Wind turbine clutter classification identification method, device and storage medium

Abstract

The application discloses a wind power generator clutter grading identification method and device and a storage medium, and relates to the field of meteorological radar signal processing.The method comprises the following steps: obtaining an amplitude spectrum and a power spectrum by pre-processing radar IQ data; obtaining a radial noise threshold and a noise / signal marking matrix according to the power spectrum; removing discrete points and correcting misjudgments to obtain matrices C and E; multiplying the amplitude spectrum with the C and E points to obtain F and G; then performing grading judgment: if the maximum value of the F column is 0, marking no pollution, if the maximum values of the F and G columns are equal, marking serious pollution; otherwise, counting the number of E noise position power exceeding the standard, if more than half, calculating the fourth-order central moment ratio and the total power ratio, and marking moderate pollution or noise anomaly according to the fourth-order central moment ratio and the total power ratio; otherwise, detecting isolated strong points to mark slight pollution; and finally reducing isolated noise anomaly or slight pollution to no pollution.The application can finely classify wind power generator clutter into five levels, and provides a basis for subsequent differential suppression.

Description

Technical Field

[0001] This invention belongs to the field of meteorological radar signal processing technology, and in particular relates to a method, device and storage medium for classifying and identifying wind turbine clutter in phased array weather radar. It is especially suitable for classifying and identifying the degree of wind turbine clutter pollution, providing a basis for subsequent differentiated clutter suppression. Background Technology

[0002] In weather radar detection, the received echo signals are typically a mixture of meteorological targets (such as precipitation and clouds) and non-meteorological targets (such as ground features, organisms, and wind turbines). Non-meteorological target echoes significantly affect the estimation accuracy of key meteorological parameters such as reflectivity factor, radial velocity, spectral width, differential reflectivity, differential phase shift, correlation coefficient, and differential phase shift rate. Among these, wind turbine clutter interference (composed of nacelle, tower, rotor, and blades) is particularly prominent: the nacelle and tower are stationary relative to the radar, generating fixed echoes similar to ground clutter; while the blades rotate with the wind, generating dynamic, non-uniform Doppler frequencies, whose spectral characteristics are sometimes highly similar to weather echoes. Therefore, accurately identifying wind turbine clutter is a challenge in weather radar signal processing.

[0003] Currently, wind turbine clutter identification methods are mainly divided into three categories. The first category uses spatial marking based on the known geographical location of the wind turbine. However, this method struggles to accurately match the range database of high-resolution radar, and the clutter's influence range dynamically changes over time, easily leading to misclassification of uncontaminated range databases. The second category extracts features from radar base data (such as reflectivity, velocity, and spectral width) for identification. This can handle contaminated areas with obvious characteristics, but it can only perform post-processing such as subtraction and interpolation on the contaminated range database. It cannot select differentiated suppression strategies based on the degree of contamination, and not all contamination is suitable for interpolation repair. The third category uses spectral characteristics based on IQ data for real-time detection during signal processing, offering advantages such as high real-time performance and direct clutter suppression in the frequency domain.

[0004] In the third type of method, existing techniques employ clutter phase alignment (CPA) and spectral flatness. ), Fourth Central Spectral Moment The method uses features such as the hub-to-weather ratio (HWR) and fuzzy logic to distinguish wind turbine clutter. While this method can effectively differentiate areas mixed with wind turbine clutter and weather echoes, its output is only a binary decision of "contaminated" or "uncontaminated," failing to reflect the degree of clutter contamination.

[0005] However, the impact of wind turbine clutter on weather echoes varies in intensity: some distance data are severely obscured (e.g., by hub or blade main lobe contamination), some are only slightly contaminated or interfered with by sidelobes, and some only show anomalies in noise areas without affecting the weather signal. If binary classification is used, subsequent processing can only uniformly apply interpolation or spectral replacement to the contaminated distance data, easily leading to incomplete suppression of heavily contaminated areas and over-processing of lightly contaminated areas, resulting in the loss of accurate weather information. Therefore, the core technical problem facing existing technologies is the inability to classify and identify the degree of wind turbine clutter contamination, resulting in a lack of differentiated strategies for subsequent clutter suppression and making it difficult to effectively suppress clutter while preserving weather information. Summary of the Invention

[0006] To address the shortcomings of existing technologies that can only output binary results of "polluted" or "unpolluted," failing to provide fine-grained classification of pollution levels, resulting in a single clutter suppression method and difficulty in adopting differentiated suppression strategies based on actual pollution levels, thus causing incomplete suppression in heavily polluted areas and over-processing in lightly polluted areas leading to the loss of accurate weather information, this invention provides a wind turbine clutter classification and identification method, device, and storage medium. The aim is to provide a classification basis for subsequent adaptive clutter suppression by classifying the pollution levels in a distance database at multiple levels, thereby preserving and restoring accurate weather information to the greatest extent possible.

[0007] This invention solves the above-mentioned technical problems through the following technical solution: a method for classifying and identifying clutter in wind turbine generators, comprising:

[0008] The IQ data for each range library in each radial direction of the radar are preprocessed to obtain the amplitude spectrum and power spectrum of each range library. Based on the power spectrum, the radial noise threshold and noise / signal marker matrix are obtained. Discrete point removal and misjudgment correction are performed on the marker matrix sequentially to obtain matrices C and E. The amplitude spectrum is then multiplied by matrices C and E respectively to obtain matrices F and G, and then a hierarchical judgment is performed.

