Wind turbine generator blade edge trajectory abnormal fault diagnosis method and system

Abstract

The invention relates to a wind turbine generator blade edge trajectory abnormal fault diagnosis method. The method comprises the following steps: performing multi-scale analysis on a blade operation state image by adopting an improved wavelet packet to perform filtering and noise reduction; blade track edge features are obtained through invariant moment features, and translation, stretching and rotation invariance recognition is achieved; an image with large edge trajectory deviation is obtained by using a particle swarm optimization algorithm, and an abnormal state is calibrated for fault diagnosis; the method solves the problems that in the prior art, an edge detection operator threshold cannot be self-adaptive, noise resistance is poor, and faults cannot be effectively diagnosed due to the fact that blade edge track abnormal features are difficult to extract. And redundant data are screened out through an optimization algorithm, the accuracy and efficiency of intelligent identification are considered, and reliable technical support is provided for fault diagnosis of the wind turbine generator blades.

Description

Technical Field

[0001] The present application relates to the field of abnormal diagnosis of wind turbines, and in particular to a method, system, computer equipment and computer-readable storage medium for diagnosing abnormal faults of blade edge trajectories of wind turbines. Background Art

[0002] As core components of energy conversion, the operating status of wind turbine blades directly impacts the unit's safety, stability, and power generation efficiency. Diagnosing abnormal blade edge trajectory conditions is crucial for ensuring reliable operation. Currently, blade fault diagnosis relies heavily on image edge detection technology, which extracts blade edge trajectory features to identify operational anomalies.

[0003] In the existing technology, commonly used edge detection operators include Roberts operator, Sobel operator, Prewitt operator, Laplace operator and Log operator. Although they can realize image edge extraction, they have significant defects: First, the threshold cannot be adaptively adjusted according to the image, is sensitive to noise, and has poor noise resistance, which easily leads to edge detection results being distorted by noise interference; second, the extracted edges are not fine enough, and the boundaries often become widened or discontinuous, making it difficult to form a complete trajectory line, affecting the accurate identification of the blade operation status characteristics.

[0004] Although the Canny operator based on the optimization algorithm has certain advantages in signal-to-noise ratio and detection accuracy, it is still difficult to effectively solve the problem of incomplete edge feature extraction caused by image noise and blur in the complex operating environment of the blade. As a result, the abnormal state characteristics of the blade edge trajectory cannot be accurately captured, which in turn makes it difficult to effectively diagnose abnormal faults, restricting the reliability and efficiency of wind turbine blade fault diagnosis.

[0005] Therefore, there is an urgent need for a technical solution that can efficiently process noise and accurately extract edge trajectory features to improve the accuracy of blade fault diagnosis. Summary of the Invention

[0006] The embodiments of the present application provide a method, system, computer device and computer-readable storage medium for diagnosing abnormal blade edge trajectory faults of a wind turbine generator set, so as to at least solve the problem of insufficient accuracy in extracting edge trajectory features in related technologies.

[0007] In a first aspect, an embodiment of the present application provides a method for diagnosing abnormal blade edge trajectory faults of a wind turbine generator set, the method comprising: Acquire a blade image of a wind turbine, and perform multi-scale filtering on the blade image using an improved wavelet packet to obtain a filtered image; performing edge trajectory extraction on the filtered image to obtain a blade edge trajectory image, and performing edge invariant moment feature extraction on the blade edge trajectory image to output a plurality of invariant moment feature sequences; Initial abnormal samples are determined according to multiple invariant moment feature sequences, a particle swarm optimization model is constructed based on the invariant moment features of the initial abnormal samples, and an iterative optimization process is performed through the particle swarm optimization model to obtain key abnormal samples for blade fault diagnosis.

[0008] In some embodiments, performing multi-scale filtering on the leaf image using the improved wavelet packet includes: Collect images of the blade assembly from two orthogonal positions of the wind turbine to obtain blade sequence data at the two positions; After setting the two leaf sequence data to the same variable, the leaf sequence data is subjected to multi-layer wavelet packet decomposition under a three-layer wavelet basis to obtain a one-dimensional wavelet packet coefficient matrix at each scale. The function used in the wavelet packet decomposition satisfies a preset double-scale equation, the preset double-scale equation includes two sets of conjugate filter banks, and the coefficients of the conjugate filter banks have an orthogonal relationship; The one-dimensional wavelet packet coefficient matrix obtained after filtering and reconstruction is defined as two new sequence data as filtered images.

[0009] In some embodiments, performing edge trajectory extraction based on the filtered image to obtain a blade edge trajectory image includes: Converting pixel values ​​of the filtered image into grayscale values, and extracting edge points reflecting grayscale changes based on the grayscale values; Eliminating redundant boundary points from the edge points and filling boundary discontinuity points to obtain a first edge map sequence and a second edge map sequence in orthogonal directions in the same state of the blade unit; The first edge image sequence and the second edge image sequence are subtracted pixel by pixel, and the pixel points with non-zero values ​​in the subtraction result are set to 1. When the pixel points are set to 1, the position of the pixel points is defined as a trajectory point, and the trajectory point value = 1, to obtain the leaf edge trajectory image sequence.

[0010] In some embodiments, performing edge invariant moment feature extraction on the blade edge trajectory image includes: The blade edge track image is defined as a binary blade edge track, wherein the track point value = 1, and the non-track point value = 0; Based on the blade edge trajectory, a raw moment set is output by calculating the spatial distribution statistics of the blade edge point set, wherein the raw moment set includes: zero-order moment, first-order moment and high-order moment; Calculating the centroid coordinates of the blade edge trajectory image based on the original matrix set, and calculating the discretized central moment based on the centroid coordinates; The central moment is scale-normalized to obtain the normalized central moment, and based on the standardized invariant linear moment feature set studied by Hu, the normalized central distance is processed to generate rotational invariant moments of multiple dimensions, and the rotational invariant moments are defined as the invariant moment feature sequence.

[0011] In some embodiments, determining an initial abnormal sample based on a plurality of invariant moment feature sequences includes: Select the first invariant moment feature sequence and the second invariant matrix feature sequence that are continuous in time or space to construct abnormal comparison samples; Calculating the weight of each invariant moment feature based on the abnormal comparison sample; Calculating the contrast difference between the first invariant moment feature sequence and the second invariant matrix feature sequence, and the baseline difference of the first invariant moment feature sequence by combining the weighted ETD distance formula; Calculating the correlation coefficient of the abnormal comparison samples in the time series, and defining the sequence similarity of the invariant moment feature sequence based on the correlation coefficient; A difference parameter ratio is calculated according to the comparison difference and the benchmark difference, and based on the difference parameter ratio and the similarity, it is determined whether the abnormal comparison sample is an abnormal sample, and if so, it is defined as an initial abnormal sample.

