Wind power cluster short-term power prediction method and device, equipment and storage medium

Abstract

The invention provides a wind power cluster short-term power prediction method and device, equipment and a storage medium. Relates to the technical field of wind power cluster power short-term prediction. The method comprises the following steps: acquiring wind speed data of each wind power plant in a cluster to form a first data set, and dividing the wind power plant into a plurality of sub-clusters through clustering; determining a short-term correlation relation based on the distance between the wind power plants; the same sub-cluster wind power plants serve as nodes of the same type, and a same composition is constructed to represent an isomorphic space-time relation; different sub-cluster wind power plants serve as different nodes, and a heterogeneous graph is constructed to represent a heterogeneous space-time relation; dynamic graph network design: constructing a dynamic graph attention network which comprises an isomorphic module and a heterogeneous module, and extracting spatial-temporal feature vectors of each wind power plant by the isomorphic module based on the isomorphic composition; and the heterogeneous module fuses the heterogeneous graph and the time-space feature vector and outputs a power prediction result. According to the method, through clustering division and isomorphic / heterogeneous double-graph joint modeling, the complex space-time correlation of the wind power plant group is accurately captured, and the prediction precision is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wind power cluster power short-term prediction, and particularly relates to a wind power cluster short-term power prediction method, device, equipment and storage medium. BACKGROUND

[0002] With the rapid development of wind power industry, the construction mode of wind farms is gradually evolving from the early scattered, small-scale, low-penetration rate to centralized, large-scale, high-penetration rate cluster layout. The output characteristics of such large-scale wind power cluster are not only significantly affected by macro weather systems, but also face the challenge of inherent deviation of numerical weather prediction (NWP) in time and space accuracy and prediction range. Unlike isolated wind farms, wind farms within the cluster exhibit more complex spatio-temporal correlation. The traditional prediction model based on single wind farm information or simple aggregation is difficult to effectively capture the dynamic interaction within the cluster and the complex mechanism of global meteorological conditions on the overall power output, and has been unable to meet the increasingly refined operation requirements. How to fully utilize multi-dimensional NWP data, deeply mine and accurately characterize the complex spatio-temporal dependence relationship within the large-scale wind power cluster has become one of the core technical bottlenecks to be solved in the current wind power prediction field.

[0003] For the power prediction problem of wind power cluster, the existing technical paths mainly include the overall method, the cumulative method and the statistical upscaling method, etc. The overall method directly predicts the total power of the cluster by averaging the characteristics of all wind farms in the cluster, but this coarsens the temporal and spatial differences and dynamic correlations between wind farms. The cumulative method ignores the dynamic influence of the interaction between wind farms by independently modeling each wind farm and then superimposing the power. The statistical upscaling method relies on the data of a small number of benchmark wind farms to infer the power output of the entire region, but this inference method based on sparse single-point meteorological information is difficult to accurately reflect the temporal and spatial evolution of the entire cluster region. To further improve the prediction accuracy, researchers have introduced multi-input multi-output (MIMO) models such as convolutional neural network (CNN) and multi-layer perceptron (MLP). Although CNN can extract local spatio-temporal features, its fixed receptive field limits its ability to capture global information and long-distance dependencies. MLP, when dealing with high-dimensional spatio-temporal data, tends to lose key spatio-temporal structure information due to feature vectorization. Although the attention mechanism (AM) improves the model's ability to model long sequences and multi-target correlations by assigning different weights to different spatio-temporal features, its applicability and effectiveness still need to be improved when dealing with complex interactions between nodes with different characteristics (i.e., heterogeneous nodes) in large-scale wind power clusters. In recent years, graph neural networks (GNN) have been introduced into the field of wind power prediction due to their unique advantages in handling non-Euclidean space data and modeling complex topological relationships, providing a new perspective for solving the above challenges. However, current GNN-based research mostly focuses on homogeneous graphs. In such graphs, nodes (wind farms) are assumed to have a single or similar attribute, and edge relationships are relatively fixed. This setting can better represent the spatio-temporal correlations within small-scale, similar characteristic sub-clusters of wind farms, but it is difficult to effectively address the heterogeneity problem that exists in large-scale wind power clusters, where different wind farms may exhibit significantly different output characteristics and response patterns due to factors such as geographical location, micro-meteorological conditions, equipment models, and operation and maintenance strategies.

[0004] Existing methods have not yet formed a systematic theoretical framework and effective technical solution for how to construct and utilize heterogeneous graphs to capture the complex and dynamic heterogeneity between different types of wind farms and between different sub-clusters, resulting in the valuable heterogeneity information not being fully exploited and utilized.

[0005] In summary, existing wind power prediction technologies still have obvious deficiencies in dealing with large-scale, high-density wind power clusters, especially in accurately depicting the complex dynamic spatio-temporal correlations between wind farms within the cluster, where both homogeneity and heterogeneity coexist. Therefore, there is an urgent need to develop an advanced prediction method that can effectively integrate homogeneous and heterogeneous information within the cluster and dynamically capture and represent their complex interactions, in order to significantly improve the accuracy and reliability of short-term power prediction for large-scale wind power clusters. SUMMARY

[0006] The present application provides a method, device, equipment and storage medium for short-term power prediction of a wind farm cluster to overcome the problems in the prior art of insufficient characterization of the complex spatiotemporal correlations within the wind farm cluster and inability to dynamically adapt to meteorological changes.

[0007] In a first aspect, the present application provides a method for short-term power prediction of a wind power cluster, comprising:

[0008] Obtain wind speed data of each wind farm in the wind power cluster to form a first data set, and cluster the first data set to obtain multiple sub-wind power clusters; wherein the first data set is represented by X = {x1, x2, ..., x n}, x1, x2 and x n Respectively representing wind speed data of the first, second, and nth wind farms, the sub-wind farm cluster including at least one wind farm;

[0009] Determine the short-term correlation between wind farms based on the distance between them;

[0010] All wind farms in the same sub-wind power cluster are regarded as nodes of the same output type, and the spatiotemporal relationship of nodes of the same output type is regarded as the isomorphic spatiotemporal relationship to construct an isomorphic graph of the wind power cluster.

[0011] The wind farms in different sub-wind power clusters are regarded as nodes with different output types, and the spatiotemporal relationship of nodes with different output types is regarded as the heterogeneous spatiotemporal relationship to construct a heterogeneous graph of the wind power cluster.

[0012] According to the short-term correlation between wind farms, the adjacency matrix of the homogeneous graph and heterogeneous graph of the wind power cluster is determined;

[0013] Constructing a dynamic graph attention network; wherein the dynamic graph attention network includes a homogeneous module and a heterogeneous module. The homogeneous module calculates the spatiotemporal feature vector of each wind farm based on the homogeneous graph of the wind farm cluster and the wind speed data of each wind farm in the sub-wind farm cluster. The heterogeneous module outputs the predicted power of each wind farm based on the heterogeneous graph of the wind farm cluster and the spatiotemporal feature vector of each wind farm.

