Vibration control system of offshore wind turbine based on magnetorheological tuned mass damper

Abstract

Disclosed is an offshore wind turbine vibration control system based on a magnetorheological tuned mass damper. The dynamic change implicit features in the time dimension of wind loads, wave loads, and seismic loads at multiple predetermined time points within a predetermined time period are extracted through a time series encoder to obtain a global load feature vector. The magnetic field control intensity values of the MRE-TMD device and the displacement response values of the wind turbine structure of the offshore wind turbine at multiple predetermined time points are extracted through a time series encoder to obtain a magnetic field control feature vector and a displacement response feature vector. Then, the above three feature vectors are respectively fused and corrected to obtain a corrected posterior probability feature vector. Finally, the corrected posterior probability feature vector is passed through a classifier to obtain a classification result. In this way, the dynamic response of the wind turbine can be accurately vibration-controlled to ensure the safe operation of the offshore wind turbine.

Description

Technical Field

[0001] This application relates to the field of intelligent control, and more specifically, to an offshore wind turbine vibration control system based on a magnetorheological tuned mass damper. Background Art

[0002] Wind energy, as a renewable clean energy, is representative of green energy. Offshore wind power generation is a new technology for effectively developing and utilizing wind energy, and has become the mainstream development trend of wind power generation, attracting more and more attention globally. After adopting new composite materials, offshore wind turbines are larger and lighter than before. These large flexible structures are easily affected by external vibration sources in the marine environment. In order to ensure the safe operation of offshore wind turbine units, it is necessary to perform vibration control on the dynamic response of the wind turbine units. Currently, domestic and foreign scholars mainly adopt passive tuned mass dampers (TMDs) installed at the top of the offshore wind turbine tower (i.e., inside the nacelle) for vibration control of the offshore wind turbine tower / blade. By suppressing the fundamental vibration mode of the tower with the largest displacement at the tower top, the vibration reduction effect on the wind turbine tower is achieved. This method is applicable to the vibration control of offshore wind turbines under low-frequency loads of wind and waves. At this time, the main excitation mode of the wind turbine tower is the first-order main vibration mode of the tower top displacement.

[0003] Many existing or under-construction offshore wind turbine units in the world are located in earthquake-prone areas. Under the action of multiple loads of wind, waves, and earthquakes, the high-order mode vibration modes of offshore wind turbines are excited, and the system exhibits vibration characteristics with multiple different orders of natural frequencies. If the passive TMD designed for conventional working conditions (wind / wave loads) is still used to reduce the vibration of the wind turbine, the effect will be greatly reduced or even the vibration suppression will fail.

[0004] Therefore, an optimized vibration control scheme for offshore wind turbines is expected. Summary of the Invention

[0005] In order to solve the above technical problems, this application is proposed. The embodiments of this application provide an offshore wind turbine vibration control system based on a magnetorheological tuned mass damper. It extracts the dynamic change implicit features of wind loads, wave loads, and seismic loads at multiple predetermined time points within a predetermined time period in the time dimension through a time series encoder to obtain a global load feature vector, and extracts the dynamic implicit correlation features of the magnetic field control intensity values of the MRE-TMD device and the displacement response values of the wind turbine structure of the offshore wind turbine at multiple predetermined time points in the time series dimension to obtain a magnetic field control feature vector and a displacement response feature vector. Then, the above three feature vectors are respectively fused and corrected to obtain a corrected posterior probability feature vector. Finally, the corrected posterior probability feature vector is passed through a classifier to obtain a classification result. In this way, the dynamic response of the wind turbine unit can be accurately vibration-controlled to ensure the safe operation of the offshore wind turbine unit.

[0006] According to one aspect of the present application, a vibration control system for an offshore wind turbine based on a magnetorheological tuned mass damper is provided, which includes:

[0007] A data acquisition module, configured to obtain wind loads, wave loads, and seismic loads at multiple predetermined time points within a predetermined time period, the magnetic field control intensity values of the MRE-TMD device at the multiple predetermined time points, and the displacement response values of the wind turbine structure of the offshore wind turbine at the multiple predetermined time points;

[0008] A load data time series feature extraction module, configured to arrange the wind loads, wave loads, and seismic loads at the multiple predetermined time points within the predetermined time period as input vectors respectively, and then obtain first to third load feature vectors through a time series encoder including a one-dimensional convolutional layer;

[0009] A load feature fusion module, configured to fuse the first to third load feature vectors to obtain a global load feature vector;

[0010] A magnetic field data and displacement data encoding module, configured to arrange the magnetic field control intensity values of the MRE-TMD device at the multiple predetermined time points and the displacement response values of the wind turbine structure of the offshore wind turbine at the multiple predetermined time points as input vectors respectively, and then obtain a magnetic field control feature vector and a displacement response feature vector through the time series encoder including the one-dimensional convolutional layer;

[0011] A Bayesian inference module, configured to use a Bayesian probability model to fuse the magnetic field control feature vector, the displacement response feature vector, and the global load feature vector to obtain a posterior probability feature vector;

[0012] A posterior distribution correction module, configured to correct the eigenvalue at each position in the posterior probability feature vector based on the mean and variance of the eigenvalue set at all positions in the posterior probability vector to obtain a corrected posterior probability feature vector; and

[0013] A vibration control result generation module, configured to pass the corrected posterior probability feature vector through a classifier to obtain a classification result, where the classification result is used to indicate whether the magnetic field control intensity value of the MRE-TMD device at the current time point should be increased or decreased.

[0014] According to another aspect of the present application, a vibration control method for an offshore wind turbine based on a magnetorheological tuned mass damper is provided, which includes:

[0015] Obtain wind loads, wave loads, and seismic loads at multiple predetermined time points within a predetermined time period, the magnetic field control intensity values of the MRE-TMD device at the multiple predetermined time points, and the displacement response values of the wind turbine structure of the offshore wind turbine at the multiple predetermined time points;

[0016] After arranging the wind loads, wave loads, and seismic loads at multiple predetermined time points within the predetermined time period as input vectors respectively, use a time series encoder including a one-dimensional convolutional layer to obtain the first to third load feature vectors;

[0017] Fuse the first to third load feature vectors to obtain a global load feature vector;

[0018] After arranging the magnetic field control intensity values of the MRE-TMD device at the multiple predetermined time points and the displacement response values of the wind turbine structure of the offshore wind turbine at the multiple predetermined time points as input vectors respectively, use the time series encoder including the one-dimensional convolutional layer to obtain a magnetic field control feature vector and a displacement response feature vector;

[0019] Use a Bayesian probability model to fuse the magnetic field control feature vector, the displacement response feature vector, and the global load feature vector to obtain a posterior probability feature vector;

[0020] Based on the mean and variance of the set of eigenvalue at all positions in the posterior probability vector, correct the eigenvalue at each position in the posterior probability feature vector to obtain a corrected posterior probability feature vector; and

[0021] Pass the corrected posterior probability feature vector through a classifier to obtain a classification result, and the classification result is used to indicate whether the magnetic field control intensity value of the MRE-TMD device at the current time point should be increased or decreased.

