Short-term power prediction method and system for regional wind power cluster

Abstract

The invention discloses a short-term power prediction method and system for a regional wind power cluster, and the method comprises the steps: S1, collecting and processing wind power plant data of each wind power generator; s2, calculating wind power plant data of non-observation points of the wind power plant by using a spatial interpolation method, obtaining a wind speed and wind direction distribution diagram of the whole wind power plant based on the data of known observation points, and extracting the wind power plant data; s3, performing similarity capture on the fans, and classifying the similar fans into one group; s4, integrating the wind power plant data, and enhancing the wind power plant data; a graph attention mechanism advanced feature learning method is utilized to deeply excavate the mutual relation among the fans in the wind power plant; s5, in combination with a FoecastNet deep learning model, constructing a deep learning model; and S6, based on wind power plant data, performing short-term regional wind power cluster power prediction by using the constructed deep learning model.

Description

Technical Field

[0001] The present invention relates to a short-term power prediction method and system for a wind power cluster, and in particular to a short-term power prediction method and system for a regional wind power cluster. Background Art

[0002] As the global demand for renewable energy continues to increase, wind energy, as a clean and renewable energy form, has received widespread attention and application. As an important form of wind energy development and utilization, regional wind power clusters are composed of multiple wind farms, which have scale effects and complementary advantages, and are of great significance for improving wind energy utilization and grid stability.

[0003] In the field of short-term power forecasting of wind power clusters, existing technologies mainly include forecasting methods based on physical models, forecasting methods based on statistics, and forecasting methods based on artificial intelligence.

[0004] Prediction method based on physical model: This method predicts wind power by constructing a physical model of the wind farm and combining meteorological data and wind turbine performance parameters.

[0005] Forecasting methods based on statistics: including time series analysis, regression analysis, etc. These methods use historical data to establish statistical models and predict future wind power through the trends and patterns of historical data.

[0006] Traditional wind power prediction methods are often based on data from a single wind farm, ignoring the mutual influence and relationship between the different locations of wind turbines within the wind power cluster, which may result in inaccurate prediction accuracy. Summary of the invention

[0007] The present invention overcomes the deficiencies of the prior art and provides a short-term power prediction method and system for a regional wind power cluster.

[0008] To achieve the above object, the technical solution adopted by the present invention is: a short-term power prediction method and system for a regional wind power cluster, comprising: S1, collecting wind farm data of each wind turbine and processing it;

[0009] S2. Calculate the wind farm data of unobserved points of the wind farm using the spatial interpolation method, and obtain the wind speed and wind direction distribution map of the entire wind farm based on the data of known observation points, and extract the wind farm data;

[0010] S3, capturing similarities of wind turbines and grouping similar wind turbines into one group;

[0011] S4. Integrate wind farm data and enhance wind farm data;

[0012] Using the Graph Attention Mechanism (GATs) advanced feature learning method, we can deeply explore the relationship between wind turbines in the wind farm.

[0013] S5. Build a deep learning model based on the data of the graph attention mechanism and combined with the ForecastNet deep learning model;

[0014] S6. Based on wind farm data, use the constructed deep learning model to perform short-term regional wind power cluster power forecasting.

[0015] In a preferred embodiment of the present invention, in step S1, the wind turbine data includes longitude and latitude data, wind speed and wind direction data.

[0016] In a preferred embodiment of the present invention, in step S2, the spatial interpolation method is an inverse distance weighted interpolation method, which is used to estimate the wind speed and wind direction of unobserved points in the wind farm.

[0017] In a preferred embodiment of the present invention, the inverse distance weighted interpolation method is:

[0018] Suppose there is a set of known data points p = {(x 1 ,y 1 ,z 1 ),(x 2 ,y 2 ,z 2 ),…(x n ,y n ,z n )}, where (x i ,y i ) is the two-dimensional space coordinate of the i-th data point, z i It is the attribute value corresponding to the point (such as wind speed, wind direction, etc.).

[0019] Assume the coordinates of the point to be interpolated are (x 0 ,y 0 ), we need to estimate the attribute value z of this point 0 .

