A sampling-based finite element model updating method and system for offshore wind power structures

Abstract

The application discloses a kind of based on sampling offshore wind power structure finite element model updating method and system, belong to wind power generation technical field.System includes sensing monitoring module and updating module;Sensing monitoring module is used to collect the data of the bearing position of offshore wind power structure characterized by the finite element model to be updated;Updating module is used to obtain a plurality of modal parameters for finite element model updating based on the data collected by sensing monitoring module;Updating module is equipped with the finite element model to be updated, updating module is based on finite element model and enhanced transition Markov chain Monte Carlo method, the preset prior distribution of modal parameter and several parameters to be updated of finite element model is handled, obtains the posterior distribution of parameter to be updated, and then obtains the posterior mean of each parameter to be updated based on posterior distribution, and updates corresponding parameter in finite element model, completes the updating of finite element model.The present application provides a data basis for the reliability analysis of wind power structure.

Description

Technical Field

[0001] This invention belongs to the field of wind power generation technology, specifically relating to a sampling-based finite element model update method and system for offshore wind power structures. Background Technology

[0002] Offshore wind turbine support structures are subjected to long-term wind, wave, current, and cyclic loads during operation, posing serious challenges to their structural safety and service life. High-fidelity finite element modeling is the physical foundation for accurately sensing and assessing the condition and safety of support structures. By establishing accurate finite element models, their mechanical behavior can be reproduced in virtual space, providing support for understanding the stress distribution and dynamic response of structures under complex loads.

[0003] Due to numerous uncertainties such as manufacturing tolerances, installation deviations, and structural damage, there is an inevitable difference between the initially constructed finite element model and the actual state of the structure. Therefore, dynamic model updates based on field monitoring data become a crucial step in ensuring the effectiveness of the digital twin. By integrating measured vibration, strain, and displacement response data, and continuously calibrating key parameters such as elasticity, stiffness, and damping in the finite element model, the model can gradually approach the actual mechanical state of the structure, thereby supporting early damage identification, accurate assessment of load-bearing capacity, and prediction of remaining life.

[0004] Offshore wind power field measurements are subject to interference from time-varying operating conditions and complex marine environments, including significant measurement noise. Traditional deterministic model update methods directly output point estimates of model parameters. Their update accuracy is severely affected by noise, which may lead to misjudgments of structural health and potentially mask potential damage risks.

[0005] Bayesian Model Update (BMU), as a probabilistic model update framework that can effectively integrate prior parameter knowledge, model physical information and field observation data, can not only provide optimal estimates of parameters, but also quantify the uncertainty of the estimation results. It is a key technology for building reliable digital twins of wind power structures.

[0006] However, applying Bayesian model sampling (BMU) to complex systems such as offshore wind power structures faces two fundamental challenges: First, high-fidelity finite element models typically contain a large number of parameters to be updated, resulting in a high-dimensional parameter space and a complex posterior distribution. Second, in engineering practice, precise prior knowledge of the model parameters is often lacking, forcing the use of broad and sparse prior distributions based on experience. These challenges lead to problems such as low exploration efficiency, long computation time, and poor update accuracy in traditional Bayesian model update sampling methods.

[0007] Therefore, in response to the actual engineering needs of offshore wind power, there is an urgent need for a new sampling method that can break through the limitations of broad prior space and efficiently explore high-dimensional posterior space, so as to provide core support for the dynamic updating of high-precision and high-reliability digital twin models of offshore wind power structures. Summary of the Invention

[0008] To address the problems in the existing technology, this invention provides a sampling-based finite element model update method and system for offshore wind power structures.

[0009] The technical solution of the present invention is as follows:

[0010] In a first aspect, the present invention discloses a sampling-based finite element model update system for offshore wind power structures, including a sensing and monitoring module and an update module;

[0011] The sensing and monitoring module includes multiple strain gauges and multiple vibration sensors. Both strain gauges and vibration sensors are deployed on the load-bearing parts of the offshore wind power structure represented by the finite element model to be updated. The strain gauges are used to collect strain data at their deployment locations in real time, and the vibration sensors are used to collect vibration acceleration at their deployment locations in real time.

