Structural reliability evaluation and prediction method and system based on long-term monitoring data

Abstract

The invention relates to the technical field of civil engineering structure data monitoring, in particular to a structure reliability evaluation and prediction method and system based on long-term monitoring data. According to the technical scheme, the method comprises the following steps: data preprocessing: preprocessing long-term monitoring data to improve data quality; data distribution self-adaptive modeling is carried out, and a self-adaptive data distribution model is constructed to adapt to complex data features; reliability index calculation and evaluation: calculating a structure reliability index based on the data distribution model and completing evaluation; and reliability dynamic prediction: carrying out reliability dynamic prediction according to the time characteristic of the data. According to the method, the quality of long-term monitoring data can be effectively improved, key features can be accurately extracted, a solid foundation is laid for subsequent complex data modeling based on a mixed Copula function model, the structural reliability is dynamically evaluated and predicted in combination with time-varying features, and the reliability is improved by fusing multi-source data.

Description

Technical Field

[0001] The present invention 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 Art

[0002] In the assessment and prediction of structural reliability of long-term monitoring data in civil engineering, there is a prominent problem in the assessment and prediction method based on traditional probability statistics: it is too dependent on the amount and distribution of data.

[0003] Traditional probability and statistics methods usually require a large amount of monitoring data to accurately estimate the probability distribution and statistical characteristics of structural parameters. In actual civil engineering, obtaining a large amount of long-term, high-quality monitoring data often faces many difficulties. On the one hand, the layout and maintenance costs of monitoring equipment are high, and it is difficult to set up a sufficient number of monitoring points at each key part of the structure and obtain data stably for a long time; on the other hand, the service life of the structure is limited. In some new structures or structures in specific environments, it may not be possible to accumulate enough data in a short period of time to meet the requirements of traditional methods. This leads to the fact that when the amount of data is insufficient, the evaluation and prediction results based on traditional probability statistics may have large deviations and cannot accurately reflect the true reliability of the structure.

[0004] Traditional methods generally assume that monitoring data obey a certain probability distribution, such as normal distribution, lognormal distribution, etc. However, the monitoring data of actual civil engineering structures often have complex distribution characteristics, which may be affected by a combination of multiple factors and may not necessarily conform to these assumed ideal distributions. For example, affected by environmental factors (such as climate change, seismic activity), the non-uniformity of structural materials, and the nonlinearity of structural damage development, monitoring data may show complex distribution characteristics such as asymmetry, multi-peaks, or thick tails. 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 lead to wrong conclusions, which will affect the safety assessment and maintenance decisions of the structure.

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

[0006] The purpose of the present invention is to address the problem in the background technology that the structural reliability assessment and prediction of long-term monitoring data of civil engineering projects is too dependent on the amount and distribution of data, and to propose a structural reliability assessment and prediction method and system based on long-term monitoring data.

[0007] The technical solution of the present invention is a method for evaluating and predicting structural reliability based on long-term monitoring data, comprising the following steps:

[0008] Data preprocessing: first preprocess the long-term monitoring data to improve data quality;

[0009] Data distribution adaptive modeling, building 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] Dynamic prediction of reliability: Dynamic prediction of reliability is performed based on the time characteristics of data.

[0012] Optionally, the data preprocessing specifically includes:

[0013] Data cleaning: The isolation forest algorithm is used to detect outliers in the monitoring data. Assume that the monitoring data set is D = {x 1 ,x 2 ,…,x n}, Isolation Forest constructs multiple isolated trees to score data points for isolation. Points with scores exceeding a preset threshold are considered outliers. For outliers, they are replaced with the weighted average of neighboring normal data points, with the weight calculated based on the inverse of the distance;

[0014] Data completion: For missing data, we use a method based on gradient boosting tree regression to complete the missing data. We train a gradient boosting tree model, take known data features as input and missing data as output, and predict and complete the missing values.

[0015] Data normalization: Use the Z-score normalization method to transform the data into a standard normal distribution with a mean of 0 and a standard deviation of 1. i , the normalized value in is the data mean and s is the standard deviation.

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

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

[0018] Several candidate Copula functions are selected from Gaussian Copula function, t-Copula function and Clayton Copula function, and the candidate Copula functions are evaluated and selected using Akaike information criterion and Bayesian information criterion.

