Training method of wind speed correction model, wind speed correction method and device of wind farm

Abstract

The application discloses a wind speed correction model training method, a wind speed correction method and device of a wind farm. The training method comprises the following steps: obtaining training samples, wherein the training samples comprise historical meteorological data and historical wind speed data of a target wind turbine point, and the target wind turbine point is any one of a plurality of wind turbine points of the wind farm; preprocessing the training samples to obtain a first data sequence; inputting the first data sequence into a plurality of preset wind speed correction models respectively, wherein each preset wind speed correction model has a plurality of preset parameter combinations; obtaining a first predicted wind speed sequence output by each preset wind speed correction model under each preset parameter combination; performing performance evaluation on each preset parameter combination of each preset wind speed correction model based on the first predicted wind speed sequence; and determining a preset wind speed correction model corresponding to a preset parameter combination meeting a preset requirement as a wind speed correction model of the wind turbine point.

Description

Technical Field

[0001] This disclosure generally relates to the field of wind power generation technology, and more specifically, to a training method for a wind speed correction model for the area where each wind turbine is located in a wind farm, a wind speed correction method for a wind farm, and an apparatus. Background Technology

[0002] Because strong winds can cause quality risks such as tower sweeping and blade damage to wind turbines in operation, wind farms at risk of strong winds need to issue accurate wind warnings so that on-site personnel can be notified in advance to take appropriate measures to avoid risks when a strong wind warning is issued.

[0003] However, most related technologies correct all wind speeds as a whole, without treating strong winds specially. Since the distribution characteristics of strong winds and all wind speeds are not exactly the same, correcting all wind speeds as a whole, while reducing the overall systematic bias, does not improve the correction effect for strong winds. Summary of the Invention

[0004] This disclosure provides a training method for a wind speed correction model, a wind speed correction method and apparatus for a wind farm, thereby enabling targeted training of the model or targeted correction of wind speed at various wind turbine locations in a wind farm.

[0005] In one general aspect, a training method for a wind speed correction model is provided. The training method includes: acquiring training samples, the training samples including historical meteorological data and historical wind speed data of a target wind turbine location, the target wind turbine location being any one of multiple wind turbine locations in a wind farm; preprocessing the training samples to obtain a first data sequence; inputting the first data sequence into multiple preset wind speed correction models, wherein each preset wind speed correction model has multiple preset parameter combinations; acquiring a first predicted wind speed sequence output by each preset wind speed correction model under each preset parameter combination; performing performance evaluation on each preset parameter combination of each preset wind speed correction model based on the first predicted wind speed sequence; and determining the preset wind speed correction model corresponding to the preset parameter combination whose performance evaluation results meet preset requirements as the wind speed correction model for that wind turbine location.

[0006] Optionally, the step of preprocessing the training samples to obtain the first data sequence includes: performing a first data aggregation on the historical wind speed data to aggregate the instantaneous data in the historical wind speed data into hourly data to obtain a first wind speed sequence; concatenating the first wind speed sequence with data of the same time in the historical meteorological data to obtain a second data sequence; performing a second data aggregation on the data of any hour in the second data sequence and the data of a first number of hours prior to that hour; and obtaining the first data sequence based on the second data aggregation data for each hour.

[0007] Optionally, the step of aggregating instantaneous data from the historical wind speed data into hourly data by performing a first data aggregation on the historical wind speed data to obtain a first wind speed sequence includes: performing data cleaning on the historical wind speed data to obtain a second wind speed sequence, wherein the data cleaning includes outlier handling or missing value handling; performing data scaling on the second wind speed sequence to obtain a third wind speed sequence, wherein the data scaling includes standardization or normalization; and aggregating instantaneous data from the third wind speed sequence into hourly data to obtain the first wind speed sequence.

[0008] Optionally, the performance evaluation results include a first score and a second score. The step of evaluating the performance of each preset parameter combination of each preset wind speed correction model based on the first predicted wind speed sequence includes: performing a first score on each preset parameter combination of each preset wind speed correction model based on the predicted values ​​in the first predicted wind speed sequence and the measured values ​​in the first data sequence, and acquiring strong wind events. The strong wind events include predicted strong wind events acquired through predicted values ​​or measured strong wind events acquired through measured values. The predicted strong wind events and the measured strong wind events include a first strong wind event or a second strong wind event. The first strong wind event indicates that the cumulative duration for which the predicted / measured wind speed value is greater than a first preset threshold exceeds a first duration within a preset time period, and the second strong wind event indicates that the cumulative duration for which the predicted / measured wind speed value is greater than a second preset threshold exceeds a second duration within a preset time period. A second score is performed on each preset parameter combination of each preset wind speed correction model based on a second number of predicted strong wind events and a third number of measured strong wind events.

[0009] Optionally, the step of giving a first score to each preset parameter combination of each preset wind speed correction model based on the predicted values ​​in the first predicted wind speed sequence and the measured values ​​in the first data sequence includes: calculating the absolute error rate between each measured value greater than a third preset threshold and the corresponding predicted value; and taking the average of the absolute error rates as the score of the first score.

[0010] Optionally, the step of performing a second scoring on each preset parameter combination of each preset wind speed correction model based on the second number of predicted strong wind events and the third number of measured strong wind events includes: obtaining the recall and precision of each preset wind speed correction model under each preset parameter combination based on the second number of predicted strong wind events and the third number of measured strong wind events; and performing a second scoring on each preset parameter combination of each preset wind speed correction model based on the recall and precision and the preset parameters, wherein the preset parameters are associated with the weight of the recall.

[0011] Optionally, the step of performing a second score on each preset parameter combination of each preset wind speed correction model based on the second number of predicted strong wind events and the third number of measured strong wind events includes: performing a second score on each preset parameter combination of each preset wind speed correction model based on the first strong wind event of the predicted strong wind event and the first strong wind event of the measured strong wind event to obtain a first score; and performing a second score on each preset parameter combination of each preset wind speed correction model based on the second strong wind event of the predicted strong wind event and the second strong wind event of the measured strong wind event to obtain a second score; when the fourth number of the first strong wind event in the measured strong wind event is 0, 0 is used as the score of the second score; when the fourth number is greater than 0 and less than or equal to a fourth preset threshold, the sum of the first score of the first ratio and the second score of the second ratio is used as the score of the second score, wherein the first ratio is the ratio of the fourth number to the fourth preset threshold, and the sum of the first ratio and the second ratio is 1; when the fourth number is greater than the fourth preset threshold, the first score is used as the score of the second score.

[0012] Optionally, the step of evaluating the performance of each preset parameter combination of each preset wind speed correction model based on the first predicted wind speed sequence further includes: when the fourth quantity is 0, or when the fourth quantity is greater than 0 and the score of the first score is less than or equal to the fifth preset threshold, the score of the first score is taken as the result of the performance evaluation; when the fourth quantity is greater than 0 and the score of the first score is greater than the fifth preset threshold, the score of the second score is taken as the result of the performance evaluation.

