An offshore wind turbine structure modal parameter automatic identification method

Abstract

The application provides an offshore wind turbine structure modal parameter automatic identification method, comprising: power spectrum density analysis on measured structural dynamic response of an offshore wind turbine structure, identification of harmonic frequencies by using a long short-term memory (LSTM) deep learning network; obtaining the dynamic response of the offshore wind turbine structure only with environmental excitation by using a periodic sub-signal-Kalman filter; then obtaining a Monte Carlo stability diagram by using a Monte Carlo covariance-driven stochastic subspace identification method, and further identifying stable modes of the structure; finally, obtaining modal parameters of the offshore wind turbine structure by using a density-based noise application spatial clustering method. The application can effectively solve the challenge of harmonic excitation to modal parameter identification, improve the accuracy of structural modal parameters, and improve the structural health monitoring, state evaluation and damage identification capability of the offshore wind turbine, thereby supporting the development of offshore wind energy.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of structural modal parameter identification, and relates to a modal parameter identification method, in particular to an offshore wind turbine structure modal parameter automatic identification method. BACKGROUND

[0002] With the continuous transformation of global energy structure, offshore wind turbines, as a clean and renewable energy, are gradually receiving high attention from countries around the world. Due to the special environment of offshore wind turbines (including high wind speed, large waves, marine corrosion, etc.), it is particularly important to ensure the safety of its structure and long-term stable operation.

[0003] Under this background, the accurate identification of the dynamic characteristics of offshore wind turbines, especially the modal parameters (such as natural frequency, damping ratio and mode shape), has become a key link to ensure the structural health, predict fatigue damage, optimize design and improve operation efficiency. The modal identification of this structure can effectively help engineers to timely grasp the working state of the wind turbine, provide scientific basis for subsequent maintenance and maintenance, and provide data support for design optimization, and ultimately improve the economic efficiency and safety performance of the entire wind farm.

[0004] However, the modal parameter identification of offshore wind turbines still faces many challenges. Traditional modal identification methods, such as experimental modal analysis and modal identification based on vibration data, have achieved certain results in some applications, but due to the complex working environment of offshore wind turbines, traditional methods usually rely on a large amount of experimental data, expensive equipment and long operation time, and the accuracy of the identification results is often affected by environmental noise and data quality.

[0005] In summary, the current offshore wind turbine modal parameter identification method has the problem of low precision due to high degree of human intervention, which cannot effectively guide engineering practice. SUMMARY

[0006] The purpose of the present application is to provide an offshore wind turbine structure modal parameter automatic identification method, which solves the problem of low precision of modal parameter identification due to high degree of human intervention in the existing offshore wind turbine modal parameter identification technology, which cannot effectively guide engineering practice.

[0007] In a first aspect, the application provides a method for automatically identifying modal parameters of an offshore wind turbine structure, comprising the following steps: obtaining dynamic response data of the offshore wind turbine structure in a target area; performing power spectral density analysis based on the dynamic response data of the offshore wind turbine structure to identify harmonic frequencies; filtering the harmonic frequencies to obtain dynamic response data of the offshore wind turbine structure under only environmental excitation; using a stochastic subspace algorithm to identify the dynamic response data of the offshore wind turbine structure under only environmental excitation to obtain a Monte Carlo stability diagram and stable modal parameters of the offshore wind turbine structure; clustering based on the stable modal parameters of the offshore wind turbine structure to obtain final modal parameters of the offshore wind turbine structure, thereby achieving automatic identification of the modal parameters of the offshore wind turbine structure.

[0008] In an implementation form of the first aspect, the power spectral density analysis based on the dynamic response data of the offshore wind turbine structure to identify harmonic frequencies comprises: pre-processing the dynamic response data of the offshore wind turbine structure to obtain time series segments of the dynamic response data; performing Fourier transform based on each time series segment and calculating power spectral density to obtain a relationship curve between power spectral density values and frequency values of the dynamic response data; extracting features from the relationship curve between power spectral density values and frequency values to obtain key features; the key features include harmonic frequencies, spectral peak frequencies, and power spectral density amplitudes; constructing an LSTM model based on the key features; obtaining a training set and a test set of the time series segments of the dynamic response data, and training and optimizing the LSTM model to obtain a trained LSTM model; identifying harmonic frequencies through the trained LSTM model.

[0009] In an implementation form of the first aspect, the dynamic response data of the offshore wind turbine structure includes dynamic response data of the offshore wind turbine structure under periodic excitation and dynamic response data of the offshore wind turbine structure under only environmental excitation; filtering the harmonic frequencies to obtain dynamic response data of the offshore wind turbine structure under only environmental excitation comprises: constructing a dynamic model of the offshore wind turbine structure under periodic excitation based on the harmonic frequencies; setting a periodic sub-signal Kalman filter based on the dynamic model; initializing the periodic sub-signal Kalman filter based on the harmonic frequencies and the dynamic response data of the offshore wind turbine structure; inputting the dynamic response data of the offshore wind turbine structure into the periodic sub-signal Kalman filter for filtering to obtain a periodic sub-signal corresponding to the harmonic frequencies; removing the periodic sub-signal from the dynamic response data of the offshore wind turbine structure to obtain dynamic response data of the offshore wind turbine structure under only environmental excitation.