[0020] For each known data point (x i ,y i ,z i ), calculate its value with the interpolation point (x 0 ,y 0 ) between i :

[0021]

[0022] According to the distance d i Calculate the weight w of each data point to treat the interpolation point iThe weight is the inverse of the distance raised to some power, and the power p (usually p=2) is a user-defined parameter used to adjust the sensitivity of the weight to changes in distance:

[0023]

[0024] According to the weight w i and the corresponding data point attribute value z i , calculate the attribute value z of the point to be interpolated 0

[0025]

[0026] z 0 is the wind speed and direction data of the unobserved point

[0027] in

[0028] In a preferred embodiment of the present invention, in step S3, similarity capture includes: calculating the longitude and latitude similarity using the Haversine formula; calculating the longitude and latitude similarity using the Haversine formula;

[0029] For each wind turbine, obtain its longitude and latitude coordinates (recorded as (lon1, lat1) and (lon2, lat2), and use the Haversine formula to calculate the spherical distance between the two points;

[0030]

[0031] Among them, d is the distance, r is the radius of the earth (about 6371 kilometers), Δlat=lat2-lat1, Δlon=lon2-lon1, and the smaller d is, the higher the similarity is.

[0032] Analyze the longitude and latitude and wind speed data of the wind turbines, identify wind turbine groups with similar characteristics, and judge the similarity based on longitude and latitude.

[0033] In a preferred embodiment of the present invention, in step S4, integrating wind farm data includes splicing original wind speed, wind direction, longitude and latitude data and wind speed, wind direction, longitude and latitude data after spatial interpolation processing into a multi-dimensional feature matrix.

[0034] In a preferred embodiment of the present invention, in step S4, the wind farm data is enhanced, specifically, including: using a graph attention mechanism (GATs) advanced feature learning method to deeply explore the relationship between the wind turbines in the wind farm. The graph attention mechanism (GATs) advanced feature learning method includes:

[0035] S41. Build a graph structure: treat each wind turbine as a node in the graph; define the edges between nodes based on the longitude and latitude of the wind turbine, and assign weights to the edges based on the distance similarity. Build a weighted undirected graph G = (V, E), where V is the node set (wind turbines) and E is the edge set (connections between wind turbines).

[0036] S42, Graph Attention Network (GAT) model construction;

[0037] S421, Initialize node features: Initialize the initial features h of each node v∈V v 0 It can be a vector representation of observation data such as wind speed, wind direction, longitude and latitude.

[0038] S422, graph attention layer

[0039] Multiple GAT layers are stacked to capture more complex node relationships; in each layer, for node v, the attention coefficient a between it and all neighboring nodes u∈N(v) is calculated vu ;

[0040] Use the attention coefficient to weight the features of neighboring nodes to update the feature representation of node v:

[0041]

[0042] where w l is the weight matrix of the lth layer, σ is the ReLU function;

[0043] S422. After the last layer of GAT, a fully connected layer (FC) can be added to further fuse and transform features, and finally output the power output prediction result of each node.

[0044] In a preferred embodiment of the present invention, building a deep learning model includes: combining the idea of ​​the ForecastNet deep learning model to build a deep learning model that can process multi-dimensional features (including wind farm data and time series features). The model should fully consider the wind farm data and time series features of the wind farm, gradually extract and fuse these features through a multi-layer network structure to capture the complex dynamic change process inside the wind farm; use the Adam optimization algorithm and the mean square error (MSE) algorithm to train the model to minimize the prediction error and improve the accuracy of the prediction.

[0045] In a preferred embodiment of the present invention, the mean square error (MSE) algorithm includes: assuming that there are N samples, where y i represents the true value of the i-th sample, Represents the predicted value of the i-th sample, then the calculation formula of MSE is:

[0046]

[0047] In a preferred embodiment of the present invention,

[0048] Data collection and processing module: used to collect wind farm data of each wind turbine and pre-process the collected data;

[0049] Wind farm data extraction module: Use spatial interpolation methods (such as inverse distance weighted interpolation method IDW) to calculate the wind speed and wind direction of unobserved points in the wind farm to obtain the wind speed and wind direction distribution map of the entire wind farm; extract wind farm data of the wind farm, including the spatial variation characteristics of wind speed and wind direction.

[0050] Similarity capture and grouping module: used to identify wind turbine groups with similar characteristics using the longitude and latitude data of wind turbines;

[0051] Feature integration and enhancement module: Integrate the original data and spatially interpolated data into a multi-dimensional feature matrix, and use advanced feature learning methods such as graph attention mechanisms (GATs) to deeply explore the relationships between wind turbines in the wind farm and enhance the wind farm data.