[0013] In another general aspect, a wind speed correction method for a wind farm is provided. The wind speed correction method includes: acquiring real-time meteorological data of each wind turbine location in the wind farm; inputting the real-time meteorological data of each wind turbine location into the wind speed correction model obtained by the training method described above corresponding to each wind turbine location, to obtain a second predicted wind speed sequence for each wind turbine location, and using the second predicted wind speed sequence for each wind turbine location as the corrected wind speed sequence for each wind turbine location.

[0014] Optionally, the wind speed correction method further includes: determining whether a strong wind event exists at each wind turbine location based on the corrected wind speed sequence at each wind turbine location; and issuing a strong wind warning for any wind turbine location in response to the existence of a strong wind event at any wind turbine location.

[0015] In another general aspect, a wind speed correction device for a wind farm is provided, the wind speed correction device comprising: a data acquisition unit configured to acquire real-time meteorological data of each wind turbine location in the wind farm; and a wind speed correction unit configured to input the real-time meteorological data of each wind turbine location into a wind speed correction model obtained by the training method described above corresponding to each wind turbine location, to obtain a second predicted wind speed sequence for each wind turbine location, and to use the second predicted wind speed sequence for each wind turbine location as the corrected wind speed sequence for each wind turbine location.

[0016] Optionally, the wind speed correction device further includes a gale warning unit, configured to: determine whether a gale event exists at each wind turbine location based on the corrected wind speed sequence at each wind turbine location; and issue a gale warning for any wind turbine location in response to the existence of a gale event at any wind turbine location.

[0017] In another general aspect, a computer-readable storage medium storing a computer program is provided, characterized in that, when the computer program is executed by a processor, it implements the training method for a wind speed correction model or the wind speed correction method for a wind farm as described above.

[0018] In another general aspect, a computing device is provided, the computing device comprising: a processor; and a memory storing a computer program, which, when executed by the processor, implements the training method for the wind speed correction model or the wind speed correction method for a wind farm as described above.

[0019] According to the wind speed correction model training method, wind speed correction method and apparatus of the embodiments of this disclosure, by combining the measured wind speed data of the wind turbine site and the meteorological data of the area where the wind turbine site is located for targeted model training, a wind speed correction model for each wind turbine site can be obtained. On the one hand, multiple preset models are used for combined training at each wind turbine site, which can make full use of the advantages of each preset model and maximize the effect of the wind speed correction model for each wind turbine site. On the other hand, by using a custom loss function to evaluate the performance of the model, the weight of strong wind data in calculating the model loss can be increased, thereby improving the accuracy of the model in predicting strong winds and reducing the risk of wind turbines in strong wind disaster weather.

[0020] Further aspects and / or advantages of the general concept of this disclosure will be set forth in part in the description which follows, and in part will be clear from the description or may be learned by practice of the general concept of this disclosure. Attached Figure Description

[0021] The above and other objects and features of the embodiments of this disclosure will become clearer from the following description taken in conjunction with the accompanying drawings illustrating the embodiments, wherein:

[0022] Figure 1 This is a flowchart illustrating a training method for a wind speed correction model according to an embodiment of the present disclosure;

[0023] Figure 2 This illustrates an embodiment according to the present disclosure. Figure 1 Flowchart of step S102;

[0024] Figure 3 This illustrates an embodiment according to the present disclosure. Figure 2 Flowchart of step S201;

[0025] Figure 4 This illustrates an embodiment according to the present disclosure. Figure 1 Flowchart of step S105;

[0026] Figure 5 This is a flowchart illustrating a wind speed correction method for a wind farm according to an embodiment of the present disclosure;

[0027] Figure 6 This is a schematic diagram illustrating the model training and application process according to embodiments of the present disclosure;

[0028] Figure 7 This is a schematic diagram illustrating a training data preprocessing flow according to an embodiment of the present disclosure;

[0029] Figure 8 This is a schematic diagram illustrating a model training architecture according to an embodiment of the present disclosure;

[0030] Figure 9 This illustrates an embodiment according to the present disclosure. Figure 2 Another flowchart of step S201;

[0031] Figure 10 This is another schematic diagram illustrating a model training architecture according to an embodiment of the present disclosure;

[0032] Figure 11 This is another schematic diagram illustrating the model training and application process according to embodiments of the present disclosure;

[0033] Figure 12 This is a block diagram illustrating a wind speed correction device for a wind farm according to an embodiment of the present disclosure;

[0034] Figure 13 This is a block diagram illustrating a computing device according to an embodiment of the present disclosure. Detailed Implementation

[0035] The following detailed embodiments are provided to aid the reader in gaining a comprehensive understanding of the methods, apparatus, and / or systems described herein. However, various changes, modifications, and equivalents of the methods, apparatus, and / or systems described herein will become apparent upon understanding this disclosure. For example, the order of operations described herein is merely illustrative and is not limited to those orders set forth herein, but may be changed as will become clear upon understanding this disclosure, except for operations that must occur in a specific order. Furthermore, for clarity and conciseness, descriptions of features known in the art may be omitted.

[0036] The features described herein may be implemented in different forms and should not be construed as limited to the examples described herein. Rather, the examples described herein are provided only to illustrate some of the many feasible ways of implementing the methods, apparatus, and / or systems described herein, which will become clear upon understanding the disclosure of this application.

[0037] As used herein, the term “and / or” includes any one of the associated listed items and any combination of any two or more.

[0038] Although terms such as “first,” “second,” and “third” may be used herein to describe various components, assemblies, regions, layers, or parts, these components, assemblies, regions, layers, or parts should not be limited by these terms. Rather, these terms are used only to distinguish one component, assembly, region, layer, or part from another. Thus, without departing from the teaching of the examples described herein, the first component, first assembly, first region, first layer, or first part referred to as the first component, first assembly, first region, first layer, or first part may also be referred to as the second component, second assembly, second region, second layer, or second part.

[0039] In the specification, when an element (such as a layer, region, or substrate) is described as being "on" another element, "connected to," or "bonded to" another element, the element may be directly "on" another element, directly "connected to," or "bonded to" the other element, or one or more other elements may be present in between. Conversely, when an element is described as being "directly on" another element, "directly connected to," or "directly bonded to" another element, no other elements may be present in between.

[0040] The terminology used herein is for the purpose of describing various examples only and is not intended to limit disclosure. Unless the context clearly indicates otherwise, the singular form is intended to include the plural form as well. The terms “comprising,” “including,” and “having” indicate the presence of the described features, quantities, operations, components, elements, and / or combinations thereof, but do not preclude the presence or addition of one or more other features, quantities, operations, components, elements, and / or combinations thereof.

[0041] Unless otherwise defined, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains upon understanding this disclosure. Unless expressly defined herein, terms (such as those defined in a general dictionary) shall be interpreted as having a meaning consistent with their meaning in the context of the relevant field and in this disclosure, and shall not be interpreted in an idealized or overly formalistic manner.

