Online Monitoring System and Method for the Inclination of Offshore Wind Turbine Towers

Abstract

This application relates to the field of intelligent monitoring of offshore wind turbine towers. Specifically, it discloses an online monitoring system and method for the inclination of offshore wind turbine towers. By adopting artificial intelligence monitoring technology and using a deep neural network model as a feature extractor, it dynamically and implicitly extracts associated features of the uneven settlement inclination angle of the foundation and the tower inclination angle of each offshore wind turbine in an offshore wind farm at multiple predetermined time points. And by taking the inclination dynamic characteristics of other wind turbine towers as a reference, it evaluates the overturning risk of the tower of the wind turbine to be monitored, thereby avoiding the introduction of personal cognitive biases by artificially setting thresholds, and improving the accuracy of the early warning of the inclination risk of the offshore wind turbine tower.

Description

Technical Field

[0001] The present invention relates to the field of intelligent monitoring of offshore wind turbine towers, and more specifically, to an online monitoring system and method for the inclination of offshore wind turbine towers. Background Art

[0002] With the expansion of the new energy market and the continuous growth of the installed capacity of wind power generation, the construction of wind power projects is gradually developing towards the third and fourth types of inland regions. The towers and blades of wind turbines are getting taller and longer. Major accidents such as the inclination and collapse of wind turbine towers occur frequently, posing a serious threat to the safe production of power generation enterprises. The deterioration processes such as the loosening of the foundation of wind turbines, uneven settlement, deformation of the tower, and continuous cracking are the essential hidden dangers causing the overturning accidents of the tower.

[0003] Some existing solutions use a monitoring system to monitor offshore wind turbines to determine whether dangerous accidents will occur. However, due to the diversity and instability of the systems, whether the monitoring data of the systems is consistent with the actual situation needs to be verified through long-term application, resulting in problems that cannot be predicted and discovered in advance. Therefore, an optimized online monitoring scheme for the inclination of offshore wind turbine towers is expected to generate an early warning when an overturning risk is detected. Summary of the Invention

[0004] To solve the above technical problems, the present application is proposed. Embodiments of the present application provide an online monitoring system and method for the inclination of offshore wind turbine towers. By using the monitoring technology of artificial intelligence and using a deep neural network model as a feature extractor, dynamic implicit correlation features of the foundation uneven settlement inclination angles and tower inclination angles of each offshore wind turbine in an offshore wind turbine array at multiple predetermined time points are extracted, and the tower overturning risk of the offshore wind turbine to be monitored is evaluated by referring to the inclination dynamic features of other wind turbine towers, thereby avoiding the introduction of personal cognitive biases by artificially setting thresholds and improving the accuracy of the early warning of the inclination risk of the offshore wind turbine tower.

[0005] According to one aspect of the present application, an online monitoring system for the inclination of an offshore wind turbine tower is provided, which includes:

[0006] A tower inclination monitoring data acquisition module for obtaining the foundation uneven settlement inclination angles and tower inclination angles of each offshore wind turbine in an offshore wind turbine array at multiple predetermined time points;

[0007] A tower inclination data feature extraction module for obtaining an inclination state correlation feature matrix corresponding to each offshore wind turbine by passing the foundation uneven settlement inclination angles and tower inclination angles of each offshore wind turbine at multiple predetermined time points through a deep neural network model as a feature extractor;

[0008] The object to be monitored extraction module is used to extract the tilt state correlation feature matrix of the offshore wind turbine to be monitored from the tilt state correlation feature matrices corresponding to each offshore wind turbine;

[0009] The difference module is used to calculate the differences between the tilt state correlation feature matrix of the offshore wind turbine to be monitored and the tilt state correlation feature matrices of other offshore wind turbines respectively to obtain a plurality of difference matrices;

[0010] The eigenvalue correction module is used to perform eigenvalue correction based on the maximum eigenvalue on each of the plurality of difference matrices to obtain a plurality of corrected difference feature matrices;

[0011] The aggregation encoding module is used to aggregate the plurality of corrected difference feature matrices into a three-dimensional input tensor and then pass it through a second convolutional neural network model as a feature extractor to obtain a classification feature map; and

[0012] The monitoring result generation module is used to pass the classification feature map through a classifier to obtain a classification result, and the classification result is used to indicate whether the tower barrel overturning risk of the offshore wind turbine to be monitored exceeds a predetermined threshold.

[0013] In the above-mentioned online monitoring system for the tilt of the tower barrel of an offshore wind turbine, the tower barrel tilt data feature extraction module includes: a vector construction unit for arranging the foundation uneven settlement tilt angles and tower barrel tilt angles of each offshore wind turbine at a plurality of predetermined time points in the time dimension as a first input vector and a second input vector respectively; a time series encoding unit for passing the first input vector and the second input vector through the time series encoding module including a one-dimensional convolutional layer of the deep neural network model to obtain a foundation uneven settlement tilt angle transformation feature vector and a tower barrel tilt angle change feature vector respectively; a feature vector correlation unit for calculating the product between the transposed vector of the foundation uneven settlement tilt angle transformation feature vector and the tower barrel tilt angle change feature vector to obtain a tilt state correlation matrix; and a local feature extraction unit for passing the tilt state correlation matrix through the first convolutional neural network of the deep neural network model to obtain the tilt state correlation feature matrix.

[0014] In the above-mentioned online monitoring system for the tilt of the tower barrel of an offshore wind turbine, the time series encoding unit is further used to: use the fully connected layer of the time series encoding module including a one-dimensional convolutional layer of the deep neural network model to perform fully connected encoding on the first input vector and the second input vector respectively according to the following formula to extract the high-dimensional implicit features of the eigenvalues at each position in the first input vector and the second input vector respectively, where the formula is: where X is the input vector, Y is the output vector, W is the weight matrix, and B is the bias vector, Denote matrix multiplication; use the one-dimensional convolutional layer of the time series encoding module including the one-dimensional convolutional layer of the deep neural network model to perform one-dimensional convolutional encoding on the first input vector and the second input vector respectively according to the following formula to extract the high-dimensional implicit correlation features between the eigenvalue at each position in the first input vector and the second input vector, where the formula is:

[0015]

[0016] Where a is the width of the convolutional kernel in the x direction, F is the convolutional kernel parameter vector, G is the local vector matrix for the convolutional kernel function operation, w is the size of the convolutional kernel, and X represents the input vector.

[0017] In the above-mentioned online monitoring system for the inclination of the off-shore wind turbine tower, the local feature extraction unit is further configured to: each layer of the first convolutional neural network of the deep neural network model respectively performs the following operations on the input data during the forward propagation of the layer: perform convolutional processing on the input data to obtain a convolutional feature map; perform mean pooling based on the local channel dimension on the convolutional feature map to obtain a pooled feature map; and perform non-linear activation on the pooled feature map to obtain an activation feature map; where the output of the last layer of the first convolutional neural network of the deep neural network model is the inclination state correlation feature matrix, and the input of the first layer of the first convolutional neural network of the deep neural network model is the inclination state correlation matrix.

[0018] In the above-mentioned online monitoring system for the inclination of the off-shore wind turbine tower, the difference module is further configured to: calculate the difference between the inclination state correlation feature matrix of the off-shore wind turbine to be monitored and the inclination state correlation feature matrices of other off-shore wind turbines respectively according to the following formula to obtain the multiple difference matrices;

[0019] Where the formula is:

[0020]

[0021] Where M represents the inclination state correlation feature matrix of the off-shore wind turbine to be monitored, and M i represents the inclination state correlation feature matrices of other off-shore wind turbines except for the inclination state correlation feature matrix of the off-shore wind turbine to be monitored, represents the difference by position.

[0022] In the above-mentioned online monitoring system for the inclination of the off-shore wind turbine tower, the eigenvalue correction module is further configured to: perform eigenvalue correction based on the maximum eigenvalue on each of the multiple difference matrices respectively according to the following formula to obtain the multiple corrected difference feature matrices;

[0023] Where the formula is:

[0024]

[0025] Among them, M represents each difference matrix in the multiple difference matrices, and m max represents the maximum eigenvalue of the eigenvalues at each position in the difference matrix, and ⊙ represents pointwise multiplication by position.

