A wind turbine support structure modal identification method and system

Abstract

The application provides a wind turbine support structure modal parameter identification method, acceleration data of the support structure is acquired through a vibration data acquisition system, a modal parameter identification algorithm composed of a random subspace method, a clustering method and a modal vibration mode denoising method is used to accurately extract modal parameters such as vibration frequency, damping ratio and modal vibration mode of the support structure, a prediction model between environmental parameters, operating parameters and modal parameters is constructed, and a support structure health state evaluation and early warning and correction of the prediction model are realized by using an artificial intelligence edge computing box. Finally, the operation strategy of the wind turbine and the frequency of the damping tuning vibration reduction system are adjusted in time through an operation control system, so that the safe operation of the support structure is ensured.

Description

Technical Field

[0001] This invention belongs to the field of structural health monitoring, and specifically relates to a method and system for modal identification of wind turbine support structures. Background Technology

[0002] The support structure of a wind turbine generator combines the structural characteristics of a tall structure and a load-bearing body for rotating power equipment. For onshore wind turbine generators, the support structure usually refers to the tower; for offshore wind turbine generators, the support structure consists of the tower and the foundation structure. Currently, the height of wind turbine generator support structures has exceeded 100m, and even approaches 200m. The wind turbine generator support structure bears the loads transmitted from the generator and environmental loads such as wind, waves, and currents. Under long-term load cycles, as well as the effects of seabed erosion and loosening of bolts at tower connections, the support structure may suffer structural damage, which can cause changes in the dynamic characteristics of the entire turbine mechanism and lead to catastrophic damage such as tower collapse.

[0003] Structural modal identification and analysis are core components of assessing the health and safety of wind turbine supports. Vibration data can be obtained by measuring the dynamic response of the support structure. Using structural modal identification methods, modal parameters such as vibration frequency, damping ratio, and mode shape can be extracted, supporting structural health assessment. However, unlike bridges and high-rise buildings, wind turbines in actual engineering projects are subjected to complex and coupled environmental excitations and forced vibrations. Wind, wave, and current environments, as well as turbine operating factors (including yaw and rotation), all influence structural modes to varying degrees. Engineering practice shows that traditional modal identification techniques, when directly applied to wind turbine supports, suffer from low accuracy and poor robustness. Existing modal parameter identification technologies for wind turbine support structures exhibit poor dispersion, containing many spurious modal parameters and mismatches between natural frequencies and mode shapes, making real-time monitoring and assessment of the support structure's health under complex environments difficult.

[0004] Therefore, there is an urgent need to propose a precise identification method for the structural modal parameters of the wind turbine support structure under the influence of multiple coupled factors, so as to provide a theoretical basis for the health monitoring and control of the wind turbine support structure. Summary of the Invention

[0005] Purpose of the invention: To address the problems of low modal recognition accuracy and poor robustness of wind turbine generator support under complex environmental excitation and forced vibration conditions, this invention proposes a modal recognition method for wind turbine generator support structures.

[0006] Technical solution:

[0007] This invention proposes a method for identifying modal parameters of a wind turbine generator support structure, comprising:

[0008] S1. Install acceleration sensors on the support structure of the wind turbine generator set and arrange acceleration data acquisition instruments on the platform inside the tower.

[0009] S2. Transmit the acceleration data to the AI ​​edge computing box for structural modality recognition, including:

[0010] (1) Based on acceleration data, stability diagram data is generated using a random subspace algorithm to identify the modal parameters of the unit support structure. The modal parameters include natural frequency, damping ratio and mode shape.

[0011] (2) Clustering algorithm is used to cluster the physically meaningful modal points in the stability graph data into the same class. Modal points with a frequency lower than the frequency threshold are regarded as false modal points and false modal points are deleted in the preliminary stage. The frequency threshold is determined according to the trial calculation results under each working condition.

