Method and system for cleaning abnormal data of wind speed-power curve

Abstract

Provided in the present invention are a method and system for cleaning abnormal data of a wind speed-power curve. The method comprises: importing standard wind speed-power data, and drawing a standard wind speed-power curve graph; importing actually acquired wind turbine generator data, which comprises wind speed and power, and drawing an actual wind speed-power scatter graph; on the basis of the wind speed, performing interval division on the data, and performing DBSCAN clustering processing on the data in each interval; determining a segmented dynamic threshold line, and performing secondary cleaning on the data which has been subjected to DBSCAN clustering; and dividing the cleaned data on the basis of the wind speed intervals divided in step 3, calculating the average value of the power in each interval, and obtaining wind speed-power data predicted on the basis of the actual data, so as to draw an actual wind speed-power curve. The present invention solves the technical problems in wind power prediction in the prior art of the accuracy and comprehensiveness of data cleaning operations being poor and the applicability of data cleaning to different types of data being low.

Description

Wind speed power curve abnormal data cleaning method and system TECHNICAL FIELD

[0001] The present application relates to the technical field of actual wind power curve prediction, in particular to a wind speed power curve abnormal data cleaning method and system. BACKGROUND

[0002] The wind speed-power curve of a wind turbine reflects the power information at different wind speeds, which plays an important role in evaluating the actual operation capacity and power generation capacity of the unit, and has certain reference value for optimizing the control strategy of the wind turbine. In the warranty acceptance of wind turbine, the consistency of wind power curve is an important parameter for evaluating the operation index of wind turbine. A large amount of wind speed and power data need to be collected, but due to the complex environment of wind turbine and the reasons of wind turbine itself and many other factors, there are a large number of abnormal and unreliable data in the actual data obtained, which causes the actual wind power curve obtained to be unreliable. Therefore, in order to ensure the accuracy of the final actual wind speed-power curve and the power consistency coefficient, it is particularly important to clean and select the collected data.

[0003] At present, there are many data cleaning methods, common ones include k-means clustering, DBSCAN clustering, Thomps tau method, etc. K-means clustering algorithm is an unsupervised learning algorithm, which divides the distance of data points to clusters based on Euclidean distance, and has the characteristics of simple implementation and high efficiency. However, determining the optimal k value is a relatively complex task, which needs to be determined by comparing multiple experiments. In addition, the K-means algorithm has poor effect on non-spherical cluster data. DBSCAN clustering algorithm is a density-based clustering algorithm, which divides the density of data points by defining the neighborhood radius (Eps) and the minimum number of samples in the neighborhood (Minpts). With the increase of wind power installed capacity year by year, the volatility and contingency of wind power have a great influence on grid connection. Due to the uncertainty of wind speed and wind energy, it is necessary to obtain reliable actual wind speed power curve to solve this problem, which is of great significance to ensure the stable operation of wind power system and maximize power generation efficiency. However, the existing wind speed power curve generation method has certain limitations in processing abnormal data points, so a new method is needed to remove these abnormal data points to better adapt to different wind turbine data.

[0004] The present application solves the technical problems of poor accuracy and comprehensiveness of data cleaning operation and low applicability of data cleaning for different types of data in the prior art of wind power prediction.

[0005] SUMMARY

[0006] The technical problems to be solved by the present application are that the existing wind power prediction technology has poor accuracy and comprehensiveness in data cleaning operation, and low applicability of data cleaning to different types of data.

