A structural reliability assessment prediction method and system based on long-term monitoring data

Abstract

The present application relates to the technical field of civil engineering structure data monitoring, and more particularly to a structure reliability evaluation and prediction method and system based on long-term monitoring data. The technical solution comprises the following steps: data preprocessing, long-term monitoring data is preprocessed to improve data quality; data distribution adaptive modeling, an adaptive data distribution model is constructed to adapt to complex data characteristics; reliability index calculation and evaluation, structure reliability index is calculated based on the data distribution model and evaluation is completed; reliability dynamic prediction, dynamic prediction of reliability is performed according to the time characteristics of the data. The present application can effectively improve the quality of long-term monitoring data, accurately extract key features, lay a solid foundation for subsequent complex data modeling based on a mixed Copula function model, dynamically evaluate and predict structure reliability combined with time-varying characteristics, and improve reliability by integrating multi-source data.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of civil engineering structure data monitoring, and in particular to a structure reliability assessment and prediction method and system based on long-term monitoring data. BACKGROUND

[0002] In the aspect of structure reliability assessment and prediction of civil engineering long-term monitoring data, the traditional probability statistical assessment and prediction method has a relatively prominent problem: too strong dependence on data quantity and distribution.

[0003] The traditional probability statistical method usually needs a large amount of monitoring data to accurately estimate the probability distribution and statistical characteristics of the structure parameters. In actual civil engineering, it is often difficult to obtain a large amount of long-term and high-quality monitoring data. On the one hand, the cost of monitoring equipment arrangement and maintenance is high, and it is difficult to set enough monitoring points at each key part of the structure and obtain data stably for a long time; on the other hand, the service time of the structure is limited, and in some new structures or structures in special environments, it may not be possible to accumulate enough data in a short time to meet the requirements of the traditional method. This leads to a large deviation in the assessment and prediction results based on the traditional probability statistical method in the case of insufficient data, which cannot accurately reflect the true reliability of the structure.

[0004] The traditional method generally assumes that the monitoring data follows a certain specific probability distribution, such as normal distribution, lognormal distribution, etc. However, the monitoring data of actual civil engineering structures often has complex distribution characteristics, may be influenced by multiple factors, and does not necessarily conform to these assumed ideal distributions. For example, influenced by environmental factors (such as climate change, seismic activity), non-uniformity of structural materials, and nonlinearity of structural damage development, monitoring data may exhibit complex distribution characteristics such as asymmetry, multi-peak, or thick tail. If the data is forcibly fitted to an inappropriate distribution model, it will lead to errors in the assessment and prediction of structural reliability, and may even draw incorrect conclusions, thereby affecting the safety assessment and maintenance decision of the structure.

[0005] Therefore, the present application proposes a structure reliability assessment and prediction method and system based on long-term monitoring data. SUMMARY

[0006] The purpose of the present application is to solve the problem of too strong dependence on data quantity and distribution in the aspect of structure reliability assessment and prediction of civil engineering long-term monitoring data in the background art, and to propose a structure reliability assessment and prediction method and system based on long-term monitoring data.

[0007] The technical scheme of the present application: a structure reliability assessment and prediction method based on long-term monitoring data, comprising the following steps:

[0008] Data preprocessing, first pre-process the long-term monitoring data to improve data quality;

[0009] Adaptive modeling of data distribution, construct an adaptive data distribution model to adapt to complex data characteristics;

[0010] Reliability index calculation and evaluation, calculate the structural reliability index based on the data distribution model and complete the evaluation;

[0011] Reliability dynamic prediction, according to the time characteristics of the data to predict the reliability dynamically.

[0012] Optionally, the data preprocessing specifically includes:

[0013] Data cleaning: adopt the isolation forest algorithm to detect outliers in the monitoring data, set the monitoring data set as , isolation forest constructs multiple isolated trees to score the data points, and the points with scores exceeding the preset threshold are determined as outliers. For outliers, use the weighted average of adjacent normal data points to replace, and the weight is calculated according to the distance reciprocal;

[0014] Data completion: for missing data, adopt the method based on gradient boosting tree regression to complete, train the gradient boosting tree model, take the known data characteristics as input, and the missing data as output, predict and complete the missing value;

[0015] Data normalization: use Z-score normalization method to transform the data into standard normal distribution with mean 0 and standard deviation 1. For data point , the normalized value , wherein is the data mean, is the standard deviation.