[0019] A hybrid Copula function model is constructed, and multiple screened Copula functions are combined according to certain weights, and the weights are determined by genetic algorithm optimization.

[0020] Optionally, in the reliability index calculation and evaluation stage, an important sampling method based on subset simulation is adopted:

[0021] Define the limit state function of the structure Z = g(X 1 ,X 2 ,…,X m ), where X 1 ,X 2 ,…,X m is the random variable that affects the structural reliability;

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

[0023] The importance sampling method is used in each subset to generate sampling samples according to the mixed Copula function model, and the limit state function value corresponding to the sample is calculated;

[0024] Count the number of samples that fall into each subset, calculate the failure probability of the structure step by step, and use the formula β = -Φ -1 (P f ) calculates the reliability index, where Φ -1 (·) is the inverse cumulative distribution function of the standard normal distribution.

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

[0026] Construct a GRU neural network model, take historical monitoring data and its corresponding reliability index as input, and predict the future trend of reliability index. The update gate z of the GRU model t , reset gate t and candidate hidden states The calculation formula is as follows:

[0027] z t =σ(W z [x t ,h t-1 ]+b z )

[0028] r t =σ(W r [x t ,h t-1 ]+b r )

[0029]

[0030]

[0031] Among them, xt Input for the current time, h t-1 is the hidden state at the previous moment, W is the weight matrix, b is the bias vector, σ is the sigmoid function, tanh is the hyperbolic tangent function, and ⊙ is the element-by-element multiplication;

[0032] When new monitoring data is obtained, the Bayesian update mechanism is used to correct the prediction results of the GRU model, and the prior probability distribution is updated to the posterior probability distribution according to the new data.

[0033] Optionally, after the data preprocessing stage, add a data feature extraction step:

[0034] The principal component analysis method is used to extract features from the preprocessed data, reduce the data dimension, calculate the covariance matrix of the data, solve its eigenvalues ​​and eigenvectors, and select the first k eigenvectors with larger eigenvalues ​​to form the projection matrix;

[0035] Project the original data onto the projection matrix to obtain the feature data after dimensionality reduction.

[0036] Optionally, in the calculation and evaluation of the reliability index, the time-varying characteristics of the structure are considered:

[0037] A time-dependent random process is introduced to describe the changes of structural parameters, material properties, and loads over time. The random process uses the Wiener process or the Gamma process to simulate the fatigue damage accumulation of the structure.

[0038] By adding the time variable into the limit state function, the structural reliability at different times is dynamically evaluated, and the evolution of the structural reliability over time is reflected in real time by continuously updating the random process model of the structural parameters.

[0039] Optionally, the reliability dynamic prediction stage is combined with multi-source data fusion technology:

[0040] In addition to long-term monitoring data, multi-source data of the structure is also integrated, including design data, construction records, and environmental data. Environmental data includes temperature, humidity, and wind speed;

[0041] Use evidence theory or fuzzy mathematics methods to fuse multi-source data.

[0042] In a second aspect, the present application provides a structural reliability assessment and prediction system based on long-term monitoring data, comprising:

[0043] Data preprocessing module, used to preprocess long-term monitoring data to improve data quality;

[0044] Data distribution adaptive modeling module, used to build an adaptive data distribution model to adapt to complex data characteristics;

[0045] A reliability index calculation and evaluation module, used to calculate the structural reliability index based on the data distribution model and complete the evaluation;

[0046] The reliability dynamic prediction module is used to dynamically predict the reliability based on the time characteristics of the data.

[0047] Optionally, it also includes: a user interaction module that provides a user interface for users to view data, set parameters, select models, and display results, and supports visual display of monitoring data change trends and reliability indicator change curves in the form of charts.

[0048] Compared with the prior art, the present invention has at least one of the following beneficial technical effects:

[0049] The isolation forest algorithm is used to clean outliers, gradient boosting tree regression is used to fill missing values, and Z-score normalization is used to process data, which significantly improves data quality. Subsequent principal component analysis is used to extract features, reduce dimensions, and improve computational efficiency.

