Structural reliability updating method based on domain adaptive network integration

Abstract

The invention discloses a structural reliability updating method based on domain adaptive network integration. The method comprises the following steps: constructing Monte Carlo sample pools of a source domain and a target domain; constructing a proxy model based on deep integration and active learning as a source domain proxy model; generating migration training set sample points; constructing a domain adaptive network integration agent model as a target domain agent model; selecting fine tuning training set sample points based on a learning function; calculating a sample point limit state function value to obtain a fine tuning training set; performing partial fine tuning on the target domain agent model; performing failure probability estimation by adopting a Monte Carlo method, and judging whether the failure probability meets a convergence criterion or not; and outputting a failure probability estimation value updated based on the load variable distribution change. According to the method, updating and solving of the failure probability can be accurately and efficiently completed according to the load variable distribution change condition, the safety degree of continuous service of the structure can be quantified in real time, and therefore more scientific and economical maintenance measures and overhaul plans can be formulated.

Description

Technical Field

[0001] The present invention belongs to the technical field of structural reliability analysis, and in particular relates to a structural reliability updating method based on domain adaptive network integration. Background Art

[0002] In order to deal with various uncertain factors such as load conditions, material properties, and manufacturing processes in complex engineering practical problems, probability-based structural reliability methods have become a necessary means for accurate structural safety assessment. The structural reliability method regards various uncertain factors as random variables, and quantifies various uncertainties in structural assessment through probabilistic methods, which can better balance its safety and economy. The limit state functions of complex engineering structures often have implicit, multi-dimensional, and highly nonlinear characteristics, and time-consuming finite element models need to be called to complete the calculation. In order to improve the efficiency of structural reliability calculations, fitting the complex implicit limit state functions of the structure through surrogate models to solve the failure probability has become a research hotspot.

[0003] However, during the service life of a structure, its performance does not remain static, but will be affected by a variety of time-related and environmental factors, including but not limited to load changes, environmental corrosion, material aging, and some accidental effects, which will lead to a reduction in structural strength and thus affect the overall reliability. Therefore, on the basis of comprehensively considering various uncertain factors of the structure, by incorporating real monitoring information to dynamically update and adjust the prediction model, the deviation between the evaluation prediction results and the actual situation can be effectively reduced.

[0004] Considering that some structures are not convenient for direct monitoring of stress and strain information, they can only be updated through indirect monitoring variables such as loads, that is, the distribution of input variables for structural reliability analysis will change. Changes in the distribution of input variables will lead to corresponding changes in the failure boundary, resulting in a decrease in the local fitting accuracy of the original proxy model to the failure boundary. The process of rebuilding a new proxy model takes a long time, ignoring the availability of the original proxy model and having low computational efficiency. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, an object of the present invention is to provide a structural reliability updating method based on domain adaptive network integration to solve the problem of low efficiency of structural reliability updating in the prior art.

[0006] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0007] The structural reliability updating method based on domain adaptive network integration of the present invention comprises the following steps:

[0008] 1) Determine the distribution of the load variable before and after the change, which are used as the source domain distribution and the target domain distribution respectively;

[0009] 2) Constructing the source domain Monte Carlo sample pool and the target domain Monte Carlo sample pool;

[0010] 3) Construct a proxy model based on deep integration and active learning according to the source domain distribution as the source domain proxy model;

[0011] 4) Using the Latin hypercube sampling method corresponding to the uniform distribution to generate source domain migration training set sample points and target domain migration training set sample points, the source domain migration training set sample points calculate the corresponding limit state function values ​​through the source domain proxy model;

[0012] 5) Based on the source domain proxy model and the total loss function, a domain adaptive network integration proxy model is constructed as the target domain proxy model;

[0013] 6) Select fine-tuning training set sample points in the target domain Monte Carlo sample pool based on active learning strategy;

[0014] 7) Calculate the limit state function value of the sample points of the fine-tuning training set to obtain the fine-tuning training set;

[0015] 8) Partially fine-tune the target domain proxy model and use the Monte Carlo method to estimate the probability of structural failure;

[0016] 9) Determine whether the structural failure probability meets the convergence criterion. If so, end and output the current failure probability, and complete the failure probability calculation based on the load variable distribution change update; if not, proceed to step 10);

[0017] 10) Select new sample points based on the learning function value and add them to the fine-tuning training set sample points, and return to step 7).

