional risk space of the bridge structure based on the risk level of the monitoring data change. For semi-quantitative risk analysis models, there is a certain degree of subjectivity, and it is impossible to obtain the risk space of each monitoring category of the bridge structure from an objective perspective, and thus lacks judgment on the risk level space. Summary of the invention

[0005] The present invention provides a training method for a structural risk calculation model. The method analyzes the risk situation of the structural space in real time from point to surface and from surface to body, so as to predict and prevent potential risks in the space in advance, improve the service life of the bridge and ensure its safe operation.

[0006] Methods include:

[0007] Step S101: establishing a training sample data set including a plurality of training samples, each training sample having a coordinate attribute;

[0008] Step S102: pre-processing the training samples, and calculating the change rate of each training sample by difference method with a preset time interval as a calculation cycle, and constructing a change rate matrix of the training samples;

[0009] Step S103: define a variable in the training sample change rate matrix as a dependent variable and the other variables as independent variables; reconstruct the training sample change rate matrix in a multivariate phase space, and obtain different characteristic states based on the projection of the chaotic attractor on each dimension of the reconstructed phase space;

[0010] Step S104: input the reconstructed feature state into the deep extreme learning machine, perform feature extraction through the sigmoid activation function in the first hidden layer, and extract the quantitative relationship between the variables constituting each feature state;

[0011] Step S105: obtaining an inverse function based on the quantitative relationship, using the inverse function as the activation function of the third hidden layer in the deep extreme learning machine, and calculating the degree of change of the change rates of other related monitoring data when the change rate of the target monitoring data changes;

[0012] Step S106: using a nonlinear support vector machine as an activation function of the fourth hidden layer to divide different monitoring categories to which the monitoring data change rate belongs, to obtain a set of different monitoring categories;

[0013] Step S107: Based on the classification result of the support vector machine, the Kriging interpolation function is used as the activation function in the fifth hidden layer to perform spatial interpolation on the degree of change of the change rate between the points where the change rate of the same category of monitoring data is located, so as to obtain the change rate space of the same monitoring category, that is, the monitoring category risk space;

[0014] Step S108: In the sixth hidden layer, a linear support vector machine is used as an activation function to divide the range of the change degree of the monitoring data change rate in the same category change rate space, and the risk level is defined based on the range of the change degree to obtain a monitoring category risk space with different risk levels;

[0015] Step S109: Use the weighted superposition function as the activation function of the last hidden layer in the deep extreme learning machine to superimpose the risk level space of each monitoring category, and output the risk level space after the spatial risks of all monitoring categories are superimposed.

[0016] It should be further explained that in step S102, the change rate matrix of the training samples is constructed ,include:

[0017] The training samples are preprocessed to obtain a training sample data set containing multiple variables: , and are the number of variables and The sequence length of variables; the training sample data set is converted into a training sample data matrix, in which each row represents a sample and each column represents a variable;

[0018] According to the difference method, the change rate of each training sample is calculated with the minimum time interval as the calculation period, and the change rate matrix of the training sample is obtained. , specifically expressed as follows:

[0019]

[0020] in,

[0021] ,

[0022] ,

[0023]

[0024] …

[0025] .

[0026] It should be further explained that, in step S103, the bridge structure is defined as a nonlinear chaotic system, and the chaotic characteristics of the variable sequence are determined based on the Lyapunov exponent;

[0027] Define a variable in the training sample change rate matrix is the dependent variable, denoted by , and the other variables are Independent variable, among which , the rate of change matrix of the training samples Perform multivariate phase space reconstruction and obtain different characteristic states based on the projection of chaotic attractors on each dimension of the reconstructed phase space. , Represents the dimension of the reconstructed phase space, including:

[0028] Reconstruct the multivariate phase space and solve the embedding dimension according to the CC algorithm and optimal delay time , and get the embedding dimension and optimal delay time After that, The variables are reconstructed into the new phase space respectively, and the matrix data obtained is:

[0029]

[0030] in, Represents the phase space after reconstruction dimensional space, is the number of variables; each row Represents different feature states, each of which is represented by the dependent variable and independent variables constitute.

[0031] It should be further explained that, in step S104, the network structure of the extreme learning machine autoencoder of the deep extreme learning machine includes an input layer, multiple hidden layers, and an output layer;

[0032] The reconstructed matrix data is used as the input data of the deep extreme learning machine input layer, and the sigmoid activation function is used in the first hidden layer to extract the quantitative relationship between the dependent variable and the independent variable. ,include:

[0033] Use sigmoid activation function to complete the reconstructed feature space Medium feature status Analytical extraction:

[0034] in, Represents the feature space after mapping characteristic information; Indicates the monitoring data change rate Reconstructed data; Indicates row transposition; , All are numbers greater than 0;

[0035] Calculate the dependent variable With independent variable The quantitative relationship between :

[0036]

[0037]

[0038]

[0039] in, Represents the input-output mapping relationship between hidden layers; are input features, ; is the input feature The corresponding weighting coefficient; is the dimension of the reconstructed phase space.