[0026] In the above-mentioned online monitoring system for the tilt of the tower barrel of an offshore wind turbine, the aggregation encoding module is further configured to: each layer of the second convolutional neural network model serving as a feature extractor respectively performs the following operations on the input data during the forward propagation of the layer: performing convolutional processing on the input data to obtain a convolutional feature map; performing average pooling processing on the convolutional feature map to obtain a pooled feature map; and performing non-linear activation on the pooled feature map to obtain an activation feature map; wherein, the output of the last layer of the second convolutional neural network model serving as a feature extractor is the classification feature map, and the input of the first layer of the second convolutional neural network model serving as a feature extractor is the three-dimensional input tensor.

[0027] In the above-mentioned online monitoring system for the tilt of the tower barrel of an offshore wind turbine, the monitoring result generation module is further configured to: the classifier processes the classification feature map according to the following formula to generate a classification result, where the formula is: softmax{(W n , B n ):…:(W 1 , B 1 )|Project(F)}, where Project(F) represents projecting the classification feature map array into a vector, and W 1 to W n are the weight matrices of each fully connected layer, and B 1 to B n represent the bias matrices of each fully connected layer.

[0028] According to another aspect of the present application, an online monitoring method for the tilt of the tower barrel of an offshore wind turbine includes:

[0029] Obtaining the foundation uneven settlement tilt angles and tower barrel tilt angles of each offshore wind turbine in an offshore wind turbine array at multiple predetermined time points;

[0030] Passing the foundation uneven settlement tilt angles and tower barrel tilt angles of each offshore wind turbine at multiple predetermined time points through a deep neural network model serving as a feature extractor to obtain a tilt state correlation feature matrix corresponding to each offshore wind turbine;

[0031] Extracting the tilt state correlation feature matrix of the offshore wind turbine to be monitored from the tilt state correlation feature matrices corresponding to each offshore wind turbine;

[0032] Calculate the differences between the tilt state correlation feature matrix of the offshore wind turbine to be monitored and the tilt state correlation feature matrices of other offshore wind turbines respectively to obtain a plurality of difference matrices;

[0033] Perform eigenvalue correction based on the maximum eigenvalue on each of the plurality of difference matrices to obtain a plurality of corrected difference feature matrices;

[0034] Aggregate the plurality of corrected difference feature matrices into a three-dimensional input tensor and then pass it through a second convolutional neural network model serving as a feature extractor to obtain a classification feature map; and

[0035] Pass the classification feature map through a classifier to obtain a classification result, and the classification result is used to indicate whether the tower overturning risk of the offshore wind turbine to be monitored exceeds a predetermined threshold.

[0036] In the above-mentioned online monitoring method for the tilt of the offshore wind turbine tower, passing the foundation uneven settlement tilt angle and the tower tilt angle of each offshore wind turbine at a plurality of predetermined time points through a deep neural network model serving as a feature extractor to obtain a tilt state correlation feature matrix corresponding to each offshore wind turbine includes: arranging the foundation uneven settlement tilt angle and the tower tilt angle of each offshore wind turbine at a plurality of predetermined time points into a first input vector and a second input vector respectively according to the time dimension; passing the first input vector and the second input vector through a time series encoding module including a one-dimensional convolutional layer of the deep neural network model to obtain a foundation uneven settlement tilt angle transformation feature vector and a tower tilt angle change feature vector; calculating the product between the transposed vector of the foundation uneven settlement tilt angle transformation feature vector and the tower tilt angle change feature vector to obtain a tilt state correlation matrix; and passing the tilt state correlation matrix through the first convolutional neural network of the deep neural network model to obtain the tilt state correlation feature matrix.

[0037] In the above-mentioned online monitoring method for the tilt of the offshore wind turbine tower, passing the first input vector and the second input vector through a time series encoding module including a one-dimensional convolutional layer of the deep neural network model to obtain a foundation uneven settlement tilt angle transformation feature vector and a tower tilt angle change feature vector includes: using the fully connected layer of the time series encoding module including a one-dimensional convolutional layer of the deep neural network model to perform fully connected encoding on the first input vector and the second input vector respectively according to the following formula to extract the high-dimensional implicit features of the eigenvalues at each position in the first input vector and the second input vector respectively, where the formula is: where X is the input vector, Y is the output vector, W is the weight matrix, and B is the bias vector, Denotes matrix multiplication; use the one-dimensional convolutional layer of the time series encoding module including a one-dimensional convolutional layer of the deep neural network model to perform one-dimensional convolutional encoding on the first input vector and the second input vector respectively according to the following formula to extract the high-dimensional implicit correlation features between the eigenvalue at each position in the first input vector and the second input vector, where the formula is:

[0038]

[0039] Where a is the width of the convolutional kernel in the x direction, F is the convolutional kernel parameter vector, G is the local vector matrix for the convolutional kernel function operation, w is the size of the convolutional kernel, and X represents the input vector.

[0040] In the above-mentioned online monitoring method for the tilt of the off-shore wind turbine tower, passing the tilt state correlation matrix through the first convolutional neural network of the deep neural network model to obtain the tilt state correlation feature matrix includes: each layer of the first convolutional neural network of the deep neural network model performs the following operations on the input data respectively during the forward pass of the layer: performing convolutional processing on the input data to obtain a convolutional feature map; performing mean pooling based on the local channel dimension on the convolutional feature map to obtain a pooled feature map; and performing non-linear activation on the pooled feature map to obtain an activated feature map; where the output of the last layer of the first convolutional neural network of the deep neural network model is the tilt state correlation feature matrix, and the input of the first layer of the first convolutional neural network of the deep neural network model is the tilt state correlation matrix.

[0041] In the above-mentioned online monitoring method for the tilt of the off-shore wind turbine tower, calculating the difference between the tilt state correlation feature matrix of the to-be-monitored off-shore wind turbine and the tilt state correlation feature matrices of other off-shore wind turbines respectively to obtain a plurality of difference matrices includes: calculating the difference between the tilt state correlation feature matrix of the to-be-monitored off-shore wind turbine and the tilt state correlation feature matrices of other off-shore wind turbines respectively according to the following formula to obtain the plurality of difference matrices;

[0042] Where the formula is:

[0043]

[0044] Where M represents the tilt state correlation feature matrix of the to-be-monitored off-shore wind turbine, M i represents the tilt state correlation feature matrices of other off-shore wind turbines except for the tilt state correlation feature matrix of the to-be-monitored off-shore wind turbine, represents the difference by position.

[0045] In the above-mentioned online monitoring method for the inclination of the offshore wind turbine tower, eigenvalue correction based on the maximum eigenvalue is performed on each of the multiple difference matrices to obtain multiple corrected differential eigenmatrices, including: performing eigenvalue correction based on the maximum eigenvalue on each of the multiple difference matrices according to the following formula to obtain the multiple corrected differential eigenmatrices;

[0046] wherein, the formula is:

[0047]

[0048] where M represents each of the multiple difference matrices, and m max represents the maximum eigenvalue of the eigenvalues at each position in the difference matrix, and ⊙ represents element-wise multiplication.

[0049] In the above-mentioned online monitoring method for the inclination of the offshore wind turbine tower, after aggregating the multiple corrected differential eigenmatrices into a three-dimensional input tensor, the classification feature map is obtained through the second convolutional neural network model serving as a feature extractor, including: each layer of the second convolutional neural network model serving as a feature extractor performs the following operations on the input data during the forward pass of the layer: performing convolutional processing on the input data to obtain a convolutional feature map; performing average pooling processing on the convolutional feature map to obtain a pooled feature map; and performing non-linear activation on the pooled feature map to obtain an activation feature map; wherein, the output of the last layer of the second convolutional neural network model serving as a feature extractor is the classification feature map, and the input of the first layer of the second convolutional neural network model serving as a feature extractor is the three-dimensional input tensor.

[0050] In the above-mentioned online monitoring method for the inclination of the offshore wind turbine tower, passing the classification feature map through a classifier to obtain a classification result, including: the classifier processes the classification feature map according to the following formula to generate a classification result, wherein, the formula is: softmax{(W n ,B n ):…:(W 1 ,B 1 )|Project(F)}, where Project(F) represents projecting the classification feature map matrix into a vector, and W 1 to W n are the weight matrices of each fully connected layer, and B 1 to B n represent the bias matrices of each fully connected layer.

