Method for establishing wind power typical curve based on high and low frequency signal combination reconstruction

Abstract

The present application relates to the field of wind power generation technology, especially to a wind power typical curve establishment method based on high-low frequency signal combination reconstruction, steps comprising: data cleaning stage, input original wind power data, eliminate wind power abnormal data; Normalization stage, the wind power data after cleaning is normalized; Wavelet decomposition stage, after multiple discrete wavelet transform, the normalized wind power data is decomposed into high frequency signal and low frequency signal; Combination reconstruction stage, the low frequency signal obtained after decomposition is clustered, the low frequency signal after clustering and the high frequency signal of the same layer are combined and reconstructed through inverse wavelet transform, and the typical wind power curve after combination reconstruction is output after multiple cycles; Data filling stage, according to the typical wind power curve after combination reconstruction, fill in the missing wind power data. The present application effectively mines the different periodic variation characteristics of wind power data, and ensures that the key information of wind power fluctuation form is not lost in the combination reconstruction process of high-low frequency signal.

Description

Technical Field

[0001] This invention relates to the field of wind power generation technology, and in particular to a method for establishing typical wind power curves based on high and low frequency signal combination reconstruction. Background Technology

[0002] Wind power generation, as a crucial component of the new energy industry, plays an increasingly important role in the global energy market, with annual power generation showing a year-on-year upward trend. Due to the intermittent, uncertain, and abrupt nature of wind resources, accurately grasping their changing patterns is crucial for power system operation and dispatching and electricity market transaction settlement. Influenced by geographical factors and monsoons, although wind power output varies randomly, it also exhibits a certain periodicity. Extracting the characteristics of wind power changes within a specific period to construct typical wind power curves is an effective way to understand the patterns of wind power variation.

[0003] Currently, methods for establishing typical wind power curves are mainly divided into statistical methods and clustering methods. Among them, the clustering method first performs unsupervised learning clustering on the input wind power data set, dividing the curve clusters based on curve similarity. Each wind power data point belongs to a unique curve cluster and has a high similarity to the center of that cluster. The center curves of each cluster output by the clustering method can usually serve as typical wind power curves, and can be used to fill in missing data in the wind power metering data.

[0004] However, the clustering method for establishing typical wind power curves mentioned above is mainly based on static indicators and fails to fully characterize the changing patterns of wind power. Since the detailed features and average features of wind power generation cannot be characterized simultaneously, there is a risk of missing information, which in turn affects the accuracy of clustering. Summary of the Invention

[0005] Therefore, the technical problem to be solved by the present invention is to overcome the problem in the prior art that the detailed features and average features contained in wind power generation cannot be simultaneously represented by the clustering method to establish typical wind power curves, which poses a risk of missing information and thus affects the accuracy of clustering.

[0006] To address the aforementioned technical problems, this invention provides a method for establishing typical wind power curves based on high- and low-frequency signal combination reconstruction, characterized by the following steps:

[0007] Data cleaning stage: Input the original wind power data set, remove abnormal wind power data, and obtain cleaned wind power data;

[0008] Normalization phase: The cleaned wind power data is normalized.

[0009] Wavelet decomposition stage: Discrete wavelet transform is used to reduce the dimensionality of the normalized wind power data. Through high-pass and low-pass filters, discrete wavelet transform is used to decompose each normalized wind power data point into high-frequency and low-frequency signals. The low-frequency signals obtained from each decomposition are then subjected to discrete wavelet transform again until the low-frequency signal obtained after the I-th discrete wavelet transform decomposition meets the requirements of the fluctuation measurement index, thus obtaining X. I,H ,X I-1,H ,X I-2,H ,…,X 1,H ,X I,L , where X 1～I,H X represents the high-frequency signal after each discrete wavelet transform decomposition. I,L It is a low-frequency signal;

[0010] Combined reconstruction stage: The low-frequency signals obtained after discrete wavelet transform decomposition are clustered. The clustered low-frequency signals are combined and reconstructed with the high-frequency signals of the same layer through inverse wavelet transform to obtain the low-frequency signals of the next layer. The low-frequency signals obtained after the current combined reconstruction are iterated and clustered again for a preset number of times to output the typical wind power curves of each wind power data in multiple clusters.

