A Wind Power Prediction Method and System Based on Small-Sample Data

Abstract

The present application discloses a wind power prediction method and system based on small sample data, relating to wind power and data processing technologies, including: determining the power generation information of the wind turbines corresponding to each sub-region according to the historical power generation data of the sample wind farm, and dividing the sample wind farm into multiple sub-regions according to the terrain parameters; extracting the terrain features of the sub-regions; clustering the terrain categories; selecting a sub-region sample set based on the relationship between the distance between each sub-region and the clustering center according to each clustering center; for each clustering center, training an LSTM model respectively based on the terrain features of the sub-region sample set and the power generation information of the corresponding wind turbines in different time periods; for the target wind farm to be predicted, selecting the LSTM model trained by the corresponding clustering category to perform the power generation prediction of the target wind farm. The present application constructs a classification in combination with factors such as the spatial position and geographical conditions of the wind farm to achieve the prediction of the wind power of the wind farm under the condition of small samples.

Description

Technical Field

[0001] The present application relates to the technical fields of wind power and data processing, and particularly relates to a wind power prediction method and system based on small sample data. Background Art

[0002] In recent years, the construction of wind farms has changed from decentralized and small-scale to centralized and large-scale, forming regional wind power clusters. Due to the intermittency, randomness, and volatility of wind power, regional wind power clusters have a greater impact on the peak shaving capacity and frequency stability of the power system. Therefore, the power prediction of a single wind farm can no longer meet the needs of power dispatching departments. And as wind power increasingly participates in power market transactions, more and more research has been conducted on the power prediction of regional wind power clusters. The power prediction of wind power clusters is different from that of wind turbines and power stations. The cluster power prediction needs to comprehensively analyze the information of multiple wind farms from both time and space perspectives, not only paying attention to the accuracy of deterministic prediction, but also paying more attention to the accuracy of the overall prediction trend. At the same time, a wind power cluster usually contains dozens or even hundreds of wind farms, and the huge data scale increases the difficulty of establishing a cluster power prediction model.

[0003] The methods for wind power cluster power prediction are generally divided into the summation method, the upscaling method, and the spatial resource matching method. The summation method is the earliest used method, but it has obvious limitations. This method needs to establish a separate power prediction model for each wind farm. Therefore, when the regional scale is large, data storage and model training consume huge resources. At the same time, the separate modeling of each wind farm has more stringent requirements for data quality. The upscaling method is generally divided into the physical upscaling method and the statistical upscaling method, among which the statistical upscaling method is more widely used. The statistical upscaling method needs to select representative wind farms in the region and obtain the output of the wind power cluster according to the prediction results of the representative wind farms. The advantage of the statistical upscaling method is that the selection of representative wind farms makes the prediction process not overly dependent on the data of each single field, and the factors with low correlation between fields are effectively ignored, and the dynamic adaptation ability is better. The spatial resource matching method has been more popular in recent years. This method requires less computing resources, but it is difficult to optimize model parameters, and the wind power sample data is affected by many factors. Therefore, it is difficult to construct large sample data. Summary of the Invention

[0004] Embodiments of the present application provide a wind power prediction method and system based on small sample data, which are used to construct classifications by combining factors such as the spatial location and geographical conditions of wind farms, and perform model training separately to realize the prediction of wind farm wind power under small sample conditions.

[0005] Embodiments of the present application propose a wind power prediction method based on small sample data, including:

[0006] Obtain the topographic parameters, historical power generation data, and historical wind direction data of the sample wind farm;

[0007] Determine the power generation information of the corresponding wind turbines in each sub-region according to the historical power generation data of the sample wind farm, and divide the sample wind farm into multiple sub-regions according to the topographic parameters, where at least one wind turbine is included in any one sub-region;

[0008] Based on each divided sub-region, extract the topographic features of the sub-region;

[0009] Cluster the extracted topographic features to obtain multiple clustered topographic categories;

[0010] Based on each cluster center and according to the relationship between the distance between each sub-region and the cluster center, select a specified number of sub-regions from each sub-region as the sub-region sample set, where for any one cluster center, select the corresponding number of sub-regions based on the size of the cluster radius to form the sub-region sample set;

[0011] For each cluster center, train the LSTM model respectively based on the topographic features of the sub-region sample set and the power generation information of the corresponding wind turbines in different time periods;

[0012] For the target wind farm to be predicted, divide the target wind farm into multiple sub-wind farms, and calculate the topographic features of the sub-wind farms;

[0013] Calculate the matching degree between the topographic features of the sub-wind farm and the topographic features of the sub-region of the sample wind farm;

[0014] According to the calculated matching degree and the weather forecast data, select the LSTM model trained by the corresponding clustering category to perform the power generation prediction of the target wind farm.