[0009] (1) If the maximum value of the i-th column of matrix F is 0, it is marked as no pollution level; if the maximum value of the i-th column of matrix F is not 0 and is equal to the maximum value of the i-th column of matrix G, it is marked as severely polluted level; otherwise, proceed to (2).

[0010] (2) Calculate the number of power values ​​at positions where the power spectrum is 0 in matrix E that are greater than the radial noise threshold. If this number is greater than half the total number of noise points in the distance pool, calculate the ratio of the fourth-order central moments and the total power ratio of the distance pool and its neighborhood. If the ratio of the fourth-order central moments is greater than the first threshold and the total power ratio is greater than or equal to the second threshold, it is marked as a medium pollution level. If the ratio of the fourth-order central moments is greater than the first threshold and the total power ratio is less than the second threshold, it is marked as a noise anomaly level. Otherwise, proceed to (3).

[0011] (3) For the remaining unmarked distance library, traverse its power spectrum frequency points. If there is a frequency point where the difference between the power and the average power of the surrounding ring area is greater than the third threshold, then mark it as a slightly polluted level; otherwise, mark it as a non-polluted level.

[0012] (4) For a distance library with abnormal noise or slight pollution, if there is no serious pollution in its surrounding area, it shall be downgraded to a pollution-free level.

[0013] This invention obtains optimized labeling matrices C and E by performing noise labeling, discrete point removal, and misjudgment correction on the spectrum of each distance library. Based on multi-dimensional features such as the dot product of the amplitude spectrum with C and E, noise power statistics, the ratio of fourth-order central moments, the total power ratio, and the detection of isolated strong points, the pollution level is subdivided into five levels: no pollution (P5), severe pollution (P1), moderate pollution (P2), abnormal noise (P3), and slight pollution (P4). Compared with the binary classification of existing technologies, this invention can accurately reflect the actual impact of wind turbine clutter on weather echoes, providing a reliable level basis for subsequent differentiated suppression.

[0014] In step (1), the maximum values ​​of the columns of matrices F and G are compared: if the maximum value of column F is 0, it indicates that there are no non-noise signals in the range database, and it is marked as unpolluted; if the maximum value of column F is non-zero and equal to the maximum value of column G, it indicates that the clutter power is much greater than the weather signal, and it is marked as severely polluted. The weather parameter estimates in these two cases are seriously unreliable, and subsequent strong suppression or direct removal can be performed to avoid polluting meteorological products and improve the quality of radar data.

[0015] Step (2) Calculate the fourth-order central moment ratio and the total power ratio when the noise power ratio condition is met. When the fourth-order central moment ratio is greater than the first threshold and the total power ratio is greater than or equal to the second threshold, it is marked as a moderate pollution level (P2), indicating strong clutter power and severe impact on weather signals. When the fourth-order central moment ratio is greater than the first threshold but the total power ratio is less than the second threshold, it is marked as an abnormal noise level (P3), indicating an abnormal spectrum shape but no significant deviation of the total power of the weather signal from the neighborhood, possibly only affecting the noise area. Different suppression methods such as spectral interpolation or local noise replacement can be used for P2 and P3 respectively, avoiding over-suppression or under-suppression caused by uniform processing.

[0016] Step (3) involves iterating through the power spectrum frequency points of the remaining unmarked distance database and calculating the difference between the power at each frequency point and the average power of the surrounding annular region. If the difference exceeds the third threshold, it is marked as a slight pollution level (P4). This operation can accurately locate isolated strong points or sidelobe interference caused by wind turbine clutter. These slight pollutions are often ignored or cause the entire distance database to be removed in traditional binary classification. This invention marks them separately, and subsequent lightweight processing such as local noise replacement can be used to retain effective weather information to the maximum extent.

[0017] Step (4) utilizes the physical characteristic that wind turbine clutter is often accompanied by severe pollution (P1) in space to perform a neighborhood check on the distance database that has been marked as noise anomaly (P3) or slight pollution (P4): if there is no severe pollution level (P1) in its immediate and subsequent neighborhoods, it is downgraded to a pollution-free level (P5). This post-processing step effectively eliminates misjudgments caused by noise fluctuations or atypical interference, improving the robustness and reliability of the classification results.

[0018] In summary, this invention classifies wind turbine clutter into five levels, enabling distance databases with different pollution levels to employ the most suitable suppression strategies (such as interpolation in heavily polluted areas, adaptive filtering in moderately polluted areas, local noise replacement in lightly polluted areas, and preservation of original data in unpolluted areas). This effectively suppresses clutter while preserving and restoring the true weather echo information to the greatest extent, significantly improving the estimation accuracy of meteorological parameters such as reflectivity factor, radial velocity, and spectral width.

[0019] Furthermore, the IQ data for each range library in each radial direction of the radar are preprocessed, including:

[0020] Perform an FFT along the slow time dimension on the IQ data of each radial range library of the radar to obtain the complex spectrum;

[0021] Ground clutter suppression is applied to the complex spectrum;

[0022] The amplitude spectrum and power spectrum are calculated based on the suppressed complex spectrum.