[0010] In an implementation form of the first aspect, the random subspace algorithm is adopted to identify the offshore wind turbine structure dynamic response data under the only environmental excitation, to obtain the Monte Carlo stability diagram and the stable modal of the offshore wind turbine structure, comprising: determining the value range of the control parameters based on the offshore wind turbine structure dynamic response data under the only environmental excitation; the control parameters include: the fundamental frequency, the modal number, the time delay parameter, and the system order; a plurality of groups of random control parameters are simulated and generated by using the Monte Carlo algorithm; the random subspace identification algorithm based on the covariance driving is adopted to identify the random control parameters, to obtain a plurality of groups of candidate modes; the two-stage stability checking method is adopted to identify the stable modal parameters based on a plurality of groups of the candidate modes; and the Monte Carlo stability diagram is established based on the stable modal parameters.

[0011] In an implementation form of the first aspect, the value range of the control parameters includes: the value range of the time delay parameter and the value range of the system order; the calculation formula of the value range of the time delay parameter is:

[0012]

[0013] τ max = 2 x τ min

[0014] The calculation formula of the value range of the system order is:

[0015] N min = 2 x n m

[0016] N max = 2 x N min

[0017] Wherein, τ min represents the minimum value of the time delay parameter; τ max represents the maximum value of the time delay parameter; f s represents the sampling frequency value; f f represents the basic natural frequency of the structure obtained automatically from the power spectral density diagram by using the scale space peak picking technology; N min represents the minimum value of the system order; N max represents the maximum value of the system order; n m represents the modal number.

[0018] In an implementation form of the first aspect, the two-stage stability checking method includes: the first-stage preliminary stability checking and the second-stage comprehensive stability checking; the first-stage preliminary stability checking is to compare each candidate mode with other candidate modes, to screen out the modes repeatedly appearing under different parameter combinations, and mark them as possible stable modes; the calculation formula adopted is:

[0019]

[0020] 1-MAC(φ a ,φ b )≤1%

[0021] wherein a, b represent two different modes respectively; f represents the natural frequency; ξ represents the damping ratio; φ represents the mode shape; Δf represents the relative natural frequency difference; Δξ represents the relative damping ratio difference; and MAC represents the modal assurance criterion between the mode shapes of two different modes.

[0022] In an implementation form of the first aspect, the second stage comprehensive stability check comprises: classifying parameters of the possible stable modes by using an algorithm to identify stable mode parameters; and selecting candidate modes with a stability index greater than 0.3 s as the stable result.

[0023] In an implementation form of the first aspect, the clustering based on the stable mode parameters of the offshore wind turbine structure to obtain the final offshore wind turbine structure mode parameters comprises: obtaining a stable mode parameter data set and a number of candidate modes based on the Monte Carlo stability diagram; calculating a value range of a minimum cluster size and calculating an initial value of the calculated minimum cluster size MinPts based on the number of candidate modes; performing optimal selection based on the initial value of the MinPts to obtain an optimal setting under the current MinPts; performing clustering based on the optimal setting under the current MinPts to generate a clustering cluster and calculate a silhouette coefficient of candidate modes in each clustering cluster; sequentially traversing all MinPts to perform an optimal parameter iteration to obtain a silhouette coefficient sum; and analyzing and optimizing the silhouette coefficient sum to obtain the final offshore wind turbine structure mode parameters.

[0024] In an implementation form of the first aspect, the calculation of the value range of the minimum cluster size and the calculation of the initial value of the calculated minimum cluster size MinPts based on the number of candidate modes comprises: calculating the initial value of the minimum cluster size MinPts based on the number of candidate modes; for the value of the current MinPts, calculating distances between data points in each stable mode parameter data set obtained based on the Monte Carlo stability diagram and their first MinPts-1 neighbors, and generating a K-distance diagram; connecting a first point and a last point of the K-distance diagram to form a first straight line; finding a point farthest from the first straight line and perpendicular to the straight line on the K-distance diagram as an elbow point; and the distance value corresponding to the elbow point is an optimal setting value of the maximum clustering range.

[0025] In an implementation form of the first aspect, the calculation formula of the silhouette coefficient is:

[0026]

[0027] wherein Sil(p) represents a profile coefficient; b(p) represents a minimum value of the average distance of p point to other clusters; and av(p) represents an average value of the distance of p point to other points in the same cluster.

[0028] As described above, the offshore wind turbine structure modal parameter automatic identification method has the following beneficial effects:

[0029] (1) The offshore wind turbine structure modal parameter automatic identification method provided by the present application can effectively eliminate the influence of harmonic excitation on offshore wind turbine modal parameter identification. On the basis of using power spectral density analysis in the process of measuring the dynamic response of offshore wind turbine structure, the LSTM deep learning network is used to automatically identify the harmonic frequency; then, based on the identified harmonic frequency, the periodic sub-signal Kalman filter is used to obtain the dynamic response of offshore wind turbine structure under only environmental excitation; and then the Monte Carlo covariance driven stochastic subspace identification method is used to automatically identify the modal parameters of offshore wind turbine structure.

[0030] (2) The method provided by the present application can consider the influence of system order parameters and time delay parameters on the final modal identification result by using the Monte Carlo covariance driven stochastic subspace method, thereby improving the accuracy of the identification result. In addition, the deep learning algorithm can also be used to improve the automation degree and identification efficiency of the identification process.