[0011] Data imputation phase: Based on the typical wind power curves of each wind power data in the multiple clusters output from the combined reconstruction phase, fill in the missing wind power data.

[0012] In one embodiment of the present invention, during the data cleaning stage, an original wind power data set is input, and it is determined whether each wind power data in the original wind power data set is missing. If there is a missing data, it is further determined whether the missing degree of each wind power data is greater than a preset percentage. If it is greater than the preset missing degree, the wind power data is removed. If it is less than or equal to the preset missing degree, the wind power data is retained as the object to be filled in the data filling stage. If there is no missing data, wind power data with glitch values ​​or default values ​​in the non-missing wind power data set is removed to obtain the cleaned wind power data.

[0013] In one embodiment of the present invention, the specific formula for the normalization process in the normalization stage is as follows:

[0014]

[0015] Among them, X n = <x n,1 ,x n,2 ,…,x n,R > represents the nth wind power data sequence in the cleaned wind power data, X norm Let x represent the nth normalized wind power data sequence. n,RThis represents the Rth sampling point of the nth wind power data, where R represents the total number of sampling points in the original wind power data.

[0016] In one embodiment of the present invention, the specific formula for the continuous wavelet transform (CWT) in the wavelet decomposition stage is as follows:

[0017]

[0018] Where ψ is the mother wavelet function, scale parameter a and translation parameter b are the two characteristic coefficients of the mother wavelet function, x(t) is the normalized wind power data, and t represents the t-th time.

[0019] Discretizing the scaling and translation parameters in the continuous wavelet transform transforms the continuous wavelet transform into a discrete wavelet transform. The frequency domain signal is obtained by performing a Discrete Wavelet Transform (DWT) on the wind power data. The specific formula is as follows:

[0020]

[0021] Where j is the decomposition scale, λ is the translation coefficient, and R represents the total number of sampling points of the original wind power data.

[0022] In one embodiment of the present invention, during the wavelet decomposition stage, the coefficient of variation (CV) is used as the fluctuation measure to determine the low-frequency signal X. I,L When the coefficient of variation CV is less than or equal to a preset value, the low-frequency signal X... I,L To meet the standard, the formula for the coefficient of variation (CV) is:

[0023]

[0024] Where i represents the i-th wind power data entry, r represents the r-th sampling point in each wind power data entry, R represents the total number of sampling points in the original wind power data, and X i,r This represents the data value of the r-th sampling point of the i-th wind power data. This represents the average value of the i-th wind power data point.

[0025] In one embodiment of the present invention, the specific steps of the combined reconstruction stage include:

[0026] S1. Initialize the number of loops in the combination and reconstruction phase a = -1, and set the preset total number of loops A in the combination and reconstruction phase;

[0027] S2. Let a = a + 1;

[0028] S3. The low-frequency signal X after the a-th wavelet reconstruction. I-a,L Perform the (a+1)th clustering to obtain k clusters and m average feature categories corresponding to the k clusters, where k = m;

[0029] S4, transfer the high-frequency signal X I-a,H The low-frequency signal X after the a-th wavelet reconstruction following clustering I-a,L Perform the (a+1)th reconstruction;

[0030] S5. Determine if the current loop count a is equal to the preset total loop count A; if a is not equal to A, return to S2; if a is equal to A, output the typical wind power curve after reconstruction of each wind power data in m×(A+1) clusters.

[0031] In one embodiment of the present invention, the k-means algorithm is used in S3 to reconstruct the low-frequency signal X after the a-th wavelet reconstruction. I-a,L Clustering involves the following steps:

[0032] Step 1: Use the low-frequency signal after wind power data decomposition as the sample input, set the number of clusters k according to the initial inflection point of the curve of the sum of squared errors of the cluster evaluation index, and randomly select k samples as the initial cluster centers.

[0033] Step 2: For each sample, calculate its Euclidean distance to each cluster center and assign it to the cluster containing the nearest cluster center. l Where l = 1, 2, ..., k;

[0034] Step 3: Update the Clusters for each cluster. l Cluster centers;

[0035] Step 4: Calculate the Clust value for each cluster. l If the sum of squared errors between the middle samples and the cluster centers converges or no longer changes, the clustering result is output; otherwise, return to step 2.