[0042] Furthermore, in the description of the examples, detailed descriptions of well-known related structures or functions will be omitted when it is believed that such detailed descriptions would lead to a vague interpretation of this disclosure.

[0043] For wind speed forecasting in wind farms, the commonly used method is to correct biases based on meteorological data. Meteorological data typically comes from multiple sources, including short-term forecasts from national meteorological bureaus and European Centre for Medium-Range Weather Forecasts (ECMWF). However, traditional correction methods are generally quite simple. Specifically, most traditional correction methods directly use existing machine learning models, such as neural networks, to correct meteorological data using SCADA data from wind turbines. However, there are various machine learning models, each with its own advantages and disadvantages. Furthermore, the wind distribution and trends vary across different turbine locations within a wind farm, meaning the most suitable model may differ for each turbine location. Using the same model for all turbine locations fails to maximize the wind speed correction effect for each turbine location.

[0044] The wind speed correction model training method, wind speed correction method and apparatus of the present disclosure can combine the measured wind speed data of the wind turbine site and the meteorological data of the area where the wind turbine site is located to perform targeted model training, thereby obtaining the wind speed correction model of each wind turbine site, and performing targeted wind speed correction on this basis.

[0045] The following will refer to Figures 1 to 13 The training method for the wind speed correction model, the wind speed correction method for the wind farm, and the apparatus according to embodiments of the present disclosure are described in detail.

[0046] Figure 1 This is a flowchart illustrating a training method for a wind speed correction model according to an embodiment of the present disclosure.

[0047] Reference Figure 1 In step S101, training samples can be obtained. Here, the training samples may include historical meteorological data and historical wind speed data for the target wind turbine location. The target wind turbine location can be any one of multiple wind turbine locations in the wind farm. Further, the historical meteorological data may include parameters such as temperature, air pressure, wind speed, and wind direction, and the historical wind speed data may include measured wind speed data from the unit's SCADA data.

[0048] Next, in step S102, the training samples can be preprocessed to obtain the first data sequence. See below for further details. Figure 2 Detailed description of embodiments according to this disclosure Figure 1 Step S102.

[0049] Figure 2 This illustrates an embodiment according to the present disclosure. Figure 1 The flowchart for step S102.

[0050] Reference Figure 2 In step S201, historical wind speed data can be aggregated by hour to convert instantaneous data into hourly data, thus obtaining a first wind speed sequence. This allows the historical wind speed data to be combined with hourly data from historical meteorological data for processing. (See below for further details.) Figure 3 Detailed description of embodiments according to this disclosure Figure 2 Step S201.

[0051] Figure 3 This illustrates an embodiment according to the present disclosure. Figure 2 The flowchart for step S201.

[0052] Reference Figure 3 In step S301, a second wind speed sequence can be obtained by cleaning the historical wind speed data. Here, data cleaning may include outlier handling or missing value handling. Furthermore, through data cleaning, outliers, missing values, or wind speed data from abnormal power generation states of the generator unit can be removed from the historical wind speed data.

[0053] Next, in step S302, a third wind speed sequence can be obtained by scaling the second wind speed sequence. Here, data scaling may include standardization or normalization. Further, standardization can mean converting the data to a standard normal distribution by calculating a standard score (z-score), making the average value of the data 0 and the standard deviation 1; normalization can mean transforming the data to the same dimension and mapping the data to the interval [0,1] or [-1,1]. Furthermore, data scaling can eliminate the differences between wind speed data of different orders of magnitude or different dimensions in the second wind speed sequence, allowing each wind speed value in the sequence to be processed according to a unified standard.

[0054] Next, in step S303, the instantaneous data in the third wind speed sequence can be aggregated into hourly data to obtain the first wind speed sequence. Here, the instantaneous data can be aggregated into hourly data by averaging or using the 70% to 90% quantile, but it is not limited to this method.

[0055] Return to reference Figure 2 In step S202, a second data sequence can be obtained by concatenating the first wind speed sequence and the historical meteorological data with the same time frame. Here, as an example, the first wind speed sequence and the historical meteorological data with the same time frame can be combined into vectors of a preset dimension, such as vectors [a1, a2, a3, ..., a...]. n Then, based on the vectors at each time point, a second data sequence is obtained; however, this disclosure is not limited thereto.

[0056] Next, in step S203, the data of any hour in the second data sequence can be aggregated with the data of a first number of hours preceding that hour, and a first data sequence is obtained based on the aggregated data of each hour. Here, the data of any hour can be aggregated with the data of multiple hours preceding that hour by averaging or using the 70% to 90% quantile, etc. For example, the historical wind speed data of any hour in the second data sequence after the above processing can be aggregated with the historical meteorological data of multiple hours preceding that hour, but it is not limited to this; and the value of the first number can be set by those skilled in the art according to the actual situation.

[0057] Return to reference Figure 1 In step S103, the first data sequence can be input into multiple preset wind speed correction models. Here, each preset wind speed correction model has multiple preset parameter combinations.

[0058] According to embodiments of this disclosure, as an example, the aforementioned multiple preset wind speed correction models may include a GRU (Gate Recurrent Unit) model and an XGBoost (eXtreme Gradient Boosting) model. These two models each have advantages in terms of time-series data prediction adaptability, nonlinear regression performance, and operational efficiency. Here, multiple preset parameter combinations of the GRU and XGBoost models can be obtained by combining the various optional parameter values ​​in Table 1 below.

[0059] Table 1: Summary of parameters for GRU and XGBoost models

[0060]

[0061]

[0062] As an example, for the GRU model mentioned above, to effectively capture the temporal and spatial features of meteorological data, a two-input, single-output deep learning model structure can be used. Further, the first input may include the aforementioned first data sequence (containing parameters such as temperature, air pressure, wind speed, and wind direction) to mine the spatial features between the parameters. Then, the first data sequence is passed to an NN (Neural Network) layer. The activation function of the NN layer can be selected as Sigmoid, tanh, or ReLU as shown in Table 1, and the number of hidden layer nodes can be selected as 6, 12, 24, and 48 as shown in Table 1. Simultaneously, the second input may include the wind speed sequence from the aforementioned historical meteorological data to mine the temporal features of wind speed. Then, the wind speed sequence is passed to a single-layer GRU and a single-layer NN. The activation function of the NN layer can be selected as Sigmoid, tanh, or ReLU as shown in Table 1, and the number of hidden layer nodes can be selected as 6, 12, 24, and 48 as shown in Table 1. A Dropout layer is also used to prevent overfitting of the NN layer. Based on the two inputs mentioned above, a single-layer neural network can be used to aggregate the results from both inputs, ultimately outputting the predicted wind speed sequence. Furthermore, the GRU model can be optimized using the Adam algorithm to improve training efficiency. Simultaneously, the learning rate (such as the values ​​of 0.0001, 0.001, 0.01, 0.1, and 0.15 shown in Table 1) can be dynamically adjusted based on simulated annealing or exponential learning rate strategies to balance efficiency and accuracy, avoiding getting stuck in local optima during training. Additionally, methods such as ModelCheckpoint or EarlyStopping can be used during model callbacks to reduce the probability of overfitting.