[0051] Compared with the prior art, the on-line monitoring system and method for the inclination of an off-shore wind turbine tower provided by the present application adopt an artificial intelligence monitoring technology, use a deep neural network model as a feature extractor to dynamically and implicitly extract associated features of the uneven settlement inclination angles and tower inclination angles of each off-shore wind turbine in an off-shore wind turbine array at multiple predetermined time points, and evaluate the tipping risk of the tower of the off-shore wind turbine to be monitored by taking the inclination dynamic features of other wind turbine towers as a reference, thereby avoiding the introduction of personal cognitive biases by artificially setting thresholds and improving the accuracy of the early warning of the inclination risk of the off-shore wind turbine tower. Description of the Drawings

[0052] By describing the embodiments of the present application in more detail in conjunction with the drawings, the above and other objects, features and advantages of the present application will become more obvious. The drawings are used to provide a further understanding of the embodiments of the present application, and constitute a part of the specification, and are used to explain the present application together with the embodiments of 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.

[0053] Figure 1 FIG. is an application scenario diagram of an on-line monitoring system for the inclination of an off-shore wind turbine tower according to an embodiment of the present application.

[0054] Figure 2 FIG. is a block diagram of an on-line monitoring system for the inclination of an off-shore wind turbine tower according to an embodiment of the present application.

[0055] Figure 3 FIG. is a block diagram of a tower inclination data feature extraction module in an on-line monitoring system for the inclination of an off-shore wind turbine tower according to an embodiment of the present application.

[0056] Figure 4 FIG. is a flowchart of an on-line monitoring method for the inclination of an off-shore wind turbine tower according to an embodiment of the present application.

[0057] Figure 5 FIG. is a schematic structural diagram of an on-line monitoring method for the inclination of an off-shore wind turbine tower according to an embodiment of the present application. Detailed Embodiments

[0058] Next, exemplary embodiments according to the present application will be described in detail with reference to the drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0059] Scene Overview

[0060] As mentioned above, with the expansion of the new energy market and the continuous growth of the installed capacity of wind power generation, the construction of wind power projects is gradually developing towards the third and fourth types of inland regions. The tower barrels and blades of wind turbines are getting taller and longer, and major accidents such as the inclination and collapse of wind turbine tower barrels occur frequently, posing a serious threat to the safe production of power generation enterprises. The deterioration processes such as the loosening of the foundation of wind turbines, uneven settlement, deformation of the tower barrel, and continuous cracking are the essential hidden dangers causing the tower barrel overturning accidents.

[0061] Some existing solutions are to monitor offshore wind turbines through a monitoring system to determine whether dangerous accidents will occur. However, due to the diversity and instability of the systems, whether the system monitoring data is consistent with the actual situation requires long-term application verification, resulting in problems that cannot be predicted and discovered in advance. Therefore, an optimized online monitoring scheme for the inclination of offshore wind turbine tower barrels is expected to generate an early warning when an overturning risk is detected.

[0062] Based on this, the inventors of this application considered that when monitoring the overturning risk of the tower barrel of an offshore wind turbine, the deterioration processes such as the loosening of the foundation of the wind turbine, uneven settlement, deformation of the tower barrel, and continuous cracking are the essential hidden dangers causing the tower barrel overturning accident. Therefore, in the technical solution of this application, it is expected to comprehensively judge the risk warning of the tower barrel overturning through the inclination angle of the uneven settlement of the foundation of the offshore wind turbine and the inclination angle of the tower barrel, so as to improve the accuracy of the judgment. Moreover, the risk of the tower barrel overturning of the offshore wind turbine to be monitored is also evaluated by referring to the dynamic inclination characteristics of other wind turbine tower barrels, which can avoid introducing personal cognitive biases by artificially setting thresholds, thereby improving the accuracy of risk warning.

[0063] Specifically, in the technical solution of this application, first, the inclination angle of the uneven settlement of the foundation and the inclination angle of the tower barrel of each offshore wind turbine in the offshore wind turbine array are respectively collected through a foundation horizontal sensor and an inclination sensor at multiple predetermined time points.

[0064] Then, considering that the inclination angle of the uneven settlement of the foundation and the inclination angle of the tower barrel of each offshore wind turbine have dynamic implicit correlation information in the time series dimension, therefore, in order to fully extract the deep dynamic implicit correlation features of these two inclination angles to evaluate and monitor the overturning risk of the tower barrel of the offshore wind turbine, the inclination angle of the uneven settlement of the foundation and the inclination angle of the tower barrel of each offshore wind turbine at multiple predetermined time points are further passed through a deep neural network model as a feature extractor to obtain an inclination state correlation feature matrix corresponding to each offshore wind turbine.

[0065] In particular, in a specific example of the present application, the deep neural network model serving as a feature extractor includes a time series encoding module containing a one-dimensional convolutional layer and a first convolutional neural network. Among them, the time series encoding module containing the one-dimensional convolutional layer can extract the dynamic change features of the foundation uneven settlement inclination angle and the tower barrel inclination angle of each offshore wind turbine in the time series dimension. It is composed of alternately arranged fully connected layers and one-dimensional convolutional layers. Through one-dimensional convolutional encoding, it respectively extracts the correlations of the foundation uneven settlement inclination angle and the tower barrel inclination angle of each offshore wind turbine in the time series dimension, and through fully connected encoding, it respectively extracts the high-dimensional implicit features of the foundation uneven settlement inclination angle and the tower barrel inclination angle of each offshore wind turbine. Then, the transposed vector of the obtained foundation uneven settlement inclination angle transformation feature vector and the tower barrel inclination angle change feature vector are multiplied vectorially to integrate the dynamic change features of the foundation uneven settlement inclination angle and the dynamic change features of the tower barrel inclination angle, thereby obtaining an inclination state correlation matrix. Further, the first convolutional neural network of the deep neural network model is used to extract the implicit correlation features at each position in the inclination state correlation matrix to more suitably represent the inclination state correlation feature information of each offshore wind turbine, thereby obtaining the inclination state correlation feature matrix.

[0066] When detecting the tower barrel overturning risk of the offshore wind turbine to be monitored, first, the inclination state correlation feature matrix of the offshore wind turbine to be monitored is extracted from the inclination state correlation feature matrices corresponding to each offshore wind turbine. Then, it should be understood that in order to evaluate the tower barrel overturning risk of the offshore wind turbine to be monitored with reference to the inclination dynamic features of other wind turbine towers, to avoid introducing personal cognitive biases by artificially setting thresholds, and thus to improve the accuracy of risk assessment and judgment, the differences between the inclination state correlation feature matrix of the offshore wind turbine to be monitored and the inclination state correlation feature matrices of other offshore wind turbines are further calculated respectively to obtain a plurality of difference matrices.

[0067] Furthermore, the plurality of difference matrices are aggregated into a three-dimensional input tensor and then feature extraction is performed through a second convolutional neural network model serving as a feature extractor to extract the inclination dynamic implicit correlation change features between the tower barrel of the offshore wind turbine to be monitored and the tower barrels of other wind turbines, thereby obtaining a classification feature map. Then, the classification feature map is passed through a classifier to obtain a classification result indicating whether the tower barrel overturning risk of the offshore wind turbine to be monitored exceeds a predetermined threshold.

[0068] However, since the tilt state associated feature matrix is the association matrix of the foundation uneven settlement tilt angle transformation feature vector and the tower barrel tilt angle change feature vector, that is, each position of the matrix has a phase attribute relative to the overall feature distribution. This makes it possible that there may be a negative impact on the class probability aggregation due to the phase difference between the positions of the difference matrix when calculating the difference matrix, resulting in a poor classification effect of the classification feature map obtained when performing feature extraction after splicing and aggregating it.

[0069] Therefore, in the technical solution of the present application, preferably, the eigenvalue representation aggregation is first performed on the difference matrix, specifically:

[0070]

[0071] where M represents each difference matrix in the multiple difference matrices, and m max represents the maximum eigenvalue of the eigenvalues at each position in the difference matrix, and ⊙ represents point multiplication by position.