[0023] This invention effectively eliminates the interference of ground object echoes (such as buildings, mountains, etc.) on subsequent wind turbine clutter identification and noise power estimation by first suppressing ground clutter. This avoids ground object clutter being misidentified as wind turbine clutter or causing an overestimation of the noise floor, thereby improving the accuracy of hierarchical identification. Simultaneously, it clarifies the methods for obtaining the amplitude spectrum and power spectrum, providing a correct data foundation for subsequent operations such as dot multiplication and noise statistics.

[0024] Furthermore, the radial noise threshold and noise / signal label matrix are obtained based on the power spectrum, including:

[0025] The power values ​​of each distance library are sorted in ascending order. The average of the smallest r1% power values ​​is taken as the noise power benchmark of that distance library. The noise power benchmark is multiplied by a first preset coefficient to obtain the noise threshold.

[0026] The power at each frequency point is compared with the noise threshold. If it is less than the noise threshold, it is marked as noise and 0 is filled in the corresponding position; otherwise, it is marked as signal and 1 is filled in the corresponding position, thus forming a noise / signal marking matrix.

[0027] Sort the noise power benchmarks of all current radial distance libraries in ascending order, take the smallest r2% to calculate the arithmetic mean, and then multiply it by the second preset coefficient to obtain the radial noise threshold.

[0028] This invention employs adaptive statistical sorting technology, which dynamically sets thresholds based on the actual noise level of each distance library, avoiding the deviation of fixed thresholds in non-uniform noise environments (such as ground cover leakage and radio frequency interference). By taking the minimum percentage power value, it effectively eliminates the contamination of noise estimation by signals and clutter, ensuring the purity of the noise benchmark. The radial noise threshold further smooths out noise fluctuations along the radial direction, improving the stability and consistency of the noise criterion.

[0029] Furthermore, the discrete point removal employs connected component analysis, including:

[0030] Treating the label matrix as a binary image, all connected regions with a value of 1 are labeled using 4-connected or 8-connected neighborhood rules. All pixel values ​​in connected regions with an area smaller than a preset area are changed to 0, resulting in matrix C.

[0031] The present invention employs connected component analysis to effectively remove isolated, small-area non-noise points (such as random noise spikes or false signal points caused by spectral leakage), preventing these discrete points from being misjudged as valid weather signals or wind turbine clutter, thereby improving the spatial continuity of the marker matrix and reducing the false alarm rate in subsequent classification judgments.

[0032] Furthermore, the misjudgment correction employs a sliding window method, including:

[0033] For each element in matrix C, take a window of size w×h centered on the current element; count the number of non-noise points or points with a value of 1 in the window. If the number of points is greater than half of the total number of points in the window, change the label of the current element to signal and change its value to 1; otherwise, keep the label and value of the current element unchanged, thus forming matrix E.

[0034] This invention employs a sliding window method for misjudgment correction. Through a neighborhood majority voting mechanism, signal points that were misidentified as noise due to excessive interference (such as strong clutter sidelobes or transient noise) are restored to signals, effectively filling the "holes" in the weather signal area and making the marking matrix more consistent with the physical characteristics of the continuous spatial variation of weather echoes. Simultaneously, this operation does not excessively expand the signal area, maintaining the clarity of the clutter and weather boundary, thereby improving the accuracy of subsequent classification judgments (especially the determination of severe and moderate pollution).

[0035] Furthermore, the formula for calculating the fourth-order central moment ratio is as follows:

[0036] ;

[0037] ;

[0038] in, The ratio of the fourth-order central moments; L1 is the fourth central moment of the power spectrum of the i-th distance library; L2 is the length of the comparison interval, that is, L2 distance libraries on both sides of the protection interval participate in the comparison; L1 is the length of the protection interval, that is, L1 distance libraries on both sides of the i-th distance library do not participate in the comparison; M is the number of frequency elements of the i-th distance library. The Doppler frequency value of the k-th frequency unit; Let be the first moment of the power spectrum of the i-th distance library, i.e., the average Doppler frequency;

[0039] The formula for calculating the total power ratio is:

[0040] ;

[0041] in, This is the ratio of total power. It is the sum of the power values ​​of all frequency units in the power spectrum of the i-th distance library.

[0042] The fourth-order central moment (kurtosis) measures the kurtosis of the power spectrum: the spectrum of pure weather noise is close to a Gaussian distribution, with a kurtosis of approximately 3; while wind turbine clutter, due to Doppler broadening caused by blade rotation, often exhibits a flat or bimodal spectrum, with a kurtosis deviating from 3. The fourth-order central moment ratio compares the kurtosis of the current distance library with the mean of its neighborhood (outside the protection zone), highlighting local anomalies. The total power ratio compares the difference between the total power of the current distance library and the mean of its neighborhood. Both can be used together: when... Larger and When the value is also large, it indicates that the current distance reservoir not only has an abnormal spectral shape, but also has significantly higher energy than the surrounding area, belonging to moderate pollution dominated by clutter (P2); when Larger but Under normal conditions, this indicates an abnormal spectrum shape but no significant deviation in total power, suggesting that only the noise region may be affected (P3). This joint judgment provides a precise classification basis for differentiated suppression strategies.

[0043] Furthermore, the difference between the power at a certain frequency point and the average power of the surrounding ring region is calculated using the following formula:

[0044] Let the current frequency point be located at (k, i) in the power spectrum, where k is the frequency cell index and i is the distance library index; define the frequency direction half-width of the large neighborhood window as Rf and the distance library direction half-width as Rr, and the frequency direction half-width of the inner guard window as rf and the distance library direction half-width as rr, where Rf > rf and Rr > rr; then:

[0045] The average power in the annular region is:

[0046] ;

[0047] in, This represents the number of frequency points within the ring-shaped region; Let (k,i) be the power value at the mid-frequency point (k,i) of the power spectrum; the difference is:

[0048] .