[0016] Optionally, the data distribution adaptive modeling stage adopts a hybrid Copula function model:

[0017] Fit the edge distribution of each monitoring data variable, determine the parameters of the edge distribution by maximum likelihood estimation method;

[0018] Select several candidate Copula functions from Gaussian Copula function, t-Copula function and Clayton Copula function, and evaluate and select the candidate Copula functions by Akaike information criterion and Bayesian information criterion;

[0019] Construct a hybrid Copula function model, combine the selected multiple Copula functions according to certain weights, and determine the weights by genetic algorithm optimization.

[0020] Optionally, the reliability index calculation and evaluation stage adopts an importance sampling method based on subset simulation:

[0021] Define the limit state function of the structure Wherein is a random variable affecting the reliability of the structure;

[0022] Set the initial failure probability level, and gradually divide the failure domain into multiple intermediate subsets by subset simulation;

[0023] In each subset, the importance sampling method is used for sampling, the sampling samples are generated according to the mixed Copula function model, and the limit state function value corresponding to the samples is calculated;

[0024] The number of samples falling into each subset is counted, the failure probability of the structure is gradually calculated, and the reliability index is calculated according to the formula Wherein is the inverse cumulative distribution function of the standard normal distribution.

[0025] Optionally, the reliability dynamic prediction stage adopts a gated recurrent unit neural network combined with a Bayesian updating mechanism:

[0026] A GRU neural network model is constructed, the historical monitoring data and the corresponding reliability index are taken as input, and the future change trend of the reliability index is predicted. The update gate , the reset gate and the candidate hidden state of the GRU model are calculated according to the following formulas:

[0027]

[0028]

[0029]

[0030] Wherein, is the current input, is the hidden state at the last time, is the weight matrix, is the bias vector, is the sigmoid function, is the hyperbolic tangent function, is the element-wise multiplication.

[0031] When new monitoring data is obtained, the prediction result of the GRU model is corrected by using the Bayesian updating mechanism, and the prior probability distribution is updated to the posterior probability distribution according to the new data.

[0032] Optionally, after the data preprocessing stage, a data feature extraction step is added:

[0033] The principal component analysis method is used to extract features from the preprocessed data, reduce the dimension of the data, calculate the covariance matrix of the data, solve the eigenvalues and eigenvectors, and select the first characteristic vectors with larger eigenvalues to form a projection matrix.

[0034] The original data is projected onto the projection matrix to obtain the reduced dimension feature data.

[0035] Optionally, in the reliability index calculation and evaluation, the time-varying characteristics of the structure are considered:

[0036] A time-dependent random process is introduced to describe the changes of structure parameters, material properties and loads over time. The random process uses Wiener process or Gamma process to simulate the fatigue damage accumulation of the structure.

[0037] The time variable is added to the limit state function to dynamically evaluate the reliability of the structure at different times. By continuously updating the random process model of the structure parameters, the evolution of the structure reliability over time is reflected in real time.

[0038] Optionally, in the reliability dynamic prediction stage, a multi-source data fusion technology is combined:

[0039] In addition to long-term monitoring data, multi-source data of the structure is also fused, including design data, construction records and environmental data, such as temperature, humidity and wind speed.

[0040] The evidence theory or fuzzy mathematics method is used to fuse and process the multi-source data.

[0041] In the second aspect, the application provides a structure reliability evaluation and prediction system based on long-term monitoring data, comprising:

[0042] A data preprocessing module is used to preprocess the long-term monitoring data to improve the data quality.

[0043] A data distribution adaptive modeling module is used to build an adaptive data distribution model to adapt to complex data characteristics.