[0072] Here, the eigenvalue representation aggregation introduces the wave function representation of the feature set, that is, the amplitude represents the intensity information, and the phase represents the periodic position information, to perform the aggregation of the information representation of the difference matrix in the complex function domain of the class, thereby making up for the negative impact caused by the phase difference between the positions of the difference feature matrix in the class probability aggregation (that is, in-phase enhancement and out-of-phase cancellation starting from the wave function principle). Thus, after arranging it into a three-dimensional tensor and passing it through the second convolutional neural network model as the feature extractor, the classification effect of the classification feature map is improved, and further the classification accuracy is improved.

[0073] Based on this, the present application proposes an on-line monitoring system for the inclination of an off-shore wind turbine tower, which includes: a tower inclination monitoring data acquisition module for obtaining the uneven settlement inclination angles of the foundations and the tower inclination angles of each off-shore wind turbine in an off-shore wind turbine array at multiple predetermined time points; a tower inclination data feature extraction module for passing the uneven settlement inclination angles of the foundations and the tower inclination angles of each off-shore wind turbine at multiple predetermined time points through a deep neural network model serving as a feature extractor to obtain an inclination state correlation feature matrix corresponding to each off-shore wind turbine; a to-be-monitored object extraction module for extracting the inclination state correlation feature matrix of the to-be-monitored off-shore wind turbine from the inclination state correlation feature matrices corresponding to each off-shore wind turbine; a difference module for respectively calculating the differences between the inclination state correlation feature matrix of the to-be-monitored off-shore wind turbine and the inclination state correlation feature matrices of other off-shore wind turbines to obtain a plurality of difference matrices; an eigenvalue correction module for respectively performing eigenvalue correction based on the maximum eigenvalue on each of the plurality of difference matrices to obtain a plurality of corrected difference feature matrices; an aggregation encoding module for aggregating the plurality of corrected difference feature matrices into a three-dimensional input tensor and then passing it through a second convolutional neural network model serving as a feature extractor to obtain a classification feature map; and a monitoring result generation module for passing the classification feature map through a classifier to obtain a classification result, and the classification result is used to indicate whether the tower overturning risk of the to-be-monitored off-shore wind turbine exceeds a predetermined threshold.

[0074] Figure 1 FIG. illustrates an application scenario diagram of the on-line monitoring system for the inclination of an off-shore wind turbine tower according to an embodiment of the present application. As Figure 1 shown, in this application scenario, first, the uneven settlement inclination angles of the foundations and the tower inclination angles of each off-shore wind turbine in an off-shore wind turbine array (e.g., F as shown in Figure 1 ) are respectively collected by a foundation level sensor (e.g., T1 as illustrated in Figure 1 ) and an inclination sensor (e.g., T2 as illustrated in Figure 1 ) at multiple predetermined time points. Then, the obtained uneven settlement inclination angles of the foundations and the tower inclination angles of each off-shore wind turbine at multiple predetermined time points are input into a server deployed with an on-line monitoring algorithm for the inclination of an off-shore wind turbine tower (e.g., as shown in Figure 1 the F1-Fn in ) of each off-shore wind turbine in the off-shore wind turbine array (e.g., as shown in Figure 1 the B in ) of the foundation (e.g., as shown in Figure 1 the R in ) of the tower (e.g., as shown in Figure 1The cloud server S) shown in the figure, where the server can process the foundation uneven settlement tilt angle and tower barrel tilt angle of each offshore wind turbine at multiple predetermined time points by using the offshore wind turbine tower barrel tilt online monitoring algorithm to generate a classification result indicating whether the tower barrel overturning risk of the offshore wind turbine to be monitored exceeds a predetermined threshold.

[0075] 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.

[0076] Exemplary System

[0077] Figure 2 The block diagram of the offshore wind turbine tower barrel tilt online monitoring system according to an embodiment of the present application is illustrated. As Figure 2 shown, the offshore wind turbine tower barrel tilt online monitoring system 200 according to an embodiment of the present application includes: a tower barrel tilt monitoring data acquisition module 210 for obtaining the foundation uneven settlement tilt angle and tower barrel tilt angle of each offshore wind turbine in an offshore wind turbine array at multiple predetermined time points; a tower barrel tilt data feature extraction module 220 for obtaining, through a deep neural network model as a feature extractor, the foundation uneven settlement tilt angle and tower barrel tilt angle of each offshore wind turbine at multiple predetermined time points to obtain a tilt state correlation feature matrix corresponding to each offshore wind turbine; a monitoring object extraction module 230 for extracting the tilt state correlation feature matrix of the offshore wind turbine to be monitored from the tilt state correlation feature matrix corresponding to each offshore wind turbine; a difference module 240 for respectively calculating the difference between the tilt state correlation feature matrix of the offshore wind turbine to be monitored and the tilt state correlation feature matrix of other offshore wind turbines to obtain a plurality of difference matrices; an eigenvalue correction module 250 for respectively performing eigenvalue correction based on the maximum eigenvalue on each difference matrix in the plurality of difference matrices to obtain a plurality of corrected difference feature matrices; an aggregation encoding module 260 for aggregating the plurality of corrected difference feature matrices into a three-dimensional input tensor and then passing it through a second convolutional neural network model as a feature extractor to obtain a classification feature map; and a monitoring result generation module 270 for passing the classification feature map through a classifier to obtain a classification result, the classification result being used to indicate whether the tower barrel overturning risk of the offshore wind turbine to be monitored exceeds a predetermined threshold.

[0078] Specifically, in the embodiments of the present application, the tower inclination monitoring data acquisition module 210 is configured to obtain the foundation uneven settlement inclination angles and tower inclination angles of each offshore wind turbine in an offshore wind turbine array at multiple predetermined time points. As described above, when monitoring the overturning risk of the tower of an offshore wind turbine, deterioration processes such as loosening of the wind turbine foundation, uneven settlement, deformation of the tower, and continuous cracking are the essential hidden dangers causing tower overturning accidents. Therefore, in the technical solution of the present application, it is desired to comprehensively perform risk early warning judgment of the tower overturning through the foundation uneven settlement inclination angle and tower inclination angle of the offshore wind turbine to improve the accuracy of the judgment. Moreover, the risk of tower overturning of the offshore wind turbine to be monitored is evaluated by referring to the inclination dynamic characteristics of the towers of other wind turbines, which can avoid introducing personal cognitive biases by artificially setting thresholds, thereby improving the accuracy of risk early warning.

[0079] That is, specifically, in the technical solution of the present application, first, the foundation uneven settlement inclination angles and tower inclination angles of each offshore wind turbine in the offshore wind turbine array at multiple predetermined time points are respectively collected by a foundation level sensor and an inclination sensor.

[0080] Specifically, in the embodiments of the present application, the tower inclination data feature extraction module 220 is configured to pass the foundation uneven settlement inclination angles and tower inclination angles of each offshore wind turbine at multiple predetermined time points through a deep neural network model as a feature extractor to obtain an inclination state correlation feature matrix corresponding to each offshore wind turbine. It should be understood that considering that the foundation uneven settlement inclination angles and tower inclination angles of each offshore wind turbine have dynamic implicit correlation information in the time series dimension, therefore, in the technical solution of the present application, in order to fully extract the deep dynamic implicit correlation features of these two inclination angles to evaluate and monitor the tower overturning risk of the offshore wind turbine, the foundation uneven settlement inclination angles and tower inclination angles of each offshore wind turbine at multiple predetermined time points are further passed through a deep neural network model as a feature extractor to obtain an inclination state correlation feature matrix corresponding to each offshore wind turbine.