[0036] In one embodiment of the present invention, in step 3, the cluster center C l The specific formula is:

[0037]

[0038] Where X I-a Given the input sample, 'a' represents the number of iterations in the combined reconstruction phase.

[0039] In one embodiment of the present invention, in S5, the specific formula for the typical wind power curve after reconstruction of each wind power data in m×(A+1) clusters is as follows:

[0040]

[0041] Where μ is a typical wind power curve after reconstruction of low-frequency and high-frequency signals, A is the preset total number of cycles in the reconstruction stage, a is the number of cycles in the reconstruction stage, and X... I-a,L X is the low-frequency signal after a combination and reconstruction. I-a,H The high-frequency signal is in the same layer as the low-frequency signal after a combination and reconstruction.

[0042] In one embodiment of the present invention, the specific formula for data filling in the data filling stage is as follows:

[0043]

[0044] Where μ r Y represents the typical curve fitting value at sampling point r. r X represents the data point of a typical wind power curve corresponding to sampling point r, where α and β are the sampling point numbers corresponding to the known data adjacent to sampling point r, and X... i,α X i,β Y represents the known data adjacent to sampling point r in the original wind power data. α Y β X in a typical wind power curve i,α X i,β The corresponding data points.

[0045] The technical solution of the present invention has the following advantages compared with the prior art:

[0046] This invention discloses a method for establishing typical wind power curves based on high- and low-frequency signal combination reconstruction. Addressing the issue that traditional clustering methods only reflect the average characteristics of data objects, this method utilizes discrete wavelet transform to convert wind power data from the time domain to the frequency domain and analyzes it. This decomposes the power generation curves into different frequency ranges, and clusters low-frequency signals to obtain classification results of the average characteristics of the wind power data. Based on this, high-frequency signals are combined to consider the short-term peak details of the wind power curves. The resulting typical wind power curves not only reflect the average characteristics of the wind power data but also its detailed features, effectively uncovering different periodic variation characteristics of the wind power data and ensuring that key information about wind power fluctuation patterns is not lost during the high- and low-frequency signal combination reconstruction process.

[0047] To address the issue of missing two or more wind power data points due to equipment malfunctions, signal interference, or other abnormalities during the raw data acquisition process, this invention employs a typical curve to fill in the missing data, thereby improving the accuracy of wind power data fitting and better supporting the implementation of wind power trading and settlement. Attached Figure Description

[0048] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings, wherein...

[0049] Figure 1 This is a flowchart of the method for establishing typical wind power curves based on high and low frequency signal combination reconstruction according to the present invention;

[0050] Figure 2 This is a schematic diagram of the method for establishing typical wind power curves based on the reconstruction of high and low frequency signals according to the present invention;

[0051] Figure 3 This is a flowchart of the data cleaning stage of the present invention;

[0052] Figure 4 This is a flowchart of the wavelet decomposition stage of the present invention;

[0053] Figure 5 This is a flowchart of the combined reconstruction stage of the present invention;

[0054] Figure 6 This is a flowchart of k-means clustering in the combinatorial reconstruction stage of this invention;

[0055] Figure 7 This is a flowchart of the high- and low-frequency signal reconstruction in the combined reconstruction stage of this invention;

[0056] Figure 8 This is a low-frequency signal diagram of wind power data at various scales using wavelet transform in an embodiment of the present invention;

[0057] Figure 9 This is a schematic diagram illustrating the use of typical wind power curves to fill in continuously missing data in an embodiment of the present invention. Detailed Implementation

[0058] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0059] Reference Figure 1 and Figure 2 As shown, this invention provides a method for establishing typical wind power curves based on high- and low-frequency signal combination reconstruction, including a data cleaning stage, a normalization stage, a wavelet decomposition stage, a combination reconstruction stage, and a data filling stage, which will be described in turn below.

[0060] In actual operation, wind turbines generate abnormal data due to measurement and wind curtailment, affecting the accuracy of generating typical wind power curves and making it impossible to effectively capture the typical characteristics of wind power generation. Therefore, cleaning abnormal wind power data is essential. In the data cleaning stage, this invention addresses three types of anomalies in the original wind power data: severe missing data, abnormal fluctuations, and default values.