[0050] The hybrid Copula function model combines multiple methods to accurately characterize data correlation and distribution characteristics, enhancing model adaptability and accuracy.

[0051] The reliability index is accurately calculated based on the important sampling method of subset simulation, and dynamic evaluation is achieved by considering the time-varying characteristics. The GRU neural network is combined with the Bayesian update mechanism and multi-source data fusion technology to improve the accuracy, reliability and comprehensiveness of the prediction.

[0052] The present invention can not only effectively improve the quality of long-term monitoring data, but also accurately extract key features, laying a solid foundation for subsequent complex data modeling based on the hybrid Copula function model, dynamically evaluate and predict structural reliability in combination with time-varying characteristics, and integrate multi-source data to improve reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a flow chart of a structural reliability assessment and prediction method based on long-term monitoring data. DETAILED DESCRIPTION

[0054] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0055] Example 1

[0056] like Figure 1 As shown, the present invention proposes a structural reliability assessment and prediction method based on long-term monitoring data, including data preprocessing, data distribution adaptive modeling, reliability index calculation and evaluation and reliability dynamic prediction, and each step is described in detail below.

[0057] 1. Data preprocessing: first preprocess the long-term monitoring data to improve the data quality; specifically including:

[0058] Data cleaning: The isolation forest algorithm is used to detect outliers in the monitoring data. Assume that the monitoring data set is D = {x 1 ,x 2 ,…,x n}, Isolation Forest constructs multiple isolated trees to score data points for isolation. Points with scores exceeding a preset threshold are considered outliers. For outliers, they are replaced with the weighted average of neighboring normal data points, with the weight calculated based on the inverse of the distance;

[0059] Data completion: For missing data, we use a method based on gradient boosting tree regression to complete the missing data. We train a gradient boosting tree model, take known data features as input and missing data as output, and predict and complete the missing values.

[0060] Data normalization: Use the Z-score normalization method to transform the data into a standard normal distribution with a mean of 0 and a standard deviation of 1. i , the normalized value in is the data mean and s is the standard deviation.

[0061] After the data preprocessing stage, the data feature extraction step is added:

[0062] The principal component analysis method is used to extract features from the preprocessed data, reduce the data dimension, calculate the covariance matrix of the data, solve its eigenvalues ​​and eigenvectors, and select the first k eigenvectors with larger eigenvalues ​​to form a projection matrix. The original data is projected onto the projection matrix to obtain the feature data after dimensionality reduction. Specifically, it includes:

[0063] The principal component analysis method is used to extract features from the preprocessed data and reduce the data dimension. The specific steps are as follows:

[0064] Calculate the covariance matrix: Let the preprocessed data set be X = [x ij ] n×p , where n is the number of samples, p is the number of features, and the calculation formula of the covariance matrix S is in is the mean matrix of the data set X, Each element of

[0065] Solve the eigenvalues ​​and eigenvectors: Perform eigendecomposition on the covariance matrix S, that is, solve the equation |S-λI|=0 to obtain p eigenvalues ​​λ 1 ≥λ 2 ≥…≥λ p ≥0 and the corresponding eigenvector e1 ,e 2 ,…,e p , where I is the p-order identity matrix;

[0066] Determine the number of principal components K: Use the following method to determine the value of K:

[0067] Scree plot method: draw a line graph of the eigenvalue versus the eigenvector number, observe the turning point of the eigenvalue downward trend in the line graph, after which the eigenvalue changes slowly, and select the number corresponding to the turning point as K;

[0068] Construct the projection matrix: Select the first K eigenvectors with larger eigenvalues ​​and arrange them in columns to form the projection matrix e 1 ,e 2 ,…,e K , the dimension of the projection matrix P is p×K;

[0069] Data projection: Project the original preprocessed data set X onto the projection matrix P to obtain the reduced-dimensional feature data Y=XP. The dimension of the reduced-dimensional data Y is n×K, which can reduce data redundancy and improve the efficiency of subsequent modeling and calculation.

[0070] The data quality of this embodiment is significantly improved: the data preprocessing uses the isolation forest algorithm to clean outliers and replaces them with weighted average values ​​to ensure data accuracy; the missing values ​​are filled by gradient boosting tree regression to reduce the information loss caused by missing data; Z-score normalization standardizes the data to facilitate subsequent analysis, effectively improves data quality, and provides a solid foundation for subsequent steps.

