A method and system for correcting wind speed prediction in an integrated wind farm

Abstract

The present application relates to an integrated wind farm forecast wind speed correction method and system. Compared with the traditional method, a single correction method is difficult to capture different characteristics under various time and space conditions, resulting in poor reliability of the correction results. The present application scheme establishes a variety of correction models suitable for different weather conditions, trains and optimizes the models through machine learning and correction curve methods, and obtains multiple different correction values ​​through the trained model. Then, a subjective and objective combined weighting method is used to more flexibly use multiple methods to improve the final comprehensive correction effect.

Description

Technical Field

[0001] The present application relates to the field of wind farm operation and maintenance management, and in particular to an integrated wind farm predicted wind speed correction method, system, electronic device and storage medium. Background Art

[0002] The power forecast of a wind farm is a forecast of the power output of the entire wind farm within a certain period of time in the future. The current mainstream of wind farm power forecast is distributed forecasting. Step-by-step forecasting refers to dividing power forecasting into two steps: 1. Based on relevant meteorological factors, first predict the influencing factors (wind speed, wind direction) that directly act on the wind turbine generator set; 2. Based on the influencing factors, the prediction of the unit output power is realized through the prediction model of the wind turbine generator set. Compared with direct forecasting, step-by-step forecasting is clear in the physical sense, independent and reliable in organization and implementation, and more flexible in cross-time scale research and modeling methods, so most of them use step-by-step forecasting methods.

[0003] However, due to the limitations of technologies such as numerical weather forecasting and large meteorological models, such as the uncertainty of the initial field, the lack of precision in time and space, the unavoidable nonlinear relationship problems in the calculation process, or the overfitting of large models, there are some deviations between the wind speed prediction and the actual measured wind speed results. At present, relevant researchers at home and abroad mostly use single methods such as mathematical statistics and deep learning, combined with a large number of meteorological observations in the area where the station is located, to revise the wind speed prediction results.

[0004] There are two problems with the above correction techniques:

[0005] 1. As a chaotic system, the meteorological system will present different characteristics under different environments, climates and weather conditions. It is difficult to capture the different characteristics under various time and space conditions using a single correction method.

[0006] 2. Due to the location of the observation point, as well as the accuracy and age of the observation equipment, the local observation data may deviate from the actual wind speed value that affects the power generation of the wind farm.

[0007] The above two problems will directly lead to the unsatisfactory wind speed correction effect, which cannot meet the actual production needs, and thus affect the accuracy of power prediction. Summary of the invention

[0008] The embodiments of the present application provide an integrated wind farm predicted wind speed correction method, system, computer device and computer-readable storage medium to at least solve the problem of unreliable wind farm predicted wind speed correction results in the related art.

[0009] In a first aspect, an embodiment of the present application provides a method for correcting wind speed forecast of an integrated wind farm, the method comprising:

[0010] Obtaining a whole-field power curve of the wind farm, and obtaining a whole-field wind speed at each time point in the historical period based on the whole-field power curve in the historical period;

[0011] Associating the entire wind speed and the entire wind direction time series to obtain training set data;

[0012] Based on the training set data, a plurality of wind speed correction models are respectively constructed, wherein a first wind speed correction model for normal weather and extreme weather is obtained by training a pre-constructed sLSTM (Stacked Long - Short Term Memory, abbreviated as sLSTM, stacked long short-term memory network) time series two-state model, and a second wind speed correction model for normal weather is constructed based on the correction curve;

[0013] Through multiple wind speed correction models, a forecast is made based on the weather forecast wind speed in any future target period to obtain multiple wind speed correction values, and the wind speed correction result is obtained by performing weighted assignment operation on the multiple wind speed correction values.

[0014] In some embodiments, the sLSTM time series two-state model includes: a representation layer consisting of at least two sLSTM layers, and a first MLP (Multilayer Perceptron) branch and a second MLP branch parallel to each other, wherein:

[0015] Preprocessing the training set data to obtain optimized training set data with continuity and standardization;

[0016] Through the representation layer, using the index gating mechanism and the memory mixing mechanism, the load pattern and the long-term dependency characteristics in the time series of the optimized training set data are captured to obtain the global features;

[0017] According to the weather conditions corresponding to the global features, the first MLP branch or the second MLP branch is used to perform update training based on the global features to obtain a wind speed correction result.

[0018] In some embodiments, according to the weather conditions corresponding to the global features, the first MLP branch or the second MLP branch is used to perform update training based on the global features to obtain a corrected wind speed correction value, which includes:

[0019] In the case where it is identified that the global feature corresponds to normal weather, the first MLP branch is activated and the second MLP branch is disabled,

[0020] Through the first MLP branch, the global features are nonlinearly transformed, the global features are mapped to a high-dimensional space, and the global features are nonlinearly activated using an activation function to obtain a high-level feature representation, and the high-level feature representation is mapped into a corrected result of the entire wind speed under normal weather conditions through an output layer;

[0021] In the case where it is identified that the global feature corresponds to extreme weather, the second MLP branch is activated and the first MLP branch is shielded,

[0022] The global features are nonlinearly transformed through the second MLP branch, the global features are mapped to a high-dimensional space, and the global features are nonlinearly activated using an activation function to obtain a high-level feature representation, and the high-level feature representation is mapped to the entire wind speed correction result under extreme weather conditions through an output layer.

[0023] The first MLP branch and the second MLP branch respectively include multiple processing layers, through which the global features are abstractly represented and refined for multiple iterations, and the weights between the processing layers are adjusted during the iteration process.

[0024] In some embodiments, associating the entire wind speed and the entire wind direction time series to obtain training set data includes:

[0025] Based on the whole-field power curve, obtaining the whole-field power with a time sequence relationship within any target historical period;

[0026] Based on the whole-field power curve, the whole-field wind speed at each time point in the target historical period is calculated by reverse search and difference method;

[0027] The entire wind speed and the entire wind direction at each time point in the target historical period are obtained by associating the data points at each time point through a time series, and the data points are aggregated to obtain training set data including historical time points, wind speed parameters, and wind direction parameters.

[0028] In some embodiments, constructing a second wind speed correction model for normal weather by curve fitting includes:

[0029] Obtain the weather forecast wind speed at the time when each of the data points is located;

[0030] Combining each data point and its corresponding weather forecast wind speed into a two-dimensional plane point, and dividing each two-dimensional plane point into different sectors of a circular area according to the wind direction data corresponding to the two-dimensional plane point, wherein the circular area includes multiple levels, and each level is divided into multiple sectors;

[0031] The weather forecast wind speed of each data point is taken as the X-axis, and the whole wind speed of each data point is taken as the Y-axis to construct the two-dimensional wind speed data corresponding to each data point;

[0032] By averaging the two-dimensional wind speed data on the X-axis and the Y-axis within the wind speed interval of the X-axis, the average wind speed data of each wind speed interval is obtained, and the average wind speed data of each wind speed interval is combined to obtain the wind speed correction curve corresponding to each sector.