[0007] The present application solves the above technical problems by adopting the following technical solutions: a wind speed-power curve abnormal data cleaning method comprises:

[0008] S1, import standard wind speed-power data to obtain a standard wind speed-power curve;

[0009] S2, collect and import actual wind turbine data, wherein the actual wind turbine data includes wind speed and power;

[0010] S3, perform data coarse-grained cleaning, wherein the actual wind turbine data is divided into intervals according to the wind speed in the actual wind turbine data to obtain no less than two wind speed intervals, and DBSCAN clustering is performed on the actual wind turbine data in each wind speed interval to obtain DBSCAN clustering data;

[0011] S4, perform data fine-grained cleaning, wherein the segment boundary points are determined, different fine-grained cleaning strategies are used for different interval segments, and the fine-grained cleaning strategies include dynamic threshold cleaning and static threshold cleaning; and obtain noise-removed data;

[0012] S5, according to the wind speed interval divided in step S3, the noise-removed data is divided, the power average value in each wind speed interval is calculated, the predicted wind speed-power data is obtained, and the actual wind speed-power curve is drawn according to the predicted wind speed-power data.

[0013] The present application adopts the method of DBSCAN clustering combined with dynamic threshold, which can clean abnormal data according to the characteristics of the data itself, and the dynamic threshold can better fit the trend of the original data, identify abnormal data more accurately, and clean more comprehensively.

[0014] In a more specific technical solution, in step S3, the actual wind turbine data is divided into intervals by using the following logic:

[0015] Wherein, v1, v2, v3 represent the wind speed in the standard wind speed-power data.

[0016] In a more specific technical solution, the DBSCAN clustering operation in step S3 comprises:

[0017] S31, randomly select an unvisited data point p;

[0018] S32. Perform an inspection operation centered on the unvisited data point p with a predefined radius to determine the core point and create a new cluster;

[0019] S33. For each core point, recursively check the secondary neighbor points of the core point's neighbor points, and add the secondary neighbor points to the current cluster;

[0020] S34. Repeat steps S32 to S33 until all core points are accessed within the range defined by the predefined radius.

[0021] S35. If there are unvisited non-core points within a predefined radius, mark the non-core points as noise.

[0022] In a more specific technical solution, step S32 includes:

[0023] S321. For neighboring points within a predefined radius Eps, iterate through and determine whether the number of neighbors of each neighboring point is greater than or equal to the predefined minimum number of neighbors Minpts.

[0024] S322. If so, mark the current neighbor point as the core point and create the current cluster accordingly.

[0025] In a more specific technical solution, in step S35, noise includes: noise points and boundary points.

[0026] In a more specific technical solution, step S4 includes:

[0027] S41. Based on the data characteristics when the wind speed is greater than or equal to the preset minimum wind speed at full power, determine the abnormal data cleaning threshold to serve as the dividing point between the first interval and the second interval segmented cleaning interval.

[0028] S42. In the first interval stage, calculate the noise filtering dynamic threshold line based on the ratio of all actual power to predicted power within the interval.

[0029] S43. In the second interval stage, when the wind speed is greater than the minimum wind speed for full power generation, the noise in the second interval is removed by comparing the actual power of the wind turbine with the power ratio threshold.

[0030] This invention starts with the characteristics of the data itself, automatically setting different dynamic threshold lines for different data types. Furthermore, it can import different standard wind speed-power data for different machine models, better adapting to various data from various models. This improves the applicability of the data cleaning solution.

[0031] In a more specific technical solution, in step S41, the preset minimum wind speed at full power is used as the dividing point between the first interval and the second interval when the power is constant in the statistical wind speed and power data.

[0032] In a more specific technical solution, in step S42, the actual and predicted power ratio at different powers is determined by a statistical method, and a linear function is fitted according to the wind speed and the actual and predicted power ratio, to serve as a noise filtering dynamic threshold line of the first interval.

[0033] In a more specific technical solution, the wind speed-power curve abnormal data cleaning system comprises:

[0034] A standard curve drawing module is used to import standard wind speed-power data, so as to obtain a standard wind speed-power curve;

[0035] An actual curve drawing module is used to collect and import actual wind turbine data, wherein the actual wind turbine data comprises wind speed and power, so as to draw an actual wind speed-power scatter plot;

[0036] An interval clustering module is used to perform data coarse-grained cleaning, wherein the actual wind turbine data is divided into intervals according to the wind speed in the actual wind turbine data, so as to obtain no less than two wind speed intervals, and DBSCAN clustering operation is performed on the actual wind turbine data in each wind speed interval, so as to obtain DBSCAN clustering data, and the interval clustering module is connected with the actual curve drawing module;