[0052] Model building module: Combining the ideas of deep learning models such as ForecastNet, a deep learning model that can process multi-dimensional features (including wind farm data and time series features) is constructed; Model training and optimization module: The Adam optimization algorithm and mean square error (MSE) are used to train the model to minimize the prediction error and improve the accuracy of the prediction.

[0053] Forecast output module: Use the constructed deep learning model to perform short-term regional wind power cluster power forecasts and output the forecast results, providing grid operators with reliable wind power forecast data to support their power dispatch and grid stability optimization.

[0054] The present invention solves the defects existing in the background technology and has the following beneficial effects:

[0055] (1) The present invention collects and processes wind farm data, calculates wind farm data of unobserved points in the wind farm using spatial interpolation, and obtains a wind speed and wind direction distribution map of the entire wind farm. It captures similarities between wind turbines and groups similar ones together, integrates and enhances wind farm data, and constructs a deep learning model based on the data of the graph attention mechanism method and the ForecastNet deep learning model. The model structure and algorithm parameters can be further optimized. Compared with the prior art, the present invention effectively solves the problem of significant differences in wind speed and wind direction at different locations in a wind farm.

[0056] (2) The present invention estimates the wind speed and wind direction of unobserved points in a wind farm by applying a spatial interpolation method. Based on the data of known observation points, a wind speed and wind direction distribution map of the entire wind farm is calculated by a mathematical algorithm. Compared with the prior art, the spatial variation characteristics of wind speed and wind direction in a wind farm can be accurately captured.

[0057] (3) The similarity capture of the present invention identifies wind turbine groups with similar characteristics by analyzing the geographical locations of wind turbines, and groups similar wind turbines together to provide more accurate and relevant input data for subsequent prediction models.

[0058] (4) The present invention uses the graph attention mechanism (GATs) advanced feature learning method to deeply explore the relationships between wind turbines in a wind farm; these relationships may be reflected in the mutual influence of wind speed and wind direction, energy transfer, etc.; compared with the existing technology, it helps to enhance the representation ability of features and improve the prediction accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art are briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work.

[0060] Figure 1 is a flow chart of a preferred embodiment of the present invention; DETAILED DESCRIPTION

[0061] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0062] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0063] like Figure 1 As shown, a short-term power prediction method and system for a regional wind power cluster includes:

[0064] S1. Collect wind farm data of each wind turbine and process it;

[0065] S2. Calculate the wind farm data of unobserved points of the wind farm using the spatial interpolation method, and obtain the wind speed and wind direction distribution map of the entire wind farm based on the data of known observation points, and extract the wind farm data;

[0066] S3, capturing similarities of wind turbines and grouping similar wind turbines into one group;

[0067] S4. Integrate wind farm data and enhance wind farm data;

[0068] The graph attention mechanism (GATs) advanced feature learning method is used to deeply explore the relationships between wind turbines in the wind farm.

[0069] S5. Build a deep learning model based on the data of the graph attention mechanism and combined with the ForecastNet deep learning model;

[0070] S6. Based on wind farm data, use the constructed deep learning model to perform short-term regional wind power cluster power forecasting.

[0071] Specifically, in step S1, the wind farm data includes latitude and longitude data, wind speed and wind direction data;

[0072] Wind farm data is collected through weather stations and wind turbine sensors. Weather stations can usually provide meteorological parameters such as wind speed, wind direction, temperature, humidity, and air pressure; wind turbine sensors provide real-time power output data, speed, pitch angle, and other wind turbine status information for each wind turbine.

[0073] Specifically, in step S1, the isolation forest algorithm is used to detect and remove outliers in the data to ensure data quality; for missing values ​​in the data, the Lagrange interpolation method is used to repair or fill in the missing values ​​to maintain the integrity and continuity of the data.

[0074] Specifically, in step S2, the spatial interpolation method is an inverse distance weighted interpolation method, which is used to estimate the wind speed and wind direction of unobserved points in the wind farm. The specific steps include:

[0075] Taking the horizontal plane as the plane, there is a set of known data points

[0076] p={(x 1 ,y 1 ,z 1 ),(x 2 ,y 2 ,z 2 ),…(x n ,y n ,z n )}, where (x i ,y i ) is the two-dimensional space coordinate of the i-th data point, zi It is the attribute value corresponding to the point (such as wind speed, wind direction, etc.).