[0018] Furthermore, in step 2), based on the probability density distribution of the source domain random variables and the probability density distribution of the target domain random variables, the inverse transformation method is used to obtain the variable samples that obey the distribution function, and the source domain Monte Carlo sample pool S of the random variables is constructed. MC_s , target domain Monte Carlo sample pool S MC_t .

[0019] Furthermore, the step 3) specifically includes:

[0020] According to the source domain distribution, the Latin hypercube sampling method corresponding to the uniform distribution is used to generate the initial training set sample points; the true limit state function values ​​of the training set sample points are calculated to obtain the training set; a weighted mean square error loss function is established based on the limit state function value; M artificial neural network sub-models are constructed according to the weighted mean square error loss function, and a proxy model based on deep integration and active learning is constructed through uniformly weighted mixing; the direct Monte Carlo method is used to estimate the structural failure probability; the learning function values ​​of all sample points in the source domain Monte Carlo sample pool are calculated, new sample points are selected and added to the training set sample points, and the training is iterated until the failure probability convergence criterion is met.

[0021] Furthermore, in step 4), based on the upper and lower bounds of the source domain Monte Carlo sample pool and the target domain Monte Carlo sample pool, a Latin hypercube sampling method corresponding to a uniform distribution is used to sample in the interval, and n migration training set sample points X are selected respectively. s ={x s1 ,x s2 ,x s3 ,...,x sn}, X t ={x t1 ,x t2 ,x t3 ,…,x tn}, where X s Output the corresponding limit state function value through the source domain proxy model, x sn represents the sample points of the source domain migration training set, X s For x sn The collection of x tn represents the sample points of the target domain migration training set, X t For x tn A collection of .

[0022] Furthermore, the step 5) specifically includes:

[0023] Through the constructed source domain proxy model, the network structure of the artificial neural network sub-model is read respectively to complete the construction of M domain adaptive network sub-models; the domain adaptive network sub-model consists of three parts: feature extraction module, adaptive module, and regression module: the feature extraction module is shared by the source domain and target domain networks, and is used to extract the deep features of the source domain and target domain data samples. The input layer and the first hidden layer of the artificial neural network sub-model are selected as feature extractors; the adaptive module introduces the maximum mean difference to calculate the distribution difference between the source domain deep features and the target domain deep features. The calculation expression is as follows:

[0024]

[0025] Where, the subscript H represents the reproducing kernel Hilbert space; n s ,nt are the number of samples in the source domain and the target domain respectively; φ(·) represents the nonlinear feature mapping function that maps the original data to the RHKS space; L MMD (D s ,D t ) represents the maximum mean difference between the source domain distribution and the target domain distribution;

[0026] The regression module consists of a fully connected layer network, which is used to establish the mapping relationship between deep features and predicted response values;

[0027] The total loss function is a loss function that comprehensively considers the regression prediction error and the difference between domains. The calculation expression is as follows:

[0028] L=L MSE (X L ,y)+λL MMD (D s ,D t )

[0029] Where, L MSE (X L ,y) represents the regression loss of the domain adaptation network sub-model, which represents the difference between the predicted value and the label value of the source domain training set, and uses the MSE loss function; X L is the source domain sample label value, y is the predicted value output by the source domain sample after passing through the entire domain adaptation network sub-model; L MMD (D s ,D t ) represents the distribution difference between the source domain and the target domain data; λ is a constant coefficient; L is the total loss function;

[0030] After completing the construction and training of the domain adaptive network sub-model, the M domain adaptive network sub-models are uniformly weighted mixed to obtain the domain adaptive network integrated proxy model as the target domain proxy model; the domain adaptive network integrated proxy model outputs the sample point x i The limit state function predicts the mean and standard deviation as follows:

[0031]

[0032] In the formula, μ(x i ) is the sample point x i The predicted mean of the limit state function represents the predicted output value of the domain adaptive network integration model for the sample point xi; σ(x i ) is the sample point x i The limit state function prediction standard deviation represents the domain adaptive network integration model for the sample point x i The prediction uncertainty of the output value; M is the number of domain adaptation network sub-models; represents the jth domain adaptation network sub-model For the sample point x i The predicted value of the limit state function, whose model parameter is θ j .