[0033] In some embodiments, obtaining multiple wind speed correction values ​​corresponding to the weather forecast wind speed for any target period in the future through multiple wind speed correction models includes:

[0034] In the case where the weather forecast wind speed in the target period belongs to normal weather, the weather forecast wind speed in any target period is input into the first wind speed correction model, and prediction is performed through the representation layer and the first MLP branch of the first wind speed correction model to obtain a first correction value corresponding to the entire wind speed in normal weather in the target period;

[0035] In the case where the weather forecast wind speed in the target time period belongs to extreme weather, the weather forecast wind speed in any target time period is input into the second wind speed correction model, and prediction is performed through the representation layer and the second MLP branch of the first wind speed correction model to obtain a second correction value corresponding to the entire wind speed under extreme weather in the target time period;

[0036] Input the weather forecast wind speed of any target period into the second wind speed correction model for prediction, and obtain the target sector and adjacent sectors corresponding to the weather forecast wind speed of the target period;

[0037] According to the correction curves corresponding to the target sector and the adjacent sectors, the weather forecast wind speed of the target time period is corrected respectively to obtain a third correction value corresponding to the whole wind speed.

[0038] In some embodiments, the wind speed correction result obtained by performing weighted assignment operation on multiple wind speed correction values ​​includes:

[0039] The subjective weights corresponding to each wind speed correction value are obtained by the analytic hierarchy process, and the objective weights corresponding to each wind speed correction value are obtained by the entropy weight method.

[0040] The subjective weight and the objective weight are processed by a multiplication synthesis method to obtain a combined weight;

[0041] Based on the combined weight, the first correction value, the second correction value and the third correction value are weighted and assigned to obtain the wind speed correction result.

[0042] In some embodiments, obtaining the entire wind farm power curve includes:

[0043] Obtaining a power curve of each wind turbine generator set in the wind farm;

[0044] By performing interval unification, linear interpolation, redundant point elimination, and boundary value interpolation on the power curves of each wind turbine generator set, an optimized power curve unified on the X-axis and the Y-axis is obtained;

[0045] At any target wind speed point, the power values ​​of each optimized power curve are found, and the total power of the wind farm at the target wind speed value is obtained by adding up the power values.

[0046] In a second aspect, an embodiment of the present application provides an integrated wind farm predicted wind speed correction system, the system comprising: an acquisition module, a training module and a correction module, wherein:

[0047] The acquisition module is used to acquire the whole-field power curve of the wind farm, and based on the whole-field power curve in the historical period, obtain the whole-field wind speed at each time point in the historical period;

[0048] And, associating the entire wind speed and the entire wind direction time series to obtain training set data;

[0049] The training module is used to construct a plurality of wind speed correction models based on the training set data, wherein the first wind speed correction models for normal weather and extreme weather are obtained by training the pre-constructed sLSTM time series two-state model, and the second wind speed correction model for normal weather is constructed based on the correction curve;

[0050] The correction module is used to predict the wind speed based on the weather forecast wind speed in any future target period through multiple wind speed correction models to obtain multiple wind speed correction values, and obtain the wind speed correction result by performing weighted assignment operation on the multiple wind speed correction values.

[0051] In a third aspect, an embodiment of the present application provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in the first aspect above when executing the computer program.

[0052] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method of the first aspect described above.

[0053] Compared with the related art, the integrated wind farm predicted wind speed correction method provided by the embodiment of the present application is difficult to capture different characteristics in various time and space conditions through a single correction method compared to the traditional method, resulting in the problem of poor reliability of the correction result; the present application scheme establishes a variety of correction models suitable for different weather conditions, trains and optimizes the model through machine learning and correction curve methods, and obtains multiple different correction values ​​through the trained model. Then, a combination of subjective and objective weighting methods is used to more flexibly use multiple methods to improve the final comprehensive correction effect; at the same time, compared with the traditional technology that uses resampling or downsampling when the amount of training data is small, there are deviations and distortions caused by oversampling; the sLSTM time series two-state correction model proposed in the present application can retain the original distribution of the data while fully training the model representation layer, ensuring that the model recognizes the actual frequency of occurrence of extreme weather and avoids deviations and distortions caused by oversampling. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0055] Figure 1 is a flow chart of a method for correcting wind speed forecast of an integrated wind farm according to an embodiment of the present application;

[0056] Figure 2 is a schematic diagram of the distribution of wind farm generators according to an embodiment of the present application;

[0057] Figure 3a is a schematic diagram of an original power curve according to an embodiment of the present application;

[0058] Figure 3b is a schematic diagram of normalizing the root value point intervals of the power curve on the X-axis according to an embodiment of the present application;

[0059] Figure 3c is a schematic diagram of performing linear interpolation on a power curve according to an embodiment of the present application;

[0060] Figure 3d is a schematic diagram of performing boundary value interpolation processing on a power curve according to an embodiment of the present application;

[0061] Figure 4 It is a schematic diagram of obtaining the whole field power curve by summing the power of each unit according to an embodiment of the present application;

[0062] Figure 5 is a schematic diagram of obtaining the whole-field wind speed according to the whole-field power and the whole-field power curve according to an embodiment of the present application;

[0063] Figure 6 is a schematic diagram of the structure of the sLSTM model according to an embodiment of the present application;

[0064] Figure 7 is a schematic diagram of dividing each data point into sectors according to an embodiment of the present application;

[0065] Figure 8 is a schematic diagram of a combination result based on a weather forecast wind speed and a whole-field wind speed according to an embodiment of the present application;

[0066] Fig. 9 is a schematic diagram of a curve after correction according to an embodiment of the present application;

[0067] Fig.10 is a schematic diagram of using a correction curve to correct wind speed according to an embodiment of the present application;

[0068] Fig.11 is a schematic diagram of adjacent sectors according to an embodiment of the present application;

[0069] Fig.12 is a structural block diagram of an integrated wind farm predicted wind speed correction system according to an embodiment of the present application;

[0070] Fig.13 It is a schematic diagram of the internal structure of an electronic device according to an embodiment of the present application.

[0071] Description of reference numerals:

[0072] 1. Data points in sector SEE: including time, wind direction, whole-field wind speed obtained through power curve, and corresponding meteorological forecast wind speed. DETAILED DESCRIPTION

[0073] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is described and illustrated below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. Based on the embodiments provided in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application.