[0044] A reliability index calculation and evaluation module is used to calculate and evaluate the structure reliability index based on the data distribution model.

[0045] A reliability dynamic prediction module is used to dynamically predict the reliability according to the time characteristics of the data.

[0046] Optionally, it also includes: a user interaction module that provides a user-friendly interface for users to view data, set parameters, select models, and display results, and supports the visualization of monitoring data change trends and reliability index change curves in chart form.

[0047] Compared with the prior art, this application includes at least one of the following beneficial technical effects:

[0048] By using the Isolation Forest algorithm to clean outliers, Gradient Boosting Tree Regression to fill in missing values, and Z-score normalization to process the data, the data quality is significantly improved. Subsequent principal component analysis extracts features, reduces dimensionality, and improves computational efficiency.

[0049] Hybrid Copula function models combine multiple methods to accurately characterize data correlation and distribution features, enhancing model adaptability and accuracy.

[0050] The important sampling method based on subset simulation accurately calculates the reliability index, and considers time-varying characteristics to achieve dynamic evaluation. The GRU neural network, combined with the Bayesian update mechanism and multi-source data fusion technology, improves the accuracy, reliability and comprehensiveness of the prediction.

[0051] This invention not only effectively improves the quality of long-term monitoring data, but also accurately extracts key features, laying a solid foundation for subsequent complex data modeling based on the hybrid Copula function model. It also dynamically evaluates and predicts structural reliability by combining time-varying characteristics, and integrates multi-source data to improve reliability. Attached Figure Description

[0052] Figure 1 This is a flowchart of a structural reliability assessment and prediction method based on long-term monitoring data. Detailed Implementation

[0053] The technical solution of the present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0054] Example 1

[0055] like Figure 1 As shown, the present invention proposes a structural reliability assessment and prediction method based on long-term monitoring data, which includes data preprocessing, adaptive modeling of data distribution, reliability index calculation and evaluation, and dynamic reliability prediction. Each step is described in detail below.

[0056] I. Data Preprocessing: Long-term monitoring data is preprocessed to improve data quality; this includes:

[0057] Data cleaning: The Isolation Forest algorithm is used to detect outliers in the monitoring data. Let the monitoring dataset be... , Isolation Forest scores the data points by building multiple isolated trees, and the points with scores exceeding a pre-defined threshold are determined as outliers. For outliers, the weighted average of the neighboring normal data points is used to replace the outliers, and the weights are calculated according to the reciprocal of the distance;

[0058] Data completion: For missing data, a gradient boosting tree regression-based method is used for completion. A gradient boosting tree model is trained, with known data features as input and missing data as output, to predict and complete the missing values.

[0059] Data normalization: Z-score normalization method is used to transform the data into a standard normal distribution with a mean of 0 and a standard deviation of 1. For data points , the normalized value , where is the mean of the data, is the standard deviation.

[0060] After the data preprocessing stage, a data feature extraction step is added:

[0061] Principal component analysis is used to extract features from the preprocessed data, reducing the data dimension. The covariance matrix of the data is calculated, and its eigenvalues and eigenvectors are solved. The first eigenvectors with larger eigenvalues are selected to form a projection matrix. The original data is projected onto the projection matrix to obtain the reduced dimension feature data. Specifically, it includes:

[0062] Principal component analysis is used to extract features from the preprocessed data, reducing the data dimension. The specific steps are as follows:

[0063] Calculate the covariance matrix: Let the preprocessed data set be , where is the number of samples, is the number of features, and the covariance matrix is calculated as , where is the mean matrix of the data set , and each element of ;

[0064] Solve the eigenvalues and eigenvectors: Perform eigenvalue decomposition on the covariance matrix , i.e., solve the equation , to obtain eigenvalues and corresponding eigenvectors , where is an order identity matrix;

[0065] Determine the number of principal components The following methods are used to determine: Value:

[0066] The scree plot method involves plotting a line graph of eigenvalues ​​as a function of eigenvector indices. Observe the inflection points in the decreasing trend of eigenvalues ​​on the line graph. After the inflection point, the eigenvalue changes become more gradual. Select the indices corresponding to the inflection points as the eigenvector indices. ;

[0067] Constructing the projection matrix: Select the first eigenvalues ​​with larger eigenvalues. The eigenvectors are arranged in columns to form a projection matrix. Projection matrix The dimension is ;

[0068] Data projection: Projecting the original preprocessed dataset Projected onto the projection matrix The above yields the dimensionality-reduced feature data. Dimensionally reduced data The dimension is This reduces data redundancy and improves the efficiency of subsequent modeling and calculation.