[0071] The principal component analysis method performs feature extraction after data preprocessing, reduces data dimension, removes redundant information, retains key features, improves computational efficiency, reduces data processing burden, and can better reveal the inherent laws of the data.

[0072] Second, data distribution adaptive modeling, build an adaptive data distribution model to adapt to complex data features. The data distribution adaptive modeling stage adopts a hybrid Copula function model:

[0073] First, the marginal distribution of each monitoring data variable is fitted, and the parameters of the marginal distribution are determined by the maximum likelihood estimation method; for different types of data, the normal distribution, lognormal distribution, and Weibull distribution can be selected;

[0074] Several candidate Copula functions are selected from Gaussian Copula function, t-Copula function and Clayton Copula function, and the candidate Copula functions are evaluated and selected using Akaike information criterion and Bayesian information criterion.

[0075] A hybrid Copula function model is constructed, and multiple screened Copula functions are combined according to certain weights, and the weights are determined by genetic algorithm optimization.

[0076] The specific steps are as follows:

[0077] Initialize the population: Set the number of mixed Copula functions to m, and the weight corresponding to each mixed Copula function to w i (i=1,2,…m), and satisfies

[0078]

[0079] Randomly generate N groups of weight vectors that meet the above conditions j=1,2,…N, forming the initial population, where N is the population size, N=50;

[0080] Define the fitness function: Select the negative log-likelihood function as the fitness function to measure the degree of fit of the mixed Copula function model to the data. Suppose the monitoring data is X = {X 1 ,X 2 ,…,X n}, the mixed Copula function C(u;w) is where u=(u 1 ,u 2 ,…,u d ) is the marginal distribution function value vector, C i (u; θ i ) is the i-th candidate Copula function, θ i is its parameter vector, and the fitness function F(W) is defined as:

[0081]

[0082] Among them, C i (u k θ i ) is the probability density function of the i-th candidate Copula function, u k is the marginal distribution function value vector corresponding to the kth data point;

[0083] Selection operation: Use the roulette wheel selection method to select individuals from the current population to enter the next generation, and calculate each individual W j The probability of selection:

[0084]

[0085] Then, random selection is performed according to the selection probability until N individuals are selected to form a new population;

[0086] Crossover operation: Perform a crossover operation on the selected population to generate new individuals. Randomly select two individuals. and Set the crossover probability P c =0.8, if the random number is less than P c , then a crossover operation is performed, using the linear combination crossover method to generate two new individuals W a' and W b' :

[0087] W a' =αW a +(1-α)W b

[0088] W b' =αW b +(1-α)W a

[0089] Where α is a randomly generated number in the interval (0,1). The newly generated individuals are normalized so that their weights satisfy:

[0090]

[0091] Mutation operation: Perform mutation operation on the population after crossover to increase the diversity of the population and set the mutation probability P m = 0.01, for each individual If the random number is less than P m , then for one of the weights Perform mutation and use uniform mutation method to randomly generate a new weight value in the interval [0,1] Then the individuals are normalized;

[0092] Termination condition judgment: Repeat the selection, crossover and mutation operations until the termination condition is met. The termination condition is that the fitness function value changes less than the preset threshold ∈=10 in several consecutive generations. -5 , the weight vector corresponding to the optimal individual finally obtained is the optimal weight of the mixed Copula function;

[0093] In this embodiment, the data distribution adaptive modeling adopts a mixed Copula function model, fits the marginal distribution parameters through maximum likelihood estimation, uses the Akaike information criterion and the Bayesian information criterion to screen the Copula function, and uses the genetic algorithm to optimize the weights. This can more accurately characterize the complex correlations and distribution characteristics between monitoring data and improve the adaptability and accuracy of the model.