[0014] A second data set is constructed, and the dynamic graph attention network is trained using the second data set to realize short-term power prediction of wind power clusters with the trained dynamic graph attention network; wherein the second data set includes wind speed data and wind power data of each wind farm.

[0015] In a possible design, the first data set is clustered to obtain multiple wind power sub-clusters, including:

[0016] Set k initial centroids;

[0017] The Euclidean distance between the wind farm and each centroid is calculated, and the calculation formula is:

[0018]

[0019] In the formula, d(x i ,c m ) represents the Euclidean distance between the wind speed data x i of the i th wind farm and the wind speed data of the m th centroid c m , t represents a time point, x it represents the wind speed data of the i th wind farm at the time point t, and c mt represents the wind speed data of the m th centroid c m at the time point t.

[0020] Based on the Euclidean distance between the wind farm and each centroid, the wind farm is distributed by the following formula:

[0021]

[0022] In the formula, S m represents the m th cluster, and arg min represents the minimum value.

[0023] The centroid is updated by the following formula:

[0024]

[0025] The steps of distribution and updating the centroid are repeated until a set condition is reached, and the set condition is represented as:

[0026]

[0027] or

[0028] t'≥T' (5)

[0029] In the formula, max represents the maximum value function, ε represents a set threshold, represents the wind speed data of the m th centroid at the t'+1 th and t' th iteration, t' represents the current iteration number, and T' represents the maximum iteration number.

[0030] In one possible design, the short-term correlation relationship between wind farms is determined according to the distance between wind farms, including:

[0031] The Euclidean distance between two wind farms is calculated, and the calculation formula is:

[0032]

[0033] where d(W1, W2) represents the Euclidean distance between two wind farms W1 and W2, v 1t and v 2t represent the wind speed data of two wind farms W1 and W2 at time point t;

[0034] Based on the Euclidean distance between two wind farms, the normalization is performed by the following formula:

[0035]

[0036] where d nom represents the normalized distance value, d max represents the maximum value of the Euclidean distance between all wind farms;

[0037] Based on the set distance threshold, the following formula is used to determine whether two wind farms have an STC relationship:

[0038]

[0039] where STC(W1, W2) represents the short-term correlation relationship between wind farms W1 and W2, STC(W1, W2) = 1 indicates that wind farms W1 and W2 have an STC relationship, and STC(W1, W2) = 0 indicates that wind farms W1 and W2 do not have an STC relationship.

[0040] In one possible design, the isomorphic graph representation of the wind power cluster is G Ho = {G Ho,1 ,..., G Ho,c ,..., G Ho,C}, where G Ho,1 , G Ho,c , and G Ho,C represent the isomorphic graph of the first, cth, and Cth sub-wind power cluster, respectively, G Ho,c = (V, E), V and E are the sets of nodes and edges in the isomorphic graph of the cth sub-wind power cluster, V is described as {V1, V2,..., V i}, V1, V2, and V i represent the first, second, and ith node in the same sub-wind power cluster, respectively, E ij represents the directed edge relationship between node i and node j in the same sub-wind power cluster, C n represents the set of nodes in a sub-wind power cluster;

[0041] The heterogeneous graph representation of the wind power cluster is G He = (Φ C , V', E'), Φ Crepresents the meta-path, V' and E' are the sets of nodes and edges in the heterogeneous graph, V' = {V'1,..., V'c}, where V'1, V'c represent the sets of nodes within the first, cth sub-wind power cluster, respectively, c C c C ij E' represents the directed edge relationship between node i and node j within two different sub-wind power clusters, C i and C j represent the sets of nodes within two different sub-wind power clusters.

[0042] In one possible design, according to the short-term correlation relationship between wind farms, the adjacency matrices of the isomorphic graph and the heterogeneous graph of the wind power cluster are determined by the following formula:

[0043] Adjw i,j = STC i,j i = 1,..., n; j = 1,..., n-1 (9)

[0044] In the formula, Adjw i,j represents the adjacency matrix of node i and node j, STC i,j represents the short-term correlation relationship between node i and node j, n is the number of nodes or wind farms, if node i and node j are located in the same sub-wind power cluster, Adjw i,j is the adjacency matrix of the isomorphic graph, if node i and node j are located in different sub-wind power clusters, Adjw i,j is the adjacency matrix of the heterogeneous graph.

[0045] In one possible design, the isomorphic module calculates the spatio-temporal feature vector of each wind farm based on the isomorphic graph of the wind power cluster and the wind speed data of the wind farms in each sub-wind power cluster in the following manner:

[0046] The attention coefficient of each sub-cluster is calculated by the following formula:

[0047] e(x Ci ,x Cj ) = a T [Wx Ci P Wx Cj ] j∈N Ci (13)

[0048] In the formula, e(x Ci ,x Cj ) represents the attention coefficient of node j after nonlinear transformation of node i in sub-wind power cluster C; LeakyReLU(·) is a nonlinear transformation function; a​​​​​T (·) is an attention coefficient; P is a feature concatenation operation; x Ci and x Cj are input feature vectors of node i and node j respectively, W is a shared parameter matrix of linear transformation; N Ci is the total number of neighbor nodes of node i; γ is a positive slope of the negative region of the function;

[0049] The obtained attention coefficient is normalized to obtain an attention weight, and a spatio-temporal vector is obtained by weighted summation, and the calculation process is represented as:

[0050]

[0051]

[0052] In the formula, softmax j (·) is a function of the normalized attention coefficient e Cij of neighbor node j, σ(·) is a sigmoid activation function; K is the number of heads of the multi-head attention mechanism, k is the serial number of the head of the multi-head attention mechanism; exp represents an experience function; α Cij represents an attention weight; represents a spatio-temporal vector; N i represents;

[0053] Based on the spatio-temporal vector, a spatio-temporal feature vector is calculated by the following formula:

[0054]

[0055]

[0056] In the formula, f CAM (·) represents uniformly extracting the spatial features and the temporal features of the wind power cluster, f MLP represents a multi-layer perception guided nonlinear transformation function, AvgPool represents an average pooling layer, MaxPool represents a maximum pooling layer, represents a feature vector obtained after average pooling, represents a feature vector obtained after maximum pooling, f CAM (x Ci ) represents a feature vector set of the original wind farm, e represents a Hamad operator, represents a spatio-temporal feature vector.