[0074] Obviously, the drawings described below are only some examples or embodiments of the present application. For ordinary technicians in this field, the present application can also be applied to other similar scenarios based on these drawings without creative work. In addition, it can also be understood that although the efforts made in this development process may be complicated and lengthy, for ordinary technicians in this field related to the content disclosed in this application, some changes in design, manufacturing or production based on the technical content disclosed in this application are just conventional technical means, and should not be understood as insufficient content disclosed in this application.

[0075] Reference to "embodiments" in this application means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those of ordinary skill in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0076] Unless otherwise defined, the technical terms or scientific terms involved in this application should be understood by people with ordinary skills in the technical field to which this application belongs. The words "one", "a", "a", "the" and the like involved in this application do not indicate a quantity limitation, and may indicate the singular or plural. The terms "include", "comprise", "have" and any of their variations involved in this application are intended to cover non-exclusive inclusions; for example, a process, method, system, product or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units that are not listed, or may also include other steps or units inherent to these processes, methods, products or devices. The words "connect", "connected", "coupled" and the like involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The "multiple" involved in this application refers to two or more. "And / or" describes the association relationship of associated objects, indicating that there may be three relationships, for example, "A and / or B" can mean: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects before and after are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific ordering of the objects.

[0077] The power forecast of a wind farm is a forecast of the power output of the entire wind farm within a certain period of time in the future (the ultra-short-term power forecast is 15 minutes to 4 hours, the short-term power forecast is from 0:00 the next day to the next 72 hours, the medium-term power forecast is from 0:00 the next day to the next 240 hours, and the long-term power forecast is the monthly power and total power for the next 12 months)*1.

[0078] Wind farm power generation prediction can be divided into direct prediction and step-by-step prediction based on the implementation steps. Currently, the mainstream is distributed prediction. Step-by-step prediction means that power prediction is divided into two steps:

[0079] 1. Based on relevant meteorological factors, first predict the influencing factors (wind speed and wind direction) that directly affect the wind turbine generator set;

[0080] 2. Based on the influencing factors, the prediction model of wind turbines is used to predict the output power of the units. Compared with direct prediction, step-by-step prediction is clear in the physical sense, independent and reliable in organization and implementation, and more flexible in cross-time scale research and modeling methods, so the step-by-step prediction method is currently mostly used.

[0081] The first step of the step-by-step prediction method for wind power generation prediction is crucial to the prediction accuracy of wind speed and wind direction as influencing factors: as a general understanding in the industry, the error in the first step accounts for about 50-70% of the overall error in power prediction. The actual wind speed and wind direction are easily affected by the local microclimate characteristics in complex geographical environments. The predicted wind speed and wind direction obtained by numerical weather forecasting or large meteorological modeling technology will lack resolution and accuracy due to insufficient observed data, excessive computing resources, and inadequate modeling. As a result, they are quite different from the actual wind direction and wind speed in complex geographical environments, resulting in poor power prediction results.

[0082] Wind speed is usually expressed in units of m / s, and wind direction is generally expressed in two ways: the first is due north as 0° or 360°, due east as 90°, due south as 180°, and due west as 270°; the second is due north as 0°, due east as 90°, due south as ±180°, and due west as -90°. In terms of wind direction expression, in addition to the precise angle expression method, a wind rose diagram*2 (8 directions or 16 directions) is also usually used to indicate the frequency of different wind direction ranges.

[0083] In the actual wind power generation process, since the yaw system of the wind turbine adjusts the wind direction in real time to achieve maximum power generation, the influence of wind direction on the actual power generation can be ignored to a certain extent, and only the correlation of wind speed is considered.

[0084] Due to the limitations of technologies such as numerical weather forecasting and large meteorological models, such as the uncertainty of the initial field, insufficient precision in time and space, the unavoidable problem of nonlinear relationships in the calculation process, or the overfitting of large models, there are deviations between wind speed predictions and actual measured wind speed results.

[0085] At present, relevant researchers at home and abroad mostly use single methods such as mathematical statistics and deep learning, combined with a large number of meteorological observations in the area where the station is located, to revise the wind speed forecast results.

[0086] There are two problems with the above technology:

[0087] 1. As a chaotic system, the meteorological system will present different characteristics under different environments, climates and weather conditions. It is difficult to capture the different characteristics under various time and space conditions using a single correction method.

[0088] 2. Due to the location of the observation point, as well as the accuracy and age of the observation equipment, the local observation data may deviate from the actual wind speed value that affects the power generation of the wind farm.

[0089] The above two problems will directly lead to the unsatisfactory wind speed correction effect, which cannot meet the actual production needs and affect the accuracy of power prediction.

[0090] In view of this, an embodiment of the present application provides a method for correcting wind speed prediction of an integrated wind farm. Figure 1 is a flow chart of a method for correcting wind speed forecast of an integrated wind farm according to an embodiment of the present application, such as Figure 1 As shown, the process includes the following steps:

[0091] S101, obtaining the whole-field power curve of the wind farm, and obtaining the whole-field wind speed at each time point in the historical period based on the whole-field power curve and in the historical period;

[0092] Specifically, in this embodiment, the power curve of the entire field is obtained by accumulating the power of each independent wind turbine. Specifically, this step includes the following subdivision steps:

[0093] Step 1, obtain the power curve of each wind turbine in the wind farm;

[0094] Among them, the power is usually a relationship between wind speed and generator output power, showing the power that the generator can generate at different wind speeds. This curve can be obtained by actual measurement from each unit;

[0095] In addition, if the measured power curve cannot be directly obtained, the corresponding theoretical power can be obtained according to the factory parameters of the unit.

[0096] In addition, before this, the models and quantities of all wind turbines in the wind farm can be determined, and other reference parameters such as rated power and rated wind speed can be obtained, providing more supporting information for the uncalculated overall power; Figure 2 It is a schematic diagram of the distribution of wind turbines in an electric field according to an embodiment of the present application.

[0097] For example, if there is no significant difference between generator sets of the same model, the power curve of any one of them can be used as a representative.

[0098] Step 2, by performing interval unification, linear interpolation, redundant point elimination and boundary value interpolation on the power curves of each wind turbine generator set, an optimized power curve unified on the X-axis and the Y-axis is obtained;

[0099] Figure 2 is a schematic diagram of the distribution of wind farm generators according to an embodiment of the present application, Figure 2 The properties of each signal unit in the station are shown in the following table:

[0100]

[0101] It is understandable that the wind turbines in a wind farm may come from different manufacturers or models. Even if they are of the same model, their power curves may be slightly different due to different operating conditions. The distribution of their power curves on wind speed (X-axis) and power output (Y-axis) may not be consistent. For example, the data points of some curves may be denser or sparser on the wind speed axis.