[0081] Specifically, in the embodiment of the present application, the deep neural network model serving as a feature extractor includes a time series encoding module containing a one-dimensional convolutional layer and a first convolutional neural network. Among them, the time series encoding module containing a one-dimensional convolutional layer can extract the dynamic change features of the foundation uneven settlement inclination angle and the tower barrel inclination angle of each offshore wind turbine in the time series dimension. It is composed of alternately arranged fully connected layers and one-dimensional convolutional layers. Through one-dimensional convolutional encoding, it respectively extracts the correlations of the foundation uneven settlement inclination angle and the tower barrel inclination angle of each offshore wind turbine in the time series dimension, and through fully connected encoding, it respectively extracts the high-dimensional implicit features of the foundation uneven settlement inclination angle and the tower barrel inclination angle of each offshore wind turbine. Then, the transposed vector of the obtained foundation uneven settlement inclination angle transformation feature vector and the tower barrel inclination angle change feature vector are multiplied vectorially to integrate the dynamic change features of the foundation uneven settlement inclination angle and the dynamic change features of the tower barrel inclination angle, thereby obtaining an inclination state correlation matrix. Further, the first convolutional neural network of the deep neural network model is used to extract the implicit correlation features at each position in the inclination state correlation matrix to more suitably represent the inclination state correlation feature information of each offshore wind turbine, thereby obtaining the inclination state correlation feature matrix.

[0082] More specifically, in the embodiment of the present application, the tower barrel inclination data feature extraction module includes: First, the foundation uneven settlement inclination angle and the tower barrel inclination angle of each offshore wind turbine at multiple predetermined time points are respectively arranged as a first input vector and a second input vector according to the time dimension. Then, the first input vector and the second input vector are respectively passed through the time series encoding module containing a one-dimensional convolutional layer of the deep neural network model to obtain a foundation uneven settlement inclination angle transformation feature vector and a tower barrel inclination angle change feature vector. Correspondingly, in a specific example, the fully connected layers of the time series encoding module containing a one-dimensional convolutional layer of the deep neural network model are used to perform fully connected encoding on the first input vector and the second 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 first input vector and the second input vector, where the formula is: where X is the input vector, Y is the output vector, W is the weight matrix, and B is the bias vector, represents matrix multiplication;

[0083] The one-dimensional convolutional layers of the time series encoding module containing a one-dimensional convolutional layer of the deep neural network model are used to perform one-dimensional convolutional encoding on the first input vector and the second input vector respectively according to the following formula to respectively extract the high-dimensional implicit correlation features between the feature values at each position in the first input vector and the second input vector, where the formula is:

[0084]

[0085] Among them, a is the width of the convolution kernel in the x direction, F is the convolution kernel parameter vector, G is the local vector matrix for the convolution kernel function operation, w is the size of the convolution kernel, and X represents the input vector.

[0086] Then, calculate the product between the transposed vector of the basic uneven settlement tilt angle transformation feature vector and the tower barrel tilt angle change feature vector to obtain the tilt state correlation matrix. Finally, pass the tilt state correlation matrix through the first convolutional neural network of the deep neural network model to obtain the tilt state correlation feature matrix. Correspondingly, in a specific example, each layer of the first convolutional neural network of the deep neural network model performs the following operations on the input data during the forward pass of the layer: perform convolution processing on the input data to obtain a convolution feature map; perform mean pooling based on the local channel dimension on the convolution feature map to obtain a pooled feature map; and perform non-linear activation on the pooled feature map to obtain an activation feature map; where the output of the last layer of the first convolutional neural network of the deep neural network model is the tilt state correlation feature matrix, and the input of the first layer of the first convolutional neural network of the deep neural network model is the tilt state correlation matrix.

[0087] Figure 3 The figure shows a block diagram of a tower barrel tilt data feature extraction module in an on-line monitoring system for tower barrel tilt of an offshore wind turbine according to an embodiment of the present application. As Figure 3 shown, the tower barrel tilt data feature extraction module 220 includes: a vector construction unit 221 for arranging the basic uneven settlement tilt angles and tower barrel tilt angles of each offshore wind turbine at multiple predetermined time points in the time dimension as a first input vector and a second input vector respectively; a time series encoding unit 222 for passing the first input vector and the second input vector through the time series encoding module including a one-dimensional convolutional layer of the deep neural network model to obtain a basic uneven settlement tilt angle transformation feature vector and a tower barrel tilt angle change feature vector respectively; a feature vector correlation unit 223 for calculating the product between the transposed vector of the basic uneven settlement tilt angle transformation feature vector and the tower barrel tilt angle change feature vector to obtain a tilt state correlation matrix; and a local feature extraction unit 224 for passing the tilt state correlation matrix through the first convolutional neural network of the deep neural network model to obtain the tilt state correlation feature matrix.

[0088] Specifically, in the embodiments of the present application, the object to be monitored extraction module 230 and the difference module 240 are configured to extract the tilt state correlation feature matrix of the wind turbine to be monitored from the tilt state correlation feature matrices corresponding to each offshore wind turbine, and respectively calculate the difference between the tilt state correlation feature matrix of the wind turbine to be monitored and the tilt state correlation feature matrices of other offshore wind turbines to obtain a plurality of difference matrices. That is, in the technical solution of the present application, further, when detecting the tower overturning risk of the wind turbine to be monitored, first, the tilt state correlation feature matrix of the wind turbine to be monitored is extracted from the tilt state correlation feature matrices corresponding to each offshore wind turbine. Then, it should be understood that in order to evaluate the tower overturning risk of the wind turbine to be monitored with reference to the tilt dynamic characteristics of the towers of other wind turbines, to avoid introducing personal cognitive biases by artificially setting thresholds, and thus to improve the accuracy of risk assessment and judgment, the difference between the tilt state correlation feature matrix of the wind turbine to be monitored and the tilt state correlation feature matrices of other offshore wind turbines is further calculated respectively to obtain a plurality of difference matrices.

[0089] More specifically, in the embodiments of the present application, the difference module is further configured to: calculate the difference between the tilt state correlation feature matrix of the wind turbine to be monitored and the tilt state correlation feature matrices of other offshore wind turbines respectively according to the following formula to obtain the plurality of difference matrices;

[0090] Wherein, the formula is:

[0091]

[0092] Wherein, M represents the tilt state correlation feature matrix of the wind turbine to be monitored, and M i represents the tilt state correlation feature matrices of other offshore wind turbines except the tilt state correlation feature matrix of the wind turbine to be monitored, represents differential by position.

[0093] Specifically, in the embodiments of the present application, the eigenvalue correction module 250 is used to perform eigenvalue correction based on the maximum eigenvalue on each of the multiple difference matrices to obtain multiple corrected difference feature matrices. It should be understood that since the tilt state correlation feature matrix is the correlation matrix of the foundation uneven settlement tilt angle transformation feature vector and the tower barrel tilt angle change feature vector, that is, each position of the correlation matrix has a phase attribute relative to the overall feature distribution, which makes it possible that there may be a negative impact on the class probability aggregation due to the phase difference between the positions of the difference matrix when calculating the difference matrix, resulting in a poor classification effect of the classification feature map obtained when stitching and aggregating them and then performing feature extraction. Therefore, in the technical solution of the present application, preferably, the eigenvalue representation aggregation of the difference matrix is first performed. That is, the eigenvalue representation aggregation introduces the wave function representation of the feature set, that is, the amplitude represents the intensity information, and the phase represents the periodic position information, to perform the aggregation of the information representation of the difference matrix in the complex function domain of the class, so as to make up for the negative impact caused by the phase difference between the positions of the difference feature matrix in the class probability aggregation (that is, the in-phase enhancement and out-of-phase cancellation starting from the wave function principle), so that after arranging it into a three-dimensional tensor and passing it through the second convolutional neural network model as the feature extractor, the classification effect of the classification feature map is improved, and thus the classification accuracy is improved.

[0094] More specifically, in the embodiments of the present application, the eigenvalue correction module is further configured to: perform eigenvalue correction based on the maximum eigenvalue on each of the multiple difference matrices according to the following formula to obtain the multiple corrected difference feature matrices;

[0095] Wherein, the formula is:

[0096]

[0097] Where M represents each of the multiple difference matrices, and m max represents the maximum eigenvalue of the eigenvalues of each position in the difference matrix, and ⊙ represents point multiplication by position.