[0057] In a possible design, the heterogeneous module outputs the predicted power of each wind farm based on the heterogeneous graph of the wind power cluster and the spatio-temporal feature vector of each wind farm in the following manner:

[0058] Based on the heterogeneous graph of the wind power cluster, a node pair (i, j) and a meta-path Φ connecting the node pair are obtained, and the asymmetric node attention coefficient is calculated using the following formula:

[0059]

[0060] Where, is a transformation matrix that processes the i-th sub-wind power cluster into the same feature space; Φ i is the i-th meta-path; is an asymmetric node attention coefficient; is the meta-path Φ i The total number of wind farms under is the meta-path Φ i The number of adjacent wind farms of wind farm i under ;

[0061] Normalized using softmax(·) to obtain the node-level attention vector based on the meta-path, the calculation formula is:

[0062]

[0063]

[0064] Where, is the node-level attention vector; is the meta-path Φ of node i i Output feature vector; is the meta-path Φ i The total number of wind farms under is the meta-path Φ i The number of adjacent wind farms of wind farm i under [1]; K is the total number of attention heads; k is the current attention head.

[0065] Based on the node-level attention vector, the predicted power of the wind farm is calculated using the following formula:

[0066]

[0067] Where W L is the weight matrix; A is the bias vector; q is the type-level attention vector; is the type-level attention vector; i is the i-th wind farm; Φ i is the meta-path; C is the total number of wind farms in the sub-cluster; tanh is a nonlinear function; b is the weight coefficient; is the type-level attention weight; x pre is the predicted power of the wind farm; P is the number of wind farms under the current meta-path; is the feature vector of node-level output.

[0068] In a second aspect, the present application provides a short-term power prediction device for a wind power cluster, the device comprising:

[0069] The data acquisition module is configured to acquire wind speed data of each wind farm in the wind power cluster to form a first data set, and cluster the first data set to obtain multiple sub-wind power clusters; wherein the first data set is represented by X = {x1, x2, ..., x n}, x1, x2 and x n Respectively representing wind speed data of the first, second, and nth wind farms, the sub-wind farm cluster including at least one wind farm;

[0070] a correlation calculation module configured to determine a short-term correlation relationship between wind farms based on distances between the wind farms;

[0071] The isomorphic graph construction module is configured to treat all wind farms in the same sub-wind power cluster as nodes of the same output type and to construct an isomorphic graph of the wind power cluster using the spatiotemporal relationship of the nodes of the same output type as the isomorphic spatiotemporal relationship;

[0072] A heterogeneous graph construction module is configured to construct a heterogeneous graph of the wind power cluster by treating wind farms in different sub-wind power clusters as nodes of different output types and using the spatiotemporal relationships of the nodes of different output types as heterogeneous spatiotemporal relationships;

[0073] an adjacency matrix building module, configured to determine the adjacency matrices of the homogeneous graph and the heterogeneous graph of the wind farm cluster according to the short-term correlation relationship between the wind farms;

[0074] A graph attention network module is configured to construct a dynamic graph attention network; wherein the dynamic graph attention network includes a homogeneous module and a heterogeneous module. The homogeneous module calculates the spatiotemporal feature vector of each wind farm based on the homogeneous graph of the wind farm cluster and the wind speed data of each wind farm in the sub-wind farm cluster. The heterogeneous module outputs the predicted power of each wind farm based on the heterogeneous graph of the wind farm cluster and the spatiotemporal feature vector of each wind farm.

[0075] The network training module is configured to construct a second data set, use the second data set to train the dynamic graph attention network, and use the trained dynamic graph attention network to realize short-term power prediction of wind power clusters; wherein, the second data set includes wind speed data and wind power data of each wind farm.

[0076] In a third aspect, an electronic device is provided, and the electronic device includes at least one processor and a memory. The memory stores computer-executable instructions. The at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor performs the wind power cluster short-term power prediction method according to the first aspect and various possible designs of the first aspect.

[0077] In a fourth aspect, a computer-readable storage medium is provided, and the computer-readable storage medium stores computer-executable instructions. When a processor executes the computer-executable instructions, the wind power cluster short-term power prediction method according to the first aspect and various possible designs of the first aspect is implemented.

[0078] In a fifth aspect, a computer program product is provided, and the computer program product includes a computer program. When a processor executes the computer program, the wind power cluster short-term power prediction method according to the first aspect and various possible designs of the first aspect is implemented.

[0079] The wind power cluster short-term power prediction method, device, equipment, and storage medium provided by the present application have at least the following beneficial effects:

[0080] The present application fuses homogeneous and heterogeneous information, dynamically constructs a graph structure, to improve the accuracy and reliability of wind farm cluster power prediction, and overcomes the problems of insufficient description of complex spatio-temporal correlation within the wind farm cluster, and inability to dynamically adapt to weather changes in the prior art. BRIEF DESCRIPTION OF DRAWINGS

[0081] The accompanying drawings, which are incorporated into and form a part of the specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the application.

[0082] Figure 1 A flowchart of a wind power cluster short-term power prediction method provided by an embodiment of the present application is provided.

[0083] Figure 2 A clustering flowchart of a first data set provided by an embodiment of the present application is provided.

[0084] Figure 3 A flowchart of determining a short-term correlation relationship between wind farms provided by an embodiment of the present application is provided.

[0085] Figure 4 A construction process of a homogeneous graph and a heterogeneous graph provided by an embodiment of the present application is provided.

[0086] Figure 5 A structure of a dynamic graph attention network and a data processing flowchart thereof provided by an embodiment of the present application are provided.

[0087] Figure 6 A time series diagram of a wind farm prediction result comparison provided by an embodiment of the present application is shown in the following table.

[0088] Figure 7 A structural diagram of a wind power cluster short-term power prediction device provided by an embodiment of the present application is shown in the following table.

[0089] The specific embodiments of the present application have been shown in the above drawings, and will be described in more detail hereinafter. These drawings and the written description are not intended to restrict the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0090] The exemplary embodiments will be described in detail herein below with reference to the drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments are not meant to represent all implementations consistent with the present application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the present application as detailed in the appended claims.

[0091] In the technical solution of the present application, the collection, storage, use, processing, transmission, provision and disclosure of information such as financial data or user data comply with relevant laws and regulations and do not violate public order and good customs.

[0092] It should be noted that in the embodiments of the present application, some existing industry solutions, components, models, etc. may be mentioned, which should be considered as exemplary, and the purpose is only to illustrate the feasibility of the implementation of the technical solution of the present application, but it does not mean that the applicant has or will necessarily use the solution.

[0093] The technical solution of the present application and how the technical solution of the present application solves the above technical problems will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of the present application will be described below with reference to the drawings.

[0094] Based on this, an embodiment of the present application provides a wind power cluster short-term power prediction method, as shown in the following table. Figure 1 A flowchart of a wind power cluster short-term power prediction method provided by an embodiment of the present application is shown in the following table. The wind power cluster short-term power prediction method includes the following steps S100-S700.

[0095] S100: Obtain wind speed data of each wind farm in the wind power cluster to form a first data set, and cluster the first data set to obtain a plurality of sub wind power clusters.

[0096] In this embodiment, the first data set is represented by X={x1, x2, ..., x n}, x1, x2 and x n They represent the wind speed data of the first, second and nth wind farms respectively, and the sub-wind farm cluster includes at least one wind farm.