[0063] As an example, for the XGBoost model mentioned above, since it lacks cyclic prediction capabilities, only the first data sequence is input to enable the model to extract temporal features. The impact of the input data duration on prediction accuracy can also be considered. Furthermore, a greedy algorithm can be used during XGBoost model training, attempting to split a leaf node each time, calculating the gain after the split, and then determining the optimal split feature and optimal split position based on the maximum gain.

[0064] Next, in step S104, the first predicted wind speed sequence output by each preset wind speed correction model under each preset parameter combination can be obtained. Here, a grid search can be used to traverse and select all preset parameter combinations to obtain the first predicted wind speed sequence output by each preset wind speed correction model under each preset parameter combination.

[0065] Next, in step S105, performance evaluation can be performed for each preset parameter combination of each preset wind speed correction model based on the first predicted wind speed sequence. Here, the performance evaluation results may include a first score and a second score. See below for further details. Figure 4 Detailed description of embodiments according to this disclosure Figure 1 Step S105.

[0066] Figure 4 This illustrates an embodiment according to the present disclosure. Figure 1 The flowchart for step S105.

[0067] Reference Figure 4In step S401, a first score can be performed on each preset parameter combination of each preset wind speed correction model based on the predicted values ​​in the first predicted wind speed sequence and the measured values ​​in the first data sequence, and a strong wind event can be obtained. Here, the predicted values ​​in the first predicted wind speed sequence are obtained based on the wind speeds in the aforementioned historical meteorological data, and the measured values ​​in the first data sequence are obtained based on the measured historical wind speed data. Further, a strong wind event can include a predicted strong wind event obtained through predicted values ​​or a measured strong wind event obtained through measured values; and the predicted strong wind event and the measured strong wind event can include a first strong wind event or a second strong wind event. A first strong wind event can indicate that the cumulative duration for which the predicted value / measured value of wind speed is greater than a first preset threshold exceeds a first duration within a preset time period, and a second strong wind event can indicate that the cumulative duration for which the predicted value / measured value of wind speed is greater than a second preset threshold exceeds a second duration within a preset time period. As an example, the first and second gale events can be extreme disaster warning events. The first gale event can be an event where the average wind speed is greater than 15 m / s for more than 3 hours in a continuous 24-hour period (referred to as V15), and the second gale event can be an event where the average wind speed is greater than 12 m / s for more than 6 hours in a continuous 24-hour period (referred to as V12). However, they are not limited to these. Those skilled in the art can set the first preset threshold, the first duration, the second preset threshold, and the second duration according to the actual situation.

[0068] According to embodiments of this disclosure, the absolute error rate between each measured value and the corresponding predicted value that exceeds a third preset threshold can be calculated; then, the average of the absolute error rates can be used as the score of the first score. Here, the average of the absolute error rate differences between the measured wind speeds above 12 m / s and the corresponding predicted values ​​can be calculated, i.e., the mean of high-speed-wind forecast accuracy (MHFA) when the measured wind speed at the target wind turbine location is above 12 m / s. Then, the MHFA can be used as the score of the first score, but it is not limited thereto. Those skilled in the art can set the third preset threshold according to the actual situation. Further, the MHFA can be represented by the following equation (1):

[0069]

[0070] Here, the quantity N can represent the number of measured values ​​that are greater than the third preset threshold.

[0071] Next, in step S402, a second score can be performed on each preset parameter combination of each preset wind speed correction model based on the second number of predicted strong wind events and the third number of actual strong wind events. Here, the recall rate and precision rate of each preset wind speed correction model under each preset parameter combination can be obtained based on the second number of predicted strong wind events and the third number of actual strong wind events. Then, a second score can be performed on each preset parameter combination of each preset wind speed correction model based on the recall rate, precision rate, and preset parameter (β). Further, the preset parameter is associated with the weight of the recall rate.

[0072] As an example, a second score can be obtained based on equation (2) as shown below:

[0073]

[0074] Here, F β As a quantitative indicator for evaluating the model correction effect, it can be jointly determined by precision, recall, and the β parameter. Furthermore, the weight of recall can be adjusted by changing the value of β (1, 1.5, or 2, etc.), meaning the larger the β, the higher the weight of recall. Further, recall can be represented by the following equation (3):

[0075]

[0076] Here, Count() can be used to represent counting. Therefore, recall can be expressed as the ratio of the number of times the actual strong wind events and the predicted strong wind events intersect to the number of actual strong wind events.

[0077] In addition, the accuracy can be expressed by the following equation (4):

[0078]

[0079] Here, accuracy can be expressed as the ratio of the number of times the measured and predicted gale events intersect to the number of predicted gale events.

[0080] According to embodiments of this disclosure, a second score can be performed on each preset parameter combination of each preset wind speed correction model based on the first wind event of the predicted strong wind event and the first wind event of the measured strong wind event, to obtain a first score (F). β (V15)), and based on the predicted gale event and the measured gale event, a second score is given for each preset parameter combination of each preset wind speed correction model to obtain a second score (F). β (V12)). Then, when the fourth number (N) of the first gale event in the measured gale events.15 When the fourth quantity is 0, 0 can be used as the score of the second rating (i.e., score = 0); when the fourth quantity is greater than 0 and less than or equal to the fourth preset threshold (e.g., but not limited to 10), the sum of the first score of the first ratio and the second score of the second ratio can be used as the score of the second rating. Here, the first ratio is the ratio of the fourth quantity to the fourth preset threshold, and the sum of the first ratio and the second ratio is 1. At this time, the score of the second rating can be represented by the following formula (5):

[0081] score=(N15 / 10)×F β (V15)+(1-N15 / 10)×F β (V12) (5)

[0082] Then, when the fourth quantity is greater than the fourth preset threshold, the first score can be used as the score for the second score (i.e., score = F). β (V15)).

[0083] Next, in step S403, when the fourth quantity is 0, or the fourth quantity is greater than 0 and the score of the first rating is less than or equal to the fifth preset threshold (e.g., but not limited to 0.8), the score of the first rating can be used as the result of the performance evaluation.

[0084] Next, in step S404, when the fourth quantity is greater than 0 and the score of the first rating is greater than the fifth preset threshold, the score of the second rating can be used as the result of the performance evaluation.

[0085] Return to reference Figure 1 In step S106, the preset wind speed correction model corresponding to the preset parameter combination whose performance evaluation results meet preset requirements can be determined as the wind speed correction model for the wind turbine location. Here, the preset requirement can be that the score of the first rating or the score of the second rating in the performance evaluation results is the largest. Further, when the score of the first rating is null (meaning the value is unknown), the determination of the wind speed correction model is not required.

