A wind farm wind power prediction method, medium, device and product
By performing binning and density clustering analysis on wind farm data, theoretical stress curves are generated and segmented predictions are made, solving the problem of low accuracy in wind power prediction for ultra-short-term wind farms and achieving high-precision prediction under limited resources.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUODIAN UNITED POWER TECH
- Filing Date
- 2026-05-18
- Publication Date
- 2026-07-24
Smart Images
Figure CN122456476A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of wind power prediction, specifically relating to a wind farm wind power prediction method, storage medium, electronic equipment and computer program product. Background Technology
[0002] In the context of a new power system with a high proportion of wind power and the normalized operation of the electricity spot market, ultra-short-term wind power forecasting has become a core foundation for supporting real-time grid balance, ensuring compliant grid connection of new energy sources, and improving electricity market trading revenue. Wind power is characterized by strong intermittency, volatility, and randomness. Its output is easily affected by sudden weather changes and terrain disturbances, resulting in rapid upswings and sudden drops, making it difficult for short-term forecasts to cover short-term fluctuations.
[0003] Currently, ultra-short-term forecasting models typically need to be reported every 15 minutes, which has limitations such as limited calculation cycles and limited computing resources. Field systems often directly use the latest short-term wind power forecast data or a single forecast deviation model for correction, resulting in low accuracy of ultra-short-term forecasts, which makes it difficult to meet the requirements of the spot market for real-time power output accuracy.
[0004] Therefore, existing technologies lack a technical solution for highly accurate ultra-short-term prediction of wind power in wind farms under limited computing cycles and resources. Summary of the Invention
[0005] The purpose of this application is to provide a method, medium, device and product for predicting wind power in wind farms, which can solve the problem that the accuracy of current ultra-short-term prediction models is low due to limited calculation cycle and computing resources.
[0006] In a first aspect, embodiments of this application provide a method for predicting wind power in a wind farm, the method comprising: Obtain data on grid-connected wind power, short-term wind power forecasts, and average wind speed of wind turbines; The grid-connected power of the wind farm and the average wind speed of the wind turbine are divided into boxes according to the preset wind speed interval, and density cluster analysis is performed on the data in each box to obtain the theoretical stress curve. Based on the theoretical stress curve and the short-term wind power prediction data, the prediction period is divided into multiple prediction sub-intervals; The wind power of the wind farm in the prediction period is predicted in segments according to each prediction sub-interval, and the wind power prediction result is obtained.
[0007] Optionally, the step of dividing the grid-connected power of the wind farm and the average wind speed of the wind turbine into boxes according to a preset wind speed interval includes: The average wind speed of the wind turbine is divided into multiple wind speed intervals according to a preset wind speed interval. The grid-connected power of the wind farm is divided into its respective wind speed range according to the wind speed range to which the average wind speed of the corresponding wind turbine belongs, resulting in multiple data bins corresponding to the wind speed ranges. Each data bin contains wind speed data belonging to the same wind speed range and power data corresponding to the wind speed data. The wind speed data is the average wind speed of the wind turbine, and the power data is the grid-connected power of the wind farm.
[0008] Optionally, the step of performing density clustering analysis on the data within each bin to obtain the theoretical stress curve includes: Density clustering analysis was performed on the wind speed data and the power data corresponding to the wind speed data in each data bin to obtain the sub-curve segment corresponding to each data bin. By splicing together the sub-curve segments corresponding to all data bins in ascending order of wind speed range, the theoretical stress curve is obtained.
[0009] Optionally, the step of performing density clustering analysis on the wind speed data and the corresponding power data within each data bin to obtain the sub-curve segment corresponding to each data bin includes: Density clustering analysis was performed on each data bin to obtain multiple clusters within each data bin; For each data bin, the cluster with the most data points among all clusters in the data bin is selected as the master cluster. Based on the wind speed data contained in the master cluster and the power data corresponding to the wind speed data, a sub-curve segment corresponding to each data bin is generated.
[0010] Optionally, the step of selecting the cluster with the most data points among all clusters within the data bin as the master cluster, and generating a sub-curve segment corresponding to each data bin based on the wind speed data contained in the master cluster and the power data corresponding to the wind speed data, including: The cluster with the most data points among all clusters in the data bin is selected as the master cluster. The upper power boundary, lower power boundary, and target power value are determined based on the power data in the master cluster. The power data in each data bin that are higher than the upper power boundary or lower than the lower power boundary are corrected to the target power value, thus obtaining the sub-curve segment corresponding to each data bin.
[0011] Optionally, the prediction period is divided into multiple prediction sub-intervals based on the theoretical stress curve and the short-term wind power prediction data, including: Obtain the prediction quality between the theoretical stress curve and the short-term wind power prediction data; Based on the prediction quality, the prediction period is divided into multiple consecutive prediction sub-intervals.
[0012] Optionally, the step of segmenting the wind power prediction for the wind farm during the prediction period according to each prediction sub-interval to obtain the wind power prediction result includes: Obtain the prediction accuracy for each prediction sub-interval; When the prediction accuracy of the prediction sub-interval meets the first preset condition, the wind power of the wind farm in the prediction sub-interval is predicted according to the ratio between theoretical power and short-term wind power prediction data, and the theoretical power prediction result is obtained. When the prediction accuracy of the prediction sub-interval meets the second preset condition, the short-term wind power prediction data is used as the short-term prediction result. When the prediction accuracy of the prediction sub-interval meets the third preset condition, the wind power of the wind farm in the prediction sub-interval is predicted according to the ratio between theoretical power and short-term wind power prediction data to obtain the theoretical power prediction result. The short-term wind power prediction data is used as the short-term prediction result. The theoretical power prediction result and the short-term prediction result are weighted and fused to obtain the combined prediction result. The wind power prediction result is obtained by combining the theoretical prediction result, the short-term prediction result, and the combined prediction result.
[0013] Secondly, embodiments of this application provide a storage medium that stores computer instructions, which, when executed by a computer, are used to perform the steps of the wind farm wind power prediction method as described in the first aspect.
[0014] Thirdly, embodiments of this application provide an electronic device, including at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the wind farm wind power prediction method as described in the first aspect.
[0015] Fourthly, embodiments of this application provide a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the wind farm wind power prediction method as described in the first aspect.
