A wind power cluster short-term power prediction method, system, electronic device and medium

Abstract

The application discloses a wind power cluster short-term power prediction method and system, electronic equipment and medium, and relates to the technical field of wind power prediction. The method comprises the following steps: an improved wind angle field feature conversion method is constructed based on an improved Gram angle field; for a target wind power cluster, an improved wind angle field feature conversion method is used to generate a WAF matrix; the WAF matrix is converted into a scale space by using a SIFT identification method, and continuous wind process mutation points and time sequence correlation points are identified to determine the space-time correlation relationship between wind power clusters; a graph structure of the target wind power cluster is constructed based on a dynamic graph attention network and the space-time correlation relationship, and dynamic prediction is performed by using the graph structure and daily NWP meteorological information to obtain a power prediction result of the target wind power cluster. The application can improve the short-term power prediction accuracy.

Description

Technical Field

[0001] This invention relates to the field of wind power prediction technology, and in particular to a method, system, electronic device and medium for short-term power prediction of wind power clusters. Background Technology

[0002] Wind power generation inherently relies on natural wind resources. Meteorological factors such as wind speed and direction exhibit high variability and uncertainty in both time and space, resulting in strong random fluctuations and intermittent characteristics in wind power output. The grid connection of large-scale wind power has brought unprecedented challenges to the safe and stable operation of the power system, the economic efficiency of dispatch planning, and the orderly competition in the electricity market.

[0003] Against this backdrop, achieving accurate and rapid short-term forecasting of wind power is not only a key technical support for ensuring grid acceptance capacity and improving system operational flexibility and reliability, but also a core prerequisite for optimizing dispatch decisions, reducing reserve costs, promoting wind power consumption, and participating in market transactions, especially for large-scale wind power clusters.

[0004] Currently, research on wind power forecasting has developed various technical approaches, which can be mainly summarized as follows: Statistical methods, based on historical data, utilize time series analysis models for prediction. Their advantage lies in the relatively simple model structure and convenient calculation. However, wind power series typically exhibit highly nonlinear and non-stationary characteristics, making it difficult for traditional statistical models to fully capture these complex dynamics.

[0005] Physical methods primarily rely on numerical weather prediction (NWP) systems. By modeling and simulating atmospheric physical processes, they output meteorological parameters such as wind speed and direction for future moments, which are then combined with the physical characteristics of wind turbines to convert into predicted power values. Physical methods can reflect the impact of meteorological conditions on wind power from a mechanistic perspective and have a certain theoretical foundation. However, their main bottlenecks are: the NWP models themselves are computationally intensive, requiring high computational resources; and the spatiotemporal resolution of the forecast results is relatively coarse, making it difficult to fully meet the needs of refined wind farm operation. Machine learning and deep learning methods, represented by support vector machines (SVM), artificial neural networks (ANN), long short-term memory networks (LSTM), and graph neural networks (GNN), have shown great potential in wind power prediction due to their powerful nonlinear mapping capabilities and the ability to autonomously learn complex patterns from massive amounts of data, becoming the mainstream research direction. These methods can effectively uncover hidden deep features and correlations in historical data.

[0006] However, existing machine learning and deep learning methods still face several key challenges when applied to power prediction of large-scale wind farm clusters. First, most methods rely on pre-defined, location-based static topologies or simplified similarity metrics (such as Euclidean distance and Pearson correlation coefficient) when dealing with the spatial correlation between wind farms. These approaches ignore the dynamic changes in the spatiotemporal relationships between wind farms as upstream meteorological systems evolve (e.g., wind belt movement, frontal passage). Static or quasi-static topological descriptions struggle to accurately capture the complex and time-varying interaction mechanisms within wind farm clusters, thus limiting the model's ability to grasp the overall power output characteristics of the cluster. Second, for large-scale wind farm clusters, the sheer number of farms and their complex interaction networks make it difficult to fully utilize the synergistic effects and information complementarity inherent within the cluster. This results in limited improvement in prediction accuracy and model robustness. Furthermore, traditional time-series feature extraction methods may not fully draw on advanced concepts from other fields (such as computer vision and image processing) to reveal the inherent dynamic patterns of wind power output sequences and their potential correlations with other station sequences at a deeper and more multi-dimensional level.