[0069] This embodiment significantly improves data quality: data preprocessing uses the Isolation Forest algorithm to clean outliers and replaces them with weighted averages to ensure data accuracy; gradient boosting tree regression is used to fill in missing values, reducing information loss caused by missing data; Z-score normalization standardizes the data, facilitating subsequent analysis and effectively improving data quality, providing a solid foundation for subsequent steps.

[0070] Principal component analysis (PCA) extracts features after data preprocessing, reducing data dimensionality, removing redundant information, retaining key features, improving computational efficiency, reducing data processing burden, and better revealing the inherent patterns in the data.

[0071] II. Adaptive Data Distribution Modeling: Constructing an adaptive data distribution model to suit complex data characteristics. The adaptive data distribution modeling stage employs a hybrid Copula function model.

[0072] First, the marginal distributions of each monitoring data variable are fitted, and the parameters of the marginal distributions are determined by the maximum likelihood estimation method. For different types of data, the distribution forms such as normal distribution, log-normal distribution, and Weibull distribution can be selected.

[0073] Several candidate Copula functions were selected from Gaussian Copula, t-Copula, and Clayton Copula functions. The Akaike Information Criterion and Bayesian Information Criterion were used to evaluate and select the candidate Copula functions.

[0074] A mixed Copula function model is constructed, and the screened multiple Copula functions are combined according to certain weights, and the weights are determined by genetic algorithm optimization.

[0075] The specific steps are as follows:

[0076] Initialize the population: set the number of mixed Copula functions to , the weight corresponding to each mixed Copula function is , and it satisfies

[0077]

[0078] Randomly generate a weight vector satisfying the above conditions to form the initial population, wherein is the population size, ;

[0079] Define the fitness function: select the negative log-likelihood function as the fitness function to measure the fitting degree of the mixed Copula function model to the data, and set the monitoring data as , the mixed Copula function is , wherein is the edge distribution function value vector, is the th candidate Copula function, is the parameter vector thereof, and the fitness function is defined as:

[0080]

[0081] , wherein is the probability density function of the th candidate Copula function, is the edge distribution function value vector corresponding to the th data point;

[0082] Selection operation: adopt roulette wheel selection method to select individuals from the current population into the next generation, and calculate the selection probability of each individual :

[0083]

[0084] Then, according to the selection probability, randomly select until individuals are selected to form a new population;

[0085] Cross operation: cross the selected population to generate new individuals, randomly select two individuals and , set the crossover probability , if the random number is less than , the crossover operation is performed, and a linear combination crossover method is used to generate two new individuals and :

[0086] wherein is a number randomly generated in the interval (0, 1), and the newly generated individual is normalized to make its weight satisfy:

[0087]

[0088] Mutation operation: mutation operation is performed on the population after crossover to increase the diversity of the population, and the mutation probability is set. For each individual , if the random number is less than , one of the weights is mutated, a new weight value is randomly generated in the interval using the uniform mutation method, and then the individual is normalized;

[0089] Termination condition judgment: repeat the selection, crossover and mutation operations until the termination condition is met. The termination condition is that the value of the fitness function changes less than a preset threshold in a certain number of consecutive generations , and the weight vector corresponding to the final optimal individual is the optimal weight of the mixed Copula function.

[0090] In this embodiment, the data distribution adaptive modeling adopts a mixed Copula function model, the edge distribution parameters are fitted by maximum likelihood estimation, the Akaike information criterion and the Bayesian information criterion are used to screen the Copula function, and the genetic algorithm is used to optimize the weight. The complex correlation and distribution characteristics between the monitoring data can be more accurately described, and the model adaptability and accuracy can be improved.

