Wind power abnormal data identification method based on adaptive K-means segmentation and boundary modeling

Abstract

The invention provides a wind power abnormal data identification method based on self-adaptive K-means segmentation and boundary modeling. The method comprises the steps of identifying and filtering bottom accumulation type abnormal data in a wind power data set in a rule screening mode according to the characteristic that the power value of the bottom accumulation type abnormal data is smaller than 0; adaptively dividing the wind power data set subjected to rule screening into K non-overlapping clusters by using a CH index and a k-means algorithm; dividing each cluster obtained in the step S3 into L intervals according to power values, and calculating wind speed range data and wind speed mean value data corresponding to each power interval; and according to the wind speed range data and the wind speed mean value data corresponding to each power interval, using peak correction in combination with boundary modeling to identify discrete type abnormal data and limited power type abnormal data. According to the invention, the identification accuracy of the wind power abnormal data is improved.

Description

Technical Field

[0001] The invention belongs to the technical field of wind power, and in particular relates to a method for identifying abnormal wind power data based on adaptive K-means segmentation and boundary modeling. Background Art

[0002] At present, the methods for processing abnormal wind power data are mainly divided into methods based on machine learning, methods based on statistical analysis, and methods based on image analysis.

[0003] The basic idea of ​​the abnormal wind power data processing method based on machine learning is to learn the normal pattern of wind power data through training models, and then use the training models to identify and process abnormal data. Existing machine learning methods attempt to use the nearest neighbor algorithm (KNN) to construct wind speed power curves for online detection of abnormal data; some scholars use the K-means algorithm to identify abnormal data, and solve the problem of selecting the K value in the K-means algorithm through the secondary clustering method; in addition, some scholars also use the similarity relationship between data to use support vector machines to establish multiple unit neighbor models to identify abnormal data; the common disadvantage of machine learning related algorithms is that when there are a large number of power-limited abnormal data in the data to be identified, the identification will fail. At the same time, the machine learning-based methods often contain parameters that need to be manually adjusted according to different data sets. Different parameter settings often bring completely different results, and it is difficult to obtain a suitable parameter selection method.

[0004] The basic idea of ​​the wind power abnormal data processing method based on statistical analysis is to detect and process anomalies through statistical analysis, and use statistical methods such as probability and hypothesis testing to determine whether the data is abnormal based on the statistical properties and distribution characteristics of the data. At present, the methods based on statistical analysis mainly include the 3-σ criterion, the quartile method, and the Copula function. Although this type of method takes into account power-limited abnormal data, it often requires data partitioning, resulting in a jagged recognition result, and a large amount of normal data will be mistakenly identified as abnormal data. Summary of the invention

[0005] In order to solve the problems existing in the background technology, different methods are used to identify and clean the unique characteristics of different types of data to improve the recognition accuracy of abnormal wind power data. To this end, the present invention provides a method for identifying abnormal wind power data based on adaptive K-means segmentation and boundary modeling, including:

[0006] S1: According to the distribution characteristics of different wind power abnormal data, the wind power abnormal data is divided into bottom accumulation type abnormal data, discrete type abnormal data and power limit type abnormal data; among which, the bottom accumulation type abnormal data is the wind power abnormal data with a power value less than 0;

[0007] S2: Identify and filter out the bottom accumulation type abnormal data in the wind power data set by means of rule screening based on the feature that the power value of the bottom accumulation type abnormal data is less than 0;

[0008] S3: The wind power dataset after rule screening is adaptively divided into K non-overlapping clusters using the CH index and k-means algorithm;

[0009] S4: Divide each cluster obtained in step S3 into L intervals according to the power value, and calculate the wind speed range data and wind speed mean data corresponding to each power interval;

[0010] S5: According to the wind speed range data and wind speed mean data corresponding to each power interval, peak correction combined with boundary modeling is used to identify discrete abnormal data and power-limited abnormal data.