[0007] Therefore, there is an urgent need to develop a new method for short-term power prediction. Summary of the Invention

[0008] The purpose of this invention is to provide a method, system, electronic device and medium for short-term power prediction of wind power clusters, which aims to solve or improve at least one of the above-mentioned technical problems.

[0009] To achieve the above objectives, the present invention provides the following solution:

[0010] A method for short-term power prediction of wind power clusters includes:

[0011] An improved wind angle field feature conversion method is constructed based on the improved Gram angle field; the improved wind angle field feature conversion method generates polar coordinates by processing the time series value as the radius and using the sum of the normalized inverse cosine value of the wind speed series and the wind direction value as the angle.

[0012] For the target wind farm cluster, an improved wind angle field feature transformation method is used to generate a WAF matrix;

[0013] The WAF matrix is ​​converted into a scale space using the SIFT identification method, and continuous abrupt changes in wind processes and time-series correlation points are identified to determine the spatiotemporal correlation between wind farms.

[0014] A graph structure for the target wind farm cluster is constructed based on a dynamic graph attention network and the spatiotemporal correlation. The graph structure and daily NWP meteorological information are used to make dynamic predictions to obtain the power prediction results of the target wind farm cluster.

[0015] Optionally, the step of generating a WAF matrix for the target wind farm cluster using an improved wind angle field feature transformation method specifically includes:

[0016] First, the wind speed sequence is normalized, and then the inverse cosine value of the wind speed sequence is calculated:

[0017]

[0018] Then, add the inverse cosine of the wind speed sequence to the wind direction value to obtain the angle:

[0019] cos((φ i,s +φ i,d )+(φ j,s +φ j,d ))=cos(φ i +φ j )

[0020] Next, the coordinates within the scheduling cycle are multiplied point by point to obtain the WAF matrix that can characterize the wind farm scheduling cycle:

[0021]

[0022] Where, x i,sLet x represent the normalized wind speed sequence for the i-th wind farm on the forecast day. i,d φ represents the normalized wind direction sequence for the i-th wind farm on the forecast day; i,s φ represents the inverse cosine value of the normalized wind speed sequence for the i-th wind farm. i,d This represents converting the wind direction sequence of the i-th wind farm into radians; φ j,s φ represents the inverse cosine value of the normalized wind speed sequence for the j-th wind farm. j,d The wind direction sequence of the j-th wind farm is converted into radians; φ i φ represents the angle of the i-th wind farm. j r represents the angle of the j-th wind farm; t Represents the radius in the polar coordinate system obtained through time series transformation; φ1,...,φ n The t represents the feature value of different time series; t represents the time series index; N represents the total length of the constructed window; and n represents the position of the nth time point.

[0023] Optionally, the step of using the SIFT identification method to convert the WAF matrix into a scale space and identifying continuous abrupt changes in wind processes and time-series correlation points to determine the spatiotemporal correlation between wind farms specifically includes:

[0024] First, the WAF matrix is ​​subjected to Gaussian blurring, and key points of the WAF matrix are detected by Gaussian difference calculation between different scales;

[0025] Then, feature description vectors of abrupt change points are generated based on the gradient direction and amplitude distribution around the key points, and the Euclidean distance between each feature description vector of abrupt change points is calculated. The spatiotemporal correlation between wind farms is determined based on David Lowe's distance ratio test method.

[0026] Optionally, the method based on David Lowe's distance ratio test to determine the spatiotemporal correlation between wind farms specifically includes:

[0027] Based on David Lowe's distance ratio test method, for each mutation point feature description vector, the two nearest matching points to be matched in the WAF matrix are found, and the nearest neighbor distance and the next nearest neighbor distance are calculated. If the ratio of the nearest neighbor distance to the next nearest neighbor distance is less than a set threshold, it is considered that there is a spatiotemporal correlation between the two wind farms.