[0040] It should be further explained that, in step S105, the inverse function is defined as: ;

[0041] Independent Variable The degree of change for , ;

[0042] in, Indicates the independent variable exist The value of the moment, Dependent variable exist The value of the moment, Indicates the independent variable exist The value of the moment, Dependent variable exist The value of the moment.

[0043] It should be further explained that in step S106, the nonlinear support vector machine is used as the activation function of the fourth hidden layer to divide the different monitoring categories to which the monitoring data change rate belongs, and the set of different monitoring categories is obtained. , specifically:

[0044] Select kernel function and penalty data Construct and solve the formula:

[0045]

[0046]

[0047] ,

[0048] Find the optimal solution ,choose A positive component of , the calculation method is:

[0049] Constructing a decision function

[0050] pass Implement different monitoring categories in turn Division of

[0051] in, Indicates the independent variable Category information, that is, category information of the change rate of each monitoring data.

[0052] It should be further explained that in step S107, the Kriging interpolation function is used as the activation function to calculate the change rate of the same type of monitoring data. The rate of change between Perform spatial interpolation to obtain the change rate space after interpolation of the change rate of the same category of monitoring data, that is, the risk space ;

[0053] in Indicates the number of variables, Indicates different monitoring categories;

[0054] Define the points to be interpolated Property Value Estimated value of ,

[0055] In the formula, ,express The degree of change in the monitoring data change rate of each point under the monitoring category;

[0056] The unbiased condition of the predicted value is defined as:

[0057] The calculation formula for defining the estimated variance is:

[0058]

[0059] In the formula, For two points The mean covariance between For monitoring category The number of all points in ;

[0060] Under the unbiased condition constraint, the estimated variance is Reaching the minimum weight coefficients ;

[0061] Use Kriging interpolation function to analyze each monitoring category The interpolation of the monitoring data change rate is calculated in turn to obtain the change rate space of different monitoring categories .

[0062] It should be further explained that

[0063] In step S108, a linear support vector machine is used as an activation function to sequentially calculate the change rate space of the same category. The degree of change in the monitoring data change rate The scope of the change is divided into different levels, and the risk level is determined based on the extent of the change. The definition of , obtains the risk space of monitoring category change rate with different risk levels , followed by: Select penalty data , construct and solve the convex quadratic programming formula as:

[0064]

[0065]

[0066] Find the optimal solution ,

[0067] calculate

[0068] choose A component of condition ,calculate,

[0069]

[0070] The separating hyperplane is obtained as:

[0071] The classification decision function is:

[0072] in, Represents the monitoring category The monitoring data change rate in

[0073] The change rates of different monitoring categories are divided into four appropriate ranges in turn, and the smaller intervals after division are defined as risk levels in turn. , output the risk space of monitoring category change rate with different risk levels .

[0074] It should be further explained that in step S109, the weighted superposition function is used as the activation function of the last hidden layer in the deep extreme learning machine to superimpose the risk space of each monitoring category. , output the risk level space after superposition of all monitoring category spatial risks :

[0075]

[0076] in, Represents the risk space of different monitoring categories The superposition weight of , indicating its relative importance to the final result; Indicates different monitoring categories, Represents the total number of all monitoring categories.

[0077] According to another embodiment of the present application, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method for training a structural risk calculation model when executing the program.

[0078] According to another embodiment of the present application, a storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method for training the structural risk calculation model are implemented.

[0079] It can be seen from the above technical solutions that the present invention has the following advantages:

[0080] The training method of the three-dimensional spatial risk calculation model of the structure provided in the present application is based on multivariate phase space reconstruction and deep extreme learning machine training of the spatial risk calculation model based on the monitoring data change rate. The monitoring data change rates of multiple monitoring categories in the same structural space are integrated so that the phase points in the reconstructed multidimensional phase space contain more diverse information on the monitoring data change rate. Then, the deep extreme learning machine is used to train the spatial risk calculation model based on the monitoring data change rate. By analyzing the quantitative relationship between the monitoring data change rate of a certain point in a certain monitoring category in the structural space and the change rate of the same category of monitoring data at different positions in the space and the change rate of monitoring data of different categories layer by layer, based on the change rate in the time series trend analysis, it is clear that when the monitoring data at a certain place in the structural space changes, the change trend of other related monitoring data is clarified, so that based on one point, from point to surface, from surface to body, the changing trend of the overall risk of the structure can be analyzed in real time.