[0015] Optionally, dividing the sample wind farm into multiple sub-regions according to the topographic parameters includes:

[0016] Determine the hillside and valley regions according to the topographic parameters;

[0017] Use the determined hillside and valley regions as part of the boundaries to divide the topographic parameters, and determine the distribution positions of the wind turbines in the sample wind farm;

[0018] Discard the regions that do not contain wind turbines within the partial boundaries formed by the hillside and valley regions, and divide the remaining regions into multiple sub-regions according to the specified area range of topographic division in combination with the partial boundaries.

[0019] Optionally, based on each divided sub-region, extracting the topographic features of the sub-region includes:

[0020] Configure the topographic undulation deviation for each sub-region range according to the topographic parameters, and extract the topographic features according to the topographic undulation deviation.

[0021] Optionally, configure the terrain undulation deviation for each sub-region range according to the terrain parameters, and extract terrain features according to the terrain undulation deviation, including:

[0022] Determine the position of the wind turbine in any sub-region;

[0023] According to the position of the wind turbine and the configured terrain undulation deviation, combine the terrain parameters to determine multiple paths where the terrain undulation deviation exceeds the specified deviation threshold, and any one of the paths passes through the position of the wind turbine;

[0024] Extract corresponding terrain features based on each path.

[0025] Optionally, clustering the extracted terrain features is achieved through the K-means clustering algorithm.

[0026] Optionally, based on each clustering center, select a specified number of sub-regions from each sub-region as the sub-region sample set according to the relationship between the sub-region and the clustering center, including:

[0027] For any clustering category, judge the maximum clustering radius based on the clustering center;

[0028] Based on the maximum clustering radius, divide the terrain features in any clustering category into multiple levels;

[0029] Select a specified number of sub-regions from the innermost level to the outermost level incrementally as the sub-region sample set.

[0030] Optionally, when the sample quantity is less than the preset threshold, it further includes:

[0031] Determine the wind direction distribution of the sample wind field according to the historical wind direction data;

[0032] Based on the determined wind direction distribution, determine the wind direction in different time periods for each sub-region in the sub-region sample set;

[0033] Taking the wind direction in different time periods as the reference direction, fit the terrain inverse parameters of the reference direction according to the terrain parameters, so as to amplify the sub-region sample set based on the wind direction in different time periods and the terrain inverse parameters.

[0034] Optionally, for each clustering center, train the LSTM model respectively based on the terrain features of the sub-region sample set and the power generation information of the corresponding wind turbine in different time periods, including:

[0035] Add labels to the terrain features of the sub-region sample set and the power generation information of the corresponding wind turbine in different time periods according to the historical wind direction data to train the LSTM model;

[0036] During the prediction process, marks are added to the terrain features of the sub-wind field according to the wind direction data in the weather forecast data, so as to perform the prediction by using the trained LSTM model.

[0037] An embodiment of the present application also proposes a wind power prediction system based on small sample data, including a processor and a memory. A computer program is stored on the memory, and when the computer program is executed by the processor, the steps of the wind power prediction method based on small sample data as described above are implemented.

[0038] The method of the embodiment of the present application constructs classifications in combination with factors such as the spatial location and geographical conditions of the wind farm, and performs model training separately to realize the prediction of the wind power of the wind farm under small sample conditions, and is applicable to various types of wind farms.

[0039] The above description is only an overview of the technical solution of the present application. In order to be able to understand the technical means of the present application more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features and advantages of the present application more obvious and understandable, the following specifically illustrates the specific embodiments of the present invention. Description of the Drawings

[0040] By reading the following detailed description of the preferred embodiments, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present invention. Moreover, throughout the drawings, the same reference numerals are used to represent the same components. In the drawings:

[0041] Figure 1 It is a schematic diagram of the basic process of the wind power prediction method based on small sample data in this embodiment. Detailed Embodiments

[0042] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be fully conveyed to those skilled in the art.

[0043] There is a certain output space correlation characteristic between wind farms. The spatial location and geographical conditions are important factors affecting the output correlation of wind farms. An embodiment of the present application proposes a wind power prediction method based on small sample data, as Figure 1 shown, including the following steps:

[0044] In step S101, obtain the terrain parameters, historical power generation data, and historical wind direction data of the sample wind farm. In some embodiments, a corresponding terrain model may also be established based on the terrain data of the sample wind farm to determine the terrain parameters of the sample wind farm. The historical power generation data and historical wind direction data can be obtained according to the operation logs and historical meteorological data.