[0028] Optionally, the construction of the graph structure of the target wind farm cluster based on the dynamic graph attention network and the spatiotemporal correlation specifically includes:

[0029] An initial graph structure and adjacency matrix are constructed based on the spatiotemporal correlation. The initial graph structure is represented as G = (V, E), where V and E are the node set and edge set, respectively. The adjacency matrix is ​​used to describe the relationship between any two connected nodes.

[0030] Based on the initial graph structure and the adjacency matrix, the relevant interaction coefficients between each node are dynamically calculated using a dynamic graph attention network to obtain the graph structure of the target wind farm cluster.

[0031] The present invention also provides a short-term power prediction system for wind power clusters, comprising:

[0032] An algorithm construction unit is used to construct an improved wind angle field feature conversion method based on the improved Gram angle field; the improved wind angle field feature conversion method generates polar coordinates by processing time series values ​​as radii and using the sum of the normalized inverse cosine value of the wind speed series and the wind direction value as the angle.

[0033] The matrix generation unit is used to generate a WAF matrix for a target wind farm cluster using an improved wind angle field feature transformation method.

[0034] The spatiotemporal correlation determination unit is used to convert the WAF matrix into a scale space using the SIFT recognition method, and to identify continuous wind process abrupt change points and time series correlation points to determine the spatiotemporal correlation between wind farms.

[0035] The power prediction unit is used to construct the graph structure of the target wind farm cluster based on the dynamic graph attention network and the spatiotemporal correlation, and to perform dynamic prediction using the graph structure and daily NWP meteorological information to obtain the power prediction result of the target wind farm cluster.

[0036] The present invention also provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to perform the short-term power prediction method for wind power clusters as described above.

[0037] The present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the short-term power prediction method for wind power clusters as described above.

[0038] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0039] This invention discloses a method, system, electronic device, and medium for short-term power prediction of wind farm clusters. The method includes: constructing an improved wind angle field feature transformation method based on an improved Gram angle field; generating a WAF matrix for a target wind farm cluster using the improved wind angle field feature transformation method; converting the WAF matrix into a scale space using the SIFT recognition method, and identifying continuous wind process abrupt change points and time series correlation points to determine the spatiotemporal correlation between wind farms; constructing a graph structure of the target wind farm cluster based on a dynamic graph attention network and the spatiotemporal correlation; and using the graph structure and daily NWP meteorological information for dynamic prediction to obtain the power prediction result of the target wind farm cluster. This invention can improve the accuracy of short-term power prediction. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is a flowchart illustrating the short-term power prediction method for wind power clusters of the present invention.

[0042] Figure 2 This is a schematic diagram illustrating the association discrimination process between WAF matrices in this embodiment;

[0043] Figure 3 This is a WAF diagram of a predicted day for each wind farm within the wind power cluster in this embodiment;

[0044] Figure 4 This is a time series diagram of the predicted and actual power of wind power clusters using various methods in this embodiment. Detailed Implementation

[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] The purpose of this invention is to provide a method, system, electronic device and medium for short-term power prediction of wind power clusters, which aims to solve or improve at least one of the above-mentioned technical problems.

[0047] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0048] like Figure 1 As shown, the present invention provides a short-term power prediction method for wind power clusters, comprising:

[0049] Step 100: Construct an improved wind angle field feature conversion method based on the improved Gram angle field; the improved wind angle field feature conversion method generates polar coordinates by processing the time series values ​​as the radius and using the sum of the normalized inverse cosine value of the wind speed series and the wind direction value as the angle.

[0050] Step 200: For the target wind farm cluster, generate a WAF matrix using an improved wind angle field feature transformation method.

[0051] Step 300: Use the SIFT identification method to convert the WAF matrix into a scale space, identify continuous wind process abrupt change points and time series correlation points, and determine the spatiotemporal correlation between wind farms.

[0052] Step 400: Construct the graph structure of the target wind farm cluster based on the dynamic graph attention network and the spatiotemporal correlation, and use the graph structure and daily NWP meteorological information to make dynamic predictions to obtain the power prediction results of the target wind farm cluster.