[0097] In an exemplary embodiment, the K-means clustering method is used to cluster the wind farms in the wind farm cluster based on the wind speed at 96 time points with a resolution of 15 minutes on the same day. First, the wind speed data measured every 15 minutes on the same day for each wind farm in the wind farm cluster is collected to form a data set, in which each row represents a wind farm and each column represents the wind speed at a time point, with a total of 96 columns. Let the first data set be X = {x1, x2, ..., x n}, where x i =(x i1 ,x i2 ,...,x i96 ) represents the wind speed data of the i-th wind farm, and n is the number of wind farms. Next, initialize the K-means clustering process and select k initial centroids C = {c1, c2, ..., c k}, where c m =(c m1 ,c m2 ,...,c m96 ) represents the mth centroid. Figure 2 , clustering of the first data set is achieved through the following iterative process:

[0098] S101: Allocate clusters.

[0099] For each wind farm x i , calculate its relationship with each centroid c j The Euclidean distance between:

[0100]

[0101] Where, d(x i ,c m ) represents the wind speed data x of the i-th wind farm i and the mth centroid c m The Euclidean distance of wind speed data, t represents the time point, x it represents the wind speed data of the i-th wind farm at time point t, c mt represents the mth centroid c m Wind speed data at time t;

[0102] x i Assign to the nearest centroid c m The corresponding cluster S m In Chinese, that is:

[0103]

[0104] where S m denotes the mthcluster, and arg min denotes taking the minimum value;

[0105] S102: updating the centroid.

[0106] calculating the mean value of all wind speed data of each cluster S m and updating the centroid c m :

[0107] updating the centroid by the following formula:

[0108]

[0109] S103: repeating the above steps of assigning a cluster and updating the centroid until a set condition is reached.

[0110] The set condition is that the change of the centroid is less than a set threshold ε or the maximum iteration number T' is reached, i.e.:

[0111]

[0112] or

[0113] t'≥T' (5)

[0114] where max denotes the maximum function, ε denotes the set threshold, denotes the wind speed data of the mthcentroid at the t'+1thiteration and the t'thiteration, t' denotes the current iteration number, and T' denotes the maximum iteration number.

[0115] Finally, each wind farm will be divided into a specific cluster, and the wind farms in the same cluster have high similarity in the wind speed variation trend on the same day, thereby achieving effective clustering of the wind farms in the wind power cluster.

[0116] S200: determining the short-term correlation relationship between the wind farms according to the distance between the wind farms;

[0117] In an exemplary embodiment, the distance between the wind farms is calculated based on the Euclidean distance, and it is determined whether the wind farms have the STC relationship, and whether the wind farms have the STC relationship is used to represent the short-term correlation relationship between the wind farms, please refer to Figure 3 The short-term correlation relationship between the wind farms is determined by the following steps:

[0118] S201: calculating the Euclidean distance.

[0119] Assume that the wind speed data of two wind farms W1 and W2 in a wind power cluster on a certain day are two 96-dimensional vectors v1=[v 11 ,v 12 ,...,v 196 ] and v2=[v 21 ,v 22 ,...,v 296 ], where each element represents the wind speed value at a 15-minute resolution. The Euclidean distance between the two can be calculated using the following formula:

[0120]

[0121] Where d(W1,W2) represents the Euclidean distance between two wind farms W1 and W2, and v 1t and v 2t It represents the wind speed data of two wind farms W1 and W2 at time point t.

[0122] Formula (6) calculates the square root of the sum of the squares of the wind speed differences between the two wind farms at 96 time points.

[0123] S202: Normalization processing.

[0124] In order to eliminate the influence of data scale, the calculated Euclidean distance needs to be normalized. The normalization method is usually to map the distance value to the range of 0 to 1:

[0125]

[0126] Where, d nom Represents the normalized distance value, d max Indicates the maximum value of the Euclidean distance between all wind farms.

[0127] S203: Determination of STC relationship.

[0128] When the normalized distance value d norm When the value is less than the threshold value θ = 0.25, the two wind farms are deemed to have an STC relationship and STC = 1. Otherwise, STC = 0. The judgment formula is:

[0129]

[0130] Where STC(W1, W2) represents the short-term correlation between wind farms W1 and W2. STC(W1, W2) = 1 indicates that there is an STC relationship between wind farms W1 and W2, and STC(W1, W2) = 0 indicates that there is no STC relationship between wind farms W1 and W2.

[0131] Through the above steps S201-S203, the short-term correlation relationship between wind farms in the wind power cluster can be effectively determined, thereby providing technical support for subsequent adjacency matrix.

[0132] S300: All wind farms in the same sub-wind power cluster are taken as nodes of the same output type, and the spatiotemporal relationship between nodes of the same output type is taken as the isomorphic spatiotemporal relationship, thereby constructing the isomorphic graph of the wind power cluster.

[0133] S400: Wind farms in different sub-wind power clusters are taken as nodes of different output types, and the spatiotemporal relationship between nodes of different output types is taken as the heterogeneous spatiotemporal relationship, thereby constructing the heterogeneous graph of the wind power cluster.

[0134] Traditional large-scale wind power cluster power prediction usually divides wind farms into several sub-clusters with similar output habits according to the historical output characteristics of the wind farms, and then models and predicts each sub-cluster. In order to completely construct the spatiotemporal relationship topology of the large-scale wind power cluster, the embodiment continues to refine the different node attributes and relationship attributes of the wind power cluster after dividing the large-scale wind power cluster. Two kinds of spatiotemporal relationships suitable for large-scale wind power clusters and sub-wind power clusters are designed. All wind farms in the same sub-wind power cluster are nodes of the same output type, and the spatiotemporal relationship between nodes of the same output type is isomorphic. The wind farms of different sub-wind power clusters are nodes of different output types, and the spatiotemporal relationship between nodes of different output types is heterogeneous. The homogeneous and heterogeneous spatiotemporal relationship interaction graph is shown in FIG. 1. Figure 4 The large-scale wind power cluster topology graph based on the two kinds of associations is homogeneous and heterogeneous, respectively.

[0135] Based on the above basic principles, the embodiment constructs the isomorphic graph and the heterogeneous graph of the wind power cluster through steps S300 and S400, respectively.

[0136] In an exemplary embodiment, the isomorphic graph of a sub-wind power cluster of the wind power cluster can be represented as G Ho,c =(V, E), where V and E are the sets of nodes and edges, respectively. The isomorphic graph constructed in parallel by multiple sub-wind power clusters is G Ho ={G Ho,1 ,...,G Ho,c ,...,G Ho,C}, C is the total number of sub-wind power clusters, and the V of G Ho,c can be described as {V1, V2,..., V i}, where i ∈ [1, C i ], that is, i nodes in the same sub-wind power cluster. Similarly, where E ij represents the directed edge relationship between node i and node j in the same sub-wind power cluster.