[0061] Reference Figure 3 As shown, after inputting the original wind power data, abnormal data is removed from each wind power data.

[0062] First, determine if there are serious missing data issues in the wind power data. Then, determine if each piece of wind power data in the original wind power data set is missing. If it is missing, then determine if the missing percentage of each piece of wind power data is greater than a preset percentage. If it is greater than the preset missing percentage, then remove the piece of wind power data. If it is less than or equal to the preset missing percentage, then retain the piece of wind power data as the object to be filled in the data filling stage.

[0063] If there are no missing wind power data, then it is determined whether there are any abnormal fluctuations in the wind power data. If there are long-term repeated fluctuations between zero and constant values ​​in the wind power data acquisition, that is, if the current detection data has glitch values, then the wind power data is removed.

[0064] If the wind power data is free of missing values ​​and abnormal fluctuations, then it is determined whether there is a default value issue, i.e., the data collected from multiple wind power data collection points remains unchanged. If the wind power data consistently maintains an initial constant reading, then that wind power data entry is discarded, resulting in cleaned wind power data.

[0065] After the data cleaning phase, outliers were removed, improving the quality of wind power data. To avoid large discrepancies in the size of wind power data affecting subsequent clustering results, the cleaned wind power data needs to be normalized.

[0066] In the normalization stage, the specific formula for normalization is:

[0067]

[0068] Among them, X n = <x n,1 ,x n,2 ,…,x n,R > represents the nth wind power data sequence in the cleaned wind power data, X norm Let x represent the nth normalized wind power data sequence. n,R This represents the Rth sampling point of the nth wind power data, where R represents the total number of sampling points in the original wind power data.

[0069] After normalizing the wind power data, the process proceeds to the wavelet decomposition stage. This invention uses the common db4 wavelet as the wavelet transform function and employs discrete wavelet transform to reduce the dimensionality of the normalized wind power data. Through high-pass and low-pass filters, discrete wavelet transform is used to decompose each normalized wind power data point into high-frequency and low-frequency signals. The low-frequency signals obtained from each decomposition are then subjected to discrete wavelet transform again until the low-frequency signals obtained after I-th discrete wavelet transform decomposition meet the requirements of the fluctuation measurement index, thus obtaining X. I,H ,X I-1,H ,X I-2,H ,…,X 1,H ,X I,L ;where X 1～I,H X represents the high-frequency signal, i.e., the detail features, after each discrete wavelet transform decomposition. I,L It is a low-frequency signal, i.e., an average characteristic.

[0070] Reference Figure 4 As shown, LP and HP represent low-pass and high-pass filters, respectively. The low-pass filter removes the high-frequency components of the curve, while the high-pass filter removes the low-frequency components. ↓Q represents a downsampling filter, which decomposes the power generation data into different time scales. I represents the total number of wavelet decompositions. In practice, if the lowest frequency signal obtained after I wavelet decompositions has virtually no periodic fluctuations, it is determined that the average and detail features of the input wind power data can be completely separated after I wavelet transforms.

[0071] The specific formula for discrete wavelet transform is given below.

[0072] The formula for Continuous Wavelet Transform (CWT) is:

[0073]

[0074] Where ψ is the mother wavelet function, scale parameter a and translation parameter b are the two characteristic coefficients of the mother wavelet function, x(t) is the normalized wind power data, and t represents the t-th time.

[0075] Discretizing the scaling and translation parameters in the continuous wavelet transform transforms the continuous wavelet transform into a discrete wavelet transform. The frequency domain signal is obtained by performing a Discrete Wavelet Transform (DWT) on the wind power data. The specific formula is as follows:

[0076]

[0077] Where j is the decomposition scale, λ is the translation coefficient, and R represents the total number of sampling points in the original wind power data. Generally, a0 = 2 and b0 = 1 are chosen.