[0053] As a specific implementation method, this embodiment constructs a multi-feature transformation method to generate the wind angle field (WAF) feature matrix for each wind farm. Combining the wind speed, wind direction, and temporal gradient of the wind farm on the day of measurement, the SIFT recognition method from the field of image recognition is used to identify whether there is a spatiotemporal correlation between wind farms, and a graph structure is constructed accordingly. Specifically, the following steps are included:

[0054] (1) Construction of the WAF matrix of a wind farm:

[0055] To more accurately capture the spatiotemporal correlation between wind farms, this invention proposes a wind angle field feature transformation method based on an improved Gram angle field. This method generates polar coordinates by treating time series values ​​as radii and using the sum of the normalized inverse cosine of the wind speed sequence and the wind direction value as the angle. Specifically, the wind speed sequence is first normalized (Equation (1) is the normalization formula), then its inverse cosine value is calculated and added to the wind direction value to obtain the angle. Next, the coordinates within the scheduling cycle are internally productd point by point to obtain the wind angle field that characterizes the wind farm scheduling cycle. This process not only considers the amplitude information of the time series but also introduces the gradient relationship of the time series, thus more comprehensively describing the spatiotemporal characteristics of the wind farm. The specific process is as shown in Equations (2)-(4):

[0056]

[0057]

[0058] cos((φ i,s +φ i,d )+(φ j,s +φ j,d ))=cos(φ i +φ j (3)

[0059]

[0060] The input is the normalized wind speed sequence x for the i-th wind farm on the day to be predicted. i,s and wind direction sequence x i,d ;φ i,d It is the normalized wind speed sequence x of the i-th wind farm. i,s The inverse cosine value; φ i,d This converts the wind direction sequence of the i-th wind farm into radians. φ i and φ j These are the eigenvalues ​​of different time sequences; r n It is the radius in the polar coordinate system obtained through time series transformation.

[0061] (2) Determine based on the spatiotemporal correlation between wind farms in WAF

[0062] To identify abrupt state changes in the image matrix, this scheme introduces the SIFT method to convert the WAF matrix into a scale space, thereby identifying abrupt changes in continuous wind processes and time-series correlation points. The specific process is as follows: Figure 2 As shown. First, Gaussian blurring is applied to the WAF to obtain the key points of the WAF matrix; the representation of the Gaussian blur matrix is ​​as follows:

[0063] L(x,y,σ)=G(x,y,σ)*I(x,y)(5)

[0064]

[0065] Here, L(x,y,σ) is the matrix converted to scale space. G(x,y,σ) is a Gaussian kernel with scale σ, I(x,y) is an identity matrix, * denotes convolution operation, and o and u are the x and y coordinates of the center of the Gaussian kernel, respectively. Then, keypoints are detected by calculating the difference of Gaussians between different scales.

[0066] D(x,y,σ)=[G(x,y,kσ)-G(x,y,σ)]*I(x,y)(7)

[0067] Where k is the scaling factor. This process can effectively detect stable feature points in the WAF matrix. After determining the key points, feature descriptors are generated based on the gradient direction and magnitude distribution around the key points. The formulas for calculating the gradient direction and magnitude are Equations (8)-(9).

[0068]

[0069]

[0070] Among them, f 1,i and f 2,i These are the i-th elements of the two descriptors. Generally, an empirical value of 0.75 is a commonly used threshold in David Lowe's distance ratio test. If the ratio of the nearest neighbor distance to the next nearest neighbor distance is less than 0.75, then WF is considered acceptable. i and WF j There is a spatiotemporal correlation between them.

[0071] (3) Dynamic Graph Attention Network

[0072] To refine the spatiotemporal relationships of wind farms, this invention introduces a dynamic graph attention network (DGAT) to calculate the interaction coefficients between related wind farms. The DGAT fully utilizes prior spatiotemporal dependency information to construct the spatiotemporal relationships between nodes, thus forming the spatiotemporal topology of the wind farm cluster. It also calculates the influence weights between wind farms to construct the nonlinear relationship between input features and output. Based on the spatiotemporal correlations between wind farms identified above, a graph structure G = (V, E) of the wind farm cluster is constructed, where V and E are the node set and edge set, respectively. The adjacency matrix is ​​constructed in the form Adj∈R. N×N , used to describe the relationship between any two connected nodes. If wind farm i and wind farm j have a spatiotemporal correlation, then Adjw i,j =1, otherwise Adjw i,j =0 indicates that there is no connection between them. The graph structure is dynamically updated based on daily NWP weather information.