[0086] The wind speed correction model training method according to the embodiments of this disclosure combines measured wind speed data at wind turbine sites with meteorological data of the area where the wind turbine sites are located for targeted model training. On the one hand, multiple preset models are used for combined training at each wind turbine site, which can make full use of the advantages of each preset model and maximize the effect of the wind speed correction model at each wind turbine site. On the other hand, the model performance is evaluated by using a custom loss function, which can take into account the low proportion of strong wind data in the training samples and the serious sample imbalance. By increasing the weight of strong wind data in calculating the model loss, the accuracy of the model's prediction of strong winds is improved, and the risk of wind turbines in strong wind disaster weather is reduced.

[0087] Figure 5 This is a flowchart illustrating a wind speed correction method for a wind farm according to an embodiment of the present disclosure.

[0088] Reference Figure 5 In step S501, real-time meteorological data of each wind turbine location in the wind farm can be obtained.

[0089] Next, in step S502, the real-time meteorological data of each wind turbine location can be input into the wind speed correction model obtained by the training method described above for each wind turbine location to obtain the second predicted wind speed sequence for each wind turbine location, and the second predicted wind speed sequence for each wind turbine location can be used as the corrected wind speed sequence for each wind turbine location.

[0090] According to embodiments of this disclosure, the presence of a strong wind event at each wind turbine location can be determined based on the corrected wind speed sequence at each turbine location. Then, in response to the presence of a strong wind event at any wind turbine location, a strong wind warning can be issued for that turbine location. Furthermore, the wind speed correction method for wind farms according to embodiments of this disclosure is not limited to issuing strong wind warnings; it can also be used to predict or warn of other wind characteristics.

[0091] The wind speed correction method for wind farms according to embodiments of the present disclosure can utilize the advantages of various machine learning models by differentiating the wind speed correction models for each wind turbine location, thereby adapting to the wind condition distribution and changing trends at different wind turbine locations in the wind farm.

[0092] Figure 6 This is a schematic diagram illustrating the model training and application process according to embodiments of the present disclosure.

[0093] Reference Figure 6 As an example, in the model training part, meteorological forecast data and SCADA data of the target wind turbine locations can be used (only the measured wind speed data of the turbines in the SCADA data can be used). Training samples are obtained through a pre-set data preprocessing pipeline as input data for the model. Then, the model is built based on two routes: XGBoost and GRU. Finally, the optimal model is selected as the high-wind correction model based on grid search (including searching for hyperparameters with large model influence factors and custom parameters) and model warning performance scoring (performance evaluation based on recall and precision, etc., of V12 / V15). Figure 6 As shown, as an example, in the model application part (i.e. the gale warning process), at the target wind turbine location, the real-time updated meteorological forecast data can be input into the pre-trained gale correction model to obtain the corrected forecast data, which is then used to issue a gale event warning.

[0094] Figure 7This is a schematic diagram illustrating a training data preprocessing flow according to an embodiment of the present disclosure.

[0095] Reference Figure 7 As an example, during the data preprocessing for model training, the meteorological forecast data and SCADA data for the target wind turbine location can be subjected to data cleaning (i.e., handling outliers or missing values), data scaling (i.e., standardization or normalization), data aggregation, and simultaneous concatenation of meteorological and SCADA data. This yields a sample data sequence for the target location, which may include various meteorological parameters (e.g., temperature, air pressure, wind speed, and wind direction) and wind speed data from the SCADA data. Further processing can be performed on the sample data sequence, dividing it into training and testing sets and performing time-series data processing (i.e., aggregating data from the past n hours hourly), to obtain multiple meteorological prediction parameters (which can be the first data sequence as described above). Finally, these multiple meteorological prediction parameters can be input into the model for training.

[0096] Figure 8 This is a schematic diagram illustrating a model training architecture according to an embodiment of the present disclosure.

[0097] Reference Figure 8 As an example, the multiple preset wind speed correction models mentioned above can be GRU models and XGBoost models. Figure 8 As shown, for the GRU model, a two-input, single-output deep learning model structure can be used. The first input can include the aforementioned multiple parameters of weather forecasting, which are then passed to a DNN (Deep Neural Networks) layer, combined with a Dropout layer to prevent overfitting. Simultaneously, the second input can include the predicted wind speed from historical weather data, as described above, which is then passed to a single-layer GRU and a single-layer DNN, combined with a Dropout layer to prevent overfitting. Based on these two inputs, a single-layer DNN can finally aggregate the results from both inputs to output the final wind speed prediction. Additionally, as... Figure 8 As shown, for the XGBoost model, only the aforementioned meteorological forecast parameters need to be input, and the wind speed prediction value can be obtained through XGBoost training. Finally, as mentioned above, the optimal model can be selected based on grid search and model early warning performance scoring.

[0098] Figure 9 This illustrates an embodiment according to the present disclosure. Figure 2 Another flowchart of step S201.

[0099] like Figure 9As shown, in step S901, the historical wind speed data can be aggregated according to a predetermined period to obtain a first wind speed sequence by aggregating the instantaneous data in the historical wind speed data into data of the predetermined period. In the embodiments of this disclosure, the time length of the predetermined period can be set according to actual needs; for example, the predetermined period can be set to one hour.

[0100] According to embodiments of this disclosure, references can be used Figure 7 The described method preprocesses the model training samples to obtain the first wind speed sequence.

[0101] In step S902, a second data sequence can be obtained by splicing together data from the first wind speed sequence and the historical meteorological data that are of the same time. Thus, by simultaneously splicing the first wind speed sequence and historical meteorological data, a data sequence including wind speed and meteorological parameters, i.e., the second data sequence, can be obtained.

[0102] In step S903, the data from each predetermined period to the first predetermined period before each predetermined period in the second data sequence can be subjected to a second data aggregation to obtain a first data subsequence for each predetermined period, and the data from the second predetermined period to the third predetermined period before each predetermined period in the first wind speed sequence can be subjected to a third data aggregation to obtain a second data subsequence for each predetermined period, wherein the third predetermined period is before the second predetermined period.

[0103] For each predetermined period, based on the first data subsequence obtained by the second data aggregation, information for predicting future wind speeds can be mined from wind speed data with a relatively longer time frame through the second data subsequence obtained by the second data aggregation, thereby further improving wind speed correction.

[0104] For each predetermined period, a first data subsequence can be obtained based on the second data aggregation described above, and a second data subsequence can be obtained based on the third data aggregation described above. For each predetermined period, the time period corresponding to the first data subsequence is from each predetermined period to the first predetermined period prior to each predetermined period, which can be represented as the first predetermined time period; the time period corresponding to the second data subsequence is from the second predetermined period to the third predetermined period, which can be represented as the second predetermined time period. The first predetermined time period may partially overlap with the second predetermined time period or may not overlap at all.