[0078] In the wavelet decomposition stage, this invention uses the coefficient of variation (CV) as the fluctuation measure to determine the low-frequency signal X. I,L When the coefficient of variation CV is less than or equal to a preset value, the low-frequency signal X... I,L To meet the standard, the formula for the coefficient of variation (CV) is:

[0079]

[0080] Where i represents the i-th wind power data entry, r represents the r-th sampling point in each wind power data entry, 1≤r≤R, and R represents the total number of sampling points in the original wind power data. X i,r This represents the data value of the r-th sampling point of the i-th wind power data. This represents the average value of the i-th wind power data point. This embodiment of the invention primarily targets wind power data with a time granularity of 15 minutes, meaning each data point within a day has 96 sampling points, with R = 96.

[0081] The coefficient of variation (CV) is calculated by comparing the standard deviation of the original wind power data with the mean of the original data, thus eliminating the influence of measurement scale and dimensions. When the CV is ≤ 30%, it proves that the high and low frequency signals obtained from the final decomposition can be used for subsequent signal combination and reconstruction, that is, the signal decomposition is complete.

[0082] Through the above process, the raw wind power data is decomposed into an approximate signal based on a large time scale, namely a low-frequency signal, and an approximate signal based on a small time scale, namely a high-frequency signal. Among them, the low-frequency signal has a similar overall trend to the curve of the raw wind power data, reflecting the overall change of the curve; while the high-frequency signal describes the short-term peak shape of the curve, reflecting the characteristics of the curve under different interference factors.

[0083] In the combined reconstruction stage, the low-frequency signals obtained after discrete wavelet transform decomposition are clustered. The clustered low-frequency signals and high-frequency signals of the same layer are combined and reconstructed by inverse wavelet transform to obtain the low-frequency signals of the next layer. The low-frequency signals obtained after the current combined reconstruction are iterated and clustered again for a preset number of times, and the typical wind power curves of each wind power data in multiple clusters are output.

[0084] The following is a detailed explanation.

[0085] Reference Figure 5 As shown, the specific steps of the combined reconstruction stage include:

[0086] S1. Initialize the number of loops in the combination and reconstruction phase, a = -1. Set the total number of loops to A according to the accuracy requirements of the on-site engineering, where A ≤ I-1.

[0087] S2. Let a = a + 1;

[0088] S3. The low-frequency signal X after the a-th wavelet reconstruction. I-a,L Perform the (a+1)th clustering to obtain k clusters and m average feature categories corresponding to the k clusters, where k = m;

[0089] S4, transfer the high-frequency signal X I-a,H The low-frequency signal X after the a-th wavelet reconstruction following clustering I-a,L Perform the (a+1)th reconstruction;

[0090] S5. Determine if the current loop count a is equal to the preset total loop count A; if a is not equal to A, return to S2; if a is equal to A, output the typical wind power curve after reconstruction of each wind power data in m×(A+1) clusters.

[0091] Reference Figure 6 As shown, the low-frequency signal X after the a-th wavelet reconstruction is processed using the k-means algorithm. I-a,L The steps for performing the (a+1)th clustering are as follows:

[0092] Step 1: Use the low-frequency signal after wind power data decomposition as the sample input, set the number of clusters k according to the initial inflection point of the curve of the sum of squared errors of the cluster evaluation index, and randomly select k samples as the initial cluster centers.

[0093] Step 2: For each sample, calculate its Euclidean distance to each cluster center and assign it to the cluster containing the nearest cluster center. l Where l = 1, 2, ..., k; the category Clust of the average feature is obtained. (a+1)-1 Clust (a+1)-2 Clust (a+1)-m There are m types of low-frequency signals, which characterize the contour characteristics of the wind power curve, where m = k;

[0094] Step 3: Update the Clusters for each cluster. l The cluster centers are determined by the following formula:

[0095]

[0096] Where X I-a Given the input sample, 'a' represents the number of iterations in the combined reconstruction phase.

[0097] Step 4: Calculate the Clust value for each cluster. l The sum of squared errors between the middle sample and the cluster center is given by the following formula:

[0098]

[0099] In wind power low frequency signal X I-a,L In the clustering process, the goal is to minimize the sum of squared errors (SSE) between the samples and the cluster centers. If the SSE value converges or stops changing, the clustering result is output; otherwise, return to step 2.