[0073] Adjw i,j =STC i,j i=1,...,n;j=1,...,n-1(12)

[0074] Among them, STC i,j This represents the spatiotemporal relationship matrix between wind farms.

[0075] To fully extract the spatiotemporal interaction relationships between nodes, this embodiment uses a dynamic graph attention network to dynamically calculate the correlation interaction coefficients between nodes. The specific process is as follows:

[0076]

[0077] e(x Ci ,x Cj ) = a T [Wx Ci P Wx Cj ]j∈N Ci (14)

[0078]

[0079]

[0080] Where e(x) Ci ,x Cj ) represents the nonlinear transformation attention coefficient of node j (a neighboring node of node i) with respect to node i in a wind power cluster; LeakyReLU(·) is the nonlinear transformation function; a T (·) is the attention coefficient; P is the feature concatenation operation; x Ci and x Cj These are the input feature vectors of nodes i and j, respectively. W is the shared parameter matrix of the linear transformation; N Ci γ is the total number of neighboring nodes of node i; γ is the positive slope of the negative region of the function; then, the attention weight α is obtained by normalizing the obtained attention coefficients. Cij Output vector The power of a single field from multiple heads is obtained by weighted summation of the power obtained by nonlinear fitting of the eigenvectors. `softmax(·)` is the normalization function; `softmax`... j (·) is a function that normalizes the attention coefficients of neighboring nodes j; σ(·) is the sigmoid activation function; the number of heads in the multi-head attention mechanism is K. Cij e(x) Ci ,x Cj ) represents the attention coefficient between wind farms i and j in the cluster, N i This represents the total number of wind farms in the i-th wind power cluster.

[0081] Therefore, this invention can not only quickly and effectively match wind farms with spatiotemporal relationships within a wind power cluster, but also guide the information fusion between multi-channel wind farms based on a priori constructed graph structure. As a case study, Table 1 presents the results of a wind power cluster in a province in Northeast China. The prediction model provides predictions for 96 points (at 15-minute resolution) from 00:00 to 23:45 the following day at 8:00 each day. Four differentiated models are presented, namely models 1-4. The "proposed optimized model" represents the short-term power prediction results of the wind power cluster based on graph feature relationship matching. The results in the table show that the proposed optimized model can significantly improve prediction accuracy. To illustrate the effect more intuitively, Figure 3 This demonstrates the WAF (Web Application Firewall) for a specified forecast date across the test sets of various wind farms within the wind power cluster. Figure 4 Time series plots corresponding to wind power clusters are presented. The optimization model can utilize the complementary information between wind farms to improve the prediction results.

[0082] Table 1 Forecasting Indicators for Wind Power Clusters

[0083]

[0084] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0085] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for short-term power prediction of wind power clusters, characterized in that, include: An improved wind angle field feature conversion method is constructed based on the improved Gram angle field. The improved wind angle field feature conversion method generates polar coordinates by processing time series values ​​as radii and using the sum of the normalized inverse cosine value of the wind speed series and the wind direction value as the angle. For the target wind farm cluster, an improved wind angle field feature transformation method is used to generate a WAF matrix; The WAF matrix is ​​converted into a scale space using the SIFT identification method, and continuous abrupt changes in wind processes and time-series correlation points are identified to determine the spatiotemporal correlation between wind farms. The graph structure of the target wind farm cluster is constructed based on the dynamic graph attention network and the spatiotemporal correlation, and dynamic prediction is performed using the graph structure and daily NWP meteorological information to obtain the power prediction result of the target wind farm cluster. The method for generating a WAF matrix for the target wind farm cluster using an improved wind angle field feature transformation method specifically includes: First, the wind speed sequence is normalized, and then the inverse cosine value of the wind speed sequence is calculated: Then, add the inverse cosine of the wind speed sequence to the wind direction value to obtain the angle: Next, the coordinates within the scheduling cycle are multiplied point by point to obtain the WAF matrix that can characterize the wind farm scheduling cycle: in, Indicates the first Normalized wind speed sequence for each wind farm on the forecast date Indicates the first Normalized wind direction sequence for each wind farm on the forecast date; Indicates the first The inverse cosine value of the normalized wind speed sequence of a wind farm; Indicates the first The wind direction sequence of a wind farm is converted into radians; Indicates the first The inverse cosine value of the normalized wind speed sequence of a wind farm; Indicates the first The wind direction sequence of a wind farm is converted into radians; Indicates the first From the perspective of a wind farm; Indicates the first From the perspective of a wind farm; Represents the radius in the polar coordinate system obtained through time series transformation; Characteristic values ​​representing different time series; t Indicates the time sequence number; N This indicates the total length of the constructed window. n Indicates the first n Location at each point in time.