[0045] In step S102, determine the power generation information of the wind turbines corresponding to each sub-region according to the historical power generation data of the sample wind farm, and divide the sample wind farm into multiple sub-regions according to the terrain parameters, where at least one wind turbine is included in any one sub-region. In some embodiments, dividing the sample wind farm into multiple sub-regions according to the terrain parameters includes:

[0046] Determine the hillside and valley regions according to the terrain parameters. In a specific example, the hillside and valley regions can be determined from the regions where the terrain undulation is greater than a preset threshold according to the aforementioned terrain model.

[0047] Divide the terrain parameters based on the determined hillside and valley regions as partial boundaries, and determine the distribution positions of the wind turbines in the sample wind farm.

[0048] Discard the regions that do not contain wind turbines within the partial boundaries formed by the hillside and valley regions, and divide the remaining regions into multiple sub-regions by combining the partial boundaries according to the specified area range of the terrain division. In some embodiments, particularly in general cases, wind turbines are not installed in valleys. Therefore, the valley region can be used as a partial boundary, and then further discard the hillside and valley regions, especially the regions that do not contain wind turbines within the partial boundaries formed by the valley region. For other terrain regions, divide them into multiple sub-regions according to the specified area range.

[0049] In step S103, based on each divided sub-region, extract the terrain features of the sub-region.

[0050] In step S104, cluster the extracted terrain features to obtain multiple clustered terrain categories. In some embodiments, clustering the extracted terrain features is achieved through the K-means clustering algorithm.

[0051] In step S105, based on each cluster center and according to the relationship between the distance between each sub-region and the cluster center, a specified number of sub-regions are selected from each sub-region as the sub-region sample set. For any cluster center, the corresponding number of sub-regions is selected based on the size of the cluster radius. For example, for any cluster category, the cluster radius is divided into multiple segments, and sub-regions are selected from within the circular ring of each segment radius and added to the region sample set, thereby avoiding the problem that most samples are concentrated at the cluster center and losing representativeness, improving the representativeness of the selected samples in the sample set. On the other hand, the number of samples in the sample set can be reduced to adapt to the small sample situation.

[0052] In step S106, for each cluster center, an LSTM model is trained respectively based on the terrain features of the sub-region sample set and the power generation information of the corresponding wind turbine in different time periods. In some examples, the extracted terrain features can be marked using historical meteorological data in different time periods, and the predicted power generation information is output. A loss function is constructed by comparing the predicted power generation information with the power generation information in the corresponding historical time periods, thereby training the LSTM. Training an LSTM model for each cluster can greatly improve the prediction accuracy.

[0053] In step S107, for the target wind farm to be predicted, the target wind farm is divided into multiple sub-wind farms, and the terrain features of the sub-wind farms are calculated.

[0054] In step S108, the matching degree between the terrain features of the sub-wind farm and the terrain features of the sample wind farm sub-regions is calculated. In a specific example, the relationship between the terrain features of the sub-wind farm and the cluster center can be calculated, so as to classify the terrain features of the sub-wind farm into the corresponding categories.

[0055] In step S109, according to the calculated matching degree and the weather forecast data, the LSTM model trained for the corresponding cluster category is selected to perform the power generation prediction of the target wind farm.

[0056] The method of the embodiment of the present application constructs classifications by combining factors such as the spatial location and geographical conditions of the wind farm, and performs model training respectively to realize the prediction of the wind power of the wind farm under small sample conditions. Through the method of the present application, only a few terrain categories need to be clustered to combine meteorological data to realize the power generation prediction of the wind farm. The method of the present application is applicable to various types of wind farms.

[0057] In some embodiments, based on each segmented sub-region, extracting the terrain features of the sub-region includes: configuring the terrain undulation deviation for the range of each sub-region according to the terrain parameters, and extracting the terrain features according to the terrain undulation deviation. In a specific example, for the divided sub-regions, for example, the terrain undulation deviation of each sub-region is configured based on the terrain model, and then feature extraction is performed.

[0058] In some embodiments, configuring a terrain undulation deviation for each sub-region range according to the terrain parameters, and extracting terrain features according to the terrain undulation deviation includes:

[0059] Determine the position of the wind turbine in any sub-region. In a specific example, in the foregoing example, the geographical location of the wind farm is considered when extracting terrain features, and the spatial position of the wind turbine is further determined for each sub-region.