[0137] In an exemplary embodiment, the heterogeneous graph of the wind power cluster can be represented as G He = (Φ C , V', E'), where Φ C represents the meta-path, and V' and E' are the sets of nodes and edges, respectively. The nodes in V' are described as V' = {V'1,..., V' c ,..., V' C}, where V c is the set of nodes within the cth sub-cluster of wind power, i ∈ [1, N c ] is the ith node in V c containing N ij nodes, and the sub-cluster contains n nodes. Similarly, where E' N×N represents the directed edge relationship between nodes i and j within two different sub-clusters. Different types of nodes are used to build different edge relationships between the wind farms.

[0138] In the isomorphic graph, the attributes of nodes and edges are relatively uniform, and the isomorphic GNN is used for feature extraction and relationship modeling. In the heterogeneous graph, the attributes of nodes and edges have different output characteristics and error characteristics, and the processing of the heterogeneous graph requires the heterogeneous GNN to consider the characteristics of each type of node and edge and their interactions.

[0139] S500: Determine the adjacency matrix of the isomorphic graph and the heterogeneous graph of the wind power cluster according to the short-term correlation relationship between the wind farms.

[0140] The adjacency matrix Adj ∈ R i,j is used to describe the relationship between any two connected nodes. It should be noted that if the result of graph relationship identification is STC i = 1 between wind farms WF j and WF i,j , then there is an edge between the two nodes, represented as Adjw i,j = 1. Otherwise, Adjw i,j = 0 indicates that there is no connection between them. This guarantees the sparsity of the adjacency matrix, which helps to reduce unnecessary noise interference.

[0141] In an exemplary embodiment, the adjacency matrix of the isomorphic graph and the heterogeneous graph of the wind power cluster is determined by the following formula:

[0142] Adjw i,j i=1,...,n; j=1,...,n-1 (9)

[0143] In the formula, Adjw i,j represents the adjacency matrix of node i and node j, and STCi,j represents the short-term correlation relationship between node i and node j, n is the number of nodes or wind farms, and Adjw i,j is the adjacency matrix of the homogeneous graph, and Adjw i,j is the adjacency matrix of the heterogeneous graph.

[0144] For the proposed homogeneous graph and heterogeneous graph, the spatial connection relationship of the nodes in multiple modes is determined, which is the key to the HHGAT learning of the spatial relationship. At the same time, more attention needs to be paid to the dynamics of the graph structure, rather than being highly dependent on the pre-defined structure, especially for intermittent wind power. Since the spatio-temporal dependence relationship between different sites evolves over time, in order to enable the constructed graph to dynamically capture this time-varying relationship, in some embodiments, the adjacency matrix is updated by the actual update frequency of the NWP (daily update) to adapt to dynamic and actual engineering applications. Finally, the expressions of the adjacency matrix (equations (10) and (11)) are given below, where cluster a is the homogeneous graph of the a-th sub-wind farm cluster, cluster a-o is the heterogeneous graph between the a-th sub-wind farm cluster and the o-th sub-wind farm cluster, and the edge attribute is i-j, I is the unit matrix.

[0145]

[0146] S600: Construct a dynamic graph attention network; wherein the dynamic graph attention network includes a homogeneous module and a heterogeneous module, the homogeneous module is based on the homogeneous graph of the wind farm cluster, and the spatio-temporal feature vector of each wind farm is calculated according to the wind speed data of the wind farms in each sub-wind farm cluster, and the heterogeneous module is based on the heterogeneous graph of the wind farm cluster and the spatio-temporal feature vector of each wind farm, and outputs the predicted power of each wind farm.

[0147] In this embodiment, the change process of the weight between wind farms s at different times is fitted through the homogeneous module and the heterogeneous module, so as to fully extract the interaction relationship between nodes in the spatio-temporal relationship.

[0148] In an exemplary embodiment, the structure of the dynamic graph attention network and its data processing flow diagram are as shown in Figure 5 .

[0149] The homogeneous module extracts local correlation features. In order to construct the spatio-temporal relationship within the sub-wind farm cluster, the dynamic graph attention network DGAT is used to calculate the local correlation features of each sub-cluster based on the constructed Homogeneous Graph. The specific process can be described as:

[0150]

[0151] e(xCi ,x Cj )=a T [Wx Ci P W Cj ] j∈N Ci (13)

[0152] In the formula, e(x Ci ,x Cj ) represents the attention coefficient of node j to node i in the sub-wind power cluster C after nonlinear transformation; LeakyReLU(·) is a nonlinear transformation function; a T (·) is the attention coefficient; P is the feature concatenation operation; x Ci and x Cj are the input feature vectors of nodes i and j (obtained through the wind speed data transformation of nodes i and j), W is the shared parameter matrix of the linear transformation; N Ci is the total number of neighbor nodes of node i; γ is the positive slope of the function in the negative region.

[0153] Then the obtained attention coefficient is normalized to obtain the attention weight α ij The output vector is obtained by weighted summation

[0154]

[0155]

[0156] Where, softmax j (·) is the normalized attention coefficient e of neighbor node j Cij function, σ(·) is the sigmoid activation function; K is the number of heads of the multi-head attention mechanism, k is the number of heads of the multi-head attention mechanism; exp represents the empirical function; α Cij represents the attention weight; represents the space-time vector; N i Represents the total number of neighbor nodes of node i.

[0157] Finally, through f CAM (·) Uniformly extract the spatial and temporal features of the wind farm cluster. The important node features of the original wind farm s are obtained by performing average pooling and maximum pooling on the features of the wind farm s. After MLP and sigmoid activation operations, the feature vector set f representing the original wind farm s is obtained. CAM (x Ci ), and compare it with the space-time vector extracted by DGAT The spatiotemporal feature vectors of each wind farm are obtained through fusion through isomorphic modules. The calculation process is as follows:

[0158]

[0159] where f CAM denotes the unified extraction of spatial and temporal features of wind power clusters, f MLP denotes the multi-layer perception guided nonlinear transformation function, AvgPool denotes the average pooling layer, MaxPool denotes the maximum pooling layer, denotes the feature vector obtained after average pooling, denotes the feature vector obtained after maximum pooling, f CAM (x Ci ) denotes the feature vector set of the original wind farm, e denotes the Hamada operator, denotes the spatio-temporal feature vector.

[0160] The heterogeneous module extracts global correlation features. Based on the constructed heterogeneous graph, the spatio-temporal relationship within the wind power cluster is constructed using the heterogeneous module HDGAT. The attention mechanism calculation process of the heterogeneous module HDGAT consists of node-level attention mechanism and type-level attention mechanism.

[0161] The calculation process of node-level attention is as follows:

[0162] After processing different sub-cluster wind farms into the same feature space, the self-attention mechanism is used to learn the weights for different types of nodes. Based on the constructed heterogeneous graph, given a node pair (i, j) and the meta path Φ connecting them, the node-level attention is calculated as follows:

[0163]

[0164] where, is a transformation matrix that processes the i-th sub-wind power cluster into the same feature space; Φ i is the i-th meta path; is the asymmetric node attention coefficient; is; is; T is the matrix transpose; is.