[0012] In some embodiments, calculating the weight of each invariant moment feature based on the abnormal comparison sample includes: For the invariant moment features of each dimension, calculating the absolute difference between the first invariant moment feature sequence and the second invariant matrix feature sequence; According to the absolute difference, the invariant moment features of each dimension are arranged in ascending order from small to large to obtain a sorting result; Based on the ranking result, the invariant moment features of the first k dimensions are assigned successively decreasing weights, where k is a preset threshold.

[0013] In some embodiments, based on the initial abnormal samples, an iterative optimization process is performed using a particle swarm algorithm to obtain key abnormal samples including: Constructing a particle swarm optimization model based on the invariant moment characteristics of the initial abnormal samples, wherein the invariant moment characteristic parameters of the initial abnormal samples are used as particles, and the number of particles, the maximum number of iterations, the learning factor, and the inertia weight are set; Constructing a fitness function based on the difference parameter ratio and the correlation coefficient as key indicators, wherein the fitness function value is used to reflect the significance of the abnormal characteristics of the sample; Iterative optimization is performed in a particle swarm, including: updating the position and velocity of particles through individual extreme values ​​and global extreme values, adjusting the search direction according to the current fitness function value, and screening target optimization samples with significant abnormal characteristics and representativeness. The particle swarm algorithm adopts a dynamic inertia weight strategy, and the inertia weight decays exponentially with the increase of the number of iterations; When the iteration reaches the preset maximum number or the fitness function value tends to be stable, the abnormal sample corresponding to the global optimal particle is output as the key abnormal sample for constructing blade fault diagnosis.

[0014] In a second aspect, an embodiment of the present application provides a wind turbine blade edge trajectory abnormality fault diagnosis system, the system comprising: an acquisition module, a feature extraction module and a detection module, wherein: The acquisition module is used to acquire a blade image of a wind turbine and perform multi-scale filtering on the blade image using an improved wavelet packet to obtain a filtered image; The feature extraction module is used to perform edge trajectory extraction based on the filtered image to obtain a blade edge trajectory image, and perform edge invariant moment feature extraction on the blade edge trajectory image to output a plurality of invariant moment feature sequences; The detection module is used to determine initial abnormal samples based on multiple invariant moment feature sequences, construct a particle swarm optimization model based on the invariant moment features of the initial abnormal samples, and perform iterative optimization processing through the particle swarm optimization model to obtain key abnormal samples for blade fault diagnosis.

[0015] In a third aspect, an embodiment of the present application provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in the first aspect above when executing the computer program.

[0016] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first aspect above.

[0017] Compared with related technologies, the embodiment of the present application provides a method for diagnosing abnormal faults of blade edge trajectories of wind turbines. The method uses an improved wavelet packet multi-scale analysis to filter and reduce noise on blade operating status images; obtains blade trajectory edge features through moment invariant features to achieve translation, scaling, and rotation invariance recognition; and uses a particle swarm optimization algorithm to obtain images with large edge trajectory deviations, calibrating abnormal states for fault diagnosis. This method solves the problems in the prior art of the inability to adapt the threshold of the edge detection operator, poor noise resistance, and the difficulty in extracting abnormal blade edge trajectory features, resulting in the inability to effectively diagnose faults. It improves the accuracy of fault diagnosis, and uses an optimization algorithm to screen out redundant data, taking into account both the accuracy and efficiency of intelligent identification, providing reliable technical support for wind turbine blade fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 This is a flow chart of a method for diagnosing abnormal blade edge trajectory faults of a wind turbine according to an embodiment of the present application; Figure 2 is a flow chart of wavelet packet multi-scale filtering of leaf images according to an embodiment of the present application; Figure 3 is a schematic diagram of wavelet packet decomposition according to an embodiment of the present application; Figure 4 This is a structural block diagram of a wind turbine blade edge trajectory abnormality fault diagnosis system according to an embodiment of the present application; Figure 5 Schematic diagram of the internal structure of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solutions and advantages of this application more clearly understood, the present application is described and illustrated below in conjunction with the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely used to explain this application and are not intended to limit this application. Based on the embodiments provided in this application, all other embodiments obtained by those of ordinary skill in the art without making any creative efforts are within the scope of protection of this application.

[0020] Obviously, the drawings described below are merely examples or embodiments of the present application. Those skilled in the art can, without inventive effort, apply the present application to other similar scenarios based on these drawings. Furthermore, it is also understood that, although the effort involved in such a development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this application, changes in design, manufacturing, or production based on the technical content disclosed in this application are merely conventional technical means and should not be construed as an insufficiency of the content disclosed in this application.

[0021] References to "embodiments" in this application mean that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it refer to independent or alternative embodiments that are mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described in this application may be combined with other embodiments unless there is a conflict.

[0022] Unless otherwise defined, technical or scientific terms used herein shall have the ordinary meaning as understood by persons of ordinary skill in the art to which this application belongs. The terms "a," "an," "an," "the," and similar expressions used herein do not denote quantitative limitations and may refer to either the singular or the plural. The terms "comprise," "include," "have," and any variations thereof, used herein, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or modules (units) is not limited to the listed steps or units but may also include steps or units not listed, or may include other steps or units inherent to the process, method, product, or apparatus. The terms "connected," "connected," "coupled," and similar expressions used herein are not limited to physical or mechanical connections but may include electrical connections, whether direct or indirect. As used herein, "plurality" means two or more. "And / or" describes an association between associated objects, indicating that three possible relationships exist. For example, "A and / or B" may mean: A exists alone; A and B exist simultaneously; or B exists alone. The character " / " generally indicates that the objects before and after are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific order for the objects.