[0049] This invention defines a large neighborhood window and an internal protection window, subtracting the target's own internal protection area from the annular region, and only calculating the average power of the surrounding background. (Difference) This reflects the prominence of the current frequency point relative to the background. This method can effectively detect isolated strong points (such as narrowband components scattered by blades) or sidelobe interference caused by wind turbine clutter. Furthermore, due to its universal parameter format, it can adapt to different radar parameters (such as range resolution and FFT point count) and different environmental noise levels. The third threshold can be set independently, providing an objective and quantitative basis for determining the slight pollution level (P4).

[0050] Furthermore, the frequency direction half-width Rf of the large neighborhood window is set to 10, the distance direction half-width Rr is set to 5, and the frequency direction half-width rf of the internal protection window is set to 3 and the distance direction half-width rr is set to 2.

[0051] The above set of values ​​was obtained through experimental optimization under typical weather radar parameters (such as 128–256 FFT points, 15-meter range-to-base spacing, and 1000–2000 Hz pulse repetition frequency): the large neighborhood window covers approximately 21 frequency units and 11 range bases (approximately 300 meters), which can fully statistically analyze background noise; the internal protection window covers 7 frequency units and 5 range bases (approximately 150 meters), effectively avoiding the contamination of background statistics by the spectral broadening of the wind turbine target itself. This optimized parameter achieves the best balance between detecting isolated strong points and suppressing false alarms, significantly improving the accuracy and robustness of identifying slight pollution levels.

[0052] Based on the same concept, the present invention also provides an electronic device, including a memory, a processor, and a computer program or instructions stored in the memory, wherein the processor executes the computer program or instructions to implement the wind turbine clutter classification identification method as described above.

[0053] Based on the same concept, the present invention also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implements the wind turbine clutter classification and identification method as described above. Attached Figure Description

[0054] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only one embodiment of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 This is a flowchart of the wind turbine clutter classification and identification method in an embodiment of the present invention;

[0056] Figure 2 This is a flowchart of the hierarchical judgment process in an embodiment of the present invention;

[0057] Figure 3 This is the spectral power diagram before suppression under clear sky conditions in an embodiment of the present invention;

[0058] Figure 4 This is a spectrum power diagram after suppression under clear sky conditions in an embodiment of the present invention;

[0059] Figure 5 This is a comparison diagram of radial intensity before and after suppression under clear sky conditions in an embodiment of the present invention;

[0060] Figure 6 This is the spectrum power diagram before suppression under precipitation conditions in this embodiment of the invention;

[0061] Figure 7This is a spectrum power diagram after suppression under precipitation conditions in an embodiment of the present invention;

[0062] Figure 8 This is a comparison diagram of radial intensity before and after suppression under precipitation conditions in an embodiment of the present invention. Detailed Implementation

[0063] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0065] Example 1

[0066] To address the shortcomings of existing wind turbine clutter identification methods, which only output binary results of "polluted" or "unpolluted," failing to differentiate clutter pollution levels and resulting in simplistic clutter suppression methods that lack differentiated suppression strategies based on actual pollution levels, leading to incomplete suppression in heavily polluted areas and over-processing in lightly polluted areas that loses accurate weather information, this invention provides a wind turbine clutter classification identification method. This method aims to finely classify pollution levels, providing a grading basis for subsequent adaptive suppression. Figure 1 As shown, the hierarchical identification method includes the following steps:

[0067] Step S1: Preprocess the IQ data of each range library in each radial direction of the radar to obtain the amplitude spectrum and power spectrum of each range library.

[0068] In this embodiment, the IQ data of each range library in each radial direction of the radar are preprocessed to obtain the amplitude spectrum and power spectrum of each range library, specifically including:

[0069] Step S1.1: Perform FFT on the IQ data of each radial range library of the radar along the slow time dimension to obtain the complex spectrum.

[0070] Suppose a weather radar collects n range databases in a certain radial direction, each range database corresponding to a pulse sequence, with m pulses (i.e., the slow time dimension length). For the i-th range database (i=1,2,…,n), its IQ data is a complex sequence. ,in This represents the complex baseband signal at the i-th distance at the t-th pulse time. and These represent the in-phase component and the quadrature component, respectively. j is the imaginary unit, and t is the sequence number of the pulse sequence (slow time index), t=1,2,…,m.

[0071] Perform an M-point FFT transform (usually) on the above complex sequence along the slow time dimension. (Alternatively, zeros can be added to increase the number of points), resulting in a complex spectrum sequence; arrange the complex spectra of all distance libraries column-wise to form a complex spectrum matrix S with dimensions M×n.

[0072] Step S1.2: Suppress ground clutter on the complex spectrum.

[0073] Ground clutter identification and suppression are performed on each range library in the complex spectrum matrix S. Specifically, for the i-th range library, based on its complex spectrum... The system determines whether the range database contains ground clutter. If it is determined to be ground cover, clutter suppression is performed on the range database to obtain the suppressed complex spectrum. If it is determined to be non-ground cover, the complex spectrum remains unchanged. This yields the suppressed complex spectrum matrix.