[0081] This application can realize real-time daily analysis of the overall risk situation of the structure without relying on finite element analysis software, and improve the timeliness of risk analysis from the perspective of daily management. This application can also objectively analyze the safety risk situation of the overall structure in a quantitative way. Compared with the semi-quantitative risk analysis model, it reduces certain subjective judgments, improves the reliability of risk analysis, and meets the safety management requirements for bridges. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] In order to more clearly illustrate the technical solution of the present invention, the accompanying drawings required for use in the description will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0083] Figure 1 A calculation flow chart of a training method for a structural risk calculation model;

[0084] Figure 2 The flowchart of the embodiment of calculating the three-dimensional spatial risk of a structure based on the monitoring data change rate;

[0085] Figure 3 Schematic diagram of an electronic device. DETAILED DESCRIPTION

[0086] The training method for the structural risk calculation model provided in the present application is mainly aimed at calculating the overall structural risk situation of the bridge. When the change rate of the monitoring data at any point in the structural space changes, the present application can timely and quantitatively analyze the changes in the risk level of the structural space. From the perspective of daily management, it can realize quantitative real-time daily analysis of the overall risk change situation of the structure, thereby improving the timeliness and subjective dependence of risk analysis.

[0087] The following description will specifically explain the training method of the three-dimensional spatial risk calculation model of the structure. For the purpose of explanation rather than limitation, the following takes the structural response data of the cable-stayed bridge as an example, and proposes specific details such as specific system structures and technologies to facilitate a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application can also be implemented in other embodiments without these specific details.

[0088] The phrases such as "one embodiment" or "some embodiments" described in the present application mean that the specific features, structures or characteristics described in the embodiment are included in one or more embodiments of the present application. Therefore, the phrases such as "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments" etc. that appear in different places in the present application do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways.

[0089] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0090] See also Figure 1 FIG. 1 is a flow chart of a method for training a structural risk calculation model in a specific embodiment, the method comprising:

[0091] Step S101: establishing a training sample data set including a plurality of training samples, each training sample having a coordinate attribute.

[0092] In this embodiment, taking the structural response data of a cable-stayed bridge as an example, a training sample data set including a plurality of training samples with spatial coordinate attributes is established.

[0093] It should be noted that the data set includes main beam vertical displacement monitoring data, main beam lateral displacement monitoring data, support displacement monitoring data, beam end longitudinal displacement monitoring data, tower top displacement monitoring data, tower top rotation monitoring data, beam end horizontal rotation monitoring data, beam end vertical rotation monitoring data, main beam key section strain monitoring data, cable tower key section strain monitoring data, cable force monitoring data, support reaction monitoring data, main beam vertical vibration acceleration monitoring data, main beam lateral vibration acceleration monitoring data, main beam longitudinal vibration acceleration monitoring data, tower top horizontal bidirectional vibration acceleration monitoring data, and cable vibration acceleration monitoring data.

[0094] Among them, the main beam vertical displacement monitoring data, the main beam lateral displacement monitoring data, the support displacement monitoring data, the beam end longitudinal displacement monitoring data, and the tower top displacement monitoring data belong to the displacement monitoring category. The tower top rotation angle monitoring data, the beam end horizontal rotation angle monitoring data, and the beam end vertical rotation angle monitoring data belong to the rotation angle monitoring category. The main beam key section strain monitoring data and the cable tower key section strain monitoring data belong to the strain monitoring category. The support reaction monitoring data belongs to the support reaction monitoring category. The main beam vertical vibration acceleration monitoring data, the main beam lateral vibration acceleration monitoring data, the main beam longitudinal vibration acceleration monitoring data, the tower top horizontal bidirectional vibration acceleration monitoring data, and the cable vibration acceleration monitoring data belong to the vibration monitoring category. The above data are obtained through monitoring of the corresponding existing monitoring equipment.

[0095] Step S102: pre-process the training samples, and calculate the change rate of each training sample by the difference method with the minimum time interval as the calculation cycle, and construct the change rate matrix of the training samples .

[0096] In some embodiments, the training samples are preprocessed and a multivariate spatiotemporal sequence is constructed based on the following data.

[0097] The constructed data include: main beam vertical displacement monitoring data, main beam lateral displacement monitoring data, support displacement monitoring data, beam end longitudinal displacement monitoring data, tower top displacement monitoring data, tower top rotation monitoring data, beam end horizontal rotation monitoring data, beam end vertical rotation monitoring data, main beam key section strain monitoring data, tower key section strain monitoring data, cable force monitoring data, support reaction monitoring data, main beam vertical vibration acceleration monitoring data, main beam lateral vibration acceleration monitoring data, main beam longitudinal vibration acceleration monitoring data, tower top horizontal bidirectional vibration acceleration monitoring data, cable vibration acceleration monitoring data, which can be obtained. The training sample data set of variables .

[0098] in, and are the number of variables and The sequence length of variables; the training sample data set is converted into a training sample data matrix, where each row represents a sample and each column represents a variable.

[0099] In this embodiment, multiple training samples are preprocessed, and the SPSS platform is used to calculate the change rate of each training sample by the difference method with the minimum time interval as the calculation cycle to construct the change rate matrix of the training samples. Other methods may also be used to preprocess the training samples, which is not specifically limited in this embodiment.