[0011] Preferably, step S3 comprises:

[0012] S31: For a given K value, the K-means algorithm is used to cluster the wind power data set and the clustered CH index is calculated;

[0013] S32: By trying different K values ​​and comparing their corresponding CH indexes, the K value with the highest CH index is selected as the optimal K value, and K clusters corresponding to the optimal K value are obtained.

[0014] Preferably, the CH index comprises:

[0015]

[0016]

[0017]

[0018] Among them, tr() means to find the trace of the matrix; B k represents the covariance matrix of inter-cluster data; W k represents the covariance matrix of the data within the cluster; n represents the total number of data in all clusters, K represents the number of clusters; n q represents the total number of data in the qth cluster; r q represents the center point of the qth cluster, R q represents the qth cluster; r e represents the center point of all clusters; x∈R q represents the data point in the qth cluster.

[0019] Preferably, the calculating of the wind speed range data and the wind speed mean data corresponding to each power interval includes:

[0020] The wind speed range of wind power data corresponding to each power interval is calculated using the quartile method to obtain the wind speed range data corresponding to each power interval (p c , w r ), the calculation formula is as follows:

[0021] w r =Q U -Q L +3IQR=4IQR

[0022] IQR=Q U -Q L

[0023] Among them, w r Indicates the wind speed range of wind power data corresponding to the power interval; Q U Indicates the wind speed value of the wind power data at the first 25% of the power interval; Q L Indicates the wind speed value of the wind power data at the last 25% of the power interval; p c Indicates the center point of the power interval;

[0024] The wind speed mean data corresponding to each power interval is (p c , w m ), w m Indicates the mean wind speed of the wind power data corresponding to each power interval.

[0025] Preferably, the step S5 comprises:

[0026] S51: constructing the wind speed range data and the wind speed mean data corresponding to each power interval into a wind speed range data point set and a wind speed mean data point set respectively;

[0027] S52: extracting peak data of the wind speed range data point set and peak data of the wind speed mean data point set respectively for the wind speed range data point set and the wind speed mean data point set; wherein the peak data includes: the convexity of the peak, the starting point of the peak and the end point of the peak;

[0028] S53: traversing the peak data of the wind speed range data point set and the peak data of the wind speed mean data point set, when the protrusion degree of the peak value is greater than the set threshold, the data points from the start point to the end point of the peak value are corrected by using a linear interpolation method to obtain a corrected wind speed range data point set and a corrected wind speed mean data point set;

[0029] S54: fitting the corrected wind speed range data point set using an exponential function to obtain a wind speed range curve; fitting the corrected wind speed mean data point set using a cubic function to obtain a wind speed mean curve;

[0030] S55: setting a left boundary threshold and a right boundary threshold according to the wind speed range curve and the wind speed mean curve, and identifying discrete abnormal data and power-limited abnormal data in the wind power data set based on the set left boundary threshold and right boundary threshold.

[0031] Preferably, the extracting of peak data of the wind speed range data point set and peak data of the wind speed mean data point set comprises:

[0032] S521: Traverse each data point y in the data set z =(p z , w z ), z represents the zth data point, p z Represents data point y z The corresponding power, w z Represents data point y z The corresponding wind speed range or wind speed mean, comparing the data point p z The size of its adjacent data points, if w z >w z+1 And w z >w z-1 , then the data point y z Add peak point set;

[0033] S522: Traverse each peak point y in the peak point set p , from the peak point y p Starting from the peak point y in the data set p The left and right sides are expanded synchronously to calculate the peak point y p The shortest distance to the straight line L, where the straight line L is along the peak point y p Point P obtained by expanding from the left s and along the peak point y p Point P obtained by right-side expansion e connected to form;

[0034] S523: When the peak point y p The shortest distance to the straight line L is greater than the peak point y p The corresponding peak convexity p m , then point P s and point P e As the peak point y p The starting point and end point of the corresponding peak; wherein the peak point y p The corresponding peak convexity p m include:

[0035] p m =w p -min(w s , w e )

[0036] Among them, w p Indicates the peak point y p The corresponding wind speed range or wind speed mean; w s Represents data point P s The corresponding wind speed range or wind speed mean; P e Represents data point P e The corresponding wind speed range or wind speed mean.