[0023] The present invention provides a method for diagnosing abnormal blade edge trajectory faults in a wind turbine. Figure 1 FIG. 1 is a flow chart of a method for diagnosing abnormal blade edge trajectory faults of a wind turbine according to an embodiment of the present application. Figure 1 As shown, the process includes the following steps: S101, acquiring a blade image of a wind turbine, and performing multi-scale filtering on the blade image using an improved wavelet packet to obtain a filtered image; The blade measured image signal is a linear combination of the real signal and noise, and the wavelet transform of the measured signal is also the sum of the wavelet transform of the signal and the wavelet transform of the noise. As the scale of the wavelet increases, the maximum value points of the noise decrease significantly, and the amplitude of its wavelet transform decreases with the increase of the scale. The wavelet at a large scale will mainly belong to the signal. Therefore, in this embodiment, by tracking the wavelet transform maximum at each scale using a coarse and fine strategy, the signal part is found and the noise part is removed, and the signal is reconstructed from these selected maximum points; the specific process of wavelet packet multi-scale filtering is as follows: Figure 2 shown; based on Figure 2 Based on the principle of

[15] , this embodiment constructs an improved wavelet model to purify the blade trajectory graphic features.

[0024] Figure 2 FIG. 1 is a flow chart of the leaf image wavelet packet multi-scale filtering according to an embodiment of the present application. Figure 2 As shown, the process consists of four steps: ① Select a wavelet basis function (such as db4, balancing smoothness and computational efficiency) and the number of decomposition levels N (trading off frequency resolution and computing power); ② Wavelet packet decomposition, which splits the image signal into high- and low-frequency subbands at multiple scales (more refined than traditional wavelets, preserving edge details); ③ Threshold quantization of each subband coefficient (using soft / hard thresholds to suppress small coefficients dominated by noise); and ④ Layer-by-layer reconstruction to restore the denoised image. Its core function is to filter out noise and enhance the true contour of the blade edge, resolving the "false edge / break" problem of traditional edge detection. This provides high-fidelity data for subsequent moment-invariant feature extraction, directly improving fault diagnosis accuracy.

[0025] Specifically, step S101 includes the following subdivision steps: S1, collects images of the blade unit from two orthogonal positions of the wind turbine to obtain blade sequence data at the two positions; In this embodiment, the collected blade unit images are specifically image sequences of the blade operating status collected from two different orthogonal positions of the wind turbine, and image sequence data x (n) and y (n) at time n are obtained respectively, where n is the time, and x and y are the image numbers; wherein the purpose of orthogonal position collection is to comprehensively capture the spatial characteristics of the blade from different angles, provide raw data for subsequent multi-dimensional analysis, ensure the integrity of the blade image information, and lay the foundation for subsequent filtering and edge extraction.

[0026] S2, after setting the two leaf sequence data to the same variable, performing multi-layer wavelet packet decomposition on the leaf sequence data under a three-layer wavelet basis to obtain a one-dimensional wavelet packet coefficient matrix at each scale, wherein the function used in the wavelet packet decomposition satisfies a preset two-scale equation, and the preset two-scale equation includes two sets of conjugate filter banks, and the coefficients of the conjugate filter banks have an orthogonal relationship; Specifically, the two image sequence data x(n) and y(n) obtained in step S1 are first uniformly set as the variable U(n) for unified processing; then, they are processed based on the wavelet packet function that satisfies the preset dual-scale equation, which is:

[0027] Where, is a conjugate filter bank, where , that is, the two coefficients also have an orthogonal relationship, which can ensure the stability and orthogonality of the decomposition.

[0028] U(n) is then subjected to a three-layer wavelet packet multiscale decomposition, decomposing the signal into subspaces of varying scales. Ultimately, the one-dimensional wavelet packet coefficient matrix at each scale is extracted. The purpose of the three-layer decomposition is to distinguish signal from noise (noise amplitude decreases at larger scales) through multiscale analysis, providing a coefficient basis for subsequent noise filtering.

[0029] It should also be noted that Figure 3 is a schematic diagram of wavelet packet decomposition according to an embodiment of the present application, such as Figure 3 As shown, the layer 3 of U(n) wavelet packet decomposition is marked as the signal after image filtering , then the signal Perform 3-layer wavelet packet decomposition. Wavelet space That is

[0030] Where m=0, 1, 2, ..., ; By subspace ; Subspaces are divided into Synthetic Hilbert space That is ; It shows that the entire square integrable function space can be completely covered by these wavelet subspaces. The reverse process (i.e. wavelet packet decomposition) is to transform the original signal from Decomposing into each Wj subspace provides a mathematical basis for spatial decomposition for subsequent noise removal and retention of effective signals through threshold quantization, ensuring that the three-layer decomposition can effectively separate signals and noise, laying the foundation for filtering reconstruction.

[0031] S3, defining the one-dimensional wavelet packet coefficient matrix obtained after filtering and reconstruction as two new sequence data as filtered images.

[0032] Among them, x(n) and y(n) are decomposed into wavelet packets at multiple scales under three-layer wavelet basis, and the next-dimensional wavelet packet coefficient matrix at each scale is extracted. After that, it is further "filtered": based on the difference in coefficient amplitude between noise and signal (the noise coefficient amplitude is smaller), the redundant coefficients corresponding to the noise are eliminated through threshold quantization (such as setting a reasonable threshold and setting the noise coefficient below the threshold to zero), and only the effective coefficients reflecting the blade trajectory characteristics are retained.

[0033] Then, through the “reconstruction” operation (i.e., the inverse wavelet packet decomposition process, based on subspace synthesis theory), the filtered The coefficient matrix is ​​recombined to finally obtain the denoised leaf image signal, that is, the filtered image x(n) and y(n).

[0034] Step S3 purifies the original image signal and removes noise interference, so that x(n) and y(n) more clearly retain the true characteristics of the leaf trajectory, providing high-quality preprocessing data for subsequent "leaf image edge trajectory extraction" (such as grayscale conversion and edge point extraction) to ensure the accuracy of edge detection.

[0035] In the above step S101, by analyzing the wavelet transform characteristics of the real signal and noise in the measured image signal of the blade, a method of coarse and fine tracking of the wavelet transform maximum values ​​of each scale to separate the signal and noise and reconstruct the signal is adopted to construct a wavelet model to purify the blade trajectory graphic features, thereby providing a denoised high-quality signal foundation for subsequent wavelet packet multi-scale filtering and edge trajectory extraction.