[0016] In this embodiment, the grid-connected power of the wind farm and the average wind speed of the wind turbine are divided into bins according to a preset wind speed interval. Continuous wind speed and power data are discretized according to wind speed intervals, so that the data in each bin have similar wind speed characteristics, avoiding mutual interference between data from different wind speed intervals. Subsequently, density clustering analysis is performed on the data in each bin, which can automatically identify dense areas of data point distribution, effectively filter out outliers under abnormal operating conditions, and select the most representative data clusters from each bin. The theoretical stress curve generated accordingly can truly reflect the ideal mapping relationship between wind speed and power under normal operating conditions. Furthermore, since density clustering analysis does not require pre-specifying the number of clusters and has low computational complexity, compared with the current ultra-short-term prediction models that require a large number of training iterations, it can quickly construct the theoretical stress curve with limited computing resources and adapt to the calculation cycle constraint of reporting every 15 minutes. Secondly, based on the theoretical stress curve and short-term wind power prediction data, the prediction period can be divided into multiple prediction sub-intervals with low computational load, without the need for additional numerical weather forecasts or complex meteorological analysis, further reducing the demand for computing resources. Finally, unlike existing technologies that directly use the latest short-term wind power prediction data or a single prediction mode of a single deviation correction model, the embodiments of this application use an adapted prediction method to predict the wind power of wind farms in segments for different prediction sub-intervals. This can more accurately capture the short-term fluctuation trend of wind power within a limited calculation period, avoid the accumulation of prediction deviations of a single model in all time periods, and thus effectively improve the overall accuracy of ultra-short-term prediction. Attached Figure Description
[0017] Figure 1 This is a flowchart illustrating the steps of a wind farm wind power prediction method provided in an embodiment of this application; Figure 2 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of this application. Detailed Implementation
[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0019] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.
[0020] The wind power prediction method for wind farms provided in this application will be described in detail below with reference to the accompanying drawings, through specific embodiments and application scenarios.
[0021] Reference Figure 1 This is a flowchart illustrating the steps of a wind farm wind power prediction method provided in this application embodiment, specifically including the following steps: Step 101: Obtain the grid-connected power of the wind farm, short-term wind power forecast data, and average wind speed of the wind turbine units; In this embodiment of the application, it is necessary to collect historical data of the wind farm every 15 minutes. The historical data includes, but is not limited to, the grid-connected power of the wind farm, short-term wind power prediction data updated every hour, and actual wind measurement data (average wind speed of wind turbine units).
[0022] To ensure data quality and the accuracy of subsequent analysis, historical data preprocessing is necessary. Specifically, firstly, the grid-connected power of wind farms, short-term wind power forecast data, and average wind speed of wind turbines are concatenated column-wise to form a unified data table; secondly, missing values in each data set are filled in, and if missing values are found, they are filled with the data values from the previous time point. After data preprocessing is completed, the data proceeds to step 102 for binning.
[0023] In this embodiment, historical data is divided into a training set and a test set. The training set accounts for 80% and the test set accounts for 20%. The training set is used to construct the theoretical response curve, determine the segmented prediction strategy, and optimize the weighting coefficients of the combined prediction method. The test set is used to verify the accuracy and generalization ability of the prediction method constructed in this embodiment and to evaluate the model's prediction performance on unknown data.
[0024] Step 102: Divide the grid-connected power of the wind farm and the average wind speed of the wind turbine into boxes according to the preset wind speed interval, and perform density cluster analysis on the data in each box to obtain the theoretical stress curve; In this embodiment, based on the preprocessed data, the average wind speed of the wind turbine and the grid-connected power of the wind farm are divided into multiple data bins according to a preset wind speed interval. In one embodiment, the preset wind speed interval can be set to 0.5 m / s, and each data bin corresponds to a wind speed range [v-0.25, v+0.25), where v is the center value of the wind speed of the data bin.
[0025] After binning the data, density clustering analysis needs to be performed on the data within each bin to obtain the theoretical stress curve. The theoretical stress curve is used to characterize the ideal mapping relationship between the average wind speed of the wind turbine and the grid-connected power of the wind farm under normal operating conditions.
[0026] Step 103: Based on the theoretical stress curve and the short-term wind power prediction data, divide the prediction period into multiple prediction sub-intervals; In this embodiment, the forecast period refers to an ultra-short-term forecast window within a certain future period, such as 16 forecast points within the next 4 hours, one point every 15 minutes. To more accurately capture the fluctuation pattern of wind power in different time periods, this embodiment divides the forecast period into multiple continuous forecast sub-intervals by analyzing the theoretical stress curve and short-term wind power forecast data.
[0027] Step 104: Perform segmented prediction of wind power in the wind farm during the prediction period based on each prediction sub-interval to obtain the wind power prediction result.
[0028] In the embodiments of this application, for each prediction sub-interval, a matching prediction method is selected from multiple prediction methods to perform ultra-short-term prediction of the wind power of the wind farm within the prediction sub-interval. Finally, the prediction results of each prediction sub-interval are spliced together in chronological order to obtain a complete wind power prediction result.
[0029] In this embodiment, the grid-connected power of the wind farm and the average wind speed of the wind turbine are divided into bins according to a preset wind speed interval. Continuous wind speed and power data are discretized according to wind speed intervals, so that the data in each bin have similar wind speed characteristics, avoiding mutual interference between data from different wind speed intervals. Subsequently, density clustering analysis is performed on the data in each bin, which can automatically identify dense areas of data point distribution, effectively filter out outliers under abnormal operating conditions, and select the most representative data clusters from each bin. The theoretical stress curve generated accordingly can truly reflect the ideal mapping relationship between wind speed and power under normal operating conditions. Furthermore, since density clustering analysis does not require pre-specifying the number of clusters and has low computational complexity, compared with the current ultra-short-term prediction models that require a large number of training iterations, it can quickly construct the theoretical stress curve with limited computing resources and adapt to the calculation cycle constraint of reporting every 15 minutes. Secondly, based on the theoretical stress curve and short-term wind power prediction data, the prediction period can be divided into multiple prediction sub-intervals with low computational load, without the need for additional numerical weather forecasts or complex meteorological analysis, further reducing the demand for computing resources. Finally, unlike existing technologies that directly use the latest short-term wind power prediction data or a single prediction mode of a single deviation correction model, the embodiments of this application use an adapted prediction method to predict the wind power of wind farms in segments for different prediction sub-intervals. This can more accurately capture the short-term fluctuation trend of wind power within a limited calculation period, avoid the accumulation of prediction deviations of a single model in all time periods, and thus effectively improve the overall accuracy of ultra-short-term prediction.