[0037] Preferably, the exponential function comprises:

[0038]

[0039] Where x represents power, w(x) represents wind speed range, and a2, b2, and c2 represent parameters.

[0040] Preferably, the cubic function comprises:

[0041] v(x)=a1x 3 +b1x 2 +c1x+d1

[0042] Where x represents power, v(x) represents wind speed range, and a1, b1, c1, and d1 represent parameters.

[0043] Preferably, the left boundary threshold is v(x)-0.5w(x), and the right boundary threshold is v(x)+0.5w(x).

[0044] The present invention has at least the following beneficial effects

[0045] The present invention clusters wind power data by using CH index combined with k-means clustering algorithm, adaptively determines the number of divided regions, and divides the wind speed-power scatter plot into different categories to improve the similarity of data within the same category and reduce the similarity between different categories, thereby realizing accurate recognition of abnormal data of wind power monitoring data in the full power range. In particular, by adopting peak correction combined with boundary modeling, the disadvantage that traditional methods are difficult to identify power-limited abnormal data is realized, and the dependence of traditional clustering algorithms on the number of clusters is overcome; by dividing the abnormal data of wind power monitoring into regions and obtaining wind speed range data and wind speed mean data, the boundary modeling problem is simplified to a curve fitting problem of two types of data, thereby solving the problem that the recognition effect of traditional recognition methods is difficult to be globally optimal, improving the accuracy of wind power abnormal data recognition, and providing new ideas for researchers who will conduct related research in the future. The research method has important theoretical research value and practical engineering significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0047] Figure 2 Schematic diagram of characteristic distribution of abnormal data of different types of wind power;

[0048] Figure 3 A schematic diagram of the division results using the CH index and k-means algorithm;

[0049] Figure 4 This is a schematic diagram of the principle of the quartile method;

[0050] Figure 5 It is a schematic diagram of wind speed range data and wind speed mean data;

[0051] Figure 6 Schematic diagram of boundary modeling results;

[0052] Figure 7 Schematic diagram of regional CU recognition results;

[0053] Figure 8 This is a schematic diagram comparing the abnormal data identification results of different methods. DETAILED DESCRIPTION

[0054] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0055] In this embodiment, through the investigation and analysis of the data distribution of the existing wind power data set, it is found that the power variation range of the wind turbine is relatively wide when it is running, and the wind power data in different power ranges has its own unique characteristics. The unified processing of the entire power range is often ineffective, and the data set needs to be divided into regions. The traditional division method usually directly specifies the number of categories, ignoring the differences between different data sets. A new division method is needed that can adaptively select the number of categories and improve the correlation of the same type of data, and reduce the correlation between different categories of data. Through the investigation and analysis of the existing abnormal data identification methods, it is found that abnormal wind power data can be mainly divided into three categories: bottom accumulation type abnormal data, discrete type abnormal data, and power-limited type abnormal data. The traditional abnormal data identification method uniformly identifies the actual collected original power data, ignoring that different types of abnormal data have their own unique characteristics, and it is difficult to accurately identify abnormal data. A new joint processing framework is needed to use different methods to identify and clean different types of data. For this purpose, please refer to Figure 1 The present invention provides a method for identifying abnormal wind power data based on adaptive K-means segmentation and boundary modeling, comprising:

[0056] S1: According to the distribution characteristics of different wind power abnormal data, the wind power abnormal data is divided into bottom accumulation type abnormal data, discrete type abnormal data and power limit type abnormal data; among which, the bottom accumulation type abnormal data is the wind power abnormal data with a power value less than 0;

[0057] S2: Identify and filter out the bottom accumulation type abnormal data in the wind power data set by means of rule screening based on the feature that the power value of the bottom accumulation type abnormal data is less than 0;

[0058] S3: The wind power dataset after rule screening is adaptively divided into K non-overlapping clusters using the CH index and k-means algorithm;

[0059] S4: Divide each cluster obtained in step S3 into L intervals according to the power value, and calculate the wind speed range data and wind speed mean data corresponding to each power interval;

[0060] In this embodiment, each cluster is divided into L intervals according to the power value, including: cluster CK i Divided into ten small areas along the power axis {CK (i,j) |j=0,1,2,…,9}, each power interval is equal in size.