[0022] Compared with the prior art, an offshore wind turbine vibration control system based on a magnetorheological tuned mass damper provided by the present application extracts the implicit features of the dynamic changes in the time dimension of the wind loads, wave loads, and seismic loads at multiple predetermined time points within a predetermined time period to obtain a global load feature vector, and extracts the implicit correlation features in the time series dimension of the magnetic field control intensity value of the MRE-TMD device and the displacement response value of the wind turbine structure of the offshore wind turbine at multiple predetermined time points to obtain a magnetic field control feature vector and a displacement response feature vector. Then, fuse and correct the aforementioned three feature vectors respectively to obtain a corrected posterior probability feature vector, and finally pass the corrected posterior probability feature vector through a classifier to obtain a classification result. In this way, the dynamic response of the wind turbine can be accurately vibration-controlled to ensure the safe operation of the offshore wind turbine. Description of the Drawings

[0023] The above and other objects, features, and advantages of the present application will become more apparent by describing the embodiments of the present application in more detail with reference to the accompanying drawings. The drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation to the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0024] Figure 1A and Figure 1B illustrate the Kobe earthquake acceleration time history and the Kobe earthquake acceleration spectrum.

[0025] Figure 2 illustrates an application scenario diagram of an offshore wind turbine vibration control system based on a magnetorheological tuned mass damper according to an embodiment of the present application.

[0026] Figure 3 illustrates a block diagram of an offshore wind turbine vibration control system based on a magnetorheological tuned mass damper according to an embodiment of the present application.

[0027] Figure 4 illustrates a block diagram of the load data time series feature extraction module in an offshore wind turbine vibration control system based on a magnetorheological tuned mass damper according to an embodiment of the present application.

[0028] Figure 5 illustrates a block diagram of the load feature fusion module in an offshore wind turbine vibration control system based on a magnetorheological tuned mass damper according to an embodiment of the present application.

[0029] Figure 6 illustrates a block diagram of the magnetic field data and displacement data encoding module in an offshore wind turbine vibration control system based on a magnetorheological tuned mass damper according to an embodiment of the present application.

[0030] Figure 7 illustrates a flowchart of an offshore wind turbine vibration control method based on a magnetorheological tuned mass damper according to an embodiment of the present application.

[0031] Figure 8 illustrates a schematic architecture diagram of an offshore wind turbine vibration control method based on a magnetorheological tuned mass damper according to an embodiment of the present application. Detailed Embodiments

[0032] Hereinafter, exemplary embodiments of the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0033] Scene Overview

[0034] At present, since the elastic modulus of magnetorheological elastomer (MRE) can change with the change of magnetic field and can immediately return to the initial state after the magnetic field is removed. Based on this magneto-control characteristic, MRE has been widely used in variable stiffness intelligent vibration absorbers. The TMD device consists of a mass, a damper and a spring attached to the structure. MRE-TMD is a TMD device based on MRE material, and its stiffness and damping are provided by MRE. The main feature of this device is to change the stiffness of MRE by changing the applied magnetic field, thereby adjusting the frequency of the MRE-TMD device to make its frequency follow the main frequency of the wind turbine structure, so that when an external excitation acts, the vibration energy is dissipated through the vibration of the mass block, achieving the purpose of reducing the structure vibration.

[0035] Based on this, the inventors of the present application considered changing the stiffness of MRE by regulating the magnetic field control intensity of the MRE-TMD device, thereby adjusting the frequency of the MRE-TMD device to make its frequency follow the main frequency of the wind turbine structure for vibration reduction. And, the inventors of the present application also considered that in this process, it is also necessary to excavate the implicit correlation characteristics of external wind, wave and earthquake loads in the time dimension, so as to fuse the dynamic change characteristic information of the loadings of these three in the time sequence to improve the vibration reduction effect. At the same time, the vibration change characteristic information of the offshore wind turbine structure should also be added in this process to refer to the vibration reduction result, and then the vibration control of the dynamic response of the wind turbine can be accurately carried out to ensure the safe operation of the offshore wind turbine.

[0036] Specifically, in the technical solution of the present application, first, wind loads, wave loads and earthquake loads at multiple predetermined time points within a predetermined time period are obtained, the magnetic field control intensity values of the MRE-TMD device at the multiple predetermined time points, and the displacement response values of the wind turbine structure of the offshore wind turbine at the multiple predetermined time points.

[0037] Among them, for the wind load, since the wind turbine model is simplified to a multi-degree-of-freedom lumped mass model, only the wind load acting on the wind turbine tower is considered, and the force acting on the wind turbine blades is not considered. The wind load acting on the wind turbine tower is:

[0038]

[0039] In the formula: C d is the air resistance coefficient; ρ a is the air density; A is the windward area of the tower; is the average wind speed; θ′(t) is the pulsating wind speed, which is simulated by the harmonic superposition method using the Davenport pulsating wind speed spectrum; the simulated wind load condition is the average wind speed of 25 m / s at the nacelle position.

[0040] Furthermore, for the wave load, the wave is a random process, and the energy distribution of the composed wave can be described by a wave spectrum. The wave elevation is simulated using the JONSWAP spectrum, and the JONSWAP spectrum is as follows:

[0041]

[0042] In the formula: H s is the wave height; γ is the peak factor (take γ = 3.3); g is the acceleration due to gravity; σ is the peak shape factor and α * is calculated as follows:

[0043]

[0044]

[0045] According to the Morison equation, the wave force on a unit height dz is:

[0046]

[0047] In the formula, C D and C M are the drag force coefficient and the inertia force coefficient respectively. Take C D = 0.65, C M = 2, ρ w is the seawater density, take 1030 kg / m 3 . The wave condition is that H s = 9.4 m / s, T p = 10.47 s.

[0048] Furthermore, for the seismic load, for an offshore wind turbine located in an earthquake zone, the seismic load is an extreme condition load. Using Kobe (1995) as the seismic wave input, its seismic acceleration time history and acceleration spectrum are as Figure 1A and Figure 1B shown.

[0049] Then, considering that the wind load, the wave load, and the seismic load all have the characteristic law of correlation in the time dimension. Therefore, in order to be able to fully extract the implicit dynamic change characteristics of each load, after arranging the wind load, wave load, and seismic load at multiple predetermined time points within the predetermined time period into input vectors respectively, a time series encoder containing a one-dimensional convolutional layer is used to encode the first to third input vectors of these three loads respectively, so as to extract the implicit dynamic change characteristics of the wind load in the time dimension, the implicit dynamic change characteristics of the wave load in the time dimension, and the implicit dynamic change characteristics of the seismic load in the time dimension, thereby obtaining the first to third load feature vectors.