[0030] In one embodiment of this application, the step of separating the grid-connected power of the wind farm and the average wind speed of the wind turbine according to a preset wind speed interval includes: The average wind speed of the wind turbine is divided into multiple wind speed intervals according to a preset wind speed interval. The grid-connected power of the wind farm is divided into its respective wind speed range according to the wind speed range to which the average wind speed of the corresponding wind turbine belongs, resulting in multiple data bins corresponding to the wind speed ranges. Each data bin contains wind speed data belonging to the same wind speed range and power data corresponding to the wind speed data. The wind speed data is the average wind speed of the wind turbine, and the power data is the grid-connected power of the wind farm.
[0031] In this embodiment, the preset wind speed interval can be set according to the data distribution characteristics of the actual wind farm. In one embodiment, the preset wind speed interval is set to 0.5 m / s. Based on this preset wind speed interval, the range of the average wind speed of the wind turbine is divided into multiple continuous and non-overlapping wind speed intervals. The range of each wind speed interval is [v-0.25, v+0.25), where v is the center value of the wind speed interval. Adjacent intervals are seamlessly connected at the boundary (the right boundary of the previous interval is equal to the left boundary of the next interval).
[0032] For example, when v=5.0, the corresponding wind speed range is [4.75, 5.25); when v=5.5, the corresponding wind speed range is [5.25, 5.75); and so on, until the wind speed range in all historical data is covered.
[0033] It should be noted that there is a one-to-one correspondence between the average wind speed of wind turbines and the grid-connected power of wind farms in terms of time. That is, the average wind speed of wind turbines collected at each historical moment corresponds to a grid-connected power value of wind farms at the same moment. This one-to-one correspondence is the prerequisite for correctly classifying the power data into the corresponding wind speed range.
[0034] First, the average wind speed of the wind turbine needs to be obtained to determine the wind speed range to which the wind speed value belongs. Then, the grid-connected power of the wind farm corresponding to the wind speed value is allocated to the data bin corresponding to the wind speed range. After traversing all wind speed values, a data bin is generated for each wind speed range. Each data bin contains all the wind speed data belonging to that wind speed range and the power data corresponding to each wind speed data.
[0035] Specifically, the wind speed data contained in each data bin refers to the average wind speed value of all wind turbines whose values fall within the wind speed range corresponding to that bin. The power data corresponding to this wind speed data refers to the grid-connected power value of the wind farm collected at the same historical moment as each of the aforementioned wind speed values. In other words, each data bin is essentially a data pair (wind speed-power data pair) composed of several sets of wind speed data and power data. In each data pair, the wind speed values belong to the same wind speed range, and the power value is the actual grid-connected power at the same moment corresponding to that wind speed value.
[0036] For example, assuming a wind turbine has an average wind speed of 5.0 m / s, which falls within the interval [4.75, 5.25), the corresponding grid-connected power of the wind farm (e.g., 1500 kW) is assigned to the data bin corresponding to the center value v=5.0. If another wind turbine has an average wind speed of 5.1 m / s, also falling within the interval [4.75, 5.25), its corresponding power is also assigned to the same data bin. Ultimately, the data bin corresponding to v=5.0 contains all wind speed-power data pairs within the range of 4.75 m / s to 5.25 m / s.
[0037] This embodiment of the application, through the aforementioned binning process, discretizes the originally continuously distributed wind speed and power data according to wind speed intervals, ensuring that the data within each data bin have similar wind speed characteristics. The difference in wind speed values within the same bin will not exceed a preset wind speed interval (e.g., 0.5 m / s), thereby avoiding mutual interference between data from different wind speed intervals. This provides a clear and consistent data organization foundation for subsequent density clustering analysis of each data bin.
[0038] In one embodiment of this application, the step of performing density clustering analysis on the data within each bin to obtain the theoretical stress curve includes: Density clustering analysis was performed on the wind speed data and the power data corresponding to the wind speed data in each data bin to obtain the sub-curve segment corresponding to each data bin. By splicing together the sub-curve segments corresponding to all data bins in ascending order of wind speed range, the theoretical stress curve is obtained.
[0039] This application embodiment performs density clustering analysis on the data within each data bin. Density clustering analysis is a clustering method based on the distribution density of data points. By performing density clustering analysis on the data within each data bin, this application embodiment can automatically identify densely distributed areas of data points within each wind speed range, filter out data generated by wind turbines under normal operating conditions, and exclude outlier data generated under abnormal operating conditions (such as power outages, faults, maintenance, etc.), thereby ensuring that the subsequently generated sub-curve segments can accurately reflect the ideal power mapping relationship within that wind speed range.
[0040] Density clustering analysis can generate a corresponding sub-curve segment for each data bin. Specifically, the wind speed-power data pairs within each data bin can reflect the power value range and trend within that data bin, thus forming a continuous curve segment. This sub-curve segment represents the ideal mapping relationship between wind speed and power of the wind turbine under normal operating conditions within that data bin.
[0041] Iterate through all data bins, repeating the density clustering analysis and sub-curve segment generation steps for each data bin until a corresponding sub-curve segment is generated for each data bin.
[0042] Finally, all data bins are spliced together according to their respective wind speed ranges, from smallest to largest. Since wind speed ranges are continuous and ordered (e.g., [4.75, 5.25), [5.25, 5.75), [5.75, 6.25), etc.), different wind speed ranges have a certain arrangement order based on their size. Adjacent sub-curve segments can be naturally spliced at the wind speed boundaries, forming a complete curve covering the entire wind speed range. This curve is the theoretical stress curve.
[0043] It should be noted that the theoretical wind speed curve is used to characterize the ideal mapping relationship between the average wind speed of the wind turbine and the grid-connected power of the wind farm under normal operating conditions (i.e., excluding abnormal states such as wind turbine power curtailment, faults, and maintenance). It reflects the inherent wind speed-power characteristics of the wind farm during normal operation and is an important basis for subsequent segmented prediction.