[0012] (3) The modal shapes are normalized to their maximum values, and a relationship diagram between the installation height of the accelerometer and the modal shapes after normalization is established. The test values ​​of each modal point in the relationship diagram are obtained by using Gaussian process regression. The standard deviation and average value of all test values ​​are calculated. The deviation between the test value and the average value of each modal point is calculated. Modal points with deviations higher than the standard deviation are regarded as false modal points and are deleted in the stability diagram data. The modal shapes after normalization include the first-order mode shape and the second-order mode shape.

[0013] S3. Connect the wind turbine operating parameters and wind, wave and flow environment parameters to the artificial intelligence edge computing box, and use the artificial intelligence edge computing box to analyze the relationship between the support structure modal parameters and the wind turbine operating parameters and wind, wave and flow environment parameters, and score the health status of the support structure.

[0014] Furthermore, the acceleration sensor is a dual-axis or triaxial acceleration sensor, with at least one acceleration sensor arranged at the top of the tower in the unit support structure, at least one acceleration sensor arranged at the connection between the tower and the foundation, and at least two acceleration sensors arranged in the middle of the tower.

[0015] Furthermore, the step of using a clustering algorithm to cluster physically meaningful modal points in the stable graph data into the same class includes:

[0016] The first clustering for short-term data and the second clustering for long-term data both use the same clustering algorithm; the long-term data includes acceleration data continuously monitored on a daily or weekly basis, and the short-term data includes acceleration data continuously monitored on an hourly basis.

[0017] Furthermore, the clustering algorithm uses stable graph data generated by the random subspace identification method for clustering. Each stable graph data includes several modal points, and each modal point corresponds to one natural frequency, one damping ratio, and one mode shape.

[0018] Furthermore, the clustering algorithm includes:

[0019] The absolute deviation of the natural frequency is divided by the modal confidence criterion to construct a clustering distance metric formula. This formula is used to calculate the distance between several modal points in the stability graph data of the first clustering. DBSCAN clustering is then performed based on the calculated distances to remove spurious modal points. The clustering distance metric formula is expressed as follows:

[0020]

[0021] In the formula f i Let f be the natural frequency of the i-th mode point. j Let ξ be the natural frequency of the j-th modal point. i Let ξ be the damping ratio at the i-th mode point. j Let k be the damping ratio of the j-th mode point, MAC be the measure of the correlation between the mode shape vector of the i-th mode point and the mode shape vector of the j-th mode point, and k be the damping ratio of the j-th mode point. f and k ξ This is the weight of the natural frequency and the damping ratio, and their sum is 1.

[0022] Furthermore, the wind turbine operating parameters include instantaneous wind speed and wind direction angle, average wind speed and wind direction angle, generator speed, blade pitch angle, rotor speed, yaw angle, and generator output power recorded by the wind turbine.

[0023] This invention also proposes a modal parameter identification system for wind turbine generator support structures, comprising:

[0024] The data acquisition module is used to install acceleration sensors on the support structure of the wind turbine generator set and to arrange acceleration data acquisition instruments on the platform inside the tower.

[0025] The identification module transmits acceleration data to the unit-side AI edge computing box for structural modal identification; the modal identification steps include:

[0026] (1) Based on acceleration data, stability diagram data is generated using a random subspace algorithm to identify the modal parameters of the unit support structure. The modal parameters include natural frequency, damping ratio and mode shape.

[0027] (2) Clustering algorithm is used to cluster the physically meaningful modal points in the stability graph data into the same class. Modal points with a frequency lower than the frequency threshold are regarded as false modal points and false modal points are deleted in the preliminary stage. The frequency threshold is determined according to the trial calculation results under each working condition.

[0028] (3) The modal shapes are normalized to their maximum values, and a relationship diagram between the installation height of the accelerometer and the modal shapes after normalization is established. The test values ​​of each modal point in the relationship diagram are obtained by using Gaussian process regression. The standard deviation and average value of all test values ​​are calculated. The deviation between the test value and the average value of each modal point is calculated. Modal points with deviations higher than the standard deviation are regarded as false modal points and are deleted in the stability diagram data. The modal shapes after normalization include the first-order mode shape and the second-order mode shape.