[0100] The wavelet reconstruction process is referenced. Figure 7 As shown, in the low-frequency signal X I-a,L After clustering is completed, X is added to the m cluster components. I-a,H Variables, combining high and low frequency signals, i.e., D I-a,L =[X I-a,L ,X I-a,H The high and low frequency signals are reconstructed into X using inverse wavelet transform. I-(a+1),L This forms the input object for the next iterative reconstruction phase. Similarly, when the loop variable a = A, the typical wind power curve reconstructed from each wind power data point in m × (A+1) clusters is output, with the formula:

[0101]

[0102] Where μ is the curve after combining and reconstructing the low-frequency signal clustering result with the high-frequency signal, A is the preset total number of iterations in the combination reconstruction stage, a is the number of iterations in the combination reconstruction stage, and X... I-a,L X is the low-frequency signal after a combination and reconstruction. I-a,H The high-frequency signal is in the same layer as the low-frequency signal after a combination and reconstruction.

[0103] In the combined reconstruction stage, the high and low frequency signals are combined and reconstructed (A+1) times, while simultaneously incorporating the low frequency signal X. I-a,L Contour features and high-frequency signal X I-a,H By analyzing detailed features and mining the characteristics of wind power curves at multiple scales and dimensions, a total of m×(A+1) types of typical wind power curves were extracted. As the number of times the high- and low-frequency signals of wind power are combined and reconstructed increases, the level of detail in the extracted typical wind power curves of each type becomes higher, and the accuracy of subsequent data filling for missing data also improves.

[0104] The various typical wind power curves extracted by this invention can not only reflect the average characteristics of the data object, but also reflect the detailed characteristics of the data object, effectively mining the different periodic variation characteristics of wind power data, and ensuring that the key information of wind power fluctuation pattern is not lost during the reconstruction of high and low frequency signals.

[0105] This invention obtains typical power generation curves for various types of wind power based on historical data. When the actual power generation of wind power is abnormal for various reasons, it can be used as a reference to fit and establish missing wind power data.

[0106] In the data imputation stage, data points missing two or more are filled by comparing them with typical wind power curves generated using a high- and low-frequency signal combination reconstruction method. Since the original wind power data and the typical curves exhibit consistent patterns of change over a certain time scale, the typical wind power generation curves can be compared with the changing patterns of the wind power curves. A proportional fit is then performed using typical wind power curves from the same period at the time of data loss. The specific formula for data imputation is as follows:

[0107]

[0108] Where μ r Y represents the typical curve fitting value at sampling point r. r X represents the data point of a typical wind power curve corresponding to sampling point r, where α and β are the sampling point numbers corresponding to the known data adjacent to sampling point r, and X... i,α X i,β Y represents the known data adjacent to sampling point r in the original wind power data. α Y β X in a typical wind power curve i,α X i,β The corresponding data points.

[0109] To analyze the data imputation effect under different missing data conditions, the root mean square relative error Rm was used. MSE The evaluation indicators are compared using the following formula:

[0110]

[0111] Where X i,r Let R be the power sample value of the i-th data point of wind power, R be the total number of sampling points, and r be the r-th sampling point.

[0112] The typical curves established by this invention are used to fill in missing data in a targeted manner, which improves the accuracy of wind power data fitting and thus better supports the development of wind power trading and settlement.

[0113] Example 1

[0114] This embodiment uses wind power data from a city in Shandong Province in March 2023. In the wavelet decomposition stage, the total number of decomposition iterations is set to I=8. To verify whether a low-frequency signal of any scale meeting engineering requirements exists during the 8-step wavelet decomposition process, the following results are obtained: Figure 8 The low-frequency signals at each scale are shown, and the CV values ​​of the low-frequency signals at each scale are statistically obtained as shown in Table 1.

[0115] Low-frequency signals at various scales CV / % 1 52.246 2 50.768 3 44.402 4 39.067 5 33.986 6 20.929 7 20.082 8 17.470

[0116] Table 1. CV values ​​of low-frequency signals at various scales

[0117] As shown in Table 1, the CV value decreases with increasing wavelet decomposition number, indicating that the periodic fluctuations of the low-frequency components of wind power also decrease. When the decomposition number is greater than 6, the reduction in CV value decreases significantly. Furthermore, when the decomposition number is 6, combined with… Figure 7 It can be intuitively seen that the low-frequency signal curve of the wind power data scale 6 is relatively flat and meets the fluctuation requirement of CV≤30% of this invention.