[0012] The update module is used to preprocess the data collected by the sensing and monitoring module, and obtain various modal parameters for updating the finite element model based on the preprocessed data. The update module carries the finite element model to be updated. Based on the finite element model and the enhanced transition Markov chain Monte Carlo method, the update module processes the preset prior distribution of the modal parameters and several parameters to be updated in the finite element model to obtain the posterior distribution of the parameters to be updated. Then, based on the posterior distribution, the posterior mean of each parameter to be updated is obtained, and the corresponding parameters in the finite element model are updated to complete the update of the finite element model.

[0013] Secondly, the present invention discloses a sampling-based finite element model update method for offshore wind power structures using the system, comprising the following steps:

[0014] 1) Strain gauges and vibration sensors collect strain data and vibration acceleration at their respective deployment locations in real time. The strain data and vibration acceleration are then preprocessed, and multiple modal parameters for updating the finite element model are obtained based on the preprocessed strain data and vibration acceleration.

[0015] 2) Obtain the values ​​corresponding to each parameter to be updated from the preset prior distribution of the parameters to be updated in the finite element model, and input the values ​​corresponding to each parameter to be updated into the finite element model to obtain the predicted values ​​of the modal parameters corresponding to each parameter to be updated.

[0016] 3) Based on the predicted values ​​of each parameter to be updated and the corresponding modal parameters, obtain the likelihood function values ​​of each parameter to be updated, and then obtain the log probability density function of the guided distribution based on the likelihood function values;

[0017] 4) Extract N samples from the log probability density function, which consist of all the parameters to be updated, to form the initial sample set for iteration; perform iterative transition sampling based on the initial sample set to obtain the effective sample set; then use kernel density estimation to fit the distribution of the effective sample set to obtain the posterior distribution of each parameter to be updated; then obtain the posterior mean of each parameter to be updated based on the posterior distribution, and update the corresponding parameters in the finite element model to complete the update of the finite element model.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0019] (1) By constructing a guide distribution that integrates the local gradient and global distribution information of the seed point, this invention provides the algorithm with a high-quality initial sample set and a better convergence path, fundamentally overcoming the shortcomings of traditional methods in terms of low sample quality and weak convergence ability in a wide prior space.

[0020] (2) This invention employs a dual-loop strategy combined with an early stopping mechanism, which minimizes invalid model calls while ensuring effective exploration times; and uses a parameter grouping update strategy to decompose the high-dimensional global problem into multiple low-dimensional sub-problems, significantly improving the exploration capability in the high-dimensional parameter space. The introduction of these strategies enables the e-TMCMC (enhanced transitional Markov chain Monte Carlo) method to achieve higher posterior estimation accuracy than traditional methods at the same computational cost.

[0021] (3) As an uncertainty model update method under the BMU framework, this invention not only provides the best point estimate of the model parameters, but also provides a complete posterior distribution and confidence interval, thus providing a more solid data foundation for the reliability analysis and fault early warning of wind power structures. Attached Figure Description

[0022] Figure 1 This is a flowchart of the sampling-based finite element model update method for offshore wind power structures according to the present invention;

[0023] Figure 2 This is a flowchart of the e-TMCMC (enhanced Transitional Markov Chain Monte Carlo) method of the present invention;

[0024] Figure 3 This is a schematic diagram of the sampling-based finite element model update system for offshore wind power structures according to the present invention;

[0025] Figure 4 This is a schematic diagram of the model of the catheter stent machine involved in the embodiments of the present invention;

[0026] Figure 5 This is a schematic diagram of the cosine similarity matrix of the parameter sensitivity vector involved in the embodiments of the present invention. Detailed Implementation

[0027] The present invention will be further described and illustrated below with reference to specific embodiments. The embodiments described are merely examples of the content of this disclosure and do not limit the scope of the invention. The technical features of each embodiment in the present invention can be combined accordingly, provided that there is no mutual conflict.

[0028] To fill the gaps in existing technologies and meet the needs of industrial applications, this invention provides a sampling-based finite element model update method and system for offshore wind power structures. Based on the traditional Bayesian model update framework, this method combines the overall distribution of prior samples and local gradient information to construct a guided distribution; it employs a dual-loop sampling mechanism to improve exploration efficiency; and it designs an alternating parameter group update strategy to address high-dimensional challenges, thereby achieving rapid and accurate approximation of the posterior distribution of the parameter target in a high-dimensional, wide prior space.