[0094] 3. Reliability index calculation and evaluation: Calculate the structural reliability index based on the data distribution model and complete the evaluation; in the reliability index calculation and evaluation stage, the important sampling method based on subset simulation is adopted:

[0095] Define the limit state function of the structure Z = g(X 1 ,X 2 ,…,X m ), where X 1 ,X 2 ,…,X m is the random variable that affects the structural reliability;

[0096] Set the initial failure probability level and gradually divide the failure domain into multiple intermediate subsets through subset simulation;

[0097] The importance sampling method is used in each subset to generate sampling samples according to the mixed Copula function model, and the limit state function value corresponding to the sample is calculated;

[0098] Count the number of samples that fall into each subset, calculate the failure probability of the structure step by step, and use the formula β = -Φ -1 (P f ) calculates the reliability index, where Φ -1 (·) is the inverse cumulative distribution function of the standard normal distribution.

[0099] Based on the important sampling method of subset simulation, the limit state function is defined, the failure domain is divided by subset simulation, and combined with the sampling calculation of the hybrid Copula function model, the structural failure probability and reliability index can be accurately calculated, and the random characteristics and uncertainty of the structure are fully considered.

[0100] In the calculation and evaluation of the reliability index, the time-varying characteristics of the structure are considered:

[0101] A time-dependent random process is introduced to describe the changes of structural parameters, material properties, and loads over time. The random process uses the Wiener process or the Gamma process to simulate the fatigue damage accumulation of the structure.

[0102] By adding a time variable to the limit state function, the structural reliability at different times can be dynamically evaluated, and the evolution of the structural reliability over time can be reflected in real time by continuously updating the random process model of the structural parameters. By introducing a time-related random process to describe the changes in structural parameters and adding a time variable to the limit state function, the structural reliability at different times can be dynamically evaluated, and the evolution of the structural reliability over time can be reflected in real time, providing a basis for monitoring and maintenance of the entire life cycle of the structure.

[0103] 4. Dynamic prediction of reliability: Dynamic prediction of reliability is performed based on the time characteristics of the data. The dynamic prediction stage of reliability uses a gated recurrent unit neural network combined with a Bayesian update mechanism:

[0104] Construct a GRU neural network model, take historical monitoring data and its corresponding reliability index as input, and predict the future trend of reliability index. The update gate z of the GRU model t , reset gate t and candidate hidden states The calculation formula is as follows:

[0105] z t =σ(W z [x t ,h t-1 ]+b z )

[0106] r t =σ(W r [x t ,h t-1 ]+b r )

[0107]

[0108] Among them, x t Input for the current time, h t-1 is the hidden state at the previous moment, W is the weight matrix, b is the bias vector, σ is the sigmoid function, tanh is the hyperbolic tangent function, and ⊙ is the element-by-element multiplication.

[0109] When new monitoring data is obtained, the Bayesian update mechanism is used to correct the prediction results of the GRU model, and the prior probability distribution is updated to the posterior probability distribution according to the new data.

[0110] In the reliability dynamic prediction stage, multi-source data fusion technology is used to:

[0111] In addition to long-term monitoring data, multi-source data of the structure is also integrated, including design data, construction records, and environmental data. Environmental data includes temperature, humidity, and wind speed;

[0112] Use evidence theory or fuzzy mathematics methods to fuse multi-source data.

[0113] The dynamic prediction of reliability adopts a gated recurrent unit neural network combined with a Bayesian update mechanism. The GRU neural network predicts future trends based on historical data and reliability indicators. The Bayesian update mechanism corrects the prediction results based on new data to improve the prediction accuracy. It combines multi-source data fusion technology to integrate multi-source data such as design information, construction records, and environmental data, and uses evidence theory or fuzzy mathematics methods for fusion processing to further improve the reliability and comprehensiveness of the prediction.

[0114] On the other hand, the present application provides a structural reliability assessment and prediction system based on long-term monitoring data, comprising:

[0115] Data preprocessing module, used to preprocess long-term monitoring data to improve data quality;

[0116] Data distribution adaptive modeling module, used to build an adaptive data distribution model to adapt to complex data characteristics;

[0117] A reliability index calculation and evaluation module, used to calculate the structural reliability index based on the data distribution model and complete the evaluation;

[0118] The reliability dynamic prediction module is used to dynamically predict the reliability based on the time characteristics of the data.