[0105] In one embodiment of this disclosure, the first predetermined time period corresponding to the first data subsequence may be adjacent to the second predetermined time period corresponding to the second data subsequence, and the first predetermined time period is before the second predetermined time period.

[0106] In step S904, the first data sequence can be obtained based on the first data subsequence and the second data subsequence. The first data sequence may include the first data subsequence and the second data subsequence.

[0107] In an example with a predetermined period of one hour, the historical wind speed data can be aggregated hourly to obtain a first wind speed sequence. A second data sequence is obtained by concatenating the first wind speed sequence with data of the same time from the historical meteorological data. The data from each hour (e.g., hour t corresponding to time T) in the second data sequence up to the first hour preceding that hour (e.g., hour (tn), n > 0) is then aggregated again to obtain a first data subsequence for each hour. The data from the second hour (e.g., hour (tm), m > 0) to the third hour (e.g., hour (tmp), p > 0) in the first wind speed sequence are then aggregated again to obtain a second data subsequence for each hour, where the third hour precedes the second hour. Based on the first and second data subsequences, the first data sequence is obtained. The values ​​of m, n, and p can be selected according to actual application requirements; m can be equal to n or not equal to n.

[0108] In this disclosure, only an embodiment with a predetermined period of one hour is used as an example for illustration. However, this disclosure is not limited to this. The above training samples can also be processed based on predetermined periods of different time lengths to obtain the first data sequence.

[0109] After obtaining the first data sequence by preprocessing the training samples, the first data sequence can be input into multiple preset wind speed correction models, where each preset wind speed correction model has multiple preset parameter combinations.

[0110] Figure 10 This is another schematic diagram illustrating a model training architecture according to an embodiment of the present disclosure.

[0111] Reference Figure 10 As an example, the multiple preset wind speed correction models mentioned above can be GRU models and XGBoost models. Figure 10 As shown, for the GRU model, a three-input-single-output deep learning model structure can be used.

[0112] The first input can include multiple meteorological forecast parameters (e.g., forecast wind speed, forecast wind direction, forecast air pressure, etc.) from the meteorological forecast data, which is beneficial for exploring the spatial relationships between these parameters. The first input can be flattened using a Flatten layer to transform multi-time-time data into one-dimensional or 1D tensors before being passed to the DNN (Deep Neural Networks) layer. A Dropout layer can be used to prevent overfitting. In this example, the activation function for the NN layer can be the ReLU function, and the number of hidden layer nodes can be optimized using a grid search.

[0113] The second input may include the predicted wind speed from historical meteorological data as described above, to be used to mine the temporal features of wind speed. In this embodiment, a single-layer GRU + a single-layer NN can be selected as the network structure to receive the second input. The activation function of the NN can be selected as the ReLU function, the number of hidden layer nodes can be optimized through grid search, and a Dropout layer is introduced to prevent the NN from overfitting.

[0114] The third input may include nearby measured wind speed data, such as a wind speed sequence obtained from the unit's SCADA data that is close to the time period corresponding to the first and second inputs. In this embodiment, a single-layer GRU + single-layer NN can be selected as the network structure for receiving the third input. The activation function of the NN can be selected as the ReLU function, the number of hidden layer nodes can be optimized through grid search, and a Dropout layer is introduced to prevent the NN from overfitting.

[0115] This example uses a predetermined period of one hour, but the disclosure is not limited thereto. For example, for each hour, the first input may include multiple meteorological forecast parameters from each hour (e.g., hour t corresponding to time T) to the first hour preceding each hour (e.g., hour (tn)), the second input may include the meteorological forecast wind speed from each hour (e.g., hour t corresponding to time T) to the first hour preceding each hour (e.g., hour (tn)), and the third input may include the wind speed sequence from the second hour (e.g., hour (tm)) to the third hour (e.g., hour (tmp)) as nearby measured wind speed data.

[0116] Based on the three inputs mentioned above, a single-layer DNN can be used to aggregate the results of the three outputs, ultimately outputting the predicted wind speed value. In this embodiment, the activation function of the NN can be selected as the ReLU function, and the number of hidden layer nodes can be optimized through grid search.

[0117] In addition, such as Figure 10 As shown, the meteorological forecast parameters and nearby measured wind speed data described above can be input into the XGBoost model to obtain wind speed prediction values ​​through XGBoost training.

[0118] Based on the outputs of the GRU model and the XGBoost model, the optimal model can be selected based on grid search and model early warning performance score.

[0119] exist Figure 10 In the described embodiments, the multiple preset wind speed correction models may include the GRU model and the XGBoost model. These two models each have advantages in terms of time-series data prediction adaptability, nonlinear regression performance, and operational efficiency. Here, multiple preset parameter combinations of the GRU model and the XGBoost model can be obtained by combining the various parameter selectable values ​​in Table 2 below.

[0120] Table 2: Summary of parameters for GRU and XGBoost models

[0121]

[0122]

[0123] References can be used Figures 1 to 4 The performance evaluation method described here evaluates the performance for each preset parameter combination of each preset wind speed correction model, and will not be elaborated further for the sake of brevity.

[0124] The following reference Figure 11 Describe the process of model training and application. Figure 11 This is another schematic diagram illustrating the model training and application process according to embodiments of the present disclosure.

[0125] Reference Figure 11 As an example, in the model training part, meteorological forecast data and SCADA data (e.g., measured wind speed data in SCADA data) of the target wind turbine location can be used to obtain training samples as input data for the model through a pre-set data preprocessing pipeline. Then, the model is built based on two routes: XGBoost and GRU. Finally, the optimal model is selected as the gale correction model based on grid search (including the search for hyperparameters with large model influence factors and custom parameters) and model warning performance score (performance evaluation based on recall and precision of V12 / V15, etc.).

[0126] like Figure 11 As shown, as an example, in the model application part (i.e. the gale warning process), real-time updated meteorological forecast data and real-time updated nearby measured wind speed data can be input into the pre-trained gale correction model at the target wind turbine location to obtain the corrected forecast data as wind turbine prediction data, which is used to issue a gale event warning.

[0127] and Figure 6Compared to the illustrated embodiment, in addition to using real-time updated weather forecast data as model input, it also uses real-time updated nearby measured wind speed data as model input. This allows for further mining of the wind turbine's own measured data for wind speed correction, rather than being limited to weather forecast data.

[0128] Figure 12 This is a block diagram illustrating a wind speed correction device for a wind farm according to an embodiment of the present disclosure. The wind speed correction device for a wind farm according to an embodiment of the present disclosure can be implemented in a computing device with sufficient computing power.

[0129] Reference Figure 12 The wind speed correction device 900 for a wind farm may include a data acquisition unit 910 and a wind speed correction unit 920.

[0130] The data acquisition unit 910 can acquire real-time meteorological data for each wind turbine location in the wind farm.