[0029] The analysis module is used to connect the wind turbine operating parameters and wind, wave and flow environment parameters to the artificial intelligence edge computing box. The artificial intelligence edge computing box is used to analyze the relationship between the modal parameters of the support structure and the wind turbine operating parameters and wind, wave and flow environment parameters, and to score the health status of the support structure.

[0030] Furthermore, the acceleration sensor is a dual-axis or triaxial acceleration sensor, with at least one acceleration sensor arranged at the top of the tower in the unit support structure, at least one acceleration sensor arranged at the connection between the tower and the foundation, and at least two acceleration sensors arranged in the middle of the tower.

[0031] Furthermore, the clustering algorithm is used to group physically meaningful modal points in the stability graph data into the same class, including: a first clustering for short-term data and a second clustering for long-term data, with the same clustering algorithm used for both clusterings; the long-term data includes acceleration data continuously monitored on a daily or weekly basis, and the short-term data includes acceleration data continuously monitored on an hourly basis.

[0032] Furthermore, the wind turbine operating parameters include instantaneous wind speed and wind direction angle, average wind speed and wind direction angle, generator speed, blade pitch angle, rotor speed, yaw angle, and generator output power recorded by the wind turbine.

[0033] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:

[0034] This invention designs a modal parameter identification algorithm. The algorithm accurately extracts modal parameters such as vibration frequency, damping ratio and mode shape of the supporting structure through random subspace method and clustering method. It removes obvious noise points through two clustering, and performs preliminary false modal point removal, retaining modal points with physical significance, which effectively improves the efficiency and accuracy of modal identification.

[0035] This invention establishes a relationship diagram between the installation height of the accelerometer and the normalized modal shape after the maximum value. Since the normalized first-order modal shape parameters, second-order modal shape parameters and tower height have relatively clear shapes, advanced false modal points can be eliminated based on the relationship diagram. False modal points with similar natural frequencies but low modal shape similarity are eliminated, thereby achieving accurate extraction of the modal parameters of the support structure.

[0036] This invention also constructs a prediction model between environmental parameters, operating parameters, and modal parameters. It utilizes the computing power of an artificial intelligence edge computing box, combines the prediction model for correction, and further incorporates expert knowledge to achieve health status assessment and early warning of supporting structures. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of the system composition according to an embodiment of the present invention;

[0038] Figure 2 This is a schematic diagram illustrating the advanced modality point deletion process after the second clustering.

[0039] Figure 3 A graph showing the relationship between the installation height of the accelerometer and the normalized mode shape after the maximum value;

[0040] Figure 4 This is a schematic diagram of the sliding window algorithm. Detailed Implementation

[0041] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments. The present invention proposes a method for modal identification of a wind turbine generator support structure, specifically including the following steps:

[0042] S1: Install acceleration sensors on the support structure (tower and foundation) of the wind turbine generator set, and arrange acceleration data acquisition instruments on the platform inside the tower.

[0043] like Figure 1 The diagram shows the system composition in this embodiment. Accelerometers are placed at different heights along the same vertical line on the tower surface to accurately characterize the vibration modes of the supporting structure. Each accelerometer is connected to a rear-end accelerometer data acquisition unit, which transmits the collected data to a back-end AI edge computing box.

[0044] The acceleration sensor is a biaxial or triaxial acceleration sensor. Preferably, at least one acceleration sensor is arranged at the top of the tower in the unit support structure, at least one acceleration sensor is arranged at the connection between the tower and the foundation, and at least two acceleration sensors are arranged in the middle of the tower.

[0045] S2: Transmit acceleration data to the unit-side AI edge computing box and perform structural modal recognition.