[0165] The softmax(·) is used for normalization to obtain the neighbor weight parameter based on the meta path According to the neighbor weight parameter, the feature vector output by node i through the meta path Φ i is calculated, and the calculation process is shown in formulas (20) and (21).

[0166]

[0167] where, is the node-level attention vector; is the feature vector output by node i through the meta path Φi Output feature vector The total number of wind farms under the meta path i The number of adjacent wind farms of the wind farm i under the meta path i

[0168] Node-level attention is used to capture the importance of wind farms with different output characteristics and reduce the weight of noisy nodes.

[0169] The calculation process of type-level attention is as follows:

[0170] Type-level attention aggregates the attention vectors of C groups of nodes obtained in the last step based on multiple meta paths As input, the node embeddings of different output feature sub-clusters are transformed using a nonlinear function. Then these edges are aggregated according to the weights to obtain the final representation x pre .

[0171]

[0172] In the formula, W L is the weight matrix; A is the bias vector; q is the type-level attention vector; is the type-level attention vector; i is the i-th wind farm; F i is the i-th meta path; C is the total number of wind farms in the sub-cluster; tanh is a nonlinear function; b is the weight coefficient; is the type-level attention weight; x pre is the predicted power of the wind farm; P is the number of wind farms under the current meta path; is the node-level output feature vector.

[0173] Type-level attention learns the weights of adjacent sub-clusters of different output feature types and weights the feature vectors.

[0174] S700: Construct a second data set, and train the dynamic graph attention network using the second data set to realize short-term power prediction of the wind farm cluster by the trained dynamic graph attention network; wherein the second data set includes wind speed data and wind power data of each wind farm.

[0175] In this embodiment, the data included in the second data set is divided into NWP data and wind power data p = [p1, p2,..., p n ] T ​​where n represents the number of samples, q represents the number of types of NWP meteorological elements adopted, according to engineering experience, wind speed, wind direction, temperature, pressure and the like are commonly used as explanatory variables of a statistical method model in wind power prediction, and p is the explained variable of the statistical method model as the prediction target. In particular, the predicted wind speed in X will be used as a key meteorological element for dividing wind processes in the following, denoted as v. The data set D = (X, p) composed of the explanatory variables X and the corresponding explained variables p is divided into a training set and a test set according to a certain proportion, the training set data (historical data) is used for fitting of model parameters, and the test set data is used as independent data to verify the prediction effect of the model.

[0176] In the training process, both the homogeneous attention vector a and the heterogeneous attention vector q can be learned through gradient back propagation. So far, the spatiotemporal relationship modeling has integrated the spatiotemporal features in the entire large-scale wind power cluster, representing the future state of all nodes in the wind power cluster.

[0177] The method can quickly and effectively realize the identification and matching of wind processes, and perform targeted integration optimization on the identified different wind processes, thereby improving the accuracy of wind power prediction. To prove that the method can achieve the above effects, the method is tested in a wind power cluster together with five existing models. The five models are as follows:

[0178] Model 1: advanced model Transformer with a large number of parameters.

[0179] Model 2: G-GAT based on the construction of an adjacency matrix based on geographical location.

[0180] Model 3: CBAM-LSTM based on the attention mechanism of the convolution module.

[0181] Model 4: TCN-LSTM based on time series convolution.

[0182] Model 5: linear model LASSO.

[0183] Table 1: prediction index of wind farm 1

[0184] As shown in Table 1, the prediction results of three wind farms are given. The prediction model gives the prediction results of 96 points (with a resolution of 15 minutes) from 00:00 to 23:45 of the next day at 8:00 every day. Among them, each station gives three differentiated models, i.e. models 1-5, and the "proposed optimization model" is the wind power prediction result based on the method. According to the results in Table 1, the proposed optimization model can significantly improve the prediction accuracy. To visually illustrate the effect, Figure 6A time series diagram comparing wind farm prediction results is given. It can be found that the proposed method effectively extracts the spatial characteristics of wind farms to a certain extent and applies them to predict power.

[0185] The present application also provides a short-term power prediction device for a wind power cluster, such as Figure 7 As shown, the wind power cluster short-term power prediction device includes:

[0186] The data acquisition module 701 is configured to acquire the wind speed data of each wind farm in the wind power cluster, forming

[0187] The first data set is clustered to obtain a plurality of wind power sub-clusters; wherein the first data set is represented by X={x1, x2, ..., x n}, x1, x2 and x n Respectively representing wind speed data of the first, second, and nth wind farms, the sub-wind farm cluster including at least one wind farm;

[0188] The correlation calculation module 702 is configured to determine a short-term correlation relationship between wind farms based on the distances between the wind farms;

[0189] The isomorphic graph construction module 703 is configured to construct an isomorphic graph of the wind power cluster by treating all wind farms in the same sub-wind power cluster as nodes of the same output type and using the spatiotemporal relationship of the nodes of the same output type as the isomorphic spatiotemporal relationship;

[0190] The heterogeneous graph construction module 704 is configured to construct a heterogeneous graph of the wind power cluster by treating wind farms in different sub-wind power clusters as nodes of different output types and using the spatiotemporal relationship of the nodes of different output types as a heterogeneous spatiotemporal relationship;

[0191] The adjacency matrix construction module 705 is configured to determine the adjacency matrix of the homogeneous graph and the heterogeneous graph of the wind farm cluster according to the short-term correlation relationship between the wind farms;

[0192] The graph attention network module 706 is configured to construct a dynamic graph attention network; wherein the dynamic graph attention network includes a homogeneous module and a heterogeneous module. The homogeneous module calculates the spatiotemporal feature vector of each wind farm based on the homogeneous graph of the wind farm cluster and the wind speed data of each wind farm in the sub-wind farm cluster. The heterogeneous module outputs the predicted power of each wind farm based on the heterogeneous graph of the wind farm cluster and the spatiotemporal feature vector of each wind farm.

[0193] The network training module 707 is configured to construct a second data set, train the dynamic graph attention network by using the second data set, and realize short-term power prediction of a wind power cluster by using the trained dynamic graph attention network; wherein the second data set comprises wind speed data and wind power data of each wind farm.

[0194] The electronic device provided in the embodiments of the present application can include a processor, a memory, wherein the processor and the memory can communicate; for example, the processor and the memory communicate through a communication bus.

[0195] The processor executes computer-executed instructions stored in the memory, so that the processor executes the solutions in the above embodiments. The processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; and can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component.

[0196] The communication bus can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The system bus can be divided into an address bus, a data bus, a control bus, etc. The transceiver is used to realize communication between the database access device and other computers (for example, a client, a read-write library and a read-only library). The memory can include a random access memory (RAM), and can also include a non-volatile memory.

[0197] The electronic device provided in the embodiments of the present application can be a terminal device of the above embodiments.