[0029] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0030] The technical solution of the present invention is as follows:

[0031] like Figure 3 As shown, it is a schematic diagram of the sampling-based finite element model update system for offshore wind power structures of the present invention. The sampling-based finite element model update system for offshore wind power structures of the present invention includes a sensing and monitoring module, a modal feature extractor, a numerical model library, and a model update module; the feature extraction module, the numerical model library, and the model update module constitute the update module.

[0032] The sensing and monitoring module includes a sensor array deployed on the load-bearing parts of the offshore wind power structure represented by the finite element model to be updated, such as the tower, pile foundation, and duct machine. The sensor array is used to collect the dynamic response signals of the load-bearing parts under external loads in real time or periodically. The sensor array includes multiple strain gauges and multiple vibration sensors. The strain gauges are used to collect strain data at their deployment locations in real time, and the vibration sensors are used to collect vibration acceleration at their deployment locations in real time.

[0033] The modal feature extractor connects to the sensing and monitoring module to preprocess the strain and vibration acceleration data acquired by the module. It then extracts various modal parameters for model updates from the preprocessed strain and vibration acceleration data using modal analysis or time-frequency domain analysis methods. These modal parameters include natural frequencies, damping ratios, and elastic moduli. The preprocessing includes denoising and windowing filtering. Specifically, the preprocessing steps involve first denoising the strain and vibration acceleration data, and then applying windowing filtering to the denoised data to obtain the preprocessed strain and vibration acceleration data.

[0034] The numerical model library stores finite element models that need to be updated. The input to the finite element model is the model parameters to be updated (i.e., the parameters to be updated), and the output is the predicted values ​​of all modal parameters corresponding to the model parameters to be updated. Among them, the model parameters to be updated include structural stiffness, elastic modulus, mass distribution and / or damping coefficient, etc.; modal parameters include natural frequencies, mode shapes and damping.

[0035] The numerical model library also sets prior distributions for the parameters of the finite element model to be updated.

[0036] The model update module integrates an enhanced transitional Markov chain Monte Carlo method (e-TMCMC method). The module receives modal parameters from the feature extraction module and the prior distribution of the parameters to be updated from the numerical model library. It then calls the finite element model to be updated from the numerical model library to obtain the predicted values ​​of the modal parameters associated with the parameters to be updated, and calculates the likelihood function with the modal parameters from the feature extraction module. The e-TMCMC method is then executed to obtain an effective sample set composed of the parameters to be updated. Kernel density estimation is then used to fit the distribution of the effective sample set to obtain the posterior distribution of each parameter to be updated. Based on the posterior distribution, the posterior mean of each parameter to be updated is obtained, and the corresponding parameters in the finite element model are updated using the posterior mean, thus completing the update of the finite element model.

[0037] In one embodiment of the present invention, the system further includes a decision support module, which visually displays the model update results and corresponding confidence levels in the form of charts; it performs preliminary identification and early warning of potential anomalies based on the degree to which the posterior distribution of parameters deviates from the historical benchmark, and generates concise status reports and maintenance suggestions to provide a reference for operation and maintenance decisions.

[0038] In this embodiment, the posterior distribution of each model parameter to be updated is obtained, including:

[0039] 1) Obtain the values ​​corresponding to each model parameter to be updated from the preset prior distribution of the model parameters to be updated, and input the values ​​corresponding to each model parameter to be updated into the finite element model to obtain the predicted values ​​of all modal parameters corresponding to each model parameter to be updated;

[0040] 2) Based on the predicted values ​​obtained in step 1) and the modal parameters obtained by the update module, the posterior distribution of each model parameter to be updated is obtained through the enhanced transition Markov chain Monte Carlo method.

[0041] Further, step 2) includes:

[0042] 2.1) First, based on the predicted values ​​of each model parameter to be updated and the corresponding modal parameters, the likelihood function values ​​of each model parameter to be updated are obtained. Then, the likelihood function values ​​are sorted from largest to smallest, and the top-ranked values ​​are extracted. Each likelihood function value corresponds to a model parameter to be updated;

[0043] 2.2) Based on the model parameters to be updated and the corresponding likelihood function values ​​extracted in step 2.1), obtain the Fisher information matrix, then obtain the bandwidth matrix based on the Fisher matrix, and finally obtain the log probability density function of the guided distribution based on the Fisher information matrix, the bandwidth matrix and the model parameters to be updated extracted in step 2.1).