[0031] (3) In the present application, the effectiveness and accuracy of the method are verified by numerical simulation and field measurement, which can provide an effective tool for modal parameter identification of offshore wind turbine structure under operating conditions, thereby improving the structural health monitoring, state evaluation and damage identification ability of offshore wind turbine. BRIEF DESCRIPTION OF DRAWINGS

[0032] Figure 1 A schematic diagram of the hardware application scene in an embodiment of the offshore wind turbine structure modal parameter automatic identification method described in the present application is shown.

[0033] Figure 2 A flowchart of the offshore wind turbine structure modal parameter automatic identification method described in the present application is shown.

[0034] Figure 3 A flowchart of S2 in the offshore wind turbine structure modal parameter automatic identification method described in the present application is shown.

[0035] Figure 4 A schematic diagram of the LSTM deep learning model in the present application is shown.

[0036] Figure 5 A schematic diagram of the offshore wind turbine and its dynamic response time history in the present application is shown.

[0037] Figure 6 S3 flowchart of the offshore wind turbine structure modal parameter automatic identification method described in the present application is shown.

[0038] Figure 7 The offshore wind turbine structure acceleration and power spectral density analysis chart of the present application is shown.

[0039] Figure 8 S4 flowchart of the offshore wind turbine structure modal parameter automatic identification method described in the present application is shown.

[0040] Figure 9 The structure response modal identification stability chart of the offshore wind turbine structure modal parameter automatic identification method described in the present application is shown.

[0041] Figure 10 S5 flowchart of the offshore wind turbine structure modal parameter automatic identification method described in the present application is shown.

[0042] Figure 11 The offshore wind turbine stable candidate modal damping ratio-frequency relationship chart of the present application is shown.

[0043] Element number explanation

[0044] 11 Frequency analysis module

[0045] 12 Data processing module

[0046] 13 Spatial identification module

[0047] 14 Spatial clustering module DETAILED DESCRIPTION

[0048] The implementation of the present application is described below through specific, concrete examples, and other advantages and effects of the present application can be easily understood by those skilled in the art from the disclosure of the present specification. The present application can also be implemented or applied through other different specific implementations, and each detail in the present specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that the following examples and features in the examples can be combined with each other without conflict.

[0049] It should be noted that the diagrams provided in the following examples only illustrate the basic concept of the present application in a schematic manner, and only the components related to the present application are shown in the diagrams, not the number, shape and size of the components when actually implemented. The actual implementation of each component can be a random change in type, number and proportion, and the component layout type can also be more complex.

[0050] The following embodiments of the application provide an offshore wind turbine structure modal parameter automatic identification method, which solves the problems of low precision of modal parameter identification and inability to effectively guide engineering practice due to excessive human intervention in existing offshore wind turbine modal parameter identification technology.

[0051] As shown in Figure 1 The hardware application scene diagram of the offshore wind turbine structure modal parameter automatic identification method in an embodiment of the application specifically includes a frequency analysis module 11, a data processing module 12, a spatial identification module 13, and a spatial clustering module 14. Specifically, in the frequency analysis module 11, power spectral density analysis is performed on the offshore wind turbine structure dynamic response, and the LSTM (Long Short-Term Memory) deep learning network is used to identify the harmonic frequency; through the data processing module 12, the identified harmonic frequency is used to obtain the dynamic response of the offshore wind turbine structure under only environmental excitation by using the periodic sub-signal Kalman filter; in the spatial identification module 13, the Monte Carlo covariance-driven random subspace identification method is applied to the structure dynamic response under only environmental excitation to obtain the Monte Carlo stability diagram, thereby identifying the stable modal of the structure; in the spatial clustering module 14, the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) method is used to automatically obtain the modal parameters of the offshore wind turbine structure.

[0052] The offshore wind turbine structure modal parameter automatic identification method provided in the embodiments of the application will be described in detail below with reference to the accompanying drawings in the embodiments of the application.

[0053] Please refer to Figure 2 , which shows the flowchart of the offshore wind turbine structure modal parameter automatic identification method in an embodiment of the application. As shown in Figure 2 The embodiment provides an offshore wind turbine structure modal parameter automatic identification method.

[0054] The offshore wind turbine structure modal parameter automatic identification method specifically includes the following steps:

[0055] S1, obtaining offshore wind turbine structure dynamic response data in a target area.

[0056] In this embodiment, first, the specific position and range of the offshore wind turbine for obtaining the dynamic response data are determined. Then, according to the monitoring requirements, appropriate sensors and monitoring devices (such as acceleration sensors, displacement sensors, force sensors, etc.) are selected to collect the structural dynamic response data of the wind turbine.

[0057] Then, the selected sensors and monitoring devices are installed on the offshore wind turbine in the target area to ensure that the devices can stably and accurately collect data.

[0058] Finally, according to the monitoring requirements and data analysis requirements, set appropriate sampling frequency, sampling duration and other parameters; start the monitoring equipment installed on the fan, start collecting the structural dynamic response data of the fan; at the same time, record the wind speed, wind direction, wave and other environmental parameters in the target area, so as to facilitate subsequent analysis.

[0059] S2, based on the offshore wind turbine structural dynamic response data, power spectrum density analysis is carried out, and the harmonic frequency is identified. Please refer to Figure 3 , which is the S2 process diagram of the offshore wind turbine structural modal parameter automatic identification method described in the present application. As Figure 3 shown, the S2 comprises the following steps:

[0060] S21, pre-processing the offshore wind turbine structural dynamic response data to obtain time series segments of dynamic response data;

[0061] S22, Fourier transform is carried out based on each time series segment, and the power spectrum density is calculated to obtain the relationship curve of power spectrum density value and frequency value of dynamic response data;

[0062] S23, feature extraction is carried out on the relationship curve of power spectrum density value and frequency value to obtain key features; the key features include harmonic frequency, spectral peak frequency and power spectrum density amplitude;

[0063] S24, constructing an LSTM model based on the key features;

[0064] S25, obtaining the training set and test set of the time series segments of the dynamic response data, and training and optimizing the LSTM model to obtain the trained LSTM model;

[0065] S26, identifying the harmonic frequency through the trained LSTM model.