[0198] The embodiments of the present application further provide a computer-readable storage medium, which stores computer instructions, and when the computer instructions run on a computer, the computer executes the technical solutions of the short-term power prediction method of the wind power cluster in the above embodiments.

[0199] The embodiment of the present application further provides a computer program product, which comprises a computer program stored in a computer readable storage medium, at least one processor can read the computer program from the computer readable storage medium, and the at least one processor executes the computer program to realize the technical scheme of the wind power cluster short-term power prediction method in the above embodiment.

[0200] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other manners. For example, the above described device embodiments are merely illustrative. For example, the division of the modules is only a logical function division, and there can be another division manner in actual implementation. For example, a plurality of modules or features can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections between different modules can be indirect couplings or communication connections through some interfaces, devices or modules, and can be electrical, mechanical or in other forms.

[0201] The modules illustrated as separated components can or can not be physically separated, and the components illustrated as modules can or can not be physical units, i.e., can be located in one place, or can be distributed on a plurality of network units. Some or all of the modules can be selected according to actual needs to implement the embodiments of the present application.

[0202] In addition, each functional module in each embodiment of the present application can be integrated in one processing unit, or each module can be physically present alone, or two or more modules can be integrated in one unit. The unit formed by the above modules can be realized in the form of hardware, or in the form of hardware plus software function unit.

[0203] The integrated modules realized in the form of software function modules can be stored in a computer readable storage medium. The software function modules stored in the storage medium include a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute some steps of the method of each embodiment of the present application.

[0204] It should be appreciated that the above processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in combination with the application can be directly embodied as hardware processor execution, or executed by a combination of hardware and software modules in the processor.

[0205] The memory can include a high-speed RAM memory, and can also include a non-volatile storage NVM, for example at least one disk memory, and can also be a U disk, a mobile hard disk, a read-only memory, a magnetic disk or an optical disk, etc.

[0206] The bus can be an industry standard architecture (ISA) bus, a peripheral component interconnect (PCI) bus, or an extended industry standard architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc.

[0207] The above storage medium can be realized by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk. The storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0208] An exemplary storage medium is coupled to the processor, so that the processor can read information from the storage medium and can write information to the storage medium. Of course, the storage medium can also be an integral part of the processor. The processor and the storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the storage medium can also exist as discrete components in an electronic control unit or a host device.

[0209] Those skilled in the art can understand that all or part of the steps of the above-mentioned method embodiments can be completed by program instruction related hardware. The foregoing program can be stored in a computer readable storage medium. The program executes to perform the steps of the above-mentioned method embodiments; and the foregoing storage medium includes various storage media that can store program codes, such as ROM, RAM, magnetic disk or optical disk.

[0210] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for short-term power prediction of a wind power cluster, characterized in that: The method comprises: Obtain wind speed data of each wind farm in the wind power cluster to form a first data set, and cluster the first data set to obtain multiple sub-wind power clusters; wherein the first data set is represented by X = {x1, x2, ..., x n }, x1, x2 and x n Respectively representing wind speed data of the first, second, and nth wind farms, the sub-wind farm cluster including at least one wind farm; Determine the short-term correlation between wind farms based on the distance between them; All wind farms in the same sub-wind power cluster are regarded as nodes of the same output type, and the spatiotemporal relationship of nodes of the same output type is regarded as the isomorphic spatiotemporal relationship to construct an isomorphic graph of the wind power cluster. The wind farms in different sub-wind power clusters are regarded as nodes with different output types, and the spatiotemporal relationship of nodes with different output types is regarded as the heterogeneous spatiotemporal relationship to construct a heterogeneous graph of the wind power cluster. According to the short-term correlation between wind farms, the adjacency matrix of the homogeneous graph and heterogeneous graph of the wind power cluster is determined; Constructing a dynamic graph attention network; wherein the dynamic graph attention network includes a homogeneous module and a heterogeneous module. The homogeneous module calculates the spatiotemporal feature vector of each wind farm based on the homogeneous graph of the wind farm cluster and the wind speed data of each wind farm in the sub-wind farm cluster. The heterogeneous module outputs the predicted power of each wind farm based on the heterogeneous graph of the wind farm cluster and the spatiotemporal feature vector of each wind farm. A second data set is constructed, and the dynamic graph attention network is trained using the second data set to realize short-term power prediction of wind power clusters with the trained dynamic graph attention network; wherein the second data set includes wind speed data and wind power data of each wind farm.

2. The method for short-term power prediction of a wind power cluster according to claim 1, characterized in that: Clustering the first data set to obtain multiple wind power sub-clusters, including: Set k initial centroids; Calculate the Euclidean distance between the wind farm and each centroid using the following formula: Where, d(x i ,c m ) represents the wind speed data x of the i-th wind farm i and the mth centroid c m The Euclidean distance of wind speed data, t represents the time point, x it represents the wind speed data of the i-th wind farm at time point t, c mt represents the mth centroid c m Wind speed data at time t; Based on the Euclidean distance between the wind farm and each centroid, the wind farm is allocated using the following formula: Where S m represents the mth cluster, arg min represents the minimum value; Update the centroid using the following formula: The steps of allocating and updating the centroid are repeated until a set condition is reached, which is expressed as: or t'≥T' (5) In the formula, max represents the maximum value function, ε Indicates the set threshold value, Represents the wind speed data of the mth centroid at the t'+1th and t'th iterations, t' represents the current iteration number, and T' represents the maximum iteration number.

3. The method for short-term power prediction of a wind power cluster according to claim 1, characterized in that: Determine the short-term correlation between wind farms based on their distances, including: Calculate the Euclidean distance between two wind farms using the following formula: Where d(W1,W2) represents the Euclidean distance between two wind farms W1 and W2, and v 1t and v 2t represents the wind speed data of two wind farms W1 and W2 at time point t; Based on the Euclidean distance between two wind farms, normalization is performed using the following formula: Where, d nom Represents the normalized distance value, d max represents the maximum value of the Euclidean distance between all wind farms; Based on the set distance threshold, the following formula is used to determine whether two wind farms have an STC relationship: Where STC(W1, W2) represents the short-term correlation between wind farms W1 and W2. STC(W1, W2) = 1 indicates that there is an STC relationship between wind farms W1 and W2, and STC(W1, W2) = 0 indicates that there is no STC relationship between wind farms W1 and W2.

4. The method for short-term power prediction of a wind power cluster according to claim 1, characterized in that: The isomorphic graph of the wind power cluster is represented by G Ho ={G Ho,1 ,...,G Ho,c ,...,G Ho,C }, where G Ho,1 , G Ho,c and G Ho,C They represent the isomorphic graphs of the first, cth and Cth sub-wind power clusters respectively, G Ho,c =(V,E), where V and E are the sets of nodes and edges in the isomorphic graph of the c-th wind farm cluster, and V is described as {V1,V2,...,V i }, V1, V2 and V i Respectively represent the first, second and i-th nodes in the same sub-wind power cluster, E ij represents the directed edge relationship between node i and node j in the same sub-wind power cluster, C n Represents the node set in a sub-wind power cluster; The heterogeneous graph of the wind power cluster is represented as G He =(Φ C ,V',E'),Φ C represents the meta-path, V' and E' are the sets of nodes and edges in the heterogeneous graph, V'={V'1,...,V' c ,...,V' C }, where V'1, V' c and V' C Respectively represent the node sets within the first, cth and Cth sub-wind power clusters, where E' ij represents the directed edge relationship between nodes i and j in two different sub-wind power clusters, C i and C j Represents the internal node sets of two different sub-wind power clusters.