[0119] Example 2

[0120] This embodiment, based on Embodiment 1, further includes:

[0121] In the reliability dynamic prediction step, transfer learning is introduced to improve the generalization ability of the model, including:

[0122] Source and target domain selection:

[0123] Determine the source domain and target domain. Assuming that the target structure is a cable-stayed bridge, select multiple cable-stayed bridges of the same type that have been built and have long-term monitoring data for the source domain. Collect monitoring data of the cable-stayed bridge in the source domain, including cable force, main beam deflection, and tower deflection data, as well as the preliminary monitoring data of the target cable-stayed bridge.

[0124] The data of the source domain and the target domain are preprocessed using the same method as the original data preprocessing module. The isolation forest algorithm is used to clean outliers, the missing values ​​are filled based on the gradient boosting tree regression, and the 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.

[0125] Transfer learning model construction:

[0126] A transfer learning strategy based on pre-training and fine-tuning is adopted. First, the GRU neural network is pre-trained 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 as m according to the number of input data features, the number of hidden layer nodes is set to h, and the number of output layer nodes is 1 (output reliability index prediction value). The Adam optimizer is used, the learning rate is set to 0.001, and the number of training rounds is T. 1 =100.

[0127] The first k layers of the pre-trained GRU model, k = 2, are transferred to the target domain model, and then fine-tuned on the target domain data. During the fine-tuning process, the first k parameters of the pre-training are fixed, and only the parameters of the subsequent layers (such as the output layer and the layer close to the output layer) are adjusted. The learning rate of fine-tuning is set to 0.0001, and the number of training rounds is T. 2 = 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 prediction of the reliability of the target structure.

[0128] Hybrid prediction method combining deep learning and physical models

[0129] Physical model construction:

[0130] According to the mechanical principles and material properties of the structure, a physical model of the structure is established. Taking the high-rise building structure as an example, a bar system model is used. Based on the displacement method and force method theory in structural mechanics, the mechanical equilibrium equation of the structure under horizontal loads (such as wind loads and earthquake effects) and vertical loads is established. Assume that the mass of each layer of the structure is m i , the interlayer stiffness is k i , the horizontal displacement is x i , according to D'Alembert's principle and Hooke's law, the equation of motion is established:

[0131]

[0132] where c i is the damping coefficient, F i (t) is the external force acting on the i-th layer, and are acceleration and velocity respectively.

[0133] Combined with the constitutive relationship of the material, the stress-strain relationship of concrete adopts the curve model recommended by the specification, and the constitutive relationship of the steel bar adopts the bilinear strengthening model to determine the mechanical response of the structure under different stress states.

[0134] Fusion of deep learning models and physical models:

[0135] Combine the GRU neural network with the physical model. The stress σ of the key parts of the structure calculated by the physical model phy , strain ε phy Physical quantity, and monitoring data displacement monitoring value d meas , stress monitoring value σ meas Together as the input of the GRU model. The input data dimension of the GRU model is n, then the physical model calculation results and monitoring data are spliced ​​to form an input vector [σ phy ,ε phy ,d meas ,σmeas ,…].

[0136] At the same time, the prediction results of the deep learning model are used to modify the physical model. The GRU model is used to predict the load changes of the structure in the future. Assuming that the predicted wind load σ at the future time t is meas , using it as the input load of the physical model to recalculate the mechanical response of the structure, thereby making a more accurate prediction of the reliability of the structure.

[0137] Example 3

[0138] In this embodiment, based on the first embodiment, the structural reliability assessment and prediction system based on long-term monitoring data also includes a user interaction module, which provides a user interface for users to view data, set parameters, select models, and display results. It also supports the visualization of monitoring data change trends and reliability index change curves in the form of charts. The system realizes accurate assessment and prediction of the reliability of civil engineering structures, and provides support for structural safety monitoring and maintenance decisions.

[0139] Example 4

[0140] This embodiment provides a structural reliability assessment and prediction system based on long-term monitoring data, including:

[0141] Data acquisition devices are used to collect long-term monitoring data of civil engineering structures from various sensors and monitoring equipment, covering stress, strain, displacement, and vibration physical quantity data, and support multiple data transmission protocols to stably and accurately transmit data to the next link;

[0142] The data preprocessing module includes a data cleaning unit, which uses the isolation forest algorithm and a statistical threshold-based method to identify and mark abnormal data points, and corrects them through interpolation and mean filling methods; a data completion unit, which uses machine learning algorithms such as gradient boosting tree regression to complete missing data; and a data normalization unit, which uses Z-score normalization, minimum-maximum normalization and other methods to unify the scale of data.