[0033] Furthermore, in step 6), the active learning strategy is used to adaptively select sample points of the fine-tuning training set, and the target domain Monte Carlo sample pool S is calculated. MC_t The learning function values ​​U(x) of all sample points in the training set are obtained, and the k sample points with the smallest U(x) values ​​are selected as the sample points T of the fine-tuning training set. t ; The calculation expression of U(x) is:

[0034]

[0035] In the formula, μ(x i ) and σ(x i ) represent the sample points x i The predicted mean and predicted standard deviation of the limit state function.

[0036] Furthermore, in step 7), the limit state function value G of the fine-tuning training set sample point is calculated by calling the limit state function or the finite element model. t , get the fine-tuning training set {T t ,G t}.

[0037] Furthermore, the step 8) specifically includes:

[0038] For each domain adaptive network sub-model in the target domain proxy model, the model input layer and the first hidden layer network parameters are first fixed, and the output layer parameters are retrained based on the fine-tuning training set; then the parameter freeze is unfrozen and all network parameters are fine-tuned. At this time, the model parameters are only slightly adjusted; the estimated value of the structural failure probability is calculated based on the Monte Carlo method The expression is as follows:

[0039]

[0040] Where N F is the number of sample points whose limit state function value is less than 0, and N is the sample pool S MC_t The number of sample points.

[0041] Furthermore, the convergence criterion expression in step 9) is:

[0042]

[0043] in, represents the estimated failure probability of the τth iteration, represents the estimated failure probability of the τ-1th iteration; ec is the threshold and set e c =0.005; the convergence condition is that the convergence criterion is met for three consecutive iterations.

[0044] Furthermore, the step 10) specifically includes:

[0045] Calculate the target domain sample pool S MC_t The learning function values ​​U(x) of all sample points in , and select the minimum value U min The corresponding sample points are added to the fine-tuning training set as new sample points.

[0046] Beneficial effects of the present invention:

[0047] The present invention introduces the maximum mean difference to calculate the inter-domain difference, constructs a domain adaptive network model to complete the update of the proxy model, and completes the network training by comprehensively considering the total loss function of the regression loss and the maximum mean difference, thereby minimizing the prediction error and the inter-domain difference.

[0048] The present invention combines active learning strategies to complete adaptive sampling of the best points for model fine-tuning effects, thereby reducing the sample points required for model fine-tuning.

[0049] The present invention can fully utilize the original proxy model to complete the efficient and accurate update of the structural failure probability when the input variable distribution changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 Flow chart of the method of the present invention.

[0051] Figure 2 Schematic diagram of the cantilever cylinder structure used in the embodiment.

[0052] Figure 3 Schematic diagram of the domain adaptation network sub-model. DETAILED DESCRIPTION

[0053] In order to facilitate the understanding of those skilled in the art, the present invention is further described below in conjunction with embodiments and drawings. The contents mentioned in the implementation modes are not intended to limit the present invention.

[0054] This embodiment is a reliability analysis problem of a cantilever cylinder structure with 9-dimensional random variables;

[0055] like Figure 2 As shown, considering the failure of structural yield strength, its limit state function is as follows:

[0056] g(x)=S-σ max

[0057] Where S represents the structural yield strength; σ maxIt represents the maximum Misses stress of the structure;

[0058] Specifically:

[0059]

[0060] τ zx =(Td / 4I)

[0061] Among them, θ1=5°, θ2=10°; σ x and τ zx are normal stress and torsional stress respectively; P, F1, F2 are concentrated forces on the outside of the structure; T is the torque on the structure; M is the bending moment; A is the cross-sectional area of ​​the tube; I is the moment of inertia of the section; the relevant parameters can be calculated by the following formula:

[0062]

[0063] In the structural reliability update problem, the distribution of monitoring variables before the change and other variables are used as the source domain data distribution, and the distribution of monitoring variables after the change and other variables are used as the target domain data distribution;

[0064] Table 1 below shows the load variable distribution before and after the change. The parameter table of all random variables is as follows:

[0065] Table 1

[0066]

[0067] Reference Figures 1 to 3 As shown, a structural reliability updating method based on domain adaptive network integration of the present invention comprises the following steps:

[0068] 1) Determine the distribution of the load variable before and after the change, and use them as the source domain distribution and target domain distribution, as shown in Table 1.

[0069] 2) Constructing the source domain Monte Carlo sample pool and the target domain Monte Carlo sample pool;

[0070] Based on the probability density distribution of the source domain random variables and the probability density distribution of the target domain random variables, the inverse transformation method is used to obtain the variable samples that obey the distribution function, and the source domain Monte Carlo sample pool S of the random variables is constructed. MC_s , target domain Monte Carlo sample pool S MC_t ;

[0071] 3) Construct a proxy model based on deep integration and active learning according to the source domain distribution as the source domain proxy model;

[0072] According to the source domain distribution, the Latin hypercube sampling method corresponding to the uniform distribution is used to generate the initial training set sample points; the true limit state function values ​​of the training set sample points are calculated to obtain the training set; a weighted mean square error loss function is established based on the limit state function value; M artificial neural network sub-models are constructed according to the weighted mean square error loss function, and a proxy model based on deep integration and active learning is constructed through uniformly weighted mixing; the direct Monte Carlo method is used to estimate the structural failure probability; the learning function values ​​of all sample points in the source domain Monte Carlo sample pool are calculated, new sample points are selected and added to the training set sample points, and the training is iterated until the failure probability convergence criterion is met.

[0073] 4) Using the Latin hypercube sampling method corresponding to the uniform distribution to generate source domain migration training set sample points and target domain migration training set sample points, the source domain migration training set sample points calculate the corresponding limit state function values ​​through the source domain proxy model;

[0074] Wherein, the step 4) is based on the upper and lower bounds of the source domain Monte Carlo sample pool and the target domain Monte Carlo sample pool, respectively, and the Latin hypercube sampling method corresponding to the uniform distribution is used to sample in the interval, and n migration training set sample points X are selected respectively. s ={x s1 ,x s2 ,x s3 ,...,x sn}, X t ={x t1 ,x t2 ,x t3 ,...,x tn}, where X s Output the corresponding limit state function value through the source domain proxy model, x sn represents the sample points of the source domain migration training set, X s For x sn The collection of x tn represents the sample points of the target domain migration training set, X t For x tn In this embodiment, n is 1000.

[0075] 5) Based on the source domain proxy model and the total loss function, a domain adaptive network integration proxy model is constructed as the target domain proxy model;

[0076] Through the constructed source domain proxy model, the network structure of the artificial neural network sub-model is read respectively to complete the construction of M domain adaptive network sub-models; the domain adaptive network sub-model consists of three parts: feature extraction module, adaptive module, and regression module: the feature extraction module is shared by the source domain and target domain networks, and is used to extract the deep features of the source domain and target domain data samples. The input layer and the first hidden layer of the artificial neural network sub-model are selected as feature extractors; the adaptive module introduces the maximum mean difference to calculate the distribution difference between the source domain deep features and the target domain deep features. The calculation expression is as follows:

[0077]

[0078] Where, the subscript H represents the reproducing kernel Hilbert space; n s ,n t are the number of samples in the source domain and the target domain respectively; φ(·) represents the nonlinear feature mapping function that maps the original data to the RHKS space; L MMD (D s ,D t ) represents the maximum mean difference between the source domain distribution and the target domain distribution;

[0079] The regression module consists of a fully connected layer network, which is used to establish the mapping relationship between deep features and predicted response values;

[0080] The total loss function is a loss function that comprehensively considers the regression prediction error and the difference between domains. The calculation expression is as follows:

[0081] L=L MSE (X L ,y)+λL MMD (D s ,D t )