[0102] Furthermore, the actual measured power curve may have missing data points. If these missing points are not processed, the accuracy of subsequent calculations will be affected. The original power curve may contain too many detailed data points, which may be repeated or irrelevant to the calculation.

[0103] Therefore, in order to perform effective calculations, the power curves of each wind turbine generator set are normalized at the wind speed value point intervals on the X-axis in turn, and each power curve is unified at the same wind speed interval, and the power output of each generator is compared and added at the same wind speed.

[0104] In addition, by linear interpolation, missing power data points are filled to obtain a continuous power curve, ensuring that there is a corresponding power value at each uniform wind speed point. By eliminating redundant value points, the power curve is simplified, making the calculation more efficient and reducing the consumption of computing resources.

[0105] Finally, near the minimum or maximum wind speed, the power curve may lack sufficient data points or have outliers. Through boundary value interpolation, we can ensure that the power curve is reasonable and continuous in these extreme cases and avoid abrupt calculation errors at the boundaries.

[0106] It can be understood that after the above optimization steps, on the basis of the unification of the X-axis (wind speed), it is ensured that each wind turbine has a corresponding power value at the same wind speed. Since the power curves of all generator sets go through the same processing steps, the power value of the final optimized power curve on the Y-axis will correspond to the same wind speed point. The X-axis represents the wind speed, and the Y-axis parameter directly represents the power value. Through the processing of Step 2, although the original power curves of different generators may be different, after optimization, at the same wind speed, the power values ​​of all generators can be directly compared and added.

[0107] Step 3, at any target wind speed point, find the power value of each optimized power curve, and obtain the total power of the wind farm at the target wind speed value by adding up the power values;

[0108] Figure 3a is a schematic diagram of an original power curve according to an embodiment of the present application; Figure 3b is a schematic diagram of normalizing the root value point intervals of the power curve on the X-axis according to an embodiment of the present application; Figure 3c is a schematic diagram of performing linear interpolation on a power curve according to an embodiment of the present application; Figure 3d is a schematic diagram of performing boundary value interpolation processing on a power curve according to an embodiment of the present application;

[0109] like Figure 3a , Figure 3b , Figure 3c and Figure 3d As shown, the X-axis wind speed and Y-power coordinate axes of the power curves of all wind turbines are unified by normalizing the intervals of wind speed value points on the X-axis, linear interpolation or elimination of redundant value points, and boundary value interpolation based on the model with the largest rated wind speed.

[0110] Specifically, for each wind speed parameter on the X-axis, the power value of each wind turbine at the wind speed is found through the optimized power curve, and the power values ​​of all turbines at the target wind speed are added together to obtain the total output power of the entire wind farm at the wind speed. Figure 4 It is a schematic diagram of obtaining the power curve of the entire field by summing up the power of each unit according to an embodiment of the present application.

[0111] Through the above steps, for extreme weather conditions such as strong winds and gusty winds, the threshold method is used to filter out the data under extreme weather conditions in the initial data set, and the weather forecast data set is labeled with the state label of normal weather or extreme weather to facilitate model training. In addition, the historical wind speed and wind direction data of the entire field and the weather forecast wind speed and wind direction data of the corresponding time period, as well as the data identified according to the standard, are used as data sets to train the sLSTM time series dual-state multiple correction model.

[0112] S102, associating the entire wind speed and the entire wind direction time series to obtain training set data;

[0113] The purpose of this step is to associate the entire wind speed and the entire wind direction in the time series to obtain input data suitable for LSTM model training. Specifically, this step includes the following subdivision steps:

[0114] Step 1, based on the whole field power curve, obtain the whole field power with a time series relationship in any target historical period;

[0115] The whole-field power curve is used to collect the actual power generation data of the wind farm in the past period of time. These data are usually recorded in the form of time series, with each time point (such as every hour or every minute) corresponding to a power generation value. These historical power data will be used as the basic data in subsequent calculations to infer the wind speed at these time points.

[0116] Step 2, based on the whole power curve, calculate the whole wind speed at each time point in the historical period through reverse search and difference method;

[0117] It can be understood that this step is based on the power curve. Under the premise of knowing the Y-axis parameter power, the entire wind speed can be obtained by obtaining the corresponding X-axis coordinate value in the curve; Figure 5 It is a schematic diagram of obtaining the whole-field wind speed according to the whole-field power and the whole-field power curve according to an embodiment of the present application.

[0118] Step 3, associate the entire wind speed and the entire wind direction at each time point through time series to obtain the data point at each time point, and aggregate the data points to obtain the training set data containing historical time points, wind speed parameters and wind direction parameters.

[0119] The time series predicted wind direction at the typical location of the wind farm within the time period is further obtained as the wind direction of the entire field; the wind direction data is also saved in the form of a time series, which corresponds one-to-one with the power data at each time point; it should be noted that in this embodiment, the wind direction is not directly corrected, but under complex terrain conditions, different wind directions have a greater impact on wind speed, so wind direction needs to be used as an input variable of the model.

[0120] Finally, the entire wind speed and the entire wind direction are correlated through time series to form a complete historical time-wind speed-wind direction dataset for training the wind speed correction model.

[0121] S103, based on the training set data, constructing a plurality of wind speed correction models respectively, wherein a first wind speed correction model for normal weather and extreme weather is obtained by training a pre-constructed sLSTM time series two-state model, and a second wind speed correction model for normal weather is constructed based on the correction curve;

[0122] Specifically, in step 1, the first wind speed correction model under normal weather and extreme weather is obtained by training based on the training set data, including:

[0123] Step 1.1, pre-process the training set data to obtain optimized training set data with continuity and standardization;

[0124] Among them, the training set data is preprocessed to ensure the continuity and standardization of the data. This step is to optimize the quality of the data and eliminate noise, missing values ​​or outliers in the data so that the model can better learn and extract time series features.

[0125] Specifically, when dealing with missing points or discontinuous time series in the data, it may be necessary to fill in the missing data through interpolation, smoothing, or other methods; it is also possible to transform input features of different scales into the same range so that the sLSTM model can process and learn more effectively. It is also possible to detect and remove outliers to ensure the quality of input data.

[0126] Optionally, in addition to the weather forecast wind speed data, weather forecast temperature, humidity and air pressure data may also be used as optional data to improve the model accuracy.

[0127] Step 1.2, input the optimized training set data into the pre-built sLSTM model;

[0128] Among them, sLSTM is a deep learning model that can capture and learn long-term and short-term dependencies in time series data. In this embodiment, the preprocessed data is input into a pre-built sLSTM model. These data contain time series information, such as wind speed, wind direction, power and other features.