[0044] This application's embodiments utilize density clustering analysis, processing each data bin independently. This automatically adapts to the data distribution characteristics across different wind speed ranges. Furthermore, by performing density clustering analysis separately on each data bin, mutual interference between data from different wind speed ranges is avoided. This ensures that the generated theoretical stress curve accurately reflects the inherent wind speed-power mapping relationship of a wind farm under normal operating conditions. Based on this, the curve segments from each bin are stitched together sequentially according to wind speed ranges to form a complete theoretical stress curve covering the entire wind speed range. This provides a reliable theoretical benchmark for subsequent segmented predictions, effectively improving the accuracy of ultra-short-term wind power prediction with limited computing resources.
[0045] In one embodiment of this application, the step of performing density clustering analysis on the wind speed data and the power data corresponding to the wind speed data in each data bin to obtain the sub-curve segment corresponding to each data bin includes: Density clustering analysis was performed on each data bin to obtain multiple clusters within each data bin; For each data bin, the cluster with the most data points among all clusters in the data bin is selected as the master cluster. Based on the wind speed data contained in the master cluster and the power data corresponding to the wind speed data, a sub-curve segment corresponding to each data bin is generated.
[0046] In practical applications, distance-based clustering methods (such as K-means clustering) are commonly used for data clustering analysis. However, these methods require pre-specifying the number of clusters, which is often unknown in real-world data. Furthermore, distance-based clustering is sensitive to the selection of initial cluster centers; different initial values can lead to different clustering results. Additionally, distance-based clustering is typically based on Euclidean distance, making it difficult to identify non-convex clusters, and it is highly sensitive to noise and outliers, easily forcing abnormal data into a particular cluster, thus affecting the accuracy of the clustering results. For wind farm data, outliers generated by abnormal operating conditions (such as turbine power curtailment, faults, and maintenance) are common. Traditional distance-based clustering methods can mix these outliers with normal data, causing the generated theoretical stress curve to deviate from the actual wind speed-power mapping relationship.
[0047] Therefore, by performing density clustering analysis on the wind speed-power data pairs (wind speed data and its corresponding power data) in each data bin, this application embodiment can automatically identify dense areas in the data space without pre-specifying the number of clusters, divide densely connected data points into the same cluster, and effectively eliminate isolated noise points in low-density areas.
[0048] Specifically, the wind speed data and corresponding power data in each data box need to be mapped to data points in a two-dimensional space. The horizontal axis of the two-dimensional space is the average wind speed of the wind turbine, and the vertical axis is the corresponding grid-connected power value of the wind farm. Each data point represents a set of wind speed-power data pairs.
[0049] Secondly, determine the similarity between different data points. As an example, the similarity between different data points can be measured based on the Euclidean distance between the data points. The smaller the distance between two data points, the higher their similarity; conversely, the larger the distance, the lower the similarity.
[0050] Then, for each data bin, all data points are analyzed based on the similarity between different data points. Specifically, for any data point, if the number of data points within its neighborhood radius is greater than or equal to a preset number, the data point is marked as a core point; if a data point is within the neighborhood of a core point but is not itself a core point, it is marked as a boundary point; if a data point is neither a core point nor within the neighborhood of any core point, it is marked as a noise point. Through this method, all core points and their density-connected boundary points are grouped into the same cluster, while noise points are excluded. Ultimately, the data points within each data bin are divided into multiple clusters, along with several excluded isolated noise points.
[0051] It should be noted that while the similarity between different data points can be calculated based on Euclidean distance, which is similar to the aforementioned distance-based clustering analysis method in terms of distance measurement, distance-based clustering directly uses distance for partitioning, requires pre-specifying the number of clusters, and each data point must belong to a specific cluster, making it unable to identify noise points. In contrast, the density-based clustering method used in this application does not directly partition based on distance. Instead, it determines whether to form a cluster by judging whether the density of points in the neighborhood of a data point reaches a threshold. It does not require pre-specifying the number of clusters and can automatically identify and exclude points in low-density areas as noise points. In other words, distance measurement in this application is only used as a basic tool for judging similarity, not as a direct basis for clustering.
[0052] For example, suppose a data bin contains 200 data points with a neighborhood of 0.3 and a preset number of points of 5. After density clustering analysis, it is found that 150 data points are densely distributed in one area, with the Euclidean distance between these points all within 0.3, and each point has at least 5 other points in its neighborhood. These 150 points are classified into one cluster. Another 30 points are distributed in another area, but with a lower density, and are classified into another smaller cluster. The remaining 20 points are isolated and do not have a sufficient number of neighboring points. They are identified as noise points and excluded.
[0053] To select the data cluster that best represents the wind speed-power mapping relationship under normal operating conditions from each data bin, the cluster with the most data points within each data bin needs to be selected as the master cluster. The master cluster represents the most representative and frequently occurring wind speed-power mapping relationship within that wind speed range, because the largest number of data points means that this operating condition has occurred most frequently in historical data, best reflecting the normal output characteristics of the wind turbine in that wind speed range. Therefore, based on the wind speed data and corresponding power data contained in the master cluster, a sub-curve segment corresponding to each data bin can be generated.
[0054] This application embodiment uses density clustering analysis to automatically adapt to the characteristics of data distribution in different wind speed ranges. It does not require pre-setting the number of clusters, effectively identifies dense areas of data points in each sub-box and eliminates abnormal noise points, thereby ensuring that the generated sub-curve segments can truly reflect the wind speed-power mapping relationship of the wind turbine under normal operating conditions.
[0055] In one embodiment of this application, the step of selecting the cluster with the most data points among all clusters within a data bin as the master cluster, and generating a sub-curve segment corresponding to each data bin based on the wind speed data contained in the master cluster and the power data corresponding to the wind speed data, includes: The cluster with the most data points among all clusters in the data bin is selected as the master cluster. The upper power boundary, lower power boundary, and target power value are determined based on the power data in the master cluster. The power data in each data bin that are higher than the upper power boundary or lower than the lower power boundary are corrected to the target power value, thus obtaining the sub-curve segment corresponding to each data bin.