[0046] The artificial intelligence edge computing box used in this invention is a small processor with a computing power of no less than 16 TOPS, and the processor has built-in artificial intelligence algorithms. The structural modality recognition includes the following sub-steps:

[0047] (1) Based on acceleration data, a stability diagram is generated using a random subspace algorithm to identify modal parameters such as the natural frequency, damping ratio, and mode shape of the unit support structure.

[0048] Acceleration data monitored continuously over a long period is divided into several short time intervals. The long period includes acceleration data monitored continuously on a daily or weekly basis, and the short time intervals include acceleration data monitored continuously on an hourly basis. The acceleration data from these short time intervals is analyzed using the random subspace method to generate stability diagram data. Each stability diagram data includes several modal points, each corresponding to one natural frequency, one damping ratio, and one mode shape.

[0049] (2) A clustering algorithm is used to cluster physically meaningful modal points in the stable graph data into the same class. Modal points with a frequency lower than the frequency threshold are regarded as false modal points and are removed. The frequency threshold is manually specified based on the trial calculation results under each working condition.

[0050] The clustering algorithm includes primary clustering for short time periods and secondary clustering for long time periods, specifically:

[0051] (2.1) First clustering:

[0052] The first clustering is performed on stable graph data generated using the random subspace identification method for short-term data. Its core purpose is noise reduction and classification.

[0053] First, the absolute deviation of the natural frequency is divided by the modal confidence criterion to construct a clustering distance metric formula. This formula is then used to calculate the distance between several modal points within each set of stability map data generated over a short time period. DBSCAN clustering is then performed to conduct the first preliminary deletion of false modal points. The clustering distance metric formula, considering both the natural frequency and mode shape, is expressed as:

[0054]

[0055] In the formula f i Let f be the natural frequency of the i-th mode point. j Let ξ be the natural frequency of the j-th modal point. i Let ξ be the damping ratio at the i-th mode point. jLet k be the damping ratio of the j-th mode point, MAC be the measure of the correlation between the mode shape vector of the i-th mode point and the mode shape vector of the j-th mode point, and k be the damping ratio of the j-th mode point. f and k ξ This is the weight of the natural frequency and the damping ratio, and their sum is 1.

[0056] During the analysis, the ratio of the relative deviation of the natural frequency to the MAC index was used, and the MAC index was used as an amplification factor to improve the influence of the mode shape on the clustering quality. The normalized natural frequency was used to adapt to the characteristics of low-frequency vibration of the wind turbine support structure and to distinguish similar frequencies in the downwind and crosswind directions.

[0057] (2.2) Second clustering:

[0058] The short-term stability graph data after clustering is re-divided into long-term data for a second clustering. Further denoising is then performed on the long-term stability graph data to remove infrequent spurious modes, preserving modes with more stable physical meaning to the greatest extent possible. The most crucial indicator for determining the physical significance of a mode point is its mode shape. Using mode shape denoising, anomalous mode points with similar natural frequencies but low mode shape similarity are eliminated, completing the second preliminary spurious mode point removal and achieving accurate extraction of the support structure's modal parameters.

[0059] The clustering diagram obtained after the second clustering in this embodiment is as follows: Figure 2 As shown, after calculation, the circular part is identified as a false mode, and the corresponding mode point is removed from the corresponding stability plot data along the horizontal axis (frequency) in the figure.

[0060] (3) The clustered stability graph data includes several modal point clusters, each modal point in the cluster corresponding to a mode shape. Max-normalization is performed on the mode shapes, and the installation height of each accelerometer on the support structure is collected. Based on the first and second mode shapes of the support structure, a relationship graph between the accelerometer installation height and the mode shape after max-normalization is established.

[0061] For wind turbine support structures, the first and second modal parameters are relatively more important, and the normalized first and second modal parameters are relatively well-defined in relation to the tower height. The first modal parameters include the natural frequency and mode shape. The natural frequency represents the structure's ability to resist deformation during free vibration. The second modal parameters refer to the second natural frequency (i.e., the circular frequency or natural frequency), the second mode shape, and the second damping ratio.