[0074] Step S1.3: Calculate the amplitude spectrum and power spectrum based on the suppressed complex spectrum.

[0075] For the i-th distance library, its amplitude spectrum Its power spectrum For all distances in the current radial direction, the amplitude spectrum and power spectrum are calculated, and the amplitude spectrum matrix A and power spectrum matrix P are output.

[0076] Step S2: Obtain the radial noise threshold and noise / signal marker matrix based on the power spectrum.

[0077] The power spectrum matrix P output in step S1 has dimensions M×n, where M is the number of frequency elements (FFT points) and n is the current radial distance unit. This represents the power value of the i-th distance unit and the k-th frequency unit.

[0078] For the i-th distance column, retrieve the M power values ​​from that column, sort them in ascending order, and take the smallest r1% power values, where r1=25. A power value is calculated. The arithmetic mean of these power values ​​is used as the noise power reference for the distance library. The noise power reference is multiplied by a first preset coefficient (which is 2 in this embodiment) to obtain the noise threshold T. i .

[0079] like If the position is marked as noise, fill the corresponding position in the noise / signal marking matrix B with 0; otherwise, mark it as a signal and fill it with 1.

[0080] After traversing all distance libraries and all frequency units, a binary label matrix B is obtained, with dimensions M×n, where 0 represents noise and 1 represents non-noise.

[0081] For the current radial direction, collect the noise power references for each distance library obtained in the above steps, resulting in n values. Sort these values ​​in ascending order from smallest to largest, and select the smallest r2% of the noise power references, where r2=50, i.e., the top [number] values. A set of values ​​are calculated. The arithmetic mean of these values ​​is then multiplied by a second preset coefficient (which is 2 in this embodiment) to obtain the radial noise threshold.

[0082] In this embodiment, r1=25, r2=50, and both the first and second preset coefficients are 2. These values ​​are optimal values ​​obtained based on the noise statistical characteristics of typical weather radars and experimental optimization. In practical applications, if radar operating parameters (such as pulse repetition frequency, receiver noise figure, etc.) change significantly, these parameters can be adjusted appropriately.

[0083] Step S3: Perform discrete point removal and misjudgment correction on the marker matrix in sequence to obtain matrices C and E.

[0084] The noise / signal label matrix B (a binary matrix, where 0 represents noise and 1 represents no noise) generated in step S2 is treated as a binary image. All connected regions with a value of 1 are labeled using the 8-connectivity neighborhood rule. 8-connectivity means that the neighborhood of each pixel includes its eight neighboring pixels in the directions above, below, left, right, and four diagonals.

[0085] The specific operation is as follows: traverse each element of matrix B. When encountering a pixel with a value of 1 that is not marked, use that pixel as the seed point and recursively or iteratively visit all pixels with a value of 1 that are reachable through 8-connected paths using a depth-first search method, assigning them the same connected region number. Continue scanning until all pixels with a value of 1 are marked.

[0086] For each connected region, its area (i.e., the number of pixels contained within that region) is calculated, and this area is compared with a preset area. In this embodiment, the preset area is 3 pixels.

[0087] If the area of ​​a connected region is smaller than a preset area, the connected region is considered to be an isolated discrete point caused by noise or spectral leakage. The value of all pixels in the connected region is changed from 1 to 0 (i.e., from non-noise to noise). After traversing all connected regions, a binary matrix C after removing discrete points is obtained. Its dimension is the same as B, still M×n, where 0 represents noise and 1 represents non-noise.

[0088] Correcting sliding window misjudgments in matrix C: The size of the sliding window is set to w×h. In this embodiment, w=3 and h=3, meaning the window size is 3×3 pixels. For each element (center point) in matrix C, a 3×3 window is taken with its current position as the center. When the center point is located on the matrix boundary, zero padding or only pixels within the effective range can be used (in this embodiment, boundary points are not processed, or are only processed when the window is completely within the matrix; to simplify the operation, boundary points can be ignored because the boundary distance database usually does not contain effective meteorological targets).

[0089] Count the number of pixels with a value of 1 within the window (i.e., the number of non-noise points). If the number of non-noise points is greater than half of the total number of points in the window (3×3 / 2), that is, the number of non-noise points is ≥5, and the current center point value is 0 (noise), then the center point is considered to be misidentified as noise and should be corrected to a signal, that is, the value of the center point is changed to 1; otherwise, the original value of the center point remains unchanged.

[0090] The above judgment and correction operations are performed sequentially on all elements of matrix C (boundary points can be ignored or processed according to the same rules), and finally the misjudgment corrected binary matrix E is obtained. The dimension of E is still M×n, and the value meaning is the same as that of C (0 is noise, 1 is non-noise), but it fills the signal gaps that were misjudged as noise due to excessive local interference.

[0091] Step S4: Multiply the amplitude spectrum by matrices C and E respectively to obtain matrices F and G.

[0092] From step S1, we obtain the amplitude spectrum matrix A. From step S3, we obtain two binary label matrices C and E. We then perform an element-wise dot product between the amplitude spectrum matrix A and matrix C, that is, multiply the two values ​​at corresponding positions to obtain matrix F. Matrix F retains all amplitude values ​​in the amplitude spectrum that are marked as non-noise positions by C, while setting the amplitude values ​​at noise positions to zero. Similarly, we perform an element-wise dot product between the amplitude spectrum matrix A and matrix E to obtain matrix G. Matrix G retains all amplitude values ​​in the amplitude spectrum that are marked as non-noise positions by E, while setting the amplitude values ​​at noise positions to zero.