[0056] Within each data bin obtained through density clustering analysis, the cluster with the most data points (the primary cluster) contains most of the data points generated by the wind turbines under normal operating conditions within that wind speed range. The power values of these data points are not entirely equal, but fluctuate within a certain range around a central value. The maximum power value in the primary cluster is defined as the upper power boundary, and the minimum value as the lower power boundary. These upper and lower power boundaries together define the reasonable fluctuation range of the wind turbine's power output under normal operating conditions within that wind speed range. The target power value is a representative value of the power data in the primary cluster, used to replace abnormal power data that exceeds the reasonable fluctuation range.
[0057] In one embodiment, the target power value can be the average, median, or most frequently occurring power value of all power data in the main cluster. Preferably, the average power data in the main cluster is used as the target power value because the average value can better reflect the overall level of the cluster.
[0058] For example, suppose that the cluster (main cluster) with the most data points in a data bin contains 150 data points, and the power values of these data points are distributed between 1400kW and 1600kW, with an upper power limit of 1600kW, a lower power limit of 1400kW, and an average power value of 1500kW. Then the upper power limit of this data bin is 1600kW, the lower power limit is 1400kW, and the target power value is 1500kW.
[0059] In actual wind farm operation, due to factors such as power curtailment, malfunctions, maintenance, and sudden weather changes, some historical data points may deviate from the power output range under normal operating conditions. If these abnormal data points are directly used to generate the theoretical response curve, the curve will deviate from the actual wind speed-power mapping relationship, thus affecting the accuracy of subsequent ultra-short-term forecasts. By correcting abnormal power data that are above the upper power boundary or below the lower power boundary to the target power value, the interference of abnormal operating conditions on the theoretical response curve can be effectively eliminated, making the data in each data bin more concentrated in reflecting the power output characteristics under normal operating conditions.
[0060] Within a data bin, besides the main cluster (representing normal operating conditions), there may be other smaller clusters (representing other atypical operating conditions) and isolated data points identified as noise by density clustering algorithms (representing abnormal operating conditions). Among these atypical clusters and noisy data points, the power values of some data points may fall outside the power boundary range of normal operating conditions (i.e., above the upper power boundary or below the lower power boundary), while others may fall within the boundary range. This application only corrects atypical and abnormal data points that exceed the boundary range, pulling their power values back to the target power value, thus placing them within the normal range in terms of power. For data points that, although not belonging to the main cluster, already have power values within the boundary range, their original power values can be retained, or correction can be decided based on actual needs.
[0061] Specifically, by traversing all data points within each data bin, data points with power values higher than the upper power boundary are modified to the target power value; data points with power values lower than the lower power boundary are also modified to the target power value; and data points with power values between the upper and lower power boundaries are left unchanged. After this correction, the abnormal data points within each data bin are adjusted to the normal power range, resulting in a more concentrated and standardized data distribution within that data bin.
[0062] Based on the corrected data (all corrected clusters within each data bin), a corresponding sub-curve segment is generated for each data bin. This sub-curve segment represents the ideal mapping relationship between wind speed and power of the wind turbine under normal operating conditions within the wind speed range.
[0063] For example, suppose a data bin contains 150 data points in its largest cluster (primary cluster A). The power values of these data points range from 1400kW to 1600kW, with an upper power limit of 1600kW, a lower power limit of 1400kW, and an average power of 1500kW. Simultaneously, this data bin also contains a smaller cluster (cluster B) with 30 data points, whose power values range from 1200kW to 1350kW, all below the lower power limit of 1400kW. There are also 20 noise points, 15 of which have power values between 1650kW and 1700kW (above the upper power limit), and 5 have power values between 1300kW and 1380kW (below the lower power limit). After a correction operation, the power values of all 30 data points in cluster B, along with the 15 noise points above the upper limit and the 5 noise points below the lower limit, are corrected to the target power value of 1500kW. After correction, all data points outside cluster A within the data bin were adjusted to the normal range in terms of power dimension, making the sub-curve segments generated based on the bin more accurately reflect the ideal power mapping relationship within the wind speed range.
[0064] This application's embodiments, by determining the upper power boundary, lower power boundary, and target power value, and correcting abnormal power data exceeding the reasonable fluctuation range within each data bin, can effectively eliminate the interference of abnormal operating conditions on the theoretical stress curve. This makes the generated theoretical stress curve more accurately reflect the inherent wind speed-power mapping relationship of the wind farm under normal operating conditions, providing a reliable theoretical benchmark for subsequent segmented predictions, thereby improving the accuracy of ultra-short-term wind power prediction.
[0065] In one embodiment of this application, the division of the prediction period into multiple prediction sub-intervals based on the theoretical stress curve and the short-term wind power prediction data includes: Obtain the prediction quality between the theoretical stress curve and the short-term wind power prediction data; Based on the prediction quality, the prediction period is divided into multiple consecutive prediction sub-intervals.
[0066] In this embodiment, prediction quality is used to measure the consistency between the theoretical stress curve and short-term wind power prediction data, as well as the reliability of the short-term wind power prediction data itself. Specifically, prediction quality can be quantified through two dimensions: the correlation between the theoretical stress curve and the short-term wind power prediction data, and the data accuracy of the short-term wind power prediction data itself.
[0067] It should be noted that the theoretical response curve is a continuous curve covering the entire wind speed range, reflecting the ideal mapping relationship between wind speed and power under normal operating conditions. Based on the actual wind speed at the current moment, the theoretical power value at the current moment can be interpolated on the theoretical response curve. Based on the predicted wind speed (derived from the wind speed prediction portion of the short-term wind power prediction data) within the future prediction period (including n prediction points), the theoretical response power value for each prediction point in the future can be interpolated on the theoretical response curve. Each prediction point represents a prediction data pair consisting of the predicted wind speed and the power value to be predicted, i.e., a predicted wind speed-power data pair. In other words, although the prediction point and the aforementioned data point are not directly related, they are indirectly related through the theoretical response curve. The data point belongs to the historical period and is used to construct the theoretical response curve, while the prediction point belongs to the future period, and its predicted wind speed is interpolated on the curve to obtain the theoretical response power value (the power value to be predicted).