[0066] Please refer to Figure 4 and Figure 5 .

[0067] In this embodiment, the power spectrum density analysis of the measured structural dynamic response of offshore wind turbine structure is carried out, and the LSTM deep learning model is used to identify the harmonic frequency. In order to realize efficient automatic identification, a deep learning framework is developed. According to the framework, the whole process is divided into six layers, a large number of samples are obtained based on field measurement and numerical simulation data, the LSTM deep learning network is trained, the difference of predicted harmonic frequency is small, and the subsequent automatic structural analysis is ensured.

[0068] The developed deep learning framework consists of 6 layers: input layer, convolutional neural network layer, long short-term memory network layer, attention layer, fully connected layer and output layer.

[0069] The input layer represents a power spectral density curve of the structural response under harmonic and random excitations, and the output layer represents a predicted harmonic frequency.

[0070] Specifically, it is preferred to develop an LSTM-based deep learning framework that can automatically identify and predict the harmonic frequency in the structural dynamic response of offshore wind turbines, with a small prediction difference to ensure the accuracy of subsequent automated structural analysis. That is, the deep learning framework includes an input layer, a convolutional neural network layer, a long short-term memory network layer, an attention layer, a fully connected layer, and an output layer. The input layer receives the original data of the structural dynamic response of offshore wind turbines, such as time series data of acceleration, displacement, etc. The convolutional neural network layer extracts features of the data, such as frequency components and power distribution, using power spectral density analysis methods. The long short-term memory network layer, as the core of the deep learning model, is used to capture long-term dependencies in time series data and identify harmonic frequencies. The fully connected layer maps the output of the LSTM layer to the final predicted value, i.e., the harmonic frequency. Finally, the output is the predicted harmonic frequency.

[0071] For example, the measured acceleration data of a certain offshore wind turbine located in the Yangjiang wind farm in China can be analyzed. That is, acceleration sensors are installed at three different heights (e.g., 1#, 2#, 3#) on the offshore wind turbine structure to obtain dynamic responses and perform power spectral density analysis. Using the LSTM deep learning model, the power spectral density analysis graph is obtained, and the identified harmonic frequency can prove the effectiveness of the deep learning model.

[0072] S3, filtering the harmonic frequency to obtain offshore wind turbine structural dynamic response data only under environmental excitation. Please refer to Figure 6 and Figure 7 , respectively showing the S3 process schematic diagram in the offshore wind turbine structural modal parameter automatic identification method described in the present application and the offshore wind turbine structural acceleration and power spectral density analysis graph. As shown in Figure 6 and Figure 7 , the S3 includes the following steps:

[0073] By determining the harmonic frequency, a periodic sub-signal Kalman filter is used to obtain the dynamic response of the offshore wind turbine structure only under environmental excitation. The offshore wind turbine structural dynamic response data includes the offshore wind turbine structural response under periodic excitation and the structural response data only under environmental excitation.

[0074] S31, constructing a dynamic model of the offshore wind turbine structure under periodic excitation according to the harmonic frequency;

[0075] S32, setting a periodic sub-signal Kalman filter based on the dynamic model;

[0076] S33, initializing the periodic sub-signal Kalman filter according to the harmonic frequency and the offshore wind turbine structure dynamic response data;

[0077] S34, inputting the offshore wind turbine structure dynamic response data into the periodic sub-signal Kalman filter for filtering to obtain a periodic sub-signal corresponding to the harmonic frequency;

[0078] S35, removing the periodic sub-signal from the offshore wind turbine structure dynamic response data to obtain offshore wind turbine structure dynamic response data under only environmental excitation.

[0079] In the embodiment, the state of the system is estimated from the total structure response under mixed periodic and random excitation by the Kalman filter; the harmonic state is identified according to the obtained harmonic frequency, and the periodic sub-signal is constructed from the harmonic state; and the structure dynamic response under random environmental excitation is obtained by subtracting the identified periodic sub-signal from the total structure response.

[0080] Firstly, the harmonic frequency obtained by the power spectral density analysis is directly determined, the periodic sub-signal Kalman filter is adopted, the harmonic sub-signal is identified from the total structure dynamic response graph and the power spectral density graph, and the structure response under environmental excitation can be easily obtained by subtracting the identified harmonic sub-signal from the structure response. The value of the state vector of the initialized Kalman filter can estimate the initial amplitude and phase of the signal by the preliminary spectral analysis. The noise covariance matrix is set according to the noise characteristics of the system, and the amplitude of the environmental noise is estimated by the calibration data. At each time, the state vector is updated according to the current observation value and the predicted value, and the amplitude and phase of the periodic signal are gradually extracted. After the Kalman filter processing, the output periodic signal will contain the structure response part caused by the environmental excitation, while the periodic component produced by the structure inherent characteristics is removed.