[0077] Assume the coordinates of the point to be interpolated are (x 0 ,y 0 ), we need to estimate the attribute value z of this point 0 .

[0078] For each known data point (x i ,y i ,z i ), calculate its value with the interpolation point (x 0 ,y 0 ) between i :

[0079]

[0080] According to the distance d i Calculate the weight w of each data point to treat the interpolation point i The weight is the inverse of the distance raised to some power, and the power p (usually p=2) is a user-defined parameter used to adjust the sensitivity of the weight to changes in distance:

[0081]

[0082] According to the weight w i and the corresponding data point attribute value z i , calculate the attribute value z of the point to be interpolated 0

[0083]

[0084] Specifically, in step S2, a wind speed and wind direction distribution map of the entire wind farm is obtained by creating a two-dimensional grid in the wind farm area, and assigning the interpolated wind speed and wind direction values ​​to corresponding positions of the grid to form a gridded wind speed and wind direction data set;

[0085] Select the Matplotlib tool, use the gridded wind speed data, and draw a wind speed distribution map on the map, using color depth or different colors to indicate the size of the wind speed; the wind direction distribution map can be represented by drawing small arrows on the grid points. The direction of the arrows indicates the wind direction, and the length or thickness indicates the wind speed; superimpose the wind speed distribution map and the wind direction distribution map together, or draw them separately on different layers, so that the distribution of wind speed and wind direction can be viewed at the same time.

[0086] Specifically, in step S3, similarity capture includes: using the Haversine formula to calculate the longitude and latitude similarity;

[0087] For each wind turbine, obtain its longitude and latitude coordinates (recorded as (lon1, lat1) and (lon2, lat2), and use the Haversine formula to calculate the spherical distance between the two points;

[0088]

[0089] Among them, d is the distance, r is the radius of the earth (about 6371 kilometers), Δlat=lat2-lat1, Δlon=lon2-lon1, and the smaller d is, the higher the similarity is.

[0090] Analyze the longitude and latitude and wind speed data of wind turbines, identify wind turbine groups with similar characteristics, and judge similarity based on longitude and latitude;

[0091] Specifically, in step S4, integrating the wind farm data includes splicing the original wind speed, wind direction, longitude and latitude data and the wind speed, wind direction, longitude and latitude data after spatial interpolation processing into a multi-dimensional feature matrix.

[0092] Specifically, in step S4, the wind farm data is enhanced, specifically including: using the graph attention mechanism (GATs) advanced feature learning method to deeply explore the relationship between the wind turbines in the wind farm. The graph attention mechanism (GATs) advanced feature learning method includes:

[0093] S41. Build a graph structure: treat each wind turbine as a node in the graph; define the edges between nodes based on the longitude and latitude of the wind turbine, and assign weights to the edges based on the distance similarity. Build a weighted undirected graph G = (V, E), where V is the node set (wind turbines) and E is the edge set (connections between wind turbines).

[0094] S42, Graph Attention Network (GAT) model construction;

[0095] S421, Initialize node features: Initialize the initial features h of each node v∈V v 0 It can be a vector representation of observation data such as wind speed, wind direction, longitude and latitude.

[0096] S422, graph attention layer

[0097] Multiple GAT layers are stacked to capture more complex node relationships; in each layer, for node v, the attention coefficient a between it and all neighboring nodes u∈N(v) is calculated vu ;

[0098] Use the attention coefficient to weight the features of neighboring nodes to update the feature representation of node v:

[0099]

[0100] where w l is the weight matrix of the lth layer, σ is the ReLU function;

[0101] S422. After the last layer of GAT, a fully connected layer (FC) can be added to further fuse and transform features, and finally output the power output prediction result of each node.

[0102] Specifically, in step S5, building a deep learning model specifically includes: combining the idea of ​​the ForecastNet deep learning model, building a deep learning model that can process multi-dimensional features (including wind farm data and time series features). The model should fully consider the wind farm data and time series features of the wind farm, gradually extract and fuse these features through a multi-layer network structure to capture the complex dynamic change process inside the wind farm; using the Adam optimization algorithm and the mean square error (MSE) algorithm, the model is trained to minimize the prediction error and improve the accuracy of the prediction.