[0068] Specifically, the correlation between the theoretical stress curve and short-term wind power forecast data over the entire forecast period can be calculated using the following Pearson correlation coefficient formula:
[0069] Where r represents the correlation between the theoretical stress curve and the short-term wind power forecast data, and x represents the sequence of theoretical stress at each forecast point (x1, x2, x3, ..., x...). n ), where y is a sequence of short-term wind power prediction data for each prediction point (y1, y2, y3, ..., y...). n Cov(x, y) is the covariance of x and y, Var(x) is the variance of x, Var(y) is the variance of y, and the correlation coefficient r ranges from [-1, 1]. The closer the value is to 1, the stronger the positive correlation.
[0070] The correlation between the theoretical stress curve and short-term wind power forecast data reflects the degree of agreement between the inherent power output characteristics of the wind farm represented by the theoretical stress curve and the future trend information reflected by the short-term wind power forecast data. A higher correlation indicates a greater consistency between the short-term wind power forecast data and the actual operating characteristics of the wind farm, and thus a higher reliability of the short-term wind power forecast data.
[0071] Specifically, the accuracy of the short-term wind power forecast data over the entire forecast period can be calculated using the following root mean square error formula:
[0072] Where z is the data accuracy of the short-term wind power prediction data itself within the entire prediction period, n is the number of prediction points within the entire prediction period, and y i For the short-term wind power prediction data at the i-th prediction point, Let Cap be the grid-connected power of the wind farm at the i-th prediction point, and Cap be the installed capacity of the wind farm, which is derived from the rated installed capacity parameter of the wind farm and is a basic design parameter of the wind farm.
[0073] The accuracy of short-term wind power forecast data over the entire forecast period reflects the degree of closeness between the short-term wind power forecast data and the actual grid-connected power of wind farms, and is an important indicator for evaluating the performance of short-term forecast models. Higher data accuracy indicates more reliable short-term wind power forecast data, and a lower risk of directly using short-term wind power forecast data for ultra-short-term forecasts.
[0074] Specifically, according to the "Technical Requirements for Dispatch-Side Wind Power or Photovoltaic Power Prediction System" (GB / T 40607-2021), an accuracy rate of 83% is considered compliant, and a rate higher than 83% is considered relatively high. A stronger correlation indicates a higher degree of statistical association between the two variables; a correlation value exceeding 0.8 is generally considered highly correlated. Therefore, a correlation greater than 0.8 can be defined as a high correlation between the theoretical wind power curve and short-term wind power prediction data, and a data accuracy rate greater than 83% can be defined as a relatively high short-term prediction accuracy. Under the premise of high accuracy and high correlation, it is believed that ultra-short-term predictions can be segmented and adjusted by referring to the latest short-term prediction trend and combining it with real-time power data. That is, the prediction period can be divided into multiple prediction sub-intervals, and different prediction methods can be used for segmented predictions in different sub-intervals.
[0075] This application embodiment obtains the correlation between the theoretical stress curve and short-term wind power prediction data, as well as the data accuracy of the short-term prediction data itself. When both the correlation and the data accuracy reach a preset threshold, the short-term wind power prediction data is determined to have high credibility. At this time, segmented prediction can be performed based on the trend information of the short-term wind power prediction data. By dividing the prediction period into multiple sub-intervals and adopting differentiated prediction strategies for different sub-intervals, the accuracy of ultra-short-term wind power prediction can be effectively improved under limited computing resources.
[0076] In one embodiment of this application, the prediction quality between the theoretical wind power curve and short-term wind power prediction data can be statistically analyzed monthly, and the boundary can be dynamically adjusted based on the latest historical data, so that the segmented prediction strategy can adapt to the wind power variation patterns under different seasons and meteorological conditions.
[0077] In one embodiment of this application, the step of segmenting and predicting the wind power of the wind farm during the prediction period according to each prediction sub-interval to obtain the wind power prediction result includes: Obtain the prediction accuracy for each prediction sub-interval; When the prediction accuracy of the prediction sub-interval meets the first preset condition, the wind power of the wind farm in the prediction sub-interval is predicted according to the ratio between theoretical power and short-term wind power prediction data, and the theoretical power prediction result is obtained. When the prediction accuracy of the prediction sub-interval meets the second preset condition, the short-term wind power prediction data is used as the short-term prediction result. When the prediction accuracy of the prediction sub-interval meets the third preset condition, the wind power of the wind farm in the prediction sub-interval is predicted according to the ratio between theoretical power and short-term wind power prediction data to obtain the theoretical power prediction result. The short-term wind power prediction data is used as the short-term prediction result. The theoretical power prediction result and the short-term prediction result are weighted and fused to obtain the combined prediction result. The wind power prediction result is obtained by combining the theoretical prediction result, the short-term prediction result, and the combined prediction result.
[0078] In this embodiment, the prediction accuracy is calculated separately for multiple prediction methods. Each prediction method (theoretical prediction method, short-term prediction method, and combined prediction method) corresponds to a prediction accuracy at each prediction point in the training set. Specifically, by comparing the prediction results of each prediction method with the grid-connected power of the wind farm, the historical performance of each prediction method at different prediction points can be quantitatively evaluated. The higher the prediction accuracy, the better the prediction effect of the prediction method at that prediction point.
[0079] To determine which prediction method should be used for each prediction sub-interval, it is necessary to compare the prediction accuracy of different methods at each prediction point. Specifically, on the training set, each prediction point is traversed, and the prediction accuracy of each of the three methods at that point is calculated. Then, the three accuracies are compared, and the prediction method with the highest accuracy is determined as the optimal prediction method for that prediction point. Subsequently, consecutive prediction points with the same optimal prediction method are merged into the same prediction sub-interval, with each sub-interval corresponding to one prediction method, thus completing the division of prediction sub-intervals and determining the prediction method corresponding to each sub-interval.
[0080] Specifically, when the prediction accuracy of the theoretical stress prediction method is higher than that of the short-term prediction method and the combined prediction method within a prediction sub-interval, the first preset condition is met, and the theoretical stress prediction method is used for prediction in that prediction sub-interval. The theoretical stress prediction method is calculated by obtaining the theoretical power value at the current moment and the short-term wind power prediction data at the current moment, calculating the ratio between the two, and multiplying this ratio by the short-term wind power prediction data for each prediction point within the prediction sub-interval to obtain the theoretical stress prediction result for each prediction point within the prediction sub-interval. Here, the theoretical power value is the power value interpolated onto the theoretical stress curve based on the wind speed at the prediction point.