[0061] S5: According to the wind speed range data and wind speed mean data corresponding to each power interval, peak correction combined with boundary modeling is used to identify discrete abnormal data and power-limited abnormal data.

[0062] In this embodiment, the wind power abnormal data are divided into three categories according to the distribution characteristics of different wind power abnormal data: bottom accumulation type abnormal data, discrete type abnormal data and power limit type abnormal data: bottom accumulation type abnormal data: located at the bottom of the data, characterized by high density, band distribution and power value near zero, mainly caused by shutdown for maintenance, wind power curtailment and self-consumption of internal measurement and control system when the fan blades are not running; discrete type abnormal data: characterized by random distribution around the standard power curve, the farther away from the curve, the less, mainly caused by measurement error, signal noise, extreme weather, etc.; that is, power limit type abnormal data, located in the middle of the data, characterized by a horizontal dense power band, mainly caused by power limit such as Figure 2 The bottom accumulation type abnormal data is identified and cleaned by rule screening based on the characteristic that its power value is less than zero.

[0063] Preferably, step S3 comprises:

[0064] S31: For a given K value, the K-means algorithm is used to cluster the wind power data set and the clustered CH index is calculated;

[0065] S32: By trying different K values ​​and comparing their corresponding CH indexes, the K value with the highest CH index is selected as the optimal K value, and K clusters corresponding to the optimal K value are obtained.

[0066] Preferably, the CH index comprises:

[0067]

[0068]

[0069]

[0070] Among them, tr() means to find the trace of the matrix; B k represents the covariance matrix of inter-cluster data; W k represents the covariance matrix of the data within the cluster; n represents the total number of data in all clusters, K represents the number of clusters; n q represents the total number of data in the qth cluster; r q represents the center point of the qth cluster, R q represents the qth cluster; r e Represents the data center point of all clusters; x∈Rq represents the data point in the qth cluster.

[0071] In this embodiment, the Calinski-Harabasz (CH) index and k-means algorithm are used to adaptively divide the wind power data set after rule screening into K non-overlapping intervals, so that the similarity between data points in the same cluster is maximized, while the similarity between different clusters is minimized. Each classification interval can be adaptively determined, and the interval containing power-limited abnormal data is separated from the normal data interval, which is convenient for subsequent processing. The CH index is used to adaptively determine the number of divisions K. The basic idea of ​​the CH index is to measure the similarity within a class and the similarity between classes, so as to increase the similarity within a class and reduce the similarity between classes. The wind power data set is divided into K intervals, denoted as {CK i |i=1,2,...,K}, the partition result is as follows Figure 3 As shown;

[0072] The present invention provides a preferred real-time method, before dividing the wind power data set into K intervals, performing the following operations:

[0073] Traverse each data point in the wind power data set to determine whether its wind speed value is greater than 1.2 times the rated wind speed value. If so, remove the data point from the wind power data set and add the CU area. After the traversal is completed, the preprocessed wind power data set and the CU area data set are obtained; the preprocessed wind power data set is used to perform step S3 and subsequent abnormal data identification. For the data point d to be processed in the CU area data set, if its power value p does not meet Q cuL -1.5IQR cuL ≤p≤Q cuU +1.5IQR cuL , IQR cuL =Q cuU -Q cuL , then d is considered to be discrete abnormal data, where the data points in the CU area data set are arranged in ascending order of power value; Q cuL Indicates the value of the first 25% position data in the CU area dataset, Q cuU Represents the value of the last 25% of the position data in the CU area data set, and its recognition result diagram Figure 7 shown.