[0037] A data cleaning module is used to perform data fine-grained cleaning, wherein the segment boundary points are determined, different fine-grained cleaning strategies are used for different interval segments, the fine-grained cleaning strategies comprise dynamic threshold cleaning and static threshold cleaning, and filtered noise data is obtained, and the data cleaning module is connected with the interval clustering module;

[0038] An actual wind speed-power curve drawing module is used to divide the filtered noise data according to the wind speed intervals divided in step S3, calculate the power average value in each wind speed interval, so as to obtain predicted wind speed-power data, and draw an actual wind speed-power curve, and the actual wind speed-power curve drawing module is connected with the data cleaning module and the interval clustering module.

[0039] Compared with the prior art, the present application has the following advantages:

[0040] The present application adopts the method of DBSCAN clustering combined with dynamic threshold, which can clean abnormal data according to the characteristics of the data itself, and the dynamic threshold can better fit the trend of the original data, so that the identification of abnormal data is more accurate and the cleaning is more comprehensive.

[0041] The present application starts from the characteristics of the data itself, automatically formulates different dynamic threshold lines for different data, and then imports different standard wind speed-power data for different machine types, so as to better adapt to various data of various machine types, and improve the applicability of the data cleaning scheme.

[0042] The present application solves the technical problems of poor accuracy and comprehensiveness of data cleaning operation and low applicability of data cleaning for different types of data in the prior art of wind power prediction. BRIEF DESCRIPTION OF DRAWINGS

[0043] FIG. 1 is a schematic diagram of the basic steps of the wind speed-power curve abnormal data cleaning method according to the embodiment of the present application;

[0044] FIG. 2 is a schematic diagram of the specific implementation steps of the DBSCAN clustering algorithm according to the embodiment of the present application;

[0045] FIG. 3 is a schematic diagram of the specific implementation steps of the wind speed-power curve abnormal data cleaning method according to the embodiment of the present application;

[0046] FIG. 4 is a standard wind speed-power curve diagram according to the embodiment of the present application;

[0047] FIG. 5 is a wind speed-power scatter plot according to the embodiment of the present application;

[0048] FIG. 6a is a first-stage wind speed-power curve diagram according to the embodiment of the present application;

[0049] FIG. 6b is a second-stage wind speed-power curve diagram according to the embodiment of the present application;

[0050] FIG. 7 is an actual wind speed-power curve diagram according to the embodiment of the present application. DETAILED DESCRIPTION

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in a clear and complete manner in conjunction with the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0052] Embodiment 1

[0053] As shown in FIG. 1, the wind speed-power curve abnormal data cleaning method provided by the present application includes the following basic steps:

[0054] Step S1, import standard wind speed-power data, and calculate a standard wind speed-power curve;

[0055] In this embodiment, the standard wind speed-power data v={v1, v2, v3…} and s={s1, s2, s3…} are imported, and a standard wind speed-power curve diagram is drawn;

[0056] Step S2, import the actual acquired wind turbine data;

[0057] In the embodiment, the actually acquired wind turbine data includes but is not limited to: wind speed, power, and a scatter plot of actual wind speed-power is drawn;

[0058] Step S3: interval division is performed on the data according to the wind speed;

[0059] DBSCAN clustering processing is performed on the data in each interval;

[0060] As shown in FIG. 2, in the embodiment, the DBSCAN clustering algorithm further includes the following specific implementation steps:

[0061] S31: when the algorithm starts, a data point p that has not been visited is randomly selected;

[0062] S32: all neighbor points within a predefined radius (Eps) range centered on the point p are checked, and if the number of neighbors of the point is greater than or equal to a predefined minimum neighbor number (Minpts), the point is marked as a core point, and a new cluster is created;

[0063] S33: for each core point, the neighbors of the neighbors are recursively checked, and the neighbors are added to the current cluster;

[0064] S34: the foregoing steps S32 to S33 are repeated until all core points are visited;

[0065] S35: if there are unvisited non-core points, the points are marked as noise or boundary points and do not belong to any cluster.