[0100] The SPSS platform of this embodiment is the full name of SPSS, which stands for Statistical Package for the Social Sciences. The SPSS platform provides text analysis, a large number of machine learning algorithms, data analysis models, advanced statistical analysis functions, etc. The SPSS platform supports functions such as data management, description, visualization, statistical analysis, and predictive modeling.

[0101] Step S103: Define a variable in the training sample change rate matrix is the dependent variable, denoted by , and the other variables are Independent variable, among which , the rate of change matrix of the training samples Perform multivariate phase space reconstruction and obtain different characteristic states based on the projection of chaotic attractors on each dimension of the reconstructed phase space. .

[0102] In this embodiment, the operational risk situation of the bridge structure is a nonlinear chaotic system. When judging the chaotic characteristics of the variable sequence, the Lyapunov index can be used as a basis. When the Lyapunov index is less than 0, the sequence changes stably. When the Lyapunov index is equal to 0, the sequence changes are at the stable boundary. When the Lyapunov index is greater than 0, the sequence changes are unstable.

[0103] After determining the chaotic characteristics of the variable sequence, the phase space of multiple variables is further reconstructed.

[0104] This embodiment solves the embedding dimension according to the CC algorithm and optimal delay time , and get the embedding dimension and optimal delay time After that, The variables are reconstructed into the new phase space respectively, and the matrix data obtained is:

[0105]

[0106] in, Represents the phase space after reconstruction dimensional space, is the number of variables. Each row Represents different feature states, each of which is represented by the dependent variable and independent variables constitute.

[0107] Specifically, the CC algorithm uses correlation integral to simultaneously estimate the optimal delay time and embedded window width , then by Find the embedding dimension .

[0108] In this embodiment, in each phase point after the reconstruction of the multivariable phase space, not only the change rate sequence of the main beam vertical displacement monitoring data is retained, but also the change rate sequences of other monitoring data of the same monitoring category or different monitoring categories are integrated into the system, thereby better restoring the inherent laws of the nonlinear chaotic system of the risk situation composed of all monitoring data in the bridge structure space.

[0109] Step S104: Reconstruct the feature state Input the deep extreme learning machine, perform feature extraction through the sigmoid activation function in the first hidden layer, and extract the quantitative relationship between the variables of each feature state .

[0110] In some embodiments, the feature state obtained after reconstruction Input the deep extreme learning machine, perform feature extraction through the sigmoid activation function in the first hidden layer, and extract the quantitative relationship between the variables of each feature state .

[0111] It should be noted that the deep extreme learning machine is a deep network model formed by cascading multiple extreme learning machine-automatic encoders (ELMAE). Compared with the traditional ELM, the deep extreme learning machine DELM (Deep Extreme Learning Machine) can more comprehensively capture the mapping relationship between data, thereby improving the nonlinear fitting ability and prediction performance, and DELM does not have a reverse fine-tuning process, which greatly shortens the network training time. The network structure of DELM includes an input layer, multiple hidden layers, and an output layer.

[0112] In this embodiment, during the training process of the deep extreme learning machine DELM, the reconstructed matrix data is used as the input data of the input layer of the deep extreme learning machine, and the sigmoid activation function is used in the first hidden layer to extract features, and the quantitative relationship between the change rate sequence of the main beam vertical displacement monitoring data and the change rate sequence of other monitoring data is extracted as follows: .

[0113] Use sigmoid activation function to complete the reconstructed feature space Medium feature status The analysis and extraction method is: .

[0114] in, Represents the feature space after mapping characteristic information; Indicates the monitoring data change rate Reconstructed data; Indicates row transposition; , All are numbers greater than 0.

[0115] Calculate the change rate sequence of the main beam vertical displacement monitoring data With other monitoring data change rate series The quantitative relationship between , and calculate the input-output mapping relationship between the hidden layers, specifically:

[0116]

[0117]

[0118]

[0119] in, are input features, ; is the input feature The corresponding weighting coefficient; is the dimension of the reconstructed phase space.

[0120] Each neuron node in the hidden layer of this embodiment receives the reconstructed data output by the input layer as the input value of the neuron, and uses the sigmoid function to introduce nonlinear factors into the neuron to extract the reconstructed Dimensional Space The feature information of each dimension in the hidden layer is extracted by each neuron node, so that the output of the hidden layer can arbitrarily approximate any nonlinear function.

[0121] Step S105: Based on the above quantitative relationship Get its inverse function ,Will As the activation function of the third hidden layer in the deep extreme learning machine, the dependent variable, i.e. the rate of change of the target monitoring data, is calculated When changes occur, the independent variable, i.e. the change rate of other related monitoring data The degree of change .