[0044] 2.3) Extract N samples from the log probability density function of step 2.2) to form the initial sample set for iteration; perform iterative transition sampling based on the initial sample set to obtain the effective sample set;

[0045] 2.4) Use kernel density estimation to fit the distribution of the effective sample set to obtain the posterior distribution of each model parameter to be updated.

[0046] Further, in step 2.3), the iterative transition sampling based on the initial iterative sample set to obtain an effective sample set includes:

[0047] 2.3.1) First, the temperature parameters are initialized. Then, the likelihood function and probability density function of each sample in the initial sample set are calculated. Then, the weight of each sample is generated by combining the temperature parameters. Based on the weight of each sample, the samples in the initial sample set are weighted and sampled. Finally, the resampled initial sample set is updated in a double loop and the updated initial sample set is used as the sample set for the next iteration.

[0048] 2.3.2) In the qth iteration, the temperature parameter is first updated with the goal of gradually transitioning to 1 during the iteration process. Then, the weights of each sample in the sample set of the qth iteration are generated based on the updated temperature parameter. Then, the weights of each sample in the sample set are weighted and sampled based on the weights of each sample. Finally, the resampled sample set is updated in a double loop, and the finally updated sample set is used as the sample set for the next iteration.

[0049] 2.3.3) Repeat step 2.3.2) until the temperature parameter transitions to 1, and a valid sample set is obtained.

[0050] Further, in step 2.3.2), the double-loop update of the resampled sample set includes:

[0051] 2.3.2.1) At the mean point of the prior distribution of the model parameters to be updated, a Jacobian matrix for the q-th iteration is generated based on the model parameters to be updated and the modal parameters. The Jacobian matrix is ​​composed of the gradient matrices of each model parameter to be updated. Then, based on the cosine similarity between the gradient matrices of every two model parameters to be updated in the Jacobian matrix, the model parameters to be updated with a cosine similarity greater than 0.5 are divided into the same parameter group, and it is ensured that the cosine similarity between different parameter groups is less than 0.5.

[0052] 2.3.2.2) Based on the MH criterion, update the corresponding model parameters to be updated in each sample of the resampled sample set in turn using the model parameters to be updated in each parameter group to obtain the updated sample set;

[0053] 2.3.2.3) Repeat step 2.3.2.2) until the preset number of rounds is reached to obtain the final updated sample set.

[0054] like Figure 2 As shown, in the model update module, the e-TMCMC method is used to obtain the posterior distribution of the model parameters to be updated. The e-TMCMC method approximates the posterior distribution of the model parameters to be updated by sampling. The e-TMCMC method includes two steps: wizard distribution construction and iterative transition sampling.

[0055] 1) Wizard-based distributed construction

[0056] From the model parameter set to be updated prior distribution Extract N from prior For each of the following model parameters to be updated, calculate the likelihood function value:

[0057]

[0058] in, , Let i be the i-th model parameter to be updated; For the model parameters to be updated In modal parameter set The likelihood function value under the given conditions; For modal parameter set, , This represents the j-th modal parameter; The number of modal parameters; Indicates the first The model parameters to be updated, drawn from the prior distribution, are the inputs to the finite element model. The finite element model predicts the output of the corresponding j-th model parameter. It represents the standard deviation of the likelihood function preset for the j-th model parameter.

[0059] Based on the calculated likelihood function values, the parameters of each model to be updated are sorted in descending order. The top-ranked parameters are then selected. The model parameters to be updated corresponding to each likelihood function value are used as "seed points".

[0060] Next, to further obtain the covariance matrix of the guide distribution, the Fisher information matrix of the "seed point" is first calculated to obtain local curvature information near the "seed point". The Fisher information matrix is ​​a matrix whose number of rows and columns is equal to the number of parameters to be updated in the finite element model. A two-dimensional matrix, where the elements are expressed as:

[0061]

[0062] in, This refers to the element located in the m-th row and n-th column of the Fisher information matrix; Indicates the first Likelihood function values ​​of the model parameters to be updated The logarithm of the m-th model parameter to be updated The partial derivatives; Indicates the first Likelihood function values ​​of the model parameters to be updated The logarithm of the nth model parameter to be updated The partial derivatives of .