[0066] Step S4: a segment cleaning boundary point is determined, and the data after DBSCAN clustering is cleaned again;

[0067] In the embodiment, in the first interval stage, when the wind speed is less than the minimum wind speed for full power, the actual and predicted power ratio at different power is determined by using a statistical method, and according to the wind speed and the actual and predicted power ratio, a linear function is fitted to serve as a segment dynamic threshold line of the first interval. Meanwhile, according to the result after DBSCAN clustering, a wind speed-power curve is fitted, the corresponding predicted power when the actual wind speed is less than the minimum wind speed for full power is calculated, the actual power and the predicted power are compared (the small ratio is multiplied by the large one), the ratio of the two is obtained, and the ratio is compared with the threshold value; the ratio less than the threshold value is marked as noise; in the second interval stage, when the wind speed is greater than the minimum wind speed for full power, the threshold value is equal to (0.96*full power), and the actual power less than the threshold value is marked as noise;

[0068] In the embodiment, in the determination process of the segmentation threshold, in the standard wind speed-power curve, when the current wind speed is greater than the preset full-load minimum wind speed, the power curve is constant and the power value is constant. Therefore, the full-load minimum wind speed can be found by statistical method according to the standard wind speed-power data.

[0069] In the embodiment, in the determination process of the data cleaning threshold, according to the determined segmentation threshold, the data is divided into two intervals, less than the preset full-load minimum wind speed and greater than the preset full-load minimum wind speed, which are referred to as the first interval and the second interval. Since the data distribution of the two intervals is significantly different, two methods of removing outliers are used for the two different intervals.

[0070] In the embodiment, before implementing the two methods of removing outliers, first, polynomial features are generated for the complete actual wind speed data and the wind speed data cleaned by DBSCAN. At this time, two polynomial features are obtained, which are referred to as the first feature and the second feature. Then, a polynomial regression model is fitted by using the second feature and the corresponding power data. Finally, the corresponding power data of the first feature is predicted by using the regression model as the predicted power, and the two methods of removing outliers are performed. The common points of the two methods include:

[0071] The specific value of the first method of removing outliers is dynamically set according to the wind speed, and the specific value of the second method of removing outliers is a constant value.

[0072] In the first interval threshold determination method of the embodiment, the method determines the ratio of different power prediction values to actual values by using statistical method, in which the value is large as the denominator. A linear function is fitted as the dynamic threshold line of interval 1 according to the wind speed and the ratio.

[0073] In the second interval threshold determination method of the embodiment, since the standard power is constant when the wind speed is greater than the segmentation threshold, according to the actual data, when the wind speed is greater than the segmentation threshold, the actual power also has the trend of approaching to a constant value. Therefore, a static threshold is set for interval 2.

[0074] Step S5, the cleaned data is divided according to the wind speed interval divided in step S3, the average value of the power in each interval is calculated, the wind speed-power data of the actual data prediction {(v1, s1'), (v2, s2'), (v3, s3')…} is obtained, and the actual wind speed-power curve is drawn according to the data.

[0075] Embodiment 2

[0076] As shown in FIG. 3, in the embodiment, the wind speed-power curve outlier cleaning method further includes the following specific implementation steps:

[0077] Step S1', as shown in Figure 4, import standard wind speed-power data, v={3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25}, s={35, 144, 310, 545, 872, 1312, 1816, 2286, 2461, 2500, 2500, 2500, 2500, 2500, 2500, 2500, 2500, 2500, 2500, 2500, 2500, 2500, 2500}, draw standard wind speed-power curve;

[0078] Step S2', as shown in Figure 5, import the actual acquired wind turbine data, including wind speed and power, draw the actual wind speed-power scatter plot;