[0122] In some embodiments, based on the above quantitative relationship Get its inverse function ,Will As the activation function of the third hidden layer in the deep extreme learning machine, the change rate sequence of the main beam vertical displacement monitoring data is calculated. When changes occur, other monitoring data change rate sequences The degree of change The calculation method is:

[0123]

[0124]

[0125]

[0126] in, Indicates the change rate sequence of other monitoring data exist The value of the moment, Represents the change rate sequence of the main beam vertical displacement monitoring data exist The value of the moment, Indicates the change rate sequence of other monitoring data exist The value of the moment, Represents the change rate sequence of the main beam vertical displacement monitoring data exist The value of the moment.

[0127] Step S106: Use the nonlinear support vector machine as the activation function of the fourth hidden layer to divide the different monitoring categories to which the monitoring data change rate belongs, and obtain a set of different monitoring categories. .

[0128] In some embodiments, an appropriate kernel function is selected and appropriate penalty data , construct and solve the optimization method as follows:

[0129]

[0130] ,

[0131] Find the optimal solution ,

[0132] choose A positive component of , calculated as

[0133] Constructing a decision function

[0134] pass Realize the monitoring categories of displacement, rotation angle, strain, cable force, support reaction force, vibration, etc. of which, Indicates the change rate of each monitoring data Category information.

[0135] Step S107: Based on the classification results of the support vector machine, the fifth hidden layer uses the Kriging interpolation function as an activation function to perform spatial interpolation on the change rate between the points where the change rate of the same category of monitoring data is located, and obtains the risk space .

[0136] In some embodiments, based on the classification results of different monitoring categories such as displacement, rotation angle, strain, cable force, support reaction force, vibration, etc., the fifth hidden layer uses the Kriging interpolation function as the activation function to spatially interpolate the change rate between the points where the change rate of the same category of monitoring data is located, and obtains the displacement change rate risk space , Corner change rate risk space , strain change rate risk space , risk space of cable force change rate , Risk space of support reaction force change rate , Vibration change rate risk space .

[0137] Specifically, the Kriging interpolation function is used as the activation function to calculate the change rate of the same type of monitoring data. The rate of change between Perform spatial interpolation to obtain the change rate space after interpolation of the change rate of the same category of monitoring data, that is, the risk space ,in Indicates the number of variables, Indicates different monitoring categories.

[0138] In this embodiment, a point to be interpolated is defined Property Value Estimated value of Use the change rate of several training sample points within its surrounding influence range The linear combination of is estimated as follows:

[0139]

[0140] In the formula, ,express The degree of change in the monitoring data change rate of each point under the monitoring category.

[0141] In order to make the prediction value unbiased, the unbiased condition is defined as:

[0142]

[0143] The formula for calculating the estimated variance is:

[0144]

[0145] In the formula, For two points The covariance mean between , similarly, and And so on. For this monitoring category The number of all points in .

[0146] Under the unbiased condition constraint, the estimated variance is Reaching the minimum weight coefficients Solve the following system of equations

[0147]

[0148] or,

[0149]

[0150]

[0151] In the formula, is the variogram, obtained from the hypothesis theory or training sample data; and To find the coefficients, solve the above system of equations.

[0152] Use Kriging interpolation function to analyze each monitoring category The interpolation of the monitoring data change rate is calculated in turn to obtain the change rate space of different monitoring categories, that is, the risk space .

[0153] Step S108: In the sixth hidden layer, a linear support vector machine is used as an activation function to transform the space of the same category change rate The degree of change in the monitoring data change rate The scope of the change is divided into different levels, and the risk level is determined based on the extent of the change. The definition of , obtains the risk space of monitoring category change rate with different risk levels .

[0154] In some embodiments, a linear support vector machine is used as the activation function in the sixth hidden layer to classify the space of the same category change rate. The degree of change in the monitoring data change rate The scope of the change is divided into different levels, and the risk level is determined based on the extent of the change. The definition of displacement monitoring data change rate risk space with different risk levels is obtained , Risk space of corner monitoring data change rate , risk space of change rate of strain monitoring data , Risk space of cable force monitoring data change rate , Risk space of change rate of support reaction monitoring data , risk space of vibration monitoring data change rate .

[0155] Specifically, choose the penalty data , the method to construct and solve the convex quadratic programming is:

[0156]

[0157]

[0158]

[0159] Find the optimal solution ,

[0160] calculate

[0161] choose A component of Suitable conditions , the calculation method is:

[0162]

[0163] The separating hyperplane is obtained as:

[0164] The classification decision function is:

[0165] in, Represents the monitoring category The monitoring data change rate in .

[0166] In this embodiment, the change rates of different monitoring categories are divided into four appropriate ranges in turn, and the smaller intervals after division are defined as risk levels in turn. , output the risk space based on the change rate of monitoring categories with different risk levels .

[0167] Step S109: Use the weighted superposition function as the activation function of the last hidden layer in the deep extreme learning machine to superimpose the risk space of each monitoring category , output the risk level space after superposition of all monitoring category spatial risks .