[0050] In this way, the dynamic correlation features of these three loads in time are further fused to obtain a global load feature vector. It should be understood that since the wind load, the wave load, and the seismic load have different degrees of influence on the vibration of the offshore wind turbine, in order to pay attention to the degree of vibration influence of each load feature during feature fusion, the first to third load feature vectors are further pre-classified through a pre-classifier to obtain the first to third probability values. In this way, the obtained first to third probability values can be used as weights to weight the first to third load feature vectors, and then the weighted sum between the weighted load feature vectors is calculated to obtain the global load feature vector.

[0051] Regarding the magnetic field control intensity values of the MRE-TMD device at the multiple predetermined time points and the displacement response values of the wind turbine structure of the offshore wind turbine at the multiple predetermined time points, similarly, since they also have dynamic correlation implicit feature information in time series, the magnetic field control intensity values of the MRE-TMD device at the multiple predetermined time points and the displacement response values of the wind turbine structure of the offshore wind turbine at the multiple predetermined time points are respectively arranged as input vectors and then encoded through the time series encoder including a one-dimensional convolutional layer to respectively extract the dynamic implicit correlation features of the magnetic field control intensity value of the MRE-TMD device and the displacement response value of the wind turbine structure of the offshore wind turbine in the time series dimension, so as to obtain a magnetic field control feature vector and a displacement response feature vector.

[0052] It should be understood that in the technical solution of the present application, considering that when the wind load, the wave load, and the seismic load change, the vibration is reduced by regulating the magnetic field control intensity of the MRE-TMD device, and in this process, the vibration change characteristic information represented by the displacement response value of the offshore wind turbine structure is introduced to represent the vibration change characteristic to refer to the vibration reduction result. Therefore, using the magnetic field control feature vector as the prior probability, the purpose in the technical solution of the present application is to update the prior probability to obtain the posterior probability with new evidence, that is, when there are new changes in the wind load, the wave load, and the seismic load. Then according to Bayes' formula, the posterior probability is the prior probability multiplied by the event probability divided by the evidence probability. Therefore, in the technical solution of the present application, a Bayesian probability model is used to fuse the magnetic field control feature vector, the displacement response feature vector, and the global load feature vector to obtain a posterior probability feature vector, where the magnetic field control feature vector is used as the prior, the displacement response feature vector is used as the event, and the global load feature vector is used as the evidence. In this way, the posterior probability feature vector can be passed through a classifier to obtain whether the magnetic field control intensity value of the MRE-TMD device at the current time point should be increased or decreased.

[0053] In particular, in the technical solution of the present application, when the Bayesian probability model is used to fuse the magnetic field control feature vector, the displacement response feature vector and the global load feature vector, since the global load feature vector is the load-related feature in the time series direction obtained by a convolutional neural network as a filter, it may not be completely aligned with the magnetic field control feature vector and the displacement response feature vector in the time series direction. Due to the position-by-position calculation characteristics of the Bayesian probability model, the obtained posterior probability feature vector may contain special feature value points caused by misaligned features, thereby affecting the classification performance of the feature distribution of the posterior probability feature vector.

[0054] Therefore, the information statistics of the adaptive instance are normalized on the posterior probability feature vector, specifically:

[0055]

[0056] f i is the posterior probability feature vector, for example, denoted as the eigenvalue of V, and μ and σ are the feature set f i ∈V, L is the length of the posterior probability feature vector, and α is a hyperparameter.

[0057] Here, the information statistical normalization of the adaptive instance takes the feature value set of the posterior probability feature vector as the adaptive instance, and uses the essential intrinsic prior information of its statistical characteristics to dynamically generate information normalization for a single feature value, and at the same time uses the normalized modulus information of the feature set as a bias to describe the invariance within the set distribution domain, so that the optimization of the feature distribution that shields the disturbance distribution of special instances as much as possible is achieved, thereby improving the accuracy of classification. In this way, the dynamic response of the wind turbine can be accurately controlled to ensure the safe operation of the offshore wind turbine.

[0058] Based on this, the present application proposes an offshore wind turbine vibration control system based on a magnetorheological tuned mass damper, which includes: a data acquisition module for obtaining wind loads, wave loads, and seismic loads at multiple predetermined time points within a predetermined time period, the magnetic field control intensity values of the MRE-TMD device at the multiple predetermined time points, and the displacement response values of the wind turbine structure of the offshore wind turbine at the multiple predetermined time points; a load data time series feature extraction module for arranging the wind loads, wave loads, and seismic loads at the multiple predetermined time points within the predetermined time period as input vectors respectively and passing them through a time series encoder including a one-dimensional convolutional layer to obtain first to third load feature vectors; a load feature fusion module for fusing the first to third load feature vectors to obtain a global load feature vector; a magnetic field data and displacement data encoding module for arranging the magnetic field control intensity values of the MRE-TMD device at the multiple predetermined time points and the displacement response values of the wind turbine structure of the offshore wind turbine at the multiple predetermined time points as input vectors respectively and passing them through the time series encoder including the one-dimensional convolutional layer to obtain a magnetic field control feature vector and a displacement response feature vector; a Bayesian inference module for using a Bayesian probability model to fuse the magnetic field control feature vector, the displacement response feature vector, and the global load feature vector to obtain a posterior probability feature vector; a posterior distribution correction module for correcting the feature values at each position in the posterior probability feature vector based on the mean and variance of the set of feature values at all positions in the posterior probability vector to obtain a corrected posterior probability feature vector; and a vibration control result generation module for passing the corrected posterior probability feature vector through a classifier to obtain a classification result, and the classification result is used to indicate whether the magnetic field control intensity value of the MRE-TMD device at the current time point should be increased or decreased.

[0059] Figure 2 FIG. illustrates an application scenario diagram of an offshore wind turbine vibration control system based on a magnetorheological tuned mass damper according to an embodiment of the present application. As Figure 2 shown, in this application scenario, first, multiple sensors (for example, C as Figure 2 illustrated) are used to obtain wind loads, wave loads, and seismic loads at multiple predetermined time points within a predetermined time period, the magnetic field control intensity values of the MRE-TMD device (for example, M as Figure 2 illustrated), and the displacement response values of the wind turbine structure of the offshore wind turbine (for example, U as Figure 2 illustrated) at the multiple predetermined time points; then, the obtained wind loads, wave loads, seismic loads, magnetic field control intensity values of the MRE-TMD device, and displacement response values of the wind turbine structure of the offshore wind turbine are input into a server (for example, as Figure 2In the schematic diagram (S), the server processes the wind load, wave load, seismic load, magnetic field control intensity value of the MRE-TMD device, and displacement response value of the fan structure of the offshore wind turbine by using an offshore wind turbine vibration control algorithm based on a magnetorheological tuned mass damper, so as to generate a classification result indicating whether the magnetic field control intensity value of the MRE-TMD device at the current time point should be increased or decreased.

[0060] After introducing the basic principle of the present application, various non-limiting embodiments of the present application will be specifically introduced with reference to the accompanying drawings.