[0091] III. Reliability index calculation and evaluation, structural reliability index is calculated based on data distribution model and evaluation is completed; in the reliability index calculation and evaluation stage, an important sampling method based on subset simulation is used:

[0092] The limit state function of the structure is defined , wherein is a random variable affecting the reliability of the structure;

[0093] An initial failure probability level is set, and the failure domain is gradually divided into multiple intermediate subsets by subset simulation;

[0094] Importance sampling method is used in each subset, and the sampling sample is generated according to the mixed Copula function model, and the limit state function value corresponding to the sample is calculated;

[0095] The number of samples falling into each subset is counted, the failure probability of the structure is calculated step by step, and the formula is The reliability index is calculated, wherein is the inverse cumulative distribution function of the standard normal distribution.

[0096] Based on the importance sampling method of subset simulation, the limit state function is defined, the failure domain is divided by subset simulation, and the mixed Copula function model is combined with sampling calculation, so that the failure probability and reliability index of the structure can be accurately calculated, and the random characteristics and uncertainty of the structure are fully considered.

[0097] In the reliability index calculation and evaluation, the time-varying characteristics of the structure are considered:

[0098] The time-dependent random process is introduced to describe the change of structure parameters, material performance and load with time, and the Wiener process or Gamma process is used to simulate the fatigue damage accumulation of the structure;

[0099] The time variable is added to the limit state function, and the dynamic evaluation of the reliability of the structure at different times is carried out, and the evolution of the reliability of the structure with time is reflected in real time by continuously updating the random process model of the structure parameters. The time-dependent random process is introduced to describe the change of structure parameters, and the time variable is added to the limit state function, so that the reliability of the structure at different times can be dynamically evaluated, and the evolution of the reliability of the structure with time can be reflected in real time. Provide basis for structure life cycle monitoring and maintenance.

[0100] IV. Dynamic prediction of reliability, dynamic prediction of reliability according to time characteristics of data, dynamic prediction of reliability stage adopts gated recurrent unit neural network combined with bayesian update mechanism:

[0101] The GRU neural network model is constructed, the historical monitoring data and the corresponding reliability index are taken as input, and the future change trend of the reliability index is predicted, and the update gate , reset gate and candidate hidden state of GRU model are calculated as follows:

[0102]

[0103]

[0104]

[0105] wherein, is the input at the current time, is the hidden state at the previous time, is the weight matrix, is the bias vector, is the sigmoid function, is the hyperbolic tangent function, is the element-wise multiplication.

[0106] When new monitoring data is obtained, the prediction result of the GRU model is corrected by using a Bayesian updating mechanism, and the prior probability distribution is updated to the posterior probability distribution according to the new data.

[0107] In the reliability dynamic prediction stage, a multi-source data fusion technology is combined:

[0108] In addition to long-term monitoring data, multi-source data of the structure is also fused, and the multi-source data includes design data, construction records, and environmental data, and the environmental data includes temperature, humidity, and wind speed.

[0109] The multi-source data is fused and processed by using an evidence theory or a fuzzy mathematics method.

[0110] The reliability dynamic prediction adopts a gated recurrent unit neural network combined with a Bayesian updating mechanism. The GRU neural network predicts future trends according to historical data and reliability indexes, the Bayesian updating mechanism corrects the prediction result according to new data, and the prediction accuracy is improved. In combination with the multi-source data fusion technology, multi-source data such as design data, construction records, and environmental data is fused, and the evidence theory or the fuzzy mathematics method is used for fusion processing, so as to further improve the reliability and comprehensiveness of the prediction.

[0111] On the other hand, the application provides a structure reliability evaluation and prediction system based on long-term monitoring data, comprising:

[0112] A data preprocessing module is configured to preprocess the long-term monitoring data to improve data quality.

[0113] A data distribution adaptive modeling module is configured to construct an adaptive data distribution model to adapt to complex data characteristics.