[0168] In some embodiments, a weighted superposition function is used as the activation function of the last hidden layer in the deep extreme learning machine to superimpose the risk space of each monitoring category. , output the risk level space after superposition of all monitoring category spatial risks , thus clarifying that when the change rate of the vertical displacement monitoring data of a main beam in the bridge structure exceeds the limit, the risk change trend of the overall bridge structure is:

[0169]

[0170] in, Represents the risk space of different monitoring categories The superposition weight of , indicating its relative importance to the final result; Indicates different monitoring categories.

[0171] The monitoring method for spatial risk calculation model based on monitoring data change rate provided by the present application integrates the monitoring data change rates of multiple monitoring categories in the same structural space so that the phase points in the reconstructed multidimensional phase space contain more diverse information on the monitoring data change rate, and then uses a deep extreme learning machine to train the spatial risk calculation model based on the monitoring data change rate, and analyzes the quantitative relationship between the monitoring data change rate of a certain point in a certain monitoring category in the structural space and the monitoring data change rate of the same category at different positions in the space and the monitoring data change rate of different categories layer by layer, and based on the change rate in the time series trend analysis, it is clear that when the monitoring data at a certain place in the structural space changes, the change trend of other related monitoring data, so that the change trend of the structural risk can be analyzed in real time from point to surface and from surface to body based on one point. Moreover, it can realize the real-time daily analysis of the overall risk situation of the structure without relying on finite element analysis software, and improve the timeliness of risk analysis from the perspective of daily management; in addition, this can objectively analyze the safety risk situation of the overall structure in a quantitative way, which reduces certain subjective judgments and improves the reliability of analysis compared with the semi-quantitative risk analysis model.

[0172] Based on the above-mentioned monitoring method for the three-dimensional spatial risk calculation model of the structure based on the monitoring data change rate, the embodiment of the present application also provides a three-dimensional spatial risk analysis method of the structure based on the monitoring data change rate, such as Figure 2 As shown, the method includes:

[0173] Step S201: Acquire monitoring data of all points under long-term monitoring in a certain structural space.

[0174] In this embodiment, various sensors and monitoring equipment, such as displacement sensors, stress sensors, vibration sensors, etc., can be installed on the bridge to perform long-term monitoring of various key points in the bridge structure space. The continuity and accuracy of data collection can be ensured to obtain the state information of the structure space at different times and under different conditions. The collected monitoring data is then stored in a database to facilitate subsequent data processing and analysis.

[0175] Step S202: Calculate the monitoring data according to the training method of the spatial risk calculation model based on the monitoring data change rate, and obtain the change of the risk level of the three-dimensional space of the structure when the monitoring data change rate at any point in the space changes.

[0176] In this embodiment, key features are extracted from the monitoring data, such as the rate of change, fluctuation range, trend, etc. of the monitoring data, for use in the subsequent risk calculation model.

[0177] The collected monitoring data is also annotated, including information such as data change rate and risk level, so that the model can learn the correlation between data change and risk level. The annotated data is input into the large language model module and machine learning algorithm for training. By adjusting the parameters and structure of the model, the model can accurately predict the risk level changes in the three-dimensional space of the structure.

[0178] This embodiment uses a trained model to calculate the rate of change of monitoring data at any point in the three-dimensional space of the structure, and determines the change trend of the bridge in the same monitoring category and related monitoring categories based on the results output by the model, and then evaluates the risk level status of the entire three-dimensional space of the structure through quantitative change analysis of the overall structural response of the bridge.

[0179] Step S203: Based on the risk level change of the three-dimensional space of the structure, potential risks and hidden dangers can be predicted and prevented in advance.

[0180] In this embodiment, the potential risk hazards in the structural space are predicted in advance according to the risk level change calculated in step S202. By analyzing the change trend and law of the risk level, the potential structural hazards that may exist can be identified.

[0181] In this embodiment, according to the spatial risk analysis method based on the monitoring data change rate, it can be calculated that when the change rate of the vertical displacement monitoring data of the main beam at any point in the bridge structure space changes, the overall risk level of the bridge structure changes. Therefore, based on the risk level change in the three-dimensional space of the structure, potential risks in the bridge structure can be predicted and prevented in advance.

[0182] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0183] The present application also provides an electronic device, such as Figure 3 As shown, the electronic device includes: a memory 100, a processor 101, and a computer program stored in the memory 100 and executable on the processor 101, and the processor implements the steps of the training method for the structural risk calculation model when executing the program.

[0184] In embodiments of the present invention, electronic devices include, but are not limited to, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the embodiments of the present application described and / or required herein.

[0185] In the embodiment of the present application, the processor 101 can be implemented by using at least one of an application-specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a processor, a controller, a microcontroller, a microprocessor, and an electronic unit designed to perform the functions described herein. In some cases, such an implementation can be implemented in a controller. For software implementation, implementations such as processes or functions can be implemented with separate software modules that allow execution of at least one function or operation. The software code can be implemented by a software application (or program) written in any appropriate programming language, and the software code can be stored in a memory and executed by a controller.