[0036] S102, performing edge trajectory extraction based on the filtered image to obtain a blade edge trajectory image, and performing edge invariant moment feature extraction on the blade edge trajectory image to output multiple invariant moment feature sequences; The edge trajectory extraction is performed based on the filtered image to obtain the leaf edge trajectory image, which specifically includes the following subdivision steps: S1, converting the pixel values ​​of the filtered image into grayscale values, and extracting edge points reflecting grayscale changes based on the grayscale values; It can be understood that in step S1, based on the leaf image signals x(n) and y(n) obtained after filtering, the RGB values ​​of the image pixels are converted into grayscale values ​​using the color image RGB to grayscale formula (gray = R * 0.299 + G * 0.587 + B * 0.114), thereby simplifying the image from color to grayscale and reducing the data dimension to improve the efficiency of subsequent processing.

[0037] On this basis, edge points are extracted according to the changing characteristics of grayscale values. The location of grayscale value mutation is the potential location of the leaf edge. This step locates the edge by capturing the grayscale change, providing an initial edge point set for subsequent edge refinement processing.

[0038] S2, removing redundant edge points from the edge points and filling the boundary discontinuity points, to obtain a first edge map sequence and a second edge map sequence in orthogonal directions in the same state of the blade unit; Considering that the initial edge points extracted in step S1 may contain redundant boundary points (non-real edges) and boundary discontinuities (edge ​​discontinuities) due to residual noise or image blur, this step optimizes edge accuracy and continuity by removing redundant points (e.g., filtering out pseudo-edges with insignificant grayscale changes) and filling discontinuities (e.g., connecting adjacent edge segments with consistent grayscale change trends).

[0039] At the same time, for the orthogonal filtered images x(n) and y(n) of the leaf in the same state, the corresponding first edge map sequence E1 (x, y) and second edge map sequence E2 (z, y) are generated respectively, and the edge features are doubly captured from the orthogonal directions to ensure the integrity of the edge information.

[0040] S3, performing pixel-by-pixel subtraction on the first edge image sequence and the second edge image sequence, setting the pixel points with non-zero values ​​in the subtraction result to 1, and while setting the pixel points to 1, defining the positions of the pixel points as trajectory points, and the trajectory point values ​​= 1, to obtain the leaf edge trajectory image sequence.

[0041] Specifically, the first edge map sequence E1 and the second edge map sequence E2 obtained in step S2 are subtracted pixel by pixel. The redundant edges (such as the edges of the background or non-critical structures) overlapping in the two edge maps can be filtered out through the difference operation. The numerical part of the subtraction result is set to 1, which is the trajectory point. The core of this step is to suppress non-valid edges with slight differences, which can further highlight the valid edges with significant grayscale changes, and finally obtain the refined leaf edge trajectory image sequence E3 (x, y), that is, the leaf edge trajectory image f (x, y) output by the edge trajectory extraction, which provides a clear and accurate edge trajectory basis for the subsequent invariant moment feature extraction.

[0042] Through the above steps S1 to S3, the filtered image is converted to grayscale, edge points are extracted and optimized, and edge map subtraction and refinement processing are performed to extract a clear and accurate blade edge trajectory image, which provides a reliable edge trajectory basis for subsequent blade trajectory invariant moment feature extraction and abnormal fault diagnosis.

[0043] Furthermore, edge invariant moment feature extraction is performed on the blade edge trajectory image, and multiple invariant moment feature sequences are output, including the following subdivision steps: S1, the leaf edge trajectory image is defined as a binary leaf edge trajectory, where the trajectory point value = 1 and the non-track point value = 0; Specifically, the leaf edge trajectory image f(x,y) is defined as a binary trajectory: the grayscale value of each coordinate point passed by the leaf edge trajectory is set to 1, and the grayscale value of each coordinate point not passed by the trajectory is set to 0. Binarization significantly simplifies the image data dimension, retaining only the spatial position information of the edge trajectory and filtering out irrelevant grayscale fluctuations. This provides a clear and unified input basis for subsequent moment feature calculations, ensuring that feature extraction focuses on the geometric distribution of the trajectory itself.

[0044] S2, based on the blade edge trajectory, calculates the spatial distribution statistics of the blade edge point set and outputs the original moment set, where the original moment set includes zero-order moment, first-order moment and high-order moment; Specifically, based on the binary leaf edge trajectory, the (p+q) order raw moment is calculated, which is expressed as , which includes the zero-order moment (p=0,q=0, reflecting the overall existence of the trajectory), the first-order moment (p+q=1, such as , reflecting the basic distribution related to the trajectory's center of gravity) and higher-order moments (p+q≥2, reflecting the detailed characteristics of the trajectory's shape). These raw moments, as statistics of the spatial distribution of edge point sets, fully characterize the position, range, and morphological characteristics of the trajectory and serve as the basic data for the subsequent calculation of invariant moments.

[0045] S3, based on the original matrix set, calculate the centroid coordinates of the blade edge trajectory image, and calculate the discretized invariant line moment based on the centroid coordinates; First, the centroid coordinates of the blade edge trajectory image are calculated based on the original moment: ,in is the zero-order moment, is the first-order moment), and the barycentric coordinates reflect the geometric center position of the trajectory.

[0046] Furthermore, the discretized central moment is calculated with the center of gravity as the origin: It can be understood that the central moment eliminates the influence of trajectory translation on the features (no matter where the trajectory translates, the moment with the center of gravity as the origin remains unchanged), laying the foundation for subsequent scale and rotation invariance processing.

[0047] S3, scale normalization is performed on the central moment to obtain the normalized invariant linear moment, and based on the standardized invariant linear moment feature set studied by Hu, the normalized central distance is processed to generate rotational invariant moments in multiple dimensions, and multiple rotational invariant moments are defined as an invariant moment feature sequence.

[0048] In this step, the invariant moments are scaled normalized by Eliminate the impact of trajectory scale changes (regardless of whether the trajectory is enlarged or reduced, the normalized moment value remains stable).

[0049] Then, based on the normalized invariant moment theory studied by Hu, 7 rotational invariant moments were derived using the second-order and third-order normalized central moments ( μ 1 to μ 7), whose expression ensures that the moment value remains unchanged when the trajectory rotates. The seven rotational invariant moments generated ultimately constitute an invariant moment feature sequence, which achieves invariant recognition of the translation, extension, and rotation changes of the blade edge trajectory. Each trajectory can be marked by a unique feature sequence, providing a stable feature benchmark for subsequent anomaly determination. The seven Hu invariant moments are defined as follows: (4.8)

[0050] Through the above steps S1 to S3, seven invariant line moment features of the blade image edge are obtained, so that each blade trajectory figure is marked by a unique edge invariant moment. Therefore, the recognition of the edge invariant moment completes the recognition of the blade trajectory figure.