[0074] Preferably, the calculating of the wind speed range data and the wind speed mean data corresponding to each power interval includes:

[0075] The wind speed range of wind power data corresponding to each power interval is calculated using the quartile method to obtain the wind speed range data corresponding to each power interval (p c , w r ), the calculation formula is as follows:

[0076] w r =Q U -Q L +3IQR=4IQR

[0077] IQR=Q U -Q L

[0078] Among them, w r Indicates the wind speed range of wind power data corresponding to the power interval; Q U Indicates the wind speed value of the wind power data at the first 25% of the power interval; Q L Indicates the wind speed value of the wind power data at the last 25% of the power interval; p c Indicates the center point of the power interval;

[0079] The wind speed mean data corresponding to each power interval is (p c ,w m ), w m Indicates the mean wind speed of the wind power data corresponding to each power interval.

[0080] In this embodiment, the principle of the quartile method is as follows Figure 4 As shown, a set of data is arranged in ascending order and divided into four equal parts, Q L is the lower quartile, indicating the value of the data in the first 25% of the positions, Q U is the upper quartile, which indicates the value of the data in the last 25%. The interquartile range (IQR) reflects the concentration of the data. This paper uses the quartile method to calculate the CK (i,j) The discrete data recognition threshold Q L -1.5IQR and Q U The difference of +1.5IQR is used as the wind speed range parameter corresponding to the power interval, and the corresponding power parameter is the mean value of the power interval, that is, the center point.

[0081] Preferably, the step S5 comprises:

[0082] S51: constructing the wind speed range data and the wind speed mean data corresponding to each power interval into a wind speed range data point set and a wind speed mean data point set respectively;

[0083] In this embodiment, due to the distribution characteristics of the power-limited abnormal data, the wind speed range data point set and the wind speed mean data point set will have unsmooth protrusion data, and the power value of the data is the same as the power interval of the power-limited abnormal data in the original wind power scatter diagram. Therefore, it is necessary to correct the protrusion data in the wind speed range data point set and the wind speed mean data point set, such as Figure 5 shown.

[0084] S52: extracting peak data of the wind speed range data point set and peak data of the wind speed mean data point set respectively for the wind speed range data point set and the wind speed mean data point set; wherein the peak data includes: the convexity of the peak, the starting point of the peak and the end point of the peak;

[0085] S53: traversing the peak data of the wind speed range data point set and the peak data of the wind speed mean data point set, when the protrusion degree of the peak value is greater than the set threshold, the data points from the start point to the end point of the peak value are corrected by using a linear interpolation method to obtain a corrected wind speed range data point set and a corrected wind speed mean data point set;

[0086] S54: fitting the corrected wind speed range data point set using an exponential function to obtain a wind speed range curve; fitting the corrected wind speed mean data point set using a cubic function to obtain a wind speed mean curve;

[0087] S55: setting a left boundary threshold and a right boundary threshold according to the wind speed range curve and the wind speed mean curve, and identifying discrete abnormal data and power-limited abnormal data in the wind power data set based on the set left boundary threshold and right boundary threshold.

[0088] Preferably, the extracting of peak data of the wind speed range data point set and peak data of the wind speed mean data point set comprises:

[0089] S521: Traverse each data point y in the data set z =(p z , w z ), z represents the zth data point, p z Represents data point y z The corresponding power, w z Represents data point y z The corresponding wind speed range or wind speed mean, comparing the data point p z The size of its adjacent data points, if w z >w z+1 And w z >w z-1 , then the data point y z Add peak point set;

[0090] S522: Traverse each peak point y in the peak point set p , from the peak point y p Starting from the peak point y in the data set p The left and right sides are expanded synchronously to calculate the peak point y p The shortest distance to the straight line L, where the straight line L is along the peak point y p Point P obtained by expanding from the left s and along the peak point y p Point P obtained by right-side expansion e connected to form;

[0091] In this embodiment, the peak point y in the data set is p The left and right sides are expanded synchronously including: Assuming that along the peak point y p The corresponding power coordinate is f, then the point P obtained after the first left expansion is s The power coordinate is f-1, and the point P obtained after the first right expansion is e The power coordinate is f+1. Similarly, the point P obtained after the tth left expansion s The power coordinate is ft, and the point P obtained after the tth side expansion is e The power coordinate of is f+t.