[0061] Exemplary System

[0062] Figure 3 The block diagram of an offshore wind turbine vibration control system based on a magnetorheological tuned mass damper according to an embodiment of the present application is illustrated. As Figure 3 shown, the offshore wind turbine vibration control system 100 based on a magnetorheological tuned mass damper according to an embodiment of the present application includes: a data acquisition module 110, configured to obtain the wind load, wave load, and seismic load at a plurality of predetermined time points within a predetermined time period, the magnetic field control intensity value of the MRE-TMD device at the plurality of predetermined time points, and the displacement response value of the fan structure of the offshore wind turbine at the plurality of predetermined time points; a load data time series feature extraction module 120, configured to arrange the wind load, wave load, and seismic load at the plurality of predetermined time points within the predetermined time period as input vectors respectively, and then obtain first to third load feature vectors through a time series encoder including a one-dimensional convolutional layer; a load feature fusion module 130, configured to fuse the first to third load feature vectors to obtain a global load feature vector; a magnetic field data and displacement data encoding module 140, configured to arrange the magnetic field control intensity value of the MRE-TMD device at the plurality of predetermined time points and the displacement response value of the fan structure of the offshore wind turbine at the plurality of predetermined time points as input vectors respectively, and then obtain a magnetic field control feature vector and a displacement response feature vector through the time series encoder including the one-dimensional convolutional layer; a Bayesian inference module 150, configured to use a Bayesian probability model to fuse the magnetic field control feature vector, the displacement response feature vector, and the global load feature vector to obtain a posterior probability feature vector; a posterior distribution correction module 160, configured to correct the feature values at each position in the posterior probability feature vector based on the mean and variance of the set of feature values at all positions in the posterior probability vector to obtain a corrected posterior probability feature vector; and a vibration control result generation module 170, configured to pass the corrected posterior probability feature vector through a classifier to obtain a classification result, where the classification result is used to indicate whether the magnetic field control intensity value of the MRE-TMD device at the current time point should be increased or decreased.

[0063] Specifically, in the embodiment of the present application, the data acquisition module 110 is used to obtain the wind load, wave load and seismic load at multiple predetermined time points within a predetermined time period, the magnetic field control intensity value of the MRE-TMD device at the multiple predetermined time points, and the displacement response value of the wind turbine structure of the offshore wind turbine at the multiple predetermined time points. As mentioned above, since the elastic modulus of the magnetorheological elastomer (MRE) can change with the change of the magnetic field, it can immediately return to the initial state when the magnetic field is removed. Based on this magnetic control characteristic, magnetorheological elastomer has been widely used in variable stiffness intelligent vibration absorbers. The TMD device consists of a mass, a damper and a spring attached to the structure. The MRE-TMD is a TMD device based on MRE material, and its stiffness and damping are provided by the MRE. The main feature of the device is to change the stiffness of the MRE by changing the external magnetic field and then adjust the frequency of the MRE-TMD device so that its frequency follows the main frequency of the wind turbine structure, thereby achieving the purpose of reducing structural vibration through the vibration energy consumption of the mass block when external excitation acts.

[0064] Based on this, the inventors of the present application considered changing the stiffness of the MRE by adjusting the magnetic field control intensity of the MRE-TMD device, and then adjusting the frequency of the MRE-TMD device so that its frequency follows the main frequency of the wind turbine structure to reduce vibration. In addition, the inventors of the present application also considered that in this process, it is also necessary to explore the implicit correlation characteristics of the external wind, wave, and earthquake loads in the time dimension, so as to integrate the dynamic change characteristic information of the three loads in the time series to improve the effect of vibration reduction. At the same time, the vibration change characteristic information of the offshore wind turbine structure should also be added in this process to refer to the results of vibration reduction, so as to accurately control the vibration of the dynamic response of the wind turbine and ensure the safe operation of the offshore wind turbine.

[0065] Therefore, in the technical solution of the present application, firstly, the wind load, wave load and seismic load at multiple predetermined time points within a predetermined time period, the magnetic field control intensity value of the MRE-TMD device at the multiple predetermined time points, and the displacement response value of the wind turbine structure of the offshore wind turbine at the multiple predetermined time points are obtained.

[0066] Specifically, in the embodiments of the present application, the load data time-series feature extraction module 120 is configured to arrange the wind load, wave load, and seismic load at multiple predetermined time points within the predetermined time period into input vectors respectively, and then use a time-series encoder including a one-dimensional convolutional layer to obtain the first to third load feature vectors. It can be understood that the wind load, the wave load, and the seismic load are all correlated in the time dimension. In a specific example of the present application, the time-series encoder is composed of alternately arranged fully connected layers and one-dimensional convolutional layers, and it extracts the correlations of the wind load, the wave load, and the seismic load in the time series dimension through one-dimensional convolutional encoding and extracts the high-dimensional implicit features of the wind load, the wave load, and the seismic load through fully connected encoding.

[0067] In order to fully extract the dynamic change implicit features of each load, after arranging the wind load, wave load, and seismic load at multiple predetermined time points within the predetermined time period into input vectors respectively, a time-series encoder including a one-dimensional convolutional layer is used to encode the first to third input vectors of these three loads respectively, so as to extract the dynamic change implicit features of the wind load in the time dimension, the dynamic change implicit features of the wave load in the time dimension, and the dynamic change implicit features of the seismic load in the time dimension respectively, thereby obtaining the first to third load feature vectors.

[0068] More specifically, in the embodiments of the present application, Figure 4 The figure illustrates a block diagram of the load data time-series feature extraction module in the offshore wind turbine vibration control system based on a magnetorheological tuned mass damper according to an embodiment of the present application, as Figure 4 shown, the load data time-series feature extraction module 120 includes: an input vector construction unit 210, configured to arrange the wind load, wave load, and seismic load at multiple predetermined time points within the predetermined time period into input vectors according to the time dimension respectively to obtain a first input vector, a second input vector, and a third input vector; a fully connected encoding unit 220, configured to use the fully connected layer of the time-series encoder to perform fully connected encoding on the first input vector, the second input vector, and the third input vector respectively according to the following formula to extract the high-dimensional implicit features of the feature values at each position in the first input vector, the second input vector, and the third input vector respectively, where the formula is: where X is the first input vector, the second input vector, and the third input vector, Y is the output vector, W is the weight matrix, and B is the bias vector. representing matrix multiplication; and, a one-dimensional convolutional encoding unit 230, configured to respectively perform one-dimensional convolutional encoding on the first input vector, the second input vector, and the third input vector by using the one-dimensional convolutional layer of the temporal encoder to respectively extract high-dimensional implicit correlation features between eigenvalue positions in the first input vector, the second input vector, and the third input vector, where the formula is:

[0069]

[0070] where a is the width of the convolutional kernel in the x direction, F(a) is the convolutional kernel parameter vector, G(x - a) is the local vector matrix operating with the convolutional kernel function, w is the size of the convolutional kernel, and X represents the first input vector, the second input vector, and the third input vector.

[0071] Specifically, in the embodiments of the present application, the load feature fusion module 130 is configured to fuse the first to third load feature vectors to obtain a global load feature vector. That is, after obtaining the first to third load feature vectors, further fuse the dynamic correlation features of these three loads in time to obtain a global load feature vector.