[0063] The bandwidth matrix of the covariance matrix is ​​calculated based on the Fisher information matrix. :

[0064]

[0065] in, The eigenvector matrix representing the Fisher information matrix; This represents a scaling matrix where the base scaling factor is on the diagonal and all other values ​​are 0. The number of rows and columns of this matrix are equal to the number of parameters to be updated in the finite element model. The d-th scaling factor The definition of ,in Based on the scaling factor, It is the result of geometrically normalizing the eigenvalues ​​of the Fisher information matrix, calculated as follows:

[0066]

[0067] in, This represents the d-th eigenvalue of the Fisher information matrix. This represents the d-th eigenvalue of the Fisher information matrix. Power of 1.

[0068] Then the covariance matrix of the guided distribution is obtained. for:

[0069]

[0070] in, This is the inverse of the Fisher information matrix; for The covariance matrix of the parameter distribution of the model to be updated is obtained by directly estimating the kernel density (KDE) using conventional methods. The set of model parameters to be updated is obtained by distribution fitting. It is a balancing factor used to adjust the weights of local gradient information and overall distribution information. .

[0071] Finally, the log probability density function of the guided distribution is obtained. The expression is:

[0072]

[0073] in, Indicates the set of model parameters to be updated Compared to The first of the model parameters to be updated One model parameter to be updated The Mahalanobis distance is calculated as follows:

[0074] .

[0075] 2) Iterative transition sampling

[0076] Log probability density function from the guide distribution Extraction An initial sample set is constructed from samples consisting of all the model parameters to be updated. , , Represents the initial sample set The first in The initial sample set will participate in subsequent iterative sampling, and will successively approximate a series of transition distributions during the iteration process. , ,in, This represents the transition distribution in the q-th iteration; This indicates the use of the model parameter set to be updated. and modal parameter set The calculated likelihood function Power; The log probability density function of the guided distribution. Power; This represents the temperature parameter, which gradually transitions from 0 to 1 during the iteration process.

[0077] In the q-th iteration, the e-TMCMC method performs the following steps sequentially:

[0078] Step 1: Update the temperature parameters. First, update the temperature parameters. The update target is: to find , so that:

[0079]

[0080] in, Represents the coefficient of variation. Denotes the sample set of the q-th iteration. The Middle The weights of each sample, ,in, Indicates the first Sample In modal parameter set The likelihood function value under the given conditions. Indicates the first Sample The probability density function of the guide distribution.

[0081] Step 2: Normalize the weights of each sample in the current stage:

[0082]

[0083] in, Indicates the first The normalized weights of each sample.

[0084] The covariance matrix of the update process is calculated based on the normalized weights:

[0085]

[0086] in, This represents the covariance matrix during the q-stage sample update process, with 1 / 2 rows and 1 / 2 columns. ; This is the scaling factor; This represents the weighted mean of all samples in the q-th iteration. .

[0087] Step 3: Based on For the sample set of the qth iteration Each sample in the dataset is weighted and resampled to obtain a resampled set with the same number of points. , , , For the first Each sample is resampled. This process has a greater... Points with smaller values ​​will have a higher probability of being sampled multiple times, while those with smaller values ​​will have a higher probability of being sampled multiple times. The points that are selected will have a higher probability of not being selected.

[0088] Step 4: [Regarding...] The samples in the loop are updated using a double loop. In each outer loop, a total of [number] updates are performed. This is a total update, and within each total update, the most frequent operations are performed on different parameter groups sequentially. The inner loop update is described in Step 4.1, and the parameter group partitioning rules are described in Step 4.2. The inner loop update process is executed independently on each parameter group, ensuring that each parameter group completes one valid update. The outer loop controls the overall number of rounds of the entire parameter update process, and each round of the outer loop includes one inner loop update for all parameter groups. By increasing the number of outer loop rounds, the number of valid updates for each parameter group can be increased, thereby improving the performance of parameter space exploration, but this will also result in more computation time.

[0089] Step 4.1: In the inner loop, determine whether to accept a new sample using the Metropolis–Hastings (MH) criterion:

[0090]

[0091] in, express and The smaller one, This indicates the q-th resampling result. Sample The corresponding transition distribution, Indicates Centered on the covariance matrix during the q-stage sample update process, Candidate samples obtained by moving by step size The corresponding transitional distribution; The range is The uniform distribution.

[0092] The inner loop of each sample executes at most once. Next, if it reaches A valid update was completed before the next inner loop, meaning the candidate sample was successfully accepted. If the current sample's inner loop is terminated immediately, the accepted candidate samples are used as the output of the current sample's inner loop, and the sample set of the q-th iteration is then processed based on this. The inner loop executes on all subsequent samples.