[0129] Step 1.3, the sLSTM model is composed of a representation layer through multiple slstm layers, and uses the exponential gating mechanism and the memory mixing mechanism to capture the load pattern and long-term dependency features in the time series of the optimized training set data layer by layer to obtain the global features.

[0130] Figure 6is a schematic diagram of the structure of the sLSTM model according to an embodiment of the present application, such as Figure 6 As shown in the figure, specifically, the sLSTM model extracts time series features from the input data layer by layer through multiple feature extraction layers (i.e., sLSTM units). Each layer can capture different levels of temporal dependencies, thereby constructing higher-level global features. Through these hierarchical feature extractions, the final output is a high-level feature representation of time series data, which contains complex dynamic patterns and trends in the time series.

[0131] In this embodiment, the sLSTM model introduces an exponential gating mechanism, which can more finely control the flow of information in the cell state. Compared with the gating mechanism of the traditional LSTM, the exponential gating can respond more sensitively to changes in the time series;

[0132] Correspondingly, in this scheme, in wind speed prediction, when the wind speed changes suddenly, the exponential gating can adjust the cell state more quickly to adapt to this change, thereby better capturing the short-term fluctuations in the wind speed time series; for other variables with time series characteristics such as wind direction and power, the exponential gating mechanism can also dynamically adjust the model's attention to different features according to the speed and amplitude of their changes, thereby improving the model's adaptability to complex time series data.

[0133] In addition, long-term dependency is a key issue when processing time series data. Traditional LSTM may suffer from information decay or memory fuzziness when processing data with a long time span. This application uses sLSTM to effectively retain long-term dependency data through a memory hybrid mechanism, especially in extreme weather forecasting, and maintains effective memory of historical data.

[0134] In Step 1.4, the sLSTM model is updated and trained based on the global features output in Step 1.3 through the parallel first MLP branch, second MLP branch and output layer according to the weather conditions corresponding to the global features, to obtain the correction result of the entire wind speed.

[0135] In this embodiment, the global features generated by the representation layer are then passed to two MLP branches, one corresponding to normal weather patterns and the other corresponding to extreme weather patterns. When the weather state of the sample (normal weather or extreme weather) is read, the model activates the corresponding branch and calculates the loss of the branch, while shielding the updates of other branches to ensure that each branch specifically learns the data pattern under this state. Through the fully connected layer and the output layer, the model converts the global feature representation into the output of the corresponding branch pattern, thereby achieving accurate prediction under different weather conditions.

[0136] Specifically, it can be determined whether each sample in the data is in normal weather or extreme weather based on established standards. Optionally, the wind speed can be converted according to the national standard "Wind Force Level", and extreme weather is defined as an average wind speed of above Level 6 or a gust of above Level 7 within 1 hour.

[0137] (1) For the normal weather mode branch:

[0138] When the model determines that the input sample is in normal weather conditions, this branch is activated. It receives global features generated by the sLSTM representation layer. These features contain the time-dependent features of the input data extracted by the sLSTM layer under normal weather conditions.

[0139] The first MLP uses its multiple fully connected layers to further nonlinearly transform and process these features. Through the weight adjustment and activation function between different layers, more advanced and representative features are gradually extracted, which are optimized for common weather patterns.

[0140] After multiple layers of processing and output layers, the high-level feature representation is converted into output corresponding to normal weather conditions through the fully connected layer and output layer. This output can be the predicted value of some meteorological parameters related to normal weather, such as the reasonable range or trend of temperature, humidity, air pressure, etc. under normal weather conditions.

[0141] (2) For extreme weather pattern branches

[0142] When the model determines that the input sample is in extreme weather conditions, this branch is activated. It receives global features generated by the sLSTM representation layer, but at this time these features reflect the data characteristics under extreme weather conditions.

[0143] The second MLP branch processes the characteristics of extreme weather. Since extreme weather is very different from normal weather in terms of meteorological characteristics, MLP needs to learn different patterns and laws. For example, in extreme weather, parameters such as wind speed and air pressure may have greater fluctuations and outliers, and the second MLP needs to be able to capture these characteristics and process them accordingly. By adjusting the weights of different layers and selecting activation functions, the second MLP gradually extracts high-level features suitable for extreme weather patterns.

[0144] Finally, the processed features are converted into outputs corresponding to extreme weather conditions through the fully connected layer and the output layer. This output may include predictions of the intensity and duration of extreme weather phenomena such as strong winds, heavy rains, and blizzards.

[0145] Throughout the process, the two branches are independent of each other. When one branch is activated, the update of the other branch is blocked, ensuring that each branch specifically learns the data pattern under the corresponding state. In addition, each branch includes multiple processing layers, and each processing layer repeats the operation of the previous step to further linearly transform and nonlinearly activate the output of the previous layer. Each layer can be regarded as an abstraction and refinement of features, gradually extracting more advanced and representative features. By continuously adjusting the weights between layers, the model can better adapt to the data distribution under normal weather conditions, thereby improving the prediction accuracy of the model under different weather conditions.

[0146] Among them, the wind speed forecast value generated by the sLSTM model through the fully connected layer and the output layer based on the extraction of time series features is the "corrected wind speed forecast value for the entire field". The "correction" here usually refers to the model continuously adjusting parameters by optimizing the loss function during the training process to make the predicted wind speed value as close to the actual value as possible. Specifically, the correction process may involve multiple factors, such as the influence of wind direction under complex terrain, error correction in historical data, etc. The model improves the accuracy of the prediction by learning the relationship between these factors and wind speed.

[0147] It should be noted that during the model training process, the gradient of the loss function relative to the model parameters is calculated through the back propagation algorithm, and the model parameters are updated using the gradient descent method. The goal is to minimize the loss function so that the error between the model's predicted results and the actual values ​​is as small as possible; the loss function measures the error between the model's predicted values ​​and the actual values, and common loss functions include mean square error (MSE) and so on.

[0148] In addition, in traditional technology, in order to fully train the extreme weather correction model with a small amount of original data, a resampling method is generally used, which will cause the frequency of extreme weather in training to far exceed the actual situation, resulting in deviations and distortions caused by oversampling. In this regard, the sLSTM time series two-state correction model proposed in this application scheme can retain the original distribution of the data while fully training the model representation layer, and at the same time ensure that the model recognizes the actual frequency of extreme weather, avoiding deviations and distortions caused by oversampling.

[0149] In addition, the exponential gating mechanism of sLSTM improves dynamic response capabilities and quickly adapts to meteorological changes. The memory hybrid mechanism effectively retains long-term dependent data, especially in extreme weather forecasting. At the same time, the standardized state mechanism solves the problem of gradient explosion or disappearance, ensuring the stability and training efficiency of the model in large-scale meteorological data.