[0051] S103, determining initial abnormal samples according to multiple invariant moment feature sequences, constructing a particle swarm optimization model based on the invariant moment features of the initial abnormal samples, and performing iterative optimization processing through the particle swarm optimization model to obtain key abnormal samples for blade fault diagnosis.

[0052] Determining the initial abnormal samples based on multiple invariant moment feature sequences includes the following subdivision steps: S1, selecting a first invariant moment feature sequence and a second invariant matrix feature sequence that are continuous in time or space, constructing an abnormal comparison sample, and calculating the weight of each invariant moment feature based on the abnormal comparison sample.

[0053] Among them, from the invariant moment feature sequences output in step S102, firstly select two sets of invariant moment feature sequences that are continuous in time or space, namely A i (i=1,…,7) and B i+1 (i=1,…,7) as abnormal comparison samples, where A i and B i+1 Both contain 7-dimensional invariant moment features (μ1 to μ7).

[0054] Furthermore, calculating the weight of each invariant moment feature includes: S1.1, for each dimension of the invariant moment feature, calculate the absolute difference between the first invariant moment feature sequence and the second invariant matrix feature sequence; S1.2, based on absolute interpolation, sort the invariant moment features of each dimension in ascending order from small to large to obtain the sorting result; S1.3, based on the sorting results, assign descending weights to the invariant moment features of the first k dimensions, where k is a preset threshold.

[0055] In this embodiment, when calculating the weights of the invariant moment features of each dimension, the absolute difference between Ai and Bi+1 is first calculated for each dimension (such as μ1 to μ7) to reflect the difference between the two groups of sequences in that dimension. The absolute differences of each dimension are then sorted from small to large. According to the principle of "near large and far small", the first k dimensions with smaller differences after sorting are assigned successively decreasing weights. The smaller the difference, the more representative the dimension of the trajectory feature and the higher the weight. Finally, the weight set W={wi}(i=1,…,7) is obtained, which provides the weight basis for subsequent difference calculations.

[0056] S2, calculates the contrast difference between the first invariant moment feature sequence and the second invariant matrix feature sequence, and the baseline difference between the first invariant moment feature sequence by combining the weighted ETD distance formula; Among them, the weight W={wi} obtained by combining S1 is used to calculate the difference between the two groups of sequences using the defined ETD distance formula: Here, "contrast difference" refers to the ETD distance between the first invariant moment feature sequence (Ai) and the second invariant moment feature sequence (Bi+1), that is, D(A,B). The formula uses the weight wi to weight the differences in each dimension, highlighting the impact of important dimensions (dimensions with high weights) on the overall difference; Specifically, the ETD distance from A to B is defined as follows: :

[0057] Among them, the weight function is selected according to the principle of large near and small far: ,in, The predicted value of the N+1th point of the S sequence is: ; Additionally, "baseline difference" refers to the ETD distance between the first invariant moment feature sequence (Ai) and itself, namely D(A,A), which serves as a baseline value for measuring differences (theoretically, the self-comparison difference is 0, but in practice it is used to normalize the ratio). In this embodiment, the ETD distance calculation converts high-dimensional invariant moment feature differences into quantifiable values, providing data support for subsequent judgments.

[0058] S3, calculate the correlation coefficient of the abnormal comparison samples in the time series, and define the sequence similarity of the invariant moment feature sequence based on the correlation coefficient; Based on the differences in each dimension obtained by S2, the following correlation coefficient formula is used:

[0059] Calculate the correlation coefficient of the two time series A and B. The correlation coefficient reflects the overall similarity of the two series: the closer the value is to 1, the more consistent the trend of the series changes (high similarity); the lower the value, the greater the trend difference (low similarity). This correlation coefficient provides a basis for sequence correlation for anomaly determination.

[0060] S4, calculating a difference parameter ratio according to the comparison difference and the benchmark difference, and judging whether the abnormal comparison sample is an abnormal sample based on the difference parameter ratio and similarity, and if so, defining it as an initial abnormal sample.

[0061] It can be understood that this step first calculates the difference parameter ratio, that is, the ratio of the comparison difference to the benchmark difference D (A, B) / D (A, A), and then combines it with the comprehensive judgment of sequence similarity obtained in S3: If the ratio is ≥80% and the correlation coefficient shows a high similarity, it means that the two groups of sequence characteristics are similar and are judged to be normal; If the ratio is < 80%, or the correlation coefficient shows low similarity, it means that the sequence characteristics of the two groups are significantly different and are judged to be abnormal. The comparison sample is defined as the initial abnormal sample.

[0062] Steps S1 to S3 achieve preliminary screening of abnormal states of blade edge trajectories through quantified difference ratios and sequence similarities, laying the foundation for the subsequent optimization algorithm to further accurately identify abnormal samples.

[0063] In an exemplary embodiment, the following Table 1 is a table of edge trajectory values ​​of abnormal blade samples:

[0064] Table 2 below is a table of edge trajectory values ​​for unknown leaf samples:

[0065] Table 1 lists the values ​​of seven invariant moments (μ1-μ7) for four known normal blades. Their core function is to serve as a "benchmark feature library." The μ values ​​of normal samples generally have a small range (e.g., μ1 between 0.0108 and 0.0145, and μ2 between 0.0597 and 0.1026). These values ​​are stable, with no significant positive or negative fluctuations. These values ​​reflect the typical invariant moment characteristics of normal blade edge trajectories and provide a "normal state reference" for comparison with unknown samples. Table 2 Edge trajectory values ​​of unknown blade samples The table contains 7 invariant moment values ​​of 16 samples to be diagnosed, which are the objects of abnormal screening. The μ values ​​of these samples are significantly different from those of normal samples.

[0066] For example, the μ1 of unknown sample 1 is 0.7099 (approximately 49 times the maximum μ1 of a normal sample), and μ3 is -1.8390 (μ3 of a normal sample is always positive and ≤0.1184). The overall numerical range is larger, and the positive and negative fluctuations are obvious. By calculating the D(A,B) / D(A,A) ratio compared with normal samples, we can preliminarily determine which samples deviate from normal characteristics.