[0060] According to the position of the wind turbine and the configured terrain undulation deviation, and in combination with the terrain parameters, determine multiple paths where the terrain undulation deviation exceeds a specified deviation threshold, and any one of the paths passes through the position of the wind turbine. In a specific example, for sub-regions in the sample with terrain undulations, according to the position of the wind turbine and the terrain undulation relationship, determine the terrain paths, so as to associate the spatial position of the wind turbine with the terrain it is in. For flat regions, directly configure the position of the wind turbine according to the representative sub-region to extract features.

[0061] Extract corresponding terrain features based on each path and the terrain undulation deviation of the sub-region.

[0062] In some embodiments, selecting a specified number of sub-regions from each sub-region as a sub-region sample set based on the relationship between each clustering center and the distance between the sub-region and the clustering center includes:

[0063] For any clustering category, judge the maximum clustering radius based on the clustering center.

[0064] Based on the maximum clustering radius, divide the terrain features in any clustering category into multiple levels, that is, segment according to the maximum clustering radius.

[0065] Select a specified number of sub-regions from the innermost level to the outermost level incrementally as the sub-region sample set. In a specific example, by segmenting according to the maximum radius and selecting incrementally, on the one hand, the sample size can be further reduced. On the other hand, since the deviation of the sample closest to the clustering center is small, and the samples farther from the clustering center should be more worthy of attention in training. The method of this application can effectively improve the representativeness of the selected samples and the prediction effect of the trained model by selecting a specified number of sub-regions incrementally from the innermost level to the outermost level.

[0066] In some embodiments, when the sample quantity is less than a preset threshold, it further includes:

[0067] Determine the wind direction distribution of the sample wind farm according to historical wind direction data;

[0068] Based on the determined wind direction distribution, determine the wind direction in different time periods of each sub-region in the sub-region sample set;

[0069] Taking the wind direction in different time periods as the reference direction, fitting the terrain inverse parameter of the reference direction according to the terrain parameters, and amplifying the sub-region sample set based on the wind direction in different time periods and the terrain inverse parameter. For example, in the case where the overall selected sample quantity is insufficient, in the embodiments of the present application, the terrain inverse parameter is further fitted based on the historical wind direction in different time periods according to the terrain parameters. For example, the deviation of the sub-region in the reference direction is adjusted in the terrain model according to the terrain undulation deviation, so as to obtain the terrain inverse parameter. And the power output of the wind turbine based on the forward wind direction or the reverse wind direction can be considered to be approximately the same, thereby expanding the sub-region sample set.

[0070] In some embodiments, for each clustering center, training the LSTM model based on the terrain features of the sub-region sample set and the power generation information of the corresponding wind turbine in different time periods includes:

[0071] Adding labels to the terrain features of the sub-region sample set and the power generation information of the corresponding wind turbine in different time periods according to the historical wind direction data to train the LSTM model;

[0072] During the prediction process, adding labels to the terrain features of the sub-wind farm according to the wind direction data in the weather forecast data to perform prediction by using the trained LSTM model.

[0073] In some examples, calculating the matching degree between the terrain features of the sub-wind farm and the terrain features of the sample wind farm sub-region can be to match the terrain features of the sub-wind farm to the corresponding clustering category, so as to use the corresponding LSTM model to perform wind power prediction.

[0074] On the one hand, through the reasonable sample selection method of the method of the present application, the prediction accuracy of the model is improved. On the other hand, through the expansion of the samples, the sample quantity can be further reduced, and the method of the present application has universal applicability.

[0075] The embodiments of the present application also propose a wind power prediction system based on small sample data, including a processor and a memory. A computer program is stored on the memory, and when the computer program is executed by the processor, the steps of the wind power prediction method based on small sample data as described above are implemented.

[0076] In addition, although the exemplary embodiments have been described herein, their scope includes any and all embodiments based on the present disclosure having equivalent elements, modifications, omissions, combinations (e.g., schemes of cross-combination of various embodiments), adaptations or changes. It is not limited to the examples described in this specification or during the implementation of the present application, and the examples will be interpreted as non-exclusive.

[0077] The above description is intended to be illustrative rather than restrictive. For example, the above examples (or one or more aspects thereof) may be used in combination with each other. For example, those of ordinary skill in the art may use other embodiments when reading the above description.

[0078] The above embodiments are only exemplary embodiments of the present disclosure. Those skilled in the art can make various modifications or equivalent substitutions to the present invention within the essence and scope of protection of the present disclosure, and such modifications or equivalent substitutions should also be regarded as falling within the scope of protection of the present invention.