[0150] Specifically, in step 3, the second wind speed correction model under normal weather conditions is obtained by training based on the training set data, including:

[0151] Step 3.1, obtain the weather forecast wind speed at each data point;

[0152] This step is to obtain the weather forecast wind speed and wind direction data in the corresponding time period from the training set data. It should be noted that, in combination with the above step S101, it can be known that the wind direction in the training set data is consistent with the wind direction data in the weather forecast.

[0153] Step 3.2, combine each data point and its corresponding weather forecast wind speed into a two-dimensional plane point, and divide each data point into different sectors of the circular area according to the wind direction data corresponding to the data point; Figure 7 is a schematic diagram of dividing each data point into sectors according to an embodiment of the present application, such as Figure 7 As shown, the correction circle includes multiple levels, and each level is equally divided into multiple sectors;

[0154] Step 3.3, construct the two-dimensional wind speed data corresponding to each data point by taking the weather forecast wind speed of each data point as the X-axis and the whole wind speed of each data point as the Y-axis;

[0155] In each wind direction sector, the forecast wind speed is used as the X-axis and the actual wind speed is used as the Y-axis to form two-dimensional data points. These two-dimensional data points describe the relationship between the forecast wind speed and the actual wind speed in each wind speed interval.

[0156] If there are insufficient wind speed data points in a sector, data points from two adjacent sectors can be used to supplement them to ensure sufficient data volume, thereby improving the reliability of the correction curve.

[0157] Step 3.4, by averaging the two-dimensional wind speed data on the X-axis and the Y-axis within the wind speed range of the X-axis, the average wind speed data of each wind speed range is obtained; and based on the average wind speed data of each wind speed range, the wind speed correction curve corresponding to each sector is obtained.

[0158] Optionally, the x-axis (weather forecast wind speed) can be divided into multiple 2 m / s intervals, such as 0-2 m / s, 2-4 m / s, etc. Further, in each 2 m / s interval, all two-dimensional data points are averaged on the x-axis and y-axis respectively, and the average x value (weather forecast wind speed) and average y value (actual wind speed) in the interval are calculated. The average data point of each wind speed interval is calculated to represent the average relationship between the forecast wind speed and the actual wind speed in the wind speed interval.

[0159] Finally, a correction curve is formed by connecting the average data points within each interval.

[0160] Figure 8is a schematic diagram of a combination result based on a weather forecast wind speed and a whole-field wind speed according to an embodiment of the present application, Fig. 9 is a schematic diagram of a curve after correction according to an embodiment of the present application; Figure 8 and Fig. 9 As shown, the curve shows the relationship between the weather forecast wind speed and the actual wind speed throughout the field, which can be used to correct the forecast wind speed.

[0161] It can be understood that the model based on the correction curve established by the method of the above step Step 3 can correct the weather forecast wind speed to a value closer to the actual wind speed, thereby improving the accuracy of wind speed prediction. The wind speed correction through the correction curve is mainly applicable to normal weather conditions, but when there is less historical training data, the prediction results obtained by the ordinary weather branch of the first correction model are more reliable. When there is more historical training data, the processing effect for linear relationships can be similar to that of the first correction model, but the processing effect for nonlinear relationships is also weak.

[0162] Through the above step S103, the sLSTM time series two-state model meets the needs of different weather conditions through a dual-branch design. Under normal weather conditions, the general weather branch of the model can make stable prediction corrections based on the conventional wind speed pattern to ensure accuracy and efficiency. In extreme weather, the extreme weather branch can quickly capture sudden changes in wind speed and influence the final prediction correction results through higher weights, thereby improving the response speed and prediction accuracy of the model under severe weather conditions. This flexible dual-branch structure ensures that the model can provide targeted predictions under different weather conditions to meet the needs of conventional and extreme environments. Secondly, the correction curve model has better reliability when there is less training data available, and can be applied to newly established stations or stations where historical data lacks good management. S104, through multiple wind speed correction models, predictions are made based on the weather forecast wind speed of any target period in the future, and multiple wind speed correction values ​​are obtained, and the wind speed correction result is obtained by weighted operation of multiple wind speed correction values.

[0163] Specifically, the correction results obtained by the first correction model under normal weather or extreme weather include:

[0164] Step 1.1. Take the entire wind speed as the prediction target and the real-time weather forecast wind speed data as the input of the first correction model. If weather forecast temperature, humidity or air pressure data is used in the model training phase, such data also needs to be used as data input in the prediction phase.

[0165] Step 1.2, in the prediction stage, the trained first correction model receives the data input of the above steps in real time, and generates the corrected wind speed forecast value for the corresponding forecast period. Among them, under normal weather conditions, the normal weather branch of the model can make stable prediction corrections based on the conventional wind speed pattern. In extreme weather, the extreme weather branch can quickly capture sudden wind speed changes and influence the final prediction correction results with higher weights. In this way, the model can use historical power data and real-time meteorological forecast wind speed to improve the accuracy of the entire wind speed forecast.

[0166] In addition, the correction effect evaluation index of the wind speed correction model is the root mean square error RMSE:

[0167]

[0168] Among them, RMSE is the root mean square error, Oi is the whole-field wind speed data obtained at time i using real-time power and the whole-field power curve, and Pi is the wind speed forecast data at time i after correction for the whole field.

[0169] Finally, the correction results under normal weather conditions obtained through the second correction model include:

[0170] Step 2.1, in the prediction stage, the real-time weather forecast wind speed is corrected by the sector where the real-time weather forecast wind direction is located and the correction curve of the corresponding sector.

[0171] Fig.10 is a schematic diagram of using a correction curve to correct wind speed according to an embodiment of the present application;

[0172] Step 2.2, in order to improve the continuity of wind speed correction in different adjacent sectors, in addition to the correction curve in the corresponding sector, the correction curve of the closest sector can be used for correction, and the two correction results can be corrected by the inverse distance weighted method. For example, in the current weather forecast data, the wind direction data falls in sector SEE, and the closest sector is sector SE.

[0173] Fig.11 is a schematic diagram of adjacent sectors according to an embodiment of the present application. It can be understood that, in combination with the above multiple sectors, the final correction result can be obtained by the following formula:

[0174]

[0175] Among them, V ID is the revised wind speed result obtained by the inverse distance weighted method, in m / s; V SEE is the corrected wind speed result obtained by the correction curve in the sector SEE, in m / s, V SE Similarly, DSEE is the angle between the weather forecast wind direction and the central axis of the sector SEE, in degrees (can also be expressed in radians), D SE Same reason.