[0082] Where, L MSE (X L ,y) represents the regression loss of the domain adaptation network sub-model, which represents the difference between the predicted value and the label value of the source domain training set, and uses the MSE loss function; X L is the source domain sample label value, y is the predicted value output by the source domain sample after passing through the entire domain adaptation network sub-model; L MMD (D s ,D t ) represents the distribution difference between the source domain and the target domain data; λ is a constant coefficient; L is the total loss function;

[0083] After completing the construction and training of the domain adaptive network sub-model, the M domain adaptive network sub-models are uniformly weighted mixed to obtain the domain adaptive network integrated proxy model as the target domain proxy model; the domain adaptive network integrated proxy model outputs the sample point x iThe limit state function predicts the mean and standard deviation as follows:

[0084]

[0085]

[0086] In the formula, μ(x i ) is the sample point x i The predicted mean of the limit state function represents the predicted output value of the domain adaptive network integration model for the sample point xi; σ(x i ) is the sample point x i The limit state function prediction standard deviation represents the domain adaptive network integration model for the sample point x i The prediction uncertainty of the output value; M is the number of domain adaptation network sub-models; represents the jth domain adaptation network sub-model For the sample point x i The predicted value of the limit state function, whose model parameter is θ j .

[0087] 6) Select fine-tuning training set sample points in the target domain Monte Carlo sample pool based on active learning strategy;

[0088] Among them, in step 6), the active learning strategy is used to adaptively select the fine-tuning training set sample points, and the target domain Monte Carlo sample pool S is calculated. MC_t The learning function values ​​U(x) of all sample points in the training set are obtained, and the k sample points with the smallest U(x) values ​​are selected as the sample points T of the fine-tuning training set. t ; In this embodiment, k is 10; the calculation expression of U(x) is:

[0089]

[0090] In the formula, μ(x i ) and σ(x i ) represent the sample points x i The predicted mean and predicted standard deviation of the limit state function.

[0091] 7) Calculate the limit state function value of the sample points of the fine-tuning training set to obtain the fine-tuning training set;

[0092] By calling the limit state function or finite element model, the limit state function value G of the fine-tuning training set sample point is calculated. t , get the fine-tuning training set {T t ,G t}.

[0093] 8) Partially fine-tune the target domain proxy model and use the Monte Carlo method to estimate the probability of structural failure;

[0094] For each domain adaptive network sub-model in the target domain proxy model, the model input layer and the first hidden layer network parameters are first fixed, and the output layer parameters are retrained based on the fine-tuning training set; then the parameter freeze is unfrozen and all network parameters are fine-tuned. At this time, the model parameters are only slightly adjusted; the estimated value of the structural failure probability is calculated based on the Monte Carlo method The expression is as follows:

[0095]

[0096] Where N F is the number of sample points whose limit state function value is less than 0, and N is the sample pool S MC_t The number of sample points.

[0097] 9) Determine whether the structural failure probability meets the convergence criterion. If so, end and output the current failure probability, and complete the failure probability calculation based on the load variable distribution change update; if not, proceed to step 10);

[0098] Wherein, the convergence criterion expression in step 9) is:

[0099]

[0100] in, represents the estimated failure probability of the τth iteration, represents the estimated failure probability of the τ-1th iteration; e c is the threshold and set e c =0.005; the convergence condition is that the convergence criterion is met for three consecutive iterations.

[0101] 10) Select new sample points based on the learning function value and add them to the fine-tuning training set sample points, and return to step 7).

[0102] The calculation results of reconstructing the target domain proxy model by using the direct Monte Carlo method (MCS), the active learning Kriging method (AK-MCS+U), the proxy model method based on deep integration and active learning (DE-AL), and updating the target domain proxy model by using the method of the present invention are shown in Table 2. The calculation results obtained by the direct Monte Carlo method are used as the reference for the true value of the structural reliability problem. The estimated value of the failure probability obtained by the method of the present invention is closest to the reference value, and the relative error is only -0.72%, which is smaller than the other three methods. At the same time, the number of calls to the limit state function is only 7 times, which is significantly lower than other calculation methods. The results show that the method of the present invention can ensure the high accuracy of the estimated value of the failure probability after the input variable changes with a small number of calls to the limit state function; Table 2 is a comparison of the calculation results as follows:

[0103] Table 2

[0104]

[0105] The present invention has many specific application paths. The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements can be made without departing from the principle of the present invention. These improvements should also be regarded as the protection scope of the present invention.