[0098] Specifically, in the embodiments of the present application, the aggregation encoding module 260 and the monitoring result generation module 270 are configured to aggregate the multiple corrected differential feature matrices into a three-dimensional input tensor, and then pass the three-dimensional input tensor through a second convolutional neural network model serving as a feature extractor to obtain a classification feature map, and pass the classification feature map through a classifier to obtain a classification result, where the classification result is used to indicate whether the tower overturning risk of the offshore wind turbine to be monitored exceeds a predetermined threshold. That is, in the technical solution of the present application, further, the multiple corrected differential matrices are aggregated into a three-dimensional input tensor and then feature extraction is performed through a second convolutional neural network model serving as a feature extractor to extract the implicit dynamic correlation change features of the inclination of the tower of the offshore wind turbine to be monitored and the towers of other wind turbines, so as to obtain a classification feature map. In this way, the classification result indicating whether the tower overturning risk of the offshore wind turbine to be monitored exceeds a predetermined threshold can be obtained by passing the classification feature map through the classifier. Correspondingly, in a specific example, the classifier processes the classification feature map according to the following formula to generate a classification result, where the formula is: softmax{(W n , B n ):...:(W 1 , B 1 )|Project(F)}, where Project(F) represents projecting the classification feature map matrix into a vector, and W 1 to W n are the weight matrices of each fully connected layer, and B 1 to B n represent the bias matrices of each fully connected layer.

[0099] More specifically, in the embodiments of the present application, the aggregation encoding module is further configured to: in the forward pass of each layer of the second convolutional neural network model serving as a feature extractor, respectively perform the following operations on the input data: perform convolutional processing on the input data to obtain a convolutional feature map; perform average pooling processing on the convolutional feature map to obtain a pooled feature map; and perform non-linear activation on the pooled feature map to obtain an activation feature map; where the output of the last layer of the second convolutional neural network model serving as a feature extractor is the classification feature map, and the input of the first layer of the second convolutional neural network model serving as a feature extractor is the three-dimensional input tensor.

[0100] In summary, the online monitoring system 200 for the tilt of the offshore wind turbine tower based on the embodiments of the present application is elucidated. By adopting the monitoring technology of artificial intelligence and using the deep neural network model as a feature extractor, it dynamically and implicitly extracts the associated features of the uneven settlement tilt angle of the foundation and the tower tilt angle of each offshore wind turbine in the offshore wind turbine array at multiple predetermined time points, and evaluates the tipping risk of the tower of the offshore wind turbine to be monitored by referring to the tilt dynamic features of the towers of other wind turbines, thereby avoiding the introduction of personal cognitive biases by artificially setting thresholds, and improving the accuracy of the early warning of the tilt risk of the offshore wind turbine tower.

[0101] As described above, the online monitoring system 200 for the tilt of the offshore wind turbine tower according to the embodiments of the present application can be implemented in various terminal devices, such as a server for the online monitoring algorithm of the offshore wind turbine tower tilt. In one example, the online monitoring system 200 for the tilt of the offshore wind turbine tower according to the embodiments of the present application can be integrated into the terminal device as a software module and / or a hardware module. For example, the online monitoring system 200 for the tilt of the offshore wind turbine tower can be a software module in the operating system of the terminal device, or can be an application program developed for the terminal device; of course, the online monitoring system 200 for the tilt of the offshore wind turbine tower can also be one of the many hardware modules of the terminal device.

[0102] Alternatively, in another example, the online monitoring system 200 for the tilt of the offshore wind turbine tower and the terminal device can also be separate devices, and the online monitoring system 200 for the tilt of the offshore wind turbine tower 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.

[0103] Exemplary Method

[0104] Figure 4 The flowchart of the online monitoring method for the tilt of the offshore wind turbine tower is illustrated. As Figure 4As shown, the online monitoring method for the tilt of an offshore wind turbine tower according to an embodiment of the present application includes the steps of: S110, obtaining the uneven settlement tilt angles of the foundations and the tower tilt angles of each offshore wind turbine in an offshore wind turbine array at multiple predetermined time points; S120, passing the uneven settlement tilt angles of the foundations and the tower tilt angles of each offshore wind turbine at multiple predetermined time points through a deep neural network model serving as a feature extractor to obtain an inclination state correlation feature matrix corresponding to each offshore wind turbine; S130, extracting the inclination state correlation feature matrix of the offshore wind turbine to be monitored from the inclination state correlation feature matrix corresponding to each offshore wind turbine; S140, respectively calculating the differences between the inclination state correlation feature matrix of the offshore wind turbine to be monitored and the inclination state correlation feature matrices of other offshore wind turbines to obtain a plurality of difference matrices; S150, respectively performing eigenvalue correction based on the maximum eigenvalue on each difference matrix in the plurality of difference matrices to obtain a plurality of corrected difference feature matrices; S160, aggregating the plurality of corrected difference feature matrices into a three-dimensional input tensor and then passing it through a second convolutional neural network model serving as a feature extractor to obtain a classification feature map; and S170, passing the classification feature map through a classifier to obtain a classification result, where the classification result is used to indicate whether the tower overturning risk of the offshore wind turbine to be monitored exceeds a predetermined threshold.

[0105] Figure 5 The figure shows a schematic architecture diagram of an online monitoring method for the tilt of an offshore wind turbine tower according to an embodiment of the present application. As Figure 5 shown, in the network architecture of the online monitoring method for the tilt of the offshore wind turbine tower, first, the obtained uneven settlement tilt angles of the foundations (e.g., P1 as shown in Figure 5 ) and the tower tilt angles (e.g., P2 as shown in Figure 5 ) of each offshore wind turbine at multiple predetermined time points are passed through a deep neural network model (e.g., CNN1 as shown in Figure 5 ) serving as a feature extractor to obtain an inclination state correlation feature matrix (e.g., MF1 as shown in Figure 5 ) corresponding to each offshore wind turbine; then, the inclination state correlation feature matrix of the offshore wind turbine to be monitored is extracted from the inclination state correlation feature matrix corresponding to each offshore wind turbine (e.g., MF2 as shown in Figure 5 ); then, the differences between the inclination state correlation feature matrix of the offshore wind turbine to be monitored and the inclination state correlation feature matrices of other offshore wind turbines are respectively calculated to obtain a plurality of difference matrices (e.g., M1 as shown in Figure 5 ); then, eigenvalue correction based on the maximum eigenvalue is respectively performed on each difference matrix in the plurality of difference matrices to obtain a plurality of corrected difference feature matrices (e.g., as shown in Figure 5the M2) shown; then, aggregate the multiple corrected differential feature matrices into a three-dimensional input tensor (e.g., the T shown in Figure 5 ), and then pass it through a second convolutional neural network model as a feature extractor (e.g., the CNN2 shown in Figure 5 ) to obtain a classification feature map (e.g., the FC shown in Figure 5 ); and finally, pass the classification feature map through a classifier (e.g., the classifier shown in Figure 5 ) to obtain a classification result, and the classification result is used to indicate whether the tower overturning risk of the offshore wind turbine to be monitored exceeds a predetermined threshold.

[0106] More specifically, in steps S110 and S120, obtain the foundation uneven settlement inclination angles and tower inclination angles of each offshore wind turbine in the offshore wind turbine array at multiple predetermined time points, and pass the foundation uneven settlement inclination angles and tower inclination angles of each offshore wind turbine at multiple predetermined time points through a deep neural network model as a feature extractor to obtain an inclination state correlation feature matrix corresponding to each offshore wind turbine. It should be understood that when monitoring the tower overturning risk of an offshore wind turbine, the deterioration processes such as the loosening of the wind turbine foundation, uneven settlement, tower deformation, and continuous cracking are the essential hidden dangers causing tower overturning accidents. Therefore, in the technical solution of this application, it is desired to comprehensively perform risk early warning judgment on the tower overturning through the foundation uneven settlement inclination angle and tower inclination angle of the offshore wind turbine to improve the accuracy of the judgment. And, the tower overturning risk of the offshore wind turbine to be monitored is also evaluated by referring to the inclination dynamic characteristics of other wind turbine towers, which can avoid introducing personal cognitive biases by artificially setting thresholds, thereby improving the accuracy of risk early warning. That is, specifically, in the technical solution of this application, first, collect the foundation uneven settlement inclination angles and tower inclination angles of each offshore wind turbine in the offshore wind turbine array at multiple predetermined time points through a foundation level sensor and an inclination sensor respectively.