[0081] Specifically, firstly, the dynamic model of the offshore wind turbine structure under periodic excitation is established according to the extracted harmonic frequency, and this model should be able to reflect the vibration characteristics and response mode of the structure under harmonic excitation; the periodic sub-signal Kalman filter is designed according to the dynamic model of the offshore wind turbine structure and the harmonic frequency identification result; the state vector of the filter is determined, including the amplitude, phase and other parameters of the harmonic signal; the state transition matrix and the observation matrix are designed to reflect the dynamic characteristics and observation characteristics of the offshore wind turbine structure dynamic response.

[0082] Then, the state vector of the Kalman filter is initialized according to the harmonic frequency identification result and the preliminary dynamic response data analysis; the process noise covariance matrix and the measurement noise covariance matrix are set to reflect the uncertainty of the system dynamics and the measurement process.

[0083] Then, the collected dynamic response data is input into the designed Kalman filter; the filter recursively updates the state vector according to the state transition matrix and the observation matrix, combined with the statistical characteristics of the process noise and the measurement noise.

[0084] Then, the periodic sub-signals corresponding to the harmonic frequencies are extracted through the output of the Kalman filter; these periodic sub-signals reflect the response characteristics of the offshore wind turbine structure under harmonic excitation.

[0085] Finally, the extracted periodic sub-signals are subtracted from the original dynamic response data; the result obtained is the dynamic response of the offshore wind turbine structure affected only by environmental excitation. The dynamic response data after removing the periodic sub-signals is verified to ensure that it reflects the structural response caused only by environmental excitation; if the verification result is not ideal, the parameters of the Kalman filter can be adjusted or the harmonic frequency identification can be performed again to improve the extraction accuracy.

[0086] It should be noted that in-depth analysis of the dynamic response data affected only by environmental excitation can help understand the vibration characteristics and fatigue damage of the offshore wind turbine structure under environmental excitation. The processed dynamic response data can be used for health monitoring and early warning system of the offshore wind turbine structure to timely discover potential safety hazards and handle them. At the same time, according to the analysis results of the dynamic response data, the design of the offshore wind turbine structure can be optimized and improved to improve its ability to resist wind, waves and other environmental excitations.

[0087] S4, using the random subspace algorithm, identifying the offshore wind turbine structure dynamic response data under only environmental excitation to obtain the Monte Carlo stability diagram and the stable modal parameters of the offshore wind turbine structure. Please refer to Figure 8 and Figure 9 , respectively showing the flowchart of S4 in the offshore wind turbine structure modal parameter automatic identification method described in the present application and the structural response modal identification stability diagram in the offshore wind turbine structure modal parameter automatic identification method described in the present application. As shown in Figure 8 and Figure 9 , the S4 comprises the following steps:

[0088] S41, determining the value range of the control parameters based on the offshore wind turbine structure dynamic response data under only environmental excitation; the control parameters include: fundamental frequency, modal number, time delay parameter, system order; the value range of the control parameters includes: the value range of the time delay parameter and the value range of the system order;

[0089] S42, generating a plurality of groups of random control parameters using the Monte Carlo algorithm;

[0090] S43, the random control parameters are identified by using a covariance-driven stochastic subspace identification algorithm to obtain a plurality of sets of candidate modalities.

[0091] S44, the stable modal parameters are obtained by using a two-stage stability checking method based on the plurality of sets of candidate modalities.

[0092] In this embodiment, the two-stage stability checking method includes a first-stage preliminary stability checking and a second-stage comprehensive stability checking.

[0093] The first-stage preliminary stability checking is to compare each candidate modality with other candidate modalities to screen out modalities that repeatedly appear under different parameter combinations and mark them as possible stable modalities.

[0094] The second-stage comprehensive stability checking includes classifying the parameters of the possible stable modalities by using an algorithm to identify stable modality parameters; and selecting candidate modalities with a stability index greater than 0.3s as stable results.

[0095] S45, a Monte Carlo stability diagram is established based on the stable modality parameters.

[0096] In this embodiment, the Monte Carlo covariance-driven stochastic subspace identification method is applied to the structure response under environmental excitation to obtain a Monte Carlo stability diagram to identify the stable modalities of the structure.

[0097] Specifically, the SSI-COV method based on Monte Carlo is used; by considering the fundamental frequency f f and the number of modalities n m of the structure, the value range of two control parameters is determined, and then s=100 sets of random control parameters are generated by using Monte Carlo simulation; the covariance-driven stochastic subspace method is used based on the generated control parameters and the modalities of the structure under random excitation; then, by using the two-stage stability checking, a Monte Carlo-based stability diagram is established to identify the modalities of the structure.

[0098] The calculation formula of the value range of the time delay parameter τ is:

[0099]

[0100] τ max = 2 x τ min (2)

[0101] The calculation formula of the value range of the system order O is:

[0102] N min = 2 x n m (3)

[0103] N max= 2 x N min (4)

[0104] where τ min represents the minimum value of the time delay parameter; τ max represents the maximum value of the time delay parameter; f s represents the sampling frequency value; f f represents the structural fundamental natural frequency automatically obtained from the power spectral density plot using the scale-space peak picking technique; N min represents the minimum value of the system order; N max represents the maximum value of the system order; n m represents the number of modes.

[0105] Then, s = 100 sets of random numbers about the system order parameter and the time delay parameter can be generated using the Monte Carlo simulation, and a total of s candidate modes can be obtained using the generated control parameters and candidate modal parameters based on the covariance-driven stochastic subspace identification method.