[0072] It should be understood that since the wind load, the wave load, and the seismic load have different degrees of influence on the vibration of the offshore wind turbine, in order to pay attention to the vibration influence degree of each load feature during feature fusion, the first to third load feature vectors are further pre-classified through a pre-classifier to obtain the first to third probability values. In this way, the obtained first to third probability values can be used as weights to weight the first to third load feature vectors, and then calculate the weighted sum between the weighted load feature vectors to obtain the global load feature vector.

[0073] More specifically, in the embodiments of the present application, Figure 5 The figure illustrates a block diagram of the load feature fusion module in the offshore wind turbine vibration control system based on a magnetorheological tuned mass damper according to an embodiment of the present application, as Figure 5 shown, the load feature fusion module 130 includes: a pre-classification unit 310, configured to respectively pass the first to third load feature vectors through a pre-classifier to obtain the first to third probability values; and, a fusion unit 320, calculating the position-wise weighted sum of the first to third weighted load feature vectors with the first to third probability values as weights to obtain the global load feature vector.

[0074] Specifically, in the embodiments of the present application, the magnetic field data and displacement data encoding module 140 is configured to arrange the magnetic field control intensity values of the MRE-TMD device at the multiple predetermined time points and the displacement response values of the fan structure of the offshore wind turbine at the multiple predetermined time points as input vectors respectively, and then obtain a magnetic field control feature vector and a displacement response feature vector through the time series encoder including a one-dimensional convolutional layer. For the magnetic field control intensity values of the MRE-TMD device at the multiple predetermined time points and the displacement response values of the fan structure of the offshore wind turbine at the multiple predetermined time points, similarly, they also have dynamic correlation implicit feature information in terms of time series.

[0075] In a specific example of the present application, the time series encoder is composed of alternately arranged fully connected layers and one-dimensional convolutional layers. It extracts the correlation in the time series dimension of the magnetic field control intensity values of the MRE-TMD device at the multiple predetermined time points and the displacement response values of the fan structure of the offshore wind turbine at the multiple predetermined time points through one-dimensional convolutional encoding, and extracts the high-dimensional implicit features of the magnetic field control intensity values of the MRE-TMD device at the multiple predetermined time points and the displacement response values of the fan structure of the offshore wind turbine at the multiple predetermined time points through fully connected encoding.

[0076] Furthermore, the magnetic field control intensity values of the MRE-TMD device at the multiple predetermined time points and the displacement response values of the fan structure of the offshore wind turbine at the multiple predetermined time points are respectively arranged as input vectors and then encoded through the time series encoder including a one-dimensional convolutional layer to respectively extract the dynamic implicit correlation features in the time series dimension of the magnetic field control intensity value of the MRE-TMD device and the displacement response value of the fan structure of the offshore wind turbine, so as to obtain a magnetic field control feature vector and a displacement response feature vector.

[0077] More specifically, in the embodiments of the present application, Figure 6 The block diagram of the magnetic field data and displacement data encoding module in the offshore wind turbine vibration control system based on a magnetorheological tuned mass damper according to the embodiments of the present application is illustrated, as Figure 6 shown, the magnetic field data and displacement data encoding module 140 includes: an arrangement unit 410, configured to arrange the magnetic field control intensity values of the MRE-TMD device at the multiple predetermined time points and the displacement response values of the fan structure of the offshore wind turbine at the multiple predetermined time points as input vectors according to the time dimension respectively to obtain a control intensity input vector and a displacement response input vector; an implicit feature extraction unit 420, configured to use the fully connected layer of the time series encoder to perform fully connected encoding on the control intensity input vector and the displacement response input vector respectively according to the following formula to respectively extract the high-dimensional implicit features of the feature values at each position in the control intensity input vector and the displacement response input vector, where the formula is: where X is the control intensity input vector and the displacement response input vector, Y is the output vector, W is the weight matrix, and B is the bias vector. denotes matrix multiplication; and an associated feature extraction unit 430, configured to use the one-dimensional convolutional layer of the time series encoder to perform one-dimensional convolutional encoding on the control intensity input vector and the displacement response input vector respectively according to the following formula to extract the high-dimensional implicit association features between the eigenvalue positions in the control intensity input vector and the displacement response input vector respectively, where the formula is:

[0078]

[0079] where a is the width of the convolutional kernel in the x direction, F(a) is the convolutional kernel parameter vector, G(x - a) is the local vector matrix operating with the convolutional kernel function, w is the size of the convolutional kernel, and X represents the control intensity input vector and the displacement response input vector.

[0080] Specifically, in the embodiments of the present application, the Bayesian inference module 150 is configured to use a Bayesian probability model to fuse the magnetic field control feature vector, the displacement response feature vector, and the global load feature vector to obtain a posterior probability feature vector. It should be understood that in the technical solution of the present application, considering that when the wind load, wave load, and seismic load change, the magnetic field control intensity of the MRE-TMD device is adjusted for vibration reduction, and the vibration change characteristic information represented by the displacement response value of the offshore wind turbine structure is introduced during this process to represent the vibration change characteristic for reference of the vibration reduction result.

[0081] Therefore, using the magnetic field control feature vector as the prior probability, in the technical solution of the present application, when there are new changes in the wind load, wave load, and seismic load, the prior probability is updated to obtain the posterior probability. Among them, the magnetic field control feature vector is the prior information, the displacement response feature vector is the event probability, and the global load feature vector is the evidence probability.

[0082] According to Bayes' formula, the posterior probability is the prior probability multiplied by the event probability divided by the evidence probability. The purpose in the technical solution of the present application is to use a Bayesian probability model to fuse the magnetic field control feature vector, the displacement response feature vector, and the global load feature vector to obtain a posterior probability feature vector when there is new evidence, that is, when there are new changes in the wind load, wave load, and seismic load. Among them, the magnetic field control feature vector is used as the prior, the displacement response feature vector is used as the event, and the global load feature vector is used as the evidence. In this way, the posterior probability feature vector can be passed through a classifier to obtain the value indicating whether the magnetic field control intensity of the MRE-TMD device at the current time point should be increased or decreased.

[0083] More specifically, a Bayesian probability model is used to fuse the magnetic field control feature vector, the displacement response feature vector, and the global load feature vector according to the following formula to obtain the posterior probability feature vector; where the formula is:

[0084] qi = pi * ai / bi

[0085] Where pi is the eigenvalue at each position in the magnetic field control feature vector, ai and bi are the eigenvalues at each position in the displacement response feature vector and the global load feature vector respectively, and qi is the eigenvalue at each position in the posterior probability feature vector.

[0086] Specifically, in the embodiment of the present application, the posterior distribution correction module 160 is configured to correct the eigenvalues at each position in the posterior probability feature vector based on the mean and variance of the set of eigenvalues at all positions in the posterior probability vector to obtain the corrected posterior probability feature vector.