[0093] Step 4.2: In each outer loop update, the e-TMCMC method does not directly update all... Instead of updating the parameters of the individual finite element models, an alternating inner loop update is used. To achieve sample partitioning, the Jacobian matrix for the update process is first calculated near the prior Gaussian distribution mean point of the parameters to be updated.

[0094]

[0095] in, The number of rows equals the number of modal parameters. The number of columns equals the number of model parameters to be updated. A two-dimensional matrix, For the model parameters to be updated The corresponding gradient matrix, , This indicates that the modal parameters are near the mean point of the prior Gaussian distribution of the model parameters to be updated. To update model parameters The partial derivatives of .

[0096] To identify the correlation between different model parameters to be updated during the finite element model update process, the cosine similarity between the gradient matrices of each model parameter to be updated is calculated:

[0097]

[0098] in, Indicates the model parameters to be updated and model parameters to be updated The cosine similarity between the gradient matrices; Represents modal parameters To update model parameters The partial derivatives; Represents modal parameters To update model parameters The partial derivatives of .

[0099] Based on the cosine similarity calculation results, the parameters of the model to be updated that have a cosine similarity greater than 0.5 are grouped into the same group, and it is ensured that the cosine similarity between different parameter groups is less than 0.5.

[0100] When temperature parameter After updating to 1, the e-TMCMC method executes the remaining iteration steps of the current stage, terminates the iteration after completion, and obtains a valid sample set.

[0101] Then, using standard KDE, the distribution of each parameter to be updated in the model is fitted to the effective sample set, thus obtaining the posterior distribution of each parameter. The mean of the fitted parameters is then used as the posterior estimate mean of the Bayesian model update, and the standard deviation of the fitted parameters is used as the posterior standard deviation of the Bayesian model update. The posterior estimate mean of each parameter to be updated is input into the finite element model, and the corresponding parameters in the finite element model are updated using the posterior estimate mean, finally obtaining the updated finite element model, thus completing the finite element model update.

[0102] In one embodiment of the present invention, the present invention also provides a sampling-based finite element model update method for offshore wind power structures, such as... Figure 1 As shown, the method includes the following steps:

[0103] 1) Strain gauges and vibration sensors collect strain data and vibration acceleration at their respective deployment locations in real time. The strain data and vibration acceleration are then preprocessed, and multiple modal parameters for updating the finite element model are obtained based on the preprocessed strain data and vibration acceleration.

[0104] 2) Obtain the values ​​corresponding to each parameter to be updated from the preset prior distribution of the parameters to be updated in the finite element model, and input the values ​​corresponding to each parameter to be updated into the finite element model to obtain the predicted values ​​of all modal parameters corresponding to each parameter to be updated.

[0105] 3) Based on the predicted values ​​of each parameter to be updated and the corresponding modal parameters, obtain the likelihood function values ​​of each parameter to be updated. Then, sort the likelihood function values ​​from largest to smallest and extract the top-ranked values. The parameters to be updated correspond to each likelihood function value; then, based on the extracted model parameters to be updated and the corresponding likelihood function values, the log probability density function of the guided distribution is obtained.

[0106] 4) Draw N samples from the logarithmic probability density function, consisting of all the parameters of the model to be updated, to form the initial sample set for iteration; perform iterative transition sampling based on the initial sample set to obtain the effective sample set; then use kernel density estimation to fit the distribution of the effective sample set to obtain the posterior distribution of each parameter of the model to be updated; then obtain the posterior mean of each parameter of the model to be updated based on the posterior distribution, and update the corresponding parameters in the finite element model to complete the update of the finite element model. Example: Soil-structure interaction (SSI) parameter update of a 10 MW jacket-type offshore wind turbine.

[0107] 1. Problem Definition

[0108] Observational data and modal parameters: This case study was conducted on a 1:100 scale model of a 10 MW jacket-type offshore wind turbine. The observational data consisted of vibration accelerations collected by vibration sensors on the turbine. The modal parameters used for updating the finite element model included the first three structural modal frequencies obtained from modal analysis of the vibration accelerations. Damping ratio There are a total of 6 indicators.