5. The method for short-term power prediction of a wind power cluster according to claim 1 or 3, characterized in that: According to the short-term correlation between wind farms, the adjacency matrix of the homogeneous graph and heterogeneous graph of the wind farm cluster is determined by the following formula: Adjw i,j =STC i,j i=1,...,n;j=1,...,n-1 (9) Where, Adjw i,j Represents the adjacency matrix of node i and node j, STC i,j Represents the short-term correlation between node i and node j, n is the number of nodes or wind farms, if node i and node j are located in the same sub-wind power cluster, then Adjw i,j is the adjacency matrix of the isomorphic graph. If node i and node j are located in different sub-wind power clusters, then Adjw i,j is the adjacency matrix of the heterogeneous graph.

6. The method for short-term power prediction of a wind power cluster according to claim 1, characterized in that: The isomorphic module calculates the spatiotemporal characteristic vector of each wind farm based on the isomorphic graph of the wind farm cluster and the wind speed data of each wind farm in the sub-wind farm cluster. include: The attention coefficient of each sub-cluster is calculated using the following formula: e(x Ci ,x Cj )=a T [Wx Ci PWx Cj ]j∈N Ci (13) In the formula, e(x Ci ,x Cj ) represents the attention coefficient of node j to node i in the sub-wind power cluster C after nonlinear transformation; LeakyReLU(·) is a nonlinear transformation function; a T (·) is the attention coefficient; P is the feature concatenation operation; x Ci and x Cj are the input feature vectors of nodes i and j respectively, W is the shared parameter matrix of linear transformation; N Ci is the total number of neighbor nodes of node i; γ is the positive slope of the negative region of the function; The obtained attention coefficient is normalized to obtain the attention weight, and the space-time vector is obtained by weighted summation. The calculation process is expressed as: Where, softmax j (·) is the normalized attention coefficient e of neighbor node j Cij function, σ(·) is the sigmoid activation function; K is the number of heads of the multi-head attention mechanism, k is the number of heads of the multi-head attention mechanism; exp represents the empirical function; α Cij represents the attention weight; represents the space-time vector; N Ci is the total number of neighbor nodes of node i; Based on the space-time vector, the space-time feature vector is calculated using the following formula: Where, f CAM (·) represents the unified extraction of spatial and temporal characteristics of wind power clusters, f MLP Represents the nonlinear transformation function guided by the multi-layer perceptron, AvgPool represents the average pooling layer, MaxPool represents the maximum pooling layer, represents the feature vector obtained after average pooling, Represents the feature vector obtained after maximum pooling, f CAM (x Ci ) represents the eigenvector set of the original wind farm, e represents the Hamada operator, represents the spatiotemporal feature vector.

7. The method for short-term power prediction of a wind power cluster according to claim 6, characterized in that: The heterogeneous module outputs the predicted power of each wind farm based on the heterogeneous graph of the wind farm cluster and the spatiotemporal feature vector of each wind farm in the following manner: Based on the heterogeneous graph of the wind power cluster, a node pair (i, j) and a meta-path Φ connecting the node pair are obtained, and the asymmetric node attention coefficient is calculated using the following formula: Where, is a transformation matrix that processes the i-th sub-wind power cluster into the same feature space; Φ i is the i-th meta-path; is an asymmetric node attention coefficient; is the metapath Φ i The i-th wind farm under ; is the metapath Φ i The j-th wind farm under ; T is the matrix transpose; is the metapath Φ i The attention vector under Normalized using softmax(·) to obtain the node-level attention vector based on the meta-path, the calculation formula is: Where, is the node-level attention vector; is the meta-path Φ of node i i Output feature vector; is the meta-path Φ i The total number of wind farms under is the meta-path Φ i The number of adjacent wind farms of wind farm i under ; Based on the node-level attention vector, the predicted power of the wind farm is calculated using the following formula: Where W L is the weight matrix; A is the bias vector; q is the type-level attention vector; is the type-level attention vector; i is the i-th wind farm; Φ i is the i-th element path; C is the total number of wind farms in the sub-cluster; tanh is a nonlinear function; b is the weight coefficient; is the type-level attention weight; x pre is the predicted power of the wind farm; P is the number of wind farms under the current meta-path; is the feature vector of node-level output.

8. A short-term power prediction device for a wind power cluster, characterized in that: The device comprises: The data acquisition module is configured to acquire wind speed data of each wind farm in the wind power cluster to form a first data set, and cluster the first data set to obtain multiple sub-wind power clusters; wherein the first data set is represented by X = {x1, x2, ..., x n }, x1, x2 and x n Respectively representing wind speed data of the first, second, and nth wind farms, the sub-wind farm cluster including at least one wind farm; a correlation calculation module configured to determine a short-term correlation relationship between wind farms based on distances between the wind farms; The isomorphic graph construction module is configured to treat all wind farms in the same sub-wind power cluster as nodes of the same output type and to construct an isomorphic graph of the wind power cluster using the spatiotemporal relationship of the nodes of the same output type as the isomorphic spatiotemporal relationship; A heterogeneous graph construction module is configured to construct a heterogeneous graph of the wind power cluster by treating wind farms in different sub-wind power clusters as nodes of different output types and using the spatiotemporal relationships of the nodes of different output types as heterogeneous spatiotemporal relationships; an adjacency matrix building module, configured to determine the adjacency matrices of the homogeneous graph and the heterogeneous graph of the wind farm cluster according to the short-term correlation relationship between the wind farms; A graph attention network module is configured to construct a dynamic graph attention network; wherein the dynamic graph attention network includes a homogeneous module and a heterogeneous module. The homogeneous module calculates the spatiotemporal feature vector of each wind farm based on the homogeneous graph of the wind farm cluster and the wind speed data of each wind farm in the sub-wind farm cluster. The heterogeneous module outputs the predicted power of each wind farm based on the heterogeneous graph of the wind farm cluster and the spatiotemporal feature vector of each wind farm. The network training module is configured to construct a second data set, use the second data set to train the dynamic graph attention network, and use the trained dynamic graph attention network to realize short-term power prediction of wind power clusters; wherein, the second data set includes wind speed data and wind power data of each wind farm.

9. An electronic device, characterized in that: include: a processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the wind power cluster short-term power prediction method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method for short-term power prediction of a wind power cluster according to any one of claims 1 to 7.

Images