2. The short-term power prediction method for wind power clusters according to claim 1, characterized in that, The process of converting the WAF matrix into a scale space using the SIFT recognition method, identifying continuous abrupt changes in wind processes and time-series correlation points, and determining the spatiotemporal correlation between wind farms specifically includes: First, the WAF matrix is ​​subjected to Gaussian blurring, and key points of the WAF matrix are detected by Gaussian difference calculation between different scales; Then, feature description vectors of abrupt change points are generated based on the gradient direction and amplitude distribution around the key points, and the Euclidean distance between each feature description vector of abrupt change points is calculated. The spatiotemporal correlation between wind farms is determined based on David Lowe's distance ratio test method.

3. The short-term power prediction method for wind power clusters according to claim 2, characterized in that, The David Lowe distance ratio test method for determining the spatiotemporal correlation between wind farms specifically includes: Based on David Lowe's distance ratio test method, for each mutation point feature description vector, the two nearest matching points to be matched in the WAF matrix are found, and the nearest neighbor distance and the next nearest neighbor distance are calculated. If the ratio of the nearest neighbor distance to the next nearest neighbor distance is less than a set threshold, it is considered that there is a spatiotemporal correlation between the two wind farms.

4. The short-term power prediction method for wind power clusters according to claim 1, characterized in that, The construction of the graph structure of the target wind farm cluster based on the dynamic graph attention network and the spatiotemporal correlation specifically includes: An initial graph structure and adjacency matrix are constructed based on the spatiotemporal relationships; the initial graph structure is represented as follows: ,in, and These are the node set and the edge set, respectively; the adjacency matrix is ​​used to describe the relationship between any two connected nodes. Based on the initial graph structure and the adjacency matrix, the relevant interaction coefficients between each node are dynamically calculated using a dynamic graph attention network to obtain the graph structure of the target wind farm cluster.

5. A short-term power prediction system for wind power clusters, using the method as described in any one of claims 1-4, characterized in that, include: Algorithm building unit, used to construct an improved wind angle field feature conversion method based on the improved Gram angle field; The improved wind angle field feature conversion method generates polar coordinates by processing time series values ​​as radii and using the sum of the normalized inverse cosine value of the wind speed series and the wind direction value as the angle. The matrix generation unit is used to generate a WAF matrix for a target wind farm cluster using an improved wind angle field feature transformation method. The spatiotemporal correlation determination unit is used to convert the WAF matrix into a scale space using the SIFT recognition method, and to identify continuous wind process abrupt change points and time series correlation points to determine the spatiotemporal correlation between wind farms. The power prediction unit is used to construct the graph structure of the target wind farm cluster based on the dynamic graph attention network and the spatiotemporal correlation, and to perform dynamic prediction using the graph structure and daily NWP meteorological information to obtain the power prediction result of the target wind farm cluster.

6. An electronic device, characterized in that, The device includes a memory and a processor, the memory being used to store a computer program, and the processor running the computer program to enable the electronic device to perform the short-term power prediction method for wind power clusters according to any one of claims 1-4.

7. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the short-term power prediction method for wind power clusters as described in any one of claims 1-4.

Images