[0109] Numerical Model and Parameter Priors: A finite element model of the jacket-type unit was established at a scaled-down scale. The py curve method was used to uniformly construct three equivalent horizontal soil springs at three different depths (top, middle, and bottom) on the support piles to simulate the SSI effect of the unit. The stiffness of the three springs was... and damping As parameters to be updated in the model, the mean of the prior (Gaussian) distribution of the six parameters is the initial value obtained through the Py curve formula, and all are normalized to 1. The values ​​of the standard deviation of the prior Gaussian distribution are shown in the table below:

[0110]

[0111] Algorithm Setup: To determine parameter grouping during the update process, the Jacobian matrix at the prior center point is calculated, and the cosine similarity between the gradient matrices corresponding to each parameter of the model to be updated is calculated to identify the coupling relationship between the parameters of the model to be updated. Figure 5 As shown. The analysis results indicate that the stiffness parameter The similarity between the gradient matrices remains at a high level, while the damping parameters... The similarity between the gradient matrices of the stiffness and damping parameters remained at a high level, while the similarity between the gradient matrices of the stiffness and damping parameters was low. Therefore, the six model parameters to be updated were divided into two groups: Other key parameter settings include: the number of samples participating in the iteration. Guided distribution balance factor Covariance scaling factor during the update process Maximum number of inner loop iterations Number of outer loops Draw from the prior distribution From the given sample points, select the ones with the highest likelihood. A sample is used as a seed point to construct a guide distribution.

[0112] Results Analysis and Comparison: To verify the effectiveness of the method, a set of "true values" were artificially set for each parameter of the finite element model to be updated. The finite element model output under the "true values" was used as the index for model updating. The updating results of the e-TMCMC method and the traditional TMCMC method are shown in the table below, where μ post (error) represents the posterior mean and the corresponding estimation error, σ post This represents the posterior standard deviation. From this, we can see that compared to the traditional TMCMC method, the e-TMCMC method has a significant improvement in parameter estimation accuracy. The e-TMCMC method reduces the estimation error by more than 60% for parameters that are difficult to identify. Meanwhile, the e-TMCMC method performs roughly at the same level as the TMCMC method in terms of posterior standard deviation estimation, indicating that it maintains a reliable level of uncertainty quantification while improving estimation accuracy. This result confirms the superiority of the e-TMCMC method in the current task.

[0113]

[0114] The above-described embodiments are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. Those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A sampling-based offshore wind structure finite element model updating system, characterized in that, The method comprises a sensing monitoring module and an updating module. The sensing monitoring module comprises a plurality of strain gauges and a plurality of vibration sensors, the strain gauges and the vibration sensors are arranged at load-bearing positions of an offshore wind power structure represented by the finite element model to be updated, the strain gauges are used to collect strain data at the positions where the strain gauges are arranged, and the vibration sensors are used to collect vibration acceleration at the positions where the vibration sensors are arranged. The updating module is used to preprocess the data collected by the sensing monitoring module, and obtain a plurality of modal parameters for updating the finite element model based on the preprocessed data. The updating module is loaded with the finite element model to be updated, and the updating module processes the preset prior distribution of the modal parameters and a plurality of parameters to be updated of the finite element model based on the finite element model and an enhanced transition Markov chain Monte Carlo method, obtains the posterior distribution of the parameters to be updated, obtains the posterior mean of each parameter to be updated based on the posterior distribution, and updates the corresponding parameters in the finite element model to complete the updating of the finite element model. The method comprises the following steps: 1) obtaining the values of the parameters to be updated from the preset prior distribution of the parameters to be updated, and inputting the values of the parameters to be updated into the finite element model to obtain the predicted values of all the modal parameters corresponding to the parameters to be updated; 2) obtaining the posterior distribution of the parameters to be updated based on the predicted values obtained in step 1) and the modal parameters obtained by the updating module through the enhanced transition Markov chain Monte Carlo method; Step 2) comprises the following steps: 2.1) first, based on the predicted value of each to-be-updated parameter and the corresponding modal parameter, obtain the likelihood function value of each to-be-updated parameter, then sort the likelihood function values from large to small, and extract the to-be-updated parameters corresponding to the first K likelihood function values respectively; likelihood function values 2.2) obtaining a Fisher information matrix based on the parameters to be updated and the corresponding likelihood function values extracted in step 2.1), obtaining a bandwidth matrix based on the Fisher matrix, and obtaining a log probability density function of a guide distribution based on the Fisher information matrix, the bandwidth matrix and the parameters to be updated extracted in step 2.1); 2.3) extracting N samples composed of all the parameters to be updated from the log probability density function in step 2.2) to form an initial iteration sample set, and performing iterative transition sampling based on the initial iteration sample set to obtain an effective sample set; 2.4) fitting the distribution of the effective sample set by kernel density estimation to obtain the posterior distribution of each parameter to be updated.