[0106] For example, within the value range of the time delay parameter τ and the system order N, s sets of random integers [τ, N] are generated as control parameters using the Monte Carlo simulation. Then, for each set of random control parameters [τ, N], the covariance-driven stochastic subspace identification algorithm (SSI-COV) is used to analyze the structural dynamic response under only environmental excitation to obtain the candidate modal parameters corresponding to each set of control parameters, including the natural frequency, damping ratio, and mode shape, etc.

[0107] Next, by adopting a two-stage stability check, a two-stage stability diagram screening method is proposed to eliminate false modes to obtain a high-quality clear stability diagram.

[0108] The first stage is to compare and screen each candidate mode with other candidate modes from other groups using the following criteria, i.e.:

[0109] The calculation formula used is:

[0110]

[0111] 1 - MAC (φ a , φ b ) ≤ 1% (7)

[0112] where a and b represent two different modes; f represents the natural frequency; ξ represents the damping ratio; φ represents the mode shape; Δf represents the relative natural frequency difference; Δξ represents the relative damping ratio difference; and MAC represents the modal assurance criterion between the mode shapes of two different modes.

[0113] The second stage: further verify the preliminarily screened modes to ensure their stability.

[0114] In the two-stage stability chart screening method, the second stage considers only the candidate modes with a stability index greater than 0.3s = 30 as stable results.

[0115] Finally, the stray modes are effectively identified by the two-stage stability checking method. For easy comparison, the traditional stability chart is provided.

[0116] That is, based on the results of the two-stage stability check, a Monte Carlo-based stability chart is drawn. The x-axis of the stability chart is the natural frequency, and the y-axis is the serial number (or group number) of the random control parameters [τ,Ο] generated by Monte Carlo simulation.

[0117] Therefore, due to the interference of harmonic excitation, it is difficult to accurately identify the basic structural modal and misidentify several pseudo-modes as stable results, making it challenging to obtain accurate modal identification results. However, the Monte Carlo-based stability chart greatly reduces the workload of structural modal parameter identification due to the absence of harmonic excitation interference.

[0118] It should be noted that during the data preprocessing stage, more appropriate filtering methods and parameters can be selected based on the characteristics of the data and the identification requirements; during the random subspace algorithm selection and training stage, different algorithm and parameter combinations can be tried to find the best identification scheme; during the two-stage stability check, the consistency of the test conditions should be ensured to accurately evaluate the stability of the modal parameters. The stability index and threshold should be adjusted according to the specific application scenario and requirements. During the Monte Carlo simulation and stability chart generation stage, the simulation times and the value range of the control parameters can be adjusted as needed.

[0119] In addition, the generation of the Monte Carlo stability chart and the determination of the stable modal parameters of the offshore wind turbine structure may require the use of professional software and tools, such as programming languages like MATLAB and Python, and data analysis software. These software and tools provide rich data processing, algorithm implementation, and visualization functions, which can help to more efficiently complete the identification task.

[0120] S5, based on the stable modal parameters of the offshore wind turbine structure, clustering is performed to obtain the final offshore wind turbine structure modal parameters, thereby realizing the automated identification of the offshore wind turbine structure modal parameters. Please refer to Figure 10 , which shows the flowchart of S5 in the offshore wind turbine structure modal parameter automated identification method described in the present application. As Figure 4 indicated, S5 includes the following steps:

[0121] S51, based on the Monte Carlo stability chart, obtaining a stable modal parameter data set and the number of candidate modes;

[0122] S52, based on the number of candidate modalities, calculating the value range of the minimum group size, and calculating the initial value of the minimum group size MinPts; including: based on the number of candidate modalities, calculating the initial value of the minimum group size MinPts; for the current value of MinPts, calculating the distance between each data point in the stable modal parameter data set obtained by the Monte Carlo stable figure and its first MinPts-1 neighbor, and generating a K-distance graph; connecting the first point and the last point of the K-distance graph to form a first straight line; finding the point farthest from the first straight line and perpendicular to the straight line on the K-distance graph as the elbow point; the distance value corresponding to the elbow point is the optimal setting value of the maximum cluster range;

[0123] S53, based on the initial value of MinPts, performing optimal selection to obtain the optimal setting under the current MinPts;

[0124] S54, based on the optimal setting under the current MinPts, performing clustering to generate clustering clusters, and calculating the profile coefficient of each candidate modality in the clustering cluster;

[0125] The calculation formula of the profile coefficient is:

[0126]

[0127] Wherein, Sil(p) represents the profile coefficient; b(p) represents the minimum value of the average distance of p point to other clusters; av(p) represents the average value of the distance of p point to other points in the same cluster;

[0128] S55, sequentially traversing all MinPts to perform parameter iteration optimization to obtain the total profile coefficient;

[0129] S56, analyzing and optimizing the total profile coefficient to obtain the final offshore wind turbine structure modal parameters.

[0130] In the embodiment, the structural modal parameters of the offshore wind turbine structure are obtained by a density-based spatial clustering method with noise application. A method of automatically optimally selecting two control parameters required by the DBSCAN density clustering method combining the profile coefficient and the elbow rule is used for processing.