[0186] The memory 100 may include one or more computer-readable storage media, which may be non-transitory. The memory 100 may also include a high-speed random access memory and a non-volatile memory, such as one or more disk storage devices and flash memory storage devices. The resources stored in the memory 100 may also include an operating system and data, etc., and the storage method may be temporary storage or permanent storage. Among them, the operating system may include Windows, Unix, Linux, etc. The data may include but is not limited to the relevant data involved in the above method, etc.

[0187] The electronic device may further include a display module 102 for displaying the execution process and monitoring results of the method for monitoring the spatial risk calculation model based on the data change rate.

[0188] Those skilled in the art will understand that Figure 3 The structure shown in the figure does not constitute a limitation on the electronic device, and may include more or fewer components than those shown in the figure.

[0189] The present application also provides a storage medium, in which a program product capable of implementing a training method for a structural risk calculation model is stored.

[0190] The specific method includes: establishing a training sample data set containing multiple training samples; preprocessing the training samples, and calculating the change rate of each training sample based on the difference method with the minimum time interval as the calculation cycle, and constructing the change rate matrix of the training samples; reconstructing the change rate matrix of the training samples in a multivariate phase space, and obtaining different characteristic states based on the projection of the attractor on each dimension; inputting each characteristic state into a deep extreme learning machine for feature extraction, and obtaining a quantitative relationship between the change rate of the target monitoring data and the change rate of other monitoring data; calculating the inverse function of the above quantitative relationship and using it as the activation function of the next hidden layer in the deep extreme learning machine, and calculating the degree of change of the change rate of other related monitoring data when the change rate of the target monitoring data changes; using a nonlinear support vector machine as an activation function to divide the monitoring categories of each monitoring data; using Kriging interpolation activation function to calculate the risk change between the points where the change rate of each category of monitoring data is located, and completing the risk space calculation of a certain monitoring category; using a linear support vector machine as an activation function to divide the degree of change of the change rate of the same category of monitoring data, and define the risk level; using a weighted superposition function as an activation function, superimposing the risk space of each monitoring category after interpolation, and outputting the risk space after the superposition of all monitoring category risks. The embodiment of the present invention can analyze the degree of change of the change rate of other related monitoring points in the three-dimensional space of the structure where the monitoring data is located based on the change rate of a certain monitoring data in the space, thereby clarifying the risk situation of the three-dimensional space of the structure where the point is located based on the risk change of the point.

[0191] The storage medium can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can be, for example, but not limited to, a system, device or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0192] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for training a structural risk calculation model, characterized in that: Methods include: Step S101: establishing a training sample data set including a plurality of training samples, each training sample having a coordinate attribute; Step S102: pre-processing the training samples, and calculating the change rate of each training sample by difference method with a preset time interval as a calculation cycle, and constructing a change rate matrix of the training samples; Step S103: define a variable in the training sample change rate matrix as a dependent variable and the other variables as independent variables; reconstruct the training sample change rate matrix in a multivariate phase space, and obtain different characteristic states based on the projection of the chaotic attractor on each dimension of the reconstructed phase space; Step S104: input the reconstructed feature state into the deep extreme learning machine, perform feature extraction through the sigmoid activation function in the first hidden layer, and extract the quantitative relationship between the variables constituting each feature state; Step S105: obtaining an inverse function based on the quantitative relationship, using the inverse function as the activation function of the third hidden layer in the deep extreme learning machine, and calculating the degree of change of the change rates of other related monitoring data when the change rate of the target monitoring data changes; Step S106: using a nonlinear support vector machine as an activation function of the fourth hidden layer to divide different monitoring categories to which the monitoring data change rate belongs, to obtain a set of different monitoring categories; Step S107: Based on the classification result of the support vector machine, the Kriging interpolation function is used as the activation function in the fifth hidden layer to perform spatial interpolation on the degree of change of the change rate between the points where the change rate of the same category of monitoring data is located, and the change rate space of the same monitoring category is obtained, which is defined as the monitoring category risk space; Step S108: In the sixth hidden layer, a linear support vector machine is used as an activation function to divide the range of the change degree of the monitoring data change rate in the same category change rate space, and the risk level is defined based on the range of the change degree to obtain a monitoring category risk space with different risk levels; Step S109: Use the weighted superposition function as the activation function of the last hidden layer in the deep extreme learning machine to superimpose the risk level space of each monitoring category, and output the risk level space after the spatial risks of all monitoring categories are superimposed.