[0067] It can be understood that the values ​​in Tables 1 and 2 are the input data for the subsequent particle swarm optimization of marginal invariant moments. By comparing the invariant moment values ​​of 16 unknown samples with those of 4 normal samples, samples with significant differences from normal samples (i.e., ratios < 80%) are screened out. This provides a basis for the subsequent particle swarm algorithm to further purify key abnormal samples (such as the 8 samples retained in the final optimization of the document), ultimately achieving the elimination of redundant data and improving the accuracy and efficiency of blade fault diagnosis.

[0068] Furthermore, based on the initial abnormal samples, the particle swarm algorithm is used to perform iterative optimization processing to obtain key abnormal samples, which includes the following subdivision steps: S1, constructing a particle swarm optimization model based on the invariant moment characteristics of the initial abnormal samples, wherein the invariant moment characteristic parameters of the initial abnormal samples are used as particles, and the number of particles, maximum number of iterations, learning factor and inertia weight are set; Specifically, the seven invariant moment feature parameters (μ1 to μ7) of the initial abnormal sample are used as "particles", and each particle is represented by a triplet (xi, vi, pbest_i), where xi is the current position of the particle (corresponding to the subscript i=1,2,…,7 of the invariant moment feature), vi is the current velocity of the particle (corresponding to the numerical value of the invariant moment feature), and pbest_i is the optimal position searched by the particle itself (that is, the historically optimal combination of invariant moment features of the particle).

[0069] Furthermore, key parameters are set: the number of particles (determined by the number of initial outlier samples), the maximum number of iterations, the learning factors C1 and C2 (positive constants used to adjust the weights of particles learning towards individual and global extremes), and the initial and final values ​​of the inertia weight Wmax and Wmin (used to balance the particles' global exploration and local exploitation capabilities). This step provides a basic model framework for particle swarm optimization, ensuring that the optimization process revolves around the invariant moment characteristics of outlier samples.

[0070] S2, constructing a fitness function based on the difference parameter ratio and correlation coefficient as key indicators, where the fitness function value is used to reflect the significance of the sample's abnormal characteristics; Specifically, the fitness function is constructed with the "difference parameter ratio (D (A,B) / D (A,A))" and "correlation coefficient (N (A,B))" as core indicators; the fitness function value directly reflects the significance of the sample's abnormal characteristics: if the difference parameter ratio of a sample is smaller (the further it deviates from the normal sample) and the correlation coefficient is lower (the greater the difference from the normal sample trend), the fitness function value is smaller (or larger according to the setting), indicating that the abnormal characteristics of the sample are more significant.

[0071] The optimization goal of this embodiment is to "obtain a sample set Xi with the smallest difference from the data to be identified" (here "minimum difference" means the highest match with the abnormal characteristics). Therefore, the design of the fitness function needs to prioritize screening out samples with a difference parameter ratio < 80% and a low correlation coefficient, providing a quantitative evaluation standard for subsequent iterative optimization.

[0072] S3, iterative optimization in the particle swarm, including: updating the position and velocity of the particles through individual extreme values ​​and global extreme values, adjusting the search direction according to the current fitness function value, screening out target optimization samples with significant abnormal characteristics and representativeness. Among them, the particle swarm algorithm adopts a dynamic inertia weight strategy, and the inertia weight decays exponentially with the increase of the number of iterations; The core of this step is to filter representative abnormal samples by iteratively updating the particle state: Specifically, during the iteration process, each particle tracks the "individual extreme value pbest_i" (its own historical optimal position) and the "global extreme value gbest" (the historical optimal position of the entire particle swarm), and updates its speed and position according to the formula: the speed update combines the learning factor, random number and extreme value deviation, and the position update is the current position plus the new speed, ensuring that the particle searches for a better abnormal feature area.

[0073] A dynamic inertia weight strategy is adopted: the inertia weight w decays as the number of iterations increases according to the formula w=Wmax - (Wmax - Wmin)*t / tmax (t is the current number of iterations, tmax is the maximum number of iterations), and the search direction is adjusted according to the fitness function value at each iteration: if the fitness function value of the current particle is better than the historical extreme value, pbest_i or gbest is updated to gradually screen out the target samples with the most significant and representative abnormal characteristics.

[0074] S4, when the iteration reaches the preset maximum number or the fitness function value tends to be stable, the abnormal sample corresponding to the global optimal particle is output as the key abnormal sample for constructing blade fault diagnosis.

[0075] As you can understand, when the maximum number of iterations is reached, or when the fitness function value changes less than a threshold (approaches stability) over multiple consecutive iterations, the optimization process stops. The outlier samples represented by the particles corresponding to the global extreme value gbest are output as "key outlier samples." These samples are removed after redundant data is removed (for example, only 8 of the 16 unknown samples in the document are ultimately retained during optimization), and they fully encompass the characteristics of all outlier states ("The optimization sample library fully encompasses all four leaf edge trajectory samples of all outlier states").

[0076] Those skilled in the art will know that the output of key abnormal samples provides high-quality training data for subsequent blade fault diagnosis models (such as BP neural networks), taking into account the accuracy and efficiency of intelligent identification, and ultimately improving the reliability of wind turbine blade fault diagnosis.

[0077] In an exemplary embodiment, a specific optimization process includes the following steps: Step 1: Each research entity in the search space is called a "particle". Each particle has its own position and velocity, which can be expressed by a triplet Indicates that Indicates the current position of the particle, Indicates the current velocity of the particle. represents the best position that the particle itself has searched; Step 2: Repeatedly iterate the particle's current position and velocity, using D(A,B) / D(A,A) as the final abnormality determination condition. If the ratio is greater than or equal to 80%, the results are considered similar. If the ratio is less than 80%, the results are considered similar.

[0078] If a better solution is found, it will be used as a basis to find the next solution until the optimal solution that meets the conditions is found. In each iteration, the particle updates itself by tracking two "extreme values": the first is the best solution found by the particle itself, which is called the individual extreme point (denoted by indicates its location).

[0079] Step 3: The other extreme point is the best solution found by the entire population so far, called the global extreme point (using Indicates its position); Assume that the search space of the problem is a square in two-dimensional space, represented by S, and S=a1, b1×...×aD, bD. After finding the two best solutions in each iteration, the particle Update your speed and position according to the following formula:

[0080] Where: is the optimal position that particle i passes through, is the best position that all particles in the group have passed; A positive constant is the learning factor, is a random number in the interval [0,1]; thus the optimal particle swarm that meets the conditions is obtained.