[0062] The relationship diagram obtained in this embodiment is as follows: Figure 3As shown in the figure, hollow symbols (triangles, quadrilaterals, pentagons) represent modal parameters at different times, and solid dots represent the actual vibration mode at each measuring point. The measuring points are the installation points of the accelerometers. Figure 3 In the first-order vibration mode diagram on the left, the curve formed by connecting the pentagonal points and the curve formed by connecting the solid points have significantly different directions; similarly, in Figure 3 In the second-order mode shape relationship diagram on the right, the curve formed by connecting the triangular points and the curve formed by connecting the solid points have significantly different directions, indicating that the mode shape of the corresponding mode point is inconsistent with the true mode shape. Therefore, these spurious mode points need to be deleted from the stability diagram data.

[0063] Specifically, the method for distinguishing the above-mentioned false modal points includes: using Gaussian process regression to obtain the test value of the mode shape of each measurement point in the relationship graph, further calculating the standard deviation / mean of all test values, calculating the deviation of each test value from the mean, and when the deviation is higher than the standard deviation, the corresponding modal point is considered to be a false mode and the corresponding false modal point is deleted.

[0064] In particular, for extreme conditions such as earthquakes and typhoons, the duration of extreme loads is short, but their impact on the health of the supporting structure is significant. Therefore, the sliding window method is combined with a two-stage clustering modal identification method to expand the data set and improve the accuracy of modal parameter identification. Specifically, this combination includes: firstly, using the sliding window algorithm to extract data over a specific period, as follows... Figure 4 As shown, the rectangular portion represents the data set to be expanded, the straight arrow represents the sliding window, and k is the window size. This algorithm divides one hour of data into (120 / k-2) k-minute data sets, thus achieving the goal of expanding the data set. Automatic runtime modal analysis based on two clustering operations is used to identify the expanded data sets and obtain modal parameters.

[0065] Through the above sub-steps (1)-(3), the modal identification method proposed in this invention undergoes two rounds of false modal point elimination, which can eliminate false modal points to the maximum extent, thereby accurately obtaining modal parameters such as the natural frequency, damping ratio, and mode shape of the turbine support structure. In actual generation, using the above sub-steps (1)-(3), the average first-order natural frequency of the turbine support structure (monopile foundation) of a certain offshore wind farm in Guangdong was successfully and accurately identified as approximately 0.295Hz, and the first-order mode shape was a single-phase bending mode shape, which is higher than the designed natural frequency of 0.267Hz, indicating that the wind turbine is operating in a safe state and the support structure design has a certain degree of safety redundancy. The average first-order natural frequency of the turbine support structure (monopile foundation) of a certain offshore wind farm in Liaoning was approximately 0.292Hz, and the first-order mode shape was a single-phase bending mode shape.

[0066] S3: Connect the SCADA data of wind turbine operation and marine environmental monitoring data to the artificial intelligence edge computing box, analyze the relationship between the modal parameters of the support structure, the environmental parameters of wind, wave and current, and the operating parameters of the wind turbine, and give a health status score of the support structure by constructing a prediction model between environmental parameters, operating parameters and modal parameters and through the built-in support structure health status evaluation index.

[0067] SCADA data and environmental parameters include instantaneous wind speed and wind direction angle, average wind speed and wind direction angle, generator speed, blade pitch angle, rotor speed, yaw angle, generator output power, and wind, wave, and flow environmental data recorded by the wind turbine. The AI ​​edge computing box used in this embodiment has a computing power greater than 16 TOPS, and the specific modules used include:

[0068] The feature extraction module uses random subspace identification (SSI) and fast Fourier transform (FFT) to extract modal parameters from vibration, acceleration, and other signals.