[0081] When the prediction accuracy of the short-term forecasting method within a given prediction sub-interval is higher than that of the theoretical wind power prediction method and the combined prediction method, the second preset condition is met, and the short-term forecasting method is used for prediction in that prediction sub-interval. The short-term forecasting method is calculated by directly using the short-term wind power prediction data of each prediction point within the prediction sub-interval as the prediction result, without the need for additional calculations.
[0082] When the prediction accuracy of the combined prediction method within a given prediction sub-interval is higher than that of the theoretical payoff prediction method and the short-term prediction method, the third preset condition is met, and the combined prediction method is used for prediction in that prediction sub-interval. The calculation method of the combined prediction method is to weight and fuse the theoretical payoff prediction results and the short-term prediction results for each prediction point within the prediction sub-interval, that is, the combined prediction result = a × theoretical payoff prediction result + b × short-term prediction result, where a and b are weighting coefficients, and a + b = 1.
[0083] To obtain the optimal weighting coefficients, this embodiment employs an iterative search method to optimize a and b. Specifically, on the training set, the range of a is set to [0, 1], and the traversal is performed with a step size of 0.1, i.e., a = 0, 0.1, 0.2, ..., 0.9, 1.0, with corresponding b = 1 - a. For each set of (a, b) values, the root mean square accuracy of the combined prediction method at each prediction point is calculated, and the set of (a, b) that yields the highest root mean square accuracy is selected as the optimal weighting coefficient for the current training set. The weighting coefficients can be dynamically updated monthly. The historical data of the previous month is used as the training set, and the above iterative search process is re-executed to obtain the optimal weighting coefficients for the next month. Furthermore, the weighting coefficients can be further refined and optimized by combining different coefficients or linear combinations of coefficient iterations, or by forward or backward moving average methods, thereby further improving prediction accuracy while maintaining existing computational speed.
[0084] After completing the predictions for each prediction sub-interval, the prediction results within each sub-interval are concatenated in chronological order. Specifically, assuming the prediction period contains M prediction points, the above division results in K consecutive prediction sub-intervals, each covering several consecutive prediction points. The prediction results of the first sub-interval are arranged in chronological order, and then the second, third, and so on, until the prediction results of the Kth sub-interval are concatenated, ultimately forming a complete prediction sequence covering the entire prediction period. This sequence is the final wind power prediction result.
[0085] For example, assuming the forecast period is 4 hours and there are 16 forecast points, the forecast is divided into three sub-intervals. Points 1-4 use the theoretical forecast method, points 5-11 use the combined forecast method, and points 12-16 use the short-term forecast method. Then, by sequentially splicing the theoretical forecast results of points 1-4, the combined forecast results of points 5-11, and the short-term forecast results of points 12-16, the wind power forecast results for the 16 points in the next 4 hours can be obtained.
[0086] It should be noted that although the above calculation process involves iteratively calculating the accuracy of the three prediction methods at each prediction point, this calculation is performed offline using historical training data. Once the segmentation strategy is determined, there is no need to repeat the calculation in the actual ultra-short-term prediction process; prediction can be performed directly according to the determined segmentation strategy. The computational overhead in the online prediction phase is extremely low, which can meet the calculation cycle constraint of reporting once every 15 minutes. At the same time, the computational load itself is relatively small. The prediction period is usually 16 prediction points in the next 4 hours. Each prediction method only needs to perform 16 prediction accuracy calculations, for a total of 48 calculations for the three methods. Each calculation only involves simple arithmetic operations (such as multiplication, division, summation, square root, etc.) and does not involve complex iterative training or large-scale matrix operations. Even if the segmentation strategy is dynamically updated once a month, the computational load is within an acceptable range. In addition, since the calculations at each prediction point are independent of each other, batch calculation or parallel processing can be used to further shorten the calculation time.
[0087] Unlike existing technologies that directly use the latest short-term wind power forecast data or a single prediction model with a single bias correction model, this application's embodiments calculate the prediction accuracy of multiple prediction methods at each prediction point. Based on the accuracy comparison results, the optimal prediction method is selected for each prediction sub-interval. This enables a differentiated segmented prediction strategy with limited computing resources, effectively avoiding the accumulation of prediction biases of a single prediction method across all time periods. It ensures that the most suitable prediction method can be used for wind power periods with different fluctuation characteristics, thereby significantly improving the overall accuracy of ultra-short-term wind power prediction. Simultaneously, offline calculation and the pre-determining of the segmented strategy guarantee the real-time requirements of the online prediction phase.
[0088] Based on the partitioning results determined in the training set, the prediction accuracy is verified using the test set. Specifically, the segmentation strategy (i.e., the partitioning boundaries of each prediction sub-interval and the prediction method corresponding to each prediction sub-interval), determined by comparing the prediction accuracy of the three prediction methods at each prediction point in the training set, is applied to the test set for prediction. The prediction results are then compared with the actual on-grid power of wind farms in the test set to calculate the overall prediction accuracy of the segmented prediction method, thereby verifying the generalization ability and effectiveness of the segmentation strategy.
[0089] In one embodiment, the segmented prediction method uses the theoretical prediction method for the first 4 points (prediction points 1-4), the combined prediction method for the 5th to 11th points (prediction points 5-11), and the short-term prediction method for the 12th to 16th points (prediction points 12-16). The specific prediction results are shown in Table 1 below.
[0090] Table 1
[0091] As shown in Table 1 above, the average accuracy of the segmented forecasting method is 87.3%, which is higher than that of the single theoretical event forecasting method (83.8%), the single combined forecasting method (86.7%), and the single short-term forecasting method (86.2%). This indicates that the segmented forecasting method proposed in this application effectively integrates the advantages of each individual forecasting method by using differentiated forecasting methods in different forecasting periods, avoiding the accumulation of forecasting biases of individual forecasting methods across all periods, thus achieving a better overall forecasting effect than any single forecasting method.
[0092] The technical solutions provided in this application are not limited to wind farms, but can also be extended to new energy power plants such as photovoltaic power plants, providing stable segmentation standards and segmentation prediction methods for different types of new energy power generation, and their application locations are not limited.
[0093] It should be noted that, for the sake of simplicity, the method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments of this application are not limited to the described order of actions, because according to the embodiments of this application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of this application.