[0143] The data distribution adaptive modeling module is equipped with a marginal distribution fitting unit, which provides a variety of distribution function options such as normal distribution, lognormal distribution, Weibull distribution, etc. for each monitoring data variable, and determines the parameters through the maximum likelihood estimation method; the Copula function screening unit selects suitable functions from various Copula functions such as Gaussian Copula, t-Copula, Clayton Copula, etc. according to AIC (Akaike Information Criterion) and BIC (Bayesian Information Criterion); the mixed Copula function construction unit combines the screened Copula functions according to the weights determined by genetic algorithm optimization to construct a mixed Copula function model.

[0144] The reliability index calculation and evaluation module is equipped with a limit state function definition unit, which supports users to customize or select preset limit state functions according to the mechanical characteristics and failure modes of civil engineering structures; the importance sampling calculation unit adopts the importance sampling method based on subset simulation, generates sampling samples according to the hybrid Copula function model, and substitutes them into the limit state function to calculate the failure probability and reliability index; the reliability evaluation unit evaluates the structural reliability according to the reliability index and the preset structural health status classification standard and outputs the evaluation results.

[0145] The reliability dynamic prediction module includes a time series model construction unit, which uses a gated recurrent unit (GRU) neural network and an autoregressive integrated moving average model (ARIMA) to construct a prediction model with historical monitoring data and its corresponding reliability indicators as input; the prediction result update unit uses a Bayesian update mechanism combined with multi-source data fusion technology to update the model when acquiring new monitoring data, and outputs the future structural reliability prediction value.

[0146] The user interaction module provides a friendly interface for users to view data, set parameters, select models, display results, and other operations. It also supports the visualization of monitoring data change trends and reliability index change curves in the form of charts. The system can accurately evaluate and predict the reliability of civil engineering structures, and provide support for structural safety monitoring and maintenance decisions.

[0147] The above specific embodiments are only several optional embodiments of the present invention. Based on the technical solutions of the present invention and the relevant inspirations of the above embodiments, those skilled in the art can make various alternative improvements and combinations to the above specific embodiments.

Claims

1. A structural reliability assessment and prediction method based on long-term monitoring data, characterized in that: The following steps are involved: Data preprocessing: first preprocess the long-term monitoring data to improve data quality; Data distribution adaptive modeling, building an adaptive data distribution model to adapt to complex data characteristics; Reliability index calculation and evaluation: calculate the structural reliability index based on the data distribution model and complete the evaluation; Dynamic prediction of reliability: Dynamic prediction of reliability is performed based on the time characteristics of data.

2. A structural reliability assessment and prediction method based on long-term monitoring data according to claim 1, characterized in that: The data preprocessing specifically includes: Data cleaning: The isolation forest algorithm is used to detect outliers in the monitoring data. Assume that the monitoring data set is D = {x1, x2, ..., x n }, Isolation Forest constructs multiple isolated trees to score data points for isolation. Points with scores exceeding a preset threshold are considered outliers. For outliers, they are replaced with the weighted average of neighboring normal data points, with the weight calculated based on the inverse of the distance; Data completion: For missing data, we use a method based on gradient boosting tree regression to complete the missing data. We train a gradient boosting tree model, take known data features as input and missing data as output, and predict and complete the missing values. Data normalization: Use the Z-score normalization method to transform the data into a standard normal distribution with a mean of 0 and a standard deviation of 1. i , the normalized value in is the data mean and s is the standard deviation.

3. The structural reliability assessment and prediction method based on long-term monitoring data according to claim 1 is characterized in that: The data distribution adaptive modeling stage adopts a mixed Copula function model: Fit the marginal distribution of each monitoring data variable and determine the parameters of the marginal distribution by maximum likelihood estimation method; Several candidate Copula functions are selected from Gaussian Copula function, t-Copula function and Clayton Copula function, and the candidate Copula functions are evaluated and selected using Akaike information criterion and Bayesian information criterion. A hybrid Copula function model is constructed, and multiple screened Copula functions are combined according to certain weights, and the weights are determined by genetic algorithm optimization.