[0118] Therefore, this example demonstrates that setting the total decomposition order I to 6 is sufficient to meet engineering requirements. This example can be applied to other engineering applications; the total wavelet decomposition order I can be set according to the actual input data and site conditions.

[0119] During the data imputation phase, refer to Figure 9 As shown, the left figure is a typical wind power curve, and the right figure shows the data after filling in the missing data.

[0120] like Figure 9 As shown, wind power data is continuously missing at sampling points 31, 32, 33, and 34. Compared with the typical wind power curve generated using the model of this invention, the curves of the original wind power data and the typical wind power curves show consistent variation patterns over a certain time scale. Therefore, the missing data can be generated proportionally based on the ratio of the power data of the two curves using the typical wind power curves.

[0121] The curves obtained by filling in missing data using typical wind power curves are compared with actual wind power data, and R is calculated. MSE The value is 0.12478, which confirms that the filled curve closely matches the real curve, proving that the typical curve of this type obtained by reconstructing wind power data based on the high and low frequency signal combination method proposed in this invention can effectively fill in this type of missing data.

[0122] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0123] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0124] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0125] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0126] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for establishing typical wind power curves based on high- and low-frequency signal combination reconstruction, characterized in that the steps are as follows: include: Data cleaning stage: Input the original wind power data set, remove abnormal wind power data, and obtain cleaned wind power data; Normalization phase: The cleaned wind power data is normalized. Wavelet decomposition stage: Discrete wavelet transform is used to reduce the dimensionality of the normalized wind power data. Through high-pass and low-pass filters, discrete wavelet transform is used to decompose each normalized wind power data into high-frequency and low-frequency signals. Each low-frequency signal obtained from the decomposition is then subjected to a discrete wavelet transform until the low-frequency signal obtained after the I-th discrete wavelet transform decomposition meets the requirements of the fluctuation measurement index, thus obtaining X. I，H X I-1，H X I-2，H , ..., X 1，H X I，L , where X 1～I，H X represents the high-frequency signal after each discrete wavelet transform decomposition. I，L It is a low-frequency signal; Combined Reconstruction Stage: The low-frequency signals obtained after discrete wavelet transform decomposition are clustered. The clustered low-frequency signals are then combined and reconstructed with the high-frequency signals of the same layer using inverse wavelet transform to obtain the low-frequency signals of the next layer. The low-frequency signals obtained after the current combined reconstruction are then iteratively clustered and reconstructed a preset number of times, outputting typical wind power curves for each wind power data point in multiple clusters. Data imputation phase: Based on the typical wind power curves of each wind power data in the multiple clusters output from the combined reconstruction phase, fill in the missing wind power data.

2. The method for establishing typical wind power curves based on high- and low-frequency signal combination reconstruction according to claim 1, characterized in that, During the data cleaning phase, the original wind power data set is input, and it is determined whether there are any missing data points for each wind power data set in the original wind power data set; If missing data exists, the system continues to determine whether the missing percentage of each wind power data is greater than a preset percentage. If it is greater than the preset percentage, the wind power data is removed. If it is less than or equal to the preset percentage, the wind power data is retained as the data to be filled in the data filling stage. If there is no missing data, the wind power data with spurious or missing values ​​in the non-missing wind power data set is removed to obtain the cleaned wind power data.

3. The method for establishing typical wind power curves based on high- and low-frequency signal combination reconstruction according to claim 1, characterized in that, In the normalization stage, the specific formula for the normalization process is as follows: Among them, X n = <x n，1 x n，2 , ..., x n，R > represents the nth wind power data sequence in the cleaned wind power data, X norm Let x represent the nth normalized wind power data sequence. n，R This represents the Rth sampling point of the nth wind power data, where R represents the total number of sampling points in the original wind power data.

4. The method for establishing typical wind power curves based on high- and low-frequency signal combination reconstruction according to claim 1, characterized in that, In the wavelet decomposition stage, the specific formula for the continuous wavelet transform (CWT) is as follows: Where ψ is the mother wavelet function, scale parameter a and translation parameter b are the two characteristic coefficients of the mother wavelet function, x(t) is the normalized wind power data, and t represents the t-th time. The scaling and translation parameters in the continuous wavelet transform are discretized, transforming the continuous wavelet transform into a discrete wavelet transform. This Discrete Wavelet Transform (DWT) is then applied to the wind power data to obtain the frequency domain signal. The specific formula is as follows: Where j is the decomposition scale, λ is the translation coefficient, and R represents the total number of sampling points of the original wind power data.