[0114] A reliability index calculation and evaluation module is configured to calculate and evaluate the structure reliability index based on the data distribution model.

[0115] A reliability dynamic prediction module is configured to dynamically predict the reliability according to the time characteristics of the data.

[0116] Embodiment 2

[0117] The embodiment further includes the following on the basis of the embodiment 1:

[0118] The generalization ability of the model is improved by introducing transfer learning in the reliability dynamic prediction step, which includes:

[0119] Source domain and target domain selection:

[0120] Determine the source domain and the target domain. Assume that the target structure is a cable-stayed bridge. The source domain is selected as multiple built cable-stayed bridges of the same type with long-term monitoring data. Collect monitoring data of the source domain cable-stayed bridges, including cable force, main girder deflection, tower deflection data, and early monitoring data of the target cable-stayed bridge.

[0121] Preprocess the data of the source domain and the target domain. Use the same method as the original data preprocessing module, use the Isolation Forest algorithm to clean outliers, use the Gradient Boosting Tree regression to fill in missing values, and use the Z-score normalization method to convert the data to a standard normal distribution with a mean of 0 and a standard deviation of 1.

[0122] Transfer learning model construction:

[0123] Use a transfer learning strategy based on pre-training and fine-tuning. First, pre-train the GRU neural network on the source domain data. The structure of the GRU neural network includes an input layer, a hidden layer, and an output layer. Assume that the number of input layer nodes is determined according to the number of input data features, which is , the number of hidden layer nodes is set to , and the number of output layer nodes is 1 (output reliability index prediction value). Use the Adam optimizer with a learning rate of 0.001 and a training round of T1=100.

[0124] Migrate the first k layers of the pre-trained GRU model to the target domain model, k=2, and then fine-tune on the target domain data. During fine-tuning, fix the pre-trained first k parameters and adjust the parameters of the subsequent layers (such as the output layer and the layer close to the output layer). Set the fine-tuning learning rate to 0.0001 and the training round to T2=50. In this way, the knowledge in the source domain data is used to accelerate the training of the target domain model and improve the generalization ability of the model in the target structure reliability prediction.

[0125] Hybrid prediction method combining deep learning and physical model

[0126] Physical model construction:

[0127] According to the mechanical principles and material properties of the structure, establish the physical model of the structure. Take a high-rise building structure as an example, use a truss model, and based on the displacement method and force method theory in structural mechanics, establish the mechanical equilibrium equation of the structure under horizontal load (such as wind load, earthquake action) and vertical load. Assume that the mass of each layer of the structure is , the interlayer stiffness is , and the horizontal displacement is According to the D'Alembert principle and Hooke's law, the motion equation is established:

[0128]

[0129] wherein is a damping coefficient, is an external force acting on the first layer, and are acceleration and velocity, respectively.

[0130] Combined with the constitutive relation of the material, the stress-strain relation of the concrete adopts the curve model recommended by the specification, and the constitutive relation of the steel bar adopts the bilinear hardening model, so as to determine the mechanical response of the structure under different stress states.

[0131] Fusion of deep learning model and physical model:

[0132] The GRU neural network is combined with the physical model. The stress and strain physical quantities of the key parts of the structure calculated by the physical model are taken as the inputs of the GRU model together with the displacement monitoring value and stress monitoring value . The dimension of the input data of the GRU model is , so the calculation results of the physical model and the monitoring data are spliced to form the input vector .

[0133] Meanwhile, the prediction results of the deep learning model are used to correct the physical model. The load change of the structure in the future period of time is predicted by the GRU model, and the wind load at the future moment predicted is taken as the input load of the physical model, so as to recalculate the mechanical response of the structure, thereby more accurately predicting the reliability of the structure.

[0134] Embodiment 3

[0135] Based on the structure reliability evaluation and prediction system based on long-term monitoring data in Embodiment 1, the system further comprises a user interaction module, provides a friendly interface, and is used for data viewing, parameter setting, model selection, result display by a user, and supports visual display of monitoring data change trend and reliability index change curve in a chart form. The system realizes accurate evaluation and prediction of the reliability of the civil engineering structure, and provides support for structure safety monitoring and maintenance decision.