[0087] In particular, in the technical solution of the present application, when using the Bayesian probability model to fuse the magnetic field control feature vector, the displacement response feature vector, and the global load feature vector, since the global load feature vector is obtained by passing the load correlation feature in the time series direction through a convolutional neural network as a filter, it may not be completely aligned with the magnetic field control feature vector and the displacement response feature vector in the time series direction. And due to the per-position calculation characteristic of the Bayesian probability model, the obtained posterior probability feature vector may contain special eigenvalue points due to misaligned features, thus affecting the classification performance of the feature distribution of the posterior probability feature vector.

[0088] Therefore, information statistical normalization of adaptive instances is performed on the posterior probability feature vector, that is, based on the mean and variance of the set of eigenvalues at all positions in the posterior probability vector, the eigenvalues at each position in the posterior probability feature vector are corrected according to the following formula to obtain the corrected posterior probability feature vector; where the formula is:

[0089]

[0090] Where f i represents the eigenvalue at each position in the posterior probability feature vector, and μ and σ represent the mean and variance of the set of eigenvalues at all positions in the posterior probability vector respectively, L is the length of the posterior probability feature vector, and α is a hyperparameter.

[0091] Here, the information statistics normalization of the adaptive instance uses the set of eigenvalues of the posterior probability feature vector as the adaptive instance, and uses the essential intrinsic prior information of its statistical features to perform dynamic generative information normalization on a single eigenvalue. At the same time, the normalized modulus length information of the feature set is used as a bias to serve as an invariance description within the set distribution domain. In this way, the optimization of the feature distribution that shields the perturbation distribution of special instances as much as possible is achieved, thereby improving the classification accuracy. In this way, the dynamic response of the wind turbine can be accurately vibration-controlled to ensure the safe operation of the offshore wind turbine.

[0092] Specifically, in the embodiment of the present application, the vibration control result generation module 170 is configured to pass the corrected posterior probability feature vector through a classifier to obtain a classification result, and the classification result is used to indicate whether the magnetic field control intensity value of the MRE-TMD device at the current time point should be increased or decreased. That is, the corrected posterior probability feature vector is input into a classification function to obtain a classification function value, where the classification function value is the classification result, and the classification result is used to indicate whether the magnetic field control intensity value of the MRE-TMD device at the current time point should be increased or decreased.

[0093] Further, the classifier is used to process the corrected posterior probability feature vector according to the following formula to obtain the classification result, where the formula is:

[0094] softmax{(W n ,B n ):…:(W1,B1)|X}, where W1 to W n are weight matrices, B1 to B n are bias vectors, and X is the corrected posterior probability feature vector.

[0095] In summary, the offshore wind turbine vibration control system 100 based on the magnetorheological tuned mass damper according to the embodiment of the present application is elucidated. It extracts the dynamic change implicit features in the time dimension of the wind load, wave load, and seismic load at multiple predetermined time points within a predetermined time period through a time series encoder to obtain a global load feature vector, and extracts the dynamic implicit correlation features in the time series dimension of the magnetic field control intensity value of the MRE-TMD device and the displacement response value of the wind turbine structure of the offshore wind turbine at multiple predetermined time points to obtain a magnetic field control feature vector and a displacement response feature vector. Then, the above three feature vectors are respectively fused and corrected to obtain a corrected posterior probability feature vector, and finally the corrected posterior probability feature vector is passed through a classifier to obtain a classification result. In this way, the dynamic response of the wind turbine can be accurately vibration-controlled to ensure the safe operation of the offshore wind turbine.

[0096] As described above, the offshore wind turbine vibration control system 100 based on a magnetorheological tuned mass damper according to an embodiment of the present application can be implemented in various terminal devices, such as a server for offshore wind turbine vibration control based on a magnetorheological tuned mass damper. In one example, the offshore wind turbine vibration control system 100 based on a magnetorheological tuned mass damper according to an embodiment of the present application can be integrated into the terminal device as a software module and / or a hardware module. For example, the offshore wind turbine vibration control system 100 based on a magnetorheological tuned mass damper can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the offshore wind turbine vibration control system 100 based on a magnetorheological tuned mass damper can also be one of many hardware modules of the terminal device.

[0097] Alternatively, in another example, the offshore wind turbine vibration control system 100 based on a magnetorheological tuned mass damper and the terminal device can also be separate devices, and the offshore wind turbine vibration control system 100 based on a magnetorheological tuned mass damper can be connected to the terminal device through a wired and / or wireless network and transmit interaction information in accordance with a predefined data format.

[0098] Exemplary Method

[0099] Figure 7 The flowchart of the offshore wind turbine vibration control method based on a magnetorheological tuned mass damper according to an embodiment of the present application is illustrated. As Figure 7As shown, the method for controlling the vibration of an offshore wind turbine based on a magneto-rheological tuned mass damper according to an embodiment of the present application includes: S110, obtaining the wind load, wave load, and seismic load at a plurality of predetermined time points within a predetermined time period, the magnetic field control intensity values of the MRE-TMD device at the plurality of predetermined time points, and the displacement response values of the wind turbine structure of the offshore wind turbine at the plurality of predetermined time points; S120, arranging the wind load, wave load, and seismic load at the plurality of predetermined time points within the predetermined time period as input vectors respectively, and obtaining first to third load feature vectors through a time series encoder including a one-dimensional convolutional layer; S130, fusing the first to third load feature vectors to obtain a global load feature vector; S140, arranging the magnetic field control intensity values of the MRE-TMD device at the plurality of predetermined time points and the displacement response values of the wind turbine structure of the offshore wind turbine at the plurality of predetermined time points as input vectors respectively, and obtaining a magnetic field control feature vector and a displacement response feature vector through the time series encoder including the one-dimensional convolutional layer; S150, using a Bayesian probability model to fuse the magnetic field control feature vector, the displacement response feature vector, and the global load feature vector to obtain a posterior probability feature vector; S160, based on the mean and variance of the set of eigenvalue at all positions in the posterior probability vector, correcting the eigenvalue at each position in the posterior probability feature vector to obtain a corrected posterior probability feature vector; and S170, passing the corrected posterior probability feature vector through a classifier to obtain a classification result, where the classification result is used to indicate whether the magnetic field control intensity value of the MRE-TMD device at the current time point should be increased or decreased.