[0081] Step 4: By comparing the edge moment data of the 16 acquired patterns with the four measured patterns, we found that the values ​​of the corresponding positions of the invariant moments of similar blade edge trajectory images are extremely similar. Therefore, when the sample invariant moments are considered as a set Ai, where i = 1, 2, …, 7, and the invariant moments of the patterns to be identified are considered as a set Bi, where i = 1, 2, …, 7, we studied the data set and found that similar blade edge trajectories exhibit approximately equal maxima and minima at the same invariant moment position i, Ai and Bi. The greater the difference in the blade edge trajectory images, the greater the difference in the corresponding positions of the maximum and minimum values ​​of the invariant moments.

[0082] Based on the above principles, this paper proposes an improved particle swarm global optimization algorithm; first, the invariant moment is regarded as a particle in a particle swarm, the position of the particle is the subscript corresponding to the invariant moment value in the seven invariant moments (i.e., i=1,2,…,7), and the particle velocity is the invariant moment value (i.e., Ai or Bi); secondly, the velocities of the corresponding positions (i.e., i=1,2,…,7) in the two groups of particle swarms (i.e., Ai and Bi) are compared, and particles with seven positions and velocities close to each other are found as the optimization particles, and the particles obtained by the optimization replace the original particles Ai, and are recorded as Xi; then, the objective function value of the new particles is calculated. If f(Xi)≤f(Ai), Xi is set to Ai, and the inertia weight is updated by formula (4.10); finally, it is determined whether the termination condition is met. If so, the calculation is stopped and the optimal solution is output. Otherwise, the number of iterations is updated and the calculation is continued, and it is repeated until the global optimum is reached; thus, the algorithm obtains a sample set Xi with the smallest difference from the data to be identified. Based on this, the accuracy and efficiency of the following BP neural network recognition algorithm are greatly improved.

[0083] The weight factor for global optimization is: in: are the initial and final values ​​of the inertia weight, is the maximum number of iterations of the algorithm, and t is the number of iterations.

[0084] The data after optimization are as follows Table 3: The numerical table of leaf sample edge trajectory after particle swarm optimization:

[0085] The above optimization results show that 8 samples were obtained in the actual measurement, and the excessive redundant data was screened out so that the intelligent recognition algorithm could achieve a balance between accuracy and efficiency. The key samples obtained after optimization are sample 1, sample 2, sample 6, sample 8, sample 10, sample 11, sample 15, and sample 16. By comparing the optimized graphic samples with the graphic samples in the known sample library, it is found that the optimized sample library completely contains all the four leaf graphic edge trajectory samples in all abnormal states.

[0086] Through the above steps S101 to S103, blade fault diagnosis is achieved by processing the blade image through wavelet packet multi-scale filtering, extracting invariant moment features (including Hu invariant moments) to capture trajectory characteristics, and combining particle swarm optimization to screen key abnormal samples. This effectively solves the problem of traditional edge detection operator thresholds being unable to adapt and having poor noise resistance. Specifically, wavelet packet multi-scale analysis accurately separates signal and noise, making blade image features clearer and improving edge extraction accuracy, laying a high-quality data foundation for subsequent diagnosis. Furthermore, the translation, scaling, and rotational invariance of invariant moment features ensures stable recognition of blade trajectory features under different states, avoiding feature distortion caused by changes in blade motion posture. In addition, the particle swarm optimization algorithm eliminates redundant data, and the key abnormal samples screened out are highly representative, greatly improving the efficiency of fault diagnosis while ensuring the integrity of abnormal features.

[0087] This application takes into account both diagnostic accuracy and intelligent identification efficiency, providing reliable technical support for the timely discovery and accurate determination of wind turbine blade failures, helping to reduce operation and maintenance costs and improve the safety and stability of unit operation.

[0088] In a second aspect, the present application also provides a wind turbine blade edge trajectory abnormality fault diagnosis system. Figure 4 This is a structural block diagram of a wind turbine blade edge trajectory abnormality fault diagnosis system according to an embodiment of the present application. Figure 4 As shown, the system includes: an acquisition module 40, a feature extraction module 41 and a detection module 42, wherein: The acquisition module 40 is used to acquire the blade image of the wind turbine and perform multi-scale filtering on the blade image using an improved wavelet packet to obtain a filtered image; The feature extraction module 41 is used to extract edge tracks based on the filtered image to obtain a blade edge track image, and to extract edge invariant moment features from the blade edge track image to output multiple invariant moment feature sequences; The detection module 42 is used to determine the initial abnormal samples according to multiple invariant moment feature sequences, construct a particle swarm optimization model based on the invariant moment features of the initial abnormal samples, and perform iterative optimization processing through the particle swarm optimization model to obtain key abnormal samples for blade fault diagnosis.

[0089] This system uses wavelet packet multi-scale analysis to accurately separate signals and noise, making blade image features clearer and improving edge extraction accuracy, laying a high-quality data foundation for subsequent diagnosis; further, relying on the translation, scaling, and rotation invariance of the invariant moment feature, it ensures stable recognition of blade trajectory features under different states and avoids feature distortion caused by changes in blade motion posture; in addition, the particle swarm optimization algorithm eliminates redundant data, and the key abnormal samples screened out are highly representative, which greatly improves the efficiency of fault diagnosis while ensuring the integrity of abnormal features; this application scheme takes into account both diagnostic accuracy and intelligent recognition efficiency, and provides reliable technical support for the timely discovery and accurate determination of wind turbine blade faults, which helps to reduce operation and maintenance costs and improve the safety and stability of unit operation.

[0090] In one embodiment, Figure 5 is a schematic diagram of the internal structure of an electronic device according to an embodiment of the present application, such as Figure 5 As shown, an electronic device is provided, which may be a server, and its internal structure diagram may be as shown in FIG. Figure 5 As shown. This electronic device includes a processor, a network interface, an internal memory, and a non-volatile memory connected via an internal bus, wherein the non-volatile memory stores an operating system, a computer program, and a database. The processor is used to provide computing and control capabilities, the network interface is used to communicate with external terminals via a network connection, the internal memory is used to provide an environment for the operation of the operating system, the computer program, when executed by the processor, implements a method for diagnosing abnormal blade edge trajectory faults of a wind turbine, and the database is used to store data.