[0093] F eliminates the influence of isolated, discrete, non-noise points, while E, compared to C, fills signal holes that are mistakenly identified as noise. Therefore, comparing the maximum values ​​of F and G in the same column can determine whether the clutter is severe enough to dominate the entire spectrum.

[0094] Step S5: Grading judgment.

[0095] After completing steps S1 to S4 and obtaining the amplitude spectrum matrix A, power spectrum matrix P, radial noise threshold, label matrices C and E, and dot product result matrices F and G, as follows: Figure 2 As shown, each distance database is classified and judged according to the following sub-steps.

[0096] (1) Identification of severe pollution and pure noise.

[0097] For the i-th distance matrix, extract the i-th column of matrix F to obtain the column vector F(:,i); extract the i-th column of matrix G to obtain the column vector G(:,i), and calculate the maximum value F of column vector F(:,i). imax .

[0098] If F imax If the value is 0, the distance database is marked as pollution-free level P5, the evaluation of the distance database ends, and the process moves to the next distance database.

[0099] Otherwise (F imax >0), calculate the maximum value G of column vector G(:,i). imax If F imax =G imax If the distance is marked as severely polluted (P1), the evaluation of that distance database ends, and the process moves to the next distance database.

[0100] Otherwise, proceed to step (2) to continue the judgment.

[0101] (2) Identification of moderate pollution and noise anomalies.

[0102] Get the i-th column E(:,i) of matrix E, find all positions in the column with a value of 0 (i.e., noise positions or noise points), and let the set of row indices of these noise positions be Kno={k|E(k,i)=0}, and the total number of noise points in this distance library be Num1=|Kno|.

[0103] For each k∈Kno, extract the power value P(k,i) at the same position (k,i) in the power spectrum matrix P, and compare it with the radial noise threshold. Count the number of cases where P(k,i) > radial noise threshold, denoted as count.

[0104] If count > Num1 / 2, then calculate the fourth-order central moment ratio and the total power ratio; otherwise, proceed to step (3).

[0105] Due to the influence of pulse compression, the location of the target will affect multiple nearby distance databases. Therefore, a protection interval is set. Considering the relatively discrete installation of wind turbines and the widening effect after pulse compression, all distances within the comparison interval are calculated with the current distance database as the center. In this embodiment, the distance database spacing is 15 meters, the protection interval is 150 meters, and the comparison interval is from 150 meters to 300 meters. Therefore, the length of the protection interval is set to L1 = 5 distance databases, and the length of the comparison interval is L2 = 5 distance databases. The protection interval consists of L1 distance databases on each side of the current distance database (i.e., indices i−L1 to i−1 and i+1 to i+L1). These distance databases are not included in the mean calculation. The comparison interval consists of L2 distance databases outside the protection interval. That is, the left comparison interval index is i−L1−L2 to i−L1−1 (a total of L2), and the right comparison interval index is i+L1+1 to i+L1+L2 (a total of L2). These distance databases are included in the calculation. Therefore, the total neighborhood range is L1+L2=10 distance libraries before and after.

[0106] For the current distance library i and each distance library l (i.e., the neighboring distance library) within the comparison interval, its fourth-order central moment can be calculated according to the following formula:

[0107] ;

[0108] in, Let M be the fourth central moment of the power spectrum of the i-th distance library; M is the number of frequency elements in the i-th distance library. The Doppler frequency value of the k-th frequency unit; Let be the first moment of the power spectrum of the i-th distance library, i.e., the average Doppler frequency.

[0109] Calculate the arithmetic mean of the fourth-order central moments of all distance libraries within the comparison interval:

[0110] ;

[0111] The ratio of the fourth-order central moments is: .

[0112] Calculate the total power of the current distance library Calculate the arithmetic mean of the total power of all distance libraries within the comparison interval:

[0113] ;

[0114] Then the total power ratio .

[0115] If the ratio of the fourth-order central moments > First threshold and total power ratio If the pollution level is greater than or equal to the second threshold, it is marked as medium pollution level P2;

[0116] If the ratio of the fourth-order central moments > First threshold and total power ratio If the value is less than the second threshold, it is marked as noise anomaly level P3;

[0117] Otherwise proceed to step (3).

[0118] In this embodiment, the first threshold ε1 is set to 2.1 and the second threshold ε2 is set to 20.

[0119] (3) Identification of minor contamination.

[0120] For distances that still do not have a grade label after steps (1) and (2), traverse all frequency points in their power spectrum. For the current frequency point (k,i), calculate the difference between its power and the average power of the surrounding ring region.

[0121] Define the annular region:

[0122] Large neighborhood window: the half-width in the frequency direction is Rf (i.e., the frequency range k ± Rf), and the half-width in the distance to the library direction is Rr (the distance to the library range i ± Rr).

[0123] Internal protection window: half-width in the frequency direction is rf, and half-width in the distance direction is rr.

[0124] In this embodiment, the frequency direction half-width Rf is set to 10, the distance to the library direction half-width Rr is set to 5, the frequency direction half-width rf is set to 3, and the distance to the library direction half-width rr is set to 2. Then the number of frequency points in the annular area is 21×11-7×5=196.