[0079] Step S3', according to the wind speed, the data is divided into intervals, {(2.5, 3.5), (3.5, 4.5), (4.5, 5.5), (5.5, 6.5), (6.5, 7.5)…}, DBSCAN clustering processing is carried out for the data in each interval;

[0080] In the DBSCAN clustering algorithm used in the embodiment, at the beginning of the algorithm, an unvisited data point p is randomly selected; check all neighbor points within the predefined radius (Eps=0.5) range with p point as the center, if the number of neighbors of the point is greater than or equal to the predefined minimum neighbor number (Minpts=2), the point is marked as a core point, and a new cluster is created; for each core point, recursively check the neighbors of its neighbors and add them to the current cluster; the foregoing operation is executed in a loop until all core points are visited; if there are unvisited non-core points, these points are marked as noise or boundary points, and do not belong to any cluster;

[0081] Step S4', as shown in Figures 6a and 6b, a piece of segmented dynamic threshold line is determined to perform secondary cleaning on the data clustered by DBSCAN; in the first interval stage of the embodiment, when the wind speed is less than 12 m / s, it is determined through experiments that when the threshold value is equal to 0.7, the removal effect is good when the wind speed is greater than 12 m / s, and the removal effect is good when the threshold value is equal to 0.05 and the wind speed is less than 2.5 m / s, a straight line is determined as a dynamic threshold line, and a wind speed-power curve is fitted according to the results after DBSCAN clustering, the corresponding predicted power when the actual wind speed is less than the minimum wind speed for full power is calculated, the actual power and the predicted power are compared (the smaller ratio is multiplied by the larger one), and the ratio is compared with the threshold value, and the ratio less than the threshold value is marked as noise; in the second interval stage of the embodiment, when the wind speed is greater than the minimum wind speed for full power, the threshold value is equal to (0.96*2500), and the actual power less than the threshold value is marked as noise.

[0082] Step S5, as shown in Figure 7, the cleaned data is divided according to the wind speed interval divided in the foregoing step S3', the average value of the power in each interval is calculated, the wind speed-power data of the actual data prediction {(3, 35.35), (4, 107.04), (5, 235.32), (6, 428.21), …} is obtained, and the actual wind speed-power curve is drawn according to the data.

[0083] The method of DBSCAN clustering combined with dynamic threshold value can clean the abnormal data according to the characteristics of the data itself, the dynamic threshold value can better fit the trend of the original data, the abnormal data can be more accurately identified, and the cleaning can be more comprehensive.

[0084] The present application automatically formulates different dynamic threshold lines for different data according to the characteristics of the data itself, and can import different standard wind speed-power data for different models, so as to better adapt to various data of various models, and improve the applicability of the data cleaning scheme.

[0085] The present application solves the technical problems of poor accuracy and comprehensiveness of data cleaning operation and low applicability of data cleaning for different types of data in the prior art of wind power prediction.

[0086] The above embodiments are only used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for cleaning abnormal data of a wind speed-power curve, characterized in that, The method comprises: S1, importing standard wind speed-power data, according to which a standard wind speed-power curve is obtained; S2, collecting and importing actual wind turbine data, wherein the actual wind turbine data comprises wind speed and power, according to which an actual wind speed-power scatter plot is drawn; S3, performing data coarse-grained cleaning, wherein the actual wind turbine data is divided into intervals according to the wind speed in the actual wind turbine data, so as to obtain not less than two wind speed intervals, and DBSCAN clustering operation is performed on the actual wind turbine data in each wind speed interval, so as to obtain DBSCAN clustering data; S4, performing data fine-grained cleaning, wherein the segment boundary points are determined, different fine-grained cleaning strategies are used for different interval segments, the fine-grained cleaning strategies comprise dynamic threshold cleaning and static threshold cleaning, and the DBSCAN clustering data is cleaned to obtain data filtered out of noise; S5, according to the wind speed intervals divided in the step S3, the filtered noise data is divided to calculate the average power in each wind speed interval, so as to obtain predicted wind speed-power data, according to which an actual wind speed-power curve is drawn.