[0100] Figure 8 FIG. illustrates a schematic architecture diagram of a method for controlling the vibration of an offshore wind turbine based on a magneto-rheological tuned mass damper according to an embodiment of the present application. As Figure 8As shown, in the network architecture of the offshore wind turbine vibration control method based on the magnetorheological tuned mass damper, first, wind loads, wave loads, and seismic loads at multiple predetermined time points within a predetermined time period are obtained, the magnetic field control intensity values of the MRE-TMD device at the multiple predetermined time points, and the displacement response values of the wind turbine structure of the offshore wind turbine at the multiple predetermined time points; then, the wind loads, wave loads, and seismic loads at the multiple predetermined time points within the predetermined time period are respectively arranged as input vectors and passed through a time series encoder including a one-dimensional convolutional layer to obtain first to third load feature vectors; then, the first to third load feature vectors are fused to obtain a global load feature vector; then, the magnetic field control intensity values of the MRE-TMD device at the multiple predetermined time points and the displacement response values of the wind turbine structure of the offshore wind turbine at the multiple predetermined time points are respectively arranged as input vectors and passed through the time series encoder including the one-dimensional convolutional layer to obtain a magnetic field control feature vector and a displacement response feature vector; then, a Bayesian probability model is used to fuse the magnetic field control feature vector, the displacement response feature vector, and the global load feature vector to obtain a posterior probability feature vector; then, based on the mean and variance of the eigenvalue sets at all positions in the posterior probability vector, the eigenvalues at each position in the posterior probability feature vector are corrected to obtain a corrected posterior probability feature vector; and finally, the corrected posterior probability feature vector is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether the magnetic field control intensity value of the MRE-TMD device at the current time point should be increased or decreased.

[0101] In an embodiment of the present application, in the above-mentioned offshore wind turbine vibration control method based on the magnetorheological tuned mass damper, the step of arranging the wind loads, wave loads, and seismic loads at multiple predetermined time points within the predetermined time period as input vectors respectively and passing them through a time series encoder including a one-dimensional convolutional layer to obtain first to third load feature vectors includes: arranging the wind loads, wave loads, and seismic loads at multiple predetermined time points within the predetermined time period as input vectors along the time dimension respectively to obtain a first input vector, a second input vector, and a third input vector; using the fully connected layer of the time series encoder to perform fully connected encoding on the first input vector, the second input vector, and the third input vector respectively according to the following formula to extract the high-dimensional hidden features of the eigenvalues at each position in the first input vector, the second input vector, and the third input vector, where the formula is: where X is the first input vector, the second input vector, and the third input vector, Y is the output vector, W is the weight matrix, and B is the bias vector. represent matrix multiplication; and, use the one-dimensional convolutional layer of the said time series encoder to perform one-dimensional convolutional encoding on the said first input vector, the said second input vector, and the said third input vector respectively according to the following formula to extract the high-dimensional implicit correlation features between the eigenvalue at each position in the said first input vector, the said second input vector, and the said third input vector respectively, where the said formula is:

[0102]

[0103] where a is the width of the convolutional kernel in the x direction, F(a) is the convolutional kernel parameter vector, G(x - a) is the local vector matrix operating with the convolutional kernel function, w is the size of the convolutional kernel, and X represents the said first input vector, the said second input vector, and the said third input vector.

[0104] In an embodiment of the present application, in the above-mentioned vibration control method of an offshore wind turbine based on a magnetorheological tuned mass damper, the step of fusing the first to third load feature vectors to obtain a global load feature vector includes: passing the first to third load feature vectors through a pre-classifier respectively to obtain first to third probability values; and, using the first to third probability values as weights to calculate the position-wise weighted sum of the first to third weighted load feature vectors to obtain the global load feature vector.

[0105] In an embodiment of the present application, in the above-mentioned vibration control method of an offshore wind turbine based on a magnetorheological tuned mass damper, the step of arranging the magnetic field control intensity values of the MRE-TMD device at the said multiple predetermined time points and the displacement response values of the wind turbine structure of the offshore wind turbine at the said multiple predetermined time points as input vectors respectively and then passing them through the time series encoder including a one-dimensional convolutional layer to obtain a magnetic field control feature vector and a displacement response feature vector includes: arranging the magnetic field control intensity values of the MRE-TMD device at the said multiple predetermined time points and the displacement response values of the wind turbine structure of the offshore wind turbine at the said multiple predetermined time points as input vectors respectively according to the time dimension to obtain a control intensity input vector and a displacement response input vector; using the fully connected layer of the time series encoder to perform fully connected encoding on the control intensity input vector and the displacement response input vector respectively according to the following formula to extract the high-dimensional implicit features of the eigenvalue at each position in the control intensity input vector and the displacement response input vector respectively, where the said formula is: where X is the control intensity input vector and the displacement response input vector, Y is the output vector, W is the weight matrix, and B is the bias vector, Denote matrix multiplication; and, use the one-dimensional convolutional layer of the timing encoder to perform one-dimensional convolutional encoding on the control intensity input vector and the displacement response input vector respectively according to the following formula to extract the high-dimensional implicit correlation features between the eigenvalues at each position in the control intensity input vector and the displacement response input vector, where the formula is:

[0106]

[0107] where a is the width of the convolutional kernel in the x direction, F(a) is the convolutional kernel parameter vector, G(x - a) is the local vector matrix operating with the convolutional kernel function, w is the size of the convolutional kernel, and X represents the control intensity input vector and the displacement response input vector.

[0108] In an embodiment of the present application, in the above-mentioned offshore wind turbine vibration control method based on a magnetorheological tuned mass damper, the use of the Bayesian probability model to fuse the magnetic field control feature vector, the displacement response feature vector, and the global load feature vector to obtain the posterior probability feature vector includes: using the Bayesian probability model according to the following formula to fuse the magnetic field control feature vector, the displacement response feature vector, and the global load feature vector to obtain the posterior probability feature vector; where the formula is:

[0109] qi = pi * ai / bi

[0110] where pi are the eigenvalues at each position in the magnetic field control feature vector, ai and bi are the eigenvalues at each position in the displacement response feature vector and the global load feature vector respectively, and qi are the eigenvalues at each position in the posterior probability feature vector.

[0111] In an embodiment of the present application, in the above-mentioned offshore wind turbine vibration control method based on a magnetorheological tuned mass damper, the correction of the eigenvalues at each position in the posterior probability feature vector based on the mean and variance of the eigenvalue set at all positions in the posterior probability vector to obtain the corrected posterior probability feature vector includes: based on the mean and variance of the eigenvalue set at all positions in the posterior probability vector, using the following formula to correct the eigenvalues at each position in the posterior probability feature vector to obtain the corrected posterior probability feature vector; where the formula is:

[0112]

[0113] where f i represents the eigenvalues at each position in the posterior probability feature vector, and μ and σ represent the mean and variance of the eigenvalue set at all positions in the posterior probability vector respectively, L is the length of the posterior probability feature vector, and α is a hyperparameter.

[0114] In one embodiment of the present application, in the above-mentioned vibration control method of an offshore wind turbine based on a magnetorheological tuned mass damper, the step of obtaining a classification result by passing the corrected posterior probability feature vector through a classifier, where the classification result is used to indicate whether the magnetic field control intensity value of the MRE-TMD device at the current time point should be increased or decreased, includes: processing the corrected posterior probability feature vector using the classifier according to the following formula to obtain the classification result, where the formula is:

[0115] softmax{(W n ,B n ):…:(W1,B1)|X}, where W1 to W n are weight matrices, B1 to B n are bias vectors, and X is the corrected posterior probability feature vector.