Claims

1. A wind power prediction method based on small sample data, characterized in that: include: Obtain terrain parameters, historical power generation data, and historical wind direction data of the sample wind farm; Determine the power generation information of the wind turbines corresponding to each sub-area according to the historical power generation data of the sample wind farm, and divide the sample wind farm into a plurality of sub-areas according to the terrain parameters, wherein any sub-area contains at least one wind turbine; Based on the segmented sub-regions, extracting the terrain features of the sub-regions; Clustering the extracted terrain features to obtain multiple clustered terrain categories; Based on the distance relationship between each sub-region and the cluster center, a specified number of sub-regions are selected from each sub-region as a sub-region sample set, wherein for any cluster center, a corresponding number of sub-regions are selected based on the size of the cluster radius to form the sub-region sample set; For each cluster center, the LSTM model is trained based on the terrain characteristics of the sub-region sample set and the power generation information of the corresponding wind turbine in different time periods; For the target wind field to be predicted, the target wind field is divided into multiple sub-wind fields, and the terrain characteristics of the sub-wind fields are calculated; The matching degree between the terrain characteristics of the sub-wind farm and the terrain characteristics of the sub-region of the sample wind farm is calculated according to the terrain characteristics of the sub-wind farm; According to the calculated matching degree and weather forecast data, the LSTM model trained by the corresponding cluster category is selected to perform power generation prediction of the target wind farm; Configuring a terrain relief deviation for each sub-region according to the terrain parameters, and extracting terrain features according to the terrain relief deviation includes: Determine the location of the fans in any sub-area; According to the wind turbine position and the configured terrain undulation deviation, a plurality of paths whose terrain undulation deviation exceeds a specified deviation threshold are determined in combination with the terrain parameters, any one of which passes through the wind turbine position; Corresponding terrain features are extracted based on each path.

2. The wind power prediction method based on small sample data according to claim 1, characterized in that: Dividing the sample wind field into a plurality of sub-areas according to the terrain parameters comprises: Determining hillside and valley areas based on the terrain parameters; Dividing the terrain parameters according to the determined hillside and valley areas as partial boundaries, and determining the wind turbine distribution locations of the sample wind farm; The area that does not contain wind turbines within the partial boundary formed by the hillside and valley area is discarded, and the remaining area is divided into multiple sub-areas according to the area range of the specified terrain segmentation combined with the partial boundary segmentation.

3. The wind power prediction method based on small sample data according to claim 2, characterized in that: Based on the segmented sub-regions, the terrain features of the sub-regions are extracted, including: A terrain relief deviation is configured for each sub-region according to the terrain parameters, and terrain features are extracted according to the terrain relief deviation.

4. The method for predicting wind power based on small sample data according to claim 3, characterized in that: Clustering of the extracted terrain features is achieved through the K-means clustering algorithm.

5. The method for predicting wind power based on small sample data according to claim 1, characterized in that: Based on the distance relationship between each sub-region and the cluster center, a specified number of sub-regions are selected from each sub-region as the sub-region sample set including: For any cluster category, the maximum cluster radius is determined based on the cluster center; Based on the maximum cluster radius, the terrain features in any cluster category are divided into multiple levels; A specified number of sub-regions are selected step by step from the innermost level to the outermost level as the sub-region sample set.

6. The wind power prediction method based on small sample data according to claim 1, characterized in that: When the number of samples is less than the preset threshold, it also includes: Determining the wind direction distribution of the sample wind field according to historical wind direction data; Determine the time-division wind direction of each sub-region in the sub-region sample set based on the determined wind direction distribution; The time-divided wind direction is used as a reference direction, and a terrain inverse parameter of the reference direction is fitted according to the terrain parameters, so as to expand the sub-region sample set based on the time-divided wind direction and the terrain inverse parameter.

7. The method for predicting wind power based on small sample data according to claim 1, characterized in that: For each cluster center, the LSTM model is trained based on the terrain characteristics of the sub-region sample set and the power generation information of the corresponding wind turbine in different time periods, including: According to the historical wind direction data, the terrain features of the sub-region sample set and the power generation information of the corresponding wind turbines in different time periods are marked to train the LSTM model; During the prediction process, the terrain features of the sub-wind farm are labeled according to the wind direction data in the weather forecast data to perform prediction using the trained LSTM model.

8. A wind power prediction system based on small sample data, characterized in that: The method comprises a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the steps of the method for predicting wind power based on small sample data as described in any one of claims 1 to 7 are implemented.

Images