[0092] S523: When the peak point y p The shortest distance to the straight line L is greater than the peak point y p The corresponding peak convexity p m , then point P s and point P e As the peak point y pThe starting point and end point of the corresponding peak; wherein the peak point y p The corresponding peak convexity p m include:

[0093] p m =w p -min(w s , w e )

[0094] Among them, w p Indicates the peak point y p The corresponding wind speed range or wind speed mean; w s Represents data point P s The corresponding wind speed range or wind speed mean; P e Represents data point P e The corresponding wind speed range or wind speed mean.

[0095] In step S5 of this embodiment, a wind turbine power limit type abnormal data recognition framework is proposed, which solves the problem that the wind turbine power limit type abnormal data is difficult to recognize.

[0096] Preferably, the exponential function comprises:

[0097]

[0098] Where x represents power, w(x) represents wind speed range, and a2, b2, and c2 represent parameters.

[0099] After analysis in this embodiment, due to the wind speed distribution characteristics and the unit operating characteristics, the wind speed width parameter defined in this article will mutate in the rated power area. Therefore, the wind speed range data is theoretically approximate to an exponential curve. Therefore, an exponential function is selected to fit the corrected wind speed range data point set.

[0100] Preferably, the cubic function comprises:

[0101] v(x)=a1x 3 +b1x 2 +c1x+d1

[0102] Where x represents power, v(x) represents wind speed range, and a1, b1, c1, and d1 represent parameters.

[0103] In this embodiment, after analysis, due to the wind speed distribution characteristics and the unit working characteristics, the wind speed mean data is theoretically approximate to a cubic function, so a cubic function is selected to fit the corrected wind speed mean data point set.

[0104] Preferably, the left boundary threshold is v(x)-0.5w(x), and the right boundary threshold is v(x)+0.5w(x).

[0105] Finally, the remaining data points in the wind power data set are identified based on the obtained left boundary threshold v(x)-0.5w(x) and the right boundary threshold v(x)+0.5w(x). That is, the power value of the data point is substituted into the function w(x) and v(x), and it is determined whether the wind speed value of the data point is greater than or equal to v(x)-0.5w(x) and less than or equal to v(x)+0.5w(x). If so, the data point is normal data, otherwise it is abnormal data. Discrete abnormal data and power-limited abnormal data are identified, such as Figure 6 and Figure 8 shown.

[0106] In order to verify the effectiveness and versatility of the method in this paper, the actual operation data of the wind turbines in a domestic wind farm from March 2014 to March 2015 were selected to verify and analyze the identification method in this paper. The wind turbines in this wind farm record data once a minute. The basic parameters of the wind turbines are as follows: cut-in wind speed is 1m / s, rated wind speed is 12m / s, cut-out wind speed is 25m / s, and rated power is 1500kW. This paper uses wind turbine No. 1 as an example. First, rule screening is used to identify bottom accumulation type abnormal data, and then the CH index combined with the k-means clustering algorithm is used to divide the wind power data in the SCADA system of unit No. 1 into regions (see for details). Figure 2 ), the wind speed range data and wind speed mean data were obtained using the quartile method (see Figure 3 ), discrete abnormal data and power-limited abnormal data are identified through peak correction combined with boundary modeling.

[0107] In order to further verify the effectiveness of the wind power abnormal data recognition method based on adaptive K-means segmentation and boundary modeling proposed in this paper in identifying abnormal data of wind turbine power, the recognition results of this method are compared with the abnormal data recognition results of the DBSCAN algorithm based on density clustering. The neighborhood radius Eps and the minimum number of included points Minpts are its main parameters. For different data sets, the Eps and Minpts that should be selected are quite different. For the data in this paper, Eps = 0.01 and Minpts = 20 are selected this time. For details, see the comparison Figure 8 ,According to the simulation diagram, it can be seen that the ,comparison algorithm has poor recognition effect on the ,limited power type abnormal data, while the proposed method is ,not affected by the density of the limited power type abnormal ,data and can accurately identify the limited power type abnormal ,data.