[0069] The environmental compensation module is used to eliminate environmental impacts by normalizing the environment through partial least squares regression (PLS), principal component analysis (PCA), and Gaussian process regression (GPR).

[0070] The modal feature input module converts the environmentally compensated modal features into the standard input format for machine learning models;

[0071] The health assessment module uses a machine learning model to score the current health status of the supporting structure based on the input modal features. In this embodiment, the machine learning model is a supervised multilayer perceptron (MLP) model, which learns the relationship between input and output through multilayer nonlinear mapping.

[0072] The prediction model includes a modal parameter-environmental parameter prediction model and a modal parameter-operational parameter prediction model. During the operation of the wind turbine, the prediction model is further adjusted using an AI edge computing box. The adjustment of the model by the AI ​​edge computing box specifically includes the following steps:

[0073] (1) Improve prediction accuracy by correcting the prediction model through self-learning;

[0074] (2) By integrating and analyzing the modal parameters of the supporting structure with monitoring data such as strain and stress of the whole structure, a rapid assessment of the health status of the supporting structure can be achieved;

[0075] (3) Predict the trend of environmental parameter changes, and use the modal parameter prediction model to predict the trend of modal parameter changes, so as to realize the prediction of the health status of the supporting structure.

[0076] The wind turbine generator's operation control system can adjust the wind turbine generator's operation strategy based on the changes in the support structure modal parameters derived from this invention. At the same time, it can adjust the frequency of the damping tuned vibration reduction system to make it close to the natural frequency of the support structure, thereby achieving the best vibration reduction effect.

Claims

1. A method for identifying modal parameters of a wind turbine generator support structure, characterized in that, include: S1. Install acceleration sensors on the support structure of the wind turbine generator set and arrange acceleration data acquisition instruments on the platform inside the tower. S2. Transmit the acceleration data to the AI ​​edge computing box for structural modality recognition, including: (1) Based on acceleration data, stability diagram data is generated using the random subspace algorithm to identify the modal parameters of the unit support structure, including natural frequency, damping ratio and mode shape; (2) A clustering algorithm is used to cluster the physically meaningful modal points in the stability graph data into the same class. Modal points with a frequency lower than the frequency threshold are regarded as false modal points, and false modal points are initially deleted in the stability graph data. The frequency threshold is determined based on the trial calculation results under each working condition. (3) The modal shapes are normalized to their maximum values, and a relationship diagram between the installation height of the accelerometer and the modal shapes after normalization is established. The test values ​​of each modal point in the relationship diagram are obtained by using Gaussian process regression. The standard deviation and average value of all test values ​​are calculated. The deviation between the test value and the average value of each modal point is calculated. Modal points with deviations higher than the standard deviation are regarded as false modal points and are deleted in the stability diagram data. The modal shapes after normalization include the first-order mode shape and the second-order mode shape. S3. Connect the wind turbine operating parameters and wind, wave and flow environment parameters to the artificial intelligence edge computing box, and use the artificial intelligence edge computing box to analyze the relationship between the support structure modal parameters and the wind turbine operating parameters and wind, wave and flow environment parameters, and score the health status of the support structure.

2. The modal parameter identification method according to claim 1, characterized in that, The acceleration sensor is a dual-axis or triaxial acceleration sensor. At least one acceleration sensor is arranged at the top of the tower in the unit support structure, at least one acceleration sensor is arranged at the connection between the tower and the foundation, and at least two acceleration sensors are arranged in the middle of the tower.

3. The modal parameter identification method according to claim 2, characterized in that, The method of using a clustering algorithm to group physically meaningful modal points in the stability graph data into the same class includes: The first clustering for short-term data and the second clustering for long-term data both use the same clustering algorithm; the long-term data includes acceleration data continuously monitored on a daily or weekly basis, and the short-term data includes acceleration data continuously monitored on an hourly basis.