[0125] The average power in the annular region is:

[0126] ;

[0127] in, This represents the number of frequency points within the ring-shaped region; Let be the power value at frequency point (k,i) in the power spectrum.

[0128] The difference between the power at the current frequency and the average power of the surrounding ring region is: .

[0129] If there exists a difference corresponding to any frequency point If the distance is greater than the third threshold ε3 (e.g., 5.6), then the distance library is marked as slightly polluted level P4; otherwise, it is marked as unpolluted level P5.

[0130] (4) Anomaly detection.

[0131] For a distance database already marked as noise anomaly level P3 or slight pollution level P4, check if a distance database of severe pollution level P1 exists within five distance databases before and after it. If no distance database of severe pollution level P1 exists, downgrade the level of that distance database to no pollution level P5; otherwise, keep the level of that distance database unchanged.

[0132] The final output is the pollution level of each distance library: P1 (severe pollution), P2 (medium pollution), P3 (abnormal noise), P4 (slight pollution), and P5 (no pollution).

[0133] To verify the effectiveness of the invention, experiments were conducted using measured wind turbine clutter data under clear sky and precipitation conditions, respectively.

[0134] Using wind turbine clutter data scanned by a weather radar under clear skies, a specific radial data point is selected. After hierarchical identification through steps S1-S5 of this invention, statistical noise is used to suppress the clutter in the range library identified as contamination. A comparison of the spectrum and intensity before and after suppression is provided. Figures 3 to 5 As shown.

[0135] Figure 3 The image shows the power spectrum before suppression. The horizontal yellow band near 0Hz represents the residual ground clutter after suppression, while the vertical yellow band represents the clutter spectrum caused by the wind turbine. The clutter impact is greatest near a distance of 251 Hz, but due to pulse compression, there are interference sidelobes with low clutter-to-noise ratios and indistinct clutter characteristics on both sides.

[0136] Figure 4 After identification using the method of this invention, statistical noise was used to replace the suppressed spectral power diagram for the pollution distance database. It can be seen that the clutter in the distance database affected by wind turbine interference is essentially suppressed.

[0137] Figure 5 This is a comparison of radial intensity before and after suppression, where the blue curve represents the radial intensity distribution before suppression, and the orange curve represents the radial intensity distribution after suppression. From Figure 5 As can be seen, at a distance of 251 from the wind turbine interference, the noise-to-speech ratio is extremely high before suppression, with sidelobes on both sides. After suppression, the intensity spikes caused by the clutter are effectively eliminated. If traditional one-dimensional or two-dimensional interpolation frequency domain methods are used, they are easily affected by interference near the main signal; however, this invention uses a hierarchical identification method. After identifying that the area is clear-sky data, statistical noise is directly used to replace the original spectrum, which can suppress wind turbine clutter simply and quickly in the frequency domain.

[0138] Using wind turbine clutter data scanned during precipitation, a specific radial data point is selected. After hierarchical identification in steps S1-S5 of this invention, the pollution distance database is suppressed (statistical noise is used in this embodiment). A comparison of the spectrum and intensity before and after suppression is obtained, as shown below. Figures 6 to 8 As shown.

[0139] Figure 6 The spectrum data before suppression shows obvious bright bands, indicating wind turbine clutter pollution.

[0140] Figure 7 The bright band region is effectively suppressed in the spectral data identified and suppressed using the method of the present invention.

[0141] Figure 8 This is a comparison of radial intensity before and after suppression, where the blue curve represents the radial intensity distribution before suppression, and the orange curve represents the radial intensity distribution after suppression. From Figure 8 It is evident that the suppressed radial intensity is closer to the intensity of real weather echoes, and the intensity jitter caused by clutter is significantly reduced.

[0142] The above experimental results show that the method of the present invention can effectively identify and classify the distance database of wind turbine clutter pollution, and can provide accurate classification basis for subsequent suppression under different weather conditions, thereby achieving good clutter suppression effect and providing a more reliable data basis for weather parameter estimation.

[0143] Example 2

[0144] This invention also provides an electronic device, which includes a memory, a processor, and a computer program or instructions stored in the memory. The processor executes the computer program or instructions to implement the wind turbine clutter classification and identification method of this invention.

[0145] Although not shown, the electronic device includes a processor that can perform various appropriate operations and processes based on programs and / or data stored in read-only memory (ROM) or loaded from a storage portion into random access memory (RAM). The processor can be a multi-core processor or may contain multiple processors. In some embodiments, the processor may include a general-purpose main processor and one or more specialized coprocessors, such as a central processing unit, graphics processing unit (GPU), neural network processor (NPU), digital signal processor (DSP), etc. Various programs and data required for device operation are also stored in RAM. The processor, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0146] The processor and memory described above are used together to execute programs / instructions stored in the memory. When the program / instructions are executed by the computer, they can implement the methods, steps, or functions described in the above embodiments.

[0147] Although not shown, embodiments of the present invention also provide a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implements the wind turbine clutter classification and identification method of the present invention.