[0103] More specifically, MSE calculates the prediction error of each sample (i.e., the difference between the true value and the predicted value), then sums the squares of the errors of all samples and divides it by the number of samples N to get the average error. The smaller the MSE, the closer the predicted value is to the true value, and the better the model performance.

[0104] Assume there are N samples, where y ii represents the true value of the ii-th sample, Represents the predicted value of the ii-th sample, then the calculation formula of MSE is:

[0105]

[0106] A short-term power prediction system for a regional wind power cluster, comprising:

[0107] Data collection and processing module: used to collect wind farm data of each wind turbine and pre-process the collected data;

[0108] Wind farm data extraction module: Use spatial interpolation methods (such as inverse distance weighted interpolation method IDW) to calculate the wind speed and wind direction of unobserved points in the wind farm to obtain the wind speed and wind direction distribution map of the entire wind farm; extract wind farm data of the wind farm, including the spatial variation characteristics of wind speed and wind direction.

[0109] Similarity capture and grouping module: used to identify wind turbine groups with similar characteristics using the longitude and latitude data of wind turbines;

[0110] Feature integration and enhancement module: Integrate the original data and spatially interpolated data into a multi-dimensional feature matrix, and use advanced feature learning methods such as graph attention mechanisms (GATs) to deeply explore the relationships between wind turbines in the wind farm and enhance the wind farm data.

[0111] Model building module: Combining the ideas of deep learning models such as ForecastNet, a deep learning model that can process multi-dimensional features (including wind farm data and time series features) is constructed; Model training and optimization module: The Adam optimization algorithm and mean square error (MSE) are used to train the model to minimize the prediction error and improve the accuracy of the prediction.

[0112] Forecast output module: Use the constructed deep learning model to perform short-term regional wind power cluster power forecasts and output the forecast results, providing grid operators with reliable wind power forecast data to support their power dispatch and grid stability optimization.

[0113] The above is based on the ideal embodiment of the present invention. Through the above description, relevant personnel can make various changes and modifications without departing from the technical concept of the present invention. The technical scope of the present invention is not limited to the content in the specification, and the technical scope must be determined according to the scope of the claims.

Claims

1. A short-term power prediction method for a regional wind power cluster, characterized in that: The following steps are involved: S1. Collect wind farm data of each wind turbine and perform preprocessing; S2, using the spatial interpolation method to calculate the wind farm data of the unobserved points of the wind farm, and based on the data of the known observation points, obtain the wind speed and wind direction distribution map of the entire wind farm; S3, capturing similarities of wind turbines and grouping similar wind turbines into one group; S4. Integrate wind farm data and enhance wind farm data; Using graph attention mechanism and advanced feature learning methods, we can deeply explore the relationship between wind turbines in the wind farm. S5. Build a deep learning model based on the data of the graph attention mechanism and combined with the ForecastNet deep learning model; S6. Based on the wind farm data in S2, use the constructed deep learning model to perform short-term regional wind power cluster power forecasting.

2. The short-term power prediction method for a regional wind power cluster according to claim 1 is characterized by: In step S1 , the wind turbine data includes latitude and longitude data, wind speed and wind direction data.

3. The short-term power prediction method for a regional wind power cluster according to claim 1 is characterized in that: In step S2, the spatial interpolation method is an inverse distance weighted interpolation method, which is used to estimate the wind speed and wind direction of unobserved points in the wind farm.

4. The short-term power prediction method for a regional wind power cluster according to claim 3 is characterized by: The inverse distance weighted interpolation method is: Suppose there is a set of known data points p = {(x1,y1,z1),(x2,y2,z2),…(x n ,y n ,z n )}, where (x i ,y i ) is the two-dimensional space coordinate of the i-th data point, z i is the attribute value corresponding to the point; Assume the coordinates of the point to be interpolated are (x0, y0), and the attribute value z0 of the point needs to be estimated; For each known data point (x i ,y i ,z i ), calculate the Euclidean distance d between it and the point to be interpolated (x0, y0) i : According to the distance d i Calculate the weight w of each data point to treat the interpolation point i , the weight is a power of the inverse of the distance, and the power p is a user-defined parameter used to adjust the sensitivity of the weight to changes in distance: According to the weight w i and the corresponding data point attribute value z i , calculate the attribute value z0 of the point to be interpolated Among them, z0 is the wind speed and direction data of the unobserved point.