4. The structural reliability assessment and prediction method based on long-term monitoring data according to claim 1 is characterized in that: In the reliability index calculation and evaluation stage, an important sampling method based on subset simulation is adopted: Define the limit state function of the structure Z = g(X1, X2, ..., X m ), where X1, X2, …, X m is the random variable that affects the structural reliability; Set the initial failure probability level and gradually divide the failure domain into multiple intermediate subsets through subset simulation; The importance sampling method is used in each subset to generate sampling samples according to the mixed Copula function model, and the limit state function value corresponding to the sample is calculated; Count the number of samples that fall into each subset, calculate the failure probability of the structure step by step, and use the formula β = -Φ -1 (P f ) calculates the reliability index, where Φ -1 (·) is the inverse cumulative distribution function of the standard normal distribution.

5. The structural reliability assessment and prediction method based on long-term monitoring data according to claim 1 is characterized in that: The reliability dynamic prediction stage adopts a gated recurrent unit neural network combined with a Bayesian update mechanism: Construct a GRU neural network model, take historical monitoring data and its corresponding reliability index as input, and predict the future trend of reliability index. The update gate z of the GRU model t , reset gate t and candidate hidden states The calculation formula is as follows: z t =σ(W z [x t ,h t-1 ]+b z ) r t =σ(W r [x t ,h t-1 ]+b r ) Among them, x t Input for the current time, h t-1 is the hidden state at the previous moment, W is the weight matrix, b is the bias vector, σ is the sigmoid function, tanh is the hyperbolic tangent function, and ⊙ is the element-by-element multiplication; When new monitoring data is obtained, the Bayesian update mechanism is used to correct the prediction results of the GRU model, and the prior probability distribution is updated to the posterior probability distribution according to the new data.

6. The structural reliability assessment and prediction method based on long-term monitoring data according to claim 1 is characterized in that: After the data preprocessing stage, the data feature extraction step is added: The principal component analysis method is used to extract features from the preprocessed data, reduce the data dimension, calculate the covariance matrix of the data, solve its eigenvalues ​​and eigenvectors, and select the first k eigenvectors with larger eigenvalues ​​to form the projection matrix; Project the original data onto the projection matrix to obtain the feature data after dimensionality reduction.

7. The structural reliability assessment and prediction method based on long-term monitoring data according to claim 1 is characterized in that: In the calculation and evaluation of the reliability index, the time-varying characteristics of the structure are considered: A time-dependent random process is introduced to describe the changes of structural parameters, material properties, and loads over time. The random process uses the Wiener process or the Gamma process to simulate the fatigue damage accumulation of the structure. By adding the time variable into the limit state function, the structural reliability at different times is dynamically evaluated, and the evolution of the structural reliability over time is reflected in real time by continuously updating the random process model of the structural parameters.

8. The structural reliability assessment and prediction method based on long-term monitoring data according to claim 1 is characterized in that: The reliability dynamic prediction stage combines multi-source data fusion technology: In addition to long-term monitoring data, multi-source data of the structure is also integrated, including design data, construction records, and environmental data. Environmental data includes temperature, humidity, and wind speed; Use evidence theory or fuzzy mathematics methods to fuse multi-source data.

9. A structural reliability assessment and prediction system based on long-term monitoring data, characterized in that: include: Data preprocessing module, used to preprocess long-term monitoring data to improve data quality; Data distribution adaptive modeling module, used to build an adaptive data distribution model to adapt to complex data characteristics; A reliability index calculation and evaluation module, used to calculate the structural reliability index based on the data distribution model and complete the evaluation; The reliability dynamic prediction module is used to dynamically predict the reliability based on the time characteristics of the data.

10. The structural reliability assessment and prediction system based on long-term monitoring data according to claim 9, further comprising: The user interaction module provides a friendly interface for users to view data, set parameters, select models, and display results. It also supports the visualization of monitoring data change trends and reliability indicator change curves in the form of charts.

Images