[0148] Readable storage media include both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0149] The above description only discloses specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or modifications that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for classifying and identifying clutter in wind turbine generators, characterized in that, include: The IQ data of each range library in each radial direction of the radar are preprocessed to obtain the amplitude spectrum and power spectrum of each range library; The radial noise threshold and noise / signal marker matrix are obtained from the power spectrum. Discrete point removal and misjudgment correction are performed on the marker matrix sequentially to obtain matrices C and E. The amplitude spectrum is then multiplied by matrices C and E respectively to obtain matrices F and G, followed by a graded judgment. (1) If the maximum value of the i-th column of matrix F is 0, it is marked as no pollution level; if the maximum value of the i-th column of matrix F is not 0 and is equal to the maximum value of the i-th column of matrix G, it is marked as severely polluted level; otherwise, proceed to (2). (2) Calculate the number of power values ​​at positions where the power spectrum is 0 in matrix E that are greater than the radial noise threshold. If this number is greater than half the total number of noise points in the distance pool, calculate the ratio of the fourth-order central moments and the total power ratio of the distance pool and its neighborhood. If the ratio of the fourth-order central moments is greater than the first threshold and the total power ratio is greater than or equal to the second threshold, it is marked as a medium pollution level. If the ratio of the fourth-order central moments is greater than the first threshold and the total power ratio is less than the second threshold, it is marked as a noise anomaly level. Otherwise, proceed to (3). (3) For the remaining unmarked distance library, traverse its power spectrum frequency points. If there is a frequency point where the difference between the power and the average power of the surrounding ring area is greater than the third threshold, then mark it as a slightly polluted level; otherwise, mark it as a non-polluted level. (4) For a distance library with abnormal noise or slight pollution, if there is no serious pollution in its surrounding area, it shall be downgraded to a pollution-free level. The radial noise threshold and noise / signal label matrix are obtained based on the power spectrum, including: The power values ​​of each distance library are sorted in ascending order. The average of the smallest r1% power values ​​is taken as the noise power benchmark of that distance library. The noise power benchmark is multiplied by a first preset coefficient to obtain the noise threshold. The power at each frequency point is compared with the noise threshold. If it is less than the noise threshold, it is marked as noise and 0 is filled in the corresponding position; otherwise, it is marked as signal and 1 is filled in the corresponding position, thus forming a noise / signal marking matrix. Sort the noise power benchmarks of all current radial distance libraries in ascending order, take the smallest r2% to calculate the arithmetic mean, and then multiply it by the second preset coefficient to obtain the radial noise threshold.

2. The wind turbine clutter classification and identification method according to claim 1, characterized in that, Preprocessing of IQ data for each range library in each radial direction of the radar is performed, including: Perform an FFT along the slow time dimension on the IQ data of each radial range library of the radar to obtain the complex spectrum; Ground clutter suppression is applied to the complex spectrum; The amplitude spectrum and power spectrum are calculated based on the suppressed complex spectrum.

3. The wind turbine clutter classification and identification method according to claim 1, characterized in that, The discrete point removal employs connected component analysis, including: Treating the label matrix as a binary image, all connected regions with a value of 1 are labeled using 4-connected or 8-connected neighborhood rules. All pixel values ​​in connected regions with an area smaller than a preset area are changed to 0, resulting in matrix C.

4. The wind turbine clutter classification and identification method according to claim 1, characterized in that, The misjudgment correction adopts the sliding window method, including: For each element in matrix C, take a window of size w×h centered on the current element; count the number of non-noise points or points with a value of 1 in the window. If the number of points is greater than half of the total number of points in the window, change the label of the current element to signal and change its value to 1; otherwise, keep the label and value of the current element unchanged, thus forming matrix E.

5. The wind turbine clutter classification and identification method according to claim 1, characterized in that, The formula for calculating the fourth-order central moment ratio is: ； ； in, The ratio of the fourth-order central moments; L1 is the fourth central moment of the power spectrum of the i-th distance library; L2 is the length of the comparison interval, that is, L2 distance libraries on both sides of the protection interval participate in the comparison; L1 is the length of the protection interval, that is, L1 distance libraries on both sides of the i-th distance library do not participate in the comparison; M is the number of frequency elements of the i-th distance library. The Doppler frequency value of the k-th frequency unit; Let be the first moment of the power spectrum of the i-th distance library, i.e., the average Doppler frequency; The formula for calculating the total power ratio is: ； in, This is the ratio of total power. It is the sum of the power values ​​of all frequency units in the power spectrum of the i-th distance library.

6. The wind turbine clutter classification and identification method according to any one of claims 1 to 5, characterized in that, The difference between the power at a certain frequency point and the average power of the surrounding ring region is calculated using the following formula: Let the current frequency point be located at (k, i) in the power spectrum, where k is the frequency cell index and i is the distance library index; define the frequency direction half-width of the large neighborhood window as Rf and the distance library direction half-width as Rr, and the frequency direction half-width of the inner guard window as rf and the distance library direction half-width as rr, where Rf > rf and Rr > rr; then: The average power in the annular region is: ； in, This represents the number of frequency points within the ring-shaped region; Let (k,i) be the power value at the mid-frequency point (k,i) of the power spectrum; the difference is: 。 7. The wind turbine clutter classification and identification method according to claim 6, characterized in that, The frequency direction half-width Rf of the large neighborhood window is set to 10, the distance direction half-width Rr is set to 5, and the frequency direction half-width rf of the internal protection window is set to 3 and the distance direction half-width rr is set to 2.

8. An electronic device comprising a memory, a processor, and a computer program or instructions stored in the memory, characterized in that, The processor executes the computer program or instructions to implement the wind turbine clutter classification and identification method as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by the processor, they implement the wind turbine clutter classification and identification method as described in any one of claims 1 to 7.

Images