[0136] Embodiment 4

[0137] The embodiment provides a structure reliability evaluation and prediction system based on long-term monitoring data, comprising:

[0138] Data acquisition device for collecting long-term monitoring data of civil engineering structures from various sensors and monitoring devices, covering stress, strain, displacement, vibration physical quantity data, and supporting multiple data transmission protocols to transmit data to the next link stably and accurately.

[0139] Data preprocessing module, including data cleaning unit, using isolated forest algorithm and statistical threshold-based method to identify and mark abnormal data points, and correcting by interpolation method and mean filling method; data completion unit, using machine learning algorithms such as gradient boosting tree regression to complete missing data; data normalization unit, using Z-score normalization and minimum-maximum normalization to unify the scale of data.

[0140] Data distribution adaptive modeling module, with edge distribution fitting unit, providing normal distribution, lognormal distribution, Weibull distribution and other distribution function selection for each monitoring data variable, and determining parameters by maximum likelihood estimation method; Copula function screening unit, screening appropriate function from Gaussian Copula, t-Copula, Clayton Copula and other Copula functions according to AIC (Akaike Information Criterion) and BIC (Bayesian Information Criterion); mixed Copula function construction unit, combining the selected Copula functions according to the weight determined by genetic algorithm optimization to construct a mixed Copula function model.

[0141] Reliability index calculation and evaluation module, with limit state function definition unit, supporting users to define or select preset limit state functions according to the mechanical properties and failure modes of civil engineering structures; important sampling calculation unit, using subset simulation-based important sampling method to generate sampling samples according to the mixed Copula function model and substitute them into the limit state function to calculate failure probability and reliability index; reliability evaluation unit, evaluating structure reliability according to reliability index and preset structure health state classification standard and outputting evaluation results.

[0142] Reliability dynamic prediction module, including time series model construction unit, using gated recurrent unit (GRU) neural network and autoregressive integrated moving average model (ARIMA) to construct prediction model with historical monitoring data and corresponding reliability index as input; prediction result updating unit, using Bayesian updating mechanism combined with multi-source data fusion technology to update model and output future structure reliability prediction value when new monitoring data is obtained.

[0143] A user interaction module provides a friendly interface, facilitates the user to perform data viewing, parameter setting, model selection, result display and the like, and supports visual display of monitoring data change trend and reliability index change curve in a chart form, so that the system realizes accurate evaluation and prediction of the reliability of the civil engineering structure, and provides support for structural safety monitoring and maintenance decision.

[0144] The above specific embodiments are only several optional embodiments of the present application, and based on the technical solutions of the present application and the related inspiration of the above embodiments, those skilled in the art can make various alternative improvements and combinations on the above specific embodiments.

Claims

1. A method for predicting structural reliability assessment based on long-term monitoring data, characterized by, The method comprises the following steps: Data preprocessing, pre-processing long-term monitoring data to improve data quality; Adaptive modeling of data distribution, constructing an adaptive data distribution model to adapt to complex data characteristics; Reliability index calculation and evaluation, calculating and evaluating the structural reliability index based on the data distribution model; Dynamic prediction of reliability, dynamic prediction of reliability based on the time characteristics of the data; The data distribution adaptive modeling stage adopts a hybrid Copula function model: Fit the edge distribution of each monitoring data variable, and determine the parameters of the edge distribution by maximum likelihood estimation method; Select several candidate Copula functions from Gaussian Copula function, t-Copula function and Clayton Copula function, and evaluate and select the candidate Copula functions by Akaike information criterion and Bayesian information criterion; Construct a hybrid Copula function model, combine the selected multiple Copula functions according to certain weights, and determine the weights by genetic algorithm optimization; In the reliability index calculation and evaluation, the time-varying characteristics of the structure are considered: Introduce a time-dependent random process to describe the changes of structural parameters, material properties and loads with time, and use Wiener process or Gamma process to simulate the fatigue damage accumulation of the structure; Add a time variable to the limit state function to dynamically evaluate the reliability of the structure at different times, and reflect the evolution of the structural reliability with time by continuously updating the random process model of the structural parameters.