[0131] The wind speed correction unit 920 inputs the real-time meteorological data of each wind turbine location into the wind speed correction model obtained by the training method described above, corresponding to each wind turbine location, to obtain the second predicted wind speed sequence of each wind turbine location, and uses the second predicted wind speed sequence of each wind turbine location as the corrected wind speed sequence of each wind turbine location.

[0132] Optionally, the wind speed correction device 900 also includes a gale warning unit. The gale warning unit can determine whether a gale event exists at each wind turbine location based on the corrected wind speed sequence at each wind turbine location; and in response to the existence of a gale event at any wind turbine location, it issues a gale warning for that wind turbine location.

[0133] Figure 13 This is a block diagram illustrating a computing device according to an embodiment of the present disclosure.

[0134] Reference Figure 13 The computing device 1000 according to embodiments of the present disclosure may include a processor 1010 and a memory 1020. The processor 1010 may include (but is not limited to) a central processing unit (CPU), a digital signal processor (DSP), a microcomputer, a field-programmable gate array (FPGA), a system-on-a-chip (SoC), a microprocessor, an application-specific integrated circuit (ASIC), etc. The memory 1020 stores computer programs to be executed by the processor 1010. The memory 1020 includes high-speed random access memory and / or a non-volatile computer-readable storage medium. When the processor 1010 executes the computer program stored in the memory 1020, the wind speed correction model training method or the wind speed correction method for a wind farm as described above can be implemented.

[0135] The wind speed correction model training method or wind speed correction method for a wind farm according to embodiments of this disclosure can be programmed into a computer program and stored on a computer-readable storage medium. When the computer program is executed by a processor, the wind speed correction model training method or wind speed correction method for a wind farm as described above can be implemented. Examples of computer-readable storage media include: read-only memory (ROM), random access programmable read-only memory (PROM), electrically erasable programmable read-only memory (EEPROM), random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), flash memory, non-volatile memory, CD-ROM, CD-R, CD+R, CD-RW, CD+RW, DVD-ROM, DVD-R, DVD+R, DVD-RW, DVD+RW, DVD-RAM, BD-ROM, BD-R, BD-R LTH, BD-RE, Blu-ray or optical disc storage, hard disk drive (HDD), solid-state drive (SSD), card storage (such as multimedia cards, secure digital (SD) cards, or ultra-fast digital (XD) cards), magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid-state drive, and any other device configured to store a computer program and any associated data, data files, and data structures in a non-transitory manner and to provide the computer program and any associated data, data files, and data structures to a processor or computer so that the processor or computer can execute the computer program. In one example, the computer program and any associated data, data files, and data structures are distributed across a networked computer system, such that the computer program and any associated data, data files, and data structures are stored, accessed, and executed in a distributed manner through one or more processors or computers.

[0136] According to the wind speed correction model training method, wind speed correction method and apparatus of the embodiments of this disclosure, by combining the measured wind speed data of the wind turbine site and the meteorological data of the area where the wind turbine site is located for targeted model training, a wind speed correction model for each wind turbine site can be obtained. On the one hand, multiple preset models are used for combined training at each wind turbine site, which can make full use of the advantages of each preset model and maximize the effect of the wind speed correction model for each wind turbine site. On the other hand, by using a custom loss function to evaluate the performance of the model, the weight of strong wind data in calculating the model loss can be increased, thereby improving the accuracy of the model in predicting strong winds and reducing the risk of wind turbines in strong wind disaster weather.

[0137] The training method for the wind speed correction model according to embodiments of this disclosure, on the one hand, integrates multiple machine learning models for combined training to mine the spatiotemporal correlation between parameters such as wind speed, wind direction, and air pressure in meteorological data and strong wind events at target wind turbine locations, effectively improving the accuracy and generalization ability of strong wind warnings; on the other hand, it establishes an automated meteorological data preprocessing pipeline that integrates data cleaning, data scaling, data aggregation, multi-attribute combination of data, and time-series data processing, significantly improving data processing efficiency and ensuring data standardization of training samples; and on the other hand, it proposes a method that integrates strong wind MHFA, precision / recall rate, and F... β The model performance evaluation criteria can comprehensively assess wind speed prediction performance and gale warning effectiveness, and more comprehensively and accurately reflect the model's prediction performance, ensuring that the selected models can effectively respond to project goals and needs. In addition, it can further explore the measured data of the wind turbine itself to further improve wind speed correction, without being limited by meteorological forecast data.

[0138] While some embodiments of this disclosure have been shown and described, those skilled in the art will understand that modifications may be made to these embodiments without departing from the principles and spirit of this disclosure, which are defined by the claims and their equivalents.

Claims

1. A training method for a wind speed correction model, characterized in that, The training method includes: Acquire training samples, which include historical meteorological data and historical wind speed data of the target wind turbine location, where the target wind turbine location is any one of multiple wind turbine locations in the wind farm. The training samples are preprocessed to obtain a first data sequence; The first data sequence is input into multiple preset wind speed correction models, wherein each preset wind speed correction model has multiple preset parameter combinations; Obtain the first predicted wind speed sequence output by each preset wind speed correction model under each preset parameter combination; Based on the first predicted wind speed sequence, performance evaluation is performed for each preset parameter combination of each preset wind speed correction model; The preset wind speed correction model corresponding to the preset parameter combination whose performance evaluation results meet the preset requirements is determined as the wind speed correction model for this wind turbine location. The performance evaluation results include a first score and a second score. The step of evaluating the performance of each preset parameter combination of each preset wind speed correction model based on the first predicted wind speed sequence includes: performing a first score on each preset parameter combination of each preset wind speed correction model based on the predicted values ​​in the first predicted wind speed sequence and the measured values ​​in the first data sequence, and acquiring strong wind events. These strong wind events include predicted strong wind events acquired through predicted values ​​or measured strong wind events acquired through measured values. The predicted and measured strong wind events include either a first strong wind event or a second strong wind event. The first strong wind event indicates that the cumulative duration for which the predicted / measured wind speed value is greater than a first preset threshold exceeds a first duration within a preset time period; the second strong wind event indicates that the cumulative duration for which the predicted / measured wind speed value is greater than a second preset threshold exceeds a second duration within a preset time period. A second score is then performed on each preset parameter combination of each preset wind speed correction model based on a second number of predicted strong wind events and a third number of measured strong wind events.

2. The training method as described in claim 1, characterized in that, The steps of preprocessing the training samples to obtain the first data sequence include: By performing a first data aggregation on the historical wind speed data, the instantaneous data in the historical wind speed data is aggregated into hourly data to obtain a first wind speed sequence; A second data sequence is obtained by splicing together the first wind speed sequence and the historical meteorological data that have the same time. The data of any hour in the second data sequence is aggregated with the data of a first number of hours preceding that hour, and the first data sequence is obtained based on the aggregated data for each hour.