[0107] Then, it should be understood that considering that the foundation uneven settlement inclination angles and tower inclination angles of each offshore wind turbine have dynamic implicit correlation information in the time series dimension, therefore, in the technical solution of this application, in order to fully extract the deep dynamic implicit correlation features of these two inclination angles to evaluate and monitor the tower overturning risk of the offshore wind turbine, further pass the foundation uneven settlement inclination angles and tower inclination angles of each offshore wind turbine at multiple predetermined time points through a deep neural network model as a feature extractor to obtain an inclination state correlation feature matrix corresponding to each offshore wind turbine.

[0108] In particular, in the embodiments of the present application, the deep neural network model serving as a feature extractor includes a time series encoding module containing a one-dimensional convolutional layer and a first convolutional neural network. Among them, the time series encoding module containing the one-dimensional convolutional layer can extract the dynamic change features of the foundation uneven settlement inclination angle and the tower barrel inclination angle of each offshore wind turbine in the time series dimension. It is composed of alternately arranged fully connected layers and one-dimensional convolutional layers. It respectively extracts the correlation of the foundation uneven settlement inclination angle and the tower barrel inclination angle of each offshore wind turbine in the time series dimension through one-dimensional convolutional encoding and extracts the high-dimensional implicit features of the foundation uneven settlement inclination angle and the tower barrel inclination angle of each offshore wind turbine through fully connected encoding. Then, the transposed vector of the obtained foundation uneven settlement inclination angle transformation feature vector and the tower barrel inclination angle change feature vector are multiplied vectorially to integrate the dynamic change features of the foundation uneven settlement inclination angle and the dynamic change features of the tower barrel inclination angle, thereby obtaining an inclination state correlation matrix. Further, the first convolutional neural network of the deep neural network model is used to extract the implicit correlation features at each position in the inclination state correlation matrix to more suitably represent the inclination state correlation feature information of each offshore wind turbine, thereby obtaining the inclination state correlation feature matrix.

[0109] More specifically, in steps S130 and S140, the inclination state correlation feature matrix of the offshore wind turbine to be monitored is extracted from the inclination state correlation feature matrices corresponding to each offshore wind turbine, and the differences between the inclination state correlation feature matrix of the offshore wind turbine to be monitored and the inclination state correlation feature matrices of other offshore wind turbines are calculated respectively to obtain a plurality of difference matrices. That is, in the technical solution of the present application, further, when detecting the tower barrel overturning risk of the offshore wind turbine to be monitored, first, the inclination state correlation feature matrix of the offshore wind turbine to be monitored is extracted from the inclination state correlation feature matrices corresponding to each offshore wind turbine. Then, it should be understood that in order to evaluate the tower barrel overturning risk of the offshore wind turbine to be monitored with reference to the inclination dynamic features of the tower barrels of other wind turbines, to avoid introducing personal cognitive biases by artificially setting thresholds, and further improve the accuracy of risk assessment and judgment, the differences between the inclination state correlation feature matrix of the offshore wind turbine to be monitored and the inclination state correlation feature matrices of other offshore wind turbines are calculated respectively to obtain a plurality of difference matrices.

[0110] More specifically, in step S150, eigenvalue correction based on the maximum eigenvalue is performed on each of the plurality of difference matrices to obtain a plurality of corrected difference feature matrices. It should be understood that since the tilt state correlation feature matrix is the correlation matrix of the foundation uneven settlement tilt angle transformation feature vector and the tower barrel tilt angle change feature vector, that is, each position of the correlation matrix has a phase attribute relative to the overall feature distribution. This makes it possible that there may be a negative impact on class probability aggregation due to the phase difference between the positions of the difference matrix when calculating the difference matrix, resulting in a poor classification effect of the classification feature map obtained when performing feature extraction after splicing and aggregating it. Therefore, in the technical solution of this application, preferably, eigenvalue representation aggregation is first performed on the difference matrix. That is, the eigenvalue representation aggregation introduces the wave function representation of the feature set, that is, the amplitude represents the intensity information, and the phase represents the periodic position information, to perform aggregation in the complex function domain of the information representation of the difference matrix, thereby compensating for the negative impact caused by the phase difference between the positions of the difference feature matrix on class probability aggregation (that is, in-phase enhancement and out-of-phase cancellation based on the wave function principle). Thus, after arranging it into a three-dimensional tensor and passing it through the second convolutional neural network model as the feature extractor, the classification effect of the classification feature map is improved, and further, the classification accuracy is improved.

[0111] More specifically, in steps S160 and S170, the plurality of corrected difference feature matrices are aggregated into a three-dimensional input tensor and then passed through the second convolutional neural network model as the feature extractor to obtain a classification feature map, and the classification feature map is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether the tower barrel overturning risk of the offshore wind turbine to be monitored exceeds a predetermined threshold. That is, in the technical solution of this application, further, the plurality of corrected difference matrices are aggregated into a three-dimensional input tensor and then feature extraction is performed in the second convolutional neural network model as the feature extractor to extract the implicit correlation change feature of the tilt dynamics between the tower barrel of the offshore wind turbine to be monitored and the tower barrels of other wind turbines, so as to obtain a classification feature map. In this way, the classification result that can be used to indicate whether the tower barrel overturning risk of the offshore wind turbine to be monitored exceeds a predetermined threshold can be obtained by passing the classification feature map through the classifier.

[0112] In summary, the online monitoring method for the tilt of the off - shore wind turbine tower based on the embodiments of the present application is elucidated. By adopting the monitoring technology of artificial intelligence and using the deep neural network model as a feature extractor, it dynamically and implicitly extracts the associated features of the uneven settlement tilt angle of the foundation and the tower tilt angle of each off - shore wind turbine in the off - shore wind turbine array at multiple predetermined time points. And by taking the tilt dynamic features of other wind turbine towers as a reference, it evaluates the overturning risk of the tower of the wind turbine to be monitored, thereby avoiding the introduction of personal cognitive biases by artificially setting thresholds, and improving the accuracy of the early warning of the tilt risk of the off - shore wind turbine tower.

[0113] The basic principles of the present application have been described above in conjunction with specific embodiments. However, it should be noted that the advantages, benefits, effects, etc. mentioned in the present application are only examples and not limitations. It cannot be considered that these advantages, benefits, effects, etc. are essential for each embodiment of the present application. In addition, the above - disclosed specific details are only for illustrative and easy - to - understand purposes, rather than limitations. These details do not limit the present application to necessarily adopt these specific details for implementation.

[0114] The block diagrams of the devices, apparatuses, equipment, and systems involved in the present application are only illustrative examples and do not intend to require or imply that they must be connected, arranged, and configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, equipment, and systems can be connected, arranged, and configured in any way. Words such as "including", "comprising", "having", etc. are open - ended terms, meaning "including but not limited to", and can be used interchangeably with each other. The word "or" and "and" used herein refer to the word "and / or" and can be used interchangeably with it, unless the context clearly indicates otherwise. The word "such as" used herein refers to the phrase "such as but not limited to" and can be used interchangeably with it.

[0115] It should also be noted that in the devices, equipment, and methods of the present application, each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be regarded as equivalent solutions of the present application.

[0116] The above description of the disclosed aspects enables any person skilled in the art to make or use the present application. Various modifications to these aspects are very obvious to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of the present application. Therefore, the present application is not intended to be limited to the aspects shown herein, but rather to the broadest scope consistent with the principles and novel features disclosed herein.

[0117] The foregoing description has been presented for purposes of illustration and description. In addition, the description is not intended to limit embodiments of the present application to the form disclosed herein. Although several example aspects and embodiments have been discussed above, those skilled in the art will recognize some of their variations, modifications, alterations, additions, and subcombinations.