[0176] After obtaining the various correction results, in this embodiment, the final wind speed correction result is obtained by weighting these correction results, which specifically includes the following steps:

[0177] Step 1: Obtain the subjective weight corresponding to each wind speed correction value through the hierarchical analysis method, and obtain the objective weight corresponding to each wind speed correction value through the entropy weight method;

[0178] In this embodiment, a judgment matrix is ​​constructed based on the applicability, advantages and disadvantages of the above methods to compare the relative importance of each pair of wind speed correction methods. The element values ​​of the matrix are based on experience and judgment, reflecting the importance of each method relative to other methods. The subjective weight of each method is determined by calculating the eigenvector of the judgment matrix.

[0179] In addition, the actual performance data of each wind speed correction method under various weather conditions are collected to construct a decision matrix. For each column in the matrix (i.e., each method), the information entropy is calculated, where the information entropy reflects the degree of discreteness of the data. The higher the entropy value, the more uncertain the information; finally, the objective weight of each method is calculated based on the entropy value. The lower the entropy value, the higher the weight, indicating that the objective performance of the method in wind speed correction is more important.

[0180] Step 2, process the subjective weight and objective weight through multiplication synthesis method to obtain the combined weight;

[0181]

[0182] Where Wi is the combined weight of the i-th wind speed correction method, is the subjective weight, is the objective weight.

[0183] Step 3, based on the combined weight, perform weighted assignment operation on the first correction value, the second correction value and the third correction value to obtain the wind speed correction result.

[0184] Specifically, the weighted calculation can be performed through the following formula:

[0185]

[0186] Among them, Vfinal is the final wind speed correction result, Wi is the combination weight of the i-th method, and Vi is the correction result of the i-th method.

[0187] Through the above steps S101 to S104, compared with the traditional method, it is difficult to capture different features in multiple spatiotemporal conditions through a single correction method, resulting in poor reliability of the correction results; the present application scheme establishes a variety of correction models suitable for different weather conditions, trains and optimizes the model through machine learning and correction curve methods, and obtains multiple different correction values ​​through the trained model, and uses a combination of subjective and objective weighting methods to use multiple methods more flexibly to improve the final comprehensive correction effect; at the same time, compared with the traditional technology, when the amount of training data is small, the resampling or downsampling methods used lead to the frequency of extreme weather in training far exceeding the actual situation, and then the deviation and distortion caused by oversampling occur. In this regard, the sLSTM time series two-state correction model proposed in this embodiment can retain the original distribution of the data while fully training the model representation layer, ensuring that the model recognizes the actual frequency of extreme weather and avoids the deviation and distortion caused by oversampling.

[0188] It should be noted that the steps shown in the above process or the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0189] The present application also provides an integrated wind farm forecast wind speed correction system, Fig.12 is a structural block diagram of an integrated wind farm predicted wind speed correction system according to an embodiment of the present application, such as Fig.12 As shown, the system comprises: an acquisition module 10, a training module 20 and a correction module 30, wherein:

[0190] The acquisition module 10 is used to acquire the whole-field power curve of the wind farm, and obtain the whole-field wind speed at each time point in the historical period based on the whole-field power curve in the historical period;

[0191] And, the whole wind speed and the whole wind direction time series are associated to obtain the training set data;

[0192] The training module 20 is used to construct a plurality of wind speed correction models based on the training set data, wherein a first wind speed correction model for normal weather and extreme weather is obtained by training a pre-constructed sLSTM time series two-state model, and a second wind speed correction model for normal weather is constructed based on the correction curve;

[0193] The correction module 30 is used to predict the wind speed in the weather forecast for any target period in the future through multiple wind speed correction models to obtain multiple wind speed correction values, and obtain the wind speed correction result by performing weighted assignment operation on the multiple wind speed correction values.

[0194] Through the above system, compared with the traditional method, it is difficult to capture different characteristics in various time and space conditions through a single correction method, resulting in poor reliability of the correction results; this system establishes a variety of correction models suitable for different weather conditions, trains and optimizes the models through machine learning and correction curve methods, and obtains multiple different correction values ​​through the trained models. Through the subjective and objective combined weighting method, multiple methods can be used more flexibly to improve the final comprehensive correction effect.

[0195] In one embodiment, Fig.13 is a schematic diagram of the internal structure of an electronic device according to an embodiment of the present application, such as Fig.13 As shown, an electronic device is provided, which may be a server, and its internal structure diagram may be as shown in Fig.13 The electronic device includes a processor, a memory, a network interface and a database connected through a system bus. The processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the electronic device is used to store data. The network interface of the electronic device is used to communicate with an external terminal through a network connection. The computer program is executed by the processor to implement a method for correcting wind speed forecasts for an integrated wind farm.

[0196] Those skilled in the art will understand that Fig.13 The structure shown in the figure is merely a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the electronic device to which the solution of the present application is applied. The specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0197] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0198] Those skilled in the art should understand that the various technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the various technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0199] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A method for correcting wind speed forecast of an integrated wind farm, characterized in that: The method comprises: Obtaining a whole-field power curve of the wind farm, and obtaining a whole-field wind speed at each time point in the historical period based on the whole-field power curve in the historical period; The entire wind speed and the entire wind direction are time-series-correlated to obtain the training set data including: Based on the whole-field power curve, obtaining the whole-field power with a time sequence relationship within any target historical period; Based on the whole-field power curve, the whole-field wind speed at each time point in the target historical period is calculated by reverse search and difference method; Acquire the data points of each time point by associating the whole wind speed and the whole wind direction at each time point in the target historical period through time series, and collect the data points to obtain training set data including historical time points, wind speed parameters and wind direction parameters; Based on the training set data, a plurality of wind speed correction models are respectively constructed, wherein a first wind speed correction model for normal weather and extreme weather is obtained by training a pre-constructed sLSTM time series two-state model, and a second wind speed correction model for normal weather is constructed based on the correction curve, wherein the sLSTM time series two-state model includes: a representation layer composed of at least two sLSTM layers, and a first MLP branch and a second MLP branch parallel to each other, the training set data is preprocessed to obtain optimized training set data with continuity and standardization; through the representation layer, the load pattern and long-term dependence characteristics in the time series of the optimized training set data are captured by using an exponential gating mechanism and a memory mixing mechanism to obtain global features; according to the weather conditions corresponding to the global features, the first MLP branch or the second MLP branch is used to perform update training based on the global features to obtain wind speed correction results; Obtain the weather forecast wind speed at the time when each of the data points is located; combine each of the data points and its corresponding weather forecast wind speed into a two-dimensional plane point, and divide each of the two-dimensional plane points into different sectors of a circular area according to the wind direction data corresponding to the two-dimensional plane point, wherein the circular area includes multiple levels, and each level is divided into multiple sectors; construct the two-dimensional wind speed data corresponding to each data point with the weather forecast wind speed of each data point as the X-axis and the whole wind speed of each data point as the Y-axis; obtain the average wind speed data of each wind speed interval by averaging the two-dimensional wind speed data on the X-axis and the Y-axis within the wind speed interval of the X-axis, and obtain the correction curve corresponding to each sector by combining the average wind speed data of each wind speed interval; Through multiple wind speed correction models, a forecast is made based on the weather forecast wind speed in any future target period to obtain multiple wind speed correction values, and the wind speed correction result is obtained by performing weighted assignment operation on the multiple wind speed correction values.