Claims

1. A structural reliability updating method based on domain adaptive network integration, characterized in that: Here are the steps: 1) Determine the distribution of the load variable before and after the change, which are used as the source domain distribution and the target domain distribution respectively; 2) Constructing the source domain Monte Carlo sample pool and the target domain Monte Carlo sample pool; 3) Construct a proxy model based on deep integration and active learning according to the source domain distribution as the source domain proxy model; 4) Using the Latin hypercube sampling method corresponding to the uniform distribution to generate source domain migration training set sample points and target domain migration training set sample points, the source domain migration training set sample points calculate the corresponding limit state function values ​​through the source domain proxy model; 5) Based on the source domain proxy model and the total loss function, a domain adaptive network integration proxy model is constructed as the target domain proxy model; 6) Select fine-tuning training set sample points in the target domain Monte Carlo sample pool based on active learning strategy; 7) Calculate the limit state function value of the sample points of the fine-tuning training set to obtain the fine-tuning training set; 8) Partially fine-tune the target domain proxy model and use the Monte Carlo method to estimate the probability of structural failure; 9) Determine whether the structural failure probability meets the convergence criterion. If so, end and output the current failure probability, and complete the failure probability calculation based on the load variable distribution change update; if not, proceed to step 10); 10) Select new sample points based on the learning function value and add them to the fine-tuning training set sample points, and return to step 7).

2. The structural reliability updating method based on domain adaptive network integration according to claim 1 is characterized in that: In the step 2), based on the probability density distribution of the source domain random variables and the probability density distribution of the target domain random variables, the inverse transformation method is used to obtain the variable samples that obey the distribution function, and the source domain Monte Carlo sample pool S of the random variables is constructed. MC_s , target domain Monte Carlo sample pool S MC_t .

3. The structural reliability updating method based on domain adaptive network integration according to claim 1 is characterized in that: The step 3) specifically includes: According to the source domain distribution, the Latin hypercube sampling method corresponding to the uniform distribution is used to generate the initial training set sample points; the true limit state function values ​​of the training set sample points are calculated to obtain the training set; a weighted mean square error loss function is established based on the limit state function value; M artificial neural network sub-models are constructed according to the weighted mean square error loss function, and a proxy model based on deep integration and active learning is constructed through uniformly weighted mixing; the direct Monte Carlo method is used to estimate the structural failure probability; the learning function values ​​of all sample points in the source domain Monte Carlo sample pool are calculated, new sample points are selected and added to the training set sample points, and the training is iterated until the failure probability convergence criterion is met.

4. The structural reliability updating method based on domain adaptive network integration according to claim 1 is characterized in that: In step 4), based on the upper and lower bounds of the source domain Monte Carlo sample pool and the target domain Monte Carlo sample pool, Latin hypercube sampling method corresponding to uniform distribution is used to sample in the interval, and n migration training set sample points X are selected respectively. s ={x s1 ,x s2 ,x s3 ,...,x sn }, X t ={x t1 ,x t2 ,x t3 ,...,x tn }, where X s Output the corresponding limit state function value through the source domain proxy model, x sn represents the sample points of the source domain migration training set, X s For x sn The collection of x tn represents the sample points of the target domain migration training set, X t For x tn A collection of .