Claims

1. An on-line monitoring system for the inclination of an off-shore wind turbine tower Characterized in that Comprising: A tower inclination monitoring data acquisition module for obtaining the foundation uneven settlement inclination angles and tower inclination angles of each off-shore wind turbine in an off-shore wind turbine array at multiple predetermined time points; A tower inclination data feature extraction module for obtaining an inclination state correlation feature matrix corresponding to each off-shore wind turbine by using the foundation uneven settlement inclination angles and tower inclination angles of each off-shore wind turbine at multiple predetermined time points through a deep neural network model serving as a feature extractor; A to-be-monitored object extraction module for extracting the inclination state correlation feature matrix of the to-be-monitored off-shore wind turbine from the inclination state correlation feature matrices corresponding to each off-shore wind turbine; A difference module for respectively calculating the differences between the inclination state correlation feature matrix of the to-be-monitored off-shore wind turbine and the inclination state correlation feature matrices of other off-shore wind turbines to obtain a plurality of difference matrices; An eigenvalue correction module for respectively performing eigenvalue correction based on the maximum eigenvalue on each difference matrix in the plurality of difference matrices to obtain a plurality of corrected difference feature matrices; An aggregation coding module for aggregating the plurality of corrected difference feature matrices into a three-dimensional input tensor and then passing it through a second convolutional neural network model serving as a feature extractor to obtain a classification feature map; And A monitoring result generation module for obtaining a classification result by passing the classification feature map through a classifier, and the classification result is used to indicate whether the tower overturning risk of the to-be-monitored off-shore wind turbine exceeds a predetermined threshold.

2. The on-line monitoring system for the inclination of an off-shore wind turbine tower according to claim 1, Characterized in that The tower inclination data feature extraction module includes: A vector construction unit for respectively arranging the foundation uneven settlement inclination angles and tower inclination angles of each off-shore wind turbine at multiple predetermined time points into a first input vector and a second input vector according to the time dimension; A time series coding unit for respectively passing the first input vector and the second input vector through a time series coding module including a one-dimensional convolutional layer of the deep neural network model to obtain a foundation uneven settlement inclination angle transformation feature vector and a tower inclination angle change feature vector; A feature vector correlation unit for calculating the product between the transposed vector of the foundation uneven settlement inclination angle transformation feature vector and the tower inclination angle change feature vector to obtain an inclination state correlation matrix; and A local feature extraction unit for passing the inclination state correlation matrix through the first convolutional neural network of the deep neural network model to obtain the inclination state correlation feature matrix.

3. The on-line monitoring system for the inclination of an off-shore wind turbine tower according to claim 2, Characterized in that The time series coding unit is further configured to: Use the fully connected layer of the time series encoding module including a one-dimensional convolutional layer of the deep neural network model to perform fully connected encoding on the first input vector and the second input vector respectively according to the following formula to extract the high-dimensional implicit features of the eigenvalue at each position in the first input vector and the second input vector respectively, where the formula is: where X is the input vector, Y is the output vector, W is the weight matrix, and B is the bias vector, represents matrix multiplication; Use the one-dimensional convolutional layer of the time series coding module including a one-dimensional convolutional layer of the deep neural network model to respectively perform one-dimensional convolutional coding on the first input vector and the second input vector according to the following formula to respectively extract the high-dimensional implicit correlation features between the eigenvalues at each position in the first input vector and the second input vector, where the formula is: Wherein, a is the width of the convolution kernel in the x direction, F is the convolution kernel parameter vector, G is the local vector matrix for the convolution kernel function operation, w is the size of the convolution kernel, and X represents the input vector.

4. The on-line monitoring system for the inclination of the tower barrel of an offshore wind turbine according to claim 3, characterized in that the local feature extraction unit is further configured to: in the forward propagation of each layer of the first convolutional neural network of the deep neural network model, respectively perform on the input data: perform convolution processing on the input data to obtain a convolution feature map; perform mean pooling based on the local channel dimension on the convolution feature map to obtain a pooled feature map; and perform non-linear activation on the pooled feature map to obtain an activation feature map; wherein, the output of the last layer of the first convolutional neural network of the deep neural network model is the inclination state correlation feature matrix, and the input of the first layer of the first convolutional neural network of the deep neural network model is the inclination state correlation matrix.

5. The on-line monitoring system for the inclination of the tower barrel of an offshore wind turbine according to claim 4, characterized in that the difference module is further configured to: calculate the difference between the inclination state correlation feature matrix of the offshore wind turbine to be monitored and the inclination state correlation feature matrices of other offshore wind turbines respectively according to the following formula to obtain the multiple difference matrices; wherein, the formula is: where M represents the tilt state correlation feature matrix of the offshore wind turbine to be monitored, and M i represents the tilt state correlation feature matrix of other offshore wind turbines except the tilt state correlation feature matrix of the offshore wind turbine to be monitored, represents differential by position.

6. The on-line monitoring system for the inclination of the tower barrel of an offshore wind turbine according to claim 5, characterized in that the eigenvalue correction module is further configured to: perform eigenvalue correction based on the maximum eigenvalue on each of the multiple difference matrices respectively according to the following formula to obtain the multiple corrected difference feature matrices; wherein, the formula is: where M represents each of the plurality of difference matrices, and m max represents the maximum eigenvalue of the eigenvalues at each position in the difference matrix, and ⊙ represents pointwise multiplication by position.

7. The on-line monitoring system for the inclination of the tower barrel of an offshore wind turbine according to claim 6, characterized in that the aggregation encoding module is further configured to: in the forward propagation of each layer of the second convolutional neural network model as a feature extractor, respectively perform on the input data: perform convolution processing on the input data to obtain a convolution feature map; perform mean pooling processing on the convolution feature map to obtain a pooled feature map; and perform non-linear activation on the pooled feature map to obtain an activation feature map; wherein, the output of the last layer of the second convolutional neural network model as a feature extractor is the classification feature map, and the input of the first layer of the second convolutional neural network model as a feature extractor is the three-dimensional input tensor.

8. The on-line monitoring system for the inclination of the tower barrel of an offshore wind turbine according to claim 7, characterized in that The monitoring result generation module is further configured to: the classifier processes the classification feature map with the following formula to generate a classification result, where the formula is: softmax{(W n ,B n ):…:(W 1 ,B 1 )|Project(F)}, where Project(F) represents projecting the classification feature map array into a vector, and W 1 to W n are the weight matrices of each fully connected layer, and B 1 to B n represent the bias matrices of each fully connected layer.

9. An on-line monitoring method for the inclination of the tower barrel of an offshore wind turbine, characterized in that comprises: obtaining the foundation uneven settlement inclination angles and tower barrel inclination angles of each offshore wind turbine in an offshore wind turbine array at a plurality of predetermined time points; passing the foundation uneven settlement inclination angles and tower barrel inclination angles of each offshore wind turbine at a plurality of predetermined time points through a deep neural network model as a feature extractor to obtain an inclination state correlation feature matrix corresponding to each offshore wind turbine; Extract the tilt state correlation feature matrix of the offshore wind turbine to be monitored from the corresponding tilt state correlation feature matrices of each offshore wind turbine; Calculate the differences between the tilt state correlation feature matrix of the offshore wind turbine to be monitored and the tilt state correlation feature matrices of other offshore wind turbines respectively to obtain a plurality of difference matrices; Perform eigenvalue correction based on the maximum eigenvalue on each of the plurality of difference matrices to obtain a plurality of corrected difference feature matrices; Aggregate the plurality of corrected difference feature matrices into a three-dimensional input tensor and then pass it through a second convolutional neural network model as a feature extractor to obtain a classification feature map; And Pass the classification feature map through a classifier to obtain a classification result, and the classification result is used to indicate whether the tower overturning risk of the offshore wind turbine to be monitored exceeds a predetermined threshold.

10. The online monitoring method for the tilt of the tower barrel of an offshore wind turbine according to claim 9, wherein, The step of obtaining the tilt state correlation feature matrix corresponding to each offshore wind turbine by passing the foundation uneven settlement tilt angle and the tower barrel tilt angle of each offshore wind turbine at a plurality of predetermined time points through a deep neural network model as a feature extractor includes: Arrange the foundation uneven settlement tilt angle and the tower barrel tilt angle of each offshore wind turbine at a plurality of predetermined time points into a first input vector and a second input vector respectively according to the time dimension; Pass the first input vector and the second input vector through the time series encoding module including a one-dimensional convolutional layer of the deep neural network model respectively to obtain a foundation uneven settlement tilt angle transformation feature vector and a tower barrel tilt angle change feature vector; Calculate the product between the transposed vector of the foundation uneven settlement tilt angle transformation feature vector and the tower barrel tilt angle change feature vector to obtain a tilt state correlation matrix; and Pass the tilt state correlation matrix through the first convolutional neural network of the deep neural network model to obtain the tilt state correlation feature matrix.

Images