[0131] Based on the stable number of candidate modalities obtained in the foregoing steps, the optimal value of the minimum group size MinPts is calculated according to the formula N min = 5ln(P); N max = 10ln(P), wherein P is the number of modalities in the clear stable figure based on the quasi-Monte Carlo. The possible value range of the optimal value of the minimum group size MinPts is calculated [N min , N maxThe silhouette coefficient of each candidate mode is calculated according to formula (8). When drawing the K-distance graph, the value of K is MinPts-1, a straight line is formed by connecting the first point and the last point of the K-distance graph, the point farthest from the straight line and perpendicular to the straight line on the K-distance graph is defined as the elbow of the K-distance graph, and the value is taken as the optimal setting of the maximum clustering range ε.

[0132] The method mainly includes the following five steps:

[0133] (1) MinPts=N min is set, and the corresponding value of ε can be optimally set by drawing the K-distance graph and finding the value corresponding to the elbow position of the K-distance graph;

[0134] (2) Perform a density-based clustering method to obtain a clustering result;

[0135] (3) Calculate the silhouette coefficient of each candidate mode, multiply the silhouette coefficient of the candidate mode of the outlier class by-1, and finally add the silhouette coefficients of all candidate modes to obtain a judgment index of the clustering quality of this time;

[0136] (4) Increase the value of MinPts one by one, iterate steps (1), (2) and (3) until MinPts=N max , and record the results of the silhouette coefficient sum of each candidate mode in each iteration;

[0137] (5) Select the optimal values of the two control parameters of DBSCAN: the size of the minimum group MinPts and the maximum clustering range ε according to the maximum value in the recorded silhouette coefficient sum each time.

[0138] According to the above content, each point in the clustering has at least MinPts in its neighborhood within a given radius ε, and points that do not meet this requirement are considered outliers and are removed. Using the Monte Carlo stability diagram, the DBSCAN clustering method is used to cluster the stable mode parameters, remove abnormal noise points, obtain the true mode, and finally obtain the mode parameters by averaging.

[0139] As shown in Table 1, the estimated mode parameters obtained by using the above method are compared with the mode parameters obtained by the reference method HM-SSI. Figure 11

[0140] Table 1 Offshore wind turbine automatic identification result statistical table

[0141]

[0142] The above comparison shows that there is good consistency between the two groups of results, verifying the effectiveness of the method provided in the application.​

[0143] The offshore wind turbine structure modal parameter automatic identification method provided by the application can effectively eliminate the influence of harmonic excitation on offshore wind turbine modal parameter identification. On the basis of using power spectral density analysis in the process of measuring the dynamic response of the offshore wind turbine structure, the harmonic frequency is automatically identified by using an LSTM deep learning network. Then, based on the identified harmonic frequency, the dynamic response of the offshore wind turbine structure under only environmental excitation is obtained by using a periodic sub-signal Kalman filter. Then, the modal parameters of the offshore wind turbine structure are automatically identified by using a Monte Carlo covariance-driven stochastic subspace identification method. The application can consider the influence of system order parameters and time delay parameters on the final modal identification result by using the covariance-driven stochastic subspace method based on Monte Carlo, thereby improving the accuracy of the identification result. In addition, the identification process can be improved in automation and efficiency by using a deep learning algorithm. At the same time, the effectiveness and accuracy of the method are verified by numerical simulation and field measurement, which can provide an effective tool for modal parameter identification of offshore wind turbine structures under operating conditions, thereby improving the structural health monitoring, state evaluation and damage identification capability of offshore wind turbines, effectively overcoming the shortcomings of the prior art and having high industrial utilization value.

[0144] The protection scope of the offshore wind turbine structure modal parameter automatic identification method described in the embodiments of the application is not limited to the order of steps listed in the embodiments. Any scheme realized by increasing, replacing or changing the steps of the prior art according to the principles of the application is included in the protection scope of the application.

[0145] In summary, the offshore wind turbine structure modal parameter automatic identification method provided by the application has the following beneficial effects:

[0146] The offshore wind turbine structure modal parameter automatic identification method provided by the application can effectively solve the challenge of harmonic excitation to modal parameter identification and improve the accuracy of structural modal parameters. It is expected to improve the structural health monitoring, state evaluation and damage identification capability of offshore wind turbines, thereby supporting the development of offshore wind energy. At the same time, the application effectively overcomes the shortcomings of the prior art and has high industrial utilization value.

[0147] The above embodiments only exemplarily illustrate the principles and effects of the application and are not used to limit the application. Any person skilled in the art can modify or change the above embodiments without departing from the spirit and scope of the application. Therefore, all equivalent modifications or changes made by those skilled in the art without departing from the spirit and technical thought of the application should be covered by the claims of the application.

Claims

1. An offshore wind turbine structure modal parameter automatic identification method, characterized in that, The method comprises the following steps: Obtaining offshore wind turbine structure dynamic response data in a target area; Performing power spectral density analysis based on the offshore wind turbine structure dynamic response data to identify harmonic frequencies; Filtering the harmonic frequencies to obtain offshore wind turbine structure dynamic response data under only environmental excitation; including: constructing a dynamic model of the offshore wind turbine structure under periodic excitation according to the harmonic frequencies; setting a periodic sub-signal Kalman filter based on the dynamic model; initializing the periodic sub-signal Kalman filter according to the harmonic frequencies and the offshore wind turbine structure dynamic response data; inputting the offshore wind turbine structure dynamic response data into the periodic sub-signal Kalman filter for filtering to obtain a periodic sub-signal corresponding to the harmonic frequencies; removing the periodic sub-signal from the offshore wind turbine structure dynamic response data to obtain offshore wind turbine structure dynamic response data under only environmental excitation; wherein the offshore wind turbine structure dynamic response data includes offshore wind turbine structure response under periodic excitation and structure response data under only environmental excitation; Using a random subspace algorithm to identify the offshore wind turbine structure dynamic response data under only environmental excitation to obtain a Monte Carlo stability diagram and stable modal parameters of the offshore wind turbine structure; including: determining the value range of the control parameters based on the offshore wind turbine structure dynamic response data under only environmental excitation, the control parameters including: fundamental frequency, modal number, time delay parameter, and system order; generating a plurality of groups of random control parameters using a Monte Carlo algorithm; identifying the random control parameters using a covariance-driven random subspace identification algorithm to obtain a plurality of groups of candidate modes; identifying the stable modal parameters based on a plurality of groups of the candidate modes using a two-stage stability checking method; and establishing a Monte Carlo stability diagram based on the stable modal parameters. Based on the stable modal parameters of the offshore wind turbine structure, clustering is performed to obtain the final offshore wind turbine structure modal parameters, thereby realizing automatic identification of the offshore wind turbine structure modal parameters.