[0108] In summary, the present invention utilizes the CH index combined with the k-means clustering algorithm to cluster wind power data, adaptively determines the number of divided regions, and divides the wind speed-power scatter plot into different categories to improve the similarity of data within the same category and reduce the similarity between different categories, thereby realizing accurate identification of abnormal data of wind power monitoring data in the full power range. In particular, by adopting peak correction combined with boundary modeling, the disadvantage that traditional methods are difficult to identify power-limited abnormal data is realized, and the dependence of traditional clustering algorithms on the number of clusters is overcome; by dividing the abnormal wind power monitoring data into regions and obtaining wind speed range data and wind speed mean data, the boundary modeling problem is simplified to a curve fitting problem of two types of data, thereby solving the problem that the recognition effect of traditional recognition methods is difficult to be globally optimal, improving the accuracy of wind power abnormal data recognition, and providing new ideas for researchers who will conduct related research in the future. The research method has important theoretical research value and practical engineering significance.

[0109] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A method for identifying abnormal wind power data based on adaptive K-means segmentation and boundary modeling, characterized in that: include: S1: According to the distribution characteristics of different wind power abnormal data, the wind power abnormal data is divided into bottom accumulation type abnormal data, discrete type abnormal data and power limit type abnormal data; among which, the bottom accumulation type abnormal data is the wind power abnormal data with a power value less than 0; S2: Identify and filter out the bottom accumulation type abnormal data in the wind power data set by means of rule screening based on the feature that the power value of the bottom accumulation type abnormal data is less than 0; S3: The wind power dataset after rule screening is adaptively divided into K non-overlapping clusters using the CH index and k-means algorithm; S4: Divide each cluster obtained in step S3 into L intervals according to the power value, and calculate the wind speed range data and wind speed mean data corresponding to each power interval; S5: According to the wind speed range data and wind speed mean data corresponding to each power interval, peak correction combined with boundary modeling is used to identify discrete abnormal data and power-limited abnormal data.

2. The method for identifying abnormal wind power data based on adaptive K-means segmentation and boundary modeling according to claim 1 is characterized in that: The step S3 comprises: S31: For a given K value, the K-means algorithm is used to cluster the wind power data set and the clustered CH index is calculated; S32: By trying different K values ​​and comparing their corresponding CH indexes, the K value with the highest CH index is selected as the optimal K value, and K clusters corresponding to the optimal K value are obtained.

3. The method for identifying abnormal wind power data based on adaptive K-means segmentation and boundary modeling according to claim 2 is characterized in that: The CH index includes: Among them, tr() means to find the trace of the matrix; B k represents the covariance matrix of inter-cluster data; W k represents the covariance matrix of the data within the cluster; n represents the total number of data in all clusters, K represents the number of clusters; n q represents the total number of data in the qth cluster; r q represents the center point of the qth cluster, R q represents the qth cluster; r e represents the center point of all clusters; x∈R q represents the data point in the qth cluster.

4. The method for identifying abnormal wind power data based on adaptive K-means segmentation and boundary modeling according to claim 2 is characterized in that: The calculation of the wind speed range data and the wind speed mean data corresponding to each power interval includes: The wind speed range of wind power data corresponding to each power interval is calculated using the quartile method to obtain the wind speed range data corresponding to each power interval (p c ,w r ), the calculation formula is as follows: <h2 style=";text-align:left;direction:ltr">w<h2 style=";text-align:left;direction:ltr"> r <h2 style=";text-align:left;direction:ltr"> =Q<h2 style=";text-align:left;direction:ltr"> U <h2 style=";text-align:left;direction:ltr"> -Q<h2 style=";text-align:left;direction:ltr"> L <h2 style=";text-align:left;direction:ltr"> +3IQR=4IQR IQR / Q U -Q L Among them, w r Indicates the wind speed range of wind power data corresponding to the power interval; Q U Indicates the wind speed value of the wind power data at the first 25% of the power interval; Q L Indicates the wind speed value of the wind power data at the last 25% of the power interval; p c Indicates the center point of the power interval; The wind speed mean data corresponding to each power interval is (p c ,w m ), w m Indicates the mean wind speed of the wind power data corresponding to each power interval.