2. The method of claim 1, wherein the method further comprises: The data preprocessing specifically includes: Data cleaning: Isolation Forest algorithm is used to detect outliers in monitoring data, and the monitoring data set is . Isolation Forest scores the data points by building multiple isolation trees. Points with scores exceeding the preset threshold are determined as outliers. For outliers, the weighted average of neighboring normal data points is used for replacement, and the weight is calculated according to the reciprocal of the distance. Data completion: for missing data, a method based on gradient boosting tree regression is used to complete the missing data, a gradient boosting tree model is trained, and the known data characteristics are used as input and the missing data are used as output to predict and complete the missing values; Data normalization: using Z-score normalization method, the data is transformed into a standard normal distribution with mean 0 and standard deviation 1 for data points , the normalized value , where is the data mean, is the standard deviation. 3.The structural reliability evaluation and prediction method based on long-term monitoring data according to claim 1, characterized in that, In the reliability index calculation and evaluation stage, an important sampling method based on subset simulation is used: Limit state function defining the structure wherein are random variables influencing the reliability of the structure Set an initial failure probability level, and divide the failure domain into multiple intermediate subsets by subset simulation; In each subset, important sampling method is used for sampling, sample samples are generated according to the hybrid Copula function model, and the limit state function values corresponding to the samples are calculated; The number of samples falling into each subset is counted, the failure probability of the structure is calculated step by step, and the reliability index is calculated according to the formula The reliability index is calculated, wherein is the inverse cumulative distribution function of the standard normal distribution.

4. The method of claim 1, wherein the method further comprises: In the reliability dynamic prediction stage, a gated recurrent unit neural network is combined with a Bayesian updating mechanism: A GRU neural network model is constructed, historical monitoring data and corresponding reliability indicators are taken as inputs, and future change trend of the reliability indicators is predicted , reset gate and candidate hidden state The calculation formula is as follows: ； ； ； ； wherein, is the current time input, is the previous time hidden state, is the weight matrix, is the bias vector, is the sigmoid function, is the hyperbolic tangent function, is the element-wise multiplication; When new monitoring data is obtained, the prediction result of the GRU model is corrected by the Bayesian updating mechanism, and the prior probability distribution is updated to the posterior probability distribution according to the new data.

5. The method of claim 1, wherein the method further comprises: After the data preprocessing stage, a data feature extraction step is added: The principal component analysis method is used to extract features of the preprocessed data, reduce data dimension, calculate a covariance matrix of the data, solve eigenvalues and eigenvectors of the covariance matrix, and select the first characteristic vectors with larger eigenvalues to form a projection matrix. Project the original data onto a projection matrix to obtain the feature data after dimension reduction.

6. The method of claim 1, wherein the method further comprises: In the reliability dynamic prediction stage, a multi-source data fusion technology is combined: In addition to long-term monitoring data, multi-source data of the structure is also fused, including design data, construction records and environmental data, and the environmental data includes temperature, humidity and wind speed; The multi-source data is fused by evidence theory or fuzzy mathematics method.

7. A structure reliability evaluation prediction system based on long-term monitoring data, characterized by, The method is suitable for the method of claim 1, comprising: A data preprocessing module is configured to preprocess long-term monitoring data to improve data quality; A data distribution adaptive modeling module is configured to build an adaptive data distribution model to adapt to complex data characteristics. A reliability index calculation and evaluation module is configured to calculate and evaluate the structural reliability index based on the data distribution model. A reliability dynamic prediction module is configured to dynamically predict the reliability according to the time characteristics of the data.

8. The system for structural reliability assessment and prediction based on long-term monitoring data of claim 7, further comprising: A user interaction module is configured to provide a friendly interface for users to view data, set parameters, select models, and show results, and support visual display of monitoring data trends and reliability index curves in the form of charts.

Images