2. The offshore wind turbine structure modal parameter automated identification method according to claim 1, characterized in that, The method comprises the following steps: Pretreating the offshore wind turbine structure dynamic response data to obtain time series segments of the dynamic response data; Performing Fourier transform based on each time series segment and calculating the power spectral density to obtain a relationship curve between the power spectral density value and the frequency value of the dynamic response data; Extracting features from the relationship curve between the power spectral density value and the frequency value to obtain key features; the key features include: harmonic frequency, spectral peak frequency, and power spectral density amplitude; Constructing an LSTM model based on the key features; Obtaining a training set and a test set of the time series segments of the dynamic response data, training and optimizing the LSTM model to obtain a trained LSTM model; and Identifying the harmonic frequencies through the trained LSTM model.

3. The offshore wind turbine structure modal parameter automated identification method according to claim 1, characterized in that, The value range of the control parameters includes the value range of the time delay parameter and the value range of the system order; The calculation formula of the value range of the time delay parameter is: τ max = 2 x τ min The calculation formula of the value range of the system order is: N min = 2 x n m N max = 2 x N min where τ min represents the minimum value of the time delay parameter; τ max represents the maximum value of the time delay parameter; f s represents the sampling frequency value; f f represents the fundamental natural frequency of the structure automatically obtained from the power spectral density plot using the scale space peak picking technique; N min represents the minimum value of the order of the system; N max represents the maximum value of the order of the system; n m represents the number of modes.

4. The offshore wind turbine structure modal parameter automated identification method according to claim 1, characterized in that, The two-stage stability checking method comprises: a first-stage preliminary stability checking and a second-stage comprehensive stability checking; The first-stage preliminary stability checking is to compare each candidate mode with other candidate modes to screen out modes that repeatedly appear under different parameter combinations and mark them as possible stable modes; The calculation formula used is: 1 - MAC (φ a , φ b ) ≤ 1% Wherein, a and b represent two different modes respectively; f represents the natural frequency; ξ represents the damping ratio; φ represents the mode shape; Δf represents the relative natural frequency difference; Δξ represents the relative damping ratio difference; and MAC represents the modal assurance criterion between the mode shapes of two different modes.

5. The offshore wind turbine structure modal parameter automated identification method according to claim 4, characterized in that, The second-stage comprehensive stability checking comprises: An algorithm is used to classify the parameters of the possible stable modes to identify stable mode parameters; Candidate modes with a stability index greater than 0.3s are selected as stable results.

6. The offshore wind turbine structure modal parameter automated identification method according to claim 1, characterized in that, Based on the stable mode parameters of the offshore wind turbine structure, the final offshore wind turbine structure mode parameters are obtained through clustering, including: Based on the Monte Carlo stability diagram, a stable mode parameter data set and a number of candidate modes are obtained; Based on the number of candidate modes, the value range of the minimum cluster size is calculated, and the initial value of the calculated minimum cluster size MinPts is calculated; Based on the initial value of the MinPts, the optimal setting under the current MinPts is obtained through optimal selection; Based on the optimal setting under the current MinPts, clustering clusters are generated, and the silhouette coefficients of candidate modes in each clustering cluster are calculated; All MinPts are traversed in turn to perform parameter iteration optimization to obtain the total silhouette coefficient; The total silhouette coefficient is analyzed and optimized to obtain the final offshore wind turbine structure mode parameters.

7. The offshore wind turbine structure modal parameter automated identification method according to claim 6, characterized in that, Based on the number of candidate modes, the value range of the minimum cluster size is calculated, and the initial value of the calculated minimum cluster size MinPts is calculated, including: Based on the number of candidate modes, the initial value of the minimum cluster size MinPts is calculated; For the value of the current MinPts, the distance between each data point in the Monte Carlo stability diagram and its first MinPts-1 neighbor is calculated, and a K-distance diagram is generated; The first point and the last point of the K-distance diagram are connected to form a first straight line; The elbow point farthest from the first straight line and perpendicular to the straight line is found on the K-distance diagram; the distance value corresponding to the elbow point is the optimal setting value of the maximum clustering range.

8. The offshore wind turbine structure mode parameter automatic identification method according to claim 6, wherein The calculation formula of the silhouette coefficient is: Wherein, Sil(p) represents the silhouette coefficient; b(p) represents the minimum value of the average distance of p point to other clusters; and av(p) represents the average value of the distance of p point to other points in the same cluster.

Images