[0116] Here, those skilled in the art can understand that the specific functions and operations in the above-mentioned vibration control method of an offshore wind turbine based on a magnetorheological tuned mass damper have been described in detail in the description of the offshore wind turbine vibration control system based on a magnetorheological tuned mass damper above, and therefore, the repeated description thereof will be omitted. Figures 2 to 6 Figures 2 to 6

Claims

1. An offshore wind turbine vibration control system based on a magnetorheological tuned mass damper, characterized in that, Including: A data acquisition module, configured to obtain wind loads, wave loads, and seismic loads at multiple predetermined time points within a predetermined time period, magnetic field control intensity values of an MRE-TMD device at the multiple predetermined time points, and displacement response values of a wind turbine structure of an offshore wind turbine at the multiple predetermined time points; A load data time series feature extraction module, configured to respectively arrange the wind loads, wave loads, and seismic loads at the multiple predetermined time points within the predetermined time period as input vectors, and then obtain first to third load feature vectors through a time series encoder including a one-dimensional convolutional layer; A load feature fusion module, configured to fuse the first to third load feature vectors to obtain a global load feature vector; A magnetic field data and displacement data encoding module, configured to respectively arrange the magnetic field control intensity values of the MRE-TMD device at the multiple predetermined time points and the displacement response values of the wind turbine structure of the offshore wind turbine at the multiple predetermined time points as input vectors, and then obtain a magnetic field control feature vector and a displacement response feature vector through the time series encoder including the one-dimensional convolutional layer; A Bayesian inference module, configured to use a Bayesian probability model to fuse the magnetic field control feature vector, the displacement response feature vector, and the global load feature vector to obtain a posterior probability feature vector; A posterior distribution correction module, configured to correct the feature values at each position in the posterior probability feature vector based on the mean and variance of the set of feature values at all positions in the posterior probability vector to obtain a corrected posterior probability feature vector; and A vibration control result generation module, configured to pass the corrected posterior probability feature vector through a classifier to obtain a classification result, where the classification result is used to indicate whether the magnetic field control intensity value of the MRE-TMD device at the current time point should be increased or decreased.

2. The vibration control system of an offshore wind turbine based on a magnetorheological tuned mass damper according to claim 1, wherein The load data time series feature extraction module includes: An input vector construction unit, configured to respectively arrange the wind loads, wave loads, and seismic loads at the multiple predetermined time points within the predetermined time period as input vectors according to the time dimension to obtain a first input vector, a second input vector, and a third input vector; A fully-connected encoding unit is configured to respectively perform fully-connected encoding on the first input vector, the second input vector, and the third input vector by using the fully-connected layer of the temporal encoder according to the following formula to respectively extract high-dimensional implicit features of the feature values at each position in the first input vector, the second input vector, and the third input vector, where the formula is: where X is the first input vector, the second input vector, and the third input vector, Y is the output vector, W is the weight matrix, and B is the bias vector, represents matrix multiplication; and A one-dimensional convolutional encoding unit, configured to respectively perform one-dimensional convolutional encoding on the first input vector, the second input vector, and the third input vector by using the one-dimensional convolutional layer of the time series encoder according to the following formula to respectively extract high-dimensional implicit association features between the feature values at each position in the first input vector, the second input vector, and the third input vector, where the formula is: where a is the width of the convolutional kernel in the x direction, F(a) is the convolutional kernel parameter vector, G(x - a) is the local vector matrix operating with the convolutional kernel function, w is the size of the convolutional kernel, and X represents the first input vector, the second input vector, and the third input vector.

3. The vibration control system of an offshore wind turbine based on a magnetorheological tuned mass damper according to claim 2, wherein, The load feature fusion module includes: A pre-classification unit, configured to respectively pass the first to third load feature vectors through a pre-classifier to obtain first to third probability values; and A fusion unit that calculates the position-weighted sum of the first to third weighted load feature vectors using the first to third probability values as weights to obtain the global load feature vector.

4. The vibration control system of an offshore wind turbine based on a magnetorheological tuned mass damper according to claim 3, characterized in that, The magnetic field data and displacement data encoding module includes: An arrangement unit configured to arrange the magnetic field control intensity values of the MRE-TMD device at the plurality of predetermined time points and the displacement response values of the wind turbine structure of the offshore wind turbine at the plurality of predetermined time points as input vectors in the time dimension to obtain a control intensity input vector and a displacement response input vector; An implicit feature extraction unit, which is used to respectively perform fully connected encoding on the control intensity input vector and the displacement response input vector by using the fully connected layer of the temporal encoder according to the following formula to respectively extract the high-dimensional implicit features of the eigenvalue at each position in the control intensity input vector and the displacement response input vector, where the formula is: where X is the control intensity input vector and the displacement response input vector, Y is the output vector, W is the weight matrix, and B is the bias vector, represents matrix multiplication; and An associated feature extraction unit configured to use the one-dimensional convolutional layer of the time series encoder to perform one-dimensional convolutional encoding on the control intensity input vector and the displacement response input vector respectively according to the following formula to extract the high-dimensional implicit association features between the eigenvalue at each position in the control intensity input vector and the displacement response input vector, where the formula is: where a is the width of the convolutional kernel in the x direction, F(a) is the convolutional kernel parameter vector, G(x - a) is the local vector matrix operated on by the convolutional kernel function, w is the size of the convolutional kernel, and X represents the control intensity input vector and the displacement response input vector.

5. The vibration control system of an offshore wind turbine based on a magnetorheological tuned mass damper according to claim 4, characterized in that, The Bayesian inference module is further configured to: use the Bayesian probability model to fuse the magnetic field control feature vector, the displacement response feature vector, and the global load feature vector according to the following formula to obtain the posterior probability feature vector; where the formula is: qi = pi * ai / bi where pi is the eigenvalue at each position in the magnetic field control feature vector, ai and bi are the eigenvalues at each position in the displacement response feature vector and the global load feature vector respectively, and qi is the eigenvalue at each position in the posterior probability feature vector.

6. The vibration control system of an offshore wind turbine based on a magnetorheological tuned mass damper according to claim 5, wherein The posterior distribution correction module is further configured to: based on the mean and variance of the eigenvalue set at all positions in the posterior probability vector, correct the eigenvalue at each position in the posterior probability feature vector according to the following formula to obtain the corrected posterior probability feature vector; where the formula is: where f i represents the eigenvalues at each position in the posterior probability feature vector, and μ and σ respectively represent the mean and variance of the set of eigenvalues at all positions in the posterior probability vector, L is the length of the posterior probability feature vector, and α is a hyperparameter.

7. The vibration control system of an offshore wind turbine based on a magnetorheological tuned mass damper according to claim 6, characterized in that, The vibration control result generation module is further configured to: use the classifier to process the corrected posterior probability feature vector according to the following formula to obtain the classification result, where the formula is: softmax{(W n ,B n ):…:(W1,B1)|X}, where W1 to W n are weight matrices, B1 to B n are bias vectors, and X is the corrected posterior probability feature vector.

Images