[0091] Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the electronic device to which the solution of the present application is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0092] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0093] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0094] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A method for diagnosing abnormal blade edge trajectory faults of a wind turbine generator set, characterized in that: The method comprises: Acquire a blade image of a wind turbine, and perform multi-scale filtering on the blade image using an improved wavelet packet to obtain a filtered image; performing edge trajectory extraction on the filtered image to obtain a blade edge trajectory image, and performing edge invariant moment feature extraction on the blade edge trajectory image to output a plurality of invariant moment feature sequences; Initial abnormal samples are determined according to multiple invariant moment feature sequences, a particle swarm optimization model is constructed based on the invariant moment features of the initial abnormal samples, and an iterative optimization process is performed through the particle swarm optimization model to obtain key abnormal samples for blade fault diagnosis.

2. The method according to claim 1, characterized in that Performing multi-scale filtering on the leaf image by using an improved wavelet packet includes: Collect images of the blade assembly from two orthogonal positions of the wind turbine to obtain blade sequence data at the two positions; After setting the two leaf sequence data to the same variable, the leaf sequence data is subjected to multi-layer wavelet packet decomposition under a three-layer wavelet basis to obtain a one-dimensional wavelet packet coefficient matrix at each scale. The function used in the wavelet packet decomposition satisfies a preset double-scale equation, the preset double-scale equation includes two sets of conjugate filter banks, and the coefficients of the conjugate filter banks have an orthogonal relationship; The one-dimensional wavelet packet coefficient matrix obtained after filtering and reconstruction is defined as two new sequence data as filtered images.

3. The method according to claim 1, characterized in that Extracting edge tracks according to the filtered image to obtain a blade edge track image includes: Converting pixel values ​​of the filtered image into grayscale values, and extracting edge points reflecting grayscale changes based on the grayscale values; Eliminating redundant boundary points from the edge points and filling boundary discontinuity points to obtain a first edge map sequence and a second edge map sequence in orthogonal directions in the same state of the blade unit; The first edge image sequence and the second edge image sequence are subtracted pixel by pixel, and the pixel points with non-zero values ​​in the subtraction result are set to 1. When the pixel points are set to 1, the position of the pixel points is defined as a trajectory point, and the trajectory point value = 1, to obtain the leaf edge trajectory image sequence.

4. The method according to claim 1, wherein Extracting edge invariant moment features from the blade edge trajectory image includes: The blade edge track image is defined as a binary blade edge track, wherein the track point value = 1, and the non-track point value = 0; Based on the blade edge trajectory, a raw moment set is output by calculating the spatial distribution statistics of the blade edge point set, wherein the raw moment set includes: zero-order moment, first-order moment and high-order moment; Calculating the centroid coordinates of the blade edge trajectory image based on the original matrix set, and calculating the discretized central moment based on the centroid coordinates; The central moment is scale-normalized to obtain the normalized central moment, and based on the standardized invariant linear moment feature set studied by Hu, the normalized central distance is processed to generate rotational invariant moments of multiple dimensions, and the rotational invariant moments are defined as the invariant moment feature sequence.

5. The method according to claim 4, characterized in that Determining the initial abnormal samples based on multiple invariant moment feature sequences includes: Select the first invariant moment feature sequence and the second invariant matrix feature sequence that are continuous in time or space to construct abnormal comparison samples; Calculating the weight of each invariant moment feature based on the abnormal comparison sample; Calculating the contrast difference between the first invariant moment feature sequence and the second invariant matrix feature sequence, and the baseline difference of the first invariant moment feature sequence by combining the weighted ETD distance formula; Calculating the correlation coefficient of the abnormal comparison samples in the time series, and defining the sequence similarity of the invariant moment feature sequence based on the correlation coefficient; A difference parameter ratio is calculated according to the comparison difference and the benchmark difference, and based on the difference parameter ratio and the similarity, it is determined whether the abnormal comparison sample is an abnormal sample, and if so, it is defined as an initial abnormal sample.

6. The method according to claim 5, characterized in that Based on the abnormal comparison sample, calculating the weight of each invariant moment feature includes: For the invariant moment features of each dimension, calculating the absolute difference between the first invariant moment feature sequence and the second invariant matrix feature sequence; According to the absolute difference, the invariant moment features of each dimension are arranged in ascending order from small to large to obtain a sorting result; Based on the ranking result, the invariant moment features of the first k dimensions are assigned successively decreasing weights, where k is a preset threshold.

7. The method according to claim 5, characterized in that Based on the initial abnormal samples, the particle swarm algorithm is used to perform iterative optimization processing, and the key abnormal samples obtained include: Constructing a particle swarm optimization model based on the invariant moment characteristics of the initial abnormal samples, wherein the invariant moment characteristic parameters of the initial abnormal samples are used as particles, and the number of particles, the maximum number of iterations, the learning factor, and the inertia weight are set; Constructing a fitness function based on the difference parameter ratio and the correlation coefficient as key indicators, wherein the fitness function value is used to reflect the significance of the abnormal characteristics of the sample; Iterative optimization is performed in a particle swarm, including: updating the position and velocity of particles through individual extreme values ​​and global extreme values, adjusting the search direction according to the current fitness function value, and screening target optimization samples with significant abnormal characteristics and representativeness. The particle swarm algorithm adopts a dynamic inertia weight strategy, and the inertia weight decays exponentially with the increase of the number of iterations; When the iteration reaches the preset maximum number or the fitness function value tends to be stable, the abnormal sample corresponding to the global optimal particle is output as the key abnormal sample for constructing blade fault diagnosis.

8. A wind turbine blade edge trajectory abnormality fault diagnosis system, characterized in that: The system includes: an acquisition module, a feature extraction module and a detection module, wherein: The acquisition module is used to acquire a blade image of a wind turbine and perform multi-scale filtering on the blade image using an improved wavelet packet to obtain a filtered image; The feature extraction module is used to perform edge trajectory extraction based on the filtered image to obtain a blade edge trajectory image, and perform edge invariant moment feature extraction on the blade edge trajectory image to output a plurality of invariant moment feature sequences; The detection module is used to determine initial abnormal samples based on multiple invariant moment feature sequences, construct a particle swarm optimization model based on the invariant moment features of the initial abnormal samples, and perform iterative optimization processing through the particle swarm optimization model to obtain key abnormal samples for blade fault diagnosis.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images