2. The sampled-based offshore wind structure finite element model updating system according to claim 1, wherein, The load-bearing positions comprise a tower, a pile foundation and a guide pipe of the offshore wind power structure.

3. The sampled-based offshore wind structure finite element model updating system according to claim 1, wherein, The preprocessing comprises denoising and window filtering, and the plurality of modal parameters for updating the finite element model are obtained by processing the preprocessed data through a modal analysis or a time-frequency domain analysis method.

4. The sampled-based offshore wind structure finite element model updating system according to claim 1, wherein, The modal parameters comprise natural frequency, modal shape and damping, and the parameters to be updated of the finite element model comprise structural stiffness, elastic modulus, mass distribution and / or damping coefficient.

5. The sampled-based offshore wind structure finite element model updating system according to claim 1, wherein, In step 2.3), the iterative transition sampling based on the initial iteration sample set to obtain the effective sample set comprises the following steps: 2.3.1) first initializing a temperature parameter, then calculating the likelihood function values and the probability density functions of the samples in the initial iteration sample set, combining the temperature parameter to generate the weights of the samples, performing weighted resampling on the samples in the initial iteration sample set based on the weights of the samples, and finally performing double-loop updating on the resampled initial iteration sample set, and taking the updated initial iteration sample set as the sample set in the next iteration. 2.3.2)In the qth iteration, first update the temperature parameter as the target of gradually transitioning to 1 in the iteration process, then generate the weight of each sample in the sample set of the qth iteration based on the updated temperature parameter, and then weight sample set based on the weight of each sample, and finally double-loop update the resampled sample set, and use the final updated sample set as the sample set in the next iteration; 2.3.3)Repeat step 2.3.2)until the temperature parameter transitions to 1, and obtain the effective sample set.

6. The system for updating a finite element model of an offshore wind structure based on sampling of claim 5, wherein, In step 2.3.2),the double-loop updating of the resampled sample set comprises: 2.3.2.1)At the mean point of the prior distribution of the parameters to be updated, generate the Jacobian matrix of the qth iteration process based on the parameters to be updated and the modal parameters, and the Jacobian matrix is composed of the gradient matrix of each parameter to be updated, and then based on the cosine similarity between the gradient matrix of each parameter to be updated in the Jacobian matrix, divide the part of the parameters to be updated with a cosine similarity greater than 0.5 into the same parameter group, and ensure that the cosine similarity between different parameter groups is less than 0.5; 2.3.2.2)Update the corresponding parameters to be updated in each sample of the resampled sample set based on the MH criterion in each parameter group, and obtain the updated sample set; 2.3.2.3)Repeat step 2.3.2.2)until a preset number of rounds is reached, and obtain the final updated sample set.

7. A method for updating a finite element model of a sampled-based offshore wind structure using the system of claim 1, wherein, Comprising the following steps: 1)The strain gauge and the vibration sensor respectively collect strain data and vibration acceleration at their respective deployment locations in real time, and then preprocess the strain data and vibration acceleration, and obtain a plurality of modal parameters for finite element model updating based on the preprocessed strain data and vibration acceleration; 2)Obtain the value corresponding to each parameter to be updated from the preset prior distribution of the parameters to be updated in the finite element model, and input the value corresponding to each parameter to be updated into the finite element model to obtain the predicted value of the modal parameter corresponding to each parameter to be updated; 3)Obtain the likelihood function value of each parameter to be updated based on the predicted value of each parameter to be updated and the corresponding modal parameter, and then obtain the log probability density function of the guide distribution based on the likelihood function value; 4)Extract N samples composed of all parameters to be updated from the log probability density function to form an initial iteration sample set; Based on the initial iteration sample set, perform iterative transition sampling to obtain an effective sample set, then use kernel density estimation to fit the distribution of the effective sample set to obtain the posterior distribution of each parameter to be updated, then obtain the posterior mean of each parameter to be updated based on the posterior distribution, and update the corresponding parameter in the finite element model to complete the updating of the finite element model.

Images