2. The method according to claim 1, characterized in that According to the weather conditions corresponding to the global features, the first MLP branch or the second MLP branch is used to perform update training based on the global features to obtain a wind speed correction result, including: In the case where it is identified that the global feature corresponds to normal weather, the first MLP branch is activated and the second MLP branch is disabled, Through the first MLP branch, the global features are nonlinearly transformed, the global features are mapped to a high-dimensional space, and the global features are nonlinearly activated using an activation function to obtain a high-level feature representation, and the high-level feature representation is mapped into a corrected result of the entire wind speed under normal weather conditions through an output layer; In the case where it is identified that the global feature corresponds to extreme weather, the second MLP branch is activated and the first MLP branch is shielded, The global features are nonlinearly transformed through the second MLP branch, the global features are mapped to a high-dimensional space, and the global features are nonlinearly activated using an activation function to obtain a high-level feature representation, and the high-level feature representation is mapped to the entire wind speed correction result under extreme weather conditions through an output layer. The first MLP branch and the second MLP branch respectively include multiple processing layers, through which the global features are abstractly represented and refined for multiple iterations, and the weights between the processing layers are adjusted during the iteration process.

3. The method according to any one of claims 1 to 2, characterized in that: The multiple wind speed correction values ​​corresponding to the weather forecast wind speed in any target period in the future obtained through multiple wind speed correction models include: In the case where the weather forecast wind speed in the target period belongs to normal weather, the weather forecast wind speed in any target period is input into the first wind speed correction model, and prediction is performed through the representation layer and the first MLP branch of the first wind speed correction model to obtain a first correction value corresponding to the entire wind speed in normal weather in the target period; In the case where the weather forecast wind speed in the target time period belongs to extreme weather, the weather forecast wind speed in any target time period is input into the first wind speed correction model, and prediction is performed through the representation layer and the second MLP branch of the first wind speed correction model to obtain a second correction value corresponding to the entire wind speed under extreme weather in the target time period; The weather forecast wind speed of any target time period is input into the second wind speed correction model for prediction, and the target sector and adjacent sectors corresponding to the weather forecast wind speed of the target time period are obtained. According to the correction curves corresponding to the target sector and the adjacent sectors, the weather forecast wind speed of the target time period is corrected respectively to obtain the third correction value corresponding to the entire wind speed.

4. The method according to claim 3, characterized in that By performing weighted assignment calculation on multiple wind speed correction values, the wind speed correction results include: The subjective weights corresponding to each wind speed correction value are obtained by the analytic hierarchy process, and the objective weights corresponding to each wind speed correction value are obtained by the entropy weight method. The subjective weight and the objective weight are processed by a multiplication synthesis method to obtain a combined weight; Based on the combined weight, the first correction value, the second correction value and the third correction value are weighted and assigned to obtain the wind speed correction result.

5. The method according to claim 1, characterized in that Obtaining the entire wind farm power curve includes: Obtaining a power curve of each wind turbine generator set in the wind farm; By performing interval unification, linear interpolation, redundant point elimination, and boundary value interpolation on the power curves of each wind turbine generator set, an optimized power curve unified on the X-axis and the Y-axis is obtained; At any target wind speed point, the power values ​​of each optimized power curve are found, and the total power of the wind farm at the target wind speed value is obtained by adding up the power values.

6. An integrated wind farm predicted wind speed correction system, characterized in that: The system comprises: an acquisition module, a training module and a correction module, wherein: The acquisition module is used to acquire the whole-field power curve of the wind farm, and based on the whole-field power curve in the historical period, obtain the whole-field wind speed at each time point in the historical period; And, the entire wind speed and the entire wind direction are time-series associated to obtain training set data including: Based on the whole-field power curve, obtaining the whole-field power with a time sequence relationship within any target historical period; Based on the whole-field power curve, the whole-field wind speed at each time point in the target historical period is calculated by reverse search and difference method; Acquire the entire wind speed and the entire wind direction at each time point in the target historical period by associating them through time series to obtain the data point at each time point, and collect the data points to obtain training set data including historical time points, wind speed parameters and wind direction parameters; The training module is used to construct a plurality of wind speed correction models based on the training set data, wherein the first wind speed correction models for normal weather and extreme weather are obtained by training the pre-constructed sLSTM time series two-state model, and the second wind speed correction model for normal weather is constructed based on the correction curve, wherein the sLSTM time series two-state model includes: a representation layer composed of at least two sLSTM layers, and a first MLP branch and a second MLP branch parallel to each other, pre-processing the training set data to obtain optimized training set data with continuity and standardization; through the representation layer, using the exponential gating mechanism and the memory mixing mechanism, capturing the load mode and long-term dependence characteristics in the time series of the optimized training set data to obtain the global features; according to the weather conditions corresponding to the global features, the first MLP branch or the second MLP branch is used to update the training based on the global features to obtain the wind speed correction results; Obtain the weather forecast wind speed at the time when each of the data points is located; combine each of the data points and its corresponding weather forecast wind speed into a two-dimensional plane point, and divide each of the two-dimensional plane points into different sectors of a circular area according to the wind direction data corresponding to the two-dimensional plane point, wherein the circular area includes multiple levels, and each level is divided into multiple sectors; construct the two-dimensional wind speed data corresponding to each data point with the weather forecast wind speed of each data point as the X-axis and the whole wind speed of each data point as the Y-axis; obtain the average wind speed data of each wind speed interval by averaging the two-dimensional wind speed data on the X-axis and the Y-axis within the wind speed interval of the X-axis, and obtain the correction curve corresponding to each sector by combining the average wind speed data of each wind speed interval; The correction module is used to predict the wind speed based on the weather forecast wind speed in any future target period through multiple wind speed correction models to obtain multiple wind speed correction values, and obtain wind speed correction results by performing weighted assignment operations on the multiple wind speed correction values.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 5 is implemented.

Images