5. The method for establishing typical wind power curves based on high- and low-frequency signal combination reconstruction according to claim 1, characterized in that, In the wavelet decomposition stage, the coefficient of variation (CV) is used as the fluctuation measure to determine the low-frequency signal X. I，L When the coefficient of variation CV is less than or equal to a preset value, the low-frequency signal X... I，L To meet the standard, the formula for the coefficient of variation (CV) is: Where i represents the i-th wind power data entry, r represents the r-th sampling point in each wind power data entry, R represents the total number of sampling points in the original wind power data, and X i，r This represents the data value of the r-th sampling point of the i-th wind power data. This represents the average value of the i-th wind power data point.

6. The method for establishing typical wind power curves based on high- and low-frequency signal combination reconstruction according to claim 1, characterized in that, The specific steps of the combined reconstruction stage include: S1. Initialize the number of loops in the combination and reconstruction phase a = -1, and set the preset total number of loops A in the combination and reconstruction phase; S2. Let a = a + 1; S3. The low-frequency signal X after the a-th wavelet reconstruction. I-a，L Perform the (a+1)th clustering to obtain k clusters and m average feature categories corresponding to the k clusters, where k = m; S4, transfer the high-frequency signal X I-a，H The low-frequency signal X after the a-th wavelet reconstruction following clustering I-a，L Perform the (a+1)th reconstruction; S5. Determine if the current loop count a is equal to the preset total loop count A; if a is not equal to A, return to S2; if a is equal to A, output the typical wind power curve after reconstruction of each wind power data in m×(A+1) clusters.

7. The method for establishing typical wind power curves based on high- and low-frequency signal combination reconstruction according to claim 6, characterized in that, In S3, the k-means algorithm is used to reconstruct the low-frequency signal X after the a-th wavelet reconstruction. I-a，L Clustering involves the following steps: Step 1: Use the low-frequency signal after wind power data decomposition as the sample input, set the number of clusters k according to the initial inflection point of the curve of the sum of squared errors of the cluster evaluation index, and randomly select k samples as the initial cluster centers. Step 2: For each sample, calculate its Euclidean distance to each cluster center and assign it to the cluster containing the nearest cluster center. l Where l = 1, 2, ..., k; Step 3: Update the Clusters for each cluster. l Cluster centers; Step 4: Calculate the Clust value for each cluster. l If the sum of squared errors between the middle samples and the cluster centers converges or no longer changes, the clustering result is output; otherwise, return to step 2.

8. The method for establishing typical wind power curves based on high- and low-frequency signal combination reconstruction according to claim 7, characterized in that, In step 3, cluster center C l The specific formula is: Where X I-a Given the input sample, 'a' represents the number of iterations in the combined reconstruction phase.

9. The method for establishing typical wind power curves based on high- and low-frequency signal combination reconstruction according to claim 6, characterized in that, In S5, the specific formula for the typical wind power curve after reconstruction of each wind power data point in m×(A+1) clusters is as follows: Where μ is a typical wind power curve after reconstruction of low-frequency and high-frequency signals, A is the preset total number of cycles in the reconstruction stage, a is the number of cycles in the reconstruction stage, and X... I-a，L X is the low-frequency signal after a combination and reconstruction. I-a，H The high-frequency signal is in the same layer as the low-frequency signal after a combination and reconstruction.

10. The method for establishing typical wind power curves based on high- and low-frequency signal combination reconstruction according to claim 1, characterized in that, During the data imputation stage, the specific formula for data imputation is: Where μ r Y represents the typical curve fitting value at sampling point r. r X represents the data point of a typical wind power curve corresponding to sampling point r, where α and β are the sampling point numbers corresponding to the known data adjacent to sampling point r. i，α X i，β Y represents the known data adjacent to sampling point r in the original wind power data. α Y β X in a typical wind power curve i，α X i，β The corresponding data points.

Images