3. The training method as described in claim 1, characterized in that, The steps of preprocessing the training samples to obtain the first data sequence include: By performing a first data aggregation on the historical wind speed data according to a predetermined period, the instantaneous data in the historical wind speed data is aggregated into data of a predetermined period to obtain a first wind speed sequence. A second data sequence is obtained by splicing together the first wind speed sequence and the historical meteorological data that have the same time. The data from each predetermined period in the second data sequence to the first predetermined period before each predetermined period are subjected to a second data aggregation to obtain a first data subsequence for each predetermined period, and the data from the second predetermined period to the third predetermined period before each predetermined period in the first wind speed sequence are subjected to a third data aggregation to obtain a second data subsequence for each predetermined period, wherein the third predetermined period is before the second predetermined period. The first data sequence is obtained based on the first data subsequence and the second data subsequence.

4. The training method as described in claim 3, characterized in that, The steps of preprocessing the training samples to obtain the first data sequence include: By performing a first data aggregation on the historical wind speed data, the instantaneous data in the historical wind speed data is aggregated into hourly data to obtain a first wind speed sequence; A second data sequence is obtained by splicing together the first wind speed sequence and the historical meteorological data that have the same time. The data from each hour to the first hour before each hour in the second data sequence are aggregated in the second data sequence to obtain a first data subsequence for each hour, and the data from the second hour to the third hour before each hour in the first wind speed sequence are aggregated in the third data sequence to obtain a second data subsequence for each hour, wherein the third hour is before the second hour; The first data sequence is obtained based on the first data subsequence and the second data subsequence.

5. The training method as described in claim 2 or 4, characterized in that, The steps of aggregating instantaneous data from the historical wind speed data into hourly data by performing a first data aggregation on the historical wind speed data include: A second wind speed sequence is obtained by cleaning the historical wind speed data, wherein the data cleaning includes outlier handling or missing value handling. A third wind speed sequence is obtained by scaling the second wind speed sequence, wherein the data scaling includes standardization or normalization. The first wind speed sequence is obtained by aggregating the instantaneous data in the third wind speed sequence into hourly data.

6. The training method as described in claim 1, characterized in that, The step of performing a first score on each preset parameter combination of each preset wind speed correction model based on the predicted values ​​in the first predicted wind speed sequence and the measured values ​​in the first data sequence includes: Calculate the absolute error rate between each measured value and the corresponding predicted value that exceeds the third preset threshold; The average of the absolute error rates is used as the score for the first rating.

7. The training method as described in claim 1, characterized in that, The step of performing a second scoring on each preset parameter combination for each preset wind speed correction model based on the second number of predicted strong wind events and the third number of measured strong wind events includes: Based on the second number of predicted strong wind events and the third number of measured strong wind events, the recall and precision of each preset wind speed correction model under each preset parameter combination are obtained. Based on the recall rate and precision rate, as well as preset parameters, a second score is performed on each preset parameter combination of each preset wind speed correction model, wherein the preset parameters are associated with the weight of the recall rate.

8. The training method as described in claim 7, characterized in that, The step of performing a second scoring on each preset parameter combination for each preset wind speed correction model based on the second number of predicted strong wind events and the third number of measured strong wind events includes: Based on the first strong wind event of the predicted strong wind event and the first strong wind event of the measured strong wind event, a second score is performed on each preset parameter combination of each preset wind speed correction model to obtain a first score; and based on the second strong wind event of the predicted strong wind event and the second strong wind event of the measured strong wind event, a second score is performed on each preset parameter combination of each preset wind speed correction model to obtain a second score. When the fourth number of the first gale event in the measured gale events is 0, 0 is taken as the score of the second rating; When the fourth quantity is greater than 0 and less than or equal to the fourth preset threshold, the sum of the first score of the first ratio and the second score of the second ratio is used as the score of the second rating, wherein the first ratio is the ratio of the fourth quantity to the fourth preset threshold, and the sum of the first ratio and the second ratio is 1. When the fourth quantity is greater than the fourth preset threshold, the first score is used as the score of the second rating.

9. The training method as described in claim 8, characterized in that, Based on the first predicted wind speed sequence, the step of evaluating the performance of each preset parameter combination for each preset wind speed correction model further includes: When the fourth quantity is 0, or when the fourth quantity is greater than 0 and the score of the first rating is less than or equal to the fifth preset threshold, the score of the first rating is taken as the result of the performance evaluation. When the fourth quantity is greater than 0 and the score of the first rating is greater than the fifth preset threshold, the score of the second rating is used as the result of the performance evaluation.

10. A wind speed correction method for a wind farm, characterized in that, The wind speed correction method includes: Obtain real-time meteorological data for each wind turbine location in the wind farm; The real-time meteorological data of each wind turbine location is input into the wind speed correction model obtained by the training method as described in any one of claims 1 to 9 corresponding to each wind turbine location, so as to obtain the second predicted wind speed sequence of each wind turbine location, and the second predicted wind speed sequence of each wind turbine location is used as the corrected wind speed sequence of each wind turbine location.

11. The wind speed correction method as described in claim 10, characterized in that, The wind speed correction method further includes: acquiring real-time wind speed data at each wind turbine location in the wind farm. Real-time meteorological data for each wind turbine location are input into the wind speed correction model obtained by the training method described in any one of claims 1 to 9 for each wind turbine location, to obtain a second predicted wind speed sequence for each wind turbine location, including: The real-time meteorological data and real-time wind speed data of each wind turbine location are respectively input into the wind speed correction model obtained by the training method as described in any one of claims 1 to 9 corresponding to each wind turbine location, so as to obtain the second predicted wind speed sequence of each wind turbine location.

12. The wind speed correction method as described in claim 10 or 11, characterized in that, The wind speed correction method also includes: Based on the corrected wind speed sequence of each wind turbine location, determine whether there is a strong wind event at each wind turbine location; In response to a strong wind event at any wind turbine location, a strong wind warning will be issued for that wind turbine location.

13. A wind speed correction device for a wind farm, characterized in that, The wind speed correction device includes: The data acquisition unit is configured to acquire real-time meteorological data at each wind turbine location in the wind farm. The wind speed correction unit is configured to: input the real-time meteorological data of each wind turbine location into the wind speed correction model obtained by the training method according to any one of claims 1 to 9 corresponding to each wind turbine location, obtain the second predicted wind speed sequence of each wind turbine location, and use the second predicted wind speed sequence of each wind turbine location as the corrected wind speed sequence of each wind turbine location.

14. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the training method for the wind speed correction model as described in any one of claims 1 to 9 or the wind speed correction method for the wind farm as described in any one of claims 10 to 12.

15. A computing device, characterized in that, The computing device includes: processor; and The memory stores a computer program that, when executed by a processor, implements the training method for the wind speed correction model as described in any one of claims 1 to 9 or the wind speed correction method for the wind farm as described in any one of claims 10 to 12.

Images