4. The modal parameter identification method according to claim 3, characterized in that, The clustering algorithm uses stable graph data generated by the random subspace identification method for clustering. Each stable graph data includes several modal points, and each modal point corresponds to one natural frequency, one damping ratio, and one mode shape.

5. The modal parameter identification method according to claim 4, characterized in that, The clustering algorithm includes: The absolute deviation of the natural frequency is divided by the modal confidence criterion to construct a clustering distance metric formula. This formula is used to calculate the distance between several modal points in the stability graph data of the first clustering. DBSCAN clustering is then performed based on the calculated distances to remove spurious modal points. The clustering distance metric formula is expressed as follows: In the formula f i Let f be the natural frequency of the i-th modal point. j Let ξ be the natural frequency of the j-th modal point. i Let ξ be the damping ratio at the i-th mode point. j Let k be the damping ratio of the j-th mode point, MAC be the measure of the correlation between the mode shape vector of the i-th mode point and the mode shape vector of the j-th mode point, and k be the damping ratio of the j-th mode point. f and k ξ This is the weight of the natural frequency and the damping ratio, and their sum is 1.

6. The modal parameter identification method according to claim 5, characterized in that, The wind turbine operating parameters include instantaneous wind speed and wind direction angle, average wind speed and wind direction angle, generator speed, blade pitch angle, rotor speed, and generator output power recorded by the wind turbine.

7. A system for identifying modal parameters of a wind turbine generator support structure, characterized in that, include: The data acquisition module is used to install acceleration sensors on the support structure of the wind turbine generator set and to arrange acceleration data acquisition instruments on the platform inside the tower. The recognition module transmits acceleration data to the unit-side AI edge computing box for structural modality recognition. The modality recognition steps include: (1) Based on acceleration data, stability diagram data is generated using a random subspace algorithm to identify the modal parameters of the unit support structure. The modal parameters include natural frequency, damping ratio and mode shape. (2) Clustering algorithm is used to cluster the physically meaningful modal points in the stability graph data into the same class. Modal points with a frequency lower than the frequency threshold are regarded as false modal points and false modal points are deleted in the preliminary stage. The frequency threshold is determined according to the trial calculation results under each working condition. (3) The modal shapes are normalized to their maximum values, and a relationship diagram between the installation height of the accelerometer and the modal shapes after normalization is established. The test values ​​of each modal point in the relationship diagram are obtained by using Gaussian process regression. The standard deviation and average value of all test values ​​are calculated. The deviation between the test value and the average value of each modal point is calculated. Modal points with deviations higher than the standard deviation are regarded as false modal points and are deleted in the stability diagram data. The modal shapes after normalization include the first-order mode shape and the second-order mode shape. The analysis module is used to connect the wind turbine operating parameters and wind, wave and flow environment parameters to the artificial intelligence edge computing box. The artificial intelligence edge computing box is used to analyze the relationship between the modal parameters of the support structure and the wind turbine operating parameters and wind, wave and flow environment parameters, and to score the health status of the support structure.

8. The modal parameter recognition system according to claim 7, characterized in that, The acceleration sensor is a dual-axis or triaxial acceleration sensor. At least one acceleration sensor is arranged at the top of the tower in the unit support structure, at least one acceleration sensor is arranged at the connection between the tower and the foundation, and at least two acceleration sensors are arranged in the middle of the tower.

9. The modal parameter recognition system according to claim 8, characterized in that, The method of using a clustering algorithm to group physically meaningful modal points in the stability graph data into the same class includes: The first clustering for short-term data and the second clustering for long-term data both use the same clustering algorithm; the long-term data includes acceleration data continuously monitored on a daily or weekly basis, and the short-term data includes acceleration data continuously monitored on an hourly basis.

10. The modal parameter recognition system according to claim 9, characterized in that, The wind turbine operating parameters include instantaneous wind speed and wind direction angle, average wind speed and wind direction angle, generator speed, blade pitch angle, rotor speed, and generator output power recorded by the wind turbine.

Images