5. The method for identifying abnormal wind power data based on adaptive K-means segmentation and boundary modeling according to claim 1 is characterized in that: The step S5 comprises: S51: constructing the wind speed range data and the wind speed mean data corresponding to each power interval into a wind speed range data point set and a wind speed mean data point set respectively; S52: extracting peak data of the wind speed range data point set and peak data of the wind speed mean data point set respectively for the wind speed range data point set and the wind speed mean data point set; wherein the peak data includes: the convexity of the peak, the starting point of the peak and the end point of the peak; S53: traversing the peak data of the wind speed range data point set and the peak data of the wind speed mean data point set, when the protrusion degree of the peak value is greater than the set threshold, the data points from the start point to the end point of the peak value are corrected by using a linear interpolation method to obtain a corrected wind speed range data point set and a corrected wind speed mean data point set; S54: fitting the corrected wind speed range data point set using an exponential function to obtain a wind speed range curve; fitting the corrected wind speed mean data point set using a cubic function to obtain a wind speed mean curve; S55: setting a left boundary threshold and a right boundary threshold according to the wind speed range curve and the wind speed mean curve, and identifying discrete abnormal data and power-limited abnormal data in the wind power data set based on the set left boundary threshold and right boundary threshold.

6. The method for identifying abnormal wind power data based on adaptive K-means segmentation and boundary modeling according to claim 5, characterized in that: The step of extracting peak data of a wind speed range data point set and peak data of a wind speed mean data point set comprises: S521: Traverse each data point y in the data set z =(p z , w z ), z represents the zth data point, p z Represents data point y z The corresponding power, w z Represents data point y z The corresponding wind speed range or wind speed mean, comparing the data point p z The size of its adjacent data points, if w z >w z+1 And w z >w z-1 , then the data point y z Add peak point set; S522: Traverse each peak point y in the peak point set p , from the peak point y p Starting from the peak point y in the data set p The left and right sides are expanded synchronously to calculate the peak point y p The shortest distance to the straight line L, where the straight line L is along the peak point y p Point P obtained by expanding from the left s and along the peak point y p Point P obtained by right-side expansion e connected to form; S523: When the peak point y p The shortest distance to the line L is greater than the peak point y p The corresponding peak convexity p m , then point P s and point P e As the peak point y p The starting point and end point of the corresponding peak; wherein the peak point y p The convexity P of the corresponding peak m include: p m =in p -min(in s ,In e ) Among them, w p Indicates the peak point y p The corresponding wind speed range or wind speed mean; ω s Represents data point P s The corresponding wind speed range or wind speed mean; P e Represents data point P e The corresponding wind speed range or wind speed mean.

7. The method for identifying abnormal wind power data based on adaptive K-means segmentation and boundary modeling according to claim 5, characterized in that: The exponential function includes: Where x represents power, w(x) represents wind speed range, and a2, b2, and c2 represent parameters.

8. The method for identifying abnormal wind power data based on adaptive K-means segmentation and boundary modeling according to claim 7, characterized in that: The cubic function includes: v(x)=a1x 3 +b1x 2 +c1x+d1 Where x represents power, v(x) represents wind speed range, and a1, b1, c1, and d1 represent parameters.

9. The method for identifying abnormal wind power data based on adaptive K-means segmentation and boundary modeling according to claim 8, characterized in that: The left boundary threshold is v(x)-0.5w(x), and the right boundary threshold is v(x)+0.5w(x).

Images