[0094] This application also provides a storage medium that stores computer instructions. When the computer executes the computer instructions, it is used to execute the various processes of the above-described wind farm wind power prediction method embodiments and can achieve the same technical effect. To avoid repetition, it will not be described again here.
[0095] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.
[0096] This application also provides an electronic device, including a processor 2010, a memory 209, and a program or instructions stored in the memory 209 and executable on the processor 2010. When the program or instructions are executed by the processor 2010, they implement the various processes of the above-described wind farm wind power prediction method embodiments and achieve the same technical effects. To avoid repetition, they will not be described again here.
[0097] It should be noted that the electronic devices in the embodiments of this application include the mobile electronic devices and non-mobile electronic devices described above.
[0098] Figure 2 A schematic diagram of the hardware structure of an electronic device to implement an embodiment of this application. The electronic device 200 includes, but is not limited to, components such as: radio frequency unit 201, network module 202, audio output unit 203, input unit 204, sensor 205, display unit 206, user input unit 207, interface unit 208, memory 209, and processor 2010.
[0099] Those skilled in the art will understand that the electronic device 200 may also include a power supply (such as a battery) for supplying power to various components. The power supply may be logically connected to the processor 2010 through a power management system, thereby enabling functions such as managing charging, discharging, and power consumption through the power management system. Figure 2 The electronic device structure shown does not constitute a limitation on the electronic device. The electronic device may include more or fewer components than shown, or combine certain components, or have different component arrangements, which will not be elaborated here. This application also provides a computer program product, including a computer program / instruction. When the computer program / instruction is executed by a processor, it implements the various processes of the above-described wind farm wind power prediction method embodiment and can achieve the same technical effect. To avoid repetition, it will not be described again here.
[0100] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.
[0101] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.
[0102] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.
Claims
1. A method for predicting wind power in a wind farm, characterized in that, include: Obtain data on grid-connected wind power, short-term wind power forecasts, and average wind speed of wind turbines; The grid-connected power of the wind farm and the average wind speed of the wind turbine are divided into boxes according to the preset wind speed interval, and density cluster analysis is performed on the data in each box to obtain the theoretical stress curve. Based on the theoretical stress curve and the short-term wind power prediction data, the prediction period is divided into multiple prediction sub-intervals; The wind power of the wind farm during the prediction period is predicted in segments according to each prediction sub-interval, and the wind power prediction result is obtained.
2. The method according to claim 1, characterized in that, The step of dividing the grid-connected power of the wind farm and the average wind speed of the wind turbine into boxes according to a preset wind speed interval includes: The average wind speed of the wind turbine is divided into multiple wind speed intervals according to a preset wind speed interval. The grid-connected power of the wind farm is divided into its respective wind speed range according to the wind speed range to which the average wind speed of the corresponding wind turbine belongs, resulting in multiple data bins corresponding to the wind speed ranges. Each data bin contains wind speed data belonging to the same wind speed range and power data corresponding to the wind speed data. The wind speed data is the average wind speed of the wind turbine, and the power data is the grid-connected power of the wind farm.
3. The method according to claim 2, characterized in that, The density clustering analysis performed on the data within each bin to obtain the theoretical stress curve includes: Density clustering analysis was performed on the wind speed data and the power data corresponding to the wind speed data in each data bin to obtain the sub-curve segment corresponding to each data bin. By splicing together the sub-curve segments corresponding to all data bins in ascending order of wind speed range, the theoretical stress curve is obtained.
4. The method according to claim 3, characterized in that, The density clustering analysis is performed on the wind speed data and the corresponding power data within each data bin to obtain the sub-curve segment corresponding to each data bin, including: Density clustering analysis was performed on each data bin to obtain multiple clusters within each data bin; For each data bin, the cluster with the most data points among all clusters in the data bin is selected as the master cluster. Based on the wind speed data contained in the master cluster and the power data corresponding to the wind speed data, a sub-curve segment corresponding to each data bin is generated.
5. The method according to claim 4, characterized in that, The cluster with the most data points among all clusters within the selected data bin is designated as the master cluster. Based on the wind speed data and the corresponding power data contained in the master cluster, sub-curve segments are generated for each data bin, including: The cluster with the most data points among all clusters in the data bin is selected as the master cluster. The upper power boundary, lower power boundary, and target power value are determined based on the power data in the master cluster. The power data in each data bin that are higher than the upper power boundary or lower than the lower power boundary are corrected to the target power value, thus obtaining the sub-curve segment corresponding to each data bin.
6. The method according to claim 1, characterized in that, The prediction period is divided into multiple prediction sub-intervals based on the theoretical stress curve and the short-term wind power prediction data, including: Obtain the prediction quality between the theoretical stress curve and the short-term wind power prediction data; Based on the prediction quality, the prediction period is divided into multiple consecutive prediction sub-intervals.
7. The method according to claim 1, characterized in that, The step of segmenting and predicting the wind power of the wind farm during the prediction period according to each prediction sub-interval to obtain the wind power prediction result includes: Obtain the prediction accuracy for each prediction sub-interval; When the prediction accuracy of the prediction sub-interval meets the first preset condition, the wind power of the wind farm in the prediction sub-interval is predicted according to the ratio between theoretical power and short-term wind power prediction data, and the theoretical power prediction result is obtained. When the prediction accuracy of the prediction sub-interval meets the second preset condition, the short-term wind power prediction data is used as the short-term prediction result. When the prediction accuracy of the prediction sub-interval meets the third preset condition, the wind power of the wind farm in the prediction sub-interval is predicted according to the ratio between theoretical power and short-term wind power prediction data to obtain the theoretical power prediction result. The short-term wind power prediction data is used as the short-term prediction result. The theoretical power prediction result and the short-term prediction result are weighted and fused to obtain the combined prediction result. The wind power prediction result is obtained by combining the theoretical prediction result, the short-term prediction result, and the combined prediction result.
8. A storage medium, characterized in that, The storage medium stores computer instructions, which, when executed by the computer, are used to perform a wind farm wind power prediction method as described in any one of claims 1-7.
9. An electronic device, characterized in that, Includes at least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform a wind farm wind power prediction method as described in any one of claims 1-7.
10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instruction is executed by the processor, it implements a wind power prediction method for a wind farm as described in any one of claims 1-7.