5. The structural reliability updating method based on domain adaptive network integration according to claim 1 is characterized in that: The step 5) specifically includes: Through the constructed source domain proxy model, the network structure of the artificial neural network sub-model is read respectively to complete the construction of M domain adaptive network sub-models; the domain adaptive network sub-model consists of three parts: feature extraction module, adaptive module, and regression module: the feature extraction module is shared by the source domain and target domain networks, and is used to extract the deep features of the source domain and target domain data samples. The input layer and the first hidden layer of the artificial neural network sub-model are selected as feature extractors; the adaptive module introduces the maximum mean difference to calculate the distribution difference between the source domain deep features and the target domain deep features. The calculation expression is as follows: Where, the subscript H represents the reproducing kernel Hilbert space; n s ,n t are the number of samples in the source domain and the target domain respectively; φ(·) represents the nonlinear feature mapping function that maps the original data to the RHKS space; L MMD (D s ,D t ) represents the maximum mean difference between the source domain distribution and the target domain distribution; The regression module consists of a fully connected layer network, which is used to establish the mapping relationship between deep features and predicted response values; The total loss function is a loss function that comprehensively considers the regression prediction error and the difference between domains. The calculation expression is as follows: L=L MSE (X L ,y)+λL MMD (D s ,D t ) Where, L MSE (X L ,y) represents the regression loss of the domain adaptation network sub-model, which represents the difference between the predicted value and the label value of the source domain training set, and uses the MSE loss function; X L is the source domain sample label value, y is the predicted value output by the source domain sample after passing through the entire domain adaptation network sub-model; L MMD (D s ,D t ) represents the distribution difference between the source domain and the target domain data; λ is a constant coefficient; L is the total loss function; After completing the construction and training of the domain adaptive network sub-model, the M domain adaptive network sub-models are uniformly weighted mixed to obtain the domain adaptive network integrated proxy model as the target domain proxy model; the domain adaptive network integrated proxy model outputs the sample point x i The limit state function predicts the mean and standard deviation as follows: In the formula, μ(x i ) is the sample point x i The limit state function prediction mean, which represents the domain adaptive network integration model for the sample point x i The predicted output value of σ(x i ) is the sample point x i The limit state function prediction standard deviation represents the domain adaptive network integration model for the sample point x i The prediction uncertainty of the output value; M is the number of domain adaptation network sub-models; represents the jth domain adaptation network sub-model For the sample point x i The predicted value of the limit state function, whose model parameter is θ j .

6. The structural reliability updating method based on domain adaptive network integration according to claim 1 is characterized in that: In step 6), the sample points of the fine-tuning training set are adaptively selected through the active learning strategy, and the target domain Monte Carlo sample pool S is calculated. MC_t The learning function values ​​U(x) of all sample points in the training set are obtained, and the k sample points with the smallest U(x) values ​​are selected as the sample points T of the fine-tuning training set. t ; The calculation expression of U(x) is: In the formula, μ(x i ) and σ(x i ) represent the sample points x i The predicted mean and predicted standard deviation of the limit state function.

7. The structural reliability updating method based on domain adaptive network integration according to claim 1 is characterized in that: In step 7), the limit state function value G of the sample point of the fine-tuning training set is calculated by calling the limit state function or the finite element model. t , get the fine-tuning training set {T t ,G t }.

8. The structural reliability updating method based on domain adaptive network integration according to claim 1 is characterized in that: The step 8) specifically includes: For each domain adaptive network sub-model in the target domain proxy model, the model input layer and the first hidden layer network parameters are first fixed, and the output layer parameters are retrained based on the fine-tuning training set; then the parameter freeze is unfrozen and all network parameters are fine-tuned. At this time, the model parameters are only slightly adjusted; the estimated value of the structural failure probability is calculated based on the Monte Carlo method The expression is as follows: Where N F is the number of sample points whose limit state function value is less than 0, and N is the sample pool S MC_t The number of sample points.

9. The structural reliability updating method based on domain adaptive network integration according to claim 1 is characterized in that: The convergence criterion expression in step 9) is: in, represents the estimated failure probability of the τth iteration, represents the estimated failure probability of the τ-1th iteration; e c is the threshold; the convergence condition is that three consecutive iterations meet the convergence criterion.

10. The structural reliability updating method based on domain adaptive network integration according to claim 1 is characterized in that: The step 10) specifically includes: Calculate the target domain sample pool S MC_t The learning function values ​​U(x) of all sample points in , and select the minimum value U min The corresponding sample points are added to the fine-tuning training set as new sample points.

Images