5. The short-term power prediction method for a regional wind power cluster according to claim 1, characterized in that: In step S3, similarity capture includes: calculating the longitude and latitude similarity using the Haversine formula; calculating the longitude and latitude similarity using the Haversine formula; For the longitude and latitude of each wind turbine, obtain its longitude and latitude coordinates as (lon1, lat1) and (lon2, lat2), and use the Haversine formula to calculate the spherical distance between the two points; Where d is the distance, r is the radius of the earth, Δlat = lat2-lat1, Δlon = lon2-lon1, and the smaller d is, the higher the similarity is; Analyze the longitude and latitude and wind speed data of the wind turbines, identify wind turbine groups with similar characteristics, and judge the similarity based on longitude and latitude.

6. A method for short-term power prediction of a regional wind power cluster according to claim 1, characterized in that: In step S4, integrating the wind farm data includes splicing the original wind speed, wind direction, longitude and latitude data and the wind speed, wind direction, longitude and latitude data after spatial interpolation processing into a multi-dimensional feature matrix.

7. A method for short-term power prediction of a regional wind power cluster according to claim 1, characterized in that: In step S4, the wind farm data is enhanced, including: using a graph attention mechanism advanced feature learning method to deeply explore the relationship between the wind turbines in the wind farm; wherein the graph attention mechanism advanced feature learning method includes: S41. Build a graph structure: treat each wind turbine as a node in the graph; define the edges between nodes according to the longitude and latitude of the wind turbine, and assign weights to the edges according to the distance similarity; build a weighted undirected graph G = (V, E), where V is the set of wind turbines and E is the set of connections between wind turbines; S42, graph attention network model construction; S421, Initialize node features: Initialize the initial features h of each node v∈V v 0 It can be a vector representation of observation data such as wind speed, wind direction, longitude and latitude; S422, graph attention layer Multiple GAT layers are stacked to capture more complex node relationships; in each layer, for node v, the attention coefficient a between it and all neighboring nodes u∈N(v) is calculated vu ; Use the attention coefficient to weight the features of neighboring nodes to update the feature representation of node v: where w l is the weight matrix of the lth layer, σ is the ReLU function; S43. After the last layer of GAT, a fully connected layer can be added to further fuse and transform features, and finally output the power output prediction results of each node.

8. A method for short-term power prediction of a regional wind power cluster according to claim 7, characterized in that: Building a deep learning model includes: combining the ideas of the ForecastNet deep learning model to build a deep learning model that can process multi-dimensional features; The model should fully consider the wind farm data and time series characteristics of the wind farm, gradually extract and fuse these features through a multi-layer network structure to capture the complex dynamic change process within the wind farm; the Adam optimization algorithm and the mean square error MSE algorithm are used to train the model to minimize the prediction error and improve the accuracy of the prediction.

9. A method for short-term power prediction of a regional wind power cluster according to claim 8, characterized in that: The mean square error MSE algorithm includes: assuming that there are N samples, where b ii represents the true value of the i-th sample, Represents the predicted value of the i-th sample, then the calculation formula of MSE is:

10. A short-term power prediction system for a regional wind power cluster, based on a short-term power prediction method for a regional wind power cluster according to any one of claims 1 to 9, characterized in that: include: Data collection and processing module: used to collect wind farm data of each wind turbine; And pre-process the collected data; Wind farm data extraction module: Use spatial interpolation to calculate the wind speed and wind direction of unobserved points in the wind farm to obtain the wind speed and wind direction distribution map of the entire wind farm; extract wind farm data of the wind farm, including the spatial variation characteristics of wind speed and wind direction; Similarity capture and grouping module: used to identify wind turbine groups with similar characteristics using the longitude and latitude data of wind turbines; Feature integration and enhancement module: Integrate the original data and spatially interpolated data into a multi-dimensional feature matrix, and use advanced feature learning methods such as graph attention mechanism to deeply explore the relationship between wind turbines in the wind farm and enhance the wind farm data; Model building module: Combine the ideas of deep learning models such as ForecastNet to build a deep learning model that can handle multi-dimensional features; Model training and optimization module: Use the Adam optimization algorithm and mean square error MSE to train the model to minimize the prediction error and improve the accuracy of the prediction; Prediction output module: Use the constructed deep learning model to perform short-term regional wind power cluster power forecasts and output the forecast results.

Images