2. The training method according to claim 1, characterized in that: The training samples are preprocessed, and the trend change rate of each training sample is calculated with a weekly analysis cycle to construct a change rate matrix of the training samples. ,include: The training samples are preprocessed to obtain a training sample data set containing multiple variables: ； and are the number of variables and The sequence length of variables; the training sample data set is converted into a training sample data matrix, in which each row represents a sample and each column represents a variable; According to the difference method, the change rate of each training sample is calculated with the minimum time interval as the calculation period, and the change rate matrix of the training sample is obtained. , specifically expressed as follows: in, ， ， …… 。 3. The training method according to claim 1, characterized in that: In step S103, the bridge structure is defined as a nonlinear chaotic system, and the chaotic characteristics of the variable sequence are determined based on the Lyapunov exponent; Define a variable in the training sample change rate matrix is the dependent variable, denoted by , and the other variables are Independent variable; in, , the rate of change matrix of the training samples Perform multivariate phase space reconstruction and obtain different characteristic states based on the projection of chaotic attractors on each dimension of the reconstructed phase space. , Represents the dimension of the reconstructed phase space, including: Reconstruct the multivariate phase space and solve the embedding dimension according to the CC algorithm and optimal delay time , and get the embedding dimension and optimal delay time After that, The variables are reconstructed into the new multivariable phase space, and the matrix data obtained is: in, Represents the phase space after reconstruction dimensional space, is the number of variables; each row Represents different feature states, each of which is represented by the dependent variable and independent variables constitute.

4. The training method according to claim 1, characterized in that: In step S104, the network structure of the extreme learning machine autoencoder of the deep extreme learning machine includes an input layer, multiple hidden layers, and an output layer; The reconstructed matrix data is used as the input data of the deep extreme learning machine input layer, and the sigmoid activation function is used in the first hidden layer to extract the quantitative relationship between the dependent variable and the independent variable. ,include: Use sigmoid activation function to complete the reconstructed feature space Medium feature status Analytical extraction: in, Represents the feature space after mapping characteristic information; Indicates the monitoring data change rate Reconstructed data; Indicates row transposition; , All are numbers greater than 0; Calculate the dependent variable With independent variable The quantitative relationship between : in, Represents the input-output mapping relationship between hidden layers; are input features, ; is the input feature The corresponding weighting coefficient; is the dimension of the reconstructed phase space.

5. The training method according to claim 1, characterized in that: In step S105, the inverse function is defined as: ; Independent Variable The degree of change for , ; in, Indicates the independent variable exist The value of the moment, Dependent variable exist The value of the moment, Indicates the independent variable exist The value of the moment, Dependent variable exist The value of the moment.

6. The training method according to claim 1, characterized in that: In step S106, the nonlinear support vector machine is used as the activation function of the fourth hidden layer to divide the different monitoring categories to which the monitoring data change rate belongs, and a set of different monitoring categories is obtained. , specifically: Select kernel function and penalty data Construct and solve the formula: , Find the optimal solution , choose A positive component of , the calculation method is: Constructing a decision function pass Implement different monitoring categories in turn Division of in, Indicates the independent variable Category information, that is, category information of the change rate of each monitoring data.

7. The training method according to claim 1, characterized in that: In step S107, the Kriging interpolation function is used as the activation function to calculate the change rate of the same type of monitoring data. The rate of change between Perform spatial interpolation to obtain the change rate space after interpolation of the change rate of the same category of monitoring data, that is, the risk space ; in Indicates the number of variables, Indicates different monitoring categories; Define the points to be interpolated Property Value Estimated value of , In the formula, ,express The degree of change in the monitoring data change rate of each point under the monitoring category; The unbiased condition of the predicted value is defined as: The calculation formula for defining the estimated variance is: In the formula, For two points The mean covariance between For monitoring category The number of all points in ; Under the unbiased condition constraint, the estimated variance Reaching the minimum weight coefficients ; Use Kriging interpolation function to analyze each monitoring category The interpolation of the monitoring data change rate is calculated in turn to obtain the change rate space of different monitoring categories .

8. The training method according to claim 1, characterized in that: In step S108, a linear support vector machine is used as an activation function to sequentially calculate the change rate space of the same category. The degree of change in the monitoring data change rate The scope of the change is divided into different levels, and the risk level is determined based on the extent of the change. The definition of , obtains the risk space of monitoring category change rate with different risk levels , and later: Select Penalized Data , construct and solve the convex quadratic programming formula as: Find the optimal solution , calculate choose A component of condition ,calculate, The separating hyperplane is obtained as: The classification decision function is: in, Represents the monitoring category The monitoring data change rate in The change rates of different monitoring categories are divided into four appropriate ranges in turn, and the smaller intervals after division are defined as risk levels in turn. , output the risk space of monitoring category change rate with different risk levels .

9. The training method according to claim 1, characterized in that: In step S109, the weighted superposition function is used as the activation function of the last hidden layer in the deep extreme learning machine to superimpose the risk space of each monitoring category. , output the risk level space after superposition of all monitoring category spatial risks : in, Represents the risk space of different monitoring categories The superposition weight of , indicating its relative importance to the final result; Indicates different monitoring categories, Represents the total number of all monitoring categories.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the method for training a structural risk calculation model as described in any one of claims 1 to 9 are implemented.

11. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for training a structural risk calculation model as claimed in any one of claims 1 to 9 are implemented.

Images