2. The wind speed power curve anomaly data cleaning method of claim 1, wherein, In the step S3, the actual wind turbine data is divided into intervals using the following logic: In the formula, v1, v2, v3 represent the wind speed in the standard wind speed-power data.

3. The wind speed power curve anomaly data cleaning method of claim 1, wherein, The DBSCAN clustering operation in the step S3 comprises: S31, randomly selecting an unvisited data point p; S32, checking the unvisited data point p as the center to perform checking operation according to a pre-defined radius, so as to determine core points and create new clusters; S33, for each core point, recursively checking the secondary neighbor points of the neighbor points of the core point, and adding the secondary neighbor points to the current cluster; S34, the steps S32 to S33 are executed in a loop until the visiting operation of all core points in the range determined by the pre-defined radius is completed; S35, if there is an unvisited non-core point in the range determined by the pre-defined radius, the non-core point is marked as noise.

4. The wind speed power curve anomaly data cleaning method according to claim 3, characterized in that, The step S32 comprises: S321, for the neighbor points in the range of the pre-defined radius Eps, it is judged whether the number of neighbors of the neighbor points is greater than or equal to the pre-defined minimum neighbor number Minpts; S322, if yes, the current neighbor point is marked as the core point, according to which the current cluster is created.

5. The wind speed power curve anomaly data cleaning method according to claim 3, characterized in that, In the step S35, the noise comprises noise points and boundary points.

6. The wind speed power curve anomaly data cleaning method of claim 1, wherein, The step S4 comprises: S41, according to the data characteristics when the wind speed is greater than or equal to the preset full-load minimum wind speed, an abnormal data cleaning threshold is determined as the boundary point of the cleaning interval of the first interval segment and the second interval segment; S42, in the first interval stage, a noise filtering dynamic threshold line is calculated according to the ratio of all actual power to predicted power in the interval; S43, in the second interval stage, when the wind speed is greater than the full-load minimum wind speed, the actual power of the wind turbine is compared with the power ratio threshold, according to which the second interval noise is removed.

7. The wind speed power curve anomaly data cleaning method according to claim 6, characterized in that, In the S41, the preset full-load minimum wind speed corresponding to a constant power in the wind speed power data is determined as the demarcation point of the first interval stage and the second interval stage.

8. The wind speed power curve anomaly data cleaning method of claim 6, wherein, In the S42, the actual and predicted power ratio at different powers is determined by using a statistical method, and a linear function is fitted according to the wind speed and the actual and predicted power ratio, so as to obtain the segmented dynamic threshold line of the first interval.

9. A wind speed power curve anomaly data cleaning system, characterized in that, The system comprises: a standard curve drawing module configured to import standard wind speed-power data to obtain a standard wind speed-power curve; an actual curve drawing module configured to collect and import actual wind turbine data, wherein the actual wind turbine data comprises wind speed and power, and an actual wind speed-power scatter plot is drawn according to the actual wind turbine data; an interval clustering module configured to perform data coarse-grained cleaning, wherein the actual wind turbine data is divided into wind speed intervals according to the wind speed in the actual wind turbine data to obtain no less than two wind speed intervals, and DBSCAN clustering is performed on the actual wind turbine data in each wind speed interval to obtain DBSCAN clustering data, the interval clustering module is connected with the actual curve drawing module; a data cleaning module configured to perform data fine-grained cleaning, wherein the demarcation points of segmentation are determined, different fine-grained cleaning strategies are used for different interval segments, the fine-grained cleaning strategies comprise dynamic threshold cleaning and static threshold cleaning, and the DBSCAN clustering data is cleaned to obtain noise-removed data, the data cleaning module is connected with the interval clustering module; an actual wind speed-power curve drawing module configured to divide the noise-removed data according to the wind speed intervals divided in the S3, calculate the power average in each wind speed interval, and obtain predicted wind speed-power data to draw an actual wind speed-power curve, the actual wind speed-power curve drawing module is connected